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-ipfix.h"
51 #include "ofproto-dpif-sflow.h"
52 #include "poll-loop.h"
57 #include "unaligned.h"
59 #include "vlan-bitmap.h"
62 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
64 COVERAGE_DEFINE(ofproto_dpif_expired);
65 COVERAGE_DEFINE(ofproto_dpif_xlate);
66 COVERAGE_DEFINE(facet_changed_rule);
67 COVERAGE_DEFINE(facet_revalidate);
68 COVERAGE_DEFINE(facet_unexpected);
69 COVERAGE_DEFINE(facet_suppress);
71 /* Maximum depth of flow table recursion (due to resubmit actions) in a
72 * flow translation. */
73 #define MAX_RESUBMIT_RECURSION 64
75 /* Number of implemented OpenFlow tables. */
76 enum { N_TABLES = 255 };
77 enum { TBL_INTERNAL = N_TABLES - 1 }; /* Used for internal hidden rules. */
78 BUILD_ASSERT_DECL(N_TABLES >= 2 && N_TABLES <= 255);
90 * - Do include packets and bytes from facets that have been deleted or
91 * whose own statistics have been folded into the rule.
93 * - Do include packets and bytes sent "by hand" that were accounted to
94 * the rule without any facet being involved (this is a rare corner
95 * case in rule_execute()).
97 * - Do not include packet or bytes that can be obtained from any facet's
98 * packet_count or byte_count member or that can be obtained from the
99 * datapath by, e.g., dpif_flow_get() for any subfacet.
101 uint64_t packet_count; /* Number of packets received. */
102 uint64_t byte_count; /* Number of bytes received. */
104 tag_type tag; /* Caches rule_calculate_tag() result. */
106 struct list facets; /* List of "struct facet"s. */
109 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
111 return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
114 static struct rule_dpif *rule_dpif_lookup(struct ofproto_dpif *,
115 const struct flow *);
116 static struct rule_dpif *rule_dpif_lookup__(struct ofproto_dpif *,
119 static struct rule_dpif *rule_dpif_miss_rule(struct ofproto_dpif *ofproto,
120 const struct flow *flow);
122 static void rule_get_stats(struct rule *, uint64_t *packets, uint64_t *bytes);
123 static void rule_credit_stats(struct rule_dpif *,
124 const struct dpif_flow_stats *);
125 static tag_type rule_calculate_tag(const struct flow *,
126 const struct minimask *, uint32_t basis);
127 static void rule_invalidate(const struct rule_dpif *);
129 #define MAX_MIRRORS 32
130 typedef uint32_t mirror_mask_t;
131 #define MIRROR_MASK_C(X) UINT32_C(X)
132 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
134 struct ofproto_dpif *ofproto; /* Owning ofproto. */
135 size_t idx; /* In ofproto's "mirrors" array. */
136 void *aux; /* Key supplied by ofproto's client. */
137 char *name; /* Identifier for log messages. */
139 /* Selection criteria. */
140 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
141 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
142 unsigned long *vlans; /* Bitmap of chosen VLANs, NULL selects all. */
144 /* Output (exactly one of out == NULL and out_vlan == -1 is true). */
145 struct ofbundle *out; /* Output port or NULL. */
146 int out_vlan; /* Output VLAN or -1. */
147 mirror_mask_t dup_mirrors; /* Bitmap of mirrors with the same output. */
150 int64_t packet_count; /* Number of packets sent. */
151 int64_t byte_count; /* Number of bytes sent. */
154 static void mirror_destroy(struct ofmirror *);
155 static void update_mirror_stats(struct ofproto_dpif *ofproto,
156 mirror_mask_t mirrors,
157 uint64_t packets, uint64_t bytes);
160 struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
161 struct ofproto_dpif *ofproto; /* Owning ofproto. */
162 void *aux; /* Key supplied by ofproto's client. */
163 char *name; /* Identifier for log messages. */
166 struct list ports; /* Contains "struct ofport"s. */
167 enum port_vlan_mode vlan_mode; /* VLAN mode */
168 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
169 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
170 * NULL if all VLANs are trunked. */
171 struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */
172 struct bond *bond; /* Nonnull iff more than one port. */
173 bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */
176 bool floodable; /* True if no port has OFPUTIL_PC_NO_FLOOD set. */
178 /* Port mirroring info. */
179 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
180 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
181 mirror_mask_t mirror_out; /* Mirrors that output to this bundle. */
184 static void bundle_remove(struct ofport *);
185 static void bundle_update(struct ofbundle *);
186 static void bundle_destroy(struct ofbundle *);
187 static void bundle_del_port(struct ofport_dpif *);
188 static void bundle_run(struct ofbundle *);
189 static void bundle_wait(struct ofbundle *);
190 static struct ofbundle *lookup_input_bundle(const struct ofproto_dpif *,
191 uint16_t in_port, bool warn,
192 struct ofport_dpif **in_ofportp);
194 /* A controller may use OFPP_NONE as the ingress port to indicate that
195 * it did not arrive on a "real" port. 'ofpp_none_bundle' exists for
196 * when an input bundle is needed for validation (e.g., mirroring or
197 * OFPP_NORMAL processing). It is not connected to an 'ofproto' or have
198 * any 'port' structs, so care must be taken when dealing with it. */
199 static struct ofbundle ofpp_none_bundle = {
201 .vlan_mode = PORT_VLAN_TRUNK
204 static void stp_run(struct ofproto_dpif *ofproto);
205 static void stp_wait(struct ofproto_dpif *ofproto);
206 static int set_stp_port(struct ofport *,
207 const struct ofproto_port_stp_settings *);
209 static bool ofbundle_includes_vlan(const struct ofbundle *, uint16_t vlan);
213 /* Initial values of fields of the packet that may be changed during
214 * flow processing and needed later. */
215 struct initial_vals {
216 /* This is the value of vlan_tci in the packet as actually received from
217 * dpif. This is the same as the facet's flow.vlan_tci unless the packet
218 * was received via a VLAN splinter. In that case, this value is 0
219 * (because the packet as actually received from the dpif had no 802.1Q
220 * tag) but the facet's flow.vlan_tci is set to the VLAN that the splinter
223 * This member should be removed when the VLAN splinters feature is no
229 tag_type tags; /* Tags associated with actions. */
230 enum slow_path_reason slow; /* 0 if fast path may be used. */
231 bool has_learn; /* Actions include NXAST_LEARN? */
232 bool has_normal; /* Actions output to OFPP_NORMAL? */
233 bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */
234 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
235 mirror_mask_t mirrors; /* Bitmap of associated mirrors. */
237 uint64_t odp_actions_stub[256 / 8];
238 struct ofpbuf odp_actions;
242 struct ofproto_dpif *ofproto;
244 /* Flow to which the OpenFlow actions apply. xlate_actions() will modify
245 * this flow when actions change header fields. */
248 struct initial_vals initial_vals;
250 /* The packet corresponding to 'flow', or a null pointer if we are
251 * revalidating without a packet to refer to. */
252 const struct ofpbuf *packet;
254 /* Should OFPP_NORMAL update the MAC learning table? Should "learn"
255 * actions update the flow table?
257 * We want to update these tables if we are actually processing a packet,
258 * or if we are accounting for packets that the datapath has processed, but
259 * not if we are just revalidating. */
262 /* The rule initiating translation or NULL. */
263 struct rule_dpif *rule;
265 /* The actions to translate. If 'rule' is not NULL, these may be NULL. */
266 const struct ofpact *ofpacts;
269 /* Union of the set of TCP flags seen so far in this flow. (Used only by
270 * NXAST_FIN_TIMEOUT. Set to zero to avoid updating updating rules'
274 /* If nonnull, flow translation calls this function just before executing a
275 * resubmit or OFPP_TABLE action. In addition, disables logging of traces
276 * when the recursion depth is exceeded.
278 * 'rule' is the rule being submitted into. It will be null if the
279 * resubmit or OFPP_TABLE action didn't find a matching rule.
281 * This is normally null so the client has to set it manually after
282 * calling xlate_in_init(). */
283 void (*resubmit_hook)(struct xlate_ctx *, struct rule_dpif *rule);
285 /* If nonnull, flow translation calls this function to report some
286 * significant decision, e.g. to explain why OFPP_NORMAL translation
287 * dropped a packet. */
288 void (*report_hook)(struct xlate_ctx *, const char *s);
290 /* If nonnull, flow translation credits the specified statistics to each
291 * rule reached through a resubmit or OFPP_TABLE action.
293 * This is normally null so the client has to set it manually after
294 * calling xlate_in_init(). */
295 const struct dpif_flow_stats *resubmit_stats;
298 /* Context used by xlate_actions() and its callees. */
300 struct xlate_in *xin;
301 struct xlate_out *xout;
303 struct ofproto_dpif *ofproto;
305 /* Flow at the last commit. */
306 struct flow base_flow;
308 /* Tunnel IP destination address as received. This is stored separately
309 * as the base_flow.tunnel is cleared on init to reflect the datapath
310 * behavior. Used to make sure not to send tunneled output to ourselves,
311 * which might lead to an infinite loop. This could happen easily
312 * if a tunnel is marked as 'ip_remote=flow', and the flow does not
313 * actually set the tun_dst field. */
314 ovs_be32 orig_tunnel_ip_dst;
316 /* Stack for the push and pop actions. Each stack element is of type
317 * "union mf_subvalue". */
318 union mf_subvalue init_stack[1024 / sizeof(union mf_subvalue)];
321 /* The rule that we are currently translating, or NULL. */
322 struct rule_dpif *rule;
324 int recurse; /* Recursion level, via xlate_table_action. */
325 bool max_resubmit_trigger; /* Recursed too deeply during translation. */
326 uint32_t orig_skb_priority; /* Priority when packet arrived. */
327 uint8_t table_id; /* OpenFlow table ID where flow was found. */
328 uint32_t sflow_n_outputs; /* Number of output ports. */
329 uint32_t sflow_odp_port; /* Output port for composing sFlow action. */
330 uint16_t user_cookie_offset;/* Used for user_action_cookie fixup. */
331 bool exit; /* No further actions should be processed. */
334 static void xlate_in_init(struct xlate_in *, struct ofproto_dpif *,
335 const struct flow *, const struct initial_vals *,
336 struct rule_dpif *, uint8_t tcp_flags,
337 const struct ofpbuf *);
339 static void xlate_out_uninit(struct xlate_out *);
341 static void xlate_actions(struct xlate_in *, struct xlate_out *);
343 static void xlate_actions_for_side_effects(struct xlate_in *);
345 static void xlate_table_action(struct xlate_ctx *, uint16_t in_port,
346 uint8_t table_id, bool may_packet_in);
348 static size_t put_userspace_action(const struct ofproto_dpif *,
349 struct ofpbuf *odp_actions,
351 const union user_action_cookie *,
354 static void compose_slow_path(const struct ofproto_dpif *, const struct flow *,
355 enum slow_path_reason,
356 uint64_t *stub, size_t stub_size,
357 const struct nlattr **actionsp,
358 size_t *actions_lenp);
360 static void xlate_report(struct xlate_ctx *ctx, const char *s);
362 /* A subfacet (see "struct subfacet" below) has three possible installation
365 * - SF_NOT_INSTALLED: Not installed in the datapath. This will only be the
366 * case just after the subfacet is created, just before the subfacet is
367 * destroyed, or if the datapath returns an error when we try to install a
370 * - SF_FAST_PATH: The subfacet's actions are installed in the datapath.
372 * - SF_SLOW_PATH: An action that sends every packet for the subfacet through
373 * ofproto_dpif is installed in the datapath.
376 SF_NOT_INSTALLED, /* No datapath flow for this subfacet. */
377 SF_FAST_PATH, /* Full actions are installed. */
378 SF_SLOW_PATH, /* Send-to-userspace action is installed. */
381 /* A dpif flow and actions associated with a facet.
383 * See also the large comment on struct facet. */
386 struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */
387 struct list list_node; /* In struct facet's 'facets' list. */
388 struct facet *facet; /* Owning facet. */
390 enum odp_key_fitness key_fitness;
394 long long int used; /* Time last used; time created if not used. */
395 long long int created; /* Time created. */
397 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
398 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
400 enum subfacet_path path; /* Installed in datapath? */
403 #define SUBFACET_DESTROY_MAX_BATCH 50
405 static struct subfacet *subfacet_create(struct facet *, struct flow_miss *miss,
407 static struct subfacet *subfacet_find(struct ofproto_dpif *,
408 const struct nlattr *key, size_t key_len,
410 static void subfacet_destroy(struct subfacet *);
411 static void subfacet_destroy__(struct subfacet *);
412 static void subfacet_destroy_batch(struct ofproto_dpif *,
413 struct subfacet **, int n);
414 static void subfacet_reset_dp_stats(struct subfacet *,
415 struct dpif_flow_stats *);
416 static void subfacet_update_stats(struct subfacet *,
417 const struct dpif_flow_stats *);
418 static int subfacet_install(struct subfacet *,
419 const struct ofpbuf *odp_actions,
420 struct dpif_flow_stats *);
421 static void subfacet_uninstall(struct subfacet *);
423 /* An exact-match instantiation of an OpenFlow flow.
425 * A facet associates a "struct flow", which represents the Open vSwitch
426 * userspace idea of an exact-match flow, with one or more subfacets. Each
427 * subfacet tracks the datapath's idea of the exact-match flow equivalent to
428 * the facet. When the kernel module (or other dpif implementation) and Open
429 * vSwitch userspace agree on the definition of a flow key, there is exactly
430 * one subfacet per facet. If the dpif implementation supports more-specific
431 * flow matching than userspace, however, a facet can have more than one
432 * subfacet, each of which corresponds to some distinction in flow that
433 * userspace simply doesn't understand.
435 * Flow expiration works in terms of subfacets, so a facet must have at least
436 * one subfacet or it will never expire, leaking memory. */
439 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
440 struct list list_node; /* In owning rule's 'facets' list. */
441 struct rule_dpif *rule; /* Owning rule. */
444 struct list subfacets;
445 long long int used; /* Time last used; time created if not used. */
452 * - Do include packets and bytes sent "by hand", e.g. with
455 * - Do include packets and bytes that were obtained from the datapath
456 * when a subfacet's statistics were reset (e.g. dpif_flow_put() with
457 * DPIF_FP_ZERO_STATS).
459 * - Do not include packets or bytes that can be obtained from the
460 * datapath for any existing subfacet.
462 uint64_t packet_count; /* Number of packets received. */
463 uint64_t byte_count; /* Number of bytes received. */
465 /* Resubmit statistics. */
466 uint64_t prev_packet_count; /* Number of packets from last stats push. */
467 uint64_t prev_byte_count; /* Number of bytes from last stats push. */
468 long long int prev_used; /* Used time from last stats push. */
471 uint64_t accounted_bytes; /* Bytes processed by facet_account(). */
472 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
473 uint8_t tcp_flags; /* TCP flags seen for this 'rule'. */
475 struct xlate_out xout;
477 /* Initial values of the packet that may be needed later. */
478 struct initial_vals initial_vals;
480 /* Storage for a single subfacet, to reduce malloc() time and space
481 * overhead. (A facet always has at least one subfacet and in the common
482 * case has exactly one subfacet. However, 'one_subfacet' may not
483 * always be valid, since it could have been removed after newer
484 * subfacets were pushed onto the 'subfacets' list.) */
485 struct subfacet one_subfacet;
487 long long int learn_rl; /* Rate limiter for facet_learn(). */
490 static struct facet *facet_create(const struct flow_miss *, uint32_t hash);
491 static void facet_remove(struct facet *);
492 static void facet_free(struct facet *);
494 static struct facet *facet_find(struct ofproto_dpif *,
495 const struct flow *, uint32_t hash);
496 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
497 const struct flow *, uint32_t hash);
498 static bool facet_revalidate(struct facet *);
499 static bool facet_check_consistency(struct facet *);
501 static void facet_flush_stats(struct facet *);
503 static void facet_reset_counters(struct facet *);
504 static void facet_push_stats(struct facet *, bool may_learn);
505 static void facet_learn(struct facet *);
506 static void facet_account(struct facet *);
507 static void push_all_stats(void);
509 static bool facet_is_controller_flow(struct facet *);
512 struct hmap_node odp_port_node; /* In dpif_backer's "odp_to_ofport_map". */
516 struct ofbundle *bundle; /* Bundle that contains this port, if any. */
517 struct list bundle_node; /* In struct ofbundle's "ports" list. */
518 struct cfm *cfm; /* Connectivity Fault Management, if any. */
519 struct bfd *bfd; /* BFD, if any. */
520 tag_type tag; /* Tag associated with this port. */
521 bool may_enable; /* May be enabled in bonds. */
522 long long int carrier_seq; /* Carrier status changes. */
523 struct tnl_port *tnl_port; /* Tunnel handle, or null. */
526 struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */
527 enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */
528 long long int stp_state_entered;
530 struct hmap priorities; /* Map of attached 'priority_to_dscp's. */
532 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
534 * This is deprecated. It is only for compatibility with broken device
535 * drivers in old versions of Linux that do not properly support VLANs when
536 * VLAN devices are not used. When broken device drivers are no longer in
537 * widespread use, we will delete these interfaces. */
538 uint16_t realdev_ofp_port;
542 /* Node in 'ofport_dpif''s 'priorities' map. Used to maintain a map from
543 * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
544 * traffic egressing the 'ofport' with that priority should be marked with. */
545 struct priority_to_dscp {
546 struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'priorities' map. */
547 uint32_t priority; /* Priority of this queue (see struct flow). */
549 uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */
552 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
554 * This is deprecated. It is only for compatibility with broken device drivers
555 * in old versions of Linux that do not properly support VLANs when VLAN
556 * devices are not used. When broken device drivers are no longer in
557 * widespread use, we will delete these interfaces. */
558 struct vlan_splinter {
559 struct hmap_node realdev_vid_node;
560 struct hmap_node vlandev_node;
561 uint16_t realdev_ofp_port;
562 uint16_t vlandev_ofp_port;
566 static uint32_t vsp_realdev_to_vlandev(const struct ofproto_dpif *,
567 uint32_t realdev, ovs_be16 vlan_tci);
568 static bool vsp_adjust_flow(const struct ofproto_dpif *, struct flow *);
569 static void vsp_remove(struct ofport_dpif *);
570 static void vsp_add(struct ofport_dpif *, uint16_t realdev_ofp_port, int vid);
572 static uint32_t ofp_port_to_odp_port(const struct ofproto_dpif *,
574 static uint16_t odp_port_to_ofp_port(const struct ofproto_dpif *,
577 static struct ofport_dpif *
578 ofport_dpif_cast(const struct ofport *ofport)
580 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
583 static void port_run(struct ofport_dpif *);
584 static void port_run_fast(struct ofport_dpif *);
585 static void port_wait(struct ofport_dpif *);
586 static int set_bfd(struct ofport *, const struct smap *);
587 static int set_cfm(struct ofport *, const struct cfm_settings *);
588 static void ofport_clear_priorities(struct ofport_dpif *);
589 static void run_fast_rl(void);
591 struct dpif_completion {
592 struct list list_node;
593 struct ofoperation *op;
596 /* Extra information about a classifier table.
597 * Currently used just for optimized flow revalidation. */
599 /* If either of these is nonnull, then this table has a form that allows
600 * flows to be tagged to avoid revalidating most flows for the most common
601 * kinds of flow table changes. */
602 struct cls_table *catchall_table; /* Table that wildcards all fields. */
603 struct cls_table *other_table; /* Table with any other wildcard set. */
604 uint32_t basis; /* Keeps each table's tags separate. */
607 /* Reasons that we might need to revalidate every facet, and corresponding
610 * A value of 0 means that there is no need to revalidate.
612 * It would be nice to have some cleaner way to integrate with coverage
613 * counters, but with only a few reasons I guess this is good enough for
615 enum revalidate_reason {
616 REV_RECONFIGURE = 1, /* Switch configuration changed. */
617 REV_STP, /* Spanning tree protocol port status change. */
618 REV_PORT_TOGGLED, /* Port enabled or disabled by CFM, LACP, ...*/
619 REV_FLOW_TABLE, /* Flow table changed. */
620 REV_INCONSISTENCY /* Facet self-check failed. */
622 COVERAGE_DEFINE(rev_reconfigure);
623 COVERAGE_DEFINE(rev_stp);
624 COVERAGE_DEFINE(rev_port_toggled);
625 COVERAGE_DEFINE(rev_flow_table);
626 COVERAGE_DEFINE(rev_inconsistency);
628 /* Drop keys are odp flow keys which have drop flows installed in the kernel.
629 * These are datapath flows which have no associated ofproto, if they did we
630 * would use facets. */
632 struct hmap_node hmap_node;
637 /* All datapaths of a given type share a single dpif backer instance. */
642 struct timer next_expiration;
643 struct hmap odp_to_ofport_map; /* ODP port to ofport mapping. */
645 struct simap tnl_backers; /* Set of dpif ports backing tunnels. */
647 /* Facet revalidation flags applying to facets which use this backer. */
648 enum revalidate_reason need_revalidate; /* Revalidate every facet. */
649 struct tag_set revalidate_set; /* Revalidate only matching facets. */
651 struct hmap drop_keys; /* Set of dropped odp keys. */
654 /* All existing ofproto_backer instances, indexed by ofproto->up.type. */
655 static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers);
657 static void drop_key_clear(struct dpif_backer *);
658 static struct ofport_dpif *
659 odp_port_to_ofport(const struct dpif_backer *, uint32_t odp_port);
661 static void dpif_stats_update_hit_count(struct ofproto_dpif *ofproto,
663 struct avg_subfacet_rates {
664 double add_rate; /* Moving average of new flows created per minute. */
665 double del_rate; /* Moving average of flows deleted per minute. */
667 static void show_dp_rates(struct ds *ds, const char *heading,
668 const struct avg_subfacet_rates *rates);
669 static void exp_mavg(double *avg, int base, double new);
671 struct ofproto_dpif {
672 struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
674 struct dpif_backer *backer;
676 /* Special OpenFlow rules. */
677 struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */
678 struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */
681 struct netflow *netflow;
682 struct dpif_sflow *sflow;
683 struct dpif_ipfix *ipfix;
684 struct hmap bundles; /* Contains "struct ofbundle"s. */
685 struct mac_learning *ml;
686 struct ofmirror *mirrors[MAX_MIRRORS];
688 bool has_bonded_bundles;
692 struct hmap subfacets;
693 struct governor *governor;
694 long long int consistency_rl;
697 struct table_dpif tables[N_TABLES];
699 /* Support for debugging async flow mods. */
700 struct list completions;
702 bool has_bundle_action; /* True when the first bundle action appears. */
703 struct netdev_stats stats; /* To account packets generated and consumed in
708 long long int stp_last_tick;
710 /* VLAN splinters. */
711 struct hmap realdev_vid_map; /* (realdev,vid) -> vlandev. */
712 struct hmap vlandev_map; /* vlandev -> (realdev,vid). */
715 struct sset ports; /* Set of standard port names. */
716 struct sset ghost_ports; /* Ports with no datapath port. */
717 struct sset port_poll_set; /* Queued names for port_poll() reply. */
718 int port_poll_errno; /* Last errno for port_poll() reply. */
720 /* Per ofproto's dpif stats. */
724 /* Subfacet statistics.
726 * These keep track of the total number of subfacets added and deleted and
727 * flow life span. They are useful for computing the flow rates stats
728 * exposed via "ovs-appctl dpif/show". The goal is to learn about
729 * traffic patterns in ways that we can use later to improve Open vSwitch
730 * performance in new situations. */
731 long long int created; /* Time when it is created. */
732 unsigned int max_n_subfacet; /* Maximum number of flows */
734 /* The average number of subfacets... */
735 struct avg_subfacet_rates hourly; /* ...over the last hour. */
736 struct avg_subfacet_rates daily; /* ...over the last day. */
737 long long int last_minute; /* Last time 'hourly' was updated. */
739 /* Number of subfacets added or deleted since 'last_minute'. */
740 unsigned int subfacet_add_count;
741 unsigned int subfacet_del_count;
743 /* Number of subfacets added or deleted from 'created' to 'last_minute.' */
744 unsigned long long int total_subfacet_add_count;
745 unsigned long long int total_subfacet_del_count;
747 /* Sum of the number of milliseconds that each subfacet existed,
748 * over the subfacets that have been added and then later deleted. */
749 unsigned long long int total_subfacet_life_span;
751 /* Incremented by the number of currently existing subfacets, each
752 * time we pull statistics from the kernel. */
753 unsigned long long int total_subfacet_count;
755 /* Number of times we pull statistics from the kernel. */
756 unsigned long long int n_update_stats;
758 static unsigned long long int avg_subfacet_life_span(
759 const struct ofproto_dpif *);
760 static double avg_subfacet_count(const struct ofproto_dpif *ofproto);
761 static void update_moving_averages(struct ofproto_dpif *ofproto);
762 static void dpif_stats_update_hit_count(struct ofproto_dpif *ofproto,
764 static void update_max_subfacet_count(struct ofproto_dpif *ofproto);
766 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
767 * for debugging the asynchronous flow_mod implementation.) */
770 /* All existing ofproto_dpif instances, indexed by ->up.name. */
771 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
773 static void ofproto_dpif_unixctl_init(void);
775 static struct ofproto_dpif *
776 ofproto_dpif_cast(const struct ofproto *ofproto)
778 ovs_assert(ofproto->ofproto_class == &ofproto_dpif_class);
779 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
782 static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *,
784 static struct ofport_dpif *get_odp_port(const struct ofproto_dpif *,
786 static void ofproto_trace(struct ofproto_dpif *, const struct flow *,
787 const struct ofpbuf *,
788 const struct initial_vals *, struct ds *);
790 /* Packet processing. */
791 static void update_learning_table(struct ofproto_dpif *,
792 const struct flow *, int vlan,
795 #define FLOW_MISS_MAX_BATCH 50
796 static int handle_upcalls(struct dpif_backer *, unsigned int max_batch);
798 /* Flow expiration. */
799 static int expire(struct dpif_backer *);
802 static void send_netflow_active_timeouts(struct ofproto_dpif *);
805 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
806 static size_t compose_sflow_action(const struct ofproto_dpif *,
807 struct ofpbuf *odp_actions,
808 const struct flow *, uint32_t odp_port);
809 static void compose_ipfix_action(const struct ofproto_dpif *,
810 struct ofpbuf *odp_actions,
811 const struct flow *);
812 static void add_mirror_actions(struct xlate_ctx *ctx,
813 const struct flow *flow);
814 /* Global variables. */
815 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
817 /* Initial mappings of port to bridge mappings. */
818 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
820 /* Factory functions. */
823 init(const struct shash *iface_hints)
825 struct shash_node *node;
827 /* Make a local copy, since we don't own 'iface_hints' elements. */
828 SHASH_FOR_EACH(node, iface_hints) {
829 const struct iface_hint *orig_hint = node->data;
830 struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
832 new_hint->br_name = xstrdup(orig_hint->br_name);
833 new_hint->br_type = xstrdup(orig_hint->br_type);
834 new_hint->ofp_port = orig_hint->ofp_port;
836 shash_add(&init_ofp_ports, node->name, new_hint);
841 enumerate_types(struct sset *types)
843 dp_enumerate_types(types);
847 enumerate_names(const char *type, struct sset *names)
849 struct ofproto_dpif *ofproto;
852 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
853 if (strcmp(type, ofproto->up.type)) {
856 sset_add(names, ofproto->up.name);
863 del(const char *type, const char *name)
868 error = dpif_open(name, type, &dpif);
870 error = dpif_delete(dpif);
877 port_open_type(const char *datapath_type, const char *port_type)
879 return dpif_port_open_type(datapath_type, port_type);
882 /* Type functions. */
884 static struct ofproto_dpif *
885 lookup_ofproto_dpif_by_port_name(const char *name)
887 struct ofproto_dpif *ofproto;
889 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
890 if (sset_contains(&ofproto->ports, name)) {
899 type_run(const char *type)
901 static long long int push_timer = LLONG_MIN;
902 struct dpif_backer *backer;
906 backer = shash_find_data(&all_dpif_backers, type);
908 /* This is not necessarily a problem, since backers are only
909 * created on demand. */
913 dpif_run(backer->dpif);
915 /* The most natural place to push facet statistics is when they're pulled
916 * from the datapath. However, when there are many flows in the datapath,
917 * this expensive operation can occur so frequently, that it reduces our
918 * ability to quickly set up flows. To reduce the cost, we push statistics
920 if (time_msec() > push_timer) {
921 push_timer = time_msec() + 2000;
925 if (backer->need_revalidate
926 || !tag_set_is_empty(&backer->revalidate_set)) {
927 struct tag_set revalidate_set = backer->revalidate_set;
928 bool need_revalidate = backer->need_revalidate;
929 struct ofproto_dpif *ofproto;
930 struct simap_node *node;
931 struct simap tmp_backers;
933 /* Handle tunnel garbage collection. */
934 simap_init(&tmp_backers);
935 simap_swap(&backer->tnl_backers, &tmp_backers);
937 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
938 struct ofport_dpif *iter;
940 if (backer != ofproto->backer) {
944 HMAP_FOR_EACH (iter, up.hmap_node, &ofproto->up.ports) {
947 if (!iter->tnl_port) {
951 dp_port = netdev_vport_get_dpif_port(iter->up.netdev);
952 node = simap_find(&tmp_backers, dp_port);
954 simap_put(&backer->tnl_backers, dp_port, node->data);
955 simap_delete(&tmp_backers, node);
956 node = simap_find(&backer->tnl_backers, dp_port);
958 node = simap_find(&backer->tnl_backers, dp_port);
960 uint32_t odp_port = UINT32_MAX;
962 if (!dpif_port_add(backer->dpif, iter->up.netdev,
964 simap_put(&backer->tnl_backers, dp_port, odp_port);
965 node = simap_find(&backer->tnl_backers, dp_port);
970 iter->odp_port = node ? node->data : OVSP_NONE;
971 if (tnl_port_reconfigure(&iter->up, iter->odp_port,
973 backer->need_revalidate = REV_RECONFIGURE;
978 SIMAP_FOR_EACH (node, &tmp_backers) {
979 dpif_port_del(backer->dpif, node->data);
981 simap_destroy(&tmp_backers);
983 switch (backer->need_revalidate) {
984 case REV_RECONFIGURE: COVERAGE_INC(rev_reconfigure); break;
985 case REV_STP: COVERAGE_INC(rev_stp); break;
986 case REV_PORT_TOGGLED: COVERAGE_INC(rev_port_toggled); break;
987 case REV_FLOW_TABLE: COVERAGE_INC(rev_flow_table); break;
988 case REV_INCONSISTENCY: COVERAGE_INC(rev_inconsistency); break;
991 if (backer->need_revalidate) {
992 /* Clear the drop_keys in case we should now be accepting some
993 * formerly dropped flows. */
994 drop_key_clear(backer);
997 /* Clear the revalidation flags. */
998 tag_set_init(&backer->revalidate_set);
999 backer->need_revalidate = 0;
1001 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1002 struct facet *facet, *next;
1004 if (ofproto->backer != backer) {
1008 HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &ofproto->facets) {
1010 || tag_set_intersects(&revalidate_set, facet->xout.tags)) {
1011 facet_revalidate(facet);
1018 if (timer_expired(&backer->next_expiration)) {
1019 int delay = expire(backer);
1020 timer_set_duration(&backer->next_expiration, delay);
1023 /* Check for port changes in the dpif. */
1024 while ((error = dpif_port_poll(backer->dpif, &devname)) == 0) {
1025 struct ofproto_dpif *ofproto;
1026 struct dpif_port port;
1028 /* Don't report on the datapath's device. */
1029 if (!strcmp(devname, dpif_base_name(backer->dpif))) {
1033 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
1034 &all_ofproto_dpifs) {
1035 if (simap_contains(&ofproto->backer->tnl_backers, devname)) {
1040 ofproto = lookup_ofproto_dpif_by_port_name(devname);
1041 if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
1042 /* The port was removed. If we know the datapath,
1043 * report it through poll_set(). If we don't, it may be
1044 * notifying us of a removal we initiated, so ignore it.
