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 *);
596 struct dpif_completion {
597 struct list list_node;
598 struct ofoperation *op;
601 /* Extra information about a classifier table.
602 * Currently used just for optimized flow revalidation. */
604 /* If either of these is nonnull, then this table has a form that allows
605 * flows to be tagged to avoid revalidating most flows for the most common
606 * kinds of flow table changes. */
607 struct cls_table *catchall_table; /* Table that wildcards all fields. */
608 struct cls_table *other_table; /* Table with any other wildcard set. */
609 uint32_t basis; /* Keeps each table's tags separate. */
612 /* Reasons that we might need to revalidate every facet, and corresponding
615 * A value of 0 means that there is no need to revalidate.
617 * It would be nice to have some cleaner way to integrate with coverage
618 * counters, but with only a few reasons I guess this is good enough for
620 enum revalidate_reason {
621 REV_RECONFIGURE = 1, /* Switch configuration changed. */
622 REV_STP, /* Spanning tree protocol port status change. */
623 REV_PORT_TOGGLED, /* Port enabled or disabled by CFM, LACP, ...*/
624 REV_FLOW_TABLE, /* Flow table changed. */
625 REV_INCONSISTENCY /* Facet self-check failed. */
627 COVERAGE_DEFINE(rev_reconfigure);
628 COVERAGE_DEFINE(rev_stp);
629 COVERAGE_DEFINE(rev_port_toggled);
630 COVERAGE_DEFINE(rev_flow_table);
631 COVERAGE_DEFINE(rev_inconsistency);
633 /* Drop keys are odp flow keys which have drop flows installed in the kernel.
634 * These are datapath flows which have no associated ofproto, if they did we
635 * would use facets. */
637 struct hmap_node hmap_node;
642 /* All datapaths of a given type share a single dpif backer instance. */
647 struct timer next_expiration;
648 struct hmap odp_to_ofport_map; /* ODP port to ofport mapping. */
650 struct simap tnl_backers; /* Set of dpif ports backing tunnels. */
652 /* Facet revalidation flags applying to facets which use this backer. */
653 enum revalidate_reason need_revalidate; /* Revalidate every facet. */
654 struct tag_set revalidate_set; /* Revalidate only matching facets. */
656 struct hmap drop_keys; /* Set of dropped odp keys. */
659 /* All existing ofproto_backer instances, indexed by ofproto->up.type. */
660 static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers);
662 static void drop_key_clear(struct dpif_backer *);
663 static struct ofport_dpif *
664 odp_port_to_ofport(const struct dpif_backer *, uint32_t odp_port);
666 static void dpif_stats_update_hit_count(struct ofproto_dpif *ofproto,
668 struct avg_subfacet_rates {
669 double add_rate; /* Moving average of new flows created per minute. */
670 double del_rate; /* Moving average of flows deleted per minute. */
672 static void show_dp_rates(struct ds *ds, const char *heading,
673 const struct avg_subfacet_rates *rates);
674 static void exp_mavg(double *avg, int base, double new);
676 struct ofproto_dpif {
677 struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
679 struct dpif_backer *backer;
681 /* Special OpenFlow rules. */
682 struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */
683 struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */
689 struct netflow *netflow;
690 struct dpif_sflow *sflow;
691 struct hmap bundles; /* Contains "struct ofbundle"s. */
692 struct mac_learning *ml;
693 struct ofmirror *mirrors[MAX_MIRRORS];
695 bool has_bonded_bundles;
699 struct hmap subfacets;
700 struct governor *governor;
701 long long int consistency_rl;
704 struct table_dpif tables[N_TABLES];
706 /* Support for debugging async flow mods. */
707 struct list completions;
709 bool has_bundle_action; /* True when the first bundle action appears. */
710 struct netdev_stats stats; /* To account packets generated and consumed in
715 long long int stp_last_tick;
717 /* VLAN splinters. */
718 struct hmap realdev_vid_map; /* (realdev,vid) -> vlandev. */
719 struct hmap vlandev_map; /* vlandev -> (realdev,vid). */
722 struct sset ports; /* Set of standard port names. */
723 struct sset ghost_ports; /* Ports with no datapath port. */
724 struct sset port_poll_set; /* Queued names for port_poll() reply. */
725 int port_poll_errno; /* Last errno for port_poll() reply. */
727 /* Per ofproto's dpif stats. */
731 /* Subfacet statistics.
733 * These keep track of the total number of subfacets added and deleted and
734 * flow life span. They are useful for computing the flow rates stats
735 * exposed via "ovs-appctl dpif/show". The goal is to learn about
736 * traffic patterns in ways that we can use later to improve Open vSwitch
737 * performance in new situations. */
738 long long int created; /* Time when it is created. */
739 unsigned int max_n_subfacet; /* Maximum number of flows */
741 /* The average number of subfacets... */
742 struct avg_subfacet_rates hourly; /* ...over the last hour. */
743 struct avg_subfacet_rates daily; /* ...over the last day. */
744 long long int last_minute; /* Last time 'hourly' was updated. */
746 /* Number of subfacets added or deleted since 'last_minute'. */
747 unsigned int subfacet_add_count;
748 unsigned int subfacet_del_count;
750 /* Number of subfacets added or deleted from 'created' to 'last_minute.' */
751 unsigned long long int total_subfacet_add_count;
752 unsigned long long int total_subfacet_del_count;
754 /* Sum of the number of milliseconds that each subfacet existed,
755 * over the subfacets that have been added and then later deleted. */
756 unsigned long long int total_subfacet_life_span;
758 /* Incremented by the number of currently existing subfacets, each
759 * time we pull statistics from the kernel. */
760 unsigned long long int total_subfacet_count;
762 /* Number of times we pull statistics from the kernel. */
763 unsigned long long int n_update_stats;
765 static unsigned long long int avg_subfacet_life_span(
766 const struct ofproto_dpif *);
767 static double avg_subfacet_count(const struct ofproto_dpif *ofproto);
768 static void update_moving_averages(struct ofproto_dpif *ofproto);
769 static void dpif_stats_update_hit_count(struct ofproto_dpif *ofproto,
771 static void update_max_subfacet_count(struct ofproto_dpif *ofproto);
773 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
774 * for debugging the asynchronous flow_mod implementation.) */
777 /* All existing ofproto_dpif instances, indexed by ->up.name. */
778 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
780 static void ofproto_dpif_unixctl_init(void);
782 static struct ofproto_dpif *
783 ofproto_dpif_cast(const struct ofproto *ofproto)
785 ovs_assert(ofproto->ofproto_class == &ofproto_dpif_class);
786 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
789 static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *,
791 static struct ofport_dpif *get_odp_port(const struct ofproto_dpif *,
793 static void ofproto_trace(struct ofproto_dpif *, const struct flow *,
794 const struct ofpbuf *,
795 const struct initial_vals *, struct ds *);
797 /* Packet processing. */
798 static void update_learning_table(struct ofproto_dpif *,
799 const struct flow *, int vlan,
802 #define FLOW_MISS_MAX_BATCH 50
803 static int handle_upcalls(struct dpif_backer *, unsigned int max_batch);
805 /* Flow expiration. */
806 static int expire(struct dpif_backer *);
809 static void send_netflow_active_timeouts(struct ofproto_dpif *);
812 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
813 static size_t compose_sflow_action(const struct ofproto_dpif *,
814 struct ofpbuf *odp_actions,
815 const struct flow *, uint32_t odp_port);
816 static void add_mirror_actions(struct action_xlate_ctx *ctx,
817 const struct flow *flow);
818 /* Global variables. */
819 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
821 /* Initial mappings of port to bridge mappings. */
822 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
824 /* Factory functions. */
827 init(const struct shash *iface_hints)
829 struct shash_node *node;
831 /* Make a local copy, since we don't own 'iface_hints' elements. */
832 SHASH_FOR_EACH(node, iface_hints) {
833 const struct iface_hint *orig_hint = node->data;
834 struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
836 new_hint->br_name = xstrdup(orig_hint->br_name);
837 new_hint->br_type = xstrdup(orig_hint->br_type);
838 new_hint->ofp_port = orig_hint->ofp_port;
840 shash_add(&init_ofp_ports, node->name, new_hint);
845 enumerate_types(struct sset *types)
847 dp_enumerate_types(types);
851 enumerate_names(const char *type, struct sset *names)
853 struct ofproto_dpif *ofproto;
856 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
857 if (strcmp(type, ofproto->up.type)) {
860 sset_add(names, ofproto->up.name);
867 del(const char *type, const char *name)
872 error = dpif_open(name, type, &dpif);
874 error = dpif_delete(dpif);
881 port_open_type(const char *datapath_type, const char *port_type)
883 return dpif_port_open_type(datapath_type, port_type);
886 /* Type functions. */
888 static struct ofproto_dpif *
889 lookup_ofproto_dpif_by_port_name(const char *name)
891 struct ofproto_dpif *ofproto;
893 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
894 if (sset_contains(&ofproto->ports, name)) {
903 type_run(const char *type)
905 struct dpif_backer *backer;
909 backer = shash_find_data(&all_dpif_backers, type);
911 /* This is not necessarily a problem, since backers are only
912 * created on demand. */
916 dpif_run(backer->dpif);
918 if (backer->need_revalidate
919 || !tag_set_is_empty(&backer->revalidate_set)) {
920 struct tag_set revalidate_set = backer->revalidate_set;
921 bool need_revalidate = backer->need_revalidate;
922 struct ofproto_dpif *ofproto;
923 struct simap_node *node;
924 struct simap tmp_backers;
926 /* Handle tunnel garbage collection. */
927 simap_init(&tmp_backers);
928 simap_swap(&backer->tnl_backers, &tmp_backers);
930 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
931 struct ofport_dpif *iter;
933 if (backer != ofproto->backer) {
937 HMAP_FOR_EACH (iter, up.hmap_node, &ofproto->up.ports) {
940 if (!iter->tnl_port) {
944 dp_port = netdev_vport_get_dpif_port(iter->up.netdev);
945 node = simap_find(&tmp_backers, dp_port);
947 simap_put(&backer->tnl_backers, dp_port, node->data);
948 simap_delete(&tmp_backers, node);
949 node = simap_find(&backer->tnl_backers, dp_port);
951 node = simap_find(&backer->tnl_backers, dp_port);
953 uint32_t odp_port = UINT32_MAX;
955 if (!dpif_port_add(backer->dpif, iter->up.netdev,
957 simap_put(&backer->tnl_backers, dp_port, odp_port);
958 node = simap_find(&backer->tnl_backers, dp_port);
963 iter->odp_port = node ? node->data : OVSP_NONE;
964 if (tnl_port_reconfigure(&iter->up, iter->odp_port,
966 backer->need_revalidate = REV_RECONFIGURE;
971 SIMAP_FOR_EACH (node, &tmp_backers) {
972 dpif_port_del(backer->dpif, node->data);
974 simap_destroy(&tmp_backers);
976 switch (backer->need_revalidate) {
977 case REV_RECONFIGURE: COVERAGE_INC(rev_reconfigure); break;
978 case REV_STP: COVERAGE_INC(rev_stp); break;
979 case REV_PORT_TOGGLED: COVERAGE_INC(rev_port_toggled); break;
980 case REV_FLOW_TABLE: COVERAGE_INC(rev_flow_table); break;
981 case REV_INCONSISTENCY: COVERAGE_INC(rev_inconsistency); break;
984 if (backer->need_revalidate) {
985 /* Clear the drop_keys in case we should now be accepting some
986 * formerly dropped flows. */
987 drop_key_clear(backer);
990 /* Clear the revalidation flags. */
991 tag_set_init(&backer->revalidate_set);
992 backer->need_revalidate = 0;
994 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
995 struct facet *facet, *next;
997 if (ofproto->backer != backer) {
1001 HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &ofproto->facets) {
1003 || tag_set_intersects(&revalidate_set, facet->tags)) {
1004 facet_revalidate(facet);
1010 if (timer_expired(&backer->next_expiration)) {
1011 int delay = expire(backer);
1012 timer_set_duration(&backer->next_expiration, delay);
1015 /* Check for port changes in the dpif. */
1016 while ((error = dpif_port_poll(backer->dpif, &devname)) == 0) {
1017 struct ofproto_dpif *ofproto;
1018 struct dpif_port port;
1020 /* Don't report on the datapath's device. */
1021 if (!strcmp(devname, dpif_base_name(backer->dpif))) {
1025 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
1026 &all_ofproto_dpifs) {
1027 if (simap_contains(&ofproto->backer->tnl_backers, devname)) {
1032 ofproto = lookup_ofproto_dpif_by_port_name(devname);
1033 if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
1034 /* The port was removed. If we know the datapath,
1035 * report it through poll_set(). If we don't, it may be
1036 * notifying us of a removal we initiated, so ignore it.
1037 * If there's a pending ENOBUFS, let it stand, since
1038 * everything will be reevaluated. */
1039 if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
1040 sset_add(&ofproto->port_poll_set, devname);
1041 ofproto->port_poll_errno = 0;
1043 } else if (!ofproto) {
1044 /* The port was added, but we don't know with which
1045 * ofproto we should associate it. Delete it. */
1046 dpif_port_del(backer->dpif, port.port_no);
1048 dpif_port_destroy(&port);
1054 if (error != EAGAIN) {
1055 struct ofproto_dpif *ofproto;
1057 /* There was some sort of error, so propagate it to all
1058 * ofprotos that use this backer. */
1059 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
1060 &all_ofproto_dpifs) {
1061 if (ofproto->backer == backer) {
1062 sset_clear(&ofproto->port_poll_set);
1063 ofproto->port_poll_errno = error;
1072 type_run_fast(const char *type)
1074 struct dpif_backer *backer;
1077 backer = shash_find_data(&all_dpif_backers, type);
1079 /* This is not necessarily a problem, since backers are only
1080 * created on demand. */
1084 /* Handle one or more batches of upcalls, until there's nothing left to do
1085 * or until we do a fixed total amount of work.
1087 * We do work in batches because it can be much cheaper to set up a number
1088 * of flows and fire off their patches all at once. We do multiple batches
1089 * because in some cases handling a packet can cause another packet to be
1090 * queued almost immediately as part of the return flow. Both
1091 * optimizations can make major improvements on some benchmarks and
1092 * presumably for real traffic as well. */
1094 while (work < FLOW_MISS_MAX_BATCH) {
1095 int retval = handle_upcalls(backer, FLOW_MISS_MAX_BATCH - work);
1106 type_wait(const char *type)
1108 struct dpif_backer *backer;
1110 backer = shash_find_data(&all_dpif_backers, type);
1112 /* This is not necessarily a problem, since backers are only
1113 * created on demand. */
1117 timer_wait(&backer->next_expiration);
1120 /* Basic life-cycle. */
1122 static int add_internal_flows(struct ofproto_dpif *);
1124 static struct ofproto *
1127 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
1128 return &ofproto->up;
1132 dealloc(struct ofproto *ofproto_)
1134 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1139 close_dpif_backer(struct dpif_backer *backer)
1141 struct shash_node *node;
1143 ovs_assert(backer->refcount > 0);
1145 if (--backer->refcount) {
1149 drop_key_clear(backer);
1150 hmap_destroy(&backer->drop_keys);
1152 simap_destroy(&backer->tnl_backers);
1153 hmap_destroy(&backer->odp_to_ofport_map);
1154 node = shash_find(&all_dpif_backers, backer->type);
1156 shash_delete(&all_dpif_backers, node);
1157 dpif_close(backer->dpif);
1162 /* Datapath port slated for removal from datapath. */
1163 struct odp_garbage {
1164 struct list list_node;
1169 open_dpif_backer(const char *type, struct dpif_backer **backerp)
1171 struct dpif_backer *backer;
1172 struct dpif_port_dump port_dump;
1173 struct dpif_port port;
1174 struct shash_node *node;
1175 struct list garbage_list;
1176 struct odp_garbage *garbage, *next;
1182 backer = shash_find_data(&all_dpif_backers, type);
1189 backer_name = xasprintf("ovs-%s", type);
1191 /* Remove any existing datapaths, since we assume we're the only
1192 * userspace controlling the datapath. */
1194 dp_enumerate_names(type, &names);
1195 SSET_FOR_EACH(name, &names) {
1196 struct dpif *old_dpif;
1198 /* Don't remove our backer if it exists. */
1199 if (!strcmp(name, backer_name)) {
1203 if (dpif_open(name, type, &old_dpif)) {
1204 VLOG_WARN("couldn't open old datapath %s to remove it", name);
1206 dpif_delete(old_dpif);
1207 dpif_close(old_dpif);
1210 sset_destroy(&names);
1212 backer = xmalloc(sizeof *backer);
1214 error = dpif_create_and_open(backer_name, type, &backer->dpif);
1217 VLOG_ERR("failed to open datapath of type %s: %s", type,
1223 backer->type = xstrdup(type);
1224 backer->refcount = 1;
1225 hmap_init(&backer->odp_to_ofport_map);
1226 hmap_init(&backer->drop_keys);
1227 timer_set_duration(&backer->next_expiration, 1000);
1228 backer->need_revalidate = 0;
1229 simap_init(&backer->tnl_backers);
1230 tag_set_init(&backer->revalidate_set);
1233 dpif_flow_flush(backer->dpif);
1235 /* Loop through the ports already on the datapath and remove any
1236 * that we don't need anymore. */
1237 list_init(&garbage_list);
1238 dpif_port_dump_start(&port_dump, backer->dpif);
1239 while (dpif_port_dump_next(&port_dump, &port)) {
1240 node = shash_find(&init_ofp_ports, port.name);
1241 if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
1242 garbage = xmalloc(sizeof *garbage);
1243 garbage->odp_port = port.port_no;
1244 list_push_front(&garbage_list, &garbage->list_node);
1247 dpif_port_dump_done(&port_dump);
1249 LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) {
1250 dpif_port_del(backer->dpif, garbage->odp_port);
1251 list_remove(&garbage->list_node);
1255 shash_add(&all_dpif_backers, type, backer);
1257 error = dpif_recv_set(backer->dpif, true);
1259 VLOG_ERR("failed to listen on datapath of type %s: %s",
1260 type, strerror(error));
1261 close_dpif_backer(backer);
1269 construct(struct ofproto *ofproto_)
1271 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1272 struct shash_node *node, *next;
1277 error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1282 max_ports = dpif_get_max_ports(ofproto->backer->dpif);
1283 ofproto_init_max_ports(ofproto_, MIN(max_ports, OFPP_MAX));
1285 ofproto->n_matches = 0;
1287 ofproto->netflow = NULL;
1288 ofproto->sflow = NULL;
1289 ofproto->stp = NULL;
1290 hmap_init(&ofproto->bundles);
1291 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1292 for (i = 0; i < MAX_MIRRORS; i++) {
1293 ofproto->mirrors[i] = NULL;
1295 ofproto->has_bonded_bundles = false;
1297 hmap_init(&ofproto->facets);
1298 hmap_init(&ofproto->subfacets);
1299 ofproto->governor = NULL;
1300 ofproto->consistency_rl = LLONG_MIN;
1302 for (i = 0; i < N_TABLES; i++) {
1303 struct table_dpif *table = &ofproto->tables[i];
1305 table->catchall_table = NULL;
1306 table->other_table = NULL;
1307 table->basis = random_uint32();
1310 list_init(&ofproto->completions);
1312 ofproto_dpif_unixctl_init();
1314 ofproto->has_mirrors = false;
1315 ofproto->has_bundle_action = false;
1317 hmap_init(&ofproto->vlandev_map);
1318 hmap_init(&ofproto->realdev_vid_map);
1320 sset_init(&ofproto->ports);
1321 sset_init(&ofproto->ghost_ports);
1322 sset_init(&ofproto->port_poll_set);
1323 ofproto->port_poll_errno = 0;
1325 SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1326 struct iface_hint *iface_hint = node->data;
1328 if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1329 /* Check if the datapath already has this port. */
1330 if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1331 sset_add(&ofproto->ports, node->name);
1334 free(iface_hint->br_name);
1335 free(iface_hint->br_type);
1337 shash_delete(&init_ofp_ports, node);
1341 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1342 hash_string(ofproto->up.name, 0));
1343 memset(&ofproto->stats, 0, sizeof ofproto->stats);
1345 ofproto_init_tables(ofproto_, N_TABLES);
1346 error = add_internal_flows(ofproto);
1347 ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1350 ofproto->n_missed = 0;
1352 ofproto->max_n_subfacet = 0;
1353 ofproto->created = time_msec();
1354 ofproto->last_minute = ofproto->created;
1355 memset(&ofproto->hourly, 0, sizeof ofproto->hourly);
1356 memset(&ofproto->daily, 0, sizeof ofproto->daily);
1357 ofproto->subfacet_add_count = 0;
1358 ofproto->subfacet_del_count = 0;
1359 ofproto->total_subfacet_add_count = 0;
1360 ofproto->total_subfacet_del_count = 0;
1361 ofproto->total_subfacet_life_span = 0;
1362 ofproto->total_subfacet_count = 0;
1363 ofproto->n_update_stats = 0;
1369 add_internal_flow(struct ofproto_dpif *ofproto, int id,
1370 const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1372 struct ofputil_flow_mod fm;
1375 match_init_catchall(&fm.match);
1377 match_set_reg(&fm.match, 0, id);
1378 fm.new_cookie = htonll(0);
1379 fm.cookie = htonll(0);
1380 fm.cookie_mask = htonll(0);
1381 fm.table_id = TBL_INTERNAL;
1382 fm.command = OFPFC_ADD;
1383 fm.idle_timeout = 0;
1384 fm.hard_timeout = 0;
1388 fm.ofpacts = ofpacts->data;
1389 fm.ofpacts_len = ofpacts->size;
1391 error = ofproto_flow_mod(&ofproto->up, &fm);
1393 VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
1394 id, ofperr_to_string(error));
1398 *rulep = rule_dpif_lookup__(ofproto, &fm.match.flow, TBL_INTERNAL);
1399 ovs_assert(*rulep != NULL);
1405 add_internal_flows(struct ofproto_dpif *ofproto)
1407 struct ofpact_controller *controller;
1408 uint64_t ofpacts_stub[128 / 8];
1409 struct ofpbuf ofpacts;
1413 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1416 controller = ofpact_put_CONTROLLER(&ofpacts);
1417 controller->max_len = UINT16_MAX;
1418 controller->controller_id = 0;
1419 controller->reason = OFPR_NO_MATCH;
1420 ofpact_pad(&ofpacts);
1422 error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule);
1427 ofpbuf_clear(&ofpacts);
1428 error = add_internal_flow(ofproto, id++, &ofpacts,
1429 &ofproto->no_packet_in_rule);
1434 complete_operations(struct ofproto_dpif *ofproto)
1436 struct dpif_completion *c, *next;
1438 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
1439 ofoperation_complete(c->op, 0);
1440 list_remove(&c->list_node);
1446 destruct(struct ofproto *ofproto_)
1448 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1449 struct rule_dpif *rule, *next_rule;
1450 struct oftable *table;
1453 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1454 complete_operations(ofproto);
1456 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1457 struct cls_cursor cursor;
1459 cls_cursor_init(&cursor, &table->cls, NULL);
1460 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1461 ofproto_rule_destroy(&rule->up);
1465 for (i = 0; i < MAX_MIRRORS; i++) {
1466 mirror_destroy(ofproto->mirrors[i]);
1469 netflow_destroy(ofproto->netflow);
1470 dpif_sflow_destroy(ofproto->sflow);
1471 hmap_destroy(&ofproto->bundles);
1472 mac_learning_destroy(ofproto->ml);
1474 hmap_destroy(&ofproto->facets);
1475 hmap_destroy(&ofproto->subfacets);
1476 governor_destroy(ofproto->governor);
1478 hmap_destroy(&ofproto->vlandev_map);
1479 hmap_destroy(&ofproto->realdev_vid_map);
1481 sset_destroy(&ofproto->ports);
1482 sset_destroy(&ofproto->ghost_ports);
1483 sset_destroy(&ofproto->port_poll_set);
1485 close_dpif_backer(ofproto->backer);
1489 run_fast(struct ofproto *ofproto_)
1491 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1492 struct ofport_dpif *ofport;
1494 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1495 port_run_fast(ofport);
1502 run(struct ofproto *ofproto_)
1504 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1505 struct ofport_dpif *ofport;
1506 struct ofbundle *bundle;
1510 complete_operations(ofproto);
1513 error = run_fast(ofproto_);
1518 if (ofproto->netflow) {
1519 if (netflow_run(ofproto->netflow)) {
1520 send_netflow_active_timeouts(ofproto);
1523 if (ofproto->sflow) {
1524 dpif_sflow_run(ofproto->sflow);
1527 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1530 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1535 mac_learning_run(ofproto->ml, &ofproto->backer->revalidate_set);
1537 /* Check the consistency of a random facet, to aid debugging. */
1538 if (time_msec() >= ofproto->consistency_rl
1539 && !hmap_is_empty(&ofproto->facets)
1540 && !ofproto->backer->need_revalidate) {
1541 struct facet *facet;
1543 ofproto->consistency_rl = time_msec() + 250;
1545 facet = CONTAINER_OF(hmap_random_node(&ofproto->facets),
1546 struct facet, hmap_node);
1547 if (!tag_set_intersects(&ofproto->backer->revalidate_set,
1549 if (!facet_check_consistency(facet)) {
1550 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
1555 if (ofproto->governor) {
1558 governor_run(ofproto->governor);
1560 /* If the governor has shrunk to its minimum size and the number of
1561 * subfacets has dwindled, then drop the governor entirely.
