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 uint16_t vsp_realdev_to_vlandev(const struct ofproto_dpif *,
567 uint16_t realdev_ofp_port,
569 static bool vsp_adjust_flow(const struct ofproto_dpif *, struct flow *);
570 static void vsp_remove(struct ofport_dpif *);
571 static void vsp_add(struct ofport_dpif *, uint16_t realdev_ofp_port, int vid);
573 static uint32_t ofp_port_to_odp_port(const struct ofproto_dpif *,
575 static uint16_t odp_port_to_ofp_port(const struct ofproto_dpif *,
578 static struct ofport_dpif *
579 ofport_dpif_cast(const struct ofport *ofport)
581 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
584 static void port_run(struct ofport_dpif *);
585 static void port_run_fast(struct ofport_dpif *);
586 static void port_wait(struct ofport_dpif *);
587 static int set_bfd(struct ofport *, const struct smap *);
588 static int set_cfm(struct ofport *, const struct cfm_settings *);
589 static void ofport_clear_priorities(struct ofport_dpif *);
590 static void run_fast_rl(void);
592 struct dpif_completion {
593 struct list list_node;
594 struct ofoperation *op;
597 /* Extra information about a classifier table.
598 * Currently used just for optimized flow revalidation. */
600 /* If either of these is nonnull, then this table has a form that allows
601 * flows to be tagged to avoid revalidating most flows for the most common
602 * kinds of flow table changes. */
603 struct cls_table *catchall_table; /* Table that wildcards all fields. */
604 struct cls_table *other_table; /* Table with any other wildcard set. */
605 uint32_t basis; /* Keeps each table's tags separate. */
608 /* Reasons that we might need to revalidate every facet, and corresponding
611 * A value of 0 means that there is no need to revalidate.
613 * It would be nice to have some cleaner way to integrate with coverage
614 * counters, but with only a few reasons I guess this is good enough for
616 enum revalidate_reason {
617 REV_RECONFIGURE = 1, /* Switch configuration changed. */
618 REV_STP, /* Spanning tree protocol port status change. */
619 REV_PORT_TOGGLED, /* Port enabled or disabled by CFM, LACP, ...*/
620 REV_FLOW_TABLE, /* Flow table changed. */
621 REV_INCONSISTENCY /* Facet self-check failed. */
623 COVERAGE_DEFINE(rev_reconfigure);
624 COVERAGE_DEFINE(rev_stp);
625 COVERAGE_DEFINE(rev_port_toggled);
626 COVERAGE_DEFINE(rev_flow_table);
627 COVERAGE_DEFINE(rev_inconsistency);
629 /* Drop keys are odp flow keys which have drop flows installed in the kernel.
630 * These are datapath flows which have no associated ofproto, if they did we
631 * would use facets. */
633 struct hmap_node hmap_node;
638 /* All datapaths of a given type share a single dpif backer instance. */
643 struct timer next_expiration;
644 struct hmap odp_to_ofport_map; /* ODP port to ofport mapping. */
646 struct simap tnl_backers; /* Set of dpif ports backing tunnels. */
648 /* Facet revalidation flags applying to facets which use this backer. */
649 enum revalidate_reason need_revalidate; /* Revalidate every facet. */
650 struct tag_set revalidate_set; /* Revalidate only matching facets. */
652 struct hmap drop_keys; /* Set of dropped odp keys. */
653 bool recv_set_enable; /* Enables or disables receiving packets. */
656 /* All existing ofproto_backer instances, indexed by ofproto->up.type. */
657 static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers);
659 static void drop_key_clear(struct dpif_backer *);
660 static struct ofport_dpif *
661 odp_port_to_ofport(const struct dpif_backer *, uint32_t odp_port);
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 update_max_subfacet_count(struct ofproto_dpif *ofproto);
764 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
765 * for debugging the asynchronous flow_mod implementation.) */
768 /* All existing ofproto_dpif instances, indexed by ->up.name. */
769 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
771 static void ofproto_dpif_unixctl_init(void);
773 static struct ofproto_dpif *
774 ofproto_dpif_cast(const struct ofproto *ofproto)
776 ovs_assert(ofproto->ofproto_class == &ofproto_dpif_class);
777 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
780 static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *,
782 static struct ofport_dpif *get_odp_port(const struct ofproto_dpif *,
784 static void ofproto_trace(struct ofproto_dpif *, const struct flow *,
785 const struct ofpbuf *,
786 const struct initial_vals *, struct ds *);