1045 * If there's a pending ENOBUFS, let it stand, since
1046 * everything will be reevaluated. */
1047 if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
1048 sset_add(&ofproto->port_poll_set, devname);
1049 ofproto->port_poll_errno = 0;
1051 } else if (!ofproto) {
1052 /* The port was added, but we don't know with which
1053 * ofproto we should associate it. Delete it. */
1054 dpif_port_del(backer->dpif, port.port_no);
1056 dpif_port_destroy(&port);
1062 if (error != EAGAIN) {
1063 struct ofproto_dpif *ofproto;
1065 /* There was some sort of error, so propagate it to all
1066 * ofprotos that use this backer. */
1067 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
1068 &all_ofproto_dpifs) {
1069 if (ofproto->backer == backer) {
1070 sset_clear(&ofproto->port_poll_set);
1071 ofproto->port_poll_errno = error;
1080 dpif_backer_run_fast(struct dpif_backer *backer, int max_batch)
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 < max_batch) {
1095 int retval = handle_upcalls(backer, max_batch - work);
1106 type_run_fast(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 return dpif_backer_run_fast(backer, FLOW_MISS_MAX_BATCH);
1123 static long long int port_rl = LLONG_MIN;
1124 static unsigned int backer_rl = 0;
1126 if (time_msec() >= port_rl) {
1127 struct ofproto_dpif *ofproto;
1128 struct ofport_dpif *ofport;
1130 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1132 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1133 port_run_fast(ofport);
1136 port_rl = time_msec() + 200;
1139 /* XXX: We have to be careful not to do too much work in this function. If
1140 * we call dpif_backer_run_fast() too often, or with too large a batch,
1141 * performance improves signifcantly, but at a cost. It's possible for the
1142 * number of flows in the datapath to increase without bound, and for poll
1143 * loops to take 10s of seconds. The correct solution to this problem,
1144 * long term, is to separate flow miss handling into it's own thread so it
1145 * isn't affected by revalidations, and expirations. Until then, this is
1146 * the best we can do. */
1147 if (++backer_rl >= 10) {
1148 struct shash_node *node;
1151 SHASH_FOR_EACH (node, &all_dpif_backers) {
1152 dpif_backer_run_fast(node->data, 1);
1158 type_wait(const char *type)
1160 struct dpif_backer *backer;
1162 backer = shash_find_data(&all_dpif_backers, type);
1164 /* This is not necessarily a problem, since backers are only
1165 * created on demand. */
1169 timer_wait(&backer->next_expiration);
1172 /* Basic life-cycle. */
1174 static int add_internal_flows(struct ofproto_dpif *);
1176 static struct ofproto *
1179 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
1180 return &ofproto->up;
1184 dealloc(struct ofproto *ofproto_)
1186 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1191 close_dpif_backer(struct dpif_backer *backer)
1193 struct shash_node *node;
1195 ovs_assert(backer->refcount > 0);
1197 if (--backer->refcount) {
1201 drop_key_clear(backer);
1202 hmap_destroy(&backer->drop_keys);
1204 simap_destroy(&backer->tnl_backers);
1205 hmap_destroy(&backer->odp_to_ofport_map);
1206 node = shash_find(&all_dpif_backers, backer->type);
1208 shash_delete(&all_dpif_backers, node);
1209 dpif_close(backer->dpif);
1214 /* Datapath port slated for removal from datapath. */
1215 struct odp_garbage {
1216 struct list list_node;
1221 open_dpif_backer(const char *type, struct dpif_backer **backerp)
1223 struct dpif_backer *backer;
1224 struct dpif_port_dump port_dump;
1225 struct dpif_port port;
1226 struct shash_node *node;
1227 struct list garbage_list;
1228 struct odp_garbage *garbage, *next;
1234 backer = shash_find_data(&all_dpif_backers, type);
1241 backer_name = xasprintf("ovs-%s", type);
1243 /* Remove any existing datapaths, since we assume we're the only
1244 * userspace controlling the datapath. */
1246 dp_enumerate_names(type, &names);
1247 SSET_FOR_EACH(name, &names) {
1248 struct dpif *old_dpif;
1250 /* Don't remove our backer if it exists. */
1251 if (!strcmp(name, backer_name)) {
1255 if (dpif_open(name, type, &old_dpif)) {
1256 VLOG_WARN("couldn't open old datapath %s to remove it", name);
1258 dpif_delete(old_dpif);
1259 dpif_close(old_dpif);
1262 sset_destroy(&names);
1264 backer = xmalloc(sizeof *backer);
1266 error = dpif_create_and_open(backer_name, type, &backer->dpif);
1269 VLOG_ERR("failed to open datapath of type %s: %s", type,
1275 backer->type = xstrdup(type);
1276 backer->refcount = 1;
1277 hmap_init(&backer->odp_to_ofport_map);
1278 hmap_init(&backer->drop_keys);
1279 timer_set_duration(&backer->next_expiration, 1000);
1280 backer->need_revalidate = 0;
1281 simap_init(&backer->tnl_backers);
1282 tag_set_init(&backer->revalidate_set);
1285 dpif_flow_flush(backer->dpif);
1287 /* Loop through the ports already on the datapath and remove any
1288 * that we don't need anymore. */
1289 list_init(&garbage_list);
1290 dpif_port_dump_start(&port_dump, backer->dpif);
1291 while (dpif_port_dump_next(&port_dump, &port)) {
1292 node = shash_find(&init_ofp_ports, port.name);
1293 if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
1294 garbage = xmalloc(sizeof *garbage);
1295 garbage->odp_port = port.port_no;
1296 list_push_front(&garbage_list, &garbage->list_node);
1299 dpif_port_dump_done(&port_dump);
1301 LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) {
1302 dpif_port_del(backer->dpif, garbage->odp_port);
1303 list_remove(&garbage->list_node);
1307 shash_add(&all_dpif_backers, type, backer);
1309 error = dpif_recv_set(backer->dpif, true);
1311 VLOG_ERR("failed to listen on datapath of type %s: %s",
1312 type, strerror(error));
1313 close_dpif_backer(backer);
1321 construct(struct ofproto *ofproto_)
1323 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1324 struct shash_node *node, *next;
1329 error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1334 max_ports = dpif_get_max_ports(ofproto->backer->dpif);
1335 ofproto_init_max_ports(ofproto_, MIN(max_ports, OFPP_MAX));
1337 ofproto->netflow = NULL;
1338 ofproto->sflow = NULL;
1339 ofproto->ipfix = NULL;
1340 ofproto->stp = NULL;
1341 hmap_init(&ofproto->bundles);
1342 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1343 for (i = 0; i < MAX_MIRRORS; i++) {
1344 ofproto->mirrors[i] = NULL;
1346 ofproto->has_bonded_bundles = false;
1348 hmap_init(&ofproto->facets);
1349 hmap_init(&ofproto->subfacets);
1350 ofproto->governor = NULL;
1351 ofproto->consistency_rl = LLONG_MIN;
1353 for (i = 0; i < N_TABLES; i++) {
1354 struct table_dpif *table = &ofproto->tables[i];
1356 table->catchall_table = NULL;
1357 table->other_table = NULL;
1358 table->basis = random_uint32();
1361 list_init(&ofproto->completions);
1363 ofproto_dpif_unixctl_init();
1365 ofproto->has_mirrors = false;
1366 ofproto->has_bundle_action = false;
1368 hmap_init(&ofproto->vlandev_map);
1369 hmap_init(&ofproto->realdev_vid_map);
1371 sset_init(&ofproto->ports);
1372 sset_init(&ofproto->ghost_ports);
1373 sset_init(&ofproto->port_poll_set);
1374 ofproto->port_poll_errno = 0;
1376 SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1377 struct iface_hint *iface_hint = node->data;
1379 if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1380 /* Check if the datapath already has this port. */
1381 if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1382 sset_add(&ofproto->ports, node->name);
1385 free(iface_hint->br_name);
1386 free(iface_hint->br_type);
1388 shash_delete(&init_ofp_ports, node);
1392 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1393 hash_string(ofproto->up.name, 0));
1394 memset(&ofproto->stats, 0, sizeof ofproto->stats);
1396 ofproto_init_tables(ofproto_, N_TABLES);
1397 error = add_internal_flows(ofproto);
1398 ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1401 ofproto->n_missed = 0;
1403 ofproto->max_n_subfacet = 0;
1404 ofproto->created = time_msec();
1405 ofproto->last_minute = ofproto->created;
1406 memset(&ofproto->hourly, 0, sizeof ofproto->hourly);
1407 memset(&ofproto->daily, 0, sizeof ofproto->daily);
1408 ofproto->subfacet_add_count = 0;
1409 ofproto->subfacet_del_count = 0;
1410 ofproto->total_subfacet_add_count = 0;
1411 ofproto->total_subfacet_del_count = 0;
1412 ofproto->total_subfacet_life_span = 0;
1413 ofproto->total_subfacet_count = 0;
1414 ofproto->n_update_stats = 0;
1420 add_internal_flow(struct ofproto_dpif *ofproto, int id,
1421 const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1423 struct ofputil_flow_mod fm;
1426 match_init_catchall(&fm.match);
1428 match_set_reg(&fm.match, 0, id);
1429 fm.new_cookie = htonll(0);
1430 fm.cookie = htonll(0);
1431 fm.cookie_mask = htonll(0);
1432 fm.table_id = TBL_INTERNAL;
1433 fm.command = OFPFC_ADD;
1434 fm.idle_timeout = 0;
1435 fm.hard_timeout = 0;
1439 fm.ofpacts = ofpacts->data;
1440 fm.ofpacts_len = ofpacts->size;
1442 error = ofproto_flow_mod(&ofproto->up, &fm);
1444 VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
1445 id, ofperr_to_string(error));
1449 *rulep = rule_dpif_lookup__(ofproto, &fm.match.flow, TBL_INTERNAL);
1450 ovs_assert(*rulep != NULL);
1456 add_internal_flows(struct ofproto_dpif *ofproto)
1458 struct ofpact_controller *controller;
1459 uint64_t ofpacts_stub[128 / 8];
1460 struct ofpbuf ofpacts;
1464 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1467 controller = ofpact_put_CONTROLLER(&ofpacts);
1468 controller->max_len = UINT16_MAX;
1469 controller->controller_id = 0;
1470 controller->reason = OFPR_NO_MATCH;
1471 ofpact_pad(&ofpacts);
1473 error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule);
1478 ofpbuf_clear(&ofpacts);
1479 error = add_internal_flow(ofproto, id++, &ofpacts,
1480 &ofproto->no_packet_in_rule);
1485 complete_operations(struct ofproto_dpif *ofproto)
1487 struct dpif_completion *c, *next;
1489 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
1490 ofoperation_complete(c->op, 0);
1491 list_remove(&c->list_node);
1497 destruct(struct ofproto *ofproto_)
1499 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1500 struct rule_dpif *rule, *next_rule;
1501 struct oftable *table;
1504 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1505 complete_operations(ofproto);
1507 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1508 struct cls_cursor cursor;
1510 cls_cursor_init(&cursor, &table->cls, NULL);
1511 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1512 ofproto_rule_destroy(&rule->up);
1516 for (i = 0; i < MAX_MIRRORS; i++) {
1517 mirror_destroy(ofproto->mirrors[i]);
1520 netflow_destroy(ofproto->netflow);
1521 dpif_sflow_destroy(ofproto->sflow);
1522 hmap_destroy(&ofproto->bundles);
1523 mac_learning_destroy(ofproto->ml);
1525 hmap_destroy(&ofproto->facets);
1526 hmap_destroy(&ofproto->subfacets);
1527 governor_destroy(ofproto->governor);
1529 hmap_destroy(&ofproto->vlandev_map);
1530 hmap_destroy(&ofproto->realdev_vid_map);
1532 sset_destroy(&ofproto->ports);
1533 sset_destroy(&ofproto->ghost_ports);
1534 sset_destroy(&ofproto->port_poll_set);
1536 close_dpif_backer(ofproto->backer);
1540 run_fast(struct ofproto *ofproto_)
1542 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1543 struct ofport_dpif *ofport;
1545 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1546 port_run_fast(ofport);
1553 run(struct ofproto *ofproto_)
1555 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1556 struct ofport_dpif *ofport;
1557 struct ofbundle *bundle;
1561 complete_operations(ofproto);
1564 error = run_fast(ofproto_);
1569 if (ofproto->netflow) {
1570 if (netflow_run(ofproto->netflow)) {
1571 send_netflow_active_timeouts(ofproto);
1574 if (ofproto->sflow) {
1575 dpif_sflow_run(ofproto->sflow);
1578 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1581 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1586 mac_learning_run(ofproto->ml, &ofproto->backer->revalidate_set);
1588 /* Check the consistency of a random facet, to aid debugging. */
1589 if (time_msec() >= ofproto->consistency_rl
1590 && !hmap_is_empty(&ofproto->facets)
1591 && !ofproto->backer->need_revalidate) {
1592 struct facet *facet;
1594 ofproto->consistency_rl = time_msec() + 250;
1596 facet = CONTAINER_OF(hmap_random_node(&ofproto->facets),
1597 struct facet, hmap_node);
1598 if (!tag_set_intersects(&ofproto->backer->revalidate_set,
1599 facet->xout.tags)) {
1600 if (!facet_check_consistency(facet)) {
1601 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
1606 if (ofproto->governor) {
1609 governor_run(ofproto->governor);
1611 /* If the governor has shrunk to its minimum size and the number of
1612 * subfacets has dwindled, then drop the governor entirely.
1614 * For hysteresis, the number of subfacets to drop the governor is
1615 * smaller than the number needed to trigger its creation. */
1616 n_subfacets = hmap_count(&ofproto->subfacets);
1617 if (n_subfacets * 4 < ofproto->up.flow_eviction_threshold
1618 && governor_is_idle(ofproto->governor)) {
1619 governor_destroy(ofproto->governor);
1620 ofproto->governor = NULL;
1628 wait(struct ofproto *ofproto_)
1630 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1631 struct ofport_dpif *ofport;
1632 struct ofbundle *bundle;
1634 if (!clogged && !list_is_empty(&ofproto->completions)) {
1635 poll_immediate_wake();
1638 dpif_wait(ofproto->backer->dpif);
1639 dpif_recv_wait(ofproto->backer->dpif);
1640 if (ofproto->sflow) {
1641 dpif_sflow_wait(ofproto->sflow);
1643 if (!tag_set_is_empty(&ofproto->backer->revalidate_set)) {
1644 poll_immediate_wake();
1646 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1649 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1650 bundle_wait(bundle);
1652 if (ofproto->netflow) {
1653 netflow_wait(ofproto->netflow);
1655 mac_learning_wait(ofproto->ml);
1657 if (ofproto->backer->need_revalidate) {
1658 /* Shouldn't happen, but if it does just go around again. */
1659 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1660 poll_immediate_wake();
1662 if (ofproto->governor) {
1663 governor_wait(ofproto->governor);
1668 get_memory_usage(const struct ofproto *ofproto_, struct simap *usage)
1670 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1672 simap_increase(usage, "facets", hmap_count(&ofproto->facets));
1673 simap_increase(usage, "subfacets", hmap_count(&ofproto->subfacets));
1677 flush(struct ofproto *ofproto_)
1679 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1680 struct subfacet *subfacet, *next_subfacet;
1681 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
1685 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
1686 &ofproto->subfacets) {
1687 if (subfacet->path != SF_NOT_INSTALLED) {
1688 batch[n_batch++] = subfacet;
1689 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
1690 subfacet_destroy_batch(ofproto, batch, n_batch);
1694 subfacet_destroy(subfacet);
1699 subfacet_destroy_batch(ofproto, batch, n_batch);
1704 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1705 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1707 *arp_match_ip = true;
1708 *actions = (OFPUTIL_A_OUTPUT |
1709 OFPUTIL_A_SET_VLAN_VID |
1710 OFPUTIL_A_SET_VLAN_PCP |
1711 OFPUTIL_A_STRIP_VLAN |
1712 OFPUTIL_A_SET_DL_SRC |
1713 OFPUTIL_A_SET_DL_DST |
1714 OFPUTIL_A_SET_NW_SRC |
1715 OFPUTIL_A_SET_NW_DST |
1716 OFPUTIL_A_SET_NW_TOS |
1717 OFPUTIL_A_SET_TP_SRC |
1718 OFPUTIL_A_SET_TP_DST |
1723 get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots)
1725 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1726 struct dpif_dp_stats s;
1727 uint64_t n_miss, n_no_pkt_in, n_bytes;
1730 strcpy(ots->name, "classifier");
1732 dpif_get_dp_stats(ofproto->backer->dpif, &s);
1733 rule_get_stats(&ofproto->miss_rule->up, &n_miss, &n_bytes);
1734 rule_get_stats(&ofproto->no_packet_in_rule->up, &n_no_pkt_in, &n_bytes);
1736 n_lookup = s.n_hit + s.n_missed;
1737 ots->lookup_count = htonll(n_lookup);
1738 ots->matched_count = htonll(n_lookup - n_miss - n_no_pkt_in);
1741 static struct ofport *
1744 struct ofport_dpif *port = xmalloc(sizeof *port);
1749 port_dealloc(struct ofport *port_)
1751 struct ofport_dpif *port = ofport_dpif_cast(port_);
1756 port_construct(struct ofport *port_)
1758 struct ofport_dpif *port = ofport_dpif_cast(port_);
1759 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1760 const struct netdev *netdev = port->up.netdev;
1761 struct dpif_port dpif_port;
1764 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1765 port->bundle = NULL;
1768 port->tag = tag_create_random();
1769 port->may_enable = true;
1770 port->stp_port = NULL;
1771 port->stp_state = STP_DISABLED;
1772 port->tnl_port = NULL;
1773 hmap_init(&port->priorities);
1774 port->realdev_ofp_port = 0;
1775 port->vlandev_vid = 0;
1776 port->carrier_seq = netdev_get_carrier_resets(netdev);
1778 if (netdev_vport_is_patch(netdev)) {
1779 /* By bailing out here, we don't submit the port to the sFlow module
1780 * to be considered for counter polling export. This is correct
1781 * because the patch port represents an interface that sFlow considers
1782 * to be "internal" to the switch as a whole, and therefore not an
1783 * candidate for counter polling. */
1784 port->odp_port = OVSP_NONE;
1788 error = dpif_port_query_by_name(ofproto->backer->dpif,
1789 netdev_vport_get_dpif_port(netdev),
1795 port->odp_port = dpif_port.port_no;
1797 if (netdev_get_tunnel_config(netdev)) {
1798 port->tnl_port = tnl_port_add(&port->up, port->odp_port);
1800 /* Sanity-check that a mapping doesn't already exist. This
1801 * shouldn't happen for non-tunnel ports. */
1802 if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1803 VLOG_ERR("port %s already has an OpenFlow port number",
1805 dpif_port_destroy(&dpif_port);
1809 hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1810 hash_int(port->odp_port, 0));
1812 dpif_port_destroy(&dpif_port);
1814 if (ofproto->sflow) {
1815 dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1822 port_destruct(struct ofport *port_)
1824 struct ofport_dpif *port = ofport_dpif_cast(port_);
1825 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1826 const char *dp_port_name = netdev_vport_get_dpif_port(port->up.netdev);
1827 const char *devname = netdev_get_name(port->up.netdev);
1829 if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1830 /* The underlying device is still there, so delete it. This
1831 * happens when the ofproto is being destroyed, since the caller
1832 * assumes that removal of attached ports will happen as part of
1834 if (!port->tnl_port) {
1835 dpif_port_del(ofproto->backer->dpif, port->odp_port);
1837 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1840 if (port->odp_port != OVSP_NONE && !port->tnl_port) {
1841 hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1844 tnl_port_del(port->tnl_port);
1845 sset_find_and_delete(&ofproto->ports, devname);
1846 sset_find_and_delete(&ofproto->ghost_ports, devname);
1847 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1848 bundle_remove(port_);
1849 set_cfm(port_, NULL);
1850 set_bfd(port_, NULL);
1851 if (ofproto->sflow) {
1852 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1855 ofport_clear_priorities(port);
1856 hmap_destroy(&port->priorities);
1860 port_modified(struct ofport *port_)
1862 struct ofport_dpif *port = ofport_dpif_cast(port_);
1864 if (port->bundle && port->bundle->bond) {
1865 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1870 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1872 struct ofport_dpif *port = ofport_dpif_cast(port_);
1873 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1874 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1876 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1877 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1878 OFPUTIL_PC_NO_PACKET_IN)) {
1879 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1881 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1882 bundle_update(port->bundle);
1888 set_sflow(struct ofproto *ofproto_,
1889 const struct ofproto_sflow_options *sflow_options)
1891 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1892 struct dpif_sflow *ds = ofproto->sflow;
1894 if (sflow_options) {
1896 struct ofport_dpif *ofport;
1898 ds = ofproto->sflow = dpif_sflow_create();
1899 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1900 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1902 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1904 dpif_sflow_set_options(ds, sflow_options);
1907 dpif_sflow_destroy(ds);
1908 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1909 ofproto->sflow = NULL;
1917 struct ofproto *ofproto_,
1918 const struct ofproto_ipfix_bridge_exporter_options *bridge_exporter_options,
1919 const struct ofproto_ipfix_flow_exporter_options *flow_exporters_options,
1920 size_t n_flow_exporters_options)
1922 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1923 struct dpif_ipfix *di = ofproto->ipfix;
1925 if (bridge_exporter_options || flow_exporters_options) {
1927 di = ofproto->ipfix = dpif_ipfix_create();
1929 dpif_ipfix_set_options(
1930 di, bridge_exporter_options, flow_exporters_options,
1931 n_flow_exporters_options);
1934 dpif_ipfix_destroy(di);
1935 ofproto->ipfix = NULL;
1942 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1944 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1951 struct ofproto_dpif *ofproto;
1953 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1954 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1955 ofport->cfm = cfm_create(ofport->up.netdev);
1958 if (cfm_configure(ofport->cfm, s)) {
1964 cfm_destroy(ofport->cfm);
1970 get_cfm_status(const struct ofport *ofport_,
1971 struct ofproto_cfm_status *status)
1973 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1976 status->faults = cfm_get_fault(ofport->cfm);
1977 status->remote_opstate = cfm_get_opup(ofport->cfm);
1978 status->health = cfm_get_health(ofport->cfm);
1979 cfm_get_remote_mpids(ofport->cfm, &status->rmps, &status->n_rmps);
1987 set_bfd(struct ofport *ofport_, const struct smap *cfg)
1989 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
1990 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1994 ofport->bfd = bfd_configure(old, netdev_get_name(ofport->up.netdev), cfg);
1995 if (ofport->bfd != old) {
1996 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2003 get_bfd_status(struct ofport *ofport_, struct smap *smap)
2005 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2008 bfd_get_status(ofport->bfd, smap);
2015 /* Spanning Tree. */
2018 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
2020 struct ofproto_dpif *ofproto = ofproto_;
2021 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
2022 struct ofport_dpif *ofport;
2024 ofport = stp_port_get_aux(sp);
2026 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
2027 ofproto->up.name, port_num);
2029 struct eth_header *eth = pkt->l2;
2031 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
2032 if (eth_addr_is_zero(eth->eth_src)) {
2033 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
2034 "with unknown MAC", ofproto->up.name, port_num);
2036 send_packet(ofport, pkt);
2042 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
2044 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
2046 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2048 /* Only revalidate flows if the configuration changed. */
2049 if (!s != !ofproto->stp) {
2050 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2054 if (!ofproto->stp) {
2055 ofproto->stp = stp_create(ofproto_->name, s->system_id,
2056 send_bpdu_cb, ofproto);
2057 ofproto->stp_last_tick = time_msec();
2060 stp_set_bridge_id(ofproto->stp, s->system_id);
2061 stp_set_bridge_priority(ofproto->stp, s->priority);
2062 stp_set_hello_time(ofproto->stp, s->hello_time);
2063 stp_set_max_age(ofproto->stp, s->max_age);
2064 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
2066 struct ofport *ofport;
2068 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
2069 set_stp_port(ofport, NULL);
2072 stp_destroy(ofproto->stp);
2073 ofproto->stp = NULL;
2080 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
2082 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2086 s->bridge_id = stp_get_bridge_id(ofproto->stp);
2087 s->designated_root = stp_get_designated_root(ofproto->stp);
2088 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
2097 update_stp_port_state(struct ofport_dpif *ofport)
2099 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2100 enum stp_state state;
2102 /* Figure out new state. */
2103 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
2107 if (ofport->stp_state != state) {
2108 enum ofputil_port_state of_state;
2111 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
2112 netdev_get_name(ofport->up.netdev),
2113 stp_state_name(ofport->stp_state),
2114 stp_state_name(state));
2115 if (stp_learn_in_state(ofport->stp_state)
2116 != stp_learn_in_state(state)) {
2117 /* xxx Learning action flows should also be flushed. */
2118 mac_learning_flush(ofproto->ml,
2119 &ofproto->backer->revalidate_set);
2121 fwd_change = stp_forward_in_state(ofport->stp_state)
2122 != stp_forward_in_state(state);
2124 ofproto->backer->need_revalidate = REV_STP;
2125 ofport->stp_state = state;
2126 ofport->stp_state_entered = time_msec();
2128 if (fwd_change && ofport->bundle) {
2129 bundle_update(ofport->bundle);
2132 /* Update the STP state bits in the OpenFlow port description. */
2133 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
2134 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
2135 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
2136 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
2137 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
2139 ofproto_port_set_state(&ofport->up, of_state);
2143 /* Configures STP on 'ofport_' using the settings defined in 's'. The
2144 * caller is responsible for assigning STP port numbers and ensuring
2145 * there are no duplicates. */
2147 set_stp_port(struct ofport *ofport_,
2148 const struct ofproto_port_stp_settings *s)
2150 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2151 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2152 struct stp_port *sp = ofport->stp_port;
2154 if (!s || !s->enable) {
2156 ofport->stp_port = NULL;
2157 stp_port_disable(sp);
2158 update_stp_port_state(ofport);
2161 } else if (sp && stp_port_no(sp) != s->port_num
2162 && ofport == stp_port_get_aux(sp)) {
2163 /* The port-id changed, so disable the old one if it's not
2164 * already in use by another port. */
2165 stp_port_disable(sp);
2168 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
2169 stp_port_enable(sp);
2171 stp_port_set_aux(sp, ofport);
2172 stp_port_set_priority(sp, s->priority);
2173 stp_port_set_path_cost(sp, s->path_cost);
2175 update_stp_port_state(ofport);
2181 get_stp_port_status(struct ofport *ofport_,
2182 struct ofproto_port_stp_status *s)
2184 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2185 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2186 struct stp_port *sp = ofport->stp_port;
2188 if (!ofproto->stp || !sp) {
2194 s->port_id = stp_port_get_id(sp);
2195 s->state = stp_port_get_state(sp);
2196 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
2197 s->role = stp_port_get_role(sp);
2198 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
2204 stp_run(struct ofproto_dpif *ofproto)
2207 long long int now = time_msec();
2208 long long int elapsed = now - ofproto->stp_last_tick;
2209 struct stp_port *sp;
2212 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
2213 ofproto->stp_last_tick = now;
2215 while (stp_get_changed_port(ofproto->stp, &sp)) {
2216 struct ofport_dpif *ofport = stp_port_get_aux(sp);
2219 update_stp_port_state(ofport);
2223 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
2224 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
2230 stp_wait(struct ofproto_dpif *ofproto)
2233 poll_timer_wait(1000);
2237 /* Returns true if STP should process 'flow'. */
2239 stp_should_process_flow(const struct flow *flow)
2241 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
2245 stp_process_packet(const struct ofport_dpif *ofport,
2246 const struct ofpbuf *packet)
2248 struct ofpbuf payload = *packet;
2249 struct eth_header *eth = payload.data;
2250 struct stp_port *sp = ofport->stp_port;
2252 /* Sink packets on ports that have STP disabled when the bridge has
2254 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
2258 /* Trim off padding on payload. */
2259 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
2260 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
2263 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
2264 stp_received_bpdu(sp, payload.data, payload.size);
2268 static struct priority_to_dscp *
2269 get_priority(const struct ofport_dpif *ofport, uint32_t priority)
2271 struct priority_to_dscp *pdscp;
2274 hash = hash_int(priority, 0);
2275 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) {
2276 if (pdscp->priority == priority) {
2284 ofport_clear_priorities(struct ofport_dpif *ofport)
2286 struct priority_to_dscp *pdscp, *next;
2288 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) {
2289 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
2295 set_queues(struct ofport *ofport_,
2296 const struct ofproto_port_queue *qdscp_list,
2299 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2300 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2301 struct hmap new = HMAP_INITIALIZER(&new);
2304 for (i = 0; i < n_qdscp; i++) {
2305 struct priority_to_dscp *pdscp;
2309 dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
2310 if (dpif_queue_to_priority(ofproto->backer->dpif, qdscp_list[i].queue,
2315 pdscp = get_priority(ofport, priority);
2317 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
2319 pdscp = xmalloc(sizeof *pdscp);
2320 pdscp->priority = priority;
2322 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2325 if (pdscp->dscp != dscp) {
2327 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2330 hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0));
2333 if (!hmap_is_empty(&ofport->priorities)) {
2334 ofport_clear_priorities(ofport);
2335 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2338 hmap_swap(&new, &ofport->priorities);
2346 /* Expires all MAC learning entries associated with 'bundle' and forces its
2347 * ofproto to revalidate every flow.