1563 * For hysteresis, the number of subfacets to drop the governor is
1564 * smaller than the number needed to trigger its creation. */
1565 n_subfacets = hmap_count(&ofproto->subfacets);
1566 if (n_subfacets * 4 < ofproto->up.flow_eviction_threshold
1567 && governor_is_idle(ofproto->governor)) {
1568 governor_destroy(ofproto->governor);
1569 ofproto->governor = NULL;
1577 wait(struct ofproto *ofproto_)
1579 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1580 struct ofport_dpif *ofport;
1581 struct ofbundle *bundle;
1583 if (!clogged && !list_is_empty(&ofproto->completions)) {
1584 poll_immediate_wake();
1587 dpif_wait(ofproto->backer->dpif);
1588 dpif_recv_wait(ofproto->backer->dpif);
1589 if (ofproto->sflow) {
1590 dpif_sflow_wait(ofproto->sflow);
1592 if (!tag_set_is_empty(&ofproto->backer->revalidate_set)) {
1593 poll_immediate_wake();
1595 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1598 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1599 bundle_wait(bundle);
1601 if (ofproto->netflow) {
1602 netflow_wait(ofproto->netflow);
1604 mac_learning_wait(ofproto->ml);
1606 if (ofproto->backer->need_revalidate) {
1607 /* Shouldn't happen, but if it does just go around again. */
1608 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1609 poll_immediate_wake();
1611 if (ofproto->governor) {
1612 governor_wait(ofproto->governor);
1617 get_memory_usage(const struct ofproto *ofproto_, struct simap *usage)
1619 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1621 simap_increase(usage, "facets", hmap_count(&ofproto->facets));
1622 simap_increase(usage, "subfacets", hmap_count(&ofproto->subfacets));
1626 flush(struct ofproto *ofproto_)
1628 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1629 struct subfacet *subfacet, *next_subfacet;
1630 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
1634 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
1635 &ofproto->subfacets) {
1636 if (subfacet->path != SF_NOT_INSTALLED) {
1637 batch[n_batch++] = subfacet;
1638 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
1639 subfacet_destroy_batch(ofproto, batch, n_batch);
1643 subfacet_destroy(subfacet);
1648 subfacet_destroy_batch(ofproto, batch, n_batch);
1653 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1654 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1656 *arp_match_ip = true;
1657 *actions = (OFPUTIL_A_OUTPUT |
1658 OFPUTIL_A_SET_VLAN_VID |
1659 OFPUTIL_A_SET_VLAN_PCP |
1660 OFPUTIL_A_STRIP_VLAN |
1661 OFPUTIL_A_SET_DL_SRC |
1662 OFPUTIL_A_SET_DL_DST |
1663 OFPUTIL_A_SET_NW_SRC |
1664 OFPUTIL_A_SET_NW_DST |
1665 OFPUTIL_A_SET_NW_TOS |
1666 OFPUTIL_A_SET_TP_SRC |
1667 OFPUTIL_A_SET_TP_DST |
1672 get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots)
1674 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1675 struct dpif_dp_stats s;
1677 strcpy(ots->name, "classifier");
1679 dpif_get_dp_stats(ofproto->backer->dpif, &s);
1681 ots->lookup_count = htonll(s.n_hit + s.n_missed);
1682 ots->matched_count = htonll(s.n_hit + ofproto->n_matches);
1685 static struct ofport *
1688 struct ofport_dpif *port = xmalloc(sizeof *port);
1693 port_dealloc(struct ofport *port_)
1695 struct ofport_dpif *port = ofport_dpif_cast(port_);
1700 port_construct(struct ofport *port_)
1702 struct ofport_dpif *port = ofport_dpif_cast(port_);
1703 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1704 const struct netdev *netdev = port->up.netdev;
1705 struct dpif_port dpif_port;
1708 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1709 port->bundle = NULL;
1711 port->tag = tag_create_random();
1712 port->may_enable = true;
1713 port->stp_port = NULL;
1714 port->stp_state = STP_DISABLED;
1715 port->tnl_port = NULL;
1716 hmap_init(&port->priorities);
1717 port->realdev_ofp_port = 0;
1718 port->vlandev_vid = 0;
1719 port->carrier_seq = netdev_get_carrier_resets(netdev);
1721 if (netdev_vport_is_patch(netdev)) {
1722 /* XXX By bailing out here, we don't do required sFlow work. */
1723 port->odp_port = OVSP_NONE;
1727 error = dpif_port_query_by_name(ofproto->backer->dpif,
1728 netdev_vport_get_dpif_port(netdev),
1734 port->odp_port = dpif_port.port_no;
1736 if (netdev_get_tunnel_config(netdev)) {
1737 port->tnl_port = tnl_port_add(&port->up, port->odp_port);
1739 /* Sanity-check that a mapping doesn't already exist. This
1740 * shouldn't happen for non-tunnel ports. */
1741 if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1742 VLOG_ERR("port %s already has an OpenFlow port number",
1744 dpif_port_destroy(&dpif_port);
1748 hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1749 hash_int(port->odp_port, 0));
1751 dpif_port_destroy(&dpif_port);
1753 if (ofproto->sflow) {
1754 dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1761 port_destruct(struct ofport *port_)
1763 struct ofport_dpif *port = ofport_dpif_cast(port_);
1764 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1765 const char *dp_port_name = netdev_vport_get_dpif_port(port->up.netdev);
1766 const char *devname = netdev_get_name(port->up.netdev);
1768 if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1769 /* The underlying device is still there, so delete it. This
1770 * happens when the ofproto is being destroyed, since the caller
1771 * assumes that removal of attached ports will happen as part of
1773 if (!port->tnl_port) {
1774 dpif_port_del(ofproto->backer->dpif, port->odp_port);
1776 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1779 if (port->odp_port != OVSP_NONE && !port->tnl_port) {
1780 hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1783 tnl_port_del(port->tnl_port);
1784 sset_find_and_delete(&ofproto->ports, devname);
1785 sset_find_and_delete(&ofproto->ghost_ports, devname);
1786 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1787 bundle_remove(port_);
1788 set_cfm(port_, NULL);
1789 if (ofproto->sflow) {
1790 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1793 ofport_clear_priorities(port);
1794 hmap_destroy(&port->priorities);
1798 port_modified(struct ofport *port_)
1800 struct ofport_dpif *port = ofport_dpif_cast(port_);
1802 if (port->bundle && port->bundle->bond) {
1803 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1808 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1810 struct ofport_dpif *port = ofport_dpif_cast(port_);
1811 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1812 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1814 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1815 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1816 OFPUTIL_PC_NO_PACKET_IN)) {
1817 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1819 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1820 bundle_update(port->bundle);
1826 set_sflow(struct ofproto *ofproto_,
1827 const struct ofproto_sflow_options *sflow_options)
1829 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1830 struct dpif_sflow *ds = ofproto->sflow;
1832 if (sflow_options) {
1834 struct ofport_dpif *ofport;
1836 ds = ofproto->sflow = dpif_sflow_create();
1837 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1838 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1840 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1842 dpif_sflow_set_options(ds, sflow_options);
1845 dpif_sflow_destroy(ds);
1846 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1847 ofproto->sflow = NULL;
1854 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1856 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1863 struct ofproto_dpif *ofproto;
1865 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1866 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1867 ofport->cfm = cfm_create(netdev_get_name(ofport->up.netdev));
1870 if (cfm_configure(ofport->cfm, s)) {
1876 cfm_destroy(ofport->cfm);
1882 get_cfm_fault(const struct ofport *ofport_)
1884 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1886 return ofport->cfm ? cfm_get_fault(ofport->cfm) : -1;
1890 get_cfm_opup(const struct ofport *ofport_)
1892 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1894 return ofport->cfm ? cfm_get_opup(ofport->cfm) : -1;
1898 get_cfm_remote_mpids(const struct ofport *ofport_, const uint64_t **rmps,
1901 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1904 cfm_get_remote_mpids(ofport->cfm, rmps, n_rmps);
1912 get_cfm_health(const struct ofport *ofport_)
1914 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1916 return ofport->cfm ? cfm_get_health(ofport->cfm) : -1;
1919 /* Spanning Tree. */
1922 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
1924 struct ofproto_dpif *ofproto = ofproto_;
1925 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
1926 struct ofport_dpif *ofport;
1928 ofport = stp_port_get_aux(sp);
1930 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
1931 ofproto->up.name, port_num);
1933 struct eth_header *eth = pkt->l2;
1935 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
1936 if (eth_addr_is_zero(eth->eth_src)) {
1937 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
1938 "with unknown MAC", ofproto->up.name, port_num);
1940 send_packet(ofport, pkt);
1946 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
1948 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
1950 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1952 /* Only revalidate flows if the configuration changed. */
1953 if (!s != !ofproto->stp) {
1954 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1958 if (!ofproto->stp) {
1959 ofproto->stp = stp_create(ofproto_->name, s->system_id,
1960 send_bpdu_cb, ofproto);
1961 ofproto->stp_last_tick = time_msec();
1964 stp_set_bridge_id(ofproto->stp, s->system_id);
1965 stp_set_bridge_priority(ofproto->stp, s->priority);
1966 stp_set_hello_time(ofproto->stp, s->hello_time);
1967 stp_set_max_age(ofproto->stp, s->max_age);
1968 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
1970 struct ofport *ofport;
1972 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
1973 set_stp_port(ofport, NULL);
1976 stp_destroy(ofproto->stp);
1977 ofproto->stp = NULL;
1984 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
1986 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1990 s->bridge_id = stp_get_bridge_id(ofproto->stp);
1991 s->designated_root = stp_get_designated_root(ofproto->stp);
1992 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
2001 update_stp_port_state(struct ofport_dpif *ofport)
2003 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2004 enum stp_state state;
2006 /* Figure out new state. */
2007 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
2011 if (ofport->stp_state != state) {
2012 enum ofputil_port_state of_state;
2015 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
2016 netdev_get_name(ofport->up.netdev),
2017 stp_state_name(ofport->stp_state),
2018 stp_state_name(state));
2019 if (stp_learn_in_state(ofport->stp_state)
2020 != stp_learn_in_state(state)) {
2021 /* xxx Learning action flows should also be flushed. */
2022 mac_learning_flush(ofproto->ml,
2023 &ofproto->backer->revalidate_set);
2025 fwd_change = stp_forward_in_state(ofport->stp_state)
2026 != stp_forward_in_state(state);
2028 ofproto->backer->need_revalidate = REV_STP;
2029 ofport->stp_state = state;
2030 ofport->stp_state_entered = time_msec();
2032 if (fwd_change && ofport->bundle) {
2033 bundle_update(ofport->bundle);
2036 /* Update the STP state bits in the OpenFlow port description. */
2037 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
2038 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
2039 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
2040 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
2041 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
2043 ofproto_port_set_state(&ofport->up, of_state);
2047 /* Configures STP on 'ofport_' using the settings defined in 's'. The
2048 * caller is responsible for assigning STP port numbers and ensuring
2049 * there are no duplicates. */
2051 set_stp_port(struct ofport *ofport_,
2052 const struct ofproto_port_stp_settings *s)
2054 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2055 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2056 struct stp_port *sp = ofport->stp_port;
2058 if (!s || !s->enable) {
2060 ofport->stp_port = NULL;
2061 stp_port_disable(sp);
2062 update_stp_port_state(ofport);
2065 } else if (sp && stp_port_no(sp) != s->port_num
2066 && ofport == stp_port_get_aux(sp)) {
2067 /* The port-id changed, so disable the old one if it's not
2068 * already in use by another port. */
2069 stp_port_disable(sp);
2072 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
2073 stp_port_enable(sp);
2075 stp_port_set_aux(sp, ofport);
2076 stp_port_set_priority(sp, s->priority);
2077 stp_port_set_path_cost(sp, s->path_cost);
2079 update_stp_port_state(ofport);
2085 get_stp_port_status(struct ofport *ofport_,
2086 struct ofproto_port_stp_status *s)
2088 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2089 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2090 struct stp_port *sp = ofport->stp_port;
2092 if (!ofproto->stp || !sp) {
2098 s->port_id = stp_port_get_id(sp);
2099 s->state = stp_port_get_state(sp);
2100 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
2101 s->role = stp_port_get_role(sp);
2102 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
2108 stp_run(struct ofproto_dpif *ofproto)
2111 long long int now = time_msec();
2112 long long int elapsed = now - ofproto->stp_last_tick;
2113 struct stp_port *sp;
2116 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
2117 ofproto->stp_last_tick = now;
2119 while (stp_get_changed_port(ofproto->stp, &sp)) {
2120 struct ofport_dpif *ofport = stp_port_get_aux(sp);
2123 update_stp_port_state(ofport);
2127 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
2128 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
2134 stp_wait(struct ofproto_dpif *ofproto)
2137 poll_timer_wait(1000);
2141 /* Returns true if STP should process 'flow'. */
2143 stp_should_process_flow(const struct flow *flow)
2145 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
2149 stp_process_packet(const struct ofport_dpif *ofport,
2150 const struct ofpbuf *packet)
2152 struct ofpbuf payload = *packet;
2153 struct eth_header *eth = payload.data;
2154 struct stp_port *sp = ofport->stp_port;
2156 /* Sink packets on ports that have STP disabled when the bridge has
2158 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
2162 /* Trim off padding on payload. */
2163 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
2164 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
2167 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
2168 stp_received_bpdu(sp, payload.data, payload.size);
2172 static struct priority_to_dscp *
2173 get_priority(const struct ofport_dpif *ofport, uint32_t priority)
2175 struct priority_to_dscp *pdscp;
2178 hash = hash_int(priority, 0);
2179 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) {
2180 if (pdscp->priority == priority) {
2188 ofport_clear_priorities(struct ofport_dpif *ofport)
2190 struct priority_to_dscp *pdscp, *next;
2192 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) {
2193 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
2199 set_queues(struct ofport *ofport_,
2200 const struct ofproto_port_queue *qdscp_list,
2203 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2204 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2205 struct hmap new = HMAP_INITIALIZER(&new);
2208 for (i = 0; i < n_qdscp; i++) {
2209 struct priority_to_dscp *pdscp;
2213 dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
2214 if (dpif_queue_to_priority(ofproto->backer->dpif, qdscp_list[i].queue,
2219 pdscp = get_priority(ofport, priority);
2221 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
2223 pdscp = xmalloc(sizeof *pdscp);
2224 pdscp->priority = priority;
2226 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2229 if (pdscp->dscp != dscp) {
2231 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2234 hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0));
2237 if (!hmap_is_empty(&ofport->priorities)) {
2238 ofport_clear_priorities(ofport);
2239 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2242 hmap_swap(&new, &ofport->priorities);
2250 /* Expires all MAC learning entries associated with 'bundle' and forces its
2251 * ofproto to revalidate every flow.
2253 * Normally MAC learning entries are removed only from the ofproto associated
2254 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2255 * are removed from every ofproto. When patch ports and SLB bonds are in use
2256 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2257 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2258 * with the host from which it migrated. */
2260 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2262 struct ofproto_dpif *ofproto = bundle->ofproto;
2263 struct mac_learning *ml = ofproto->ml;
2264 struct mac_entry *mac, *next_mac;
2266 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2267 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2268 if (mac->port.p == bundle) {
2270 struct ofproto_dpif *o;
2272 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2274 struct mac_entry *e;
2276 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan,
2279 mac_learning_expire(o->ml, e);
2285 mac_learning_expire(ml, mac);
2290 static struct ofbundle *
2291 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2293 struct ofbundle *bundle;
2295 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2296 &ofproto->bundles) {
2297 if (bundle->aux == aux) {
2304 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
2305 * ones that are found to 'bundles'. */
2307 bundle_lookup_multiple(struct ofproto_dpif *ofproto,
2308 void **auxes, size_t n_auxes,
2309 struct hmapx *bundles)
2313 hmapx_init(bundles);
2314 for (i = 0; i < n_auxes; i++) {
2315 struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
2317 hmapx_add(bundles, bundle);
2323 bundle_update(struct ofbundle *bundle)
2325 struct ofport_dpif *port;
2327 bundle->floodable = true;
2328 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2329 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2330 || !stp_forward_in_state(port->stp_state)) {
2331 bundle->floodable = false;
2338 bundle_del_port(struct ofport_dpif *port)
2340 struct ofbundle *bundle = port->bundle;
2342 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2344 list_remove(&port->bundle_node);
2345 port->bundle = NULL;
2348 lacp_slave_unregister(bundle->lacp, port);
2351 bond_slave_unregister(bundle->bond, port);
2354 bundle_update(bundle);
2358 bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port,
2359 struct lacp_slave_settings *lacp,
2360 uint32_t bond_stable_id)
2362 struct ofport_dpif *port;
2364 port = get_ofp_port(bundle->ofproto, ofp_port);
2369 if (port->bundle != bundle) {
2370 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2372 bundle_del_port(port);
2375 port->bundle = bundle;
2376 list_push_back(&bundle->ports, &port->bundle_node);
2377 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2378 || !stp_forward_in_state(port->stp_state)) {
2379 bundle->floodable = false;
2383 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2384 lacp_slave_register(bundle->lacp, port, lacp);
2387 port->bond_stable_id = bond_stable_id;
2393 bundle_destroy(struct ofbundle *bundle)
2395 struct ofproto_dpif *ofproto;
2396 struct ofport_dpif *port, *next_port;
2403 ofproto = bundle->ofproto;
2404 for (i = 0; i < MAX_MIRRORS; i++) {
2405 struct ofmirror *m = ofproto->mirrors[i];
2407 if (m->out == bundle) {
2409 } else if (hmapx_find_and_delete(&m->srcs, bundle)
2410 || hmapx_find_and_delete(&m->dsts, bundle)) {
2411 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2416 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2417 bundle_del_port(port);
2420 bundle_flush_macs(bundle, true);
2421 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2423 free(bundle->trunks);
2424 lacp_destroy(bundle->lacp);
2425 bond_destroy(bundle->bond);
2430 bundle_set(struct ofproto *ofproto_, void *aux,
2431 const struct ofproto_bundle_settings *s)
2433 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2434 bool need_flush = false;
2435 struct ofport_dpif *port;
2436 struct ofbundle *bundle;
2437 unsigned long *trunks;
2443 bundle_destroy(bundle_lookup(ofproto, aux));
2447 ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2448 ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2450 bundle = bundle_lookup(ofproto, aux);
2452 bundle = xmalloc(sizeof *bundle);
2454 bundle->ofproto = ofproto;
2455 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2456 hash_pointer(aux, 0));
2458 bundle->name = NULL;
2460 list_init(&bundle->ports);
2461 bundle->vlan_mode = PORT_VLAN_TRUNK;
2463 bundle->trunks = NULL;
2464 bundle->use_priority_tags = s->use_priority_tags;
2465 bundle->lacp = NULL;
2466 bundle->bond = NULL;
2468 bundle->floodable = true;
2470 bundle->src_mirrors = 0;
2471 bundle->dst_mirrors = 0;
2472 bundle->mirror_out = 0;
2475 if (!bundle->name || strcmp(s->name, bundle->name)) {
2477 bundle->name = xstrdup(s->name);
2482 if (!bundle->lacp) {
2483 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2484 bundle->lacp = lacp_create();
2486 lacp_configure(bundle->lacp, s->lacp);
2488 lacp_destroy(bundle->lacp);
2489 bundle->lacp = NULL;
2492 /* Update set of ports. */
2494 for (i = 0; i < s->n_slaves; i++) {
2495 if (!bundle_add_port(bundle, s->slaves[i],
2496 s->lacp ? &s->lacp_slaves[i] : NULL,
2497 s->bond_stable_ids ? s->bond_stable_ids[i] : 0)) {
2501 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2502 struct ofport_dpif *next_port;
2504 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2505 for (i = 0; i < s->n_slaves; i++) {
2506 if (s->slaves[i] == port->up.ofp_port) {
2511 bundle_del_port(port);
2515 ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2517 if (list_is_empty(&bundle->ports)) {
2518 bundle_destroy(bundle);
2522 /* Set VLAN tagging mode */
2523 if (s->vlan_mode != bundle->vlan_mode
2524 || s->use_priority_tags != bundle->use_priority_tags) {
2525 bundle->vlan_mode = s->vlan_mode;
2526 bundle->use_priority_tags = s->use_priority_tags;
2531 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2532 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2534 if (vlan != bundle->vlan) {
2535 bundle->vlan = vlan;
2539 /* Get trunked VLANs. */
2540 switch (s->vlan_mode) {
2541 case PORT_VLAN_ACCESS:
2545 case PORT_VLAN_TRUNK:
2546 trunks = CONST_CAST(unsigned long *, s->trunks);
2549 case PORT_VLAN_NATIVE_UNTAGGED:
2550 case PORT_VLAN_NATIVE_TAGGED:
2551 if (vlan != 0 && (!s->trunks
2552 || !bitmap_is_set(s->trunks, vlan)
2553 || bitmap_is_set(s->trunks, 0))) {
2554 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2556 trunks = bitmap_clone(s->trunks, 4096);
2558 trunks = bitmap_allocate1(4096);
2560 bitmap_set1(trunks, vlan);
2561 bitmap_set0(trunks, 0);
2563 trunks = CONST_CAST(unsigned long *, s->trunks);
2570 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2571 free(bundle->trunks);
2572 if (trunks == s->trunks) {
2573 bundle->trunks = vlan_bitmap_clone(trunks);
2575 bundle->trunks = trunks;
2580 if (trunks != s->trunks) {
2585 if (!list_is_short(&bundle->ports)) {
2586 bundle->ofproto->has_bonded_bundles = true;
2588 if (bond_reconfigure(bundle->bond, s->bond)) {
2589 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2592 bundle->bond = bond_create(s->bond);
2593 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2596 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2597 bond_slave_register(bundle->bond, port, port->bond_stable_id,
2601 bond_destroy(bundle->bond);
2602 bundle->bond = NULL;
2605 /* If we changed something that would affect MAC learning, un-learn
2606 * everything on this port and force flow revalidation. */
2608 bundle_flush_macs(bundle, false);
2615 bundle_remove(struct ofport *port_)
2617 struct ofport_dpif *port = ofport_dpif_cast(port_);
2618 struct ofbundle *bundle = port->bundle;
2621 bundle_del_port(port);
2622 if (list_is_empty(&bundle->ports)) {
2623 bundle_destroy(bundle);
2624 } else if (list_is_short(&bundle->ports)) {
2625 bond_destroy(bundle->bond);
2626 bundle->bond = NULL;
2632 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2634 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2635 struct ofport_dpif *port = port_;
2636 uint8_t ea[ETH_ADDR_LEN];
2639 error = netdev_get_etheraddr(port->up.netdev, ea);
2641 struct ofpbuf packet;
2644 ofpbuf_init(&packet, 0);
2645 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2647 memcpy(packet_pdu, pdu, pdu_size);
2649 send_packet(port, &packet);
2650 ofpbuf_uninit(&packet);
2652 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2653 "%s (%s)", port->bundle->name,
2654 netdev_get_name(port->up.netdev), strerror(error));
2659 bundle_send_learning_packets(struct ofbundle *bundle)
2661 struct ofproto_dpif *ofproto = bundle->ofproto;
2662 int error, n_packets, n_errors;
2663 struct mac_entry *e;
2665 error = n_packets = n_errors = 0;
2666 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2667 if (e->port.p != bundle) {
2668 struct ofpbuf *learning_packet;
2669 struct ofport_dpif *port;
2673 /* The assignment to "port" is unnecessary but makes "grep"ing for
2674 * struct ofport_dpif more effective. */
2675 learning_packet = bond_compose_learning_packet(bundle->bond,
2679 ret = send_packet(port, learning_packet);
2680 ofpbuf_delete(learning_packet);
2690 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2691 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2692 "packets, last error was: %s",
2693 bundle->name, n_errors, n_packets, strerror(error));
2695 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2696 bundle->name, n_packets);
2701 bundle_run(struct ofbundle *bundle)
2704 lacp_run(bundle->lacp, send_pdu_cb);
2707 struct ofport_dpif *port;
2709 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2710 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2713 bond_run(bundle->bond, &bundle->ofproto->backer->revalidate_set,
2714 lacp_status(bundle->lacp));
2715 if (bond_should_send_learning_packets(bundle->bond)) {
2716 bundle_send_learning_packets(bundle);
2722 bundle_wait(struct ofbundle *bundle)
2725 lacp_wait(bundle->lacp);
2728 bond_wait(bundle->bond);
2735 mirror_scan(struct ofproto_dpif *ofproto)
2739 for (idx = 0; idx < MAX_MIRRORS; idx++) {
2740 if (!ofproto->mirrors[idx]) {
2747 static struct ofmirror *
2748 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
2752 for (i = 0; i < MAX_MIRRORS; i++) {
2753 struct ofmirror *mirror = ofproto->mirrors[i];
2754 if (mirror && mirror->aux == aux) {
2762 /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */
2764 mirror_update_dups(struct ofproto_dpif *ofproto)
2768 for (i = 0; i < MAX_MIRRORS; i++) {
2769 struct ofmirror *m = ofproto->mirrors[i];
2772 m->dup_mirrors = MIRROR_MASK_C(1) << i;
2776 for (i = 0; i < MAX_MIRRORS; i++) {
2777 struct ofmirror *m1 = ofproto->mirrors[i];
2784 for (j = i + 1; j < MAX_MIRRORS; j++) {
2785 struct ofmirror *m2 = ofproto->mirrors[j];
2787 if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) {
2788 m1->dup_mirrors |= MIRROR_MASK_C(1) << j;
2789 m2->dup_mirrors |= m1->dup_mirrors;
2796 mirror_set(struct ofproto *ofproto_, void *aux,
2797 const struct ofproto_mirror_settings *s)
2799 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2800 mirror_mask_t mirror_bit;
2801 struct ofbundle *bundle;
2802 struct ofmirror *mirror;
2803 struct ofbundle *out;
2804 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
2805 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
2808 mirror = mirror_lookup(ofproto, aux);
2810 mirror_destroy(mirror);
2816 idx = mirror_scan(ofproto);
2818 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
2820 ofproto->up.name, MAX_MIRRORS, s->name);
2824 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
2825 mirror->ofproto = ofproto;
2828 mirror->out_vlan = -1;
2829 mirror->name = NULL;
2832 if (!mirror->name || strcmp(s->name, mirror->name)) {
2834 mirror->name = xstrdup(s->name);
2837 /* Get the new configuration. */
2838 if (s->out_bundle) {
2839 out = bundle_lookup(ofproto, s->out_bundle);
2841 mirror_destroy(mirror);
2847 out_vlan = s->out_vlan;
2849 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
2850 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
2852 /* If the configuration has not changed, do nothing. */
2853 if (hmapx_equals(&srcs, &mirror->srcs)
2854 && hmapx_equals(&dsts, &mirror->dsts)
2855 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
2856 && mirror->out == out
2857 && mirror->out_vlan == out_vlan)
2859 hmapx_destroy(&srcs);
2860 hmapx_destroy(&dsts);
2864 hmapx_swap(&srcs, &mirror->srcs);
2865 hmapx_destroy(&srcs);
2867 hmapx_swap(&dsts, &mirror->dsts);
2868 hmapx_destroy(&dsts);
2870 free(mirror->vlans);
2871 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
2874 mirror->out_vlan = out_vlan;
2876 /* Update bundles. */
2877 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2878 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
2879 if (hmapx_contains(&mirror->srcs, bundle)) {
2880 bundle->src_mirrors |= mirror_bit;
2882 bundle->src_mirrors &= ~mirror_bit;
2885 if (hmapx_contains(&mirror->dsts, bundle)) {
2886 bundle->dst_mirrors |= mirror_bit;
2888 bundle->dst_mirrors &= ~mirror_bit;
2891 if (mirror->out == bundle) {
2892 bundle->mirror_out |= mirror_bit;
2894 bundle->mirror_out &= ~mirror_bit;
2898 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2899 ofproto->has_mirrors = true;
2900 mac_learning_flush(ofproto->ml,
2901 &ofproto->backer->revalidate_set);
2902 mirror_update_dups(ofproto);
2908 mirror_destroy(struct ofmirror *mirror)
2910 struct ofproto_dpif *ofproto;
2911 mirror_mask_t mirror_bit;
2912 struct ofbundle *bundle;
2919 ofproto = mirror->ofproto;
2920 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2921 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
2923 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2924 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
2925 bundle->src_mirrors &= ~mirror_bit;
2926 bundle->dst_mirrors &= ~mirror_bit;
2927 bundle->mirror_out &= ~mirror_bit;
2930 hmapx_destroy(&mirror->srcs);
2931 hmapx_destroy(&mirror->dsts);
2932 free(mirror->vlans);
2934 ofproto->mirrors[mirror->idx] = NULL;
2938 mirror_update_dups(ofproto);
2940 ofproto->has_mirrors = false;
2941 for (i = 0; i < MAX_MIRRORS; i++) {
2942 if (ofproto->mirrors[i]) {
2943 ofproto->has_mirrors = true;
2950 mirror_get_stats(struct ofproto *ofproto_, void *aux,
2951 uint64_t *packets, uint64_t *bytes)
2953 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2954 struct ofmirror *mirror = mirror_lookup(ofproto, aux);
2957 *packets = *bytes = UINT64_MAX;
2963 *packets = mirror->packet_count;
2964 *bytes = mirror->byte_count;
2970 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2972 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2973 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2974 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
2980 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2982 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2983 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2984 return bundle && bundle->mirror_out != 0;
2988 forward_bpdu_changed(struct ofproto *ofproto_)
2990 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2991 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2995 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
2998 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2999 mac_learning_set_idle_time(ofproto->ml, idle_time);
3000 mac_learning_set_max_entries(ofproto->ml, max_entries);
3005 static struct ofport_dpif *
3006 get_ofp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
3008 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
3009 return ofport ? ofport_dpif_cast(ofport) : NULL;
3012 static struct ofport_dpif *
3013 get_odp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
3015 struct ofport_dpif *port = odp_port_to_ofport(ofproto->backer, odp_port);
3016 return port && &ofproto->up == port->up.ofproto ? port : NULL;
3020 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
3021 struct ofproto_port *ofproto_port,
3022 struct dpif_port *dpif_port)
3024 ofproto_port->name = dpif_port->name;
3025 ofproto_port->type = dpif_port->type;
3026 ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
3029 static struct ofport_dpif *
3030 ofport_get_peer(const struct ofport_dpif *ofport_dpif)
3032 const struct ofproto_dpif *ofproto;
3035 peer = netdev_vport_patch_peer(ofport_dpif->up.netdev);
3040 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
3041 struct ofport *ofport;
3043 ofport = shash_find_data(&ofproto->up.port_by_name, peer);
3044 if (ofport && ofport->ofproto->ofproto_class == &ofproto_dpif_class) {
3045 return ofport_dpif_cast(ofport);
3052 port_run_fast(struct ofport_dpif *ofport)
3054 if (ofport->cfm && cfm_should_send_ccm(ofport->cfm)) {
3055 struct ofpbuf packet;
3057 ofpbuf_init(&packet, 0);
3058 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
3059 send_packet(ofport, &packet);
3060 ofpbuf_uninit(&packet);
3065 port_run(struct ofport_dpif *ofport)
3067 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
3068 bool carrier_changed = carrier_seq != ofport->carrier_seq;
3069 bool enable = netdev_get_carrier(ofport->up.netdev);
3071 ofport->carrier_seq = carrier_seq;
3073 port_run_fast(ofport);
3075 if (ofport->tnl_port
3076 && tnl_port_reconfigure(&ofport->up, ofport->odp_port,
3077 &ofport->tnl_port)) {
3078 ofproto_dpif_cast(ofport->up.ofproto)->backer->need_revalidate = true;
3082 int cfm_opup = cfm_get_opup(ofport->cfm);
3084 cfm_run(ofport->cfm);
3085 enable = enable && !cfm_get_fault(ofport->cfm);
3087 if (cfm_opup >= 0) {
3088 enable = enable && cfm_opup;
3092 if (ofport->bundle) {
3093 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
3094 if (carrier_changed) {
3095 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
3099 if (ofport->may_enable != enable) {
3100 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3102 if (ofproto->has_bundle_action) {
3103 ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
3107 ofport->may_enable = enable;
3111 port_wait(struct ofport_dpif *ofport)
3114 cfm_wait(ofport->cfm);
3119 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
3120 struct ofproto_port *ofproto_port)
3122 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3123 struct dpif_port dpif_port;
3126 if (sset_contains(&ofproto->ghost_ports, devname)) {
3127 const char *type = netdev_get_type_from_name(devname);
3129 /* We may be called before ofproto->up.port_by_name is populated with
3130 * the appropriate ofport. For this reason, we must get the name and
3131 * type from the netdev layer directly. */
3133 const struct ofport *ofport;
3135 ofport = shash_find_data(&ofproto->up.port_by_name, devname);
3136 ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
3137 ofproto_port->name = xstrdup(devname);
3138 ofproto_port->type = xstrdup(type);
3144 if (!sset_contains(&ofproto->ports, devname)) {
3147 error = dpif_port_query_by_name(ofproto->backer->dpif,
3148 devname, &dpif_port);
3150 ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
3156 port_add(struct ofproto *ofproto_, struct netdev *netdev)
3158 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3159 const char *dp_port_name = netdev_vport_get_dpif_port(netdev);
3160 const char *devname = netdev_get_name(netdev);
3162 if (netdev_vport_is_patch(netdev)) {
3163 sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
3167 if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
3168 uint32_t port_no = UINT32_MAX;
3171 error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
3175 if (netdev_get_tunnel_config(netdev)) {
3176 simap_put(&ofproto->backer->tnl_backers, dp_port_name, port_no);
3180 if (netdev_get_tunnel_config(netdev)) {
3181 sset_add(&ofproto->ghost_ports, devname);
3183 sset_add(&ofproto->ports, devname);
3189 port_del(struct ofproto *ofproto_, uint16_t ofp_port)
3191 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3192 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
3199 sset_find_and_delete(&ofproto->ghost_ports,
3200 netdev_get_name(ofport->up.netdev));
3201 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3202 if (!ofport->tnl_port) {
3203 error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
3205 /* The caller is going to close ofport->up.netdev. If this is a
3206 * bonded port, then the bond is using that netdev, so remove it
3207 * from the bond. The client will need to reconfigure everything
3208 * after deleting ports, so then the slave will get re-added. */
3209 bundle_remove(&ofport->up);
3216 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
3218 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3223 error = netdev_get_stats(ofport->up.netdev, stats);
3225 if (!error && ofport_->ofp_port == OFPP_LOCAL) {
3226 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3228 /* ofproto->stats.tx_packets represents packets that we created
3229 * internally and sent to some port (e.g. packets sent with
3230 * send_packet()). Account for them as if they had come from
3231 * OFPP_LOCAL and got forwarded. */
3233 if (stats->rx_packets != UINT64_MAX) {
3234 stats->rx_packets += ofproto->stats.tx_packets;
3237 if (stats->rx_bytes != UINT64_MAX) {
3238 stats->rx_bytes += ofproto->stats.tx_bytes;
3241 /* ofproto->stats.rx_packets represents packets that were received on
3242 * some port and we processed internally and dropped (e.g. STP).