788 /* Packet processing. */
789 static void update_learning_table(struct ofproto_dpif *,
790 const struct flow *, int vlan,
793 #define FLOW_MISS_MAX_BATCH 50
794 static int handle_upcalls(struct dpif_backer *, unsigned int max_batch);
796 /* Flow expiration. */
797 static int expire(struct dpif_backer *);
800 static void send_netflow_active_timeouts(struct ofproto_dpif *);
803 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
804 static size_t compose_sflow_action(const struct ofproto_dpif *,
805 struct ofpbuf *odp_actions,
806 const struct flow *, uint32_t odp_port);
807 static void compose_ipfix_action(const struct ofproto_dpif *,
808 struct ofpbuf *odp_actions,
809 const struct flow *);
810 static void add_mirror_actions(struct xlate_ctx *ctx,
811 const struct flow *flow);
812 /* Global variables. */
813 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
815 /* Initial mappings of port to bridge mappings. */
816 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
818 /* Factory functions. */
821 init(const struct shash *iface_hints)
823 struct shash_node *node;
825 /* Make a local copy, since we don't own 'iface_hints' elements. */
826 SHASH_FOR_EACH(node, iface_hints) {
827 const struct iface_hint *orig_hint = node->data;
828 struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
830 new_hint->br_name = xstrdup(orig_hint->br_name);
831 new_hint->br_type = xstrdup(orig_hint->br_type);
832 new_hint->ofp_port = orig_hint->ofp_port;
834 shash_add(&init_ofp_ports, node->name, new_hint);
839 enumerate_types(struct sset *types)
841 dp_enumerate_types(types);
845 enumerate_names(const char *type, struct sset *names)
847 struct ofproto_dpif *ofproto;
850 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
851 if (strcmp(type, ofproto->up.type)) {
854 sset_add(names, ofproto->up.name);
861 del(const char *type, const char *name)
866 error = dpif_open(name, type, &dpif);
868 error = dpif_delete(dpif);
875 port_open_type(const char *datapath_type, const char *port_type)
877 return dpif_port_open_type(datapath_type, port_type);
880 /* Type functions. */
882 static struct ofproto_dpif *
883 lookup_ofproto_dpif_by_port_name(const char *name)
885 struct ofproto_dpif *ofproto;
887 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
888 if (sset_contains(&ofproto->ports, name)) {
897 type_run(const char *type)
899 static long long int push_timer = LLONG_MIN;
900 struct dpif_backer *backer;
904 backer = shash_find_data(&all_dpif_backers, type);
906 /* This is not necessarily a problem, since backers are only
907 * created on demand. */
911 dpif_run(backer->dpif);
913 /* The most natural place to push facet statistics is when they're pulled
914 * from the datapath. However, when there are many flows in the datapath,
915 * this expensive operation can occur so frequently, that it reduces our
916 * ability to quickly set up flows. To reduce the cost, we push statistics
918 if (time_msec() > push_timer) {
919 push_timer = time_msec() + 2000;
923 /* If vswitchd started with other_config:flow_restore_wait set as "true",
924 * and the configuration has now changed to "false", enable receiving
925 * packets from the datapath. */
926 if (!backer->recv_set_enable && !ofproto_get_flow_restore_wait()) {
927 backer->recv_set_enable = true;
929 error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
931 VLOG_ERR("Failed to enable receiving packets in dpif.");
934 dpif_flow_flush(backer->dpif);
935 backer->need_revalidate = REV_RECONFIGURE;
938 if (backer->need_revalidate
939 || !tag_set_is_empty(&backer->revalidate_set)) {
940 struct tag_set revalidate_set = backer->revalidate_set;
941 bool need_revalidate = backer->need_revalidate;
942 struct ofproto_dpif *ofproto;
943 struct simap_node *node;
944 struct simap tmp_backers;
946 /* Handle tunnel garbage collection. */
947 simap_init(&tmp_backers);
948 simap_swap(&backer->tnl_backers, &tmp_backers);
950 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
951 struct ofport_dpif *iter;
953 if (backer != ofproto->backer) {
957 HMAP_FOR_EACH (iter, up.hmap_node, &ofproto->up.ports) {
960 if (!iter->tnl_port) {
964 dp_port = netdev_vport_get_dpif_port(iter->up.netdev);
965 node = simap_find(&tmp_backers, dp_port);
967 simap_put(&backer->tnl_backers, dp_port, node->data);
968 simap_delete(&tmp_backers, node);
969 node = simap_find(&backer->tnl_backers, dp_port);
971 node = simap_find(&backer->tnl_backers, dp_port);