2349 * Normally MAC learning entries are removed only from the ofproto associated
2350 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2351 * are removed from every ofproto. When patch ports and SLB bonds are in use
2352 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2353 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2354 * with the host from which it migrated. */
2356 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2358 struct ofproto_dpif *ofproto = bundle->ofproto;
2359 struct mac_learning *ml = ofproto->ml;
2360 struct mac_entry *mac, *next_mac;
2362 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2363 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2364 if (mac->port.p == bundle) {
2366 struct ofproto_dpif *o;
2368 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2370 struct mac_entry *e;
2372 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan,
2375 mac_learning_expire(o->ml, e);
2381 mac_learning_expire(ml, mac);
2386 static struct ofbundle *
2387 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2389 struct ofbundle *bundle;
2391 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2392 &ofproto->bundles) {
2393 if (bundle->aux == aux) {
2400 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
2401 * ones that are found to 'bundles'. */
2403 bundle_lookup_multiple(struct ofproto_dpif *ofproto,
2404 void **auxes, size_t n_auxes,
2405 struct hmapx *bundles)
2409 hmapx_init(bundles);
2410 for (i = 0; i < n_auxes; i++) {
2411 struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
2413 hmapx_add(bundles, bundle);
2419 bundle_update(struct ofbundle *bundle)
2421 struct ofport_dpif *port;
2423 bundle->floodable = true;
2424 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2425 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2426 || !stp_forward_in_state(port->stp_state)) {
2427 bundle->floodable = false;
2434 bundle_del_port(struct ofport_dpif *port)
2436 struct ofbundle *bundle = port->bundle;
2438 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2440 list_remove(&port->bundle_node);
2441 port->bundle = NULL;
2444 lacp_slave_unregister(bundle->lacp, port);
2447 bond_slave_unregister(bundle->bond, port);
2450 bundle_update(bundle);
2454 bundle_add_port(struct ofbundle *bundle, uint16_t ofp_port,
2455 struct lacp_slave_settings *lacp)
2457 struct ofport_dpif *port;
2459 port = get_ofp_port(bundle->ofproto, ofp_port);
2464 if (port->bundle != bundle) {
2465 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2467 bundle_del_port(port);
2470 port->bundle = bundle;
2471 list_push_back(&bundle->ports, &port->bundle_node);
2472 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2473 || !stp_forward_in_state(port->stp_state)) {
2474 bundle->floodable = false;
2478 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2479 lacp_slave_register(bundle->lacp, port, lacp);
2486 bundle_destroy(struct ofbundle *bundle)
2488 struct ofproto_dpif *ofproto;
2489 struct ofport_dpif *port, *next_port;
2496 ofproto = bundle->ofproto;
2497 for (i = 0; i < MAX_MIRRORS; i++) {
2498 struct ofmirror *m = ofproto->mirrors[i];
2500 if (m->out == bundle) {
2502 } else if (hmapx_find_and_delete(&m->srcs, bundle)
2503 || hmapx_find_and_delete(&m->dsts, bundle)) {
2504 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2509 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2510 bundle_del_port(port);
2513 bundle_flush_macs(bundle, true);
2514 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2516 free(bundle->trunks);
2517 lacp_destroy(bundle->lacp);
2518 bond_destroy(bundle->bond);
2523 bundle_set(struct ofproto *ofproto_, void *aux,
2524 const struct ofproto_bundle_settings *s)
2526 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2527 bool need_flush = false;
2528 struct ofport_dpif *port;
2529 struct ofbundle *bundle;
2530 unsigned long *trunks;
2536 bundle_destroy(bundle_lookup(ofproto, aux));
2540 ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2541 ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2543 bundle = bundle_lookup(ofproto, aux);
2545 bundle = xmalloc(sizeof *bundle);
2547 bundle->ofproto = ofproto;
2548 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2549 hash_pointer(aux, 0));
2551 bundle->name = NULL;
2553 list_init(&bundle->ports);
2554 bundle->vlan_mode = PORT_VLAN_TRUNK;
2556 bundle->trunks = NULL;
2557 bundle->use_priority_tags = s->use_priority_tags;
2558 bundle->lacp = NULL;
2559 bundle->bond = NULL;
2561 bundle->floodable = true;
2563 bundle->src_mirrors = 0;
2564 bundle->dst_mirrors = 0;
2565 bundle->mirror_out = 0;
2568 if (!bundle->name || strcmp(s->name, bundle->name)) {
2570 bundle->name = xstrdup(s->name);
2575 if (!bundle->lacp) {
2576 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2577 bundle->lacp = lacp_create();
2579 lacp_configure(bundle->lacp, s->lacp);
2581 lacp_destroy(bundle->lacp);
2582 bundle->lacp = NULL;
2585 /* Update set of ports. */
2587 for (i = 0; i < s->n_slaves; i++) {
2588 if (!bundle_add_port(bundle, s->slaves[i],
2589 s->lacp ? &s->lacp_slaves[i] : NULL)) {
2593 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2594 struct ofport_dpif *next_port;
2596 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2597 for (i = 0; i < s->n_slaves; i++) {
2598 if (s->slaves[i] == port->up.ofp_port) {
2603 bundle_del_port(port);
2607 ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2609 if (list_is_empty(&bundle->ports)) {
2610 bundle_destroy(bundle);
2614 /* Set VLAN tagging mode */
2615 if (s->vlan_mode != bundle->vlan_mode
2616 || s->use_priority_tags != bundle->use_priority_tags) {
2617 bundle->vlan_mode = s->vlan_mode;
2618 bundle->use_priority_tags = s->use_priority_tags;
2623 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2624 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2626 if (vlan != bundle->vlan) {
2627 bundle->vlan = vlan;
2631 /* Get trunked VLANs. */
2632 switch (s->vlan_mode) {
2633 case PORT_VLAN_ACCESS:
2637 case PORT_VLAN_TRUNK:
2638 trunks = CONST_CAST(unsigned long *, s->trunks);
2641 case PORT_VLAN_NATIVE_UNTAGGED:
2642 case PORT_VLAN_NATIVE_TAGGED:
2643 if (vlan != 0 && (!s->trunks
2644 || !bitmap_is_set(s->trunks, vlan)
2645 || bitmap_is_set(s->trunks, 0))) {
2646 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2648 trunks = bitmap_clone(s->trunks, 4096);
2650 trunks = bitmap_allocate1(4096);
2652 bitmap_set1(trunks, vlan);
2653 bitmap_set0(trunks, 0);
2655 trunks = CONST_CAST(unsigned long *, s->trunks);
2662 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2663 free(bundle->trunks);
2664 if (trunks == s->trunks) {
2665 bundle->trunks = vlan_bitmap_clone(trunks);
2667 bundle->trunks = trunks;
2672 if (trunks != s->trunks) {
2677 if (!list_is_short(&bundle->ports)) {
2678 bundle->ofproto->has_bonded_bundles = true;
2680 if (bond_reconfigure(bundle->bond, s->bond)) {
2681 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2684 bundle->bond = bond_create(s->bond);
2685 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2688 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2689 bond_slave_register(bundle->bond, port, port->up.netdev);
2692 bond_destroy(bundle->bond);
2693 bundle->bond = NULL;
2696 /* If we changed something that would affect MAC learning, un-learn
2697 * everything on this port and force flow revalidation. */
2699 bundle_flush_macs(bundle, false);
2706 bundle_remove(struct ofport *port_)
2708 struct ofport_dpif *port = ofport_dpif_cast(port_);
2709 struct ofbundle *bundle = port->bundle;
2712 bundle_del_port(port);
2713 if (list_is_empty(&bundle->ports)) {
2714 bundle_destroy(bundle);
2715 } else if (list_is_short(&bundle->ports)) {
2716 bond_destroy(bundle->bond);
2717 bundle->bond = NULL;
2723 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2725 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2726 struct ofport_dpif *port = port_;
2727 uint8_t ea[ETH_ADDR_LEN];
2730 error = netdev_get_etheraddr(port->up.netdev, ea);
2732 struct ofpbuf packet;
2735 ofpbuf_init(&packet, 0);
2736 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2738 memcpy(packet_pdu, pdu, pdu_size);
2740 send_packet(port, &packet);
2741 ofpbuf_uninit(&packet);
2743 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2744 "%s (%s)", port->bundle->name,
2745 netdev_get_name(port->up.netdev), strerror(error));
2750 bundle_send_learning_packets(struct ofbundle *bundle)
2752 struct ofproto_dpif *ofproto = bundle->ofproto;
2753 int error, n_packets, n_errors;
2754 struct mac_entry *e;
2756 error = n_packets = n_errors = 0;
2757 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2758 if (e->port.p != bundle) {
2759 struct ofpbuf *learning_packet;
2760 struct ofport_dpif *port;
2764 /* The assignment to "port" is unnecessary but makes "grep"ing for
2765 * struct ofport_dpif more effective. */
2766 learning_packet = bond_compose_learning_packet(bundle->bond,
2770 ret = send_packet(port, learning_packet);
2771 ofpbuf_delete(learning_packet);
2781 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2782 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2783 "packets, last error was: %s",
2784 bundle->name, n_errors, n_packets, strerror(error));
2786 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2787 bundle->name, n_packets);
2792 bundle_run(struct ofbundle *bundle)
2795 lacp_run(bundle->lacp, send_pdu_cb);
2798 struct ofport_dpif *port;
2800 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2801 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2804 bond_run(bundle->bond, &bundle->ofproto->backer->revalidate_set,
2805 lacp_status(bundle->lacp));
2806 if (bond_should_send_learning_packets(bundle->bond)) {
2807 bundle_send_learning_packets(bundle);
2813 bundle_wait(struct ofbundle *bundle)
2816 lacp_wait(bundle->lacp);
2819 bond_wait(bundle->bond);
2826 mirror_scan(struct ofproto_dpif *ofproto)
2830 for (idx = 0; idx < MAX_MIRRORS; idx++) {
2831 if (!ofproto->mirrors[idx]) {
2838 static struct ofmirror *
2839 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
2843 for (i = 0; i < MAX_MIRRORS; i++) {
2844 struct ofmirror *mirror = ofproto->mirrors[i];
2845 if (mirror && mirror->aux == aux) {
2853 /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */
2855 mirror_update_dups(struct ofproto_dpif *ofproto)
2859 for (i = 0; i < MAX_MIRRORS; i++) {
2860 struct ofmirror *m = ofproto->mirrors[i];
2863 m->dup_mirrors = MIRROR_MASK_C(1) << i;
2867 for (i = 0; i < MAX_MIRRORS; i++) {
2868 struct ofmirror *m1 = ofproto->mirrors[i];
2875 for (j = i + 1; j < MAX_MIRRORS; j++) {
2876 struct ofmirror *m2 = ofproto->mirrors[j];
2878 if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) {
2879 m1->dup_mirrors |= MIRROR_MASK_C(1) << j;
2880 m2->dup_mirrors |= m1->dup_mirrors;
2887 mirror_set(struct ofproto *ofproto_, void *aux,
2888 const struct ofproto_mirror_settings *s)
2890 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2891 mirror_mask_t mirror_bit;
2892 struct ofbundle *bundle;
2893 struct ofmirror *mirror;
2894 struct ofbundle *out;
2895 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
2896 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
2899 mirror = mirror_lookup(ofproto, aux);
2901 mirror_destroy(mirror);
2907 idx = mirror_scan(ofproto);
2909 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
2911 ofproto->up.name, MAX_MIRRORS, s->name);
2915 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
2916 mirror->ofproto = ofproto;
2919 mirror->out_vlan = -1;
2920 mirror->name = NULL;
2923 if (!mirror->name || strcmp(s->name, mirror->name)) {
2925 mirror->name = xstrdup(s->name);
2928 /* Get the new configuration. */
2929 if (s->out_bundle) {
2930 out = bundle_lookup(ofproto, s->out_bundle);
2932 mirror_destroy(mirror);
2938 out_vlan = s->out_vlan;
2940 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
2941 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
2943 /* If the configuration has not changed, do nothing. */
2944 if (hmapx_equals(&srcs, &mirror->srcs)
2945 && hmapx_equals(&dsts, &mirror->dsts)
2946 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
2947 && mirror->out == out
2948 && mirror->out_vlan == out_vlan)
2950 hmapx_destroy(&srcs);
2951 hmapx_destroy(&dsts);
2955 hmapx_swap(&srcs, &mirror->srcs);
2956 hmapx_destroy(&srcs);
2958 hmapx_swap(&dsts, &mirror->dsts);
2959 hmapx_destroy(&dsts);
2961 free(mirror->vlans);
2962 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
2965 mirror->out_vlan = out_vlan;
2967 /* Update bundles. */
2968 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2969 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
2970 if (hmapx_contains(&mirror->srcs, bundle)) {
2971 bundle->src_mirrors |= mirror_bit;
2973 bundle->src_mirrors &= ~mirror_bit;
2976 if (hmapx_contains(&mirror->dsts, bundle)) {
2977 bundle->dst_mirrors |= mirror_bit;
2979 bundle->dst_mirrors &= ~mirror_bit;
2982 if (mirror->out == bundle) {
2983 bundle->mirror_out |= mirror_bit;
2985 bundle->mirror_out &= ~mirror_bit;
2989 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2990 ofproto->has_mirrors = true;
2991 mac_learning_flush(ofproto->ml,
2992 &ofproto->backer->revalidate_set);
2993 mirror_update_dups(ofproto);
2999 mirror_destroy(struct ofmirror *mirror)
3001 struct ofproto_dpif *ofproto;
3002 mirror_mask_t mirror_bit;
3003 struct ofbundle *bundle;
3010 ofproto = mirror->ofproto;
3011 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3012 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
3014 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
3015 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
3016 bundle->src_mirrors &= ~mirror_bit;
3017 bundle->dst_mirrors &= ~mirror_bit;
3018 bundle->mirror_out &= ~mirror_bit;
3021 hmapx_destroy(&mirror->srcs);
3022 hmapx_destroy(&mirror->dsts);
3023 free(mirror->vlans);
3025 ofproto->mirrors[mirror->idx] = NULL;
3029 mirror_update_dups(ofproto);
3031 ofproto->has_mirrors = false;
3032 for (i = 0; i < MAX_MIRRORS; i++) {
3033 if (ofproto->mirrors[i]) {
3034 ofproto->has_mirrors = true;
3041 mirror_get_stats(struct ofproto *ofproto_, void *aux,
3042 uint64_t *packets, uint64_t *bytes)
3044 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3045 struct ofmirror *mirror = mirror_lookup(ofproto, aux);
3048 *packets = *bytes = UINT64_MAX;
3054 *packets = mirror->packet_count;
3055 *bytes = mirror->byte_count;
3061 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
3063 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3064 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
3065 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
3071 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
3073 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3074 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
3075 return bundle && bundle->mirror_out != 0;
3079 forward_bpdu_changed(struct ofproto *ofproto_)
3081 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3082 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3086 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
3089 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3090 mac_learning_set_idle_time(ofproto->ml, idle_time);
3091 mac_learning_set_max_entries(ofproto->ml, max_entries);
3096 static struct ofport_dpif *
3097 get_ofp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
3099 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
3100 return ofport ? ofport_dpif_cast(ofport) : NULL;
3103 static struct ofport_dpif *
3104 get_odp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
3106 struct ofport_dpif *port = odp_port_to_ofport(ofproto->backer, odp_port);
3107 return port && &ofproto->up == port->up.ofproto ? port : NULL;
3111 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
3112 struct ofproto_port *ofproto_port,
3113 struct dpif_port *dpif_port)
3115 ofproto_port->name = dpif_port->name;
3116 ofproto_port->type = dpif_port->type;
3117 ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
3120 static struct ofport_dpif *
3121 ofport_get_peer(const struct ofport_dpif *ofport_dpif)
3123 const struct ofproto_dpif *ofproto;
3126 peer = netdev_vport_patch_peer(ofport_dpif->up.netdev);
3131 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
3132 struct ofport *ofport;
3134 ofport = shash_find_data(&ofproto->up.port_by_name, peer);
3135 if (ofport && ofport->ofproto->ofproto_class == &ofproto_dpif_class) {
3136 return ofport_dpif_cast(ofport);
3143 port_run_fast(struct ofport_dpif *ofport)
3145 if (ofport->cfm && cfm_should_send_ccm(ofport->cfm)) {
3146 struct ofpbuf packet;
3148 ofpbuf_init(&packet, 0);
3149 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
3150 send_packet(ofport, &packet);
3151 ofpbuf_uninit(&packet);
3154 if (ofport->bfd && bfd_should_send_packet(ofport->bfd)) {
3155 struct ofpbuf packet;
3157 ofpbuf_init(&packet, 0);
3158 bfd_put_packet(ofport->bfd, &packet, ofport->up.pp.hw_addr);
3159 send_packet(ofport, &packet);
3160 ofpbuf_uninit(&packet);
3165 port_run(struct ofport_dpif *ofport)
3167 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
3168 bool carrier_changed = carrier_seq != ofport->carrier_seq;
3169 bool enable = netdev_get_carrier(ofport->up.netdev);
3171 ofport->carrier_seq = carrier_seq;
3173 port_run_fast(ofport);
3175 if (ofport->tnl_port
3176 && tnl_port_reconfigure(&ofport->up, ofport->odp_port,
3177 &ofport->tnl_port)) {
3178 ofproto_dpif_cast(ofport->up.ofproto)->backer->need_revalidate = true;
3182 int cfm_opup = cfm_get_opup(ofport->cfm);
3184 cfm_run(ofport->cfm);
3185 enable = enable && !cfm_get_fault(ofport->cfm);
3187 if (cfm_opup >= 0) {
3188 enable = enable && cfm_opup;
3193 bfd_run(ofport->bfd);
3194 enable = enable && bfd_forwarding(ofport->bfd);
3197 if (ofport->bundle) {
3198 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
3199 if (carrier_changed) {
3200 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
3204 if (ofport->may_enable != enable) {
3205 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3207 if (ofproto->has_bundle_action) {
3208 ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
3212 ofport->may_enable = enable;
3216 port_wait(struct ofport_dpif *ofport)
3219 cfm_wait(ofport->cfm);
3223 bfd_wait(ofport->bfd);
3228 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
3229 struct ofproto_port *ofproto_port)
3231 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3232 struct dpif_port dpif_port;
3235 if (sset_contains(&ofproto->ghost_ports, devname)) {
3236 const char *type = netdev_get_type_from_name(devname);
3238 /* We may be called before ofproto->up.port_by_name is populated with
3239 * the appropriate ofport. For this reason, we must get the name and
3240 * type from the netdev layer directly. */
3242 const struct ofport *ofport;
3244 ofport = shash_find_data(&ofproto->up.port_by_name, devname);
3245 ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
3246 ofproto_port->name = xstrdup(devname);
3247 ofproto_port->type = xstrdup(type);
3253 if (!sset_contains(&ofproto->ports, devname)) {
3256 error = dpif_port_query_by_name(ofproto->backer->dpif,
3257 devname, &dpif_port);
3259 ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
3265 port_add(struct ofproto *ofproto_, struct netdev *netdev)
3267 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3268 const char *dp_port_name = netdev_vport_get_dpif_port(netdev);
3269 const char *devname = netdev_get_name(netdev);
3271 if (netdev_vport_is_patch(netdev)) {
3272 sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
3276 if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
3277 uint32_t port_no = UINT32_MAX;
3280 error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
3284 if (netdev_get_tunnel_config(netdev)) {
3285 simap_put(&ofproto->backer->tnl_backers, dp_port_name, port_no);
3289 if (netdev_get_tunnel_config(netdev)) {
3290 sset_add(&ofproto->ghost_ports, devname);
3292 sset_add(&ofproto->ports, devname);
3298 port_del(struct ofproto *ofproto_, uint16_t ofp_port)
3300 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3301 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
3308 sset_find_and_delete(&ofproto->ghost_ports,
3309 netdev_get_name(ofport->up.netdev));
3310 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3311 if (!ofport->tnl_port) {
3312 error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
3314 /* The caller is going to close ofport->up.netdev. If this is a
3315 * bonded port, then the bond is using that netdev, so remove it
3316 * from the bond. The client will need to reconfigure everything
3317 * after deleting ports, so then the slave will get re-added. */
3318 bundle_remove(&ofport->up);
3325 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
3327 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3332 error = netdev_get_stats(ofport->up.netdev, stats);
3334 if (!error && ofport_->ofp_port == OFPP_LOCAL) {
3335 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3337 /* ofproto->stats.tx_packets represents packets that we created
3338 * internally and sent to some port (e.g. packets sent with
3339 * send_packet()). Account for them as if they had come from
3340 * OFPP_LOCAL and got forwarded. */
3342 if (stats->rx_packets != UINT64_MAX) {
3343 stats->rx_packets += ofproto->stats.tx_packets;
3346 if (stats->rx_bytes != UINT64_MAX) {
3347 stats->rx_bytes += ofproto->stats.tx_bytes;
3350 /* ofproto->stats.rx_packets represents packets that were received on
3351 * some port and we processed internally and dropped (e.g. STP).
3352 * Account for them as if they had been forwarded to OFPP_LOCAL. */
3354 if (stats->tx_packets != UINT64_MAX) {
3355 stats->tx_packets += ofproto->stats.rx_packets;
3358 if (stats->tx_bytes != UINT64_MAX) {
3359 stats->tx_bytes += ofproto->stats.rx_bytes;
3366 struct port_dump_state {
3371 struct ofproto_port port;
3376 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
3378 *statep = xzalloc(sizeof(struct port_dump_state));
3383 port_dump_next(const struct ofproto *ofproto_, void *state_,
3384 struct ofproto_port *port)
3386 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3387 struct port_dump_state *state = state_;
3388 const struct sset *sset;
3389 struct sset_node *node;
3391 if (state->has_port) {
3392 ofproto_port_destroy(&state->port);
3393 state->has_port = false;
3395 sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
3396 while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
3399 error = port_query_by_name(ofproto_, node->name, &state->port);
3401 *port = state->port;
3402 state->has_port = true;
3404 } else if (error != ENODEV) {
3409 if (!state->ghost) {
3410 state->ghost = true;
3413 return port_dump_next(ofproto_, state_, port);
3420 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3422 struct port_dump_state *state = state_;
3424 if (state->has_port) {
3425 ofproto_port_destroy(&state->port);
3432 port_poll(const struct ofproto *ofproto_, char **devnamep)
3434 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3436 if (ofproto->port_poll_errno) {
3437 int error = ofproto->port_poll_errno;
3438 ofproto->port_poll_errno = 0;
3442 if (sset_is_empty(&ofproto->port_poll_set)) {
3446 *devnamep = sset_pop(&ofproto->port_poll_set);
3451 port_poll_wait(const struct ofproto *ofproto_)
3453 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3454 dpif_port_poll_wait(ofproto->backer->dpif);
3458 port_is_lacp_current(const struct ofport *ofport_)
3460 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3461 return (ofport->bundle && ofport->bundle->lacp
3462 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3466 /* Upcall handling. */
3468 /* Flow miss batching.
3470 * Some dpifs implement operations faster when you hand them off in a batch.
3471 * To allow batching, "struct flow_miss" queues the dpif-related work needed
3472 * for a given flow. Each "struct flow_miss" corresponds to sending one or
3473 * more packets, plus possibly installing the flow in the dpif.
3475 * So far we only batch the operations that affect flow setup time the most.
3476 * It's possible to batch more than that, but the benefit might be minimal. */
3478 struct hmap_node hmap_node;
3479 struct ofproto_dpif *ofproto;
3481 enum odp_key_fitness key_fitness;
3482 const struct nlattr *key;
3484 struct initial_vals initial_vals;
3485 struct list packets;
3486 enum dpif_upcall_type upcall_type;
3489 struct flow_miss_op {
3490 struct dpif_op dpif_op;
3492 uint64_t slow_stub[128 / 8]; /* Buffer for compose_slow_path() */
3493 struct xlate_out xout;
3494 bool xout_garbage; /* 'xout' needs to be uninitialized? */
3497 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
3498 * OpenFlow controller as necessary according to their individual
3499 * configurations. */
3501 send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet,
3502 const struct flow *flow)
3504 struct ofputil_packet_in pin;
3506 pin.packet = packet->data;
3507 pin.packet_len = packet->size;
3508 pin.reason = OFPR_NO_MATCH;
3509 pin.controller_id = 0;
3514 pin.send_len = 0; /* not used for flow table misses */
3516 flow_get_metadata(flow, &pin.fmd);
3518 connmgr_send_packet_in(ofproto->up.connmgr, &pin);
3521 static enum slow_path_reason
3522 process_special(struct ofproto_dpif *ofproto, const struct flow *flow,
3523 const struct ofport_dpif *ofport, const struct ofpbuf *packet)
3527 } else if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) {
3529 cfm_process_heartbeat(ofport->cfm, packet);
3532 } else if (ofport->bfd && bfd_should_process_flow(flow)) {
3534 bfd_process_packet(ofport->bfd, flow, packet);
3537 } else if (ofport->bundle && ofport->bundle->lacp
3538 && flow->dl_type == htons(ETH_TYPE_LACP)) {
3540 lacp_process_packet(ofport->bundle->lacp, ofport, packet);
3543 } else if (ofproto->stp && stp_should_process_flow(flow)) {
3545 stp_process_packet(ofport, packet);
3553 static struct flow_miss *
3554 flow_miss_find(struct hmap *todo, const struct ofproto_dpif *ofproto,
3555 const struct flow *flow, uint32_t hash)
3557 struct flow_miss *miss;
3559 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
3560 if (miss->ofproto == ofproto && flow_equal(&miss->flow, flow)) {
3568 /* Partially Initializes 'op' as an "execute" operation for 'miss' and
3569 * 'packet'. The caller must initialize op->actions and op->actions_len. If
3570 * 'miss' is associated with a subfacet the caller must also initialize the
3571 * returned op->subfacet, and if anything needs to be freed after processing
3572 * the op, the caller must initialize op->garbage also. */
3574 init_flow_miss_execute_op(struct flow_miss *miss, struct ofpbuf *packet,
3575 struct flow_miss_op *op)
3577 if (miss->flow.vlan_tci != miss->initial_vals.vlan_tci) {
3578 /* This packet was received on a VLAN splinter port. We
3579 * added a VLAN to the packet to make the packet resemble
3580 * the flow, but the actions were composed assuming that
3581 * the packet contained no VLAN. So, we must remove the
3582 * VLAN header from the packet before trying to execute the
3584 eth_pop_vlan(packet);
3587 op->xout_garbage = false;
3588 op->dpif_op.type = DPIF_OP_EXECUTE;
3589 op->dpif_op.u.execute.key = miss->key;
3590 op->dpif_op.u.execute.key_len = miss->key_len;
3591 op->dpif_op.u.execute.packet = packet;
3594 /* Helper for handle_flow_miss_without_facet() and
3595 * handle_flow_miss_with_facet(). */
3597 handle_flow_miss_common(struct rule_dpif *rule,
3598 struct ofpbuf *packet, const struct flow *flow)
3600 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3602 if (rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
3604 * Extra-special case for fail-open mode.
3606 * We are in fail-open mode and the packet matched the fail-open
3607 * rule, but we are connected to a controller too. We should send
3608 * the packet up to the controller in the hope that it will try to
3609 * set up a flow and thereby allow us to exit fail-open.
3611 * See the top-level comment in fail-open.c for more information.
3613 send_packet_in_miss(ofproto, packet, flow);
3617 /* Figures out whether a flow that missed in 'ofproto', whose details are in
3618 * 'miss', is likely to be worth tracking in detail in userspace and (usually)
3619 * installing a datapath flow. The answer is usually "yes" (a return value of
3620 * true). However, for short flows the cost of bookkeeping is much higher than
3621 * the benefits, so when the datapath holds a large number of flows we impose
3622 * some heuristics to decide which flows are likely to be worth tracking. */
3624 flow_miss_should_make_facet(struct ofproto_dpif *ofproto,
3625 struct flow_miss *miss, uint32_t hash)
3627 if (!ofproto->governor) {
3630 n_subfacets = hmap_count(&ofproto->subfacets);
3631 if (n_subfacets * 2 <= ofproto->up.flow_eviction_threshold) {
3635 ofproto->governor = governor_create(ofproto->up.name);
3638 return governor_should_install_flow(ofproto->governor, hash,
3639 list_size(&miss->packets));
3642 /* Handles 'miss', which matches 'rule', without creating a facet or subfacet
3643 * or creating any datapath flow. May add an "execute" operation to 'ops' and
3644 * increment '*n_ops'. */
3646 handle_flow_miss_without_facet(struct flow_miss *miss,
3647 struct flow_miss_op *ops, size_t *n_ops)
3649 struct rule_dpif *rule = rule_dpif_lookup(miss->ofproto, &miss->flow);
3650 long long int now = time_msec();
3651 struct ofpbuf *packet;
3652 struct xlate_in xin;
3654 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3655 struct flow_miss_op *op = &ops[*n_ops];
3656 struct dpif_flow_stats stats;
3658 COVERAGE_INC(facet_suppress);
3660 handle_flow_miss_common(rule, packet, &miss->flow);
3662 dpif_flow_stats_extract(&miss->flow, packet, now, &stats);
3663 rule_credit_stats(rule, &stats);
3665 xlate_in_init(&xin, miss->ofproto, &miss->flow, &miss->initial_vals,
3666 rule, stats.tcp_flags, packet);
3667 xin.resubmit_stats = &stats;
3668 xlate_actions(&xin, &op->xout);
3670 if (op->xout.odp_actions.size) {
3671 struct dpif_execute *execute = &op->dpif_op.u.execute;
3673 init_flow_miss_execute_op(miss, packet, op);
3674 execute->actions = op->xout.odp_actions.data;
3675 execute->actions_len = op->xout.odp_actions.size;
3676 op->xout_garbage = true;
3680 xlate_out_uninit(&op->xout);
3685 /* Handles 'miss', which matches 'facet'. May add any required datapath
3686 * operations to 'ops', incrementing '*n_ops' for each new op.