3243 * Account for them as if they had been forwarded to OFPP_LOCAL. */
3245 if (stats->tx_packets != UINT64_MAX) {
3246 stats->tx_packets += ofproto->stats.rx_packets;
3249 if (stats->tx_bytes != UINT64_MAX) {
3250 stats->tx_bytes += ofproto->stats.rx_bytes;
3257 /* Account packets for LOCAL port. */
3259 ofproto_update_local_port_stats(const struct ofproto *ofproto_,
3260 size_t tx_size, size_t rx_size)
3262 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3265 ofproto->stats.rx_packets++;
3266 ofproto->stats.rx_bytes += rx_size;
3269 ofproto->stats.tx_packets++;
3270 ofproto->stats.tx_bytes += tx_size;
3274 struct port_dump_state {
3279 struct ofproto_port port;
3284 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
3286 *statep = xzalloc(sizeof(struct port_dump_state));
3291 port_dump_next(const struct ofproto *ofproto_, void *state_,
3292 struct ofproto_port *port)
3294 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3295 struct port_dump_state *state = state_;
3296 const struct sset *sset;
3297 struct sset_node *node;
3299 if (state->has_port) {
3300 ofproto_port_destroy(&state->port);
3301 state->has_port = false;
3303 sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
3304 while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
3307 error = port_query_by_name(ofproto_, node->name, &state->port);
3309 *port = state->port;
3310 state->has_port = true;
3312 } else if (error != ENODEV) {
3317 if (!state->ghost) {
3318 state->ghost = true;
3321 return port_dump_next(ofproto_, state_, port);
3328 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3330 struct port_dump_state *state = state_;
3332 if (state->has_port) {
3333 ofproto_port_destroy(&state->port);
3340 port_poll(const struct ofproto *ofproto_, char **devnamep)
3342 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3344 if (ofproto->port_poll_errno) {
3345 int error = ofproto->port_poll_errno;
3346 ofproto->port_poll_errno = 0;
3350 if (sset_is_empty(&ofproto->port_poll_set)) {
3354 *devnamep = sset_pop(&ofproto->port_poll_set);
3359 port_poll_wait(const struct ofproto *ofproto_)
3361 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3362 dpif_port_poll_wait(ofproto->backer->dpif);
3366 port_is_lacp_current(const struct ofport *ofport_)
3368 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3369 return (ofport->bundle && ofport->bundle->lacp
3370 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3374 /* Upcall handling. */
3376 /* Flow miss batching.
3378 * Some dpifs implement operations faster when you hand them off in a batch.
3379 * To allow batching, "struct flow_miss" queues the dpif-related work needed
3380 * for a given flow. Each "struct flow_miss" corresponds to sending one or
3381 * more packets, plus possibly installing the flow in the dpif.
3383 * So far we only batch the operations that affect flow setup time the most.
3384 * It's possible to batch more than that, but the benefit might be minimal. */
3386 struct hmap_node hmap_node;
3387 struct ofproto_dpif *ofproto;
3389 enum odp_key_fitness key_fitness;
3390 const struct nlattr *key;
3392 struct initial_vals initial_vals;
3393 struct list packets;
3394 enum dpif_upcall_type upcall_type;
3395 uint32_t odp_in_port;
3398 struct flow_miss_op {
3399 struct dpif_op dpif_op;
3400 void *garbage; /* Pointer to pass to free(), NULL if none. */
3401 uint64_t stub[1024 / 8]; /* Temporary buffer. */
3404 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
3405 * OpenFlow controller as necessary according to their individual
3406 * configurations. */
3408 send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet,
3409 const struct flow *flow)
3411 struct ofputil_packet_in pin;
3413 pin.packet = packet->data;
3414 pin.packet_len = packet->size;
3415 pin.reason = OFPR_NO_MATCH;
3416 pin.controller_id = 0;
3421 pin.send_len = 0; /* not used for flow table misses */
3423 flow_get_metadata(flow, &pin.fmd);
3425 connmgr_send_packet_in(ofproto->up.connmgr, &pin);
3428 static enum slow_path_reason
3429 process_special(struct ofproto_dpif *ofproto, const struct flow *flow,
3430 const struct ofport_dpif *ofport, const struct ofpbuf *packet)
3434 } else if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) {
3436 cfm_process_heartbeat(ofport->cfm, packet);
3439 } else if (ofport->bundle && ofport->bundle->lacp
3440 && flow->dl_type == htons(ETH_TYPE_LACP)) {
3442 lacp_process_packet(ofport->bundle->lacp, ofport, packet);
3445 } else if (ofproto->stp && stp_should_process_flow(flow)) {
3447 stp_process_packet(ofport, packet);
3455 static struct flow_miss *
3456 flow_miss_find(struct hmap *todo, const struct ofproto_dpif *ofproto,
3457 const struct flow *flow, uint32_t hash)
3459 struct flow_miss *miss;
3461 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
3462 if (miss->ofproto == ofproto && flow_equal(&miss->flow, flow)) {
3470 /* Partially Initializes 'op' as an "execute" operation for 'miss' and
3471 * 'packet'. The caller must initialize op->actions and op->actions_len. If
3472 * 'miss' is associated with a subfacet the caller must also initialize the
3473 * returned op->subfacet, and if anything needs to be freed after processing
3474 * the op, the caller must initialize op->garbage also. */
3476 init_flow_miss_execute_op(struct flow_miss *miss, struct ofpbuf *packet,
3477 struct flow_miss_op *op)
3479 if (miss->flow.vlan_tci != miss->initial_vals.vlan_tci) {
3480 /* This packet was received on a VLAN splinter port. We
3481 * added a VLAN to the packet to make the packet resemble
3482 * the flow, but the actions were composed assuming that
3483 * the packet contained no VLAN. So, we must remove the
3484 * VLAN header from the packet before trying to execute the
3486 eth_pop_vlan(packet);
3490 op->dpif_op.type = DPIF_OP_EXECUTE;
3491 op->dpif_op.u.execute.key = miss->key;
3492 op->dpif_op.u.execute.key_len = miss->key_len;
3493 op->dpif_op.u.execute.packet = packet;
3496 /* Helper for handle_flow_miss_without_facet() and
3497 * handle_flow_miss_with_facet(). */
3499 handle_flow_miss_common(struct rule_dpif *rule,
3500 struct ofpbuf *packet, const struct flow *flow)
3502 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3504 ofproto->n_matches++;
3506 if (rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
3508 * Extra-special case for fail-open mode.
3510 * We are in fail-open mode and the packet matched the fail-open
3511 * rule, but we are connected to a controller too. We should send
3512 * the packet up to the controller in the hope that it will try to
3513 * set up a flow and thereby allow us to exit fail-open.
3515 * See the top-level comment in fail-open.c for more information.
3517 send_packet_in_miss(ofproto, packet, flow);
3521 /* Figures out whether a flow that missed in 'ofproto', whose details are in
3522 * 'miss', is likely to be worth tracking in detail in userspace and (usually)
3523 * installing a datapath flow. The answer is usually "yes" (a return value of
3524 * true). However, for short flows the cost of bookkeeping is much higher than
3525 * the benefits, so when the datapath holds a large number of flows we impose
3526 * some heuristics to decide which flows are likely to be worth tracking. */
3528 flow_miss_should_make_facet(struct ofproto_dpif *ofproto,
3529 struct flow_miss *miss, uint32_t hash)
3531 if (!ofproto->governor) {
3534 n_subfacets = hmap_count(&ofproto->subfacets);
3535 if (n_subfacets * 2 <= ofproto->up.flow_eviction_threshold) {
3539 ofproto->governor = governor_create(ofproto->up.name);
3542 return governor_should_install_flow(ofproto->governor, hash,
3543 list_size(&miss->packets));
3546 /* Handles 'miss', which matches 'rule', without creating a facet or subfacet
3547 * or creating any datapath flow. May add an "execute" operation to 'ops' and
3548 * increment '*n_ops'. */
3550 handle_flow_miss_without_facet(struct flow_miss *miss,
3551 struct rule_dpif *rule,
3552 struct flow_miss_op *ops, size_t *n_ops)
3554 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3555 long long int now = time_msec();
3556 struct action_xlate_ctx ctx;
3557 struct ofpbuf *packet;
3559 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3560 struct flow_miss_op *op = &ops[*n_ops];
3561 struct dpif_flow_stats stats;
3562 struct ofpbuf odp_actions;
3564 COVERAGE_INC(facet_suppress);
3566 ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub);
3568 dpif_flow_stats_extract(&miss->flow, packet, now, &stats);
3569 rule_credit_stats(rule, &stats);
3571 action_xlate_ctx_init(&ctx, ofproto, &miss->flow,
3572 &miss->initial_vals, rule, 0, packet);
3573 ctx.resubmit_stats = &stats;
3574 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len,
3577 if (odp_actions.size) {
3578 struct dpif_execute *execute = &op->dpif_op.u.execute;
3580 init_flow_miss_execute_op(miss, packet, op);
3581 execute->actions = odp_actions.data;
3582 execute->actions_len = odp_actions.size;
3583 op->garbage = ofpbuf_get_uninit_pointer(&odp_actions);
3587 ofpbuf_uninit(&odp_actions);
3592 /* Handles 'miss', which matches 'facet'. May add any required datapath
3593 * operations to 'ops', incrementing '*n_ops' for each new op.
3595 * All of the packets in 'miss' are considered to have arrived at time 'now'.
3596 * This is really important only for new facets: if we just called time_msec()
3597 * here, then the new subfacet or its packets could look (occasionally) as
3598 * though it was used some time after the facet was used. That can make a
3599 * one-packet flow look like it has a nonzero duration, which looks odd in
3600 * e.g. NetFlow statistics. */
3602 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
3604 struct flow_miss_op *ops, size_t *n_ops)
3606 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3607 enum subfacet_path want_path;
3608 struct subfacet *subfacet;
3609 struct ofpbuf *packet;
3611 subfacet = subfacet_create(facet, miss, now);
3613 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3614 struct flow_miss_op *op = &ops[*n_ops];
3615 struct dpif_flow_stats stats;
3616 struct ofpbuf odp_actions;
3618 handle_flow_miss_common(facet->rule, packet, &miss->flow);
3620 ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub);
3621 if (!subfacet->actions || subfacet->slow) {
3622 subfacet_make_actions(subfacet, packet, &odp_actions);
3625 dpif_flow_stats_extract(&facet->flow, packet, now, &stats);
3626 subfacet_update_stats(subfacet, &stats);
3628 if (subfacet->actions_len) {
3629 struct dpif_execute *execute = &op->dpif_op.u.execute;
3631 init_flow_miss_execute_op(miss, packet, op);
3632 if (!subfacet->slow) {
3633 execute->actions = subfacet->actions;
3634 execute->actions_len = subfacet->actions_len;
3635 ofpbuf_uninit(&odp_actions);
3637 execute->actions = odp_actions.data;
3638 execute->actions_len = odp_actions.size;
3639 op->garbage = ofpbuf_get_uninit_pointer(&odp_actions);
3644 ofpbuf_uninit(&odp_actions);
3648 want_path = subfacet_want_path(subfacet->slow);
3649 if (miss->upcall_type == DPIF_UC_MISS || subfacet->path != want_path) {
3650 struct flow_miss_op *op = &ops[(*n_ops)++];
3651 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
3653 subfacet->path = want_path;
3656 op->dpif_op.type = DPIF_OP_FLOW_PUT;
3657 put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
3658 put->key = miss->key;
3659 put->key_len = miss->key_len;
3660 if (want_path == SF_FAST_PATH) {
3661 put->actions = subfacet->actions;
3662 put->actions_len = subfacet->actions_len;
3664 compose_slow_path(ofproto, &facet->flow, subfacet->slow,
3665 op->stub, sizeof op->stub,
3666 &put->actions, &put->actions_len);
3672 /* Handles flow miss 'miss'. May add any required datapath operations
3673 * to 'ops', incrementing '*n_ops' for each new op. */
3675 handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops,
3678 struct ofproto_dpif *ofproto = miss->ofproto;
3679 struct facet *facet;
3683 /* The caller must ensure that miss->hmap_node.hash contains
3684 * flow_hash(miss->flow, 0). */
3685 hash = miss->hmap_node.hash;
3687 facet = facet_lookup_valid(ofproto, &miss->flow, hash);
3689 struct rule_dpif *rule = rule_dpif_lookup(ofproto, &miss->flow);
3691 if (!flow_miss_should_make_facet(ofproto, miss, hash)) {
3692 handle_flow_miss_without_facet(miss, rule, ops, n_ops);
3696 facet = facet_create(rule, &miss->flow, hash);
3701 handle_flow_miss_with_facet(miss, facet, now, ops, n_ops);
3704 static struct drop_key *
3705 drop_key_lookup(const struct dpif_backer *backer, const struct nlattr *key,
3708 struct drop_key *drop_key;
3710 HMAP_FOR_EACH_WITH_HASH (drop_key, hmap_node, hash_bytes(key, key_len, 0),
3711 &backer->drop_keys) {
3712 if (drop_key->key_len == key_len
3713 && !memcmp(drop_key->key, key, key_len)) {
3721 drop_key_clear(struct dpif_backer *backer)
3723 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3724 struct drop_key *drop_key, *next;
3726 HMAP_FOR_EACH_SAFE (drop_key, next, hmap_node, &backer->drop_keys) {
3729 error = dpif_flow_del(backer->dpif, drop_key->key, drop_key->key_len,
3731 if (error && !VLOG_DROP_WARN(&rl)) {
3732 struct ds ds = DS_EMPTY_INITIALIZER;
3733 odp_flow_key_format(drop_key->key, drop_key->key_len, &ds);
3734 VLOG_WARN("Failed to delete drop key (%s) (%s)", strerror(error),
3739 hmap_remove(&backer->drop_keys, &drop_key->hmap_node);
3740 free(drop_key->key);
3745 /* Given a datpath, packet, and flow metadata ('backer', 'packet', and 'key'
3746 * respectively), populates 'flow' with the result of odp_flow_key_to_flow().
3747 * Optionally, if nonnull, populates 'fitnessp' with the fitness of 'flow' as
3748 * returned by odp_flow_key_to_flow(). Also, optionally populates 'ofproto'
3749 * with the ofproto_dpif, and 'odp_in_port' with the datapath in_port, that
3750 * 'packet' ingressed.
3752 * If 'ofproto' is nonnull, requires 'flow''s in_port to exist. Otherwise sets
3753 * 'flow''s in_port to OFPP_NONE.
3755 * This function does post-processing on data returned from
3756 * odp_flow_key_to_flow() to help make VLAN splinters transparent to the rest
3757 * of the upcall processing logic. In particular, if the extracted in_port is
3758 * a VLAN splinter port, it replaces flow->in_port by the "real" port, sets
3759 * flow->vlan_tci correctly for the VLAN of the VLAN splinter port, and pushes
3760 * a VLAN header onto 'packet' (if it is nonnull).
3762 * Optionally, if 'initial_vals' is nonnull, sets 'initial_vals->vlan_tci'
3763 * to the VLAN TCI with which the packet was really received, that is, the
3764 * actual VLAN TCI extracted by odp_flow_key_to_flow(). (This differs from
3765 * the value returned in flow->vlan_tci only for packets received on
3766 * VLAN splinters.) Also, if received on an IP tunnel, sets
3767 * 'initial_vals->tunnel_ip_tos' to the tunnel's IP TOS.
3769 * Similarly, this function also includes some logic to help with tunnels. It
3770 * may modify 'flow' as necessary to make the tunneling implementation
3771 * transparent to the upcall processing logic.
3773 * Returns 0 if successful, ENODEV if the parsed flow has no associated ofport,
3774 * or some other positive errno if there are other problems. */
3776 ofproto_receive(const struct dpif_backer *backer, struct ofpbuf *packet,
3777 const struct nlattr *key, size_t key_len,
3778 struct flow *flow, enum odp_key_fitness *fitnessp,
3779 struct ofproto_dpif **ofproto, uint32_t *odp_in_port,
3780 struct initial_vals *initial_vals)
3782 const struct ofport_dpif *port;
3783 enum odp_key_fitness fitness;
3786 fitness = odp_flow_key_to_flow(key, key_len, flow);
3787 if (fitness == ODP_FIT_ERROR) {
3793 initial_vals->vlan_tci = flow->vlan_tci;
3794 initial_vals->tunnel_ip_tos = flow->tunnel.ip_tos;
3798 *odp_in_port = flow->in_port;
3801 if (tnl_port_should_receive(flow)) {
3802 const struct ofport *ofport = tnl_port_receive(flow);
3804 flow->in_port = OFPP_NONE;
3807 port = ofport_dpif_cast(ofport);
3809 /* We can't reproduce 'key' from 'flow'. */
3810 fitness = fitness == ODP_FIT_PERFECT ? ODP_FIT_TOO_MUCH : fitness;
3812 /* XXX: Since the tunnel module is not scoped per backer, it's
3813 * theoretically possible that we'll receive an ofport belonging to an
3814 * entirely different datapath. In practice, this can't happen because
3815 * no platforms has two separate datapaths which each support
3817 ovs_assert(ofproto_dpif_cast(port->up.ofproto)->backer == backer);
3819 port = odp_port_to_ofport(backer, flow->in_port);
3821 flow->in_port = OFPP_NONE;
3825 flow->in_port = port->up.ofp_port;
3826 if (vsp_adjust_flow(ofproto_dpif_cast(port->up.ofproto), flow)) {
3828 /* Make the packet resemble the flow, so that it gets sent to
3829 * an OpenFlow controller properly, so that it looks correct
3830 * for sFlow, and so that flow_extract() will get the correct
3831 * vlan_tci if it is called on 'packet'.
3833 * The allocated space inside 'packet' probably also contains
3834 * 'key', that is, both 'packet' and 'key' are probably part of
3835 * a struct dpif_upcall (see the large comment on that
3836 * structure definition), so pushing data on 'packet' is in
3837 * general not a good idea since it could overwrite 'key' or
3838 * free it as a side effect. However, it's OK in this special
3839 * case because we know that 'packet' is inside a Netlink
3840 * attribute: pushing 4 bytes will just overwrite the 4-byte
3841 * "struct nlattr", which is fine since we don't need that
3842 * header anymore. */
3843 eth_push_vlan(packet, flow->vlan_tci);
3845 /* We can't reproduce 'key' from 'flow'. */
3846 fitness = fitness == ODP_FIT_PERFECT ? ODP_FIT_TOO_MUCH : fitness;
3852 *ofproto = ofproto_dpif_cast(port->up.ofproto);
3857 *fitnessp = fitness;
3863 handle_miss_upcalls(struct dpif_backer *backer, struct dpif_upcall *upcalls,
3866 struct dpif_upcall *upcall;
3867 struct flow_miss *miss;
3868 struct flow_miss misses[FLOW_MISS_MAX_BATCH];
3869 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
3870 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
3880 /* Construct the to-do list.
3882 * This just amounts to extracting the flow from each packet and sticking
3883 * the packets that have the same flow in the same "flow_miss" structure so
3884 * that we can process them together. */
3887 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
3888 struct flow_miss *miss = &misses[n_misses];
3889 struct flow_miss *existing_miss;
3890 struct ofproto_dpif *ofproto;
3891 uint32_t odp_in_port;
3896 error = ofproto_receive(backer, upcall->packet, upcall->key,
3897 upcall->key_len, &flow, &miss->key_fitness,
3898 &ofproto, &odp_in_port, &miss->initial_vals);
3899 if (error == ENODEV) {
3900 struct drop_key *drop_key;
3902 /* Received packet on port for which we couldn't associate
3903 * an ofproto. This can happen if a port is removed while
3904 * traffic is being received. Print a rate-limited message
3905 * in case it happens frequently. Install a drop flow so
3906 * that future packets of the flow are inexpensively dropped
3908 VLOG_INFO_RL(&rl, "received packet on unassociated port %"PRIu32,
3911 drop_key = drop_key_lookup(backer, upcall->key, upcall->key_len);
3913 drop_key = xmalloc(sizeof *drop_key);
3914 drop_key->key = xmemdup(upcall->key, upcall->key_len);
3915 drop_key->key_len = upcall->key_len;
3917 hmap_insert(&backer->drop_keys, &drop_key->hmap_node,
3918 hash_bytes(drop_key->key, drop_key->key_len, 0));
3919 dpif_flow_put(backer->dpif, DPIF_FP_CREATE | DPIF_FP_MODIFY,
3920 drop_key->key, drop_key->key_len, NULL, 0, NULL);
3928 ofproto->n_missed++;
3929 flow_extract(upcall->packet, flow.skb_priority, flow.skb_mark,
3930 &flow.tunnel, flow.in_port, &miss->flow);
3932 /* Add other packets to a to-do list. */
3933 hash = flow_hash(&miss->flow, 0);
3934 existing_miss = flow_miss_find(&todo, ofproto, &miss->flow, hash);
3935 if (!existing_miss) {
3936 hmap_insert(&todo, &miss->hmap_node, hash);
3937 miss->ofproto = ofproto;
3938 miss->key = upcall->key;
3939 miss->key_len = upcall->key_len;
3940 miss->upcall_type = upcall->type;
3941 miss->odp_in_port = odp_in_port;
3942 list_init(&miss->packets);
3946 miss = existing_miss;
3948 list_push_back(&miss->packets, &upcall->packet->list_node);
3951 /* Process each element in the to-do list, constructing the set of
3952 * operations to batch. */
3954 HMAP_FOR_EACH (miss, hmap_node, &todo) {
3955 handle_flow_miss(miss, flow_miss_ops, &n_ops);
3957 ovs_assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
3959 /* Execute batch. */
3960 for (i = 0; i < n_ops; i++) {
3961 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
3963 dpif_operate(backer->dpif, dpif_ops, n_ops);
3966 for (i = 0; i < n_ops; i++) {
3967 free(flow_miss_ops[i].garbage);
3969 hmap_destroy(&todo);
3972 static enum { SFLOW_UPCALL, MISS_UPCALL, BAD_UPCALL }
3973 classify_upcall(const struct dpif_upcall *upcall)
3975 union user_action_cookie cookie;
3977 /* First look at the upcall type. */
3978 switch (upcall->type) {
3979 case DPIF_UC_ACTION:
3985 case DPIF_N_UC_TYPES:
3987 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
3991 /* "action" upcalls need a closer look. */
3992 memcpy(&cookie, &upcall->userdata, sizeof(cookie));
3993 switch (cookie.type) {
3994 case USER_ACTION_COOKIE_SFLOW:
3995 return SFLOW_UPCALL;
3997 case USER_ACTION_COOKIE_SLOW_PATH:
4000 case USER_ACTION_COOKIE_UNSPEC:
4002 VLOG_WARN_RL(&rl, "invalid user cookie : 0x%"PRIx64, upcall->userdata);
4008 handle_sflow_upcall(struct dpif_backer *backer,
4009 const struct dpif_upcall *upcall)
4011 struct ofproto_dpif *ofproto;
4012 union user_action_cookie cookie;
4014 uint32_t odp_in_port;
4016 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
4017 &flow, NULL, &ofproto, &odp_in_port, NULL)
4018 || !ofproto->sflow) {
4022 memcpy(&cookie, &upcall->userdata, sizeof(cookie));
4023 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
4024 odp_in_port, &cookie);
4028 handle_upcalls(struct dpif_backer *backer, unsigned int max_batch)
4030 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
4031 struct ofpbuf miss_bufs[FLOW_MISS_MAX_BATCH];
4032 uint64_t miss_buf_stubs[FLOW_MISS_MAX_BATCH][4096 / 8];
4037 ovs_assert(max_batch <= FLOW_MISS_MAX_BATCH);
4040 for (n_processed = 0; n_processed < max_batch; n_processed++) {
4041 struct dpif_upcall *upcall = &misses[n_misses];
4042 struct ofpbuf *buf = &miss_bufs[n_misses];
4045 ofpbuf_use_stub(buf, miss_buf_stubs[n_misses],
4046 sizeof miss_buf_stubs[n_misses]);
4047 error = dpif_recv(backer->dpif, upcall, buf);
4053 switch (classify_upcall(upcall)) {
4055 /* Handle it later. */
4060 handle_sflow_upcall(backer, upcall);
4070 /* Handle deferred MISS_UPCALL processing. */
4071 handle_miss_upcalls(backer, misses, n_misses);
4072 for (i = 0; i < n_misses; i++) {
4073 ofpbuf_uninit(&miss_bufs[i]);
4079 /* Flow expiration. */
4081 static int subfacet_max_idle(const struct ofproto_dpif *);
4082 static void update_stats(struct dpif_backer *);
4083 static void rule_expire(struct rule_dpif *);
4084 static void expire_subfacets(struct ofproto_dpif *, int dp_max_idle);
4086 /* This function is called periodically by run(). Its job is to collect
4087 * updates for the flows that have been installed into the datapath, most
4088 * importantly when they last were used, and then use that information to
4089 * expire flows that have not been used recently.
4091 * Returns the number of milliseconds after which it should be called again. */
4093 expire(struct dpif_backer *backer)
4095 struct ofproto_dpif *ofproto;
4096 int max_idle = INT32_MAX;
4098 /* Periodically clear out the drop keys in an effort to keep them
4099 * relatively few. */
4100 drop_key_clear(backer);
4102 /* Update stats for each flow in the backer. */
4103 update_stats(backer);
4105 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4106 struct rule *rule, *next_rule;
4109 if (ofproto->backer != backer) {
4113 /* Keep track of the max number of flows per ofproto_dpif. */
4114 update_max_subfacet_count(ofproto);
4116 /* Expire subfacets that have been idle too long. */
4117 dp_max_idle = subfacet_max_idle(ofproto);
4118 expire_subfacets(ofproto, dp_max_idle);
4120 max_idle = MIN(max_idle, dp_max_idle);
4122 /* Expire OpenFlow flows whose idle_timeout or hard_timeout
4124 LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
4125 &ofproto->up.expirable) {
4126 rule_expire(rule_dpif_cast(rule));
4129 /* All outstanding data in existing flows has been accounted, so it's a
4130 * good time to do bond rebalancing. */
4131 if (ofproto->has_bonded_bundles) {
4132 struct ofbundle *bundle;
4134 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
4136 bond_rebalance(bundle->bond, &backer->revalidate_set);
4142 return MIN(max_idle, 1000);
4145 /* Updates flow table statistics given that the datapath just reported 'stats'
4146 * as 'subfacet''s statistics. */
4148 update_subfacet_stats(struct subfacet *subfacet,
4149 const struct dpif_flow_stats *stats)
4151 struct facet *facet = subfacet->facet;
4153 if (stats->n_packets >= subfacet->dp_packet_count) {
4154 uint64_t extra = stats->n_packets - subfacet->dp_packet_count;
4155 facet->packet_count += extra;
4157 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
4160 if (stats->n_bytes >= subfacet->dp_byte_count) {
4161 facet->byte_count += stats->n_bytes - subfacet->dp_byte_count;
4163 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
4166 subfacet->dp_packet_count = stats->n_packets;
4167 subfacet->dp_byte_count = stats->n_bytes;
4169 facet->tcp_flags |= stats->tcp_flags;
4171 subfacet_update_time(subfacet, stats->used);
4172 if (facet->accounted_bytes < facet->byte_count) {
4174 facet_account(facet);
4175 facet->accounted_bytes = facet->byte_count;
4177 facet_push_stats(facet);
4180 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
4181 * about, or a flow that shouldn't be installed but was anyway. Delete it. */
4183 delete_unexpected_flow(struct ofproto_dpif *ofproto,
4184 const struct nlattr *key, size_t key_len)
4186 if (!VLOG_DROP_WARN(&rl)) {
4190 odp_flow_key_format(key, key_len, &s);
4191 VLOG_WARN("unexpected flow on %s: %s", ofproto->up.name, ds_cstr(&s));
4195 COVERAGE_INC(facet_unexpected);
4196 dpif_flow_del(ofproto->backer->dpif, key, key_len, NULL);
4199 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
4201 * This function also pushes statistics updates to rules which each facet
4202 * resubmits into. Generally these statistics will be accurate. However, if a
4203 * facet changes the rule it resubmits into at some time in between
4204 * update_stats() runs, it is possible that statistics accrued to the
4205 * old rule will be incorrectly attributed to the new rule. This could be
4206 * avoided by calling update_stats() whenever rules are created or
4207 * deleted. However, the performance impact of making so many calls to the
4208 * datapath do not justify the benefit of having perfectly accurate statistics.
4210 * In addition, this function maintains per ofproto flow hit counts. The patch
4211 * port is not treated specially. e.g. A packet ingress from br0 patched into
4212 * br1 will increase the hit count of br0 by 1, however, does not affect
4213 * the hit or miss counts of br1.
4216 update_stats(struct dpif_backer *backer)
4218 const struct dpif_flow_stats *stats;
4219 struct dpif_flow_dump dump;
4220 const struct nlattr *key;
4223 dpif_flow_dump_start(&dump, backer->dpif);
4224 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
4226 struct subfacet *subfacet;
4227 struct ofproto_dpif *ofproto;
4228 struct ofport_dpif *ofport;
4231 if (ofproto_receive(backer, NULL, key, key_len, &flow, NULL, &ofproto,
4236 ofproto->total_subfacet_count += hmap_count(&ofproto->subfacets);
4237 ofproto->n_update_stats++;
4238 update_moving_averages(ofproto);
4240 ofport = get_ofp_port(ofproto, flow.in_port);
4241 if (ofport && ofport->tnl_port) {
4242 netdev_vport_inc_rx(ofport->up.netdev, stats);
4245 key_hash = odp_flow_key_hash(key, key_len);
4246 subfacet = subfacet_find(ofproto, key, key_len, key_hash);
4247 switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
4249 /* Update ofproto_dpif's hit count. */
4250 if (stats->n_packets > subfacet->dp_packet_count) {
4251 uint64_t delta = stats->n_packets - subfacet->dp_packet_count;
4252 dpif_stats_update_hit_count(ofproto, delta);
4255 update_subfacet_stats(subfacet, stats);
4259 /* Stats are updated per-packet. */
4262 case SF_NOT_INSTALLED:
4264 delete_unexpected_flow(ofproto, key, key_len);
4268 dpif_flow_dump_done(&dump);
4271 /* Calculates and returns the number of milliseconds of idle time after which
4272 * subfacets should expire from the datapath. When a subfacet expires, we fold
4273 * its statistics into its facet, and when a facet's last subfacet expires, we
4274 * fold its statistic into its rule. */
4276 subfacet_max_idle(const struct ofproto_dpif *ofproto)
4279 * Idle time histogram.
4281 * Most of the time a switch has a relatively small number of subfacets.
4282 * When this is the case we might as well keep statistics for all of them
4283 * in userspace and to cache them in the kernel datapath for performance as
4286 * As the number of subfacets increases, the memory required to maintain
4287 * statistics about them in userspace and in the kernel becomes
4288 * significant. However, with a large number of subfacets it is likely
4289 * that only a few of them are "heavy hitters" that consume a large amount
4290 * of bandwidth. At this point, only heavy hitters are worth caching in
4291 * the kernel and maintaining in userspaces; other subfacets we can
4294 * The technique used to compute the idle time is to build a histogram with
4295 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
4296 * that is installed in the kernel gets dropped in the appropriate bucket.
4297 * After the histogram has been built, we compute the cutoff so that only
4298 * the most-recently-used 1% of subfacets (but at least
4299 * ofproto->up.flow_eviction_threshold flows) are kept cached. At least
4300 * the most-recently-used bucket of subfacets is kept, so actually an
4301 * arbitrary number of subfacets can be kept in any given expiration run
4302 * (though the next run will delete most of those unless they receive
4305 * This requires a second pass through the subfacets, in addition to the
4306 * pass made by update_stats(), because the former function never looks at
4307 * uninstallable subfacets.
4309 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
4310 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
4311 int buckets[N_BUCKETS] = { 0 };
4312 int total, subtotal, bucket;
4313 struct subfacet *subfacet;
4317 total = hmap_count(&ofproto->subfacets);
4318 if (total <= ofproto->up.flow_eviction_threshold) {
4319 return N_BUCKETS * BUCKET_WIDTH;
4322 /* Build histogram. */
4324 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
4325 long long int idle = now - subfacet->used;
4326 int bucket = (idle <= 0 ? 0
4327 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
4328 : (unsigned int) idle / BUCKET_WIDTH);
4332 /* Find the first bucket whose flows should be expired. */
4333 subtotal = bucket = 0;
4335 subtotal += buckets[bucket++];
4336 } while (bucket < N_BUCKETS &&
4337 subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100));
4339 if (VLOG_IS_DBG_ENABLED()) {
4343 ds_put_cstr(&s, "keep");
4344 for (i = 0; i < N_BUCKETS; i++) {
4346 ds_put_cstr(&s, ", drop");
4349 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
4352 VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s));
4356 return bucket * BUCKET_WIDTH;
4360 expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle)
4362 /* Cutoff time for most flows. */
4363 long long int normal_cutoff = time_msec() - dp_max_idle;
4365 /* We really want to keep flows for special protocols around, so use a more
4366 * conservative cutoff. */
4367 long long int special_cutoff = time_msec() - 10000;
4369 struct subfacet *subfacet, *next_subfacet;
4370 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
4374 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
4375 &ofproto->subfacets) {
4376 long long int cutoff;
4378 cutoff = (subfacet->slow & (SLOW_CFM | SLOW_LACP | SLOW_STP)
4381 if (subfacet->used < cutoff) {
4382 if (subfacet->path != SF_NOT_INSTALLED) {
4383 batch[n_batch++] = subfacet;
4384 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
4385 subfacet_destroy_batch(ofproto, batch, n_batch);
4389 subfacet_destroy(subfacet);
4395 subfacet_destroy_batch(ofproto, batch, n_batch);
4399 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
4400 * then delete it entirely. */
4402 rule_expire(struct rule_dpif *rule)
4404 struct facet *facet, *next_facet;
4408 if (rule->up.pending) {
4409 /* We'll have to expire it later. */
4413 /* Has 'rule' expired? */
4415 if (rule->up.hard_timeout
4416 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
4417 reason = OFPRR_HARD_TIMEOUT;
4418 } else if (rule->up.idle_timeout
4419 && now > rule->up.used + rule->up.idle_timeout * 1000) {
4420 reason = OFPRR_IDLE_TIMEOUT;
4425 COVERAGE_INC(ofproto_dpif_expired);
4427 /* Update stats. (This is a no-op if the rule expired due to an idle
4428 * timeout, because that only happens when the rule has no facets left.) */
4429 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4430 facet_remove(facet);
4433 /* Get rid of the rule. */
4434 ofproto_rule_expire(&rule->up, reason);
4439 /* Creates and returns a new facet owned by 'rule', given a 'flow'.
4441 * The caller must already have determined that no facet with an identical
4442 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
4443 * the ofproto's classifier table.
4445 * 'hash' must be the return value of flow_hash(flow, 0).
4447 * The facet will initially have no subfacets. The caller should create (at
4448 * least) one subfacet with subfacet_create(). */
4449 static struct facet *
4450 facet_create(struct rule_dpif *rule, const struct flow *flow, uint32_t hash)
4452 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4453 struct facet *facet;
4455 facet = xzalloc(sizeof *facet);
4456 facet->used = time_msec();
4457 hmap_insert(&ofproto->facets, &facet->hmap_node, hash);
4458 list_push_back(&rule->facets, &facet->list_node);
4460 facet->flow = *flow;
4461 list_init(&facet->subfacets);
4462 netflow_flow_init(&facet->nf_flow);
4463 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
4465 facet->learn_rl = time_msec() + 500;
4471 facet_free(struct facet *facet)
4476 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
4477 * 'packet', which arrived on 'in_port'. */
4479 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
4480 const struct nlattr *odp_actions, size_t actions_len,
4481 struct ofpbuf *packet)
4483 struct odputil_keybuf keybuf;
4487 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4488 odp_flow_key_from_flow(&key, flow,
4489 ofp_port_to_odp_port(ofproto, flow->in_port));
4491 error = dpif_execute(ofproto->backer->dpif, key.data, key.size,
4492 odp_actions, actions_len, packet);
4496 /* Remove 'facet' from 'ofproto' and free up the associated memory:
4498 * - If 'facet' was installed in the datapath, uninstalls it and updates its
4499 * rule's statistics, via subfacet_uninstall().
4501 * - Removes 'facet' from its rule and from ofproto->facets.
4504 facet_remove(struct facet *facet)
4506 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4507 struct subfacet *subfacet, *next_subfacet;
4509 ovs_assert(!list_is_empty(&facet->subfacets));
4511 /* First uninstall all of the subfacets to get final statistics. */
4512 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4513 subfacet_uninstall(subfacet);
4516 /* Flush the final stats to the rule.
4518 * This might require us to have at least one subfacet around so that we
4519 * can use its actions for accounting in facet_account(), which is why we
4520 * have uninstalled but not yet destroyed the subfacets. */
4521 facet_flush_stats(facet);
4523 /* Now we're really all done so destroy everything. */
4524 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
4525 &facet->subfacets) {
4526 subfacet_destroy__(subfacet);
4528 hmap_remove(&ofproto->facets, &facet->hmap_node);
4529 list_remove(&facet->list_node);
4533 /* Feed information from 'facet' back into the learning table to keep it in
4534 * sync with what is actually flowing through the datapath. */
4536 facet_learn(struct facet *facet)
4538 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4539 struct subfacet *subfacet= CONTAINER_OF(list_front(&facet->subfacets),
4540 struct subfacet, list_node);
4541 struct action_xlate_ctx ctx;
4543 if (time_msec() < facet->learn_rl) {
4547 facet->learn_rl = time_msec() + 500;
4549 if (!facet->has_learn
4550 && !facet->has_normal
4551 && (!facet->has_fin_timeout
4552 || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
4556 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
4557 &subfacet->initial_vals,
4558 facet->rule, facet->tcp_flags, NULL);
4559 ctx.may_learn = true;
4560 xlate_actions_for_side_effects(&ctx, facet->rule->up.ofpacts,
4561 facet->rule->up.ofpacts_len);
4565 facet_account(struct facet *facet)
4567 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4568 struct subfacet *subfacet;
4569 const struct nlattr *a;
4574 if (!facet->has_normal || !ofproto->has_bonded_bundles) {
4577 n_bytes = facet->byte_count - facet->accounted_bytes;
4579 /* This loop feeds byte counters to bond_account() for rebalancing to use
4580 * as a basis. We also need to track the actual VLAN on which the packet
4581 * is going to be sent to ensure that it matches the one passed to
4582 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
4585 * We use the actions from an arbitrary subfacet because they should all
4586 * be equally valid for our purpose. */
4587 subfacet = CONTAINER_OF(list_front(&facet->subfacets),
4588 struct subfacet, list_node);
4589 vlan_tci = facet->flow.vlan_tci;
4590 NL_ATTR_FOR_EACH_UNSAFE (a, left,
4591 subfacet->actions, subfacet->actions_len) {
4592 const struct ovs_action_push_vlan *vlan;
4593 struct ofport_dpif *port;
4595 switch (nl_attr_type(a)) {
4596 case OVS_ACTION_ATTR_OUTPUT:
4597 port = get_odp_port(ofproto, nl_attr_get_u32(a));
4598 if (port && port->bundle && port->bundle->bond) {
4599 bond_account(port->bundle->bond, &facet->flow,
4600 vlan_tci_to_vid(vlan_tci), n_bytes);
4604 case OVS_ACTION_ATTR_POP_VLAN:
4605 vlan_tci = htons(0);
4608 case OVS_ACTION_ATTR_PUSH_VLAN:
4609 vlan = nl_attr_get(a);
4610 vlan_tci = vlan->vlan_tci;
4616 /* Returns true if the only action for 'facet' is to send to the controller.
4617 * (We don't report NetFlow expiration messages for such facets because they
4618 * are just part of the control logic for the network, not real traffic). */
4620 facet_is_controller_flow(struct facet *facet)
4623 const struct rule *rule = &facet->rule->up;
4624 const struct ofpact *ofpacts = rule->ofpacts;
4625 size_t ofpacts_len = rule->ofpacts_len;
4627 if (ofpacts_len > 0 &&
4628 ofpacts->type == OFPACT_CONTROLLER &&
4629 ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len)) {
4636 /* Folds all of 'facet''s statistics into its rule. Also updates the
4637 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
4638 * 'facet''s statistics in the datapath should have been zeroed and folded into
4639 * its packet and byte counts before this function is called. */
4641 facet_flush_stats(struct facet *facet)
4643 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4644 struct subfacet *subfacet;
4646 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4647 ovs_assert(!subfacet->dp_byte_count);
4648 ovs_assert(!subfacet->dp_packet_count);
4651 facet_push_stats(facet);
4652 if (facet->accounted_bytes < facet->byte_count) {
4653 facet_account(facet);
4654 facet->accounted_bytes = facet->byte_count;
4657 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
4658 struct ofexpired expired;
4659 expired.flow = facet->flow;
4660 expired.packet_count = facet->packet_count;
4661 expired.byte_count = facet->byte_count;
4662 expired.used = facet->used;
4663 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4666 facet->rule->packet_count += facet->packet_count;
4667 facet->rule->byte_count += facet->byte_count;
4669 /* Reset counters to prevent double counting if 'facet' ever gets
4671 facet_reset_counters(facet);
4673 netflow_flow_clear(&facet->nf_flow);
4674 facet->tcp_flags = 0;
4677 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
4678 * Returns it if found, otherwise a null pointer.
4680 * 'hash' must be the return value of flow_hash(flow, 0).
4682 * The returned facet might need revalidation; use facet_lookup_valid()
4683 * instead if that is important. */
4684 static struct facet *
4685 facet_find(struct ofproto_dpif *ofproto,
4686 const struct flow *flow, uint32_t hash)
4688 struct facet *facet;
4690 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, hash, &ofproto->facets) {
4691 if (flow_equal(flow, &facet->flow)) {
4699 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
4700 * Returns it if found, otherwise a null pointer.
4702 * 'hash' must be the return value of flow_hash(flow, 0).
4704 * The returned facet is guaranteed to be valid. */
4705 static struct facet *
4706 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow,
4709 struct facet *facet;
4711 facet = facet_find(ofproto, flow, hash);
4713 && (ofproto->backer->need_revalidate
4714 || tag_set_intersects(&ofproto->backer->revalidate_set,
4716 facet_revalidate(facet);
4718 /* facet_revalidate() may have destroyed 'facet'. */
4719 facet = facet_find(ofproto, flow, hash);
4726 subfacet_path_to_string(enum subfacet_path path)
4729 case SF_NOT_INSTALLED:
4730 return "not installed";
4732 return "in fast path";
4734 return "in slow path";
4740 /* Returns the path in which a subfacet should be installed if its 'slow'
4741 * member has the specified value. */
4742 static enum subfacet_path
4743 subfacet_want_path(enum slow_path_reason slow)
4745 return slow ? SF_SLOW_PATH : SF_FAST_PATH;
4748 /* Returns true if 'subfacet' needs to have its datapath flow updated,
4749 * supposing that its actions have been recalculated as 'want_actions' and that
4750 * 'slow' is nonzero iff 'subfacet' should be in the slow path. */
4752 subfacet_should_install(struct subfacet *subfacet, enum slow_path_reason slow,
4753 const struct ofpbuf *want_actions)
4755 enum subfacet_path want_path = subfacet_want_path(slow);
4756 return (want_path != subfacet->path
4757 || (want_path == SF_FAST_PATH
4758 && (subfacet->actions_len != want_actions->size
4759 || memcmp(subfacet->actions, want_actions->data,
4760 subfacet->actions_len))));
4764 facet_check_consistency(struct facet *facet)
4766 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
4768 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4770 uint64_t odp_actions_stub[1024 / 8];
4771 struct ofpbuf odp_actions;
4773 struct rule_dpif *rule;
4774 struct subfacet *subfacet;
4775 bool may_log = false;
4778 /* Check the rule for consistency. */
4779 rule = rule_dpif_lookup(ofproto, &facet->flow);
4780 ok = rule == facet->rule;
4782 may_log = !VLOG_DROP_WARN(&rl);
4787 flow_format(&s, &facet->flow);
4788 ds_put_format(&s, ": facet associated with wrong rule (was "
4789 "table=%"PRIu8",", facet->rule->up.table_id);
4790 cls_rule_format(&facet->rule->up.cr, &s);
4791 ds_put_format(&s, ") (should have been table=%"PRIu8",",
4793 cls_rule_format(&rule->up.cr, &s);
4794 ds_put_char(&s, ')');
4796 VLOG_WARN("%s", ds_cstr(&s));
4801 /* Check the datapath actions for consistency. */
4802 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
4803 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4804 enum subfacet_path want_path;
4805 struct action_xlate_ctx ctx;
4808 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
4809 &subfacet->initial_vals, rule, 0, NULL);
4810 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len,
4813 if (subfacet->path == SF_NOT_INSTALLED) {
4814 /* This only happens if the datapath reported an error when we
4815 * tried to install the flow. Don't flag another error here. */
4819 want_path = subfacet_want_path(subfacet->slow);
4820 if (want_path == SF_SLOW_PATH && subfacet->path == SF_SLOW_PATH) {
4821 /* The actions for slow-path flows may legitimately vary from one
4822 * packet to the next. We're done. */
4826 if (!subfacet_should_install(subfacet, subfacet->slow, &odp_actions)) {
4830 /* Inconsistency! */
4832 may_log = !VLOG_DROP_WARN(&rl);
4836 /* Rate-limited, skip reporting. */
4841 odp_flow_key_format(subfacet->key, subfacet->key_len, &s);
4843 ds_put_cstr(&s, ": inconsistency in subfacet");
4844 if (want_path != subfacet->path) {
4845 enum odp_key_fitness fitness = subfacet->key_fitness;
4847 ds_put_format(&s, " (%s, fitness=%s)",
4848 subfacet_path_to_string(subfacet->path),
4849 odp_key_fitness_to_string(fitness));
4850 ds_put_format(&s, " (should have been %s)",
4851 subfacet_path_to_string(want_path));
4852 } else if (want_path == SF_FAST_PATH) {
4853 ds_put_cstr(&s, " (actions were: ");
4854 format_odp_actions(&s, subfacet->actions,
4855 subfacet->actions_len);
4856 ds_put_cstr(&s, ") (correct actions: ");
4857 format_odp_actions(&s, odp_actions.data, odp_actions.size);
4858 ds_put_char(&s, ')');
4860 ds_put_cstr(&s, " (actions: ");
4861 format_odp_actions(&s, subfacet->actions,
4862 subfacet->actions_len);
4863 ds_put_char(&s, ')');
4865 VLOG_WARN("%s", ds_cstr(&s));
4868 ofpbuf_uninit(&odp_actions);
4873 /* Re-searches the classifier for 'facet':
4875 * - If the rule found is different from 'facet''s current rule, moves
4876 * 'facet' to the new rule and recompiles its actions.