3688 * All of the packets in 'miss' are considered to have arrived at time 'now'.
3689 * This is really important only for new facets: if we just called time_msec()
3690 * here, then the new subfacet or its packets could look (occasionally) as
3691 * though it was used some time after the facet was used. That can make a
3692 * one-packet flow look like it has a nonzero duration, which looks odd in
3693 * e.g. NetFlow statistics. */
3695 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
3697 struct flow_miss_op *ops, size_t *n_ops)
3699 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3700 enum subfacet_path want_path;
3701 struct subfacet *subfacet;
3702 struct ofpbuf *packet;
3704 subfacet = subfacet_create(facet, miss, now);
3705 want_path = subfacet->facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
3707 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3708 struct flow_miss_op *op = &ops[*n_ops];
3709 struct dpif_flow_stats stats;
3711 handle_flow_miss_common(facet->rule, packet, &miss->flow);
3713 if (want_path != SF_FAST_PATH) {
3714 struct xlate_in xin;
3716 xlate_in_init(&xin, ofproto, &facet->flow, &facet->initial_vals,
3717 facet->rule, 0, packet);
3718 xlate_actions_for_side_effects(&xin);
3721 dpif_flow_stats_extract(&facet->flow, packet, now, &stats);
3722 subfacet_update_stats(subfacet, &stats);
3724 if (facet->xout.odp_actions.size) {
3725 struct dpif_execute *execute = &op->dpif_op.u.execute;
3727 init_flow_miss_execute_op(miss, packet, op);
3728 execute->actions = facet->xout.odp_actions.data,
3729 execute->actions_len = facet->xout.odp_actions.size;
3734 if (miss->upcall_type == DPIF_UC_MISS || subfacet->path != want_path) {
3735 struct flow_miss_op *op = &ops[(*n_ops)++];
3736 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
3738 subfacet->path = want_path;
3740 op->xout_garbage = false;
3741 op->dpif_op.type = DPIF_OP_FLOW_PUT;
3742 put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
3743 put->key = miss->key;
3744 put->key_len = miss->key_len;
3745 if (want_path == SF_FAST_PATH) {
3746 put->actions = facet->xout.odp_actions.data;
3747 put->actions_len = facet->xout.odp_actions.size;
3749 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
3750 op->slow_stub, sizeof op->slow_stub,
3751 &put->actions, &put->actions_len);
3757 /* Handles flow miss 'miss'. May add any required datapath operations
3758 * to 'ops', incrementing '*n_ops' for each new op. */
3760 handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops,
3763 struct ofproto_dpif *ofproto = miss->ofproto;
3764 struct facet *facet;
3768 /* The caller must ensure that miss->hmap_node.hash contains
3769 * flow_hash(miss->flow, 0). */
3770 hash = miss->hmap_node.hash;
3772 facet = facet_lookup_valid(ofproto, &miss->flow, hash);
3774 /* There does not exist a bijection between 'struct flow' and datapath
3775 * flow keys with fitness ODP_FIT_TO_LITTLE. This breaks a fundamental
3776 * assumption used throughout the facet and subfacet handling code.
3777 * Since we have to handle these misses in userspace anyway, we simply
3778 * skip facet creation, avoiding the problem alltogether. */
3779 if (miss->key_fitness == ODP_FIT_TOO_LITTLE
3780 || !flow_miss_should_make_facet(ofproto, miss, hash)) {
3781 handle_flow_miss_without_facet(miss, ops, n_ops);
3785 facet = facet_create(miss, hash);
3790 handle_flow_miss_with_facet(miss, facet, now, ops, n_ops);
3793 static struct drop_key *
3794 drop_key_lookup(const struct dpif_backer *backer, const struct nlattr *key,
3797 struct drop_key *drop_key;
3799 HMAP_FOR_EACH_WITH_HASH (drop_key, hmap_node, hash_bytes(key, key_len, 0),
3800 &backer->drop_keys) {
3801 if (drop_key->key_len == key_len
3802 && !memcmp(drop_key->key, key, key_len)) {
3810 drop_key_clear(struct dpif_backer *backer)
3812 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3813 struct drop_key *drop_key, *next;
3815 HMAP_FOR_EACH_SAFE (drop_key, next, hmap_node, &backer->drop_keys) {
3818 error = dpif_flow_del(backer->dpif, drop_key->key, drop_key->key_len,
3820 if (error && !VLOG_DROP_WARN(&rl)) {
3821 struct ds ds = DS_EMPTY_INITIALIZER;
3822 odp_flow_key_format(drop_key->key, drop_key->key_len, &ds);
3823 VLOG_WARN("Failed to delete drop key (%s) (%s)", strerror(error),
3828 hmap_remove(&backer->drop_keys, &drop_key->hmap_node);
3829 free(drop_key->key);
3834 /* Given a datpath, packet, and flow metadata ('backer', 'packet', and 'key'
3835 * respectively), populates 'flow' with the result of odp_flow_key_to_flow().
3836 * Optionally, if nonnull, populates 'fitnessp' with the fitness of 'flow' as
3837 * returned by odp_flow_key_to_flow(). Also, optionally populates 'ofproto'
3838 * with the ofproto_dpif, and 'odp_in_port' with the datapath in_port, that
3839 * 'packet' ingressed.
3841 * If 'ofproto' is nonnull, requires 'flow''s in_port to exist. Otherwise sets
3842 * 'flow''s in_port to OFPP_NONE.
3844 * This function does post-processing on data returned from
3845 * odp_flow_key_to_flow() to help make VLAN splinters transparent to the rest
3846 * of the upcall processing logic. In particular, if the extracted in_port is
3847 * a VLAN splinter port, it replaces flow->in_port by the "real" port, sets
3848 * flow->vlan_tci correctly for the VLAN of the VLAN splinter port, and pushes
3849 * a VLAN header onto 'packet' (if it is nonnull).
3851 * Optionally, if 'initial_vals' is nonnull, sets 'initial_vals->vlan_tci'
3852 * to the VLAN TCI with which the packet was really received, that is, the
3853 * actual VLAN TCI extracted by odp_flow_key_to_flow(). (This differs from
3854 * the value returned in flow->vlan_tci only for packets received on
3857 * Similarly, this function also includes some logic to help with tunnels. It
3858 * may modify 'flow' as necessary to make the tunneling implementation
3859 * transparent to the upcall processing logic.
3861 * Returns 0 if successful, ENODEV if the parsed flow has no associated ofport,
3862 * or some other positive errno if there are other problems. */
3864 ofproto_receive(const struct dpif_backer *backer, struct ofpbuf *packet,
3865 const struct nlattr *key, size_t key_len,
3866 struct flow *flow, enum odp_key_fitness *fitnessp,
3867 struct ofproto_dpif **ofproto, uint32_t *odp_in_port,
3868 struct initial_vals *initial_vals)
3870 const struct ofport_dpif *port;
3871 enum odp_key_fitness fitness;
3874 fitness = odp_flow_key_to_flow(key, key_len, flow);
3875 if (fitness == ODP_FIT_ERROR) {
3881 initial_vals->vlan_tci = flow->vlan_tci;
3885 *odp_in_port = flow->in_port;
3888 port = (tnl_port_should_receive(flow)
3889 ? ofport_dpif_cast(tnl_port_receive(flow))
3890 : odp_port_to_ofport(backer, flow->in_port));
3891 flow->in_port = port ? port->up.ofp_port : OFPP_NONE;
3896 /* XXX: Since the tunnel module is not scoped per backer, for a tunnel port
3897 * it's theoretically possible that we'll receive an ofport belonging to an
3898 * entirely different datapath. In practice, this can't happen because no
3899 * platforms has two separate datapaths which each support tunneling. */
3900 ovs_assert(ofproto_dpif_cast(port->up.ofproto)->backer == backer);
3902 if (vsp_adjust_flow(ofproto_dpif_cast(port->up.ofproto), flow)) {
3904 /* Make the packet resemble the flow, so that it gets sent to
3905 * an OpenFlow controller properly, so that it looks correct
3906 * for sFlow, and so that flow_extract() will get the correct
3907 * vlan_tci if it is called on 'packet'.
3909 * The allocated space inside 'packet' probably also contains
3910 * 'key', that is, both 'packet' and 'key' are probably part of
3911 * a struct dpif_upcall (see the large comment on that
3912 * structure definition), so pushing data on 'packet' is in
3913 * general not a good idea since it could overwrite 'key' or
3914 * free it as a side effect. However, it's OK in this special
3915 * case because we know that 'packet' is inside a Netlink
3916 * attribute: pushing 4 bytes will just overwrite the 4-byte
3917 * "struct nlattr", which is fine since we don't need that
3918 * header anymore. */
3919 eth_push_vlan(packet, flow->vlan_tci);
3921 /* We can't reproduce 'key' from 'flow'. */
3922 fitness = fitness == ODP_FIT_PERFECT ? ODP_FIT_TOO_MUCH : fitness;
3927 *ofproto = ofproto_dpif_cast(port->up.ofproto);
3932 *fitnessp = fitness;
3938 handle_miss_upcalls(struct dpif_backer *backer, struct dpif_upcall *upcalls,
3941 struct dpif_upcall *upcall;
3942 struct flow_miss *miss;
3943 struct flow_miss misses[FLOW_MISS_MAX_BATCH];
3944 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
3945 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
3955 /* Construct the to-do list.
3957 * This just amounts to extracting the flow from each packet and sticking
3958 * the packets that have the same flow in the same "flow_miss" structure so
3959 * that we can process them together. */
3962 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
3963 struct flow_miss *miss = &misses[n_misses];
3964 struct flow_miss *existing_miss;
3965 struct ofproto_dpif *ofproto;
3966 uint32_t odp_in_port;
3971 error = ofproto_receive(backer, upcall->packet, upcall->key,
3972 upcall->key_len, &flow, &miss->key_fitness,
3973 &ofproto, &odp_in_port, &miss->initial_vals);
3974 if (error == ENODEV) {
3975 struct drop_key *drop_key;
3977 /* Received packet on port for which we couldn't associate
3978 * an ofproto. This can happen if a port is removed while
3979 * traffic is being received. Print a rate-limited message
3980 * in case it happens frequently. Install a drop flow so
3981 * that future packets of the flow are inexpensively dropped
3983 VLOG_INFO_RL(&rl, "received packet on unassociated port %"PRIu32,
3986 drop_key = drop_key_lookup(backer, upcall->key, upcall->key_len);
3988 drop_key = xmalloc(sizeof *drop_key);
3989 drop_key->key = xmemdup(upcall->key, upcall->key_len);
3990 drop_key->key_len = upcall->key_len;
3992 hmap_insert(&backer->drop_keys, &drop_key->hmap_node,
3993 hash_bytes(drop_key->key, drop_key->key_len, 0));
3994 dpif_flow_put(backer->dpif, DPIF_FP_CREATE | DPIF_FP_MODIFY,
3995 drop_key->key, drop_key->key_len, NULL, 0, NULL);
4003 ofproto->n_missed++;
4004 flow_extract(upcall->packet, flow.skb_priority, flow.skb_mark,
4005 &flow.tunnel, flow.in_port, &miss->flow);
4007 /* Add other packets to a to-do list. */
4008 hash = flow_hash(&miss->flow, 0);
4009 existing_miss = flow_miss_find(&todo, ofproto, &miss->flow, hash);
4010 if (!existing_miss) {
4011 hmap_insert(&todo, &miss->hmap_node, hash);
4012 miss->ofproto = ofproto;
4013 miss->key = upcall->key;
4014 miss->key_len = upcall->key_len;
4015 miss->upcall_type = upcall->type;
4016 list_init(&miss->packets);
4020 miss = existing_miss;
4022 list_push_back(&miss->packets, &upcall->packet->list_node);
4025 /* Process each element in the to-do list, constructing the set of
4026 * operations to batch. */
4028 HMAP_FOR_EACH (miss, hmap_node, &todo) {
4029 handle_flow_miss(miss, flow_miss_ops, &n_ops);
4031 ovs_assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
4033 /* Execute batch. */
4034 for (i = 0; i < n_ops; i++) {
4035 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
4037 dpif_operate(backer->dpif, dpif_ops, n_ops);
4040 for (i = 0; i < n_ops; i++) {
4041 if (flow_miss_ops[i].xout_garbage) {
4042 xlate_out_uninit(&flow_miss_ops[i].xout);
4045 hmap_destroy(&todo);
4048 static enum { SFLOW_UPCALL, MISS_UPCALL, BAD_UPCALL, FLOW_SAMPLE_UPCALL,
4050 classify_upcall(const struct dpif_upcall *upcall)
4052 size_t userdata_len;
4053 union user_action_cookie cookie;
4055 /* First look at the upcall type. */
4056 switch (upcall->type) {
4057 case DPIF_UC_ACTION:
4063 case DPIF_N_UC_TYPES:
4065 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
4069 /* "action" upcalls need a closer look. */
4070 if (!upcall->userdata) {
4071 VLOG_WARN_RL(&rl, "action upcall missing cookie");
4074 userdata_len = nl_attr_get_size(upcall->userdata);
4075 if (userdata_len < sizeof cookie.type
4076 || userdata_len > sizeof cookie) {
4077 VLOG_WARN_RL(&rl, "action upcall cookie has unexpected size %zu",
4081 memset(&cookie, 0, sizeof cookie);
4082 memcpy(&cookie, nl_attr_get(upcall->userdata), userdata_len);
4083 if (userdata_len == sizeof cookie.sflow
4084 && cookie.type == USER_ACTION_COOKIE_SFLOW) {
4085 return SFLOW_UPCALL;
4086 } else if (userdata_len == sizeof cookie.slow_path
4087 && cookie.type == USER_ACTION_COOKIE_SLOW_PATH) {
4089 } else if (userdata_len == sizeof cookie.flow_sample
4090 && cookie.type == USER_ACTION_COOKIE_FLOW_SAMPLE) {
4091 return FLOW_SAMPLE_UPCALL;
4092 } else if (userdata_len == sizeof cookie.ipfix
4093 && cookie.type == USER_ACTION_COOKIE_IPFIX) {
4094 return IPFIX_UPCALL;
4096 VLOG_WARN_RL(&rl, "invalid user cookie of type %"PRIu16
4097 " and size %zu", cookie.type, userdata_len);
4103 handle_sflow_upcall(struct dpif_backer *backer,
4104 const struct dpif_upcall *upcall)
4106 struct ofproto_dpif *ofproto;
4107 union user_action_cookie cookie;
4109 uint32_t odp_in_port;
4111 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
4112 &flow, NULL, &ofproto, &odp_in_port, NULL)
4113 || !ofproto->sflow) {
4117 memset(&cookie, 0, sizeof cookie);
4118 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.sflow);
4119 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
4120 odp_in_port, &cookie);
4124 handle_flow_sample_upcall(struct dpif_backer *backer,
4125 const struct dpif_upcall *upcall)
4127 struct ofproto_dpif *ofproto;
4128 union user_action_cookie cookie;
4131 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
4132 &flow, NULL, &ofproto, NULL, NULL)
4133 || !ofproto->ipfix) {
4137 memset(&cookie, 0, sizeof cookie);
4138 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.flow_sample);
4140 /* The flow reflects exactly the contents of the packet. Sample
4141 * the packet using it. */
4142 dpif_ipfix_flow_sample(ofproto->ipfix, upcall->packet, &flow,
4143 cookie.flow_sample.collector_set_id,
4144 cookie.flow_sample.probability,
4145 cookie.flow_sample.obs_domain_id,
4146 cookie.flow_sample.obs_point_id);
4150 handle_ipfix_upcall(struct dpif_backer *backer,
4151 const struct dpif_upcall *upcall)
4153 struct ofproto_dpif *ofproto;
4156 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
4157 &flow, NULL, &ofproto, NULL, NULL)
4158 || !ofproto->ipfix) {
4162 /* The flow reflects exactly the contents of the packet. Sample
4163 * the packet using it. */
4164 dpif_ipfix_bridge_sample(ofproto->ipfix, upcall->packet, &flow);
4168 handle_upcalls(struct dpif_backer *backer, unsigned int max_batch)
4170 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
4171 struct ofpbuf miss_bufs[FLOW_MISS_MAX_BATCH];
4172 uint64_t miss_buf_stubs[FLOW_MISS_MAX_BATCH][4096 / 8];
4177 ovs_assert(max_batch <= FLOW_MISS_MAX_BATCH);
4180 for (n_processed = 0; n_processed < max_batch; n_processed++) {
4181 struct dpif_upcall *upcall = &misses[n_misses];
4182 struct ofpbuf *buf = &miss_bufs[n_misses];
4185 ofpbuf_use_stub(buf, miss_buf_stubs[n_misses],
4186 sizeof miss_buf_stubs[n_misses]);
4187 error = dpif_recv(backer->dpif, upcall, buf);
4193 switch (classify_upcall(upcall)) {
4195 /* Handle it later. */
4200 handle_sflow_upcall(backer, upcall);
4204 case FLOW_SAMPLE_UPCALL:
4205 handle_flow_sample_upcall(backer, upcall);
4210 handle_ipfix_upcall(backer, upcall);
4220 /* Handle deferred MISS_UPCALL processing. */
4221 handle_miss_upcalls(backer, misses, n_misses);
4222 for (i = 0; i < n_misses; i++) {
4223 ofpbuf_uninit(&miss_bufs[i]);
4229 /* Flow expiration. */
4231 static int subfacet_max_idle(const struct ofproto_dpif *);
4232 static void update_stats(struct dpif_backer *);
4233 static void rule_expire(struct rule_dpif *);
4234 static void expire_subfacets(struct ofproto_dpif *, int dp_max_idle);
4236 /* This function is called periodically by run(). Its job is to collect
4237 * updates for the flows that have been installed into the datapath, most
4238 * importantly when they last were used, and then use that information to
4239 * expire flows that have not been used recently.
4241 * Returns the number of milliseconds after which it should be called again. */
4243 expire(struct dpif_backer *backer)
4245 struct ofproto_dpif *ofproto;
4246 int max_idle = INT32_MAX;
4248 /* Periodically clear out the drop keys in an effort to keep them
4249 * relatively few. */
4250 drop_key_clear(backer);
4252 /* Update stats for each flow in the backer. */
4253 update_stats(backer);
4255 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4256 struct rule *rule, *next_rule;
4259 if (ofproto->backer != backer) {
4263 /* Keep track of the max number of flows per ofproto_dpif. */
4264 update_max_subfacet_count(ofproto);
4266 /* Expire subfacets that have been idle too long. */
4267 dp_max_idle = subfacet_max_idle(ofproto);
4268 expire_subfacets(ofproto, dp_max_idle);
4270 max_idle = MIN(max_idle, dp_max_idle);
4272 /* Expire OpenFlow flows whose idle_timeout or hard_timeout
4274 LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
4275 &ofproto->up.expirable) {
4276 rule_expire(rule_dpif_cast(rule));
4279 /* All outstanding data in existing flows has been accounted, so it's a
4280 * good time to do bond rebalancing. */
4281 if (ofproto->has_bonded_bundles) {
4282 struct ofbundle *bundle;
4284 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
4286 bond_rebalance(bundle->bond, &backer->revalidate_set);
4292 return MIN(max_idle, 1000);
4295 /* Updates flow table statistics given that the datapath just reported 'stats'
4296 * as 'subfacet''s statistics. */
4298 update_subfacet_stats(struct subfacet *subfacet,
4299 const struct dpif_flow_stats *stats)
4301 struct facet *facet = subfacet->facet;
4302 struct dpif_flow_stats diff;
4304 diff.tcp_flags = stats->tcp_flags;
4305 diff.used = stats->used;
4307 if (stats->n_packets >= subfacet->dp_packet_count) {
4308 diff.n_packets = stats->n_packets - subfacet->dp_packet_count;
4310 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
4314 if (stats->n_bytes >= subfacet->dp_byte_count) {
4315 diff.n_bytes = stats->n_bytes - subfacet->dp_byte_count;
4317 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
4321 subfacet->dp_packet_count = stats->n_packets;
4322 subfacet->dp_byte_count = stats->n_bytes;
4323 subfacet_update_stats(subfacet, &diff);
4325 if (facet->accounted_bytes < facet->byte_count) {
4327 facet_account(facet);
4328 facet->accounted_bytes = facet->byte_count;
4332 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
4333 * about, or a flow that shouldn't be installed but was anyway. Delete it. */
4335 delete_unexpected_flow(struct ofproto_dpif *ofproto,
4336 const struct nlattr *key, size_t key_len)
4338 if (!VLOG_DROP_WARN(&rl)) {
4342 odp_flow_key_format(key, key_len, &s);
4343 VLOG_WARN("unexpected flow on %s: %s", ofproto->up.name, ds_cstr(&s));
4347 COVERAGE_INC(facet_unexpected);
4348 dpif_flow_del(ofproto->backer->dpif, key, key_len, NULL);
4351 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
4353 * This function also pushes statistics updates to rules which each facet
4354 * resubmits into. Generally these statistics will be accurate. However, if a
4355 * facet changes the rule it resubmits into at some time in between
4356 * update_stats() runs, it is possible that statistics accrued to the
4357 * old rule will be incorrectly attributed to the new rule. This could be
4358 * avoided by calling update_stats() whenever rules are created or
4359 * deleted. However, the performance impact of making so many calls to the
4360 * datapath do not justify the benefit of having perfectly accurate statistics.
4362 * In addition, this function maintains per ofproto flow hit counts. The patch
4363 * port is not treated specially. e.g. A packet ingress from br0 patched into
4364 * br1 will increase the hit count of br0 by 1, however, does not affect
4365 * the hit or miss counts of br1.
4368 update_stats(struct dpif_backer *backer)
4370 const struct dpif_flow_stats *stats;
4371 struct dpif_flow_dump dump;
4372 const struct nlattr *key;
4373 struct ofproto_dpif *ofproto;
4376 dpif_flow_dump_start(&dump, backer->dpif);
4377 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
4379 struct subfacet *subfacet;
4382 if (ofproto_receive(backer, NULL, key, key_len, &flow, NULL, &ofproto,
4387 ofproto->total_subfacet_count += hmap_count(&ofproto->subfacets);
4388 ofproto->n_update_stats++;
4390 key_hash = odp_flow_key_hash(key, key_len);
4391 subfacet = subfacet_find(ofproto, key, key_len, key_hash);
4392 switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
4394 /* Update ofproto_dpif's hit count. */
4395 if (stats->n_packets > subfacet->dp_packet_count) {
4396 uint64_t delta = stats->n_packets - subfacet->dp_packet_count;
4397 dpif_stats_update_hit_count(ofproto, delta);
4400 update_subfacet_stats(subfacet, stats);
4404 /* Stats are updated per-packet. */
4407 case SF_NOT_INSTALLED:
4409 delete_unexpected_flow(ofproto, key, key_len);
4414 dpif_flow_dump_done(&dump);
4416 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4417 update_moving_averages(ofproto);
4422 /* Calculates and returns the number of milliseconds of idle time after which
4423 * subfacets should expire from the datapath. When a subfacet expires, we fold
4424 * its statistics into its facet, and when a facet's last subfacet expires, we
4425 * fold its statistic into its rule. */
4427 subfacet_max_idle(const struct ofproto_dpif *ofproto)
4430 * Idle time histogram.
4432 * Most of the time a switch has a relatively small number of subfacets.
4433 * When this is the case we might as well keep statistics for all of them
4434 * in userspace and to cache them in the kernel datapath for performance as
4437 * As the number of subfacets increases, the memory required to maintain
4438 * statistics about them in userspace and in the kernel becomes
4439 * significant. However, with a large number of subfacets it is likely
4440 * that only a few of them are "heavy hitters" that consume a large amount
4441 * of bandwidth. At this point, only heavy hitters are worth caching in
4442 * the kernel and maintaining in userspaces; other subfacets we can
4445 * The technique used to compute the idle time is to build a histogram with
4446 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
4447 * that is installed in the kernel gets dropped in the appropriate bucket.
4448 * After the histogram has been built, we compute the cutoff so that only
4449 * the most-recently-used 1% of subfacets (but at least
4450 * ofproto->up.flow_eviction_threshold flows) are kept cached. At least
4451 * the most-recently-used bucket of subfacets is kept, so actually an
4452 * arbitrary number of subfacets can be kept in any given expiration run
4453 * (though the next run will delete most of those unless they receive
4456 * This requires a second pass through the subfacets, in addition to the
4457 * pass made by update_stats(), because the former function never looks at
4458 * uninstallable subfacets.
4460 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
4461 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
4462 int buckets[N_BUCKETS] = { 0 };
4463 int total, subtotal, bucket;
4464 struct subfacet *subfacet;
4468 total = hmap_count(&ofproto->subfacets);
4469 if (total <= ofproto->up.flow_eviction_threshold) {
4470 return N_BUCKETS * BUCKET_WIDTH;
4473 /* Build histogram. */
4475 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
4476 long long int idle = now - subfacet->used;
4477 int bucket = (idle <= 0 ? 0
4478 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
4479 : (unsigned int) idle / BUCKET_WIDTH);
4483 /* Find the first bucket whose flows should be expired. */
4484 subtotal = bucket = 0;
4486 subtotal += buckets[bucket++];
4487 } while (bucket < N_BUCKETS &&
4488 subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100));
4490 if (VLOG_IS_DBG_ENABLED()) {
4494 ds_put_cstr(&s, "keep");
4495 for (i = 0; i < N_BUCKETS; i++) {
4497 ds_put_cstr(&s, ", drop");
4500 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
4503 VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s));
4507 return bucket * BUCKET_WIDTH;
4511 expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle)
4513 /* Cutoff time for most flows. */
4514 long long int normal_cutoff = time_msec() - dp_max_idle;
4516 /* We really want to keep flows for special protocols around, so use a more
4517 * conservative cutoff. */
4518 long long int special_cutoff = time_msec() - 10000;
4520 struct subfacet *subfacet, *next_subfacet;
4521 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
4525 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
4526 &ofproto->subfacets) {
4527 long long int cutoff;
4529 cutoff = (subfacet->facet->xout.slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP
4533 if (subfacet->used < cutoff) {
4534 if (subfacet->path != SF_NOT_INSTALLED) {
4535 batch[n_batch++] = subfacet;
4536 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
4537 subfacet_destroy_batch(ofproto, batch, n_batch);
4541 subfacet_destroy(subfacet);
4547 subfacet_destroy_batch(ofproto, batch, n_batch);
4551 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
4552 * then delete it entirely. */
4554 rule_expire(struct rule_dpif *rule)
4556 struct facet *facet, *next_facet;
4560 if (rule->up.pending) {
4561 /* We'll have to expire it later. */
4565 /* Has 'rule' expired? */
4567 if (rule->up.hard_timeout
4568 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
4569 reason = OFPRR_HARD_TIMEOUT;
4570 } else if (rule->up.idle_timeout
4571 && now > rule->up.used + rule->up.idle_timeout * 1000) {
4572 reason = OFPRR_IDLE_TIMEOUT;
4577 COVERAGE_INC(ofproto_dpif_expired);
4579 /* Update stats. (This is a no-op if the rule expired due to an idle
4580 * timeout, because that only happens when the rule has no facets left.) */
4581 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4582 facet_remove(facet);
4585 /* Get rid of the rule. */
4586 ofproto_rule_expire(&rule->up, reason);
4591 /* Creates and returns a new facet based on 'miss'.
4593 * The caller must already have determined that no facet with an identical
4594 * 'miss->flow' exists in 'miss->ofproto'.
4596 * 'hash' must be the return value of flow_hash(miss->flow, 0).
4598 * The facet will initially have no subfacets. The caller should create (at
4599 * least) one subfacet with subfacet_create(). */
4600 static struct facet *
4601 facet_create(const struct flow_miss *miss, uint32_t hash)
4603 struct ofproto_dpif *ofproto = miss->ofproto;
4604 struct xlate_in xin;
4605 struct facet *facet;
4607 facet = xzalloc(sizeof *facet);
4608 facet->used = time_msec();
4609 facet->flow = miss->flow;
4610 facet->initial_vals = miss->initial_vals;
4611 facet->rule = rule_dpif_lookup(ofproto, &facet->flow);
4612 facet->learn_rl = time_msec() + 500;
4614 hmap_insert(&ofproto->facets, &facet->hmap_node, hash);
4615 list_push_back(&facet->rule->facets, &facet->list_node);
4616 list_init(&facet->subfacets);
4617 netflow_flow_init(&facet->nf_flow);
4618 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
4620 xlate_in_init(&xin, ofproto, &facet->flow, &facet->initial_vals,
4621 facet->rule, 0, NULL);
4622 xin.may_learn = true;
4623 xlate_actions(&xin, &facet->xout);
4624 facet->nf_flow.output_iface = facet->xout.nf_output_iface;
4630 facet_free(struct facet *facet)
4633 xlate_out_uninit(&facet->xout);
4638 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
4639 * 'packet', which arrived on 'in_port'. */
4641 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
4642 const struct nlattr *odp_actions, size_t actions_len,
4643 struct ofpbuf *packet)
4645 struct odputil_keybuf keybuf;
4649 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4650 odp_flow_key_from_flow(&key, flow,
4651 ofp_port_to_odp_port(ofproto, flow->in_port));
4653 error = dpif_execute(ofproto->backer->dpif, key.data, key.size,
4654 odp_actions, actions_len, packet);
4658 /* Remove 'facet' from 'ofproto' and free up the associated memory:
4660 * - If 'facet' was installed in the datapath, uninstalls it and updates its
4661 * rule's statistics, via subfacet_uninstall().