4878 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
4879 * where it is and recompiles its actions anyway.
4881 * - If any of 'facet''s subfacets correspond to a new flow according to
4882 * ofproto_receive(), 'facet' is removed. */
4884 facet_revalidate(struct facet *facet)
4886 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4888 struct nlattr *odp_actions;
4891 struct actions *new_actions;
4893 struct action_xlate_ctx ctx;
4894 uint64_t odp_actions_stub[1024 / 8];
4895 struct ofpbuf odp_actions;
4897 struct rule_dpif *new_rule;
4898 struct subfacet *subfacet;
4901 COVERAGE_INC(facet_revalidate);
4903 /* Check that child subfacets still correspond to this facet. Tunnel
4904 * configuration changes could cause a subfacet's OpenFlow in_port to
4906 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4907 struct ofproto_dpif *recv_ofproto;
4908 struct flow recv_flow;
4911 error = ofproto_receive(ofproto->backer, NULL, subfacet->key,
4912 subfacet->key_len, &recv_flow, NULL,
4913 &recv_ofproto, NULL, NULL);
4915 || recv_ofproto != ofproto
4916 || memcmp(&recv_flow, &facet->flow, sizeof recv_flow)) {
4917 facet_remove(facet);
4922 new_rule = rule_dpif_lookup(ofproto, &facet->flow);
4924 /* Calculate new datapath actions.
4926 * We do not modify any 'facet' state yet, because we might need to, e.g.,
4927 * emit a NetFlow expiration and, if so, we need to have the old state
4928 * around to properly compose it. */
4930 /* If the datapath actions changed or the installability changed,
4931 * then we need to talk to the datapath. */
4934 memset(&ctx, 0, sizeof ctx);
4935 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
4936 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4937 enum slow_path_reason slow;
4939 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
4940 &subfacet->initial_vals, new_rule, 0, NULL);
4941 xlate_actions(&ctx, new_rule->up.ofpacts, new_rule->up.ofpacts_len,
4944 slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
4945 if (subfacet_should_install(subfacet, slow, &odp_actions)) {
4946 struct dpif_flow_stats stats;
4948 subfacet_install(subfacet,
4949 odp_actions.data, odp_actions.size, &stats, slow);
4950 subfacet_update_stats(subfacet, &stats);
4953 new_actions = xcalloc(list_size(&facet->subfacets),
4954 sizeof *new_actions);
4956 new_actions[i].odp_actions = xmemdup(odp_actions.data,
4958 new_actions[i].actions_len = odp_actions.size;
4963 ofpbuf_uninit(&odp_actions);
4966 facet_flush_stats(facet);
4969 /* Update 'facet' now that we've taken care of all the old state. */
4970 facet->tags = ctx.tags;
4971 facet->nf_flow.output_iface = ctx.nf_output_iface;
4972 facet->has_learn = ctx.has_learn;
4973 facet->has_normal = ctx.has_normal;
4974 facet->has_fin_timeout = ctx.has_fin_timeout;
4975 facet->mirrors = ctx.mirrors;
4978 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4979 subfacet->slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
4981 if (new_actions && new_actions[i].odp_actions) {
4982 free(subfacet->actions);
4983 subfacet->actions = new_actions[i].odp_actions;
4984 subfacet->actions_len = new_actions[i].actions_len;
4990 if (facet->rule != new_rule) {
4991 COVERAGE_INC(facet_changed_rule);
4992 list_remove(&facet->list_node);
4993 list_push_back(&new_rule->facets, &facet->list_node);
4994 facet->rule = new_rule;
4995 facet->used = new_rule->up.created;
4996 facet->prev_used = facet->used;
5000 /* Updates 'facet''s used time. Caller is responsible for calling
5001 * facet_push_stats() to update the flows which 'facet' resubmits into. */
5003 facet_update_time(struct facet *facet, long long int used)
5005 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5006 if (used > facet->used) {
5008 ofproto_rule_update_used(&facet->rule->up, used);
5009 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
5014 facet_reset_counters(struct facet *facet)
5016 facet->packet_count = 0;
5017 facet->byte_count = 0;
5018 facet->prev_packet_count = 0;
5019 facet->prev_byte_count = 0;
5020 facet->accounted_bytes = 0;
5024 facet_push_stats(struct facet *facet)
5026 struct dpif_flow_stats stats;
5028 ovs_assert(facet->packet_count >= facet->prev_packet_count);
5029 ovs_assert(facet->byte_count >= facet->prev_byte_count);
5030 ovs_assert(facet->used >= facet->prev_used);
5032 stats.n_packets = facet->packet_count - facet->prev_packet_count;
5033 stats.n_bytes = facet->byte_count - facet->prev_byte_count;
5034 stats.used = facet->used;
5035 stats.tcp_flags = 0;
5037 if (stats.n_packets || stats.n_bytes || facet->used > facet->prev_used) {
5038 facet->prev_packet_count = facet->packet_count;
5039 facet->prev_byte_count = facet->byte_count;
5040 facet->prev_used = facet->used;
5042 flow_push_stats(facet, &stats);
5044 update_mirror_stats(ofproto_dpif_cast(facet->rule->up.ofproto),
5045 facet->mirrors, stats.n_packets, stats.n_bytes);
5050 push_all_stats(void)
5052 static long long int rl = LLONG_MIN;
5053 struct ofproto_dpif *ofproto;
5055 if (time_msec() < rl) {
5059 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5060 struct facet *facet;
5062 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
5063 facet_push_stats(facet);
5067 rl = time_msec() + 100;
5071 rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats)
5073 rule->packet_count += stats->n_packets;
5074 rule->byte_count += stats->n_bytes;
5075 ofproto_rule_update_used(&rule->up, stats->used);
5078 /* Pushes flow statistics to the rules which 'facet->flow' resubmits
5079 * into given 'facet->rule''s actions and mirrors. */
5081 flow_push_stats(struct facet *facet, const struct dpif_flow_stats *stats)
5083 struct rule_dpif *rule = facet->rule;
5084 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5085 struct subfacet *subfacet = CONTAINER_OF(list_front(&facet->subfacets),
5086 struct subfacet, list_node);
5087 struct action_xlate_ctx ctx;
5089 ofproto_rule_update_used(&rule->up, stats->used);
5091 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
5092 &subfacet->initial_vals, rule, 0, NULL);
5093 ctx.resubmit_stats = stats;
5094 xlate_actions_for_side_effects(&ctx, rule->up.ofpacts,
5095 rule->up.ofpacts_len);
5100 static struct subfacet *
5101 subfacet_find(struct ofproto_dpif *ofproto,
5102 const struct nlattr *key, size_t key_len, uint32_t key_hash)
5104 struct subfacet *subfacet;
5106 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
5107 &ofproto->subfacets) {
5108 if (subfacet->key_len == key_len
5109 && !memcmp(key, subfacet->key, key_len)) {
5117 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
5118 * 'key_fitness', 'key', and 'key_len' members in 'miss'. Returns the
5119 * existing subfacet if there is one, otherwise creates and returns a
5122 * If the returned subfacet is new, then subfacet->actions will be NULL, in
5123 * which case the caller must populate the actions with
5124 * subfacet_make_actions(). */
5125 static struct subfacet *
5126 subfacet_create(struct facet *facet, struct flow_miss *miss,
5129 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5130 enum odp_key_fitness key_fitness = miss->key_fitness;
5131 const struct nlattr *key = miss->key;
5132 size_t key_len = miss->key_len;
5134 struct subfacet *subfacet;
5136 key_hash = odp_flow_key_hash(key, key_len);
5138 if (list_is_empty(&facet->subfacets)) {
5139 subfacet = &facet->one_subfacet;
5141 subfacet = subfacet_find(ofproto, key, key_len, key_hash);
5143 if (subfacet->facet == facet) {
5147 /* This shouldn't happen. */
5148 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
5149 subfacet_destroy(subfacet);
5152 subfacet = xmalloc(sizeof *subfacet);
5155 hmap_insert(&ofproto->subfacets, &subfacet->hmap_node, key_hash);
5156 list_push_back(&facet->subfacets, &subfacet->list_node);
5157 subfacet->facet = facet;
5158 subfacet->key_fitness = key_fitness;
5159 subfacet->key = xmemdup(key, key_len);
5160 subfacet->key_len = key_len;
5161 subfacet->used = now;
5162 subfacet->created = now;
5163 subfacet->dp_packet_count = 0;
5164 subfacet->dp_byte_count = 0;
5165 subfacet->actions_len = 0;
5166 subfacet->actions = NULL;
5167 subfacet->slow = (subfacet->key_fitness == ODP_FIT_TOO_LITTLE
5170 subfacet->path = SF_NOT_INSTALLED;
5171 subfacet->initial_vals = miss->initial_vals;
5172 subfacet->odp_in_port = miss->odp_in_port;
5174 ofproto->subfacet_add_count++;
5178 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
5179 * its facet within 'ofproto', and frees it. */
5181 subfacet_destroy__(struct subfacet *subfacet)
5183 struct facet *facet = subfacet->facet;
5184 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5186 /* Update ofproto stats before uninstall the subfacet. */
5187 ofproto->subfacet_del_count++;
5188 ofproto->total_subfacet_life_span += (time_msec() - subfacet->created);
5190 subfacet_uninstall(subfacet);
5191 hmap_remove(&ofproto->subfacets, &subfacet->hmap_node);
5192 list_remove(&subfacet->list_node);
5193 free(subfacet->key);
5194 free(subfacet->actions);
5195 if (subfacet != &facet->one_subfacet) {
5200 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
5201 * last remaining subfacet in its facet destroys the facet too. */
5203 subfacet_destroy(struct subfacet *subfacet)
5205 struct facet *facet = subfacet->facet;
5207 if (list_is_singleton(&facet->subfacets)) {
5208 /* facet_remove() needs at least one subfacet (it will remove it). */
5209 facet_remove(facet);
5211 subfacet_destroy__(subfacet);
5216 subfacet_destroy_batch(struct ofproto_dpif *ofproto,
5217 struct subfacet **subfacets, int n)
5219 struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH];
5220 struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH];
5221 struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH];
5224 for (i = 0; i < n; i++) {
5225 ops[i].type = DPIF_OP_FLOW_DEL;
5226 ops[i].u.flow_del.key = subfacets[i]->key;
5227 ops[i].u.flow_del.key_len = subfacets[i]->key_len;
5228 ops[i].u.flow_del.stats = &stats[i];
5232 dpif_operate(ofproto->backer->dpif, opsp, n);
5233 for (i = 0; i < n; i++) {
5234 subfacet_reset_dp_stats(subfacets[i], &stats[i]);
5235 subfacets[i]->path = SF_NOT_INSTALLED;
5236 subfacet_destroy(subfacets[i]);
5240 /* Composes the datapath actions for 'subfacet' based on its rule's actions.
5241 * Translates the actions into 'odp_actions', which the caller must have
5242 * initialized and is responsible for uninitializing. */
5244 subfacet_make_actions(struct subfacet *subfacet, const struct ofpbuf *packet,
5245 struct ofpbuf *odp_actions)
5247 struct facet *facet = subfacet->facet;
5248 struct rule_dpif *rule = facet->rule;
5249 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5251 struct action_xlate_ctx ctx;
5253 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
5254 &subfacet->initial_vals, rule, 0, packet);
5255 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, odp_actions);
5256 facet->tags = ctx.tags;
5257 facet->has_learn = ctx.has_learn;
5258 facet->has_normal = ctx.has_normal;
5259 facet->has_fin_timeout = ctx.has_fin_timeout;
5260 facet->nf_flow.output_iface = ctx.nf_output_iface;
5261 facet->mirrors = ctx.mirrors;
5263 subfacet->slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
5264 if (subfacet->actions_len != odp_actions->size
5265 || memcmp(subfacet->actions, odp_actions->data, odp_actions->size)) {
5266 free(subfacet->actions);
5267 subfacet->actions_len = odp_actions->size;
5268 subfacet->actions = xmemdup(odp_actions->data, odp_actions->size);
5272 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
5273 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
5274 * in the datapath will be zeroed and 'stats' will be updated with traffic new
5275 * since 'subfacet' was last updated.
5277 * Returns 0 if successful, otherwise a positive errno value. */
5279 subfacet_install(struct subfacet *subfacet,
5280 const struct nlattr *actions, size_t actions_len,
5281 struct dpif_flow_stats *stats,
5282 enum slow_path_reason slow)
5284 struct facet *facet = subfacet->facet;
5285 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5286 enum subfacet_path path = subfacet_want_path(slow);
5287 uint64_t slow_path_stub[128 / 8];
5288 enum dpif_flow_put_flags flags;
5291 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
5293 flags |= DPIF_FP_ZERO_STATS;
5296 if (path == SF_SLOW_PATH) {
5297 compose_slow_path(ofproto, &facet->flow, slow,
5298 slow_path_stub, sizeof slow_path_stub,
5299 &actions, &actions_len);
5302 ret = dpif_flow_put(ofproto->backer->dpif, flags, subfacet->key,
5303 subfacet->key_len, actions, actions_len, stats);
5306 subfacet_reset_dp_stats(subfacet, stats);
5310 subfacet->path = path;
5316 subfacet_reinstall(struct subfacet *subfacet, struct dpif_flow_stats *stats)
5318 return subfacet_install(subfacet, subfacet->actions, subfacet->actions_len,
5319 stats, subfacet->slow);
5322 /* If 'subfacet' is installed in the datapath, uninstalls it. */
5324 subfacet_uninstall(struct subfacet *subfacet)
5326 if (subfacet->path != SF_NOT_INSTALLED) {
5327 struct rule_dpif *rule = subfacet->facet->rule;
5328 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5329 struct dpif_flow_stats stats;
5332 error = dpif_flow_del(ofproto->backer->dpif, subfacet->key,
5333 subfacet->key_len, &stats);
5334 subfacet_reset_dp_stats(subfacet, &stats);
5336 subfacet_update_stats(subfacet, &stats);
5338 subfacet->path = SF_NOT_INSTALLED;
5340 ovs_assert(subfacet->dp_packet_count == 0);
5341 ovs_assert(subfacet->dp_byte_count == 0);
5345 /* Resets 'subfacet''s datapath statistics counters. This should be called
5346 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
5347 * non-null, it should contain the statistics returned by dpif when 'subfacet'
5348 * was reset in the datapath. 'stats' will be modified to include only
5349 * statistics new since 'subfacet' was last updated. */
5351 subfacet_reset_dp_stats(struct subfacet *subfacet,
5352 struct dpif_flow_stats *stats)
5355 && subfacet->dp_packet_count <= stats->n_packets
5356 && subfacet->dp_byte_count <= stats->n_bytes) {
5357 stats->n_packets -= subfacet->dp_packet_count;
5358 stats->n_bytes -= subfacet->dp_byte_count;
5361 subfacet->dp_packet_count = 0;
5362 subfacet->dp_byte_count = 0;
5365 /* Updates 'subfacet''s used time. The caller is responsible for calling
5366 * facet_push_stats() to update the flows which 'subfacet' resubmits into. */
5368 subfacet_update_time(struct subfacet *subfacet, long long int used)
5370 if (used > subfacet->used) {
5371 subfacet->used = used;
5372 facet_update_time(subfacet->facet, used);
5376 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
5378 * Because of the meaning of a subfacet's counters, it only makes sense to do
5379 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
5380 * represents a packet that was sent by hand or if it represents statistics
5381 * that have been cleared out of the datapath. */
5383 subfacet_update_stats(struct subfacet *subfacet,
5384 const struct dpif_flow_stats *stats)
5386 if (stats->n_packets || stats->used > subfacet->used) {
5387 struct facet *facet = subfacet->facet;
5389 subfacet_update_time(subfacet, stats->used);
5390 facet->packet_count += stats->n_packets;
5391 facet->byte_count += stats->n_bytes;
5392 facet->tcp_flags |= stats->tcp_flags;
5393 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
5399 static struct rule_dpif *
5400 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow)
5402 struct rule_dpif *rule;
5404 rule = rule_dpif_lookup__(ofproto, flow, 0);
5409 return rule_dpif_miss_rule(ofproto, flow);
5412 static struct rule_dpif *
5413 rule_dpif_lookup__(struct ofproto_dpif *ofproto, const struct flow *flow,
5416 struct cls_rule *cls_rule;
5417 struct classifier *cls;
5419 if (table_id >= N_TABLES) {
5423 cls = &ofproto->up.tables[table_id].cls;
5424 if (flow->nw_frag & FLOW_NW_FRAG_ANY
5425 && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
5426 /* For OFPC_NORMAL frag_handling, we must pretend that transport ports
5427 * are unavailable. */
5428 struct flow ofpc_normal_flow = *flow;
5429 ofpc_normal_flow.tp_src = htons(0);
5430 ofpc_normal_flow.tp_dst = htons(0);
5431 cls_rule = classifier_lookup(cls, &ofpc_normal_flow);
5433 cls_rule = classifier_lookup(cls, flow);
5435 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
5438 static struct rule_dpif *
5439 rule_dpif_miss_rule(struct ofproto_dpif *ofproto, const struct flow *flow)
5441 struct ofport_dpif *port;
5443 port = get_ofp_port(ofproto, flow->in_port);
5445 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, flow->in_port);
5446 return ofproto->miss_rule;
5449 if (port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN) {
5450 return ofproto->no_packet_in_rule;
5452 return ofproto->miss_rule;
5456 complete_operation(struct rule_dpif *rule)
5458 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5460 rule_invalidate(rule);
5462 struct dpif_completion *c = xmalloc(sizeof *c);
5463 c->op = rule->up.pending;
5464 list_push_back(&ofproto->completions, &c->list_node);
5466 ofoperation_complete(rule->up.pending, 0);
5470 static struct rule *
5473 struct rule_dpif *rule = xmalloc(sizeof *rule);
5478 rule_dealloc(struct rule *rule_)
5480 struct rule_dpif *rule = rule_dpif_cast(rule_);
5485 rule_construct(struct rule *rule_)
5487 struct rule_dpif *rule = rule_dpif_cast(rule_);
5488 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5489 struct rule_dpif *victim;
5492 rule->packet_count = 0;
5493 rule->byte_count = 0;
5495 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
5496 if (victim && !list_is_empty(&victim->facets)) {
5497 struct facet *facet;
5499 rule->facets = victim->facets;
5500 list_moved(&rule->facets);
5501 LIST_FOR_EACH (facet, list_node, &rule->facets) {
5502 /* XXX: We're only clearing our local counters here. It's possible
5503 * that quite a few packets are unaccounted for in the datapath
5504 * statistics. These will be accounted to the new rule instead of
5505 * cleared as required. This could be fixed by clearing out the
5506 * datapath statistics for this facet, but currently it doesn't
5508 facet_reset_counters(facet);
5512 /* Must avoid list_moved() in this case. */
5513 list_init(&rule->facets);
5516 table_id = rule->up.table_id;
5518 rule->tag = victim->tag;
5519 } else if (table_id == 0) {
5524 miniflow_expand(&rule->up.cr.match.flow, &flow);
5525 rule->tag = rule_calculate_tag(&flow, &rule->up.cr.match.mask,
5526 ofproto->tables[table_id].basis);
5529 complete_operation(rule);
5534 rule_destruct(struct rule *rule_)
5536 struct rule_dpif *rule = rule_dpif_cast(rule_);
5537 struct facet *facet, *next_facet;
5539 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
5540 facet_revalidate(facet);
5543 complete_operation(rule);
5547 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
5549 struct rule_dpif *rule = rule_dpif_cast(rule_);
5550 struct facet *facet;
5554 /* Start from historical data for 'rule' itself that are no longer tracked
5555 * in facets. This counts, for example, facets that have expired. */
5556 *packets = rule->packet_count;
5557 *bytes = rule->byte_count;
5559 /* Add any statistics that are tracked by facets. This includes
5560 * statistical data recently updated by ofproto_update_stats() as well as
5561 * stats for packets that were executed "by hand" via dpif_execute(). */
5562 LIST_FOR_EACH (facet, list_node, &rule->facets) {
5563 *packets += facet->packet_count;
5564 *bytes += facet->byte_count;
5569 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
5570 struct ofpbuf *packet)
5572 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5573 struct initial_vals initial_vals;
5574 struct dpif_flow_stats stats;
5575 struct action_xlate_ctx ctx;
5576 uint64_t odp_actions_stub[1024 / 8];
5577 struct ofpbuf odp_actions;
5579 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
5580 rule_credit_stats(rule, &stats);
5582 initial_vals.vlan_tci = flow->vlan_tci;
5583 initial_vals.tunnel_ip_tos = flow->tunnel.ip_tos;
5584 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5585 action_xlate_ctx_init(&ctx, ofproto, flow, &initial_vals,
5586 rule, stats.tcp_flags, packet);
5587 ctx.resubmit_stats = &stats;
5588 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, &odp_actions);
5590 execute_odp_actions(ofproto, flow, odp_actions.data,
5591 odp_actions.size, packet);
5593 ofpbuf_uninit(&odp_actions);
5597 rule_execute(struct rule *rule, const struct flow *flow,
5598 struct ofpbuf *packet)
5600 rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
5601 ofpbuf_delete(packet);
5606 rule_modify_actions(struct rule *rule_)
5608 struct rule_dpif *rule = rule_dpif_cast(rule_);
5610 complete_operation(rule);
5613 /* Sends 'packet' out 'ofport'.
5614 * May modify 'packet'.
5615 * Returns 0 if successful, otherwise a positive errno value. */
5617 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
5619 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
5620 uint64_t odp_actions_stub[1024 / 8];
5621 struct ofpbuf key, odp_actions;
5622 struct odputil_keybuf keybuf;
5627 flow_extract(packet, 0, 0, NULL, OFPP_LOCAL, &flow);
5628 if (netdev_vport_is_patch(ofport->up.netdev)) {
5629 struct ofproto_dpif *peer_ofproto;
5630 struct dpif_flow_stats stats;
5631 struct ofport_dpif *peer;
5632 struct rule_dpif *rule;
5634 peer = ofport_get_peer(ofport);
5639 dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
5640 netdev_vport_inc_tx(ofport->up.netdev, &stats);
5641 netdev_vport_inc_rx(peer->up.netdev, &stats);
5643 flow.in_port = peer->up.ofp_port;
5644 peer_ofproto = ofproto_dpif_cast(peer->up.ofproto);
5645 rule = rule_dpif_lookup(peer_ofproto, &flow);
5646 rule_dpif_execute(rule, &flow, packet);
5651 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5653 if (ofport->tnl_port) {
5654 struct dpif_flow_stats stats;
5656 odp_port = tnl_port_send(ofport->tnl_port, &flow);
5657 if (odp_port == OVSP_NONE) {
5661 dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
5662 netdev_vport_inc_tx(ofport->up.netdev, &stats);
5663 odp_put_tunnel_action(&flow.tunnel, &odp_actions);
5664 odp_put_skb_mark_action(flow.skb_mark, &odp_actions);
5666 odp_port = vsp_realdev_to_vlandev(ofproto, ofport->odp_port,
5668 if (odp_port != ofport->odp_port) {
5669 eth_pop_vlan(packet);
5670 flow.vlan_tci = htons(0);
5674 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5675 odp_flow_key_from_flow(&key, &flow,
5676 ofp_port_to_odp_port(ofproto, flow.in_port));
5678 compose_sflow_action(ofproto, &odp_actions, &flow, odp_port);
5680 nl_msg_put_u32(&odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port);
5681 error = dpif_execute(ofproto->backer->dpif,
5683 odp_actions.data, odp_actions.size,
5685 ofpbuf_uninit(&odp_actions);
5688 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)",
5689 ofproto->up.name, odp_port, strerror(error));
5691 ofproto_update_local_port_stats(ofport->up.ofproto, packet->size, 0);
5695 /* OpenFlow to datapath action translation. */
5697 static bool may_receive(const struct ofport_dpif *, struct action_xlate_ctx *);
5698 static void do_xlate_actions(const struct ofpact *, size_t ofpacts_len,
5699 struct action_xlate_ctx *);
5700 static void xlate_normal(struct action_xlate_ctx *);
5702 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
5703 * The action will state 'slow' as the reason that the action is in the slow
5704 * path. (This is purely informational: it allows a human viewing "ovs-dpctl
5705 * dump-flows" output to see why a flow is in the slow path.)
5707 * The 'stub_size' bytes in 'stub' will be used to store the action.
5708 * 'stub_size' must be large enough for the action.
5710 * The action and its size will be stored in '*actionsp' and '*actions_lenp',
5713 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
5714 enum slow_path_reason slow,
5715 uint64_t *stub, size_t stub_size,
5716 const struct nlattr **actionsp, size_t *actions_lenp)
5718 union user_action_cookie cookie;
5721 cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
5722 cookie.slow_path.unused = 0;
5723 cookie.slow_path.reason = slow;
5725 ofpbuf_use_stack(&buf, stub, stub_size);
5726 if (slow & (SLOW_CFM | SLOW_LACP | SLOW_STP)) {
5727 uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif, UINT32_MAX);
5728 odp_put_userspace_action(pid, &cookie, &buf);
5730 put_userspace_action(ofproto, &buf, flow, &cookie);
5732 *actionsp = buf.data;
5733 *actions_lenp = buf.size;
5737 put_userspace_action(const struct ofproto_dpif *ofproto,
5738 struct ofpbuf *odp_actions,
5739 const struct flow *flow,
5740 const union user_action_cookie *cookie)
5744 pid = dpif_port_get_pid(ofproto->backer->dpif,
5745 ofp_port_to_odp_port(ofproto, flow->in_port));
5747 return odp_put_userspace_action(pid, cookie, odp_actions);
5751 compose_sflow_cookie(const struct ofproto_dpif *ofproto,
5752 ovs_be16 vlan_tci, uint32_t odp_port,
5753 unsigned int n_outputs, union user_action_cookie *cookie)
5757 cookie->type = USER_ACTION_COOKIE_SFLOW;
5758 cookie->sflow.vlan_tci = vlan_tci;
5760 /* See http://www.sflow.org/sflow_version_5.txt (search for "Input/output
5761 * port information") for the interpretation of cookie->output. */
5762 switch (n_outputs) {
5764 /* 0x40000000 | 256 means "packet dropped for unknown reason". */
5765 cookie->sflow.output = 0x40000000 | 256;
5769 ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port);
5771 cookie->sflow.output = ifindex;
5776 /* 0x80000000 means "multiple output ports. */
5777 cookie->sflow.output = 0x80000000 | n_outputs;
5782 /* Compose SAMPLE action for sFlow. */
5784 compose_sflow_action(const struct ofproto_dpif *ofproto,
5785 struct ofpbuf *odp_actions,
5786 const struct flow *flow,
5789 uint32_t probability;
5790 union user_action_cookie cookie;
5791 size_t sample_offset, actions_offset;
5794 if (!ofproto->sflow || flow->in_port == OFPP_NONE) {
5798 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
5800 /* Number of packets out of UINT_MAX to sample. */
5801 probability = dpif_sflow_get_probability(ofproto->sflow);
5802 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
5804 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
5805 compose_sflow_cookie(ofproto, htons(0), odp_port,
5806 odp_port == OVSP_NONE ? 0 : 1, &cookie);
5807 cookie_offset = put_userspace_action(ofproto, odp_actions, flow, &cookie);
5809 nl_msg_end_nested(odp_actions, actions_offset);
5810 nl_msg_end_nested(odp_actions, sample_offset);
5811 return cookie_offset;
5814 /* SAMPLE action must be first action in any given list of actions.
5815 * At this point we do not have all information required to build it. So try to
5816 * build sample action as complete as possible. */
5818 add_sflow_action(struct action_xlate_ctx *ctx)
5820 ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto,
5822 &ctx->flow, OVSP_NONE);
5823 ctx->sflow_odp_port = 0;
5824 ctx->sflow_n_outputs = 0;
5827 /* Fix SAMPLE action according to data collected while composing ODP actions.
5828 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
5829 * USERSPACE action's user-cookie which is required for sflow. */
5831 fix_sflow_action(struct action_xlate_ctx *ctx)
5833 const struct flow *base = &ctx->base_flow;
5834 union user_action_cookie *cookie;
5836 if (!ctx->user_cookie_offset) {
5840 cookie = ofpbuf_at(ctx->odp_actions, ctx->user_cookie_offset,
5842 ovs_assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
5844 compose_sflow_cookie(ctx->ofproto, base->vlan_tci,
5845 ctx->sflow_odp_port, ctx->sflow_n_outputs, cookie);
5849 compose_output_action__(struct action_xlate_ctx *ctx, uint16_t ofp_port,
5852 const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
5853 ovs_be16 flow_vlan_tci = ctx->flow.vlan_tci;
5854 ovs_be64 flow_tun_id = ctx->flow.tunnel.tun_id;
5855 uint8_t flow_nw_tos = ctx->flow.nw_tos;
5856 struct priority_to_dscp *pdscp;
5857 uint32_t out_port, odp_port;
5859 /* If 'struct flow' gets additional metadata, we'll need to zero it out
5860 * before traversing a patch port. */
5861 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 18);
5864 xlate_report(ctx, "Nonexistent output port");
5866 } else if (ofport->up.pp.config & OFPUTIL_PC_NO_FWD) {
5867 xlate_report(ctx, "OFPPC_NO_FWD set, skipping output");
5869 } else if (check_stp && !stp_forward_in_state(ofport->stp_state)) {
5870 xlate_report(ctx, "STP not in forwarding state, skipping output");
5874 if (netdev_vport_is_patch(ofport->up.netdev)) {
5875 struct ofport_dpif *peer = ofport_get_peer(ofport);
5876 struct flow old_flow = ctx->flow;
5877 const struct ofproto_dpif *peer_ofproto;
5878 enum slow_path_reason special;
5879 struct ofport_dpif *in_port;
5882 xlate_report(ctx, "Nonexistent patch port peer");
5886 peer_ofproto = ofproto_dpif_cast(peer->up.ofproto);
5887 if (peer_ofproto->backer != ctx->ofproto->backer) {
5888 xlate_report(ctx, "Patch port peer on a different datapath");
5892 ctx->ofproto = ofproto_dpif_cast(peer->up.ofproto);
5893 ctx->flow.in_port = peer->up.ofp_port;
5894 ctx->flow.metadata = htonll(0);
5895 memset(&ctx->flow.tunnel, 0, sizeof ctx->flow.tunnel);
5896 memset(ctx->flow.regs, 0, sizeof ctx->flow.regs);
5898 in_port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
5899 special = process_special(ctx->ofproto, &ctx->flow, in_port,
5902 ctx->slow |= special;
5903 } else if (!in_port || may_receive(in_port, ctx)) {
5904 if (!in_port || stp_forward_in_state(in_port->stp_state)) {
5905 xlate_table_action(ctx, ctx->flow.in_port, 0, true);
5907 /* Forwarding is disabled by STP. Let OFPP_NORMAL and the
5908 * learning action look at the packet, then drop it. */
5909 struct flow old_base_flow = ctx->base_flow;
5910 size_t old_size = ctx->odp_actions->size;
5911 xlate_table_action(ctx, ctx->flow.in_port, 0, true);
5912 ctx->base_flow = old_base_flow;
5913 ctx->odp_actions->size = old_size;
5917 ctx->flow = old_flow;
5918 ctx->ofproto = ofproto_dpif_cast(ofport->up.ofproto);
5920 if (ctx->resubmit_stats) {
5921 netdev_vport_inc_tx(ofport->up.netdev, ctx->resubmit_stats);
5922 netdev_vport_inc_rx(peer->up.netdev, ctx->resubmit_stats);
5928 pdscp = get_priority(ofport, ctx->flow.skb_priority);
5930 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
5931 ctx->flow.nw_tos |= pdscp->dscp;
5934 odp_port = ofp_port_to_odp_port(ctx->ofproto, ofp_port);
5935 if (ofport->tnl_port) {
5936 odp_port = tnl_port_send(ofport->tnl_port, &ctx->flow);
5937 if (odp_port == OVSP_NONE) {
5938 xlate_report(ctx, "Tunneling decided against output");
5942 if (ctx->resubmit_stats) {
5943 netdev_vport_inc_tx(ofport->up.netdev, ctx->resubmit_stats);
5945 out_port = odp_port;
5946 commit_odp_tunnel_action(&ctx->flow, &ctx->base_flow,
5949 out_port = vsp_realdev_to_vlandev(ctx->ofproto, odp_port,
5950 ctx->flow.vlan_tci);
5951 if (out_port != odp_port) {
5952 ctx->flow.vlan_tci = htons(0);
5954 ctx->flow.skb_mark &= ~IPSEC_MARK;
5956 commit_odp_actions(&ctx->flow, &ctx->base_flow, ctx->odp_actions);
5957 nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_OUTPUT, out_port);
5959 ctx->sflow_odp_port = odp_port;
5960 ctx->sflow_n_outputs++;
5961 ctx->nf_output_iface = ofp_port;
5962 ctx->flow.tunnel.tun_id = flow_tun_id;
5963 ctx->flow.vlan_tci = flow_vlan_tci;
5964 ctx->flow.nw_tos = flow_nw_tos;
5968 compose_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port)
5970 compose_output_action__(ctx, ofp_port, true);
5974 xlate_table_action(struct action_xlate_ctx *ctx,
5975 uint16_t in_port, uint8_t table_id, bool may_packet_in)
5977 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
5978 struct ofproto_dpif *ofproto = ctx->ofproto;
5979 struct rule_dpif *rule;
5980 uint16_t old_in_port;
5981 uint8_t old_table_id;
5983 old_table_id = ctx->table_id;
5984 ctx->table_id = table_id;
5986 /* Look up a flow with 'in_port' as the input port. */
5987 old_in_port = ctx->flow.in_port;
5988 ctx->flow.in_port = in_port;
5989 rule = rule_dpif_lookup__(ofproto, &ctx->flow, table_id);
5992 if (table_id > 0 && table_id < N_TABLES) {
5993 struct table_dpif *table = &ofproto->tables[table_id];
5994 if (table->other_table) {
5995 ctx->tags |= (rule && rule->tag
5997 : rule_calculate_tag(&ctx->flow,
5998 &table->other_table->mask,
6003 /* Restore the original input port. Otherwise OFPP_NORMAL and
6004 * OFPP_IN_PORT will have surprising behavior. */
6005 ctx->flow.in_port = old_in_port;
6007 if (ctx->resubmit_hook) {
6008 ctx->resubmit_hook(ctx, rule);
6011 if (rule == NULL && may_packet_in) {
6013 * check if table configuration flags
6014 * OFPTC_TABLE_MISS_CONTROLLER, default.
6015 * OFPTC_TABLE_MISS_CONTINUE,
6016 * OFPTC_TABLE_MISS_DROP
6017 * When OF1.0, OFPTC_TABLE_MISS_CONTINUE is used. What to do?
6019 rule = rule_dpif_miss_rule(ofproto, &ctx->flow);
6023 struct rule_dpif *old_rule = ctx->rule;
6025 if (ctx->resubmit_stats) {
6026 rule_credit_stats(rule, ctx->resubmit_stats);
6031 do_xlate_actions(rule->up.ofpacts, rule->up.ofpacts_len, ctx);
6032 ctx->rule = old_rule;
6036 ctx->table_id = old_table_id;
6038 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
6040 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
6041 MAX_RESUBMIT_RECURSION);
6042 ctx->max_resubmit_trigger = true;
6047 xlate_ofpact_resubmit(struct action_xlate_ctx *ctx,
6048 const struct ofpact_resubmit *resubmit)
6053 in_port = resubmit->in_port;
6054 if (in_port == OFPP_IN_PORT) {
6055 in_port = ctx->flow.in_port;
6058 table_id = resubmit->table_id;
6059 if (table_id == 255) {
6060 table_id = ctx->table_id;
6063 xlate_table_action(ctx, in_port, table_id, false);
6067 flood_packets(struct action_xlate_ctx *ctx, bool all)
6069 struct ofport_dpif *ofport;
6071 HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) {
6072 uint16_t ofp_port = ofport->up.ofp_port;
6074 if (ofp_port == ctx->flow.in_port) {
6079 compose_output_action__(ctx, ofp_port, false);
6080 } else if (!(ofport->up.pp.config & OFPUTIL_PC_NO_FLOOD)) {
6081 compose_output_action(ctx, ofp_port);
6085 ctx->nf_output_iface = NF_OUT_FLOOD;
6089 execute_controller_action(struct action_xlate_ctx *ctx, int len,
6090 enum ofp_packet_in_reason reason,
6091 uint16_t controller_id)
6093 struct ofputil_packet_in pin;
6094 struct ofpbuf *packet;
6096 ctx->slow |= SLOW_CONTROLLER;
6101 packet = ofpbuf_clone(ctx->packet);
6103 if (packet->l2 && packet->l3) {
6104 struct eth_header *eh;
6106 eth_pop_vlan(packet);
6109 /* If the Ethernet type is less than ETH_TYPE_MIN, it's likely an 802.2
6110 * LLC frame. Calculating the Ethernet type of these frames is more
6111 * trouble than seems appropriate for a simple assertion. */
6112 ovs_assert(ntohs(eh->eth_type) < ETH_TYPE_MIN
6113 || eh->eth_type == ctx->flow.dl_type);
6115 memcpy(eh->eth_src, ctx->flow.dl_src, sizeof eh->eth_src);
6116 memcpy(eh->eth_dst, ctx->flow.dl_dst, sizeof eh->eth_dst);
6118 if (ctx->flow.vlan_tci & htons(VLAN_CFI)) {
6119 eth_push_vlan(packet, ctx->flow.vlan_tci);
6123 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
6124 packet_set_ipv4(packet, ctx->flow.nw_src, ctx->flow.nw_dst,
6125 ctx->flow.nw_tos, ctx->flow.nw_ttl);
6129 if (ctx->flow.nw_proto == IPPROTO_TCP) {
6130 packet_set_tcp_port(packet, ctx->flow.tp_src,
6132 } else if (ctx->flow.nw_proto == IPPROTO_UDP) {
6133 packet_set_udp_port(packet, ctx->flow.tp_src,
6140 pin.packet = packet->data;
6141 pin.packet_len = packet->size;
6142 pin.reason = reason;
6143 pin.controller_id = controller_id;
6144 pin.table_id = ctx->table_id;
6145 pin.cookie = ctx->rule ? ctx->rule->up.flow_cookie : 0;
6148 flow_get_metadata(&ctx->flow, &pin.fmd);
6150 connmgr_send_packet_in(ctx->ofproto->up.connmgr, &pin);
6151 ofpbuf_delete(packet);
6155 compose_dec_ttl(struct action_xlate_ctx *ctx, struct ofpact_cnt_ids *ids)
6157 if (ctx->flow.dl_type != htons(ETH_TYPE_IP) &&
6158 ctx->flow.dl_type != htons(ETH_TYPE_IPV6)) {
6162 if (ctx->flow.nw_ttl > 1) {
6168 for (i = 0; i < ids->n_controllers; i++) {
6169 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL,
6173 /* Stop processing for current table. */
6179 xlate_output_action(struct action_xlate_ctx *ctx,
6180 uint16_t port, uint16_t max_len, bool may_packet_in)
6182 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
6184 ctx->nf_output_iface = NF_OUT_DROP;
6188 compose_output_action(ctx, ctx->flow.in_port);
6191 xlate_table_action(ctx, ctx->flow.in_port, 0, may_packet_in);
6197 flood_packets(ctx, false);
6200 flood_packets(ctx, true);
6202 case OFPP_CONTROLLER:
6203 execute_controller_action(ctx, max_len, OFPR_ACTION, 0);
6209 if (port != ctx->flow.in_port) {
6210 compose_output_action(ctx, port);
6212 xlate_report(ctx, "skipping output to input port");
6217 if (prev_nf_output_iface == NF_OUT_FLOOD) {
6218 ctx->nf_output_iface = NF_OUT_FLOOD;
6219 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
6220 ctx->nf_output_iface = prev_nf_output_iface;
6221 } else if (prev_nf_output_iface != NF_OUT_DROP &&
6222 ctx->nf_output_iface != NF_OUT_FLOOD) {
6223 ctx->nf_output_iface = NF_OUT_MULTI;
6228 xlate_output_reg_action(struct action_xlate_ctx *ctx,
6229 const struct ofpact_output_reg *or)
6231 uint64_t port = mf_get_subfield(&or->src, &ctx->flow);
6232 if (port <= UINT16_MAX) {
6233 xlate_output_action(ctx, port, or->max_len, false);
6238 xlate_enqueue_action(struct action_xlate_ctx *ctx,
6239 const struct ofpact_enqueue *enqueue)
6241 uint16_t ofp_port = enqueue->port;
6242 uint32_t queue_id = enqueue->queue;
6243 uint32_t flow_priority, priority;
6246 /* Translate queue to priority. */
6247 error = dpif_queue_to_priority(ctx->ofproto->backer->dpif,
6248 queue_id, &priority);
6250 /* Fall back to ordinary output action. */
6251 xlate_output_action(ctx, enqueue->port, 0, false);
6255 /* Check output port. */
6256 if (ofp_port == OFPP_IN_PORT) {
6257 ofp_port = ctx->flow.in_port;
6258 } else if (ofp_port == ctx->flow.in_port) {
6262 /* Add datapath actions. */
6263 flow_priority = ctx->flow.skb_priority;
6264 ctx->flow.skb_priority = priority;
6265 compose_output_action(ctx, ofp_port);
6266 ctx->flow.skb_priority = flow_priority;
6268 /* Update NetFlow output port. */
6269 if (ctx->nf_output_iface == NF_OUT_DROP) {
6270 ctx->nf_output_iface = ofp_port;
6271 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
6272 ctx->nf_output_iface = NF_OUT_MULTI;
6277 xlate_set_queue_action(struct action_xlate_ctx *ctx, uint32_t queue_id)
6279 uint32_t skb_priority;
6281 if (!dpif_queue_to_priority(ctx->ofproto->backer->dpif,
6282 queue_id, &skb_priority)) {
6283 ctx->flow.skb_priority = skb_priority;
6285 /* Couldn't translate queue to a priority. Nothing to do. A warning
6286 * has already been logged. */
6290 struct xlate_reg_state {
6296 xlate_autopath(struct action_xlate_ctx *ctx,
6297 const struct ofpact_autopath *ap)
6299 uint16_t ofp_port = ap->port;
6300 struct ofport_dpif *port = get_ofp_port(ctx->ofproto, ofp_port);
6302 if (!port || !port->bundle) {
6303 ofp_port = OFPP_NONE;
6304 } else if (port->bundle->bond) {
6305 /* Autopath does not support VLAN hashing. */
6306 struct ofport_dpif *slave = bond_choose_output_slave(
6307 port->bundle->bond, &ctx->flow, 0, &ctx->tags);
6309 ofp_port = slave->up.ofp_port;
6312 nxm_reg_load(&ap->dst, ofp_port, &ctx->flow);
6316 slave_enabled_cb(uint16_t ofp_port, void *ofproto_)
6318 struct ofproto_dpif *ofproto = ofproto_;
6319 struct ofport_dpif *port;
6329 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
6332 port = get_ofp_port(ofproto, ofp_port);
6333 return port ? port->may_enable : false;
6338 xlate_bundle_action(struct action_xlate_ctx *ctx,
6339 const struct ofpact_bundle *bundle)
6343 port = bundle_execute(bundle, &ctx->flow, slave_enabled_cb, ctx->ofproto);
6344 if (bundle->dst.field) {
6345 nxm_reg_load(&bundle->dst, port, &ctx->flow);
6347 xlate_output_action(ctx, port, 0, false);
6352 xlate_learn_action(struct action_xlate_ctx *ctx,
6353 const struct ofpact_learn *learn)
6355 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
6356 struct ofputil_flow_mod fm;
6357 uint64_t ofpacts_stub[1024 / 8];
6358 struct ofpbuf ofpacts;
6361 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
6362 learn_execute(learn, &ctx->flow, &fm, &ofpacts);
6364 error = ofproto_flow_mod(&ctx->ofproto->up, &fm);
6365 if (error && !VLOG_DROP_WARN(&rl)) {
6366 VLOG_WARN("learning action failed to modify flow table (%s)",
6367 ofperr_get_name(error));
6370 ofpbuf_uninit(&ofpacts);
6373 /* Reduces '*timeout' to no more than 'max'. A value of zero in either case
6374 * means "infinite". */
6376 reduce_timeout(uint16_t max, uint16_t *timeout)
6378 if (max && (!*timeout || *timeout > max)) {
6384 xlate_fin_timeout(struct action_xlate_ctx *ctx,
6385 const struct ofpact_fin_timeout *oft)
6387 if (ctx->tcp_flags & (TCP_FIN | TCP_RST) && ctx->rule) {
6388 struct rule_dpif *rule = ctx->rule;
6390 reduce_timeout(oft->fin_idle_timeout, &rule->up.idle_timeout);
6391 reduce_timeout(oft->fin_hard_timeout, &rule->up.hard_timeout);
6396 may_receive(const struct ofport_dpif *port, struct action_xlate_ctx *ctx)
6398 if (port->up.pp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
6399 ? OFPUTIL_PC_NO_RECV_STP
6400 : OFPUTIL_PC_NO_RECV)) {
6404 /* Only drop packets here if both forwarding and learning are
6405 * disabled. If just learning is enabled, we need to have
6406 * OFPP_NORMAL and the learning action have a look at the packet
6407 * before we can drop it. */
6408 if (!stp_forward_in_state(port->stp_state)
6409 && !stp_learn_in_state(port->stp_state)) {
6417 tunnel_ecn_ok(struct action_xlate_ctx *ctx)
6419 if (is_ip_any(&ctx->base_flow)
6420 && (ctx->base_flow.tunnel.ip_tos & IP_ECN_MASK) == IP_ECN_CE) {
6421 if ((ctx->base_flow.nw_tos & IP_ECN_MASK) == IP_ECN_NOT_ECT) {
6422 VLOG_WARN_RL(&rl, "dropping tunnel packet marked ECN CE"
6423 " but is not ECN capable");
6426 /* Set the ECN CE value in the tunneled packet. */
6427 ctx->flow.nw_tos |= IP_ECN_CE;
6435 do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len,
6436 struct action_xlate_ctx *ctx)
6438 bool was_evictable = true;
6439 const struct ofpact *a;
6442 /* Don't let the rule we're working on get evicted underneath us. */
6443 was_evictable = ctx->rule->up.evictable;
6444 ctx->rule->up.evictable = false;
6446 OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) {
6447 struct ofpact_controller *controller;
6448 const struct ofpact_metadata *metadata;
6456 xlate_output_action(ctx, ofpact_get_OUTPUT(a)->port,
6457 ofpact_get_OUTPUT(a)->max_len, true);
6460 case OFPACT_CONTROLLER:
6461 controller = ofpact_get_CONTROLLER(a);
6462 execute_controller_action(ctx, controller->max_len,
6464 controller->controller_id);
6467 case OFPACT_ENQUEUE:
6468 xlate_enqueue_action(ctx, ofpact_get_ENQUEUE(a));
6471 case OFPACT_SET_VLAN_VID:
6472 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
6473 ctx->flow.vlan_tci |= (htons(ofpact_get_SET_VLAN_VID(a)->vlan_vid)
6477 case OFPACT_SET_VLAN_PCP:
6478 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
6479 ctx->flow.vlan_tci |= htons((ofpact_get_SET_VLAN_PCP(a)->vlan_pcp
6484 case OFPACT_STRIP_VLAN:
6485 ctx->flow.vlan_tci = htons(0);
6488 case OFPACT_PUSH_VLAN:
6489 /* XXX 802.1AD(QinQ) */
6490 ctx->flow.vlan_tci = htons(VLAN_CFI);
6493 case OFPACT_SET_ETH_SRC:
6494 memcpy(ctx->flow.dl_src, ofpact_get_SET_ETH_SRC(a)->mac,
6498 case OFPACT_SET_ETH_DST:
6499 memcpy(ctx->flow.dl_dst, ofpact_get_SET_ETH_DST(a)->mac,
6503 case OFPACT_SET_IPV4_SRC:
6504 ctx->flow.nw_src = ofpact_get_SET_IPV4_SRC(a)->ipv4;
6507 case OFPACT_SET_IPV4_DST:
6508 ctx->flow.nw_dst = ofpact_get_SET_IPV4_DST(a)->ipv4;
6511 case OFPACT_SET_IPV4_DSCP:
6512 /* OpenFlow 1.0 only supports IPv4. */
6513 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
6514 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
6515 ctx->flow.nw_tos |= ofpact_get_SET_IPV4_DSCP(a)->dscp;
6519 case OFPACT_SET_L4_SRC_PORT:
6520 ctx->flow.tp_src = htons(ofpact_get_SET_L4_SRC_PORT(a)->port);
6523 case OFPACT_SET_L4_DST_PORT:
6524 ctx->flow.tp_dst = htons(ofpact_get_SET_L4_DST_PORT(a)->port);
6527 case OFPACT_RESUBMIT:
6528 xlate_ofpact_resubmit(ctx, ofpact_get_RESUBMIT(a));
6531 case OFPACT_SET_TUNNEL:
6532 ctx->flow.tunnel.tun_id = htonll(ofpact_get_SET_TUNNEL(a)->tun_id);
6535 case OFPACT_SET_QUEUE:
6536 xlate_set_queue_action(ctx, ofpact_get_SET_QUEUE(a)->queue_id);
6539 case OFPACT_POP_QUEUE:
6540 ctx->flow.skb_priority = ctx->orig_skb_priority;
6543 case OFPACT_REG_MOVE:
6544 nxm_execute_reg_move(ofpact_get_REG_MOVE(a), &ctx->flow);
6547 case OFPACT_REG_LOAD:
6548 nxm_execute_reg_load(ofpact_get_REG_LOAD(a), &ctx->flow);
6551 case OFPACT_DEC_TTL:
6552 if (compose_dec_ttl(ctx, ofpact_get_DEC_TTL(a))) {
6558 /* Nothing to do. */
6561 case OFPACT_MULTIPATH:
6562 multipath_execute(ofpact_get_MULTIPATH(a), &ctx->flow);
6565 case OFPACT_AUTOPATH:
6566 xlate_autopath(ctx, ofpact_get_AUTOPATH(a));
6570 ctx->ofproto->has_bundle_action = true;
6571 xlate_bundle_action(ctx, ofpact_get_BUNDLE(a));
6574 case OFPACT_OUTPUT_REG:
6575 xlate_output_reg_action(ctx, ofpact_get_OUTPUT_REG(a));
6579 ctx->has_learn = true;
6580 if (ctx->may_learn) {
6581 xlate_learn_action(ctx, ofpact_get_LEARN(a));
6589 case OFPACT_FIN_TIMEOUT:
6590 ctx->has_fin_timeout = true;
6591 xlate_fin_timeout(ctx, ofpact_get_FIN_TIMEOUT(a));
6594 case OFPACT_CLEAR_ACTIONS:
6596 * Nothing to do because writa-actions is not supported for now.
6597 * When writa-actions is supported, clear-actions also must
6598 * be supported at the same time.
6602 case OFPACT_WRITE_METADATA:
6603 metadata = ofpact_get_WRITE_METADATA(a);
6604 ctx->flow.metadata &= ~metadata->mask;
6605 ctx->flow.metadata |= metadata->metadata & metadata->mask;
6608 case OFPACT_GOTO_TABLE: {
6609 /* XXX remove recursion */
6610 /* It is assumed that goto-table is last action */
6611 struct ofpact_goto_table *ogt = ofpact_get_GOTO_TABLE(a);
6612 ovs_assert(ctx->table_id < ogt->table_id);
6613 xlate_table_action(ctx, ctx->flow.in_port, ogt->table_id, true);
6621 ctx->rule->up.evictable = was_evictable;
6626 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
6627 struct ofproto_dpif *ofproto, const struct flow *flow,
6628 const struct initial_vals *initial_vals,
6629 struct rule_dpif *rule,
6630 uint8_t tcp_flags, const struct ofpbuf *packet)
6632 ovs_be64 initial_tun_id = flow->tunnel.tun_id;
6634 /* Flow initialization rules:
6635 * - 'base_flow' must match the kernel's view of the packet at the
6636 * time that action processing starts. 'flow' represents any
6637 * transformations we wish to make through actions.
6638 * - By default 'base_flow' and 'flow' are the same since the input
6639 * packet matches the output before any actions are applied.
6640 * - When using VLAN splinters, 'base_flow''s VLAN is set to the value
6641 * of the received packet as seen by the kernel. If we later output
6642 * to another device without any modifications this will cause us to
6643 * insert a new tag since the original one was stripped off by the
6645 * - Tunnel 'flow' is largely cleared when transitioning between
6646 * the input and output stages since it does not make sense to output
6647 * a packet with the exact headers that it was received with (i.e.