4663 * - Removes 'facet' from its rule and from ofproto->facets.
4666 facet_remove(struct facet *facet)
4668 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4669 struct subfacet *subfacet, *next_subfacet;
4671 ovs_assert(!list_is_empty(&facet->subfacets));
4673 /* First uninstall all of the subfacets to get final statistics. */
4674 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4675 subfacet_uninstall(subfacet);
4678 /* Flush the final stats to the rule.
4680 * This might require us to have at least one subfacet around so that we
4681 * can use its actions for accounting in facet_account(), which is why we
4682 * have uninstalled but not yet destroyed the subfacets. */
4683 facet_flush_stats(facet);
4685 /* Now we're really all done so destroy everything. */
4686 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
4687 &facet->subfacets) {
4688 subfacet_destroy__(subfacet);
4690 hmap_remove(&ofproto->facets, &facet->hmap_node);
4691 list_remove(&facet->list_node);
4695 /* Feed information from 'facet' back into the learning table to keep it in
4696 * sync with what is actually flowing through the datapath. */
4698 facet_learn(struct facet *facet)
4700 long long int now = time_msec();
4702 if (!facet->xout.has_fin_timeout && now < facet->learn_rl) {
4706 facet->learn_rl = now + 500;
4708 if (!facet->xout.has_learn
4709 && !facet->xout.has_normal
4710 && (!facet->xout.has_fin_timeout
4711 || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
4715 facet_push_stats(facet, true);
4719 facet_account(struct facet *facet)
4721 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4722 const struct nlattr *a;
4727 if (!facet->xout.has_normal || !ofproto->has_bonded_bundles) {
4730 n_bytes = facet->byte_count - facet->accounted_bytes;
4732 /* This loop feeds byte counters to bond_account() for rebalancing to use
4733 * as a basis. We also need to track the actual VLAN on which the packet
4734 * is going to be sent to ensure that it matches the one passed to
4735 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
4738 * We use the actions from an arbitrary subfacet because they should all
4739 * be equally valid for our purpose. */
4740 vlan_tci = facet->flow.vlan_tci;
4741 NL_ATTR_FOR_EACH_UNSAFE (a, left, facet->xout.odp_actions.data,
4742 facet->xout.odp_actions.size) {
4743 const struct ovs_action_push_vlan *vlan;
4744 struct ofport_dpif *port;
4746 switch (nl_attr_type(a)) {
4747 case OVS_ACTION_ATTR_OUTPUT:
4748 port = get_odp_port(ofproto, nl_attr_get_u32(a));
4749 if (port && port->bundle && port->bundle->bond) {
4750 bond_account(port->bundle->bond, &facet->flow,
4751 vlan_tci_to_vid(vlan_tci), n_bytes);
4755 case OVS_ACTION_ATTR_POP_VLAN:
4756 vlan_tci = htons(0);
4759 case OVS_ACTION_ATTR_PUSH_VLAN:
4760 vlan = nl_attr_get(a);
4761 vlan_tci = vlan->vlan_tci;
4767 /* Returns true if the only action for 'facet' is to send to the controller.
4768 * (We don't report NetFlow expiration messages for such facets because they
4769 * are just part of the control logic for the network, not real traffic). */
4771 facet_is_controller_flow(struct facet *facet)
4774 const struct rule *rule = &facet->rule->up;
4775 const struct ofpact *ofpacts = rule->ofpacts;
4776 size_t ofpacts_len = rule->ofpacts_len;
4778 if (ofpacts_len > 0 &&
4779 ofpacts->type == OFPACT_CONTROLLER &&
4780 ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len)) {
4787 /* Folds all of 'facet''s statistics into its rule. Also updates the
4788 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
4789 * 'facet''s statistics in the datapath should have been zeroed and folded into
4790 * its packet and byte counts before this function is called. */
4792 facet_flush_stats(struct facet *facet)
4794 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4795 struct subfacet *subfacet;
4797 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4798 ovs_assert(!subfacet->dp_byte_count);
4799 ovs_assert(!subfacet->dp_packet_count);
4802 facet_push_stats(facet, false);
4803 if (facet->accounted_bytes < facet->byte_count) {
4804 facet_account(facet);
4805 facet->accounted_bytes = facet->byte_count;
4808 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
4809 struct ofexpired expired;
4810 expired.flow = facet->flow;
4811 expired.packet_count = facet->packet_count;
4812 expired.byte_count = facet->byte_count;
4813 expired.used = facet->used;
4814 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4817 /* Reset counters to prevent double counting if 'facet' ever gets
4819 facet_reset_counters(facet);
4821 netflow_flow_clear(&facet->nf_flow);
4822 facet->tcp_flags = 0;
4825 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
4826 * Returns it if found, otherwise a null pointer.
4828 * 'hash' must be the return value of flow_hash(flow, 0).
4830 * The returned facet might need revalidation; use facet_lookup_valid()
4831 * instead if that is important. */
4832 static struct facet *
4833 facet_find(struct ofproto_dpif *ofproto,
4834 const struct flow *flow, uint32_t hash)
4836 struct facet *facet;
4838 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, hash, &ofproto->facets) {
4839 if (flow_equal(flow, &facet->flow)) {
4847 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
4848 * Returns it if found, otherwise a null pointer.
4850 * 'hash' must be the return value of flow_hash(flow, 0).
4852 * The returned facet is guaranteed to be valid. */
4853 static struct facet *
4854 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow,
4857 struct facet *facet;
4859 facet = facet_find(ofproto, flow, hash);
4861 && (ofproto->backer->need_revalidate
4862 || tag_set_intersects(&ofproto->backer->revalidate_set,
4864 && !facet_revalidate(facet)) {
4872 facet_check_consistency(struct facet *facet)
4874 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
4876 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4878 struct xlate_out xout;
4879 struct xlate_in xin;
4881 struct rule_dpif *rule;
4884 /* Check the rule for consistency. */
4885 rule = rule_dpif_lookup(ofproto, &facet->flow);
4886 if (rule != facet->rule) {
4887 if (!VLOG_DROP_WARN(&rl)) {
4888 struct ds s = DS_EMPTY_INITIALIZER;
4890 flow_format(&s, &facet->flow);
4891 ds_put_format(&s, ": facet associated with wrong rule (was "
4892 "table=%"PRIu8",", facet->rule->up.table_id);
4893 cls_rule_format(&facet->rule->up.cr, &s);
4894 ds_put_format(&s, ") (should have been table=%"PRIu8",",
4896 cls_rule_format(&rule->up.cr, &s);
4897 ds_put_cstr(&s, ")\n");
4904 /* Check the datapath actions for consistency. */
4905 xlate_in_init(&xin, ofproto, &facet->flow, &facet->initial_vals, rule,
4907 xlate_actions(&xin, &xout);
4909 ok = ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)
4910 && facet->xout.slow == xout.slow;
4911 if (!ok && !VLOG_DROP_WARN(&rl)) {
4912 struct ds s = DS_EMPTY_INITIALIZER;
4914 flow_format(&s, &facet->flow);
4915 ds_put_cstr(&s, ": inconsistency in facet");
4917 if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4918 ds_put_cstr(&s, " (actions were: ");
4919 format_odp_actions(&s, facet->xout.odp_actions.data,
4920 facet->xout.odp_actions.size);
4921 ds_put_cstr(&s, ") (correct actions: ");
4922 format_odp_actions(&s, xout.odp_actions.data,
4923 xout.odp_actions.size);
4924 ds_put_cstr(&s, ")");
4927 if (facet->xout.slow != xout.slow) {
4928 ds_put_format(&s, " slow path incorrect. should be %d", xout.slow);
4933 xlate_out_uninit(&xout);
4938 /* Re-searches the classifier for 'facet':
4940 * - If the rule found is different from 'facet''s current rule, moves
4941 * 'facet' to the new rule and recompiles its actions.
4943 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
4944 * where it is and recompiles its actions anyway.
4946 * - If any of 'facet''s subfacets correspond to a new flow according to
4947 * ofproto_receive(), 'facet' is removed.
4949 * Returns true if 'facet' is still valid. False if 'facet' was removed. */
4951 facet_revalidate(struct facet *facet)
4953 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4954 struct rule_dpif *new_rule;
4955 struct subfacet *subfacet;
4956 struct xlate_out xout;
4957 struct xlate_in xin;
4959 COVERAGE_INC(facet_revalidate);
4961 /* Check that child subfacets still correspond to this facet. Tunnel
4962 * configuration changes could cause a subfacet's OpenFlow in_port to
4964 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4965 struct ofproto_dpif *recv_ofproto;
4966 struct flow recv_flow;
4969 error = ofproto_receive(ofproto->backer, NULL, subfacet->key,
4970 subfacet->key_len, &recv_flow, NULL,
4971 &recv_ofproto, NULL, NULL);
4973 || recv_ofproto != ofproto
4974 || memcmp(&recv_flow, &facet->flow, sizeof recv_flow)) {
4975 facet_remove(facet);
4980 new_rule = rule_dpif_lookup(ofproto, &facet->flow);
4982 /* Calculate new datapath actions.
4984 * We do not modify any 'facet' state yet, because we might need to, e.g.,
4985 * emit a NetFlow expiration and, if so, we need to have the old state
4986 * around to properly compose it. */
4987 xlate_in_init(&xin, ofproto, &facet->flow, &facet->initial_vals, new_rule,
4989 xlate_actions(&xin, &xout);
4991 /* A facet's slow path reason should only change under dramatic
4992 * circumstances. Rather than try to update everything, it's simpler to
4993 * remove the facet and start over. */
4994 if (facet->xout.slow != xout.slow) {
4995 facet_remove(facet);
4996 xlate_out_uninit(&xout);
5000 if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
5001 LIST_FOR_EACH(subfacet, list_node, &facet->subfacets) {
5002 if (subfacet->path == SF_FAST_PATH) {
5003 struct dpif_flow_stats stats;
5005 subfacet_install(subfacet, &xout.odp_actions, &stats);
5006 subfacet_update_stats(subfacet, &stats);
5010 facet_flush_stats(facet);
5012 ofpbuf_clear(&facet->xout.odp_actions);
5013 ofpbuf_put(&facet->xout.odp_actions, xout.odp_actions.data,
5014 xout.odp_actions.size);
5017 /* Update 'facet' now that we've taken care of all the old state. */
5018 facet->xout.tags = xout.tags;
5019 facet->xout.slow = xout.slow;
5020 facet->xout.has_learn = xout.has_learn;
5021 facet->xout.has_normal = xout.has_normal;
5022 facet->xout.has_fin_timeout = xout.has_fin_timeout;
5023 facet->xout.nf_output_iface = xout.nf_output_iface;
5024 facet->xout.mirrors = xout.mirrors;
5025 facet->nf_flow.output_iface = facet->xout.nf_output_iface;
5027 if (facet->rule != new_rule) {
5028 COVERAGE_INC(facet_changed_rule);
5029 list_remove(&facet->list_node);
5030 list_push_back(&new_rule->facets, &facet->list_node);
5031 facet->rule = new_rule;
5032 facet->used = new_rule->up.created;
5033 facet->prev_used = facet->used;
5036 xlate_out_uninit(&xout);
5041 facet_reset_counters(struct facet *facet)
5043 facet->packet_count = 0;
5044 facet->byte_count = 0;
5045 facet->prev_packet_count = 0;
5046 facet->prev_byte_count = 0;
5047 facet->accounted_bytes = 0;
5051 facet_push_stats(struct facet *facet, bool may_learn)
5053 struct dpif_flow_stats stats;
5055 ovs_assert(facet->packet_count >= facet->prev_packet_count);
5056 ovs_assert(facet->byte_count >= facet->prev_byte_count);
5057 ovs_assert(facet->used >= facet->prev_used);
5059 stats.n_packets = facet->packet_count - facet->prev_packet_count;
5060 stats.n_bytes = facet->byte_count - facet->prev_byte_count;
5061 stats.used = facet->used;
5062 stats.tcp_flags = facet->tcp_flags;
5064 if (may_learn || stats.n_packets || facet->used > facet->prev_used) {
5065 struct ofproto_dpif *ofproto =
5066 ofproto_dpif_cast(facet->rule->up.ofproto);
5068 struct ofport_dpif *in_port;
5069 struct xlate_in xin;
5071 facet->prev_packet_count = facet->packet_count;
5072 facet->prev_byte_count = facet->byte_count;
5073 facet->prev_used = facet->used;
5075 in_port = get_ofp_port(ofproto, facet->flow.in_port);
5076 if (in_port && in_port->tnl_port) {
5077 netdev_vport_inc_rx(in_port->up.netdev, &stats);
5080 rule_credit_stats(facet->rule, &stats);
5081 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow,
5083 netflow_flow_update_flags(&facet->nf_flow, facet->tcp_flags);
5084 update_mirror_stats(ofproto, facet->xout.mirrors, stats.n_packets,
5087 xlate_in_init(&xin, ofproto, &facet->flow, &facet->initial_vals,
5088 facet->rule, stats.tcp_flags, NULL);
5089 xin.resubmit_stats = &stats;
5090 xin.may_learn = may_learn;
5091 xlate_actions_for_side_effects(&xin);
5096 push_all_stats__(bool run_fast)
5098 static long long int rl = LLONG_MIN;
5099 struct ofproto_dpif *ofproto;
5101 if (time_msec() < rl) {
5105 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5106 struct facet *facet;
5108 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
5109 facet_push_stats(facet, false);
5116 rl = time_msec() + 100;
5120 push_all_stats(void)
5122 push_all_stats__(true);
5126 rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats)
5128 rule->packet_count += stats->n_packets;
5129 rule->byte_count += stats->n_bytes;
5130 ofproto_rule_update_used(&rule->up, stats->used);
5135 static struct subfacet *
5136 subfacet_find(struct ofproto_dpif *ofproto,
5137 const struct nlattr *key, size_t key_len, uint32_t key_hash)
5139 struct subfacet *subfacet;
5141 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
5142 &ofproto->subfacets) {
5143 if (subfacet->key_len == key_len
5144 && !memcmp(key, subfacet->key, key_len)) {
5152 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
5153 * 'key_fitness', 'key', and 'key_len' members in 'miss'. Returns the
5154 * existing subfacet if there is one, otherwise creates and returns a
5156 static struct subfacet *
5157 subfacet_create(struct facet *facet, struct flow_miss *miss,
5160 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5161 enum odp_key_fitness key_fitness = miss->key_fitness;
5162 const struct nlattr *key = miss->key;
5163 size_t key_len = miss->key_len;
5165 struct subfacet *subfacet;
5167 key_hash = odp_flow_key_hash(key, key_len);
5169 if (list_is_empty(&facet->subfacets)) {
5170 subfacet = &facet->one_subfacet;
5172 subfacet = subfacet_find(ofproto, key, key_len, key_hash);
5174 if (subfacet->facet == facet) {
5178 /* This shouldn't happen. */
5179 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
5180 subfacet_destroy(subfacet);
5183 subfacet = xmalloc(sizeof *subfacet);
5186 hmap_insert(&ofproto->subfacets, &subfacet->hmap_node, key_hash);
5187 list_push_back(&facet->subfacets, &subfacet->list_node);
5188 subfacet->facet = facet;
5189 subfacet->key_fitness = key_fitness;
5190 subfacet->key = xmemdup(key, key_len);
5191 subfacet->key_len = key_len;
5192 subfacet->used = now;
5193 subfacet->created = now;
5194 subfacet->dp_packet_count = 0;
5195 subfacet->dp_byte_count = 0;
5196 subfacet->path = SF_NOT_INSTALLED;
5198 ofproto->subfacet_add_count++;
5202 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
5203 * its facet within 'ofproto', and frees it. */
5205 subfacet_destroy__(struct subfacet *subfacet)
5207 struct facet *facet = subfacet->facet;
5208 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5210 /* Update ofproto stats before uninstall the subfacet. */
5211 ofproto->subfacet_del_count++;
5212 ofproto->total_subfacet_life_span += (time_msec() - subfacet->created);
5214 subfacet_uninstall(subfacet);
5215 hmap_remove(&ofproto->subfacets, &subfacet->hmap_node);
5216 list_remove(&subfacet->list_node);
5217 free(subfacet->key);
5218 if (subfacet != &facet->one_subfacet) {
5223 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
5224 * last remaining subfacet in its facet destroys the facet too. */
5226 subfacet_destroy(struct subfacet *subfacet)
5228 struct facet *facet = subfacet->facet;
5230 if (list_is_singleton(&facet->subfacets)) {
5231 /* facet_remove() needs at least one subfacet (it will remove it). */
5232 facet_remove(facet);
5234 subfacet_destroy__(subfacet);
5239 subfacet_destroy_batch(struct ofproto_dpif *ofproto,
5240 struct subfacet **subfacets, int n)
5242 struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH];
5243 struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH];
5244 struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH];
5247 for (i = 0; i < n; i++) {
5248 ops[i].type = DPIF_OP_FLOW_DEL;
5249 ops[i].u.flow_del.key = subfacets[i]->key;
5250 ops[i].u.flow_del.key_len = subfacets[i]->key_len;
5251 ops[i].u.flow_del.stats = &stats[i];
5255 dpif_operate(ofproto->backer->dpif, opsp, n);
5256 for (i = 0; i < n; i++) {
5257 subfacet_reset_dp_stats(subfacets[i], &stats[i]);
5258 subfacets[i]->path = SF_NOT_INSTALLED;
5259 subfacet_destroy(subfacets[i]);
5264 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
5265 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
5266 * in the datapath will be zeroed and 'stats' will be updated with traffic new
5267 * since 'subfacet' was last updated.
5269 * Returns 0 if successful, otherwise a positive errno value. */
5271 subfacet_install(struct subfacet *subfacet, const struct ofpbuf *odp_actions,
5272 struct dpif_flow_stats *stats)
5274 struct facet *facet = subfacet->facet;
5275 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5276 enum subfacet_path path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
5277 const struct nlattr *actions = odp_actions->data;
5278 size_t actions_len = odp_actions->size;
5280 uint64_t slow_path_stub[128 / 8];
5281 enum dpif_flow_put_flags flags;
5284 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
5286 flags |= DPIF_FP_ZERO_STATS;
5289 if (path == SF_SLOW_PATH) {
5290 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
5291 slow_path_stub, sizeof slow_path_stub,
5292 &actions, &actions_len);
5295 ret = dpif_flow_put(ofproto->backer->dpif, flags, subfacet->key,
5296 subfacet->key_len, actions, actions_len, stats);
5299 subfacet_reset_dp_stats(subfacet, stats);
5303 subfacet->path = path;
5308 /* If 'subfacet' is installed in the datapath, uninstalls it. */
5310 subfacet_uninstall(struct subfacet *subfacet)
5312 if (subfacet->path != SF_NOT_INSTALLED) {
5313 struct rule_dpif *rule = subfacet->facet->rule;
5314 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5315 struct dpif_flow_stats stats;
5318 error = dpif_flow_del(ofproto->backer->dpif, subfacet->key,
5319 subfacet->key_len, &stats);
5320 subfacet_reset_dp_stats(subfacet, &stats);
5322 subfacet_update_stats(subfacet, &stats);
5324 subfacet->path = SF_NOT_INSTALLED;
5326 ovs_assert(subfacet->dp_packet_count == 0);
5327 ovs_assert(subfacet->dp_byte_count == 0);
5331 /* Resets 'subfacet''s datapath statistics counters. This should be called
5332 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
5333 * non-null, it should contain the statistics returned by dpif when 'subfacet'
5334 * was reset in the datapath. 'stats' will be modified to include only
5335 * statistics new since 'subfacet' was last updated. */
5337 subfacet_reset_dp_stats(struct subfacet *subfacet,
5338 struct dpif_flow_stats *stats)
5341 && subfacet->dp_packet_count <= stats->n_packets
5342 && subfacet->dp_byte_count <= stats->n_bytes) {
5343 stats->n_packets -= subfacet->dp_packet_count;
5344 stats->n_bytes -= subfacet->dp_byte_count;
5347 subfacet->dp_packet_count = 0;
5348 subfacet->dp_byte_count = 0;
5351 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
5353 * Because of the meaning of a subfacet's counters, it only makes sense to do
5354 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
5355 * represents a packet that was sent by hand or if it represents statistics
5356 * that have been cleared out of the datapath. */
5358 subfacet_update_stats(struct subfacet *subfacet,
5359 const struct dpif_flow_stats *stats)
5361 if (stats->n_packets || stats->used > subfacet->used) {
5362 struct facet *facet = subfacet->facet;
5364 subfacet->used = MAX(subfacet->used, stats->used);
5365 facet->used = MAX(facet->used, stats->used);
5366 facet->packet_count += stats->n_packets;
5367 facet->byte_count += stats->n_bytes;
5368 facet->tcp_flags |= stats->tcp_flags;
5374 static struct rule_dpif *
5375 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow)
5377 struct rule_dpif *rule;
5379 rule = rule_dpif_lookup__(ofproto, flow, 0);
5384 return rule_dpif_miss_rule(ofproto, flow);
5387 static struct rule_dpif *
5388 rule_dpif_lookup__(struct ofproto_dpif *ofproto, const struct flow *flow,
5391 struct cls_rule *cls_rule;
5392 struct classifier *cls;
5394 if (table_id >= N_TABLES) {
5398 cls = &ofproto->up.tables[table_id].cls;
5399 if (flow->nw_frag & FLOW_NW_FRAG_ANY
5400 && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
5401 /* For OFPC_NORMAL frag_handling, we must pretend that transport ports
5402 * are unavailable. */
5403 struct flow ofpc_normal_flow = *flow;
5404 ofpc_normal_flow.tp_src = htons(0);
5405 ofpc_normal_flow.tp_dst = htons(0);
5406 cls_rule = classifier_lookup(cls, &ofpc_normal_flow);
5408 cls_rule = classifier_lookup(cls, flow);
5410 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
5413 static struct rule_dpif *
5414 rule_dpif_miss_rule(struct ofproto_dpif *ofproto, const struct flow *flow)
5416 struct ofport_dpif *port;
5418 port = get_ofp_port(ofproto, flow->in_port);
5420 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, flow->in_port);
5421 return ofproto->miss_rule;
5424 if (port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN) {
5425 return ofproto->no_packet_in_rule;
5427 return ofproto->miss_rule;
5431 complete_operation(struct rule_dpif *rule)
5433 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5435 rule_invalidate(rule);
5437 struct dpif_completion *c = xmalloc(sizeof *c);
5438 c->op = rule->up.pending;
5439 list_push_back(&ofproto->completions, &c->list_node);
5441 ofoperation_complete(rule->up.pending, 0);
5445 static struct rule *
5448 struct rule_dpif *rule = xmalloc(sizeof *rule);
5453 rule_dealloc(struct rule *rule_)
5455 struct rule_dpif *rule = rule_dpif_cast(rule_);
5460 rule_construct(struct rule *rule_)
5462 struct rule_dpif *rule = rule_dpif_cast(rule_);
5463 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5464 struct rule_dpif *victim;
5467 rule->packet_count = 0;
5468 rule->byte_count = 0;
5470 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
5471 if (victim && !list_is_empty(&victim->facets)) {
5472 struct facet *facet;
5474 rule->facets = victim->facets;
5475 list_moved(&rule->facets);
5476 LIST_FOR_EACH (facet, list_node, &rule->facets) {
5477 /* XXX: We're only clearing our local counters here. It's possible
5478 * that quite a few packets are unaccounted for in the datapath
5479 * statistics. These will be accounted to the new rule instead of
5480 * cleared as required. This could be fixed by clearing out the
5481 * datapath statistics for this facet, but currently it doesn't
5483 facet_reset_counters(facet);
5487 /* Must avoid list_moved() in this case. */
5488 list_init(&rule->facets);
5491 table_id = rule->up.table_id;
5493 rule->tag = victim->tag;
5494 } else if (table_id == 0) {
5499 miniflow_expand(&rule->up.cr.match.flow, &flow);
5500 rule->tag = rule_calculate_tag(&flow, &rule->up.cr.match.mask,
5501 ofproto->tables[table_id].basis);
5504 complete_operation(rule);
5509 rule_destruct(struct rule *rule_)
5511 struct rule_dpif *rule = rule_dpif_cast(rule_);
5512 struct facet *facet, *next_facet;
5514 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
5515 facet_revalidate(facet);
5518 complete_operation(rule);
5522 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
5524 struct rule_dpif *rule = rule_dpif_cast(rule_);
5526 /* push_all_stats() can handle flow misses which, when using the learn
5527 * action, can cause rules to be added and deleted. This can corrupt our
5528 * caller's datastructures which assume that rule_get_stats() doesn't have
5529 * an impact on the flow table. To be safe, we disable miss handling. */
5530 push_all_stats__(false);
5532 /* Start from historical data for 'rule' itself that are no longer tracked
5533 * in facets. This counts, for example, facets that have expired. */
5534 *packets = rule->packet_count;
5535 *bytes = rule->byte_count;
5539 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
5540 struct ofpbuf *packet)
5542 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5543 struct initial_vals initial_vals;
5544 struct dpif_flow_stats stats;
5545 struct xlate_out xout;
5546 struct xlate_in xin;
5548 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
5549 rule_credit_stats(rule, &stats);
5551 initial_vals.vlan_tci = flow->vlan_tci;
5552 xlate_in_init(&xin, ofproto, flow, &initial_vals, rule, stats.tcp_flags,
5554 xin.resubmit_stats = &stats;
5555 xlate_actions(&xin, &xout);
5557 execute_odp_actions(ofproto, flow, xout.odp_actions.data,
5558 xout.odp_actions.size, packet);
5560 xlate_out_uninit(&xout);
5564 rule_execute(struct rule *rule, const struct flow *flow,
5565 struct ofpbuf *packet)
5567 rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
5568 ofpbuf_delete(packet);
5573 rule_modify_actions(struct rule *rule_)
5575 struct rule_dpif *rule = rule_dpif_cast(rule_);
5577 complete_operation(rule);
5580 /* Sends 'packet' out 'ofport'.
5581 * May modify 'packet'.
5582 * Returns 0 if successful, otherwise a positive errno value. */
5584 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
5586 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
5587 uint64_t odp_actions_stub[1024 / 8];
5588 struct ofpbuf key, odp_actions;
5589 struct dpif_flow_stats stats;
5590 struct odputil_keybuf keybuf;
5591 struct ofpact_output output;
5592 struct xlate_out xout;
5593 struct xlate_in xin;
5597 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5598 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5600 /* Use OFPP_NONE as the in_port to avoid special packet processing. */
5601 flow_extract(packet, 0, 0, NULL, OFPP_NONE, &flow);
5602 odp_flow_key_from_flow(&key, &flow, ofp_port_to_odp_port(ofproto,
5604 dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
5606 ofpact_init(&output.ofpact, OFPACT_OUTPUT, sizeof output);
5607 output.port = ofport->up.ofp_port;
5610 xlate_in_init(&xin, ofproto, &flow, NULL, NULL, 0, packet);
5611 xin.ofpacts_len = sizeof output;
5612 xin.ofpacts = &output.ofpact;
5613 xin.resubmit_stats = &stats;
5614 xlate_actions(&xin, &xout);
5616 error = dpif_execute(ofproto->backer->dpif,
5618 xout.odp_actions.data, xout.odp_actions.size,
5620 xlate_out_uninit(&xout);
5623 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %s (%s)",
5624 ofproto->up.name, netdev_get_name(ofport->up.netdev),
5628 ofproto->stats.tx_packets++;
5629 ofproto->stats.tx_bytes += packet->size;
5633 /* OpenFlow to datapath action translation. */
5635 static bool may_receive(const struct ofport_dpif *, struct xlate_ctx *);
5636 static void do_xlate_actions(const struct ofpact *, size_t ofpacts_len,
5637 struct xlate_ctx *);
5638 static void xlate_normal(struct xlate_ctx *);
5640 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
5641 * The action will state 'slow' as the reason that the action is in the slow
5642 * path. (This is purely informational: it allows a human viewing "ovs-dpctl
5643 * dump-flows" output to see why a flow is in the slow path.)
5645 * The 'stub_size' bytes in 'stub' will be used to store the action.
5646 * 'stub_size' must be large enough for the action.
5648 * The action and its size will be stored in '*actionsp' and '*actions_lenp',
5651 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
5652 enum slow_path_reason slow,
5653 uint64_t *stub, size_t stub_size,
5654 const struct nlattr **actionsp, size_t *actions_lenp)
5656 union user_action_cookie cookie;
5659 cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
5660 cookie.slow_path.unused = 0;
5661 cookie.slow_path.reason = slow;
5663 ofpbuf_use_stack(&buf, stub, stub_size);
5664 if (slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP | SLOW_STP)) {
5665 uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif, UINT32_MAX);
5666 odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
5668 put_userspace_action(ofproto, &buf, flow, &cookie,
5669 sizeof cookie.slow_path);
5671 *actionsp = buf.data;
5672 *actions_lenp = buf.size;
5676 put_userspace_action(const struct ofproto_dpif *ofproto,
5677 struct ofpbuf *odp_actions,
5678 const struct flow *flow,
5679 const union user_action_cookie *cookie,
5680 const size_t cookie_size)
5684 pid = dpif_port_get_pid(ofproto->backer->dpif,
5685 ofp_port_to_odp_port(ofproto, flow->in_port));
5687 return odp_put_userspace_action(pid, cookie, cookie_size, odp_actions);
5690 /* Compose SAMPLE action for sFlow or IPFIX. The given probability is
5691 * the number of packets out of UINT32_MAX to sample. The given
5692 * cookie is passed back in the callback for each sampled packet.