6648 * the destination IP is us). The one exception is the tun_id, which
6649 * is preserved to allow use in later resubmit lookups and loads into
6651 * - Tunnel 'base_flow' is completely cleared since that is what the
6652 * kernel does. If we wish to maintain the original values an action
6653 * needs to be generated. */
6655 ctx->ofproto = ofproto;
6657 memset(&ctx->flow.tunnel, 0, sizeof ctx->flow.tunnel);
6658 ctx->base_flow = ctx->flow;
6659 ctx->base_flow.vlan_tci = initial_vals->vlan_tci;
6660 ctx->base_flow.tunnel.ip_tos = initial_vals->tunnel_ip_tos;
6661 ctx->flow.tunnel.tun_id = initial_tun_id;
6663 ctx->packet = packet;
6664 ctx->may_learn = packet != NULL;
6665 ctx->tcp_flags = tcp_flags;
6666 ctx->resubmit_hook = NULL;
6667 ctx->report_hook = NULL;
6668 ctx->resubmit_stats = NULL;
6671 /* Translates the 'ofpacts_len' bytes of "struct ofpacts" starting at 'ofpacts'
6672 * into datapath actions in 'odp_actions', using 'ctx'. */
6674 xlate_actions(struct action_xlate_ctx *ctx,
6675 const struct ofpact *ofpacts, size_t ofpacts_len,
6676 struct ofpbuf *odp_actions)
6678 /* Normally false. Set to true if we ever hit MAX_RESUBMIT_RECURSION, so
6679 * that in the future we always keep a copy of the original flow for
6680 * tracing purposes. */
6681 static bool hit_resubmit_limit;
6683 enum slow_path_reason special;
6684 struct ofport_dpif *in_port;
6686 COVERAGE_INC(ofproto_dpif_xlate);
6688 ofpbuf_clear(odp_actions);
6689 ofpbuf_reserve(odp_actions, NL_A_U32_SIZE);
6691 ctx->odp_actions = odp_actions;
6694 ctx->has_learn = false;
6695 ctx->has_normal = false;
6696 ctx->has_fin_timeout = false;
6697 ctx->nf_output_iface = NF_OUT_DROP;
6700 ctx->max_resubmit_trigger = false;
6701 ctx->orig_skb_priority = ctx->flow.skb_priority;
6705 if (ctx->ofproto->has_mirrors || hit_resubmit_limit) {
6706 /* Do this conditionally because the copy is expensive enough that it
6707 * shows up in profiles.
6709 * We keep orig_flow in 'ctx' only because I couldn't make GCC 4.4
6710 * believe that I wasn't using it without initializing it if I kept it
6711 * in a local variable. */
6712 ctx->orig_flow = ctx->flow;
6715 if (ctx->flow.nw_frag & FLOW_NW_FRAG_ANY) {
6716 switch (ctx->ofproto->up.frag_handling) {
6717 case OFPC_FRAG_NORMAL:
6718 /* We must pretend that transport ports are unavailable. */
6719 ctx->flow.tp_src = ctx->base_flow.tp_src = htons(0);
6720 ctx->flow.tp_dst = ctx->base_flow.tp_dst = htons(0);
6723 case OFPC_FRAG_DROP:
6726 case OFPC_FRAG_REASM:
6729 case OFPC_FRAG_NX_MATCH:
6730 /* Nothing to do. */
6733 case OFPC_INVALID_TTL_TO_CONTROLLER:
6738 in_port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
6739 special = process_special(ctx->ofproto, &ctx->flow, in_port, ctx->packet);
6741 ctx->slow |= special;
6743 static struct vlog_rate_limit trace_rl = VLOG_RATE_LIMIT_INIT(1, 1);
6744 struct initial_vals initial_vals;
6745 uint32_t local_odp_port;
6747 initial_vals.vlan_tci = ctx->base_flow.vlan_tci;
6748 initial_vals.tunnel_ip_tos = ctx->base_flow.tunnel.ip_tos;
6750 add_sflow_action(ctx);
6752 if (tunnel_ecn_ok(ctx) && (!in_port || may_receive(in_port, ctx))) {
6753 do_xlate_actions(ofpacts, ofpacts_len, ctx);
6755 /* We've let OFPP_NORMAL and the learning action look at the
6756 * packet, so drop it now if forwarding is disabled. */
6757 if (in_port && !stp_forward_in_state(in_port->stp_state)) {
6758 ofpbuf_clear(ctx->odp_actions);
6759 add_sflow_action(ctx);
6763 if (ctx->max_resubmit_trigger && !ctx->resubmit_hook) {
6764 if (!hit_resubmit_limit) {
6765 /* We didn't record the original flow. Make sure we do from
6767 hit_resubmit_limit = true;
6768 } else if (!VLOG_DROP_ERR(&trace_rl)) {
6769 struct ds ds = DS_EMPTY_INITIALIZER;
6771 ofproto_trace(ctx->ofproto, &ctx->orig_flow, ctx->packet,
6772 &initial_vals, &ds);
6773 VLOG_ERR("Trace triggered by excessive resubmit "
6774 "recursion:\n%s", ds_cstr(&ds));
6779 local_odp_port = ofp_port_to_odp_port(ctx->ofproto, OFPP_LOCAL);
6780 if (!connmgr_may_set_up_flow(ctx->ofproto->up.connmgr, &ctx->flow,
6782 ctx->odp_actions->data,
6783 ctx->odp_actions->size)) {
6784 ctx->slow |= SLOW_IN_BAND;
6786 && connmgr_msg_in_hook(ctx->ofproto->up.connmgr, &ctx->flow,
6788 compose_output_action(ctx, OFPP_LOCAL);
6791 if (ctx->ofproto->has_mirrors) {
6792 add_mirror_actions(ctx, &ctx->orig_flow);
6794 fix_sflow_action(ctx);
6798 /* Translates the 'ofpacts_len' bytes of "struct ofpact"s starting at 'ofpacts'
6799 * into datapath actions, using 'ctx', and discards the datapath actions. */
6801 xlate_actions_for_side_effects(struct action_xlate_ctx *ctx,
6802 const struct ofpact *ofpacts,
6805 uint64_t odp_actions_stub[1024 / 8];
6806 struct ofpbuf odp_actions;
6808 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
6809 xlate_actions(ctx, ofpacts, ofpacts_len, &odp_actions);
6810 ofpbuf_uninit(&odp_actions);
6814 xlate_report(struct action_xlate_ctx *ctx, const char *s)
6816 if (ctx->report_hook) {
6817 ctx->report_hook(ctx, s);
6821 /* OFPP_NORMAL implementation. */
6823 static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *);
6825 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
6826 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle',
6827 * the bundle on which the packet was received, returns the VLAN to which the
6830 * Both 'vid' and the return value are in the range 0...4095. */
6832 input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid)
6834 switch (in_bundle->vlan_mode) {
6835 case PORT_VLAN_ACCESS:
6836 return in_bundle->vlan;
6839 case PORT_VLAN_TRUNK:
6842 case PORT_VLAN_NATIVE_UNTAGGED:
6843 case PORT_VLAN_NATIVE_TAGGED:
6844 return vid ? vid : in_bundle->vlan;
6851 /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'.
6852 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
6855 * 'vid' should be the VID obtained from the 802.1Q header that was received as
6856 * part of a packet (specify 0 if there was no 802.1Q header), in the range
6859 input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn)
6861 /* Allow any VID on the OFPP_NONE port. */
6862 if (in_bundle == &ofpp_none_bundle) {
6866 switch (in_bundle->vlan_mode) {
6867 case PORT_VLAN_ACCESS:
6870 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6871 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
6872 "packet received on port %s configured as VLAN "
6873 "%"PRIu16" access port",
6874 in_bundle->ofproto->up.name, vid,
6875 in_bundle->name, in_bundle->vlan);
6881 case PORT_VLAN_NATIVE_UNTAGGED:
6882 case PORT_VLAN_NATIVE_TAGGED:
6884 /* Port must always carry its native VLAN. */
6888 case PORT_VLAN_TRUNK:
6889 if (!ofbundle_includes_vlan(in_bundle, vid)) {
6891 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6892 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet "
6893 "received on port %s not configured for trunking "
6895 in_bundle->ofproto->up.name, vid,
6896 in_bundle->name, vid);
6908 /* Given 'vlan', the VLAN that a packet belongs to, and
6909 * 'out_bundle', a bundle on which the packet is to be output, returns the VID
6910 * that should be included in the 802.1Q header. (If the return value is 0,
6911 * then the 802.1Q header should only be included in the packet if there is a
6914 * Both 'vlan' and the return value are in the range 0...4095. */
6916 output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan)
6918 switch (out_bundle->vlan_mode) {
6919 case PORT_VLAN_ACCESS:
6922 case PORT_VLAN_TRUNK:
6923 case PORT_VLAN_NATIVE_TAGGED:
6926 case PORT_VLAN_NATIVE_UNTAGGED:
6927 return vlan == out_bundle->vlan ? 0 : vlan;
6935 output_normal(struct action_xlate_ctx *ctx, const struct ofbundle *out_bundle,
6938 struct ofport_dpif *port;
6940 ovs_be16 tci, old_tci;
6942 vid = output_vlan_to_vid(out_bundle, vlan);
6943 if (!out_bundle->bond) {
6944 port = ofbundle_get_a_port(out_bundle);
6946 port = bond_choose_output_slave(out_bundle->bond, &ctx->flow,
6949 /* No slaves enabled, so drop packet. */
6954 old_tci = ctx->flow.vlan_tci;
6956 if (tci || out_bundle->use_priority_tags) {
6957 tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK);
6959 tci |= htons(VLAN_CFI);
6962 ctx->flow.vlan_tci = tci;
6964 compose_output_action(ctx, port->up.ofp_port);
6965 ctx->flow.vlan_tci = old_tci;
6969 mirror_mask_ffs(mirror_mask_t mask)
6971 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
6976 ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
6978 return (bundle->vlan_mode != PORT_VLAN_ACCESS
6979 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
6983 ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan)
6985 return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan);
6988 /* Returns an arbitrary interface within 'bundle'. */
6989 static struct ofport_dpif *
6990 ofbundle_get_a_port(const struct ofbundle *bundle)
6992 return CONTAINER_OF(list_front(&bundle->ports),
6993 struct ofport_dpif, bundle_node);
6997 vlan_is_mirrored(const struct ofmirror *m, int vlan)
6999 return !m->vlans || bitmap_is_set(m->vlans, vlan);
7003 add_mirror_actions(struct action_xlate_ctx *ctx, const struct flow *orig_flow)
7005 struct ofproto_dpif *ofproto = ctx->ofproto;
7006 mirror_mask_t mirrors;
7007 struct ofbundle *in_bundle;
7010 const struct nlattr *a;
7013 in_bundle = lookup_input_bundle(ctx->ofproto, orig_flow->in_port,
7014 ctx->packet != NULL, NULL);
7018 mirrors = in_bundle->src_mirrors;
7020 /* Drop frames on bundles reserved for mirroring. */
7021 if (in_bundle->mirror_out) {
7022 if (ctx->packet != NULL) {
7023 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7024 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
7025 "%s, which is reserved exclusively for mirroring",
7026 ctx->ofproto->up.name, in_bundle->name);
7032 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
7033 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
7036 vlan = input_vid_to_vlan(in_bundle, vid);
7038 /* Look at the output ports to check for destination selections. */
7040 NL_ATTR_FOR_EACH (a, left, ctx->odp_actions->data,
7041 ctx->odp_actions->size) {
7042 enum ovs_action_attr type = nl_attr_type(a);
7043 struct ofport_dpif *ofport;
7045 if (type != OVS_ACTION_ATTR_OUTPUT) {
7049 ofport = get_odp_port(ofproto, nl_attr_get_u32(a));
7050 if (ofport && ofport->bundle) {
7051 mirrors |= ofport->bundle->dst_mirrors;
7059 /* Restore the original packet before adding the mirror actions. */
7060 ctx->flow = *orig_flow;
7065 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
7067 if (!vlan_is_mirrored(m, vlan)) {
7068 mirrors = zero_rightmost_1bit(mirrors);
7072 mirrors &= ~m->dup_mirrors;
7073 ctx->mirrors |= m->dup_mirrors;
7075 output_normal(ctx, m->out, vlan);
7076 } else if (vlan != m->out_vlan
7077 && !eth_addr_is_reserved(orig_flow->dl_dst)) {
7078 struct ofbundle *bundle;
7080 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
7081 if (ofbundle_includes_vlan(bundle, m->out_vlan)
7082 && !bundle->mirror_out) {
7083 output_normal(ctx, bundle, m->out_vlan);
7091 update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors,
7092 uint64_t packets, uint64_t bytes)
7098 for (; mirrors; mirrors = zero_rightmost_1bit(mirrors)) {
7101 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
7104 /* In normal circumstances 'm' will not be NULL. However,
7105 * if mirrors are reconfigured, we can temporarily get out
7106 * of sync in facet_revalidate(). We could "correct" the
7107 * mirror list before reaching here, but doing that would
7108 * not properly account the traffic stats we've currently
7109 * accumulated for previous mirror configuration. */
7113 m->packet_count += packets;
7114 m->byte_count += bytes;
7118 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
7119 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
7120 * indicate this; newer upstream kernels use gratuitous ARP requests. */
7122 is_gratuitous_arp(const struct flow *flow)
7124 return (flow->dl_type == htons(ETH_TYPE_ARP)
7125 && eth_addr_is_broadcast(flow->dl_dst)
7126 && (flow->nw_proto == ARP_OP_REPLY
7127 || (flow->nw_proto == ARP_OP_REQUEST
7128 && flow->nw_src == flow->nw_dst)));
7132 update_learning_table(struct ofproto_dpif *ofproto,
7133 const struct flow *flow, int vlan,
7134 struct ofbundle *in_bundle)
7136 struct mac_entry *mac;
7138 /* Don't learn the OFPP_NONE port. */
7139 if (in_bundle == &ofpp_none_bundle) {
7143 if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
7147 mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan);
7148 if (is_gratuitous_arp(flow)) {
7149 /* We don't want to learn from gratuitous ARP packets that are
7150 * reflected back over bond slaves so we lock the learning table. */
7151 if (!in_bundle->bond) {
7152 mac_entry_set_grat_arp_lock(mac);
7153 } else if (mac_entry_is_grat_arp_locked(mac)) {
7158 if (mac_entry_is_new(mac) || mac->port.p != in_bundle) {
7159 /* The log messages here could actually be useful in debugging,
7160 * so keep the rate limit relatively high. */
7161 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
7162 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
7163 "on port %s in VLAN %d",
7164 ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src),
7165 in_bundle->name, vlan);
7167 mac->port.p = in_bundle;
7168 tag_set_add(&ofproto->backer->revalidate_set,
7169 mac_learning_changed(ofproto->ml, mac));
7173 static struct ofbundle *
7174 lookup_input_bundle(const struct ofproto_dpif *ofproto, uint16_t in_port,
7175 bool warn, struct ofport_dpif **in_ofportp)
7177 struct ofport_dpif *ofport;
7179 /* Find the port and bundle for the received packet. */
7180 ofport = get_ofp_port(ofproto, in_port);
7182 *in_ofportp = ofport;
7184 if (ofport && ofport->bundle) {
7185 return ofport->bundle;
7188 /* Special-case OFPP_NONE, which a controller may use as the ingress
7189 * port for traffic that it is sourcing. */
7190 if (in_port == OFPP_NONE) {
7191 return &ofpp_none_bundle;
7194 /* Odd. A few possible reasons here:
7196 * - We deleted a port but there are still a few packets queued up
7199 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
7200 * we don't know about.
7202 * - The ofproto client didn't configure the port as part of a bundle.
7203 * This is particularly likely to happen if a packet was received on the
7204 * port after it was created, but before the client had a chance to
7205 * configure its bundle.
7208 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7210 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
7211 "port %"PRIu16, ofproto->up.name, in_port);
7216 /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or
7217 * dropped. Returns true if they may be forwarded, false if they should be
7220 * 'in_port' must be the ofport_dpif that corresponds to flow->in_port.
7221 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
7223 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
7224 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
7225 * checked by input_vid_is_valid().
7227 * May also add tags to '*tags', although the current implementation only does
7228 * so in one special case.
7231 is_admissible(struct action_xlate_ctx *ctx, struct ofport_dpif *in_port,
7234 struct ofproto_dpif *ofproto = ctx->ofproto;
7235 struct flow *flow = &ctx->flow;
7236 struct ofbundle *in_bundle = in_port->bundle;
7238 /* Drop frames for reserved multicast addresses
7239 * only if forward_bpdu option is absent. */
7240 if (!ofproto->up.forward_bpdu && eth_addr_is_reserved(flow->dl_dst)) {
7241 xlate_report(ctx, "packet has reserved destination MAC, dropping");
7245 if (in_bundle->bond) {
7246 struct mac_entry *mac;
7248 switch (bond_check_admissibility(in_bundle->bond, in_port,
7249 flow->dl_dst, &ctx->tags)) {
7254 xlate_report(ctx, "bonding refused admissibility, dropping");
7257 case BV_DROP_IF_MOVED:
7258 mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL);
7259 if (mac && mac->port.p != in_bundle &&
7260 (!is_gratuitous_arp(flow)
7261 || mac_entry_is_grat_arp_locked(mac))) {
7262 xlate_report(ctx, "SLB bond thinks this packet looped back, "
7274 xlate_normal(struct action_xlate_ctx *ctx)
7276 struct ofport_dpif *in_port;
7277 struct ofbundle *in_bundle;
7278 struct mac_entry *mac;
7282 ctx->has_normal = true;
7284 in_bundle = lookup_input_bundle(ctx->ofproto, ctx->flow.in_port,
7285 ctx->packet != NULL, &in_port);
7287 xlate_report(ctx, "no input bundle, dropping");
7291 /* Drop malformed frames. */
7292 if (ctx->flow.dl_type == htons(ETH_TYPE_VLAN) &&
7293 !(ctx->flow.vlan_tci & htons(VLAN_CFI))) {
7294 if (ctx->packet != NULL) {
7295 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7296 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
7297 "VLAN tag received on port %s",
7298 ctx->ofproto->up.name, in_bundle->name);
7300 xlate_report(ctx, "partial VLAN tag, dropping");
7304 /* Drop frames on bundles reserved for mirroring. */
7305 if (in_bundle->mirror_out) {
7306 if (ctx->packet != NULL) {
7307 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7308 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
7309 "%s, which is reserved exclusively for mirroring",
7310 ctx->ofproto->up.name, in_bundle->name);
7312 xlate_report(ctx, "input port is mirror output port, dropping");
7317 vid = vlan_tci_to_vid(ctx->flow.vlan_tci);
7318 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
7319 xlate_report(ctx, "disallowed VLAN VID for this input port, dropping");
7322 vlan = input_vid_to_vlan(in_bundle, vid);
7324 /* Check other admissibility requirements. */
7325 if (in_port && !is_admissible(ctx, in_port, vlan)) {
7329 /* Learn source MAC. */
7330 if (ctx->may_learn) {
7331 update_learning_table(ctx->ofproto, &ctx->flow, vlan, in_bundle);
7334 /* Determine output bundle. */
7335 mac = mac_learning_lookup(ctx->ofproto->ml, ctx->flow.dl_dst, vlan,
7338 if (mac->port.p != in_bundle) {
7339 xlate_report(ctx, "forwarding to learned port");
7340 output_normal(ctx, mac->port.p, vlan);
7342 xlate_report(ctx, "learned port is input port, dropping");
7345 struct ofbundle *bundle;
7347 xlate_report(ctx, "no learned MAC for destination, flooding");
7348 HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
7349 if (bundle != in_bundle
7350 && ofbundle_includes_vlan(bundle, vlan)
7351 && bundle->floodable
7352 && !bundle->mirror_out) {
7353 output_normal(ctx, bundle, vlan);
7356 ctx->nf_output_iface = NF_OUT_FLOOD;
7360 /* Optimized flow revalidation.
7362 * It's a difficult problem, in general, to tell which facets need to have
7363 * their actions recalculated whenever the OpenFlow flow table changes. We
7364 * don't try to solve that general problem: for most kinds of OpenFlow flow
7365 * table changes, we recalculate the actions for every facet. This is
7366 * relatively expensive, but it's good enough if the OpenFlow flow table
7367 * doesn't change very often.
7369 * However, we can expect one particular kind of OpenFlow flow table change to
7370 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
7371 * of CPU on revalidating every facet whenever MAC learning modifies the flow
7372 * table, we add a special case that applies to flow tables in which every rule
7373 * has the same form (that is, the same wildcards), except that the table is
7374 * also allowed to have a single "catch-all" flow that matches all packets. We
7375 * optimize this case by tagging all of the facets that resubmit into the table
7376 * and invalidating the same tag whenever a flow changes in that table. The
7377 * end result is that we revalidate just the facets that need it (and sometimes
7378 * a few more, but not all of the facets or even all of the facets that
7379 * resubmit to the table modified by MAC learning). */
7381 /* Calculates the tag to use for 'flow' and mask 'mask' when it is inserted
7382 * into an OpenFlow table with the given 'basis'. */
7384 rule_calculate_tag(const struct flow *flow, const struct minimask *mask,
7387 if (minimask_is_catchall(mask)) {
7390 uint32_t hash = flow_hash_in_minimask(flow, mask, secret);
7391 return tag_create_deterministic(hash);
7395 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
7396 * taggability of that table.
7398 * This function must be called after *each* change to a flow table. If you
7399 * skip calling it on some changes then the pointer comparisons at the end can
7400 * be invalid if you get unlucky. For example, if a flow removal causes a
7401 * cls_table to be destroyed and then a flow insertion causes a cls_table with
7402 * different wildcards to be created with the same address, then this function
7403 * will incorrectly skip revalidation. */
7405 table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
7407 struct table_dpif *table = &ofproto->tables[table_id];
7408 const struct oftable *oftable = &ofproto->up.tables[table_id];
7409 struct cls_table *catchall, *other;
7410 struct cls_table *t;
7412 catchall = other = NULL;
7414 switch (hmap_count(&oftable->cls.tables)) {
7416 /* We could tag this OpenFlow table but it would make the logic a
7417 * little harder and it's a corner case that doesn't seem worth it
7423 HMAP_FOR_EACH (t, hmap_node, &oftable->cls.tables) {
7424 if (cls_table_is_catchall(t)) {
7426 } else if (!other) {
7429 /* Indicate that we can't tag this by setting both tables to
7430 * NULL. (We know that 'catchall' is already NULL.) */
7437 /* Can't tag this table. */
7441 if (table->catchall_table != catchall || table->other_table != other) {
7442 table->catchall_table = catchall;
7443 table->other_table = other;
7444 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
7448 /* Given 'rule' that has changed in some way (either it is a rule being
7449 * inserted, a rule being deleted, or a rule whose actions are being
7450 * modified), marks facets for revalidation to ensure that packets will be
7451 * forwarded correctly according to the new state of the flow table.
7453 * This function must be called after *each* change to a flow table. See
7454 * the comment on table_update_taggable() for more information. */
7456 rule_invalidate(const struct rule_dpif *rule)
7458 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
7460 table_update_taggable(ofproto, rule->up.table_id);
7462 if (!ofproto->backer->need_revalidate) {
7463 struct table_dpif *table = &ofproto->tables[rule->up.table_id];
7465 if (table->other_table && rule->tag) {
7466 tag_set_add(&ofproto->backer->revalidate_set, rule->tag);
7468 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
7474 set_frag_handling(struct ofproto *ofproto_,
7475 enum ofp_config_flags frag_handling)
7477 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7478 if (frag_handling != OFPC_FRAG_REASM) {
7479 ofproto->backer->need_revalidate = REV_RECONFIGURE;
7487 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
7488 const struct flow *flow,
7489 const struct ofpact *ofpacts, size_t ofpacts_len)
7491 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7492 struct initial_vals initial_vals;
7493 struct odputil_keybuf keybuf;
7494 struct dpif_flow_stats stats;
7498 struct action_xlate_ctx ctx;
7499 uint64_t odp_actions_stub[1024 / 8];
7500 struct ofpbuf odp_actions;
7502 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
7503 odp_flow_key_from_flow(&key, flow,
7504 ofp_port_to_odp_port(ofproto, flow->in_port));
7506 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
7508 initial_vals.vlan_tci = flow->vlan_tci;
7509 initial_vals.tunnel_ip_tos = 0;
7510 action_xlate_ctx_init(&ctx, ofproto, flow, &initial_vals, NULL,
7511 packet_get_tcp_flags(packet, flow), packet);
7512 ctx.resubmit_stats = &stats;
7514 ofpbuf_use_stub(&odp_actions,
7515 odp_actions_stub, sizeof odp_actions_stub);
7516 xlate_actions(&ctx, ofpacts, ofpacts_len, &odp_actions);
7517 dpif_execute(ofproto->backer->dpif, key.data, key.size,
7518 odp_actions.data, odp_actions.size, packet);
7519 ofpbuf_uninit(&odp_actions);
7527 set_netflow(struct ofproto *ofproto_,
7528 const struct netflow_options *netflow_options)
7530 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7532 if (netflow_options) {
7533 if (!ofproto->netflow) {
7534 ofproto->netflow = netflow_create();
7536 return netflow_set_options(ofproto->netflow, netflow_options);
7538 netflow_destroy(ofproto->netflow);
7539 ofproto->netflow = NULL;
7545 get_netflow_ids(const struct ofproto *ofproto_,
7546 uint8_t *engine_type, uint8_t *engine_id)
7548 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7550 dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
7554 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
7556 if (!facet_is_controller_flow(facet) &&
7557 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
7558 struct subfacet *subfacet;
7559 struct ofexpired expired;
7561 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
7562 if (subfacet->path == SF_FAST_PATH) {
7563 struct dpif_flow_stats stats;
7565 subfacet_reinstall(subfacet, &stats);
7566 subfacet_update_stats(subfacet, &stats);
7570 expired.flow = facet->flow;
7571 expired.packet_count = facet->packet_count;
7572 expired.byte_count = facet->byte_count;
7573 expired.used = facet->used;
7574 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
7579 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
7581 struct facet *facet;
7583 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
7584 send_active_timeout(ofproto, facet);
7588 static struct ofproto_dpif *
7589 ofproto_dpif_lookup(const char *name)
7591 struct ofproto_dpif *ofproto;
7593 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
7594 hash_string(name, 0), &all_ofproto_dpifs) {
7595 if (!strcmp(ofproto->up.name, name)) {
7603 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
7604 const char *argv[], void *aux OVS_UNUSED)
7606 struct ofproto_dpif *ofproto;
7609 ofproto = ofproto_dpif_lookup(argv[1]);
7611 unixctl_command_reply_error(conn, "no such bridge");
7614 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
7616 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
7617 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
7621 unixctl_command_reply(conn, "table successfully flushed");
7625 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
7626 const char *argv[], void *aux OVS_UNUSED)
7628 struct ds ds = DS_EMPTY_INITIALIZER;
7629 const struct ofproto_dpif *ofproto;
7630 const struct mac_entry *e;
7632 ofproto = ofproto_dpif_lookup(argv[1]);
7634 unixctl_command_reply_error(conn, "no such bridge");
7638 ds_put_cstr(&ds, " port VLAN MAC Age\n");
7639 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
7640 struct ofbundle *bundle = e->port.p;
7641 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
7642 ofbundle_get_a_port(bundle)->odp_port,
7643 e->vlan, ETH_ADDR_ARGS(e->mac),
7644 mac_entry_age(ofproto->ml, e));
7646 unixctl_command_reply(conn, ds_cstr(&ds));
7651 struct action_xlate_ctx ctx;
7657 trace_format_rule(struct ds *result, uint8_t table_id, int level,
7658 const struct rule_dpif *rule)
7660 ds_put_char_multiple(result, '\t', level);
7662 ds_put_cstr(result, "No match\n");
7666 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
7667 table_id, ntohll(rule->up.flow_cookie));
7668 cls_rule_format(&rule->up.cr, result);
7669 ds_put_char(result, '\n');
7671 ds_put_char_multiple(result, '\t', level);
7672 ds_put_cstr(result, "OpenFlow ");
7673 ofpacts_format(rule->up.ofpacts, rule->up.ofpacts_len, result);
7674 ds_put_char(result, '\n');
7678 trace_format_flow(struct ds *result, int level, const char *title,
7679 struct trace_ctx *trace)
7681 ds_put_char_multiple(result, '\t', level);
7682 ds_put_format(result, "%s: ", title);
7683 if (flow_equal(&trace->ctx.flow, &trace->flow)) {
7684 ds_put_cstr(result, "unchanged");
7686 flow_format(result, &trace->ctx.flow);
7687 trace->flow = trace->ctx.flow;
7689 ds_put_char(result, '\n');
7693 trace_format_regs(struct ds *result, int level, const char *title,
7694 struct trace_ctx *trace)
7698 ds_put_char_multiple(result, '\t', level);
7699 ds_put_format(result, "%s:", title);
7700 for (i = 0; i < FLOW_N_REGS; i++) {
7701 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
7703 ds_put_char(result, '\n');
7707 trace_format_odp(struct ds *result, int level, const char *title,
7708 struct trace_ctx *trace)
7710 struct ofpbuf *odp_actions = trace->ctx.odp_actions;
7712 ds_put_char_multiple(result, '\t', level);
7713 ds_put_format(result, "%s: ", title);
7714 format_odp_actions(result, odp_actions->data, odp_actions->size);
7715 ds_put_char(result, '\n');
7719 trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
7721 struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx);
7722 struct ds *result = trace->result;
7724 ds_put_char(result, '\n');
7725 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
7726 trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace);
7727 trace_format_odp(result, ctx->recurse + 1, "Resubmitted odp", trace);
7728 trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule);
7732 trace_report(struct action_xlate_ctx *ctx, const char *s)
7734 struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx);
7735 struct ds *result = trace->result;
7737 ds_put_char_multiple(result, '\t', ctx->recurse);
7738 ds_put_cstr(result, s);
7739 ds_put_char(result, '\n');
7743 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
7744 void *aux OVS_UNUSED)
7746 const char *dpname = argv[1];
7747 struct ofproto_dpif *ofproto;
7748 struct ofpbuf odp_key;
7749 struct ofpbuf *packet;
7750 struct initial_vals initial_vals;
7756 ofpbuf_init(&odp_key, 0);
7759 ofproto = ofproto_dpif_lookup(dpname);
7761 unixctl_command_reply_error(conn, "Unknown ofproto (use ofproto/list "
7765 if (argc == 3 || (argc == 4 && !strcmp(argv[3], "-generate"))) {
7766 /* ofproto/trace dpname flow [-generate] */
7767 const char *flow_s = argv[2];
7768 const char *generate_s = argv[3];
7770 /* Allow 'flow_s' to be either a datapath flow or an OpenFlow-like
7771 * flow. We guess which type it is based on whether 'flow_s' contains
7772 * an '(', since a datapath flow always contains '(') but an
7773 * OpenFlow-like flow should not (in fact it's allowed but I believe
7774 * that's not documented anywhere).