5695 compose_sample_action(const struct ofproto_dpif *ofproto,
5696 struct ofpbuf *odp_actions,
5697 const struct flow *flow,
5698 const uint32_t probability,
5699 const union user_action_cookie *cookie,
5700 const size_t cookie_size)
5702 size_t sample_offset, actions_offset;
5705 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
5707 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
5709 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
5710 cookie_offset = put_userspace_action(ofproto, odp_actions, flow, cookie,
5713 nl_msg_end_nested(odp_actions, actions_offset);
5714 nl_msg_end_nested(odp_actions, sample_offset);
5715 return cookie_offset;
5719 compose_sflow_cookie(const struct ofproto_dpif *ofproto,
5720 ovs_be16 vlan_tci, uint32_t odp_port,
5721 unsigned int n_outputs, union user_action_cookie *cookie)
5725 cookie->type = USER_ACTION_COOKIE_SFLOW;
5726 cookie->sflow.vlan_tci = vlan_tci;
5728 /* See http://www.sflow.org/sflow_version_5.txt (search for "Input/output
5729 * port information") for the interpretation of cookie->output. */
5730 switch (n_outputs) {
5732 /* 0x40000000 | 256 means "packet dropped for unknown reason". */
5733 cookie->sflow.output = 0x40000000 | 256;
5737 ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port);
5739 cookie->sflow.output = ifindex;
5744 /* 0x80000000 means "multiple output ports. */
5745 cookie->sflow.output = 0x80000000 | n_outputs;
5750 /* Compose SAMPLE action for sFlow bridge sampling. */
5752 compose_sflow_action(const struct ofproto_dpif *ofproto,
5753 struct ofpbuf *odp_actions,
5754 const struct flow *flow,
5757 uint32_t probability;
5758 union user_action_cookie cookie;
5760 if (!ofproto->sflow || flow->in_port == OFPP_NONE) {
5764 probability = dpif_sflow_get_probability(ofproto->sflow);
5765 compose_sflow_cookie(ofproto, htons(0), odp_port,
5766 odp_port == OVSP_NONE ? 0 : 1, &cookie);
5768 return compose_sample_action(ofproto, odp_actions, flow, probability,
5769 &cookie, sizeof cookie.sflow);
5773 compose_flow_sample_cookie(uint16_t probability, uint32_t collector_set_id,
5774 uint32_t obs_domain_id, uint32_t obs_point_id,
5775 union user_action_cookie *cookie)
5777 cookie->type = USER_ACTION_COOKIE_FLOW_SAMPLE;
5778 cookie->flow_sample.probability = probability;
5779 cookie->flow_sample.collector_set_id = collector_set_id;
5780 cookie->flow_sample.obs_domain_id = obs_domain_id;
5781 cookie->flow_sample.obs_point_id = obs_point_id;
5785 compose_ipfix_cookie(union user_action_cookie *cookie)
5787 cookie->type = USER_ACTION_COOKIE_IPFIX;
5790 /* Compose SAMPLE action for IPFIX bridge sampling. */
5792 compose_ipfix_action(const struct ofproto_dpif *ofproto,
5793 struct ofpbuf *odp_actions,
5794 const struct flow *flow)
5796 uint32_t probability;
5797 union user_action_cookie cookie;
5799 if (!ofproto->ipfix || flow->in_port == OFPP_NONE) {
5803 probability = dpif_ipfix_get_bridge_exporter_probability(ofproto->ipfix);
5804 compose_ipfix_cookie(&cookie);
5806 compose_sample_action(ofproto, odp_actions, flow, probability,
5807 &cookie, sizeof cookie.ipfix);
5810 /* SAMPLE action for sFlow must be first action in any given list of
5811 * actions. At this point we do not have all information required to
5812 * build it. So try to build sample action as complete as possible. */
5814 add_sflow_action(struct xlate_ctx *ctx)
5816 ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto,
5817 &ctx->xout->odp_actions,
5818 &ctx->xin->flow, OVSP_NONE);
5819 ctx->sflow_odp_port = 0;
5820 ctx->sflow_n_outputs = 0;
5823 /* SAMPLE action for IPFIX must be 1st or 2nd action in any given list
5824 * of actions, eventually after the SAMPLE action for sFlow. */
5826 add_ipfix_action(struct xlate_ctx *ctx)
5828 compose_ipfix_action(ctx->ofproto, &ctx->xout->odp_actions,
5832 /* Fix SAMPLE action according to data collected while composing ODP actions.
5833 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
5834 * USERSPACE action's user-cookie which is required for sflow. */
5836 fix_sflow_action(struct xlate_ctx *ctx)
5838 const struct flow *base = &ctx->base_flow;
5839 union user_action_cookie *cookie;
5841 if (!ctx->user_cookie_offset) {
5845 cookie = ofpbuf_at(&ctx->xout->odp_actions, ctx->user_cookie_offset,
5846 sizeof cookie->sflow);
5847 ovs_assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
5849 compose_sflow_cookie(ctx->ofproto, base->vlan_tci,
5850 ctx->sflow_odp_port, ctx->sflow_n_outputs, cookie);
5854 compose_output_action__(struct xlate_ctx *ctx, uint16_t ofp_port,
5857 const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
5858 ovs_be16 flow_vlan_tci;
5859 uint32_t flow_skb_mark;
5860 uint8_t flow_nw_tos;
5861 struct priority_to_dscp *pdscp;
5862 uint32_t out_port, odp_port;
5864 /* If 'struct flow' gets additional metadata, we'll need to zero it out
5865 * before traversing a patch port. */
5866 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 20);
5869 xlate_report(ctx, "Nonexistent output port");
5871 } else if (ofport->up.pp.config & OFPUTIL_PC_NO_FWD) {
5872 xlate_report(ctx, "OFPPC_NO_FWD set, skipping output");
5874 } else if (check_stp && !stp_forward_in_state(ofport->stp_state)) {
5875 xlate_report(ctx, "STP not in forwarding state, skipping output");
5879 if (netdev_vport_is_patch(ofport->up.netdev)) {
5880 struct ofport_dpif *peer = ofport_get_peer(ofport);
5881 struct flow old_flow = ctx->xin->flow;
5882 const struct ofproto_dpif *peer_ofproto;
5883 enum slow_path_reason special;
5884 struct ofport_dpif *in_port;
5887 xlate_report(ctx, "Nonexistent patch port peer");
5891 peer_ofproto = ofproto_dpif_cast(peer->up.ofproto);
5892 if (peer_ofproto->backer != ctx->ofproto->backer) {
5893 xlate_report(ctx, "Patch port peer on a different datapath");
5897 ctx->ofproto = ofproto_dpif_cast(peer->up.ofproto);
5898 ctx->xin->flow.in_port = peer->up.ofp_port;
5899 ctx->xin->flow.metadata = htonll(0);
5900 memset(&ctx->xin->flow.tunnel, 0, sizeof ctx->xin->flow.tunnel);
5901 memset(ctx->xin->flow.regs, 0, sizeof ctx->xin->flow.regs);
5903 in_port = get_ofp_port(ctx->ofproto, ctx->xin->flow.in_port);
5904 special = process_special(ctx->ofproto, &ctx->xin->flow, in_port,
5907 ctx->xout->slow = special;
5908 } else if (!in_port || may_receive(in_port, ctx)) {
5909 if (!in_port || stp_forward_in_state(in_port->stp_state)) {
5910 xlate_table_action(ctx, ctx->xin->flow.in_port, 0, true);
5912 /* Forwarding is disabled by STP. Let OFPP_NORMAL and the
5913 * learning action look at the packet, then drop it. */
5914 struct flow old_base_flow = ctx->base_flow;
5915 size_t old_size = ctx->xout->odp_actions.size;
5916 xlate_table_action(ctx, ctx->xin->flow.in_port, 0, true);
5917 ctx->base_flow = old_base_flow;
5918 ctx->xout->odp_actions.size = old_size;
5922 ctx->xin->flow = old_flow;
5923 ctx->ofproto = ofproto_dpif_cast(ofport->up.ofproto);
5925 if (ctx->xin->resubmit_stats) {
5926 netdev_vport_inc_tx(ofport->up.netdev, ctx->xin->resubmit_stats);
5927 netdev_vport_inc_rx(peer->up.netdev, ctx->xin->resubmit_stats);
5933 flow_vlan_tci = ctx->xin->flow.vlan_tci;
5934 flow_skb_mark = ctx->xin->flow.skb_mark;
5935 flow_nw_tos = ctx->xin->flow.nw_tos;
5937 pdscp = get_priority(ofport, ctx->xin->flow.skb_priority);
5939 ctx->xin->flow.nw_tos &= ~IP_DSCP_MASK;
5940 ctx->xin->flow.nw_tos |= pdscp->dscp;
5943 if (ofport->tnl_port) {
5944 /* Save tunnel metadata so that changes made due to
5945 * the Logical (tunnel) Port are not visible for any further
5946 * matches, while explicit set actions on tunnel metadata are.
5948 struct flow_tnl flow_tnl = ctx->xin->flow.tunnel;
5949 odp_port = tnl_port_send(ofport->tnl_port, &ctx->xin->flow);
5950 if (odp_port == OVSP_NONE) {
5951 xlate_report(ctx, "Tunneling decided against output");
5952 goto out; /* restore flow_nw_tos */
5954 if (ctx->xin->flow.tunnel.ip_dst == ctx->orig_tunnel_ip_dst) {
5955 xlate_report(ctx, "Not tunneling to our own address");
5956 goto out; /* restore flow_nw_tos */
5958 if (ctx->xin->resubmit_stats) {
5959 netdev_vport_inc_tx(ofport->up.netdev, ctx->xin->resubmit_stats);
5961 out_port = odp_port;
5962 commit_odp_tunnel_action(&ctx->xin->flow, &ctx->base_flow,
5963 &ctx->xout->odp_actions);
5964 ctx->xin->flow.tunnel = flow_tnl; /* Restore tunnel metadata */
5966 odp_port = ofport->odp_port;
5967 out_port = vsp_realdev_to_vlandev(ctx->ofproto, odp_port,
5968 ctx->xin->flow.vlan_tci);
5969 if (out_port != odp_port) {
5970 ctx->xin->flow.vlan_tci = htons(0);
5972 ctx->xin->flow.skb_mark &= ~IPSEC_MARK;
5974 commit_odp_actions(&ctx->xin->flow, &ctx->base_flow,
5975 &ctx->xout->odp_actions);
5976 nl_msg_put_u32(&ctx->xout->odp_actions, OVS_ACTION_ATTR_OUTPUT, out_port);
5978 ctx->sflow_odp_port = odp_port;
5979 ctx->sflow_n_outputs++;
5980 ctx->xout->nf_output_iface = ofp_port;
5983 ctx->xin->flow.vlan_tci = flow_vlan_tci;
5984 ctx->xin->flow.skb_mark = flow_skb_mark;
5986 ctx->xin->flow.nw_tos = flow_nw_tos;
5990 compose_output_action(struct xlate_ctx *ctx, uint16_t ofp_port)
5992 compose_output_action__(ctx, ofp_port, true);
5996 tag_the_flow(struct xlate_ctx *ctx, struct rule_dpif *rule)
5998 struct ofproto_dpif *ofproto = ctx->ofproto;
5999 uint8_t table_id = ctx->table_id;
6001 if (table_id > 0 && table_id < N_TABLES) {
6002 struct table_dpif *table = &ofproto->tables[table_id];
6003 if (table->other_table) {
6004 ctx->xout->tags |= (rule && rule->tag
6006 : rule_calculate_tag(&ctx->xin->flow,
6007 &table->other_table->mask,
6013 /* Common rule processing in one place to avoid duplicating code. */
6014 static struct rule_dpif *
6015 ctx_rule_hooks(struct xlate_ctx *ctx, struct rule_dpif *rule,
6018 if (ctx->xin->resubmit_hook) {
6019 ctx->xin->resubmit_hook(ctx, rule);
6021 if (rule == NULL && may_packet_in) {
6023 * check if table configuration flags
6024 * OFPTC_TABLE_MISS_CONTROLLER, default.
6025 * OFPTC_TABLE_MISS_CONTINUE,
6026 * OFPTC_TABLE_MISS_DROP
6027 * When OF1.0, OFPTC_TABLE_MISS_CONTINUE is used. What to do?
6029 rule = rule_dpif_miss_rule(ctx->ofproto, &ctx->xin->flow);
6031 if (rule && ctx->xin->resubmit_stats) {
6032 rule_credit_stats(rule, ctx->xin->resubmit_stats);
6038 xlate_table_action(struct xlate_ctx *ctx,
6039 uint16_t in_port, uint8_t table_id, bool may_packet_in)
6041 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
6042 struct rule_dpif *rule;
6043 uint16_t old_in_port = ctx->xin->flow.in_port;
6044 uint8_t old_table_id = ctx->table_id;
6046 ctx->table_id = table_id;
6048 /* Look up a flow with 'in_port' as the input port. */
6049 ctx->xin->flow.in_port = in_port;
6050 rule = rule_dpif_lookup__(ctx->ofproto, &ctx->xin->flow, table_id);
6052 tag_the_flow(ctx, rule);
6054 /* Restore the original input port. Otherwise OFPP_NORMAL and
6055 * OFPP_IN_PORT will have surprising behavior. */
6056 ctx->xin->flow.in_port = old_in_port;
6058 rule = ctx_rule_hooks(ctx, rule, may_packet_in);
6061 struct rule_dpif *old_rule = ctx->rule;
6065 do_xlate_actions(rule->up.ofpacts, rule->up.ofpacts_len, ctx);
6066 ctx->rule = old_rule;
6070 ctx->table_id = old_table_id;
6072 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
6074 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
6075 MAX_RESUBMIT_RECURSION);
6076 ctx->max_resubmit_trigger = true;
6081 xlate_ofpact_resubmit(struct xlate_ctx *ctx,
6082 const struct ofpact_resubmit *resubmit)
6087 in_port = resubmit->in_port;
6088 if (in_port == OFPP_IN_PORT) {
6089 in_port = ctx->xin->flow.in_port;
6092 table_id = resubmit->table_id;
6093 if (table_id == 255) {
6094 table_id = ctx->table_id;
6097 xlate_table_action(ctx, in_port, table_id, false);
6101 flood_packets(struct xlate_ctx *ctx, bool all)
6103 struct ofport_dpif *ofport;
6105 HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) {
6106 uint16_t ofp_port = ofport->up.ofp_port;
6108 if (ofp_port == ctx->xin->flow.in_port) {
6113 compose_output_action__(ctx, ofp_port, false);
6114 } else if (!(ofport->up.pp.config & OFPUTIL_PC_NO_FLOOD)) {
6115 compose_output_action(ctx, ofp_port);
6119 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
6123 execute_controller_action(struct xlate_ctx *ctx, int len,
6124 enum ofp_packet_in_reason reason,
6125 uint16_t controller_id)
6127 struct ofputil_packet_in pin;
6128 struct ofpbuf *packet;
6130 ovs_assert(!ctx->xout->slow || ctx->xout->slow == SLOW_CONTROLLER);
6131 ctx->xout->slow = SLOW_CONTROLLER;
6132 if (!ctx->xin->packet) {
6136 packet = ofpbuf_clone(ctx->xin->packet);
6138 if (packet->l2 && packet->l3) {
6139 struct eth_header *eh;
6140 uint16_t mpls_depth;
6142 eth_pop_vlan(packet);
6145 memcpy(eh->eth_src, ctx->xin->flow.dl_src, sizeof eh->eth_src);
6146 memcpy(eh->eth_dst, ctx->xin->flow.dl_dst, sizeof eh->eth_dst);
6148 if (ctx->xin->flow.vlan_tci & htons(VLAN_CFI)) {
6149 eth_push_vlan(packet, ctx->xin->flow.vlan_tci);
6152 mpls_depth = eth_mpls_depth(packet);
6154 if (mpls_depth < ctx->xin->flow.mpls_depth) {
6155 push_mpls(packet, ctx->xin->flow.dl_type, ctx->xin->flow.mpls_lse);
6156 } else if (mpls_depth > ctx->xin->flow.mpls_depth) {
6157 pop_mpls(packet, ctx->xin->flow.dl_type);
6158 } else if (mpls_depth) {
6159 set_mpls_lse(packet, ctx->xin->flow.mpls_lse);
6163 if (ctx->xin->flow.dl_type == htons(ETH_TYPE_IP)) {
6164 packet_set_ipv4(packet, ctx->xin->flow.nw_src,
6165 ctx->xin->flow.nw_dst, ctx->xin->flow.nw_tos,
6166 ctx->xin->flow.nw_ttl);
6170 if (ctx->xin->flow.nw_proto == IPPROTO_TCP) {
6171 packet_set_tcp_port(packet, ctx->xin->flow.tp_src,
6172 ctx->xin->flow.tp_dst);
6173 } else if (ctx->xin->flow.nw_proto == IPPROTO_UDP) {
6174 packet_set_udp_port(packet, ctx->xin->flow.tp_src,
6175 ctx->xin->flow.tp_dst);
6181 pin.packet = packet->data;
6182 pin.packet_len = packet->size;
6183 pin.reason = reason;
6184 pin.controller_id = controller_id;
6185 pin.table_id = ctx->table_id;
6186 pin.cookie = ctx->rule ? ctx->rule->up.flow_cookie : 0;
6189 flow_get_metadata(&ctx->xin->flow, &pin.fmd);
6191 connmgr_send_packet_in(ctx->ofproto->up.connmgr, &pin);
6192 ofpbuf_delete(packet);
6196 execute_mpls_push_action(struct xlate_ctx *ctx, ovs_be16 eth_type)
6198 ovs_assert(eth_type_mpls(eth_type));
6200 if (ctx->base_flow.mpls_depth) {
6201 ctx->xin->flow.mpls_lse &= ~htonl(MPLS_BOS_MASK);
6202 ctx->xin->flow.mpls_depth++;
6207 if (ctx->xin->flow.dl_type == htons(ETH_TYPE_IPV6)) {
6208 label = htonl(0x2); /* IPV6 Explicit Null. */
6210 label = htonl(0x0); /* IPV4 Explicit Null. */
6212 tc = (ctx->xin->flow.nw_tos & IP_DSCP_MASK) >> 2;
6213 ttl = ctx->xin->flow.nw_ttl ? ctx->xin->flow.nw_ttl : 0x40;
6214 ctx->xin->flow.mpls_lse = set_mpls_lse_values(ttl, tc, 1, label);
6215 ctx->xin->flow.mpls_depth = 1;
6217 ctx->xin->flow.dl_type = eth_type;
6221 execute_mpls_pop_action(struct xlate_ctx *ctx, ovs_be16 eth_type)
6223 ovs_assert(eth_type_mpls(ctx->xin->flow.dl_type));
6224 ovs_assert(!eth_type_mpls(eth_type));
6226 if (ctx->xin->flow.mpls_depth) {
6227 ctx->xin->flow.mpls_depth--;
6228 ctx->xin->flow.mpls_lse = htonl(0);
6229 if (!ctx->xin->flow.mpls_depth) {
6230 ctx->xin->flow.dl_type = eth_type;
6236 compose_dec_ttl(struct xlate_ctx *ctx, struct ofpact_cnt_ids *ids)
6238 if (ctx->xin->flow.dl_type != htons(ETH_TYPE_IP) &&
6239 ctx->xin->flow.dl_type != htons(ETH_TYPE_IPV6)) {
6243 if (ctx->xin->flow.nw_ttl > 1) {
6244 ctx->xin->flow.nw_ttl--;
6249 for (i = 0; i < ids->n_controllers; i++) {
6250 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL,
6254 /* Stop processing for current table. */
6260 execute_set_mpls_ttl_action(struct xlate_ctx *ctx, uint8_t ttl)
6262 if (!eth_type_mpls(ctx->xin->flow.dl_type)) {
6266 set_mpls_lse_ttl(&ctx->xin->flow.mpls_lse, ttl);
6271 execute_dec_mpls_ttl_action(struct xlate_ctx *ctx)
6273 uint8_t ttl = mpls_lse_to_ttl(ctx->xin->flow.mpls_lse);
6275 if (!eth_type_mpls(ctx->xin->flow.dl_type)) {
6281 set_mpls_lse_ttl(&ctx->xin->flow.mpls_lse, ttl);
6284 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL, 0);
6286 /* Stop processing for current table. */
6292 xlate_output_action(struct xlate_ctx *ctx,
6293 uint16_t port, uint16_t max_len, bool may_packet_in)
6295 uint16_t prev_nf_output_iface = ctx->xout->nf_output_iface;
6297 ctx->xout->nf_output_iface = NF_OUT_DROP;
6301 compose_output_action(ctx, ctx->xin->flow.in_port);
6304 xlate_table_action(ctx, ctx->xin->flow.in_port, 0, may_packet_in);
6310 flood_packets(ctx, false);
6313 flood_packets(ctx, true);
6315 case OFPP_CONTROLLER:
6316 execute_controller_action(ctx, max_len, OFPR_ACTION, 0);
6322 if (port != ctx->xin->flow.in_port) {
6323 compose_output_action(ctx, port);
6325 xlate_report(ctx, "skipping output to input port");
6330 if (prev_nf_output_iface == NF_OUT_FLOOD) {
6331 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
6332 } else if (ctx->xout->nf_output_iface == NF_OUT_DROP) {
6333 ctx->xout->nf_output_iface = prev_nf_output_iface;
6334 } else if (prev_nf_output_iface != NF_OUT_DROP &&
6335 ctx->xout->nf_output_iface != NF_OUT_FLOOD) {
6336 ctx->xout->nf_output_iface = NF_OUT_MULTI;
6341 xlate_output_reg_action(struct xlate_ctx *ctx,
6342 const struct ofpact_output_reg *or)
6344 uint64_t port = mf_get_subfield(&or->src, &ctx->xin->flow);
6345 if (port <= UINT16_MAX) {
6346 xlate_output_action(ctx, port, or->max_len, false);
6351 xlate_enqueue_action(struct xlate_ctx *ctx,
6352 const struct ofpact_enqueue *enqueue)
6354 uint16_t ofp_port = enqueue->port;
6355 uint32_t queue_id = enqueue->queue;
6356 uint32_t flow_priority, priority;
6359 /* Translate queue to priority. */
6360 error = dpif_queue_to_priority(ctx->ofproto->backer->dpif,
6361 queue_id, &priority);
6363 /* Fall back to ordinary output action. */
6364 xlate_output_action(ctx, enqueue->port, 0, false);
6368 /* Check output port. */
6369 if (ofp_port == OFPP_IN_PORT) {
6370 ofp_port = ctx->xin->flow.in_port;
6371 } else if (ofp_port == ctx->xin->flow.in_port) {
6375 /* Add datapath actions. */
6376 flow_priority = ctx->xin->flow.skb_priority;
6377 ctx->xin->flow.skb_priority = priority;
6378 compose_output_action(ctx, ofp_port);
6379 ctx->xin->flow.skb_priority = flow_priority;
6381 /* Update NetFlow output port. */
6382 if (ctx->xout->nf_output_iface == NF_OUT_DROP) {
6383 ctx->xout->nf_output_iface = ofp_port;
6384 } else if (ctx->xout->nf_output_iface != NF_OUT_FLOOD) {
6385 ctx->xout->nf_output_iface = NF_OUT_MULTI;
6390 xlate_set_queue_action(struct xlate_ctx *ctx, uint32_t queue_id)
6392 uint32_t skb_priority;
6394 if (!dpif_queue_to_priority(ctx->ofproto->backer->dpif,
6395 queue_id, &skb_priority)) {
6396 ctx->xin->flow.skb_priority = skb_priority;
6398 /* Couldn't translate queue to a priority. Nothing to do. A warning
6399 * has already been logged. */
6404 slave_enabled_cb(uint16_t ofp_port, void *ofproto_)
6406 struct ofproto_dpif *ofproto = ofproto_;
6407 struct ofport_dpif *port;
6417 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
6420 port = get_ofp_port(ofproto, ofp_port);
6421 return port ? port->may_enable : false;
6426 xlate_bundle_action(struct xlate_ctx *ctx,
6427 const struct ofpact_bundle *bundle)
6431 port = bundle_execute(bundle, &ctx->xin->flow, slave_enabled_cb,
6433 if (bundle->dst.field) {
6434 nxm_reg_load(&bundle->dst, port, &ctx->xin->flow);
6436 xlate_output_action(ctx, port, 0, false);
6441 xlate_learn_action(struct xlate_ctx *ctx,
6442 const struct ofpact_learn *learn)
6444 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
6445 struct ofputil_flow_mod fm;
6446 uint64_t ofpacts_stub[1024 / 8];
6447 struct ofpbuf ofpacts;
6450 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
6451 learn_execute(learn, &ctx->xin->flow, &fm, &ofpacts);
6453 error = ofproto_flow_mod(&ctx->ofproto->up, &fm);
6454 if (error && !VLOG_DROP_WARN(&rl)) {
6455 VLOG_WARN("learning action failed to modify flow table (%s)",
6456 ofperr_get_name(error));
6459 ofpbuf_uninit(&ofpacts);
6462 /* Reduces '*timeout' to no more than 'max'. A value of zero in either case
6463 * means "infinite". */
6465 reduce_timeout(uint16_t max, uint16_t *timeout)
6467 if (max && (!*timeout || *timeout > max)) {
6473 xlate_fin_timeout(struct xlate_ctx *ctx,
6474 const struct ofpact_fin_timeout *oft)
6476 if (ctx->xin->tcp_flags & (TCP_FIN | TCP_RST) && ctx->rule) {
6477 struct rule_dpif *rule = ctx->rule;
6479 reduce_timeout(oft->fin_idle_timeout, &rule->up.idle_timeout);
6480 reduce_timeout(oft->fin_hard_timeout, &rule->up.hard_timeout);
6485 xlate_sample_action(struct xlate_ctx *ctx,
6486 const struct ofpact_sample *os)
6488 union user_action_cookie cookie;
6489 /* Scale the probability from 16-bit to 32-bit while representing
6490 * the same percentage. */
6491 uint32_t probability = (os->probability << 16) | os->probability;
6493 commit_odp_actions(&ctx->xin->flow, &ctx->base_flow,
6494 &ctx->xout->odp_actions);
6496 compose_flow_sample_cookie(os->probability, os->collector_set_id,
6497 os->obs_domain_id, os->obs_point_id, &cookie);
6498 compose_sample_action(ctx->ofproto, &ctx->xout->odp_actions, &ctx->xin->flow,
6499 probability, &cookie, sizeof cookie.flow_sample);
6503 may_receive(const struct ofport_dpif *port, struct xlate_ctx *ctx)
6505 if (port->up.pp.config & (eth_addr_equals(ctx->xin->flow.dl_dst,
6507 ? OFPUTIL_PC_NO_RECV_STP
6508 : OFPUTIL_PC_NO_RECV)) {
6512 /* Only drop packets here if both forwarding and learning are
6513 * disabled. If just learning is enabled, we need to have
6514 * OFPP_NORMAL and the learning action have a look at the packet
6515 * before we can drop it. */
6516 if (!stp_forward_in_state(port->stp_state)
6517 && !stp_learn_in_state(port->stp_state)) {
6525 tunnel_ecn_ok(struct xlate_ctx *ctx)
6527 if (is_ip_any(&ctx->base_flow)
6528 && (ctx->xin->flow.tunnel.ip_tos & IP_ECN_MASK) == IP_ECN_CE) {
6529 if ((ctx->base_flow.nw_tos & IP_ECN_MASK) == IP_ECN_NOT_ECT) {
6530 VLOG_WARN_RL(&rl, "dropping tunnel packet marked ECN CE"
6531 " but is not ECN capable");
6534 /* Set the ECN CE value in the tunneled packet. */
6535 ctx->xin->flow.nw_tos |= IP_ECN_CE;
6543 do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len,
6544 struct xlate_ctx *ctx)
6546 bool was_evictable = true;
6547 const struct ofpact *a;
6550 /* Don't let the rule we're working on get evicted underneath us. */
6551 was_evictable = ctx->rule->up.evictable;
6552 ctx->rule->up.evictable = false;
6555 do_xlate_actions_again:
6556 OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) {
6557 struct ofpact_controller *controller;
6558 const struct ofpact_metadata *metadata;
6566 xlate_output_action(ctx, ofpact_get_OUTPUT(a)->port,
6567 ofpact_get_OUTPUT(a)->max_len, true);
6570 case OFPACT_CONTROLLER:
6571 controller = ofpact_get_CONTROLLER(a);
6572 execute_controller_action(ctx, controller->max_len,
6574 controller->controller_id);
6577 case OFPACT_ENQUEUE:
6578 xlate_enqueue_action(ctx, ofpact_get_ENQUEUE(a));
6581 case OFPACT_SET_VLAN_VID:
6582 ctx->xin->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
6583 ctx->xin->flow.vlan_tci |=
6584 (htons(ofpact_get_SET_VLAN_VID(a)->vlan_vid)
6588 case OFPACT_SET_VLAN_PCP:
6589 ctx->xin->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
6590 ctx->xin->flow.vlan_tci |=
6591 htons((ofpact_get_SET_VLAN_PCP(a)->vlan_pcp << VLAN_PCP_SHIFT)
6595 case OFPACT_STRIP_VLAN:
6596 ctx->xin->flow.vlan_tci = htons(0);
6599 case OFPACT_PUSH_VLAN:
6600 /* XXX 802.1AD(QinQ) */
6601 ctx->xin->flow.vlan_tci = htons(VLAN_CFI);
6604 case OFPACT_SET_ETH_SRC:
6605 memcpy(ctx->xin->flow.dl_src, ofpact_get_SET_ETH_SRC(a)->mac,
6609 case OFPACT_SET_ETH_DST:
6610 memcpy(ctx->xin->flow.dl_dst, ofpact_get_SET_ETH_DST(a)->mac,
6614 case OFPACT_SET_IPV4_SRC:
6615 if (ctx->xin->flow.dl_type == htons(ETH_TYPE_IP)) {
6616 ctx->xin->flow.nw_src = ofpact_get_SET_IPV4_SRC(a)->ipv4;
6620 case OFPACT_SET_IPV4_DST:
6621 if (ctx->xin->flow.dl_type == htons(ETH_TYPE_IP)) {
6622 ctx->xin->flow.nw_dst = ofpact_get_SET_IPV4_DST(a)->ipv4;
6626 case OFPACT_SET_IPV4_DSCP:
6627 /* OpenFlow 1.0 only supports IPv4. */
6628 if (ctx->xin->flow.dl_type == htons(ETH_TYPE_IP)) {
6629 ctx->xin->flow.nw_tos &= ~IP_DSCP_MASK;
6630 ctx->xin->flow.nw_tos |= ofpact_get_SET_IPV4_DSCP(a)->dscp;
6634 case OFPACT_SET_L4_SRC_PORT:
6635 if (is_ip_any(&ctx->xin->flow)) {
6636 ctx->xin->flow.tp_src =
6637 htons(ofpact_get_SET_L4_SRC_PORT(a)->port);
6641 case OFPACT_SET_L4_DST_PORT:
6642 if (is_ip_any(&ctx->xin->flow)) {
6643 ctx->xin->flow.tp_dst =
6644 htons(ofpact_get_SET_L4_DST_PORT(a)->port);
6648 case OFPACT_RESUBMIT:
6649 xlate_ofpact_resubmit(ctx, ofpact_get_RESUBMIT(a));
6652 case OFPACT_SET_TUNNEL:
6653 ctx->xin->flow.tunnel.tun_id =
6654 htonll(ofpact_get_SET_TUNNEL(a)->tun_id);
6657 case OFPACT_SET_QUEUE:
6658 xlate_set_queue_action(ctx, ofpact_get_SET_QUEUE(a)->queue_id);
6661 case OFPACT_POP_QUEUE:
6662 ctx->xin->flow.skb_priority = ctx->orig_skb_priority;
6665 case OFPACT_REG_MOVE:
6666 nxm_execute_reg_move(ofpact_get_REG_MOVE(a), &ctx->xin->flow);
6669 case OFPACT_REG_LOAD:
6670 nxm_execute_reg_load(ofpact_get_REG_LOAD(a), &ctx->xin->flow);
6673 case OFPACT_STACK_PUSH:
6674 nxm_execute_stack_push(ofpact_get_STACK_PUSH(a), &ctx->xin->flow,
6678 case OFPACT_STACK_POP:
6679 nxm_execute_stack_pop(ofpact_get_STACK_POP(a), &ctx->xin->flow,
6683 case OFPACT_PUSH_MPLS:
6684 execute_mpls_push_action(ctx, ofpact_get_PUSH_MPLS(a)->ethertype);
6687 case OFPACT_POP_MPLS:
6688 execute_mpls_pop_action(ctx, ofpact_get_POP_MPLS(a)->ethertype);
6691 case OFPACT_SET_MPLS_TTL:
6692 if (execute_set_mpls_ttl_action(ctx,
6693 ofpact_get_SET_MPLS_TTL(a)->ttl)) {
6698 case OFPACT_DEC_MPLS_TTL:
6699 if (execute_dec_mpls_ttl_action(ctx)) {
6704 case OFPACT_DEC_TTL:
6705 if (compose_dec_ttl(ctx, ofpact_get_DEC_TTL(a))) {
6711 /* Nothing to do. */
6714 case OFPACT_MULTIPATH:
6715 multipath_execute(ofpact_get_MULTIPATH(a), &ctx->xin->flow);
6719 ctx->ofproto->has_bundle_action = true;
6720 xlate_bundle_action(ctx, ofpact_get_BUNDLE(a));
6723 case OFPACT_OUTPUT_REG:
6724 xlate_output_reg_action(ctx, ofpact_get_OUTPUT_REG(a));
6728 ctx->xout->has_learn = true;
6729 if (ctx->xin->may_learn) {
6730 xlate_learn_action(ctx, ofpact_get_LEARN(a));
6738 case OFPACT_FIN_TIMEOUT:
6739 ctx->xout->has_fin_timeout = true;
6740 xlate_fin_timeout(ctx, ofpact_get_FIN_TIMEOUT(a));
6743 case OFPACT_CLEAR_ACTIONS:
6745 * Nothing to do because writa-actions is not supported for now.
6746 * When writa-actions is supported, clear-actions also must
6747 * be supported at the same time.
6751 case OFPACT_WRITE_METADATA:
6752 metadata = ofpact_get_WRITE_METADATA(a);
6753 ctx->xin->flow.metadata &= ~metadata->mask;
6754 ctx->xin->flow.metadata |= metadata->metadata & metadata->mask;
6757 case OFPACT_GOTO_TABLE: {
6758 /* It is assumed that goto-table is the last action. */
6759 struct ofpact_goto_table *ogt = ofpact_get_GOTO_TABLE(a);
6760 struct rule_dpif *rule;
6762 ovs_assert(ctx->table_id < ogt->table_id);
6764 ctx->table_id = ogt->table_id;
6766 /* Look up a flow from the new table. */
6767 rule = rule_dpif_lookup__(ctx->ofproto, &ctx->xin->flow, ctx->table_id);
6769 tag_the_flow(ctx, rule);
6771 rule = ctx_rule_hooks(ctx, rule, true);
6775 ctx->rule->up.evictable = was_evictable;
6778 was_evictable = rule->up.evictable;
6779 rule->up.evictable = false;
6781 /* Tail recursion removal. */
6782 ofpacts = rule->up.ofpacts;
6783 ofpacts_len = rule->up.ofpacts_len;
6784 goto do_xlate_actions_again;
6790 xlate_sample_action(ctx, ofpact_get_SAMPLE(a));
6797 ctx->rule->up.evictable = was_evictable;
6802 xlate_in_init(struct xlate_in *xin, struct ofproto_dpif *ofproto,
6803 const struct flow *flow,
6804 const struct initial_vals *initial_vals,
6805 struct rule_dpif *rule, uint8_t tcp_flags,
6806 const struct ofpbuf *packet)
6808 xin->ofproto = ofproto;
6810 xin->packet = packet;
6811 xin->may_learn = packet != NULL;
6813 xin->ofpacts = NULL;
6814 xin->ofpacts_len = 0;
6815 xin->tcp_flags = tcp_flags;
6816 xin->resubmit_hook = NULL;
6817 xin->report_hook = NULL;
6818 xin->resubmit_stats = NULL;
6821 xin->initial_vals = *initial_vals;
6823 xin->initial_vals.vlan_tci = xin->flow.vlan_tci;
6828 xlate_out_uninit(struct xlate_out *xout)
6831 ofpbuf_uninit(&xout->odp_actions);
6835 /* Translates the 'ofpacts_len' bytes of "struct ofpacts" starting at 'ofpacts'
6836 * into datapath actions in 'odp_actions', using 'ctx'. */
6838 xlate_actions(struct xlate_in *xin, struct xlate_out *xout)
6840 /* Normally false. Set to true if we ever hit MAX_RESUBMIT_RECURSION, so
6841 * that in the future we always keep a copy of the original flow for
6842 * tracing purposes. */
6843 static bool hit_resubmit_limit;
6845 enum slow_path_reason special;
6846 const struct ofpact *ofpacts;
6847 struct ofport_dpif *in_port;
6848 struct flow orig_flow;
6849 struct xlate_ctx ctx;
6852 COVERAGE_INC(ofproto_dpif_xlate);
6854 /* Flow initialization rules:
6855 * - 'base_flow' must match the kernel's view of the packet at the
6856 * time that action processing starts. 'flow' represents any
6857 * transformations we wish to make through actions.
6858 * - By default 'base_flow' and 'flow' are the same since the input
6859 * packet matches the output before any actions are applied.
6860 * - When using VLAN splinters, 'base_flow''s VLAN is set to the value
6861 * of the received packet as seen by the kernel. If we later output
6862 * to another device without any modifications this will cause us to
6863 * insert a new tag since the original one was stripped off by the
6865 * - Tunnel metadata as received is retained in 'flow'. This allows
6866 * tunnel metadata matching also in later tables.
6867 * Since a kernel action for setting the tunnel metadata will only be
6868 * generated with actual tunnel output, changing the tunnel metadata
6869 * values in 'flow' (such as tun_id) will only have effect with a later
6870 * tunnel output action.
6871 * - Tunnel 'base_flow' is completely cleared since that is what the
6872 * kernel does. If we wish to maintain the original values an action
6873 * needs to be generated. */
6878 ctx.ofproto = xin->ofproto;
6879 ctx.rule = xin->rule;
6881 ctx.base_flow = ctx.xin->flow;
6882 ctx.base_flow.vlan_tci = xin->initial_vals.vlan_tci;
6883 memset(&ctx.base_flow.tunnel, 0, sizeof ctx.base_flow.tunnel);
6884 ctx.orig_tunnel_ip_dst = ctx.xin->flow.tunnel.ip_dst;
6888 ctx.xout->has_learn = false;
6889 ctx.xout->has_normal = false;
6890 ctx.xout->has_fin_timeout = false;
6891 ctx.xout->nf_output_iface = NF_OUT_DROP;
6892 ctx.xout->mirrors = 0;
6894 ofpbuf_use_stub(&ctx.xout->odp_actions, ctx.xout->odp_actions_stub,
6895 sizeof ctx.xout->odp_actions_stub);
6896 ofpbuf_reserve(&ctx.xout->odp_actions, NL_A_U32_SIZE);
6899 ctx.max_resubmit_trigger = false;
6900 ctx.orig_skb_priority = ctx.xin->flow.skb_priority;
6905 ofpacts = xin->ofpacts;
6906 ofpacts_len = xin->ofpacts_len;
6907 } else if (xin->rule) {
6908 ofpacts = xin->rule->up.ofpacts;
6909 ofpacts_len = xin->rule->up.ofpacts_len;
6914 ofpbuf_use_stub(&ctx.stack, ctx.init_stack, sizeof ctx.init_stack);
6916 if (ctx.ofproto->has_mirrors || hit_resubmit_limit) {
6917 /* Do this conditionally because the copy is expensive enough that it
6918 * shows up in profiles. */
6919 orig_flow = ctx.xin->flow;
6922 if (ctx.xin->flow.nw_frag & FLOW_NW_FRAG_ANY) {
6923 switch (ctx.ofproto->up.frag_handling) {
6924 case OFPC_FRAG_NORMAL:
6925 /* We must pretend that transport ports are unavailable. */
6926 ctx.xin->flow.tp_src = ctx.base_flow.tp_src = htons(0);
6927 ctx.xin->flow.tp_dst = ctx.base_flow.tp_dst = htons(0);
6930 case OFPC_FRAG_DROP:
6933 case OFPC_FRAG_REASM:
6936 case OFPC_FRAG_NX_MATCH:
6937 /* Nothing to do. */
6940 case OFPC_INVALID_TTL_TO_CONTROLLER:
6945 in_port = get_ofp_port(ctx.ofproto, ctx.xin->flow.in_port);
6946 special = process_special(ctx.ofproto, &ctx.xin->flow, in_port,
6949 ctx.xout->slow = special;
6951 static struct vlog_rate_limit trace_rl = VLOG_RATE_LIMIT_INIT(1, 1);
6952 struct initial_vals initial_vals;
6953 size_t sample_actions_len;
6954 uint32_t local_odp_port;
6956 initial_vals.vlan_tci = ctx.base_flow.vlan_tci;
6958 add_sflow_action(&ctx);
6959 add_ipfix_action(&ctx);
6960 sample_actions_len = ctx.xout->odp_actions.size;
6962 if (tunnel_ecn_ok(&ctx) && (!in_port || may_receive(in_port, &ctx))) {
6963 do_xlate_actions(ofpacts, ofpacts_len, &ctx);
6965 /* We've let OFPP_NORMAL and the learning action look at the
6966 * packet, so drop it now if forwarding is disabled. */
6967 if (in_port && !stp_forward_in_state(in_port->stp_state)) {
6968 ctx.xout->odp_actions.size = sample_actions_len;
6972 if (ctx.max_resubmit_trigger && !ctx.xin->resubmit_hook) {
6973 if (!hit_resubmit_limit) {
6974 /* We didn't record the original flow. Make sure we do from
6976 hit_resubmit_limit = true;
6977 } else if (!VLOG_DROP_ERR(&trace_rl)) {
6978 struct ds ds = DS_EMPTY_INITIALIZER;
6980 ofproto_trace(ctx.ofproto, &orig_flow, ctx.xin->packet,
6981 &initial_vals, &ds);
6982 VLOG_ERR("Trace triggered by excessive resubmit "
6983 "recursion:\n%s", ds_cstr(&ds));
6988 local_odp_port = ofp_port_to_odp_port(ctx.ofproto, OFPP_LOCAL);
6989 if (!connmgr_must_output_local(ctx.ofproto->up.connmgr, &ctx.xin->flow,
6991 ctx.xout->odp_actions.data,
6992 ctx.xout->odp_actions.size)) {
6993 compose_output_action(&ctx, OFPP_LOCAL);
6995 if (ctx.ofproto->has_mirrors) {
6996 add_mirror_actions(&ctx, &orig_flow);
6998 fix_sflow_action(&ctx);
7001 ofpbuf_uninit(&ctx.stack);
7004 /* Translates the 'ofpacts_len' bytes of "struct ofpact"s starting at 'ofpacts'
7005 * into datapath actions, using 'ctx', and discards the datapath actions. */
7007 xlate_actions_for_side_effects(struct xlate_in *xin)
7009 struct xlate_out xout;
7011 xlate_actions(xin, &xout);
7012 xlate_out_uninit(&xout);
7016 xlate_report(struct xlate_ctx *ctx, const char *s)
7018 if (ctx->xin->report_hook) {
7019 ctx->xin->report_hook(ctx, s);
7023 /* OFPP_NORMAL implementation. */
7025 static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *);
7027 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
7028 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle',
7029 * the bundle on which the packet was received, returns the VLAN to which the
7032 * Both 'vid' and the return value are in the range 0...4095. */
7034 input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid)
7036 switch (in_bundle->vlan_mode) {
7037 case PORT_VLAN_ACCESS:
7038 return in_bundle->vlan;
7041 case PORT_VLAN_TRUNK:
7044 case PORT_VLAN_NATIVE_UNTAGGED:
7045 case PORT_VLAN_NATIVE_TAGGED:
7046 return vid ? vid : in_bundle->vlan;
7053 /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'.
7054 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
7057 * 'vid' should be the VID obtained from the 802.1Q header that was received as
7058 * part of a packet (specify 0 if there was no 802.1Q header), in the range
7061 input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn)
7063 /* Allow any VID on the OFPP_NONE port. */
7064 if (in_bundle == &ofpp_none_bundle) {
7068 switch (in_bundle->vlan_mode) {
7069 case PORT_VLAN_ACCESS:
7072 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7073 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
7074 "packet received on port %s configured as VLAN "
7075 "%"PRIu16" access port",
7076 in_bundle->ofproto->up.name, vid,
7077 in_bundle->name, in_bundle->vlan);
7083 case PORT_VLAN_NATIVE_UNTAGGED:
7084 case PORT_VLAN_NATIVE_TAGGED:
7086 /* Port must always carry its native VLAN. */
7090 case PORT_VLAN_TRUNK:
7091 if (!ofbundle_includes_vlan(in_bundle, vid)) {
7093 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7094 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet "
7095 "received on port %s not configured for trunking "
7097 in_bundle->ofproto->up.name, vid,
7098 in_bundle->name, vid);
7110 /* Given 'vlan', the VLAN that a packet belongs to, and
7111 * 'out_bundle', a bundle on which the packet is to be output, returns the VID
7112 * that should be included in the 802.1Q header. (If the return value is 0,
7113 * then the 802.1Q header should only be included in the packet if there is a
7116 * Both 'vlan' and the return value are in the range 0...4095. */
7118 output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan)
7120 switch (out_bundle->vlan_mode) {
7121 case PORT_VLAN_ACCESS:
7124 case PORT_VLAN_TRUNK:
7125 case PORT_VLAN_NATIVE_TAGGED:
7128 case PORT_VLAN_NATIVE_UNTAGGED:
7129 return vlan == out_bundle->vlan ? 0 : vlan;
7137 output_normal(struct xlate_ctx *ctx, const struct ofbundle *out_bundle,
7140 struct ofport_dpif *port;
7142 ovs_be16 tci, old_tci;
7144 vid = output_vlan_to_vid(out_bundle, vlan);
7145 if (!out_bundle->bond) {
7146 port = ofbundle_get_a_port(out_bundle);
7148 port = bond_choose_output_slave(out_bundle->bond, &ctx->xin->flow,
7149 vid, &ctx->xout->tags);
7151 /* No slaves enabled, so drop packet. */
7156 old_tci = ctx->xin->flow.vlan_tci;
7158 if (tci || out_bundle->use_priority_tags) {
7159 tci |= ctx->xin->flow.vlan_tci & htons(VLAN_PCP_MASK);
7161 tci |= htons(VLAN_CFI);
7164 ctx->xin->flow.vlan_tci = tci;
7166 compose_output_action(ctx, port->up.ofp_port);
7167 ctx->xin->flow.vlan_tci = old_tci;
7171 mirror_mask_ffs(mirror_mask_t mask)
7173 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
7178 ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
7180 return (bundle->vlan_mode != PORT_VLAN_ACCESS
7181 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
7185 ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan)
7187 return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan);
7190 /* Returns an arbitrary interface within 'bundle'. */
7191 static struct ofport_dpif *
7192 ofbundle_get_a_port(const struct ofbundle *bundle)
7194 return CONTAINER_OF(list_front(&bundle->ports),
7195 struct ofport_dpif, bundle_node);
7199 vlan_is_mirrored(const struct ofmirror *m, int vlan)
7201 return !m->vlans || bitmap_is_set(m->vlans, vlan);
7205 add_mirror_actions(struct xlate_ctx *ctx, const struct flow *orig_flow)
7207 struct ofproto_dpif *ofproto = ctx->ofproto;
7208 mirror_mask_t mirrors;
7209 struct ofbundle *in_bundle;
7212 const struct nlattr *a;
7215 in_bundle = lookup_input_bundle(ctx->ofproto, orig_flow->in_port,
7216 ctx->xin->packet != NULL, NULL);
7220 mirrors = in_bundle->src_mirrors;
7222 /* Drop frames on bundles reserved for mirroring. */
7223 if (in_bundle->mirror_out) {
7224 if (ctx->xin->packet != NULL) {
7225 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7226 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
7227 "%s, which is reserved exclusively for mirroring",
7228 ctx->ofproto->up.name, in_bundle->name);
7234 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
7235 if (!input_vid_is_valid(vid, in_bundle, ctx->xin->packet != NULL)) {
7238 vlan = input_vid_to_vlan(in_bundle, vid);
7240 /* Look at the output ports to check for destination selections. */
7242 NL_ATTR_FOR_EACH (a, left, ctx->xout->odp_actions.data,
7243 ctx->xout->odp_actions.size) {
7244 enum ovs_action_attr type = nl_attr_type(a);
7245 struct ofport_dpif *ofport;
7247 if (type != OVS_ACTION_ATTR_OUTPUT) {
7251 ofport = get_odp_port(ofproto, nl_attr_get_u32(a));
7252 if (ofport && ofport->bundle) {
7253 mirrors |= ofport->bundle->dst_mirrors;
7261 /* Restore the original packet before adding the mirror actions. */
7262 ctx->xin->flow = *orig_flow;
7267 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
7269 if (!vlan_is_mirrored(m, vlan)) {
7270 mirrors = zero_rightmost_1bit(mirrors);
7274 mirrors &= ~m->dup_mirrors;
7275 ctx->xout->mirrors |= m->dup_mirrors;
7277 output_normal(ctx, m->out, vlan);
7278 } else if (vlan != m->out_vlan
7279 && !eth_addr_is_reserved(orig_flow->dl_dst)) {
7280 struct ofbundle *bundle;
7282 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
7283 if (ofbundle_includes_vlan(bundle, m->out_vlan)
7284 && !bundle->mirror_out) {
7285 output_normal(ctx, bundle, m->out_vlan);
7293 update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors,
7294 uint64_t packets, uint64_t bytes)
7300 for (; mirrors; mirrors = zero_rightmost_1bit(mirrors)) {
7303 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
7306 /* In normal circumstances 'm' will not be NULL. However,
7307 * if mirrors are reconfigured, we can temporarily get out
7308 * of sync in facet_revalidate(). We could "correct" the
7309 * mirror list before reaching here, but doing that would
7310 * not properly account the traffic stats we've currently
7311 * accumulated for previous mirror configuration. */
7315 m->packet_count += packets;
7316 m->byte_count += bytes;
7320 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
7321 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
7322 * indicate this; newer upstream kernels use gratuitous ARP requests. */
7324 is_gratuitous_arp(const struct flow *flow)
7326 return (flow->dl_type == htons(ETH_TYPE_ARP)
7327 && eth_addr_is_broadcast(flow->dl_dst)
7328 && (flow->nw_proto == ARP_OP_REPLY
7329 || (flow->nw_proto == ARP_OP_REQUEST
7330 && flow->nw_src == flow->nw_dst)));
7334 update_learning_table(struct ofproto_dpif *ofproto,
7335 const struct flow *flow, int vlan,
7336 struct ofbundle *in_bundle)
7338 struct mac_entry *mac;
7340 /* Don't learn the OFPP_NONE port. */
7341 if (in_bundle == &ofpp_none_bundle) {
7345 if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
7349 mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan);
7350 if (is_gratuitous_arp(flow)) {
7351 /* We don't want to learn from gratuitous ARP packets that are
7352 * reflected back over bond slaves so we lock the learning table. */
7353 if (!in_bundle->bond) {
7354 mac_entry_set_grat_arp_lock(mac);
7355 } else if (mac_entry_is_grat_arp_locked(mac)) {
7360 if (mac_entry_is_new(mac) || mac->port.p != in_bundle) {
7361 /* The log messages here could actually be useful in debugging,
7362 * so keep the rate limit relatively high. */
7363 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
7364 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
7365 "on port %s in VLAN %d",
7366 ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src),
7367 in_bundle->name, vlan);
7369 mac->port.p = in_bundle;
7370 tag_set_add(&ofproto->backer->revalidate_set,
7371 mac_learning_changed(ofproto->ml, mac));
7375 static struct ofbundle *
7376 lookup_input_bundle(const struct ofproto_dpif *ofproto, uint16_t in_port,
7377 bool warn, struct ofport_dpif **in_ofportp)
7379 struct ofport_dpif *ofport;
7381 /* Find the port and bundle for the received packet. */
7382 ofport = get_ofp_port(ofproto, in_port);
7384 *in_ofportp = ofport;
7386 if (ofport && ofport->bundle) {
7387 return ofport->bundle;
7390 /* Special-case OFPP_NONE, which a controller may use as the ingress
7391 * port for traffic that it is sourcing. */
7392 if (in_port == OFPP_NONE) {
7393 return &ofpp_none_bundle;
7396 /* Odd. A few possible reasons here:
7398 * - We deleted a port but there are still a few packets queued up
7401 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
7402 * we don't know about.
7404 * - The ofproto client didn't configure the port as part of a bundle.
7405 * This is particularly likely to happen if a packet was received on the
7406 * port after it was created, but before the client had a chance to
7407 * configure its bundle.
7410 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7412 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
7413 "port %"PRIu16, ofproto->up.name, in_port);
7418 /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or
7419 * dropped. Returns true if they may be forwarded, false if they should be
7422 * 'in_port' must be the ofport_dpif that corresponds to flow->in_port.
7423 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
7425 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
7426 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
7427 * checked by input_vid_is_valid().
7429 * May also add tags to '*tags', although the current implementation only does
7430 * so in one special case.
7433 is_admissible(struct xlate_ctx *ctx, struct ofport_dpif *in_port,
7436 struct ofproto_dpif *ofproto = ctx->ofproto;
7437 struct flow *flow = &ctx->xin->flow;
7438 struct ofbundle *in_bundle = in_port->bundle;
7440 /* Drop frames for reserved multicast addresses
7441 * only if forward_bpdu option is absent. */
7442 if (!ofproto->up.forward_bpdu && eth_addr_is_reserved(flow->dl_dst)) {
7443 xlate_report(ctx, "packet has reserved destination MAC, dropping");
7447 if (in_bundle->bond) {
7448 struct mac_entry *mac;
7450 switch (bond_check_admissibility(in_bundle->bond, in_port,
7451 flow->dl_dst, &ctx->xout->tags)) {
7456 xlate_report(ctx, "bonding refused admissibility, dropping");
7459 case BV_DROP_IF_MOVED:
7460 mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL);
7461 if (mac && mac->port.p != in_bundle &&
7462 (!is_gratuitous_arp(flow)
7463 || mac_entry_is_grat_arp_locked(mac))) {
7464 xlate_report(ctx, "SLB bond thinks this packet looped back, "
7476 xlate_normal(struct xlate_ctx *ctx)
7478 struct ofport_dpif *in_port;
7479 struct ofbundle *in_bundle;
7480 struct mac_entry *mac;
7484 ctx->xout->has_normal = true;
7486 in_bundle = lookup_input_bundle(ctx->ofproto, ctx->xin->flow.in_port,
7487 ctx->xin->packet != NULL, &in_port);
7489 xlate_report(ctx, "no input bundle, dropping");
7493 /* Drop malformed frames. */
7494 if (ctx->xin->flow.dl_type == htons(ETH_TYPE_VLAN) &&
7495 !(ctx->xin->flow.vlan_tci & htons(VLAN_CFI))) {
7496 if (ctx->xin->packet != NULL) {
7497 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7498 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
7499 "VLAN tag received on port %s",
7500 ctx->ofproto->up.name, in_bundle->name);
7502 xlate_report(ctx, "partial VLAN tag, dropping");
7506 /* Drop frames on bundles reserved for mirroring. */
7507 if (in_bundle->mirror_out) {
7508 if (ctx->xin->packet != NULL) {
7509 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7510 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
7511 "%s, which is reserved exclusively for mirroring",
7512 ctx->ofproto->up.name, in_bundle->name);
7514 xlate_report(ctx, "input port is mirror output port, dropping");
7519 vid = vlan_tci_to_vid(ctx->xin->flow.vlan_tci);
7520 if (!input_vid_is_valid(vid, in_bundle, ctx->xin->packet != NULL)) {
7521 xlate_report(ctx, "disallowed VLAN VID for this input port, dropping");
7524 vlan = input_vid_to_vlan(in_bundle, vid);
7526 /* Check other admissibility requirements. */
7527 if (in_port && !is_admissible(ctx, in_port, vlan)) {
7531 /* Learn source MAC. */
7532 if (ctx->xin->may_learn) {
7533 update_learning_table(ctx->ofproto, &ctx->xin->flow, vlan, in_bundle);
7536 /* Determine output bundle. */
7537 mac = mac_learning_lookup(ctx->ofproto->ml, ctx->xin->flow.dl_dst, vlan,
7540 if (mac->port.p != in_bundle) {
7541 xlate_report(ctx, "forwarding to learned port");
7542 output_normal(ctx, mac->port.p, vlan);
7544 xlate_report(ctx, "learned port is input port, dropping");
7547 struct ofbundle *bundle;
7549 xlate_report(ctx, "no learned MAC for destination, flooding");
7550 HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
7551 if (bundle != in_bundle
7552 && ofbundle_includes_vlan(bundle, vlan)
7553 && bundle->floodable
7554 && !bundle->mirror_out) {
7555 output_normal(ctx, bundle, vlan);
7558 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
7562 /* Optimized flow revalidation.
7564 * It's a difficult problem, in general, to tell which facets need to have
7565 * their actions recalculated whenever the OpenFlow flow table changes. We
7566 * don't try to solve that general problem: for most kinds of OpenFlow flow
7567 * table changes, we recalculate the actions for every facet. This is
7568 * relatively expensive, but it's good enough if the OpenFlow flow table
7569 * doesn't change very often.
7571 * However, we can expect one particular kind of OpenFlow flow table change to
7572 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
7573 * of CPU on revalidating every facet whenever MAC learning modifies the flow
7574 * table, we add a special case that applies to flow tables in which every rule
7575 * has the same form (that is, the same wildcards), except that the table is
7576 * also allowed to have a single "catch-all" flow that matches all packets. We
7577 * optimize this case by tagging all of the facets that resubmit into the table
7578 * and invalidating the same tag whenever a flow changes in that table. The
7579 * end result is that we revalidate just the facets that need it (and sometimes
7580 * a few more, but not all of the facets or even all of the facets that
7581 * resubmit to the table modified by MAC learning). */
7583 /* Calculates the tag to use for 'flow' and mask 'mask' when it is inserted
7584 * into an OpenFlow table with the given 'basis'. */
7586 rule_calculate_tag(const struct flow *flow, const struct minimask *mask,
7589 if (minimask_is_catchall(mask)) {
7592 uint32_t hash = flow_hash_in_minimask(flow, mask, secret);
7593 return tag_create_deterministic(hash);
7597 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
7598 * taggability of that table.
7600 * This function must be called after *each* change to a flow table. If you
7601 * skip calling it on some changes then the pointer comparisons at the end can
7602 * be invalid if you get unlucky. For example, if a flow removal causes a
7603 * cls_table to be destroyed and then a flow insertion causes a cls_table with
7604 * different wildcards to be created with the same address, then this function
7605 * will incorrectly skip revalidation. */
7607 table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
7609 struct table_dpif *table = &ofproto->tables[table_id];
7610 const struct oftable *oftable = &ofproto->up.tables[table_id];
7611 struct cls_table *catchall, *other;
7612 struct cls_table *t;
7614 catchall = other = NULL;
7616 switch (hmap_count(&oftable->cls.tables)) {
7618 /* We could tag this OpenFlow table but it would make the logic a
7619 * little harder and it's a corner case that doesn't seem worth it
7625 HMAP_FOR_EACH (t, hmap_node, &oftable->cls.tables) {
7626 if (cls_table_is_catchall(t)) {
7628 } else if (!other) {
7631 /* Indicate that we can't tag this by setting both tables to
7632 * NULL. (We know that 'catchall' is already NULL.) */
7639 /* Can't tag this table. */
7643 if (table->catchall_table != catchall || table->other_table != other) {
7644 table->catchall_table = catchall;
7645 table->other_table = other;
7646 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
7650 /* Given 'rule' that has changed in some way (either it is a rule being
7651 * inserted, a rule being deleted, or a rule whose actions are being
7652 * modified), marks facets for revalidation to ensure that packets will be
7653 * forwarded correctly according to the new state of the flow table.