7776 * An alternative would be to try to parse 'flow_s' both ways, but then
7777 * it would be tricky giving a sensible error message. After all, do
7778 * you just say "syntax error" or do you present both error messages?
7779 * Both choices seem lousy. */
7780 if (strchr(flow_s, '(')) {
7783 /* Convert string to datapath key. */
7784 ofpbuf_init(&odp_key, 0);
7785 error = odp_flow_key_from_string(flow_s, NULL, &odp_key);
7787 unixctl_command_reply_error(conn, "Bad flow syntax");
7791 /* The user might have specified the wrong ofproto but within the
7792 * same backer. That's OK, ofproto_receive() can find the right
7794 if (ofproto_receive(ofproto->backer, NULL, odp_key.data,
7795 odp_key.size, &flow, NULL, &ofproto, NULL,
7797 unixctl_command_reply_error(conn, "Invalid flow");
7800 ds_put_format(&result, "Bridge: %s\n", ofproto->up.name);
7804 error_s = parse_ofp_exact_flow(&flow, argv[2]);
7806 unixctl_command_reply_error(conn, error_s);
7811 initial_vals.vlan_tci = flow.vlan_tci;
7812 initial_vals.tunnel_ip_tos = flow.tunnel.ip_tos;
7815 /* Generate a packet, if requested. */
7817 packet = ofpbuf_new(0);
7818 flow_compose(packet, &flow);
7820 } else if (argc == 7) {
7821 /* ofproto/trace dpname priority tun_id in_port mark packet */
7822 const char *priority_s = argv[2];
7823 const char *tun_id_s = argv[3];
7824 const char *in_port_s = argv[4];
7825 const char *mark_s = argv[5];
7826 const char *packet_s = argv[6];
7827 uint32_t in_port = atoi(in_port_s);
7828 ovs_be64 tun_id = htonll(strtoull(tun_id_s, NULL, 0));
7829 uint32_t priority = atoi(priority_s);
7830 uint32_t mark = atoi(mark_s);
7833 msg = eth_from_hex(packet_s, &packet);
7835 unixctl_command_reply_error(conn, msg);
7839 ds_put_cstr(&result, "Packet: ");
7840 s = ofp_packet_to_string(packet->data, packet->size);
7841 ds_put_cstr(&result, s);
7844 flow_extract(packet, priority, mark, NULL, in_port, &flow);
7845 flow.tunnel.tun_id = tun_id;
7846 initial_vals.vlan_tci = flow.vlan_tci;
7847 initial_vals.tunnel_ip_tos = flow.tunnel.ip_tos;
7849 unixctl_command_reply_error(conn, "Bad command syntax");
7853 ofproto_trace(ofproto, &flow, packet, &initial_vals, &result);
7854 unixctl_command_reply(conn, ds_cstr(&result));
7857 ds_destroy(&result);
7858 ofpbuf_delete(packet);
7859 ofpbuf_uninit(&odp_key);
7863 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
7864 const struct ofpbuf *packet,
7865 const struct initial_vals *initial_vals, struct ds *ds)
7867 struct rule_dpif *rule;
7869 ds_put_cstr(ds, "Flow: ");
7870 flow_format(ds, flow);
7871 ds_put_char(ds, '\n');
7873 rule = rule_dpif_lookup(ofproto, flow);
7875 trace_format_rule(ds, 0, 0, rule);
7876 if (rule == ofproto->miss_rule) {
7877 ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
7878 } else if (rule == ofproto->no_packet_in_rule) {
7879 ds_put_cstr(ds, "\nNo match, packets dropped because "
7880 "OFPPC_NO_PACKET_IN is set on in_port.\n");
7884 uint64_t odp_actions_stub[1024 / 8];
7885 struct ofpbuf odp_actions;
7887 struct trace_ctx trace;
7890 tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
7893 ofpbuf_use_stub(&odp_actions,
7894 odp_actions_stub, sizeof odp_actions_stub);
7895 action_xlate_ctx_init(&trace.ctx, ofproto, flow, initial_vals,
7896 rule, tcp_flags, packet);
7897 trace.ctx.resubmit_hook = trace_resubmit;
7898 trace.ctx.report_hook = trace_report;
7899 xlate_actions(&trace.ctx, rule->up.ofpacts, rule->up.ofpacts_len,
7902 ds_put_char(ds, '\n');
7903 trace_format_flow(ds, 0, "Final flow", &trace);
7904 ds_put_cstr(ds, "Datapath actions: ");
7905 format_odp_actions(ds, odp_actions.data, odp_actions.size);
7906 ofpbuf_uninit(&odp_actions);
7908 if (trace.ctx.slow) {
7909 enum slow_path_reason slow;
7911 ds_put_cstr(ds, "\nThis flow is handled by the userspace "
7912 "slow path because it:");
7913 for (slow = trace.ctx.slow; slow; ) {
7914 enum slow_path_reason bit = rightmost_1bit(slow);
7918 ds_put_cstr(ds, "\n\t- Consists of CFM packets.");
7921 ds_put_cstr(ds, "\n\t- Consists of LACP packets.");
7924 ds_put_cstr(ds, "\n\t- Consists of STP packets.");
7927 ds_put_cstr(ds, "\n\t- Needs in-band special case "
7930 ds_put_cstr(ds, "\n\t (The datapath actions are "
7931 "incomplete--for complete actions, "
7932 "please supply a packet.)");
7935 case SLOW_CONTROLLER:
7936 ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages "
7937 "to the OpenFlow controller.");
7940 ds_put_cstr(ds, "\n\t- Needs more specific matching "
7941 "than the datapath supports.");
7948 if (slow & ~SLOW_MATCH) {
7949 ds_put_cstr(ds, "\nThe datapath actions above do not reflect "
7950 "the special slow-path processing.");
7957 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
7958 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
7961 unixctl_command_reply(conn, NULL);
7965 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
7966 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
7969 unixctl_command_reply(conn, NULL);
7972 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
7973 * 'reply' describing the results. */
7975 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
7977 struct facet *facet;
7981 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
7982 if (!facet_check_consistency(facet)) {
7987 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
7991 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
7992 ofproto->up.name, errors);
7994 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
7999 ofproto_dpif_self_check(struct unixctl_conn *conn,
8000 int argc, const char *argv[], void *aux OVS_UNUSED)
8002 struct ds reply = DS_EMPTY_INITIALIZER;
8003 struct ofproto_dpif *ofproto;
8006 ofproto = ofproto_dpif_lookup(argv[1]);
8008 unixctl_command_reply_error(conn, "Unknown ofproto (use "
8009 "ofproto/list for help)");
8012 ofproto_dpif_self_check__(ofproto, &reply);
8014 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
8015 ofproto_dpif_self_check__(ofproto, &reply);
8019 unixctl_command_reply(conn, ds_cstr(&reply));
8023 /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list
8024 * of the 'ofproto_shash' nodes. It is the responsibility of the caller
8025 * to destroy 'ofproto_shash' and free the returned value. */
8026 static const struct shash_node **
8027 get_ofprotos(struct shash *ofproto_shash)
8029 const struct ofproto_dpif *ofproto;
8031 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
8032 char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
8033 shash_add_nocopy(ofproto_shash, name, ofproto);
8036 return shash_sort(ofproto_shash);
8040 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
8041 const char *argv[] OVS_UNUSED,
8042 void *aux OVS_UNUSED)
8044 struct ds ds = DS_EMPTY_INITIALIZER;
8045 struct shash ofproto_shash;
8046 const struct shash_node **sorted_ofprotos;
8049 shash_init(&ofproto_shash);
8050 sorted_ofprotos = get_ofprotos(&ofproto_shash);
8051 for (i = 0; i < shash_count(&ofproto_shash); i++) {
8052 const struct shash_node *node = sorted_ofprotos[i];
8053 ds_put_format(&ds, "%s\n", node->name);
8056 shash_destroy(&ofproto_shash);
8057 free(sorted_ofprotos);
8059 unixctl_command_reply(conn, ds_cstr(&ds));
8064 show_dp_format(const struct ofproto_dpif *ofproto, struct ds *ds)
8066 const struct shash_node **ports;
8068 struct avg_subfacet_rates lifetime;
8069 unsigned long long int minutes;
8070 const int min_ms = 60 * 1000; /* milliseconds in one minute. */
8072 minutes = (time_msec() - ofproto->created) / min_ms;
8075 lifetime.add_rate = (double)ofproto->total_subfacet_add_count
8077 lifetime.del_rate = (double)ofproto->total_subfacet_del_count
8080 lifetime.add_rate = 0.0;
8081 lifetime.del_rate = 0.0;
8084 ds_put_format(ds, "%s (%s):\n", ofproto->up.name,
8085 dpif_name(ofproto->backer->dpif));
8087 "\tlookups: hit:%"PRIu64" missed:%"PRIu64"\n",
8088 ofproto->n_hit, ofproto->n_missed);
8089 ds_put_format(ds, "\tflows: cur: %zu, avg: %5.3f, max: %d,"
8090 " life span: %llu(ms)\n",
8091 hmap_count(&ofproto->subfacets),
8092 avg_subfacet_count(ofproto),
8093 ofproto->max_n_subfacet,
8094 avg_subfacet_life_span(ofproto));
8095 if (minutes >= 60) {
8096 show_dp_rates(ds, "\t\thourly avg:", &ofproto->hourly);
8098 if (minutes >= 60 * 24) {
8099 show_dp_rates(ds, "\t\tdaily avg:", &ofproto->daily);
8101 show_dp_rates(ds, "\t\toverall avg:", &lifetime);
8103 ports = shash_sort(&ofproto->up.port_by_name);
8104 for (i = 0; i < shash_count(&ofproto->up.port_by_name); i++) {
8105 const struct shash_node *node = ports[i];
8106 struct ofport *ofport = node->data;
8107 const char *name = netdev_get_name(ofport->netdev);
8108 const char *type = netdev_get_type(ofport->netdev);
8111 ds_put_format(ds, "\t%s %u/", name, ofport->ofp_port);
8113 odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
8114 if (odp_port != OVSP_NONE) {
8115 ds_put_format(ds, "%"PRIu32":", odp_port);
8117 ds_put_cstr(ds, "none:");
8120 if (strcmp(type, "system")) {
8121 struct netdev *netdev;
8124 ds_put_format(ds, " (%s", type);
8126 error = netdev_open(name, type, &netdev);
8131 error = netdev_get_config(netdev, &config);
8133 const struct smap_node **nodes;
8136 nodes = smap_sort(&config);
8137 for (i = 0; i < smap_count(&config); i++) {
8138 const struct smap_node *node = nodes[i];
8139 ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
8140 node->key, node->value);
8144 smap_destroy(&config);
8146 netdev_close(netdev);
8148 ds_put_char(ds, ')');
8150 ds_put_char(ds, '\n');
8156 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc,
8157 const char *argv[], void *aux OVS_UNUSED)
8159 struct ds ds = DS_EMPTY_INITIALIZER;
8160 const struct ofproto_dpif *ofproto;
8164 for (i = 1; i < argc; i++) {
8165 ofproto = ofproto_dpif_lookup(argv[i]);
8167 ds_put_format(&ds, "Unknown bridge %s (use dpif/dump-dps "
8168 "for help)", argv[i]);
8169 unixctl_command_reply_error(conn, ds_cstr(&ds));
8172 show_dp_format(ofproto, &ds);
8175 struct shash ofproto_shash;
8176 const struct shash_node **sorted_ofprotos;
8179 shash_init(&ofproto_shash);
8180 sorted_ofprotos = get_ofprotos(&ofproto_shash);
8181 for (i = 0; i < shash_count(&ofproto_shash); i++) {
8182 const struct shash_node *node = sorted_ofprotos[i];
8183 show_dp_format(node->data, &ds);
8186 shash_destroy(&ofproto_shash);
8187 free(sorted_ofprotos);
8190 unixctl_command_reply(conn, ds_cstr(&ds));
8195 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
8196 int argc OVS_UNUSED, const char *argv[],
8197 void *aux OVS_UNUSED)
8199 struct ds ds = DS_EMPTY_INITIALIZER;
8200 const struct ofproto_dpif *ofproto;
8201 struct subfacet *subfacet;
8203 ofproto = ofproto_dpif_lookup(argv[1]);
8205 unixctl_command_reply_error(conn, "no such bridge");
8209 update_stats(ofproto->backer);
8211 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
8212 odp_flow_key_format(subfacet->key, subfacet->key_len, &ds);
8214 ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:",
8215 subfacet->dp_packet_count, subfacet->dp_byte_count);
8216 if (subfacet->used) {
8217 ds_put_format(&ds, "%.3fs",
8218 (time_msec() - subfacet->used) / 1000.0);
8220 ds_put_format(&ds, "never");
8222 if (subfacet->facet->tcp_flags) {
8223 ds_put_cstr(&ds, ", flags:");
8224 packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags);
8227 ds_put_cstr(&ds, ", actions:");
8228 if (subfacet->slow) {
8229 uint64_t slow_path_stub[128 / 8];
8230 const struct nlattr *actions;
8233 compose_slow_path(ofproto, &subfacet->facet->flow, subfacet->slow,
8234 slow_path_stub, sizeof slow_path_stub,
8235 &actions, &actions_len);
8236 format_odp_actions(&ds, actions, actions_len);
8238 format_odp_actions(&ds, subfacet->actions, subfacet->actions_len);
8240 ds_put_char(&ds, '\n');
8243 unixctl_command_reply(conn, ds_cstr(&ds));
8248 ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn,
8249 int argc OVS_UNUSED, const char *argv[],
8250 void *aux OVS_UNUSED)
8252 struct ds ds = DS_EMPTY_INITIALIZER;
8253 struct ofproto_dpif *ofproto;
8255 ofproto = ofproto_dpif_lookup(argv[1]);
8257 unixctl_command_reply_error(conn, "no such bridge");
8261 flush(&ofproto->up);
8263 unixctl_command_reply(conn, ds_cstr(&ds));
8268 ofproto_dpif_unixctl_init(void)
8270 static bool registered;
8276 unixctl_command_register(
8278 "bridge {priority tun_id in_port mark packet | odp_flow [-generate]}",
8279 2, 6, ofproto_unixctl_trace, NULL);
8280 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
8281 ofproto_unixctl_fdb_flush, NULL);
8282 unixctl_command_register("fdb/show", "bridge", 1, 1,
8283 ofproto_unixctl_fdb_show, NULL);
8284 unixctl_command_register("ofproto/clog", "", 0, 0,
8285 ofproto_dpif_clog, NULL);
8286 unixctl_command_register("ofproto/unclog", "", 0, 0,
8287 ofproto_dpif_unclog, NULL);
8288 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
8289 ofproto_dpif_self_check, NULL);
8290 unixctl_command_register("dpif/dump-dps", "", 0, 0,
8291 ofproto_unixctl_dpif_dump_dps, NULL);
8292 unixctl_command_register("dpif/show", "[bridge]", 0, INT_MAX,
8293 ofproto_unixctl_dpif_show, NULL);
8294 unixctl_command_register("dpif/dump-flows", "bridge", 1, 1,
8295 ofproto_unixctl_dpif_dump_flows, NULL);
8296 unixctl_command_register("dpif/del-flows", "bridge", 1, 1,
8297 ofproto_unixctl_dpif_del_flows, NULL);
8300 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
8302 * This is deprecated. It is only for compatibility with broken device drivers
8303 * in old versions of Linux that do not properly support VLANs when VLAN
8304 * devices are not used. When broken device drivers are no longer in
8305 * widespread use, we will delete these interfaces. */
8308 set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid)
8310 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
8311 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
8313 if (realdev_ofp_port == ofport->realdev_ofp_port
8314 && vid == ofport->vlandev_vid) {
8318 ofproto->backer->need_revalidate = REV_RECONFIGURE;
8320 if (ofport->realdev_ofp_port) {
8323 if (realdev_ofp_port && ofport->bundle) {
8324 /* vlandevs are enslaved to their realdevs, so they are not allowed to
8325 * themselves be part of a bundle. */
8326 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
8329 ofport->realdev_ofp_port = realdev_ofp_port;
8330 ofport->vlandev_vid = vid;
8332 if (realdev_ofp_port) {
8333 vsp_add(ofport, realdev_ofp_port, vid);
8340 hash_realdev_vid(uint16_t realdev_ofp_port, int vid)
8342 return hash_2words(realdev_ofp_port, vid);
8345 /* Returns the ODP port number of the Linux VLAN device that corresponds to
8346 * 'vlan_tci' on the network device with port number 'realdev_odp_port' in
8347 * 'ofproto'. For example, given 'realdev_odp_port' of eth0 and 'vlan_tci' 9,
8348 * it would return the port number of eth0.9.
8350 * Unless VLAN splinters are enabled for port 'realdev_odp_port', this
8351 * function just returns its 'realdev_odp_port' argument. */
8353 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
8354 uint32_t realdev_odp_port, ovs_be16 vlan_tci)
8356 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
8357 uint16_t realdev_ofp_port;
8358 int vid = vlan_tci_to_vid(vlan_tci);
8359 const struct vlan_splinter *vsp;
8361 realdev_ofp_port = odp_port_to_ofp_port(ofproto, realdev_odp_port);
8362 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
8363 hash_realdev_vid(realdev_ofp_port, vid),
8364 &ofproto->realdev_vid_map) {
8365 if (vsp->realdev_ofp_port == realdev_ofp_port
8366 && vsp->vid == vid) {
8367 return ofp_port_to_odp_port(ofproto, vsp->vlandev_ofp_port);
8371 return realdev_odp_port;
8374 static struct vlan_splinter *
8375 vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port)
8377 struct vlan_splinter *vsp;
8379 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, hash_int(vlandev_ofp_port, 0),
8380 &ofproto->vlandev_map) {
8381 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
8389 /* Returns the OpenFlow port number of the "real" device underlying the Linux
8390 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
8391 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
8392 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
8393 * eth0 and store 9 in '*vid'.
8395 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
8396 * VLAN device. Unless VLAN splinters are enabled, this is what this function
8399 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
8400 uint16_t vlandev_ofp_port, int *vid)
8402 if (!hmap_is_empty(&ofproto->vlandev_map)) {
8403 const struct vlan_splinter *vsp;
8405 vsp = vlandev_find(ofproto, vlandev_ofp_port);
8410 return vsp->realdev_ofp_port;
8416 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
8417 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
8418 * 'flow->in_port' to the "real" device backing the VLAN device, sets
8419 * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is
8420 * always the case unless VLAN splinters are enabled), returns false without
8421 * making any changes. */
8423 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
8428 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port, &vid);
8433 /* Cause the flow to be processed as if it came in on the real device with
8434 * the VLAN device's VLAN ID. */
8435 flow->in_port = realdev;
8436 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
8441 vsp_remove(struct ofport_dpif *port)
8443 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
8444 struct vlan_splinter *vsp;
8446 vsp = vlandev_find(ofproto, port->up.ofp_port);
8448 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
8449 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
8452 port->realdev_ofp_port = 0;
8454 VLOG_ERR("missing vlan device record");
8459 vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid)
8461 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
8463 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
8464 && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid))
8465 == realdev_ofp_port)) {
8466 struct vlan_splinter *vsp;
8468 vsp = xmalloc(sizeof *vsp);
8469 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
8470 hash_int(port->up.ofp_port, 0));
8471 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
8472 hash_realdev_vid(realdev_ofp_port, vid));
8473 vsp->realdev_ofp_port = realdev_ofp_port;
8474 vsp->vlandev_ofp_port = port->up.ofp_port;
8477 port->realdev_ofp_port = realdev_ofp_port;
8479 VLOG_ERR("duplicate vlan device record");
8484 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
8486 const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
8487 return ofport ? ofport->odp_port : OVSP_NONE;
8490 static struct ofport_dpif *
8491 odp_port_to_ofport(const struct dpif_backer *backer, uint32_t odp_port)
8493 struct ofport_dpif *port;
8495 HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node,
8496 hash_int(odp_port, 0),
8497 &backer->odp_to_ofport_map) {
8498 if (port->odp_port == odp_port) {
8507 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
8509 struct ofport_dpif *port;
8511 port = odp_port_to_ofport(ofproto->backer, odp_port);
8512 if (port && &ofproto->up == port->up.ofproto) {
8513 return port->up.ofp_port;
8518 static unsigned long long int
8519 avg_subfacet_life_span(const struct ofproto_dpif *ofproto)
8521 unsigned long long int dc;
8522 unsigned long long int avg;
8524 dc = ofproto->total_subfacet_del_count + ofproto->subfacet_del_count;
8525 avg = dc ? ofproto->total_subfacet_life_span / dc : 0;
8531 avg_subfacet_count(const struct ofproto_dpif *ofproto)
8535 if (ofproto->n_update_stats) {
8536 avg_c = (double)ofproto->total_subfacet_count
8537 / ofproto->n_update_stats;
8544 show_dp_rates(struct ds *ds, const char *heading,
8545 const struct avg_subfacet_rates *rates)
8547 ds_put_format(ds, "%s add rate: %5.3f/min, del rate: %5.3f/min\n",
8548 heading, rates->add_rate, rates->del_rate);
8552 update_max_subfacet_count(struct ofproto_dpif *ofproto)
8554 ofproto->max_n_subfacet = MAX(ofproto->max_n_subfacet,
8555 hmap_count(&ofproto->subfacets));
8558 /* Compute exponentially weighted moving average, adding 'new' as the newest,
8559 * most heavily weighted element. 'base' designates the rate of decay: after
8560 * 'base' further updates, 'new''s weight in the EWMA decays to about 1/e
8563 exp_mavg(double *avg, int base, double new)
8565 *avg = (*avg * (base - 1) + new) / base;
8569 update_moving_averages(struct ofproto_dpif *ofproto)
8571 const int min_ms = 60 * 1000; /* milliseconds in one minute. */
8573 /* Update hourly averages on the minute boundaries. */
8574 if (time_msec() - ofproto->last_minute >= min_ms) {
8575 exp_mavg(&ofproto->hourly.add_rate, 60, ofproto->subfacet_add_count);
8576 exp_mavg(&ofproto->hourly.del_rate, 60, ofproto->subfacet_del_count);
8578 /* Update daily averages on the hour boundaries. */
8579 if ((ofproto->last_minute - ofproto->created) / min_ms % 60 == 59) {
8580 exp_mavg(&ofproto->daily.add_rate, 24, ofproto->hourly.add_rate);
8581 exp_mavg(&ofproto->daily.del_rate, 24, ofproto->hourly.del_rate);
8584 ofproto->total_subfacet_add_count += ofproto->subfacet_add_count;
8585 ofproto->total_subfacet_del_count += ofproto->subfacet_del_count;
8586 ofproto->subfacet_add_count = 0;
8587 ofproto->subfacet_del_count = 0;
8588 ofproto->last_minute += min_ms;
8593 dpif_stats_update_hit_count(struct ofproto_dpif *ofproto, uint64_t delta)
8595 ofproto->n_hit += delta;
8598 const struct ofproto_class ofproto_dpif_class = {
8633 port_is_lacp_current,
8634 NULL, /* rule_choose_table */
8641 rule_modify_actions,
8650 get_cfm_remote_mpids,
8655 get_stp_port_status,
8662 is_mirror_output_bundle,
8663 forward_bpdu_changed,
8664 set_mac_table_config,