7655 * This function must be called after *each* change to a flow table. See
7656 * the comment on table_update_taggable() for more information. */
7658 rule_invalidate(const struct rule_dpif *rule)
7660 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
7662 table_update_taggable(ofproto, rule->up.table_id);
7664 if (!ofproto->backer->need_revalidate) {
7665 struct table_dpif *table = &ofproto->tables[rule->up.table_id];
7667 if (table->other_table && rule->tag) {
7668 tag_set_add(&ofproto->backer->revalidate_set, rule->tag);
7670 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
7676 set_frag_handling(struct ofproto *ofproto_,
7677 enum ofp_config_flags frag_handling)
7679 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7680 if (frag_handling != OFPC_FRAG_REASM) {
7681 ofproto->backer->need_revalidate = REV_RECONFIGURE;
7689 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
7690 const struct flow *flow,
7691 const struct ofpact *ofpacts, size_t ofpacts_len)
7693 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7694 struct initial_vals initial_vals;
7695 struct odputil_keybuf keybuf;
7696 struct dpif_flow_stats stats;
7697 struct xlate_out xout;
7698 struct xlate_in xin;
7702 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
7703 odp_flow_key_from_flow(&key, flow,
7704 ofp_port_to_odp_port(ofproto, flow->in_port));
7706 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
7708 initial_vals.vlan_tci = flow->vlan_tci;
7709 xlate_in_init(&xin, ofproto, flow, &initial_vals, NULL, stats.tcp_flags,
7711 xin.resubmit_stats = &stats;
7712 xin.ofpacts_len = ofpacts_len;
7713 xin.ofpacts = ofpacts;
7715 xlate_actions(&xin, &xout);
7716 dpif_execute(ofproto->backer->dpif, key.data, key.size,
7717 xout.odp_actions.data, xout.odp_actions.size, packet);
7718 xlate_out_uninit(&xout);
7726 set_netflow(struct ofproto *ofproto_,
7727 const struct netflow_options *netflow_options)
7729 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7731 if (netflow_options) {
7732 if (!ofproto->netflow) {
7733 ofproto->netflow = netflow_create();
7735 return netflow_set_options(ofproto->netflow, netflow_options);
7737 netflow_destroy(ofproto->netflow);
7738 ofproto->netflow = NULL;
7744 get_netflow_ids(const struct ofproto *ofproto_,
7745 uint8_t *engine_type, uint8_t *engine_id)
7747 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7749 dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
7753 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
7755 if (!facet_is_controller_flow(facet) &&
7756 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
7757 struct subfacet *subfacet;
7758 struct ofexpired expired;
7760 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
7761 if (subfacet->path == SF_FAST_PATH) {
7762 struct dpif_flow_stats stats;
7764 subfacet_install(subfacet, &facet->xout.odp_actions, &stats);
7765 subfacet_update_stats(subfacet, &stats);
7769 expired.flow = facet->flow;
7770 expired.packet_count = facet->packet_count;
7771 expired.byte_count = facet->byte_count;
7772 expired.used = facet->used;
7773 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
7778 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
7780 struct facet *facet;
7782 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
7783 send_active_timeout(ofproto, facet);
7787 static struct ofproto_dpif *
7788 ofproto_dpif_lookup(const char *name)
7790 struct ofproto_dpif *ofproto;
7792 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
7793 hash_string(name, 0), &all_ofproto_dpifs) {
7794 if (!strcmp(ofproto->up.name, name)) {
7802 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
7803 const char *argv[], void *aux OVS_UNUSED)
7805 struct ofproto_dpif *ofproto;
7808 ofproto = ofproto_dpif_lookup(argv[1]);
7810 unixctl_command_reply_error(conn, "no such bridge");
7813 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
7815 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
7816 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
7820 unixctl_command_reply(conn, "table successfully flushed");
7824 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
7825 const char *argv[], void *aux OVS_UNUSED)
7827 struct ds ds = DS_EMPTY_INITIALIZER;
7828 const struct ofproto_dpif *ofproto;
7829 const struct mac_entry *e;
7831 ofproto = ofproto_dpif_lookup(argv[1]);
7833 unixctl_command_reply_error(conn, "no such bridge");
7837 ds_put_cstr(&ds, " port VLAN MAC Age\n");
7838 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
7839 struct ofbundle *bundle = e->port.p;
7840 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
7841 ofbundle_get_a_port(bundle)->odp_port,
7842 e->vlan, ETH_ADDR_ARGS(e->mac),
7843 mac_entry_age(ofproto->ml, e));
7845 unixctl_command_reply(conn, ds_cstr(&ds));
7850 struct xlate_out xout;
7851 struct xlate_in xin;
7857 trace_format_rule(struct ds *result, uint8_t table_id, int level,
7858 const struct rule_dpif *rule)
7860 ds_put_char_multiple(result, '\t', level);
7862 ds_put_cstr(result, "No match\n");
7866 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
7867 table_id, ntohll(rule->up.flow_cookie));
7868 cls_rule_format(&rule->up.cr, result);
7869 ds_put_char(result, '\n');
7871 ds_put_char_multiple(result, '\t', level);
7872 ds_put_cstr(result, "OpenFlow ");
7873 ofpacts_format(rule->up.ofpacts, rule->up.ofpacts_len, result);
7874 ds_put_char(result, '\n');
7878 trace_format_flow(struct ds *result, int level, const char *title,
7879 struct trace_ctx *trace)
7881 ds_put_char_multiple(result, '\t', level);
7882 ds_put_format(result, "%s: ", title);
7883 if (flow_equal(&trace->xin.flow, &trace->flow)) {
7884 ds_put_cstr(result, "unchanged");
7886 flow_format(result, &trace->xin.flow);
7887 trace->flow = trace->xin.flow;
7889 ds_put_char(result, '\n');
7893 trace_format_regs(struct ds *result, int level, const char *title,
7894 struct trace_ctx *trace)
7898 ds_put_char_multiple(result, '\t', level);
7899 ds_put_format(result, "%s:", title);
7900 for (i = 0; i < FLOW_N_REGS; i++) {
7901 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
7903 ds_put_char(result, '\n');
7907 trace_format_odp(struct ds *result, int level, const char *title,
7908 struct trace_ctx *trace)
7910 struct ofpbuf *odp_actions = &trace->xout.odp_actions;
7912 ds_put_char_multiple(result, '\t', level);
7913 ds_put_format(result, "%s: ", title);
7914 format_odp_actions(result, odp_actions->data, odp_actions->size);
7915 ds_put_char(result, '\n');
7919 trace_resubmit(struct xlate_ctx *ctx, struct rule_dpif *rule)
7921 struct trace_ctx *trace = CONTAINER_OF(ctx->xin, struct trace_ctx, xin);
7922 struct ds *result = trace->result;
7924 ds_put_char(result, '\n');
7925 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
7926 trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace);
7927 trace_format_odp(result, ctx->recurse + 1, "Resubmitted odp", trace);
7928 trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule);
7932 trace_report(struct xlate_ctx *ctx, const char *s)
7934 struct trace_ctx *trace = CONTAINER_OF(ctx->xin, struct trace_ctx, xin);
7935 struct ds *result = trace->result;
7937 ds_put_char_multiple(result, '\t', ctx->recurse);
7938 ds_put_cstr(result, s);
7939 ds_put_char(result, '\n');
7943 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
7944 void *aux OVS_UNUSED)
7946 const struct dpif_backer *backer;
7947 struct ofproto_dpif *ofproto;
7948 struct ofpbuf odp_key;
7949 struct ofpbuf *packet;
7950 struct initial_vals initial_vals;
7958 ofpbuf_init(&odp_key, 0);
7960 /* Handle "-generate" or a hex string as the last argument. */
7961 if (!strcmp(argv[argc - 1], "-generate")) {
7962 packet = ofpbuf_new(0);
7965 const char *error = eth_from_hex(argv[argc - 1], &packet);
7968 } else if (argc == 4) {
7969 /* The 3-argument form must end in "-generate' or a hex string. */
7970 unixctl_command_reply_error(conn, error);
7975 /* Parse the flow and determine whether a datapath or
7976 * bridge is specified. If function odp_flow_key_from_string()
7977 * returns 0, the flow is a odp_flow. If function
7978 * parse_ofp_exact_flow() returns 0, the flow is a br_flow. */
7979 if (!odp_flow_key_from_string(argv[argc - 1], NULL, &odp_key)) {
7980 /* If the odp_flow is the second argument,
7981 * the datapath name is the first argument. */
7983 const char *dp_type;
7984 if (!strncmp(argv[1], "ovs-", 4)) {
7985 dp_type = argv[1] + 4;
7989 backer = shash_find_data(&all_dpif_backers, dp_type);
7991 unixctl_command_reply_error(conn, "Cannot find datapath "
7996 /* No datapath name specified, so there should be only one
7998 struct shash_node *node;
7999 if (shash_count(&all_dpif_backers) != 1) {
8000 unixctl_command_reply_error(conn, "Must specify datapath "
8001 "name, there is more than one type of datapath");
8004 node = shash_first(&all_dpif_backers);
8005 backer = node->data;
8008 /* Extract the ofproto_dpif object from the ofproto_receive()
8010 if (ofproto_receive(backer, NULL, odp_key.data,
8011 odp_key.size, &flow, NULL, &ofproto, NULL,
8013 unixctl_command_reply_error(conn, "Invalid datapath flow");
8016 ds_put_format(&result, "Bridge: %s\n", ofproto->up.name);
8017 } else if (!parse_ofp_exact_flow(&flow, argv[argc - 1])) {
8019 unixctl_command_reply_error(conn, "Must specify bridge name");
8023 ofproto = ofproto_dpif_lookup(argv[1]);
8025 unixctl_command_reply_error(conn, "Unknown bridge name");
8028 initial_vals.vlan_tci = flow.vlan_tci;
8030 unixctl_command_reply_error(conn, "Bad flow syntax");
8034 /* Generate a packet, if requested. */
8036 if (!packet->size) {
8037 flow_compose(packet, &flow);
8039 ds_put_cstr(&result, "Packet: ");
8040 s = ofp_packet_to_string(packet->data, packet->size);
8041 ds_put_cstr(&result, s);
8044 /* Use the metadata from the flow and the packet argument
8045 * to reconstruct the flow. */
8046 flow_extract(packet, flow.skb_priority, flow.skb_mark, NULL,
8047 flow.in_port, &flow);
8048 initial_vals.vlan_tci = flow.vlan_tci;
8052 ofproto_trace(ofproto, &flow, packet, &initial_vals, &result);
8053 unixctl_command_reply(conn, ds_cstr(&result));
8056 ds_destroy(&result);
8057 ofpbuf_delete(packet);
8058 ofpbuf_uninit(&odp_key);
8062 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
8063 const struct ofpbuf *packet,
8064 const struct initial_vals *initial_vals, struct ds *ds)
8066 struct rule_dpif *rule;
8068 ds_put_cstr(ds, "Flow: ");
8069 flow_format(ds, flow);
8070 ds_put_char(ds, '\n');
8072 rule = rule_dpif_lookup(ofproto, flow);
8074 trace_format_rule(ds, 0, 0, rule);
8075 if (rule == ofproto->miss_rule) {
8076 ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
8077 } else if (rule == ofproto->no_packet_in_rule) {
8078 ds_put_cstr(ds, "\nNo match, packets dropped because "
8079 "OFPPC_NO_PACKET_IN is set on in_port.\n");
8083 uint64_t odp_actions_stub[1024 / 8];
8084 struct ofpbuf odp_actions;
8086 struct trace_ctx trace;
8089 tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
8092 ofpbuf_use_stub(&odp_actions,
8093 odp_actions_stub, sizeof odp_actions_stub);
8094 xlate_in_init(&trace.xin, ofproto, flow, initial_vals, rule, tcp_flags,
8096 trace.xin.resubmit_hook = trace_resubmit;
8097 trace.xin.report_hook = trace_report;
8098 xlate_actions(&trace.xin, &trace.xout);
8100 ds_put_char(ds, '\n');
8101 trace_format_flow(ds, 0, "Final flow", &trace);
8102 ds_put_cstr(ds, "Datapath actions: ");
8103 format_odp_actions(ds, trace.xout.odp_actions.data,
8104 trace.xout.odp_actions.size);
8106 if (trace.xout.slow) {
8107 ds_put_cstr(ds, "\nThis flow is handled by the userspace "
8108 "slow path because it:");
8109 switch (trace.xout.slow) {
8111 ds_put_cstr(ds, "\n\t- Consists of CFM packets.");
8114 ds_put_cstr(ds, "\n\t- Consists of LACP packets.");
8117 ds_put_cstr(ds, "\n\t- Consists of STP packets.");
8120 ds_put_cstr(ds, "\n\t- Consists of BFD packets.");
8122 case SLOW_CONTROLLER:
8123 ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages "
8124 "to the OpenFlow controller.");
8131 xlate_out_uninit(&trace.xout);
8136 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
8137 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
8140 unixctl_command_reply(conn, NULL);
8144 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
8145 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
8148 unixctl_command_reply(conn, NULL);
8151 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
8152 * 'reply' describing the results. */
8154 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
8156 struct facet *facet;
8160 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
8161 if (!facet_check_consistency(facet)) {
8166 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
8170 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
8171 ofproto->up.name, errors);
8173 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
8178 ofproto_dpif_self_check(struct unixctl_conn *conn,
8179 int argc, const char *argv[], void *aux OVS_UNUSED)
8181 struct ds reply = DS_EMPTY_INITIALIZER;
8182 struct ofproto_dpif *ofproto;
8185 ofproto = ofproto_dpif_lookup(argv[1]);
8187 unixctl_command_reply_error(conn, "Unknown ofproto (use "
8188 "ofproto/list for help)");
8191 ofproto_dpif_self_check__(ofproto, &reply);
8193 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
8194 ofproto_dpif_self_check__(ofproto, &reply);
8198 unixctl_command_reply(conn, ds_cstr(&reply));
8202 /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list
8203 * of the 'ofproto_shash' nodes. It is the responsibility of the caller
8204 * to destroy 'ofproto_shash' and free the returned value. */
8205 static const struct shash_node **
8206 get_ofprotos(struct shash *ofproto_shash)
8208 const struct ofproto_dpif *ofproto;
8210 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
8211 char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
8212 shash_add_nocopy(ofproto_shash, name, ofproto);
8215 return shash_sort(ofproto_shash);
8219 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
8220 const char *argv[] OVS_UNUSED,
8221 void *aux OVS_UNUSED)
8223 struct ds ds = DS_EMPTY_INITIALIZER;
8224 struct shash ofproto_shash;
8225 const struct shash_node **sorted_ofprotos;
8228 shash_init(&ofproto_shash);
8229 sorted_ofprotos = get_ofprotos(&ofproto_shash);
8230 for (i = 0; i < shash_count(&ofproto_shash); i++) {
8231 const struct shash_node *node = sorted_ofprotos[i];
8232 ds_put_format(&ds, "%s\n", node->name);
8235 shash_destroy(&ofproto_shash);
8236 free(sorted_ofprotos);
8238 unixctl_command_reply(conn, ds_cstr(&ds));
8243 show_dp_format(const struct ofproto_dpif *ofproto, struct ds *ds)
8245 const struct shash_node **ports;
8247 struct avg_subfacet_rates lifetime;
8248 unsigned long long int minutes;
8249 const int min_ms = 60 * 1000; /* milliseconds in one minute. */
8251 minutes = (time_msec() - ofproto->created) / min_ms;
8254 lifetime.add_rate = (double)ofproto->total_subfacet_add_count
8256 lifetime.del_rate = (double)ofproto->total_subfacet_del_count
8259 lifetime.add_rate = 0.0;
8260 lifetime.del_rate = 0.0;
8263 ds_put_format(ds, "%s (%s):\n", ofproto->up.name,
8264 dpif_name(ofproto->backer->dpif));
8266 "\tlookups: hit:%"PRIu64" missed:%"PRIu64"\n",
8267 ofproto->n_hit, ofproto->n_missed);
8268 ds_put_format(ds, "\tflows: cur: %zu, avg: %5.3f, max: %d,"
8269 " life span: %llu(ms)\n",
8270 hmap_count(&ofproto->subfacets),
8271 avg_subfacet_count(ofproto),
8272 ofproto->max_n_subfacet,
8273 avg_subfacet_life_span(ofproto));
8274 if (minutes >= 60) {
8275 show_dp_rates(ds, "\t\thourly avg:", &ofproto->hourly);
8277 if (minutes >= 60 * 24) {
8278 show_dp_rates(ds, "\t\tdaily avg:", &ofproto->daily);
8280 show_dp_rates(ds, "\t\toverall avg:", &lifetime);
8282 ports = shash_sort(&ofproto->up.port_by_name);
8283 for (i = 0; i < shash_count(&ofproto->up.port_by_name); i++) {
8284 const struct shash_node *node = ports[i];
8285 struct ofport *ofport = node->data;
8286 const char *name = netdev_get_name(ofport->netdev);
8287 const char *type = netdev_get_type(ofport->netdev);
8290 ds_put_format(ds, "\t%s %u/", name, ofport->ofp_port);
8292 odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
8293 if (odp_port != OVSP_NONE) {
8294 ds_put_format(ds, "%"PRIu32":", odp_port);
8296 ds_put_cstr(ds, "none:");
8299 if (strcmp(type, "system")) {
8300 struct netdev *netdev;
8303 ds_put_format(ds, " (%s", type);
8305 error = netdev_open(name, type, &netdev);
8310 error = netdev_get_config(netdev, &config);
8312 const struct smap_node **nodes;
8315 nodes = smap_sort(&config);
8316 for (i = 0; i < smap_count(&config); i++) {
8317 const struct smap_node *node = nodes[i];
8318 ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
8319 node->key, node->value);
8323 smap_destroy(&config);
8325 netdev_close(netdev);
8327 ds_put_char(ds, ')');
8329 ds_put_char(ds, '\n');
8335 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc,
8336 const char *argv[], void *aux OVS_UNUSED)
8338 struct ds ds = DS_EMPTY_INITIALIZER;
8339 const struct ofproto_dpif *ofproto;
8343 for (i = 1; i < argc; i++) {
8344 ofproto = ofproto_dpif_lookup(argv[i]);
8346 ds_put_format(&ds, "Unknown bridge %s (use dpif/dump-dps "
8347 "for help)", argv[i]);
8348 unixctl_command_reply_error(conn, ds_cstr(&ds));
8351 show_dp_format(ofproto, &ds);
8354 struct shash ofproto_shash;
8355 const struct shash_node **sorted_ofprotos;
8358 shash_init(&ofproto_shash);
8359 sorted_ofprotos = get_ofprotos(&ofproto_shash);
8360 for (i = 0; i < shash_count(&ofproto_shash); i++) {
8361 const struct shash_node *node = sorted_ofprotos[i];
8362 show_dp_format(node->data, &ds);
8365 shash_destroy(&ofproto_shash);
8366 free(sorted_ofprotos);
8369 unixctl_command_reply(conn, ds_cstr(&ds));
8374 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
8375 int argc OVS_UNUSED, const char *argv[],
8376 void *aux OVS_UNUSED)
8378 struct ds ds = DS_EMPTY_INITIALIZER;
8379 const struct ofproto_dpif *ofproto;
8380 struct subfacet *subfacet;
8382 ofproto = ofproto_dpif_lookup(argv[1]);
8384 unixctl_command_reply_error(conn, "no such bridge");
8388 update_stats(ofproto->backer);
8390 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
8391 struct facet *facet = subfacet->facet;
8393 odp_flow_key_format(subfacet->key, subfacet->key_len, &ds);
8395 ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:",
8396 subfacet->dp_packet_count, subfacet->dp_byte_count);
8397 if (subfacet->used) {
8398 ds_put_format(&ds, "%.3fs",
8399 (time_msec() - subfacet->used) / 1000.0);
8401 ds_put_format(&ds, "never");
8403 if (subfacet->facet->tcp_flags) {
8404 ds_put_cstr(&ds, ", flags:");
8405 packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags);
8408 ds_put_cstr(&ds, ", actions:");
8409 if (facet->xout.slow) {
8410 uint64_t slow_path_stub[128 / 8];
8411 const struct nlattr *actions;
8414 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
8415 slow_path_stub, sizeof slow_path_stub,
8416 &actions, &actions_len);
8417 format_odp_actions(&ds, actions, actions_len);
8419 format_odp_actions(&ds, facet->xout.odp_actions.data,
8420 facet->xout.odp_actions.size);
8422 ds_put_char(&ds, '\n');
8425 unixctl_command_reply(conn, ds_cstr(&ds));
8430 ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn,
8431 int argc OVS_UNUSED, const char *argv[],
8432 void *aux OVS_UNUSED)
8434 struct ds ds = DS_EMPTY_INITIALIZER;
8435 struct ofproto_dpif *ofproto;
8437 ofproto = ofproto_dpif_lookup(argv[1]);
8439 unixctl_command_reply_error(conn, "no such bridge");
8443 flush(&ofproto->up);
8445 unixctl_command_reply(conn, ds_cstr(&ds));
8450 ofproto_dpif_unixctl_init(void)
8452 static bool registered;
8458 unixctl_command_register(
8460 "[dp_name]|bridge odp_flow|br_flow [-generate|packet]",
8461 1, 3, ofproto_unixctl_trace, NULL);
8462 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
8463 ofproto_unixctl_fdb_flush, NULL);
8464 unixctl_command_register("fdb/show", "bridge", 1, 1,
8465 ofproto_unixctl_fdb_show, NULL);
8466 unixctl_command_register("ofproto/clog", "", 0, 0,
8467 ofproto_dpif_clog, NULL);
8468 unixctl_command_register("ofproto/unclog", "", 0, 0,
8469 ofproto_dpif_unclog, NULL);
8470 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
8471 ofproto_dpif_self_check, NULL);
8472 unixctl_command_register("dpif/dump-dps", "", 0, 0,
8473 ofproto_unixctl_dpif_dump_dps, NULL);
8474 unixctl_command_register("dpif/show", "[bridge]", 0, INT_MAX,
8475 ofproto_unixctl_dpif_show, NULL);
8476 unixctl_command_register("dpif/dump-flows", "bridge", 1, 1,
8477 ofproto_unixctl_dpif_dump_flows, NULL);
8478 unixctl_command_register("dpif/del-flows", "bridge", 1, 1,
8479 ofproto_unixctl_dpif_del_flows, NULL);
8482 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
8484 * This is deprecated. It is only for compatibility with broken device drivers
8485 * in old versions of Linux that do not properly support VLANs when VLAN
8486 * devices are not used. When broken device drivers are no longer in
8487 * widespread use, we will delete these interfaces. */
8490 set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid)
8492 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
8493 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
8495 if (realdev_ofp_port == ofport->realdev_ofp_port
8496 && vid == ofport->vlandev_vid) {
8500 ofproto->backer->need_revalidate = REV_RECONFIGURE;
8502 if (ofport->realdev_ofp_port) {
8505 if (realdev_ofp_port && ofport->bundle) {
8506 /* vlandevs are enslaved to their realdevs, so they are not allowed to
8507 * themselves be part of a bundle. */
8508 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
8511 ofport->realdev_ofp_port = realdev_ofp_port;
8512 ofport->vlandev_vid = vid;
8514 if (realdev_ofp_port) {
8515 vsp_add(ofport, realdev_ofp_port, vid);
8522 hash_realdev_vid(uint16_t realdev_ofp_port, int vid)
8524 return hash_2words(realdev_ofp_port, vid);
8527 /* Returns the ODP port number of the Linux VLAN device that corresponds to
8528 * 'vlan_tci' on the network device with port number 'realdev_odp_port' in
8529 * 'ofproto'. For example, given 'realdev_odp_port' of eth0 and 'vlan_tci' 9,
8530 * it would return the port number of eth0.9.
8532 * Unless VLAN splinters are enabled for port 'realdev_odp_port', this
8533 * function just returns its 'realdev_odp_port' argument. */
8535 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
8536 uint32_t realdev_odp_port, ovs_be16 vlan_tci)
8538 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
8539 uint16_t realdev_ofp_port;
8540 int vid = vlan_tci_to_vid(vlan_tci);
8541 const struct vlan_splinter *vsp;
8543 realdev_ofp_port = odp_port_to_ofp_port(ofproto, realdev_odp_port);
8544 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
8545 hash_realdev_vid(realdev_ofp_port, vid),
8546 &ofproto->realdev_vid_map) {
8547 if (vsp->realdev_ofp_port == realdev_ofp_port
8548 && vsp->vid == vid) {
8549 return ofp_port_to_odp_port(ofproto, vsp->vlandev_ofp_port);
8553 return realdev_odp_port;
8556 static struct vlan_splinter *
8557 vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port)
8559 struct vlan_splinter *vsp;
8561 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, hash_int(vlandev_ofp_port, 0),
8562 &ofproto->vlandev_map) {
8563 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
8571 /* Returns the OpenFlow port number of the "real" device underlying the Linux
8572 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
8573 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
8574 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
8575 * eth0 and store 9 in '*vid'.
8577 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
8578 * VLAN device. Unless VLAN splinters are enabled, this is what this function
8581 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
8582 uint16_t vlandev_ofp_port, int *vid)
8584 if (!hmap_is_empty(&ofproto->vlandev_map)) {
8585 const struct vlan_splinter *vsp;
8587 vsp = vlandev_find(ofproto, vlandev_ofp_port);
8592 return vsp->realdev_ofp_port;
8598 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
8599 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
8600 * 'flow->in_port' to the "real" device backing the VLAN device, sets
8601 * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is
8602 * always the case unless VLAN splinters are enabled), returns false without
8603 * making any changes. */
8605 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
8610 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port, &vid);
8615 /* Cause the flow to be processed as if it came in on the real device with
8616 * the VLAN device's VLAN ID. */
8617 flow->in_port = realdev;
8618 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
8623 vsp_remove(struct ofport_dpif *port)
8625 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
8626 struct vlan_splinter *vsp;
8628 vsp = vlandev_find(ofproto, port->up.ofp_port);
8630 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
8631 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
8634 port->realdev_ofp_port = 0;
8636 VLOG_ERR("missing vlan device record");
8641 vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid)
8643 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
8645 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
8646 && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid))
8647 == realdev_ofp_port)) {
8648 struct vlan_splinter *vsp;
8650 vsp = xmalloc(sizeof *vsp);
8651 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
8652 hash_int(port->up.ofp_port, 0));
8653 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
8654 hash_realdev_vid(realdev_ofp_port, vid));
8655 vsp->realdev_ofp_port = realdev_ofp_port;
8656 vsp->vlandev_ofp_port = port->up.ofp_port;
8659 port->realdev_ofp_port = realdev_ofp_port;
8661 VLOG_ERR("duplicate vlan device record");
8666 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
8668 const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
8669 return ofport ? ofport->odp_port : OVSP_NONE;
8672 static struct ofport_dpif *
8673 odp_port_to_ofport(const struct dpif_backer *backer, uint32_t odp_port)
8675 struct ofport_dpif *port;
8677 HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node,
8678 hash_int(odp_port, 0),
8679 &backer->odp_to_ofport_map) {
8680 if (port->odp_port == odp_port) {
8689 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
8691 struct ofport_dpif *port;
8693 port = odp_port_to_ofport(ofproto->backer, odp_port);
8694 if (port && &ofproto->up == port->up.ofproto) {
8695 return port->up.ofp_port;
8700 static unsigned long long int
8701 avg_subfacet_life_span(const struct ofproto_dpif *ofproto)
8703 unsigned long long int dc;
8704 unsigned long long int avg;
8706 dc = ofproto->total_subfacet_del_count + ofproto->subfacet_del_count;
8707 avg = dc ? ofproto->total_subfacet_life_span / dc : 0;
8713 avg_subfacet_count(const struct ofproto_dpif *ofproto)
8717 if (ofproto->n_update_stats) {
8718 avg_c = (double)ofproto->total_subfacet_count
8719 / ofproto->n_update_stats;
8726 show_dp_rates(struct ds *ds, const char *heading,
8727 const struct avg_subfacet_rates *rates)
8729 ds_put_format(ds, "%s add rate: %5.3f/min, del rate: %5.3f/min\n",
8730 heading, rates->add_rate, rates->del_rate);
8734 update_max_subfacet_count(struct ofproto_dpif *ofproto)
8736 ofproto->max_n_subfacet = MAX(ofproto->max_n_subfacet,
8737 hmap_count(&ofproto->subfacets));
8740 /* Compute exponentially weighted moving average, adding 'new' as the newest,
8741 * most heavily weighted element. 'base' designates the rate of decay: after
8742 * 'base' further updates, 'new''s weight in the EWMA decays to about 1/e
8745 exp_mavg(double *avg, int base, double new)
8747 *avg = (*avg * (base - 1) + new) / base;
8751 update_moving_averages(struct ofproto_dpif *ofproto)
8753 const int min_ms = 60 * 1000; /* milliseconds in one minute. */
8755 /* Update hourly averages on the minute boundaries. */
8756 if (time_msec() - ofproto->last_minute >= min_ms) {
8757 exp_mavg(&ofproto->hourly.add_rate, 60, ofproto->subfacet_add_count);
8758 exp_mavg(&ofproto->hourly.del_rate, 60, ofproto->subfacet_del_count);
8760 /* Update daily averages on the hour boundaries. */
8761 if ((ofproto->last_minute - ofproto->created) / min_ms % 60 == 59) {
8762 exp_mavg(&ofproto->daily.add_rate, 24, ofproto->hourly.add_rate);
8763 exp_mavg(&ofproto->daily.del_rate, 24, ofproto->hourly.del_rate);
8766 ofproto->total_subfacet_add_count += ofproto->subfacet_add_count;
8767 ofproto->total_subfacet_del_count += ofproto->subfacet_del_count;
8768 ofproto->subfacet_add_count = 0;
8769 ofproto->subfacet_del_count = 0;
8770 ofproto->last_minute += min_ms;
8775 dpif_stats_update_hit_count(struct ofproto_dpif *ofproto, uint64_t delta)
8777 ofproto->n_hit += delta;
8780 const struct ofproto_class ofproto_dpif_class = {
8815 port_is_lacp_current,
8816 NULL, /* rule_choose_table */
8823 rule_modify_actions,
8837 get_stp_port_status,
8844 is_mirror_output_bundle,
8845 forward_bpdu_changed,
8846 set_mac_table_config,