973 uint32_t odp_port = UINT32_MAX;
975 if (!dpif_port_add(backer->dpif, iter->up.netdev,
977 simap_put(&backer->tnl_backers, dp_port, odp_port);
978 node = simap_find(&backer->tnl_backers, dp_port);
983 iter->odp_port = node ? node->data : OVSP_NONE;
984 if (tnl_port_reconfigure(&iter->up, iter->odp_port,
986 backer->need_revalidate = REV_RECONFIGURE;
991 SIMAP_FOR_EACH (node, &tmp_backers) {
992 dpif_port_del(backer->dpif, node->data);
994 simap_destroy(&tmp_backers);
996 switch (backer->need_revalidate) {
997 case REV_RECONFIGURE: COVERAGE_INC(rev_reconfigure); break;
998 case REV_STP: COVERAGE_INC(rev_stp); break;
999 case REV_PORT_TOGGLED: COVERAGE_INC(rev_port_toggled); break;
1000 case REV_FLOW_TABLE: COVERAGE_INC(rev_flow_table); break;
1001 case REV_INCONSISTENCY: COVERAGE_INC(rev_inconsistency); break;
1004 if (backer->need_revalidate) {
1005 /* Clear the drop_keys in case we should now be accepting some
1006 * formerly dropped flows. */
1007 drop_key_clear(backer);
1010 /* Clear the revalidation flags. */
1011 tag_set_init(&backer->revalidate_set);
1012 backer->need_revalidate = 0;
1014 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1015 struct facet *facet, *next;
1017 if (ofproto->backer != backer) {
1021 HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &ofproto->facets) {
1023 || tag_set_intersects(&revalidate_set, facet->xout.tags)) {
1024 facet_revalidate(facet);
1031 if (!backer->recv_set_enable) {
1032 /* Wake up before a max of 1000ms. */
1033 timer_set_duration(&backer->next_expiration, 1000);
1034 } else if (timer_expired(&backer->next_expiration)) {
1035 int delay = expire(backer);
1036 timer_set_duration(&backer->next_expiration, delay);
1039 /* Check for port changes in the dpif. */
1040 while ((error = dpif_port_poll(backer->dpif, &devname)) == 0) {
1041 struct ofproto_dpif *ofproto;
1042 struct dpif_port port;
1044 /* Don't report on the datapath's device. */
1045 if (!strcmp(devname, dpif_base_name(backer->dpif))) {
1049 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
1050 &all_ofproto_dpifs) {
1051 if (simap_contains(&ofproto->backer->tnl_backers, devname)) {
1056 ofproto = lookup_ofproto_dpif_by_port_name(devname);
1057 if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
1058 /* The port was removed. If we know the datapath,
1059 * report it through poll_set(). If we don't, it may be
1060 * notifying us of a removal we initiated, so ignore it.
1061 * If there's a pending ENOBUFS, let it stand, since
1062 * everything will be reevaluated. */
1063 if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
1064 sset_add(&ofproto->port_poll_set, devname);
1065 ofproto->port_poll_errno = 0;
1067 } else if (!ofproto) {
1068 /* The port was added, but we don't know with which
1069 * ofproto we should associate it. Delete it. */
1070 dpif_port_del(backer->dpif, port.port_no);
1072 dpif_port_destroy(&port);
1078 if (error != EAGAIN) {
1079 struct ofproto_dpif *ofproto;
1081 /* There was some sort of error, so propagate it to all
1082 * ofprotos that use this backer. */
1083 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
1084 &all_ofproto_dpifs) {
1085 if (ofproto->backer == backer) {
1086 sset_clear(&ofproto->port_poll_set);
1087 ofproto->port_poll_errno = error;
1096 dpif_backer_run_fast(struct dpif_backer *backer, int max_batch)
1100 /* If recv_set_enable is false, we should not handle upcalls. */
1101 if (!backer->recv_set_enable) {
1105 /* Handle one or more batches of upcalls, until there's nothing left to do
1106 * or until we do a fixed total amount of work.
1108 * We do work in batches because it can be much cheaper to set up a number
1109 * of flows and fire off their patches all at once. We do multiple batches
1110 * because in some cases handling a packet can cause another packet to be
1111 * queued almost immediately as part of the return flow. Both
1112 * optimizations can make major improvements on some benchmarks and
1113 * presumably for real traffic as well. */
1115 while (work < max_batch) {
1116 int retval = handle_upcalls(backer, max_batch - work);
1127 type_run_fast(const char *type)
1129 struct dpif_backer *backer;
1131 backer = shash_find_data(&all_dpif_backers, type);
1133 /* This is not necessarily a problem, since backers are only
1134 * created on demand. */
1138 return dpif_backer_run_fast(backer, FLOW_MISS_MAX_BATCH);
1144 static long long int port_rl = LLONG_MIN;
1145 static unsigned int backer_rl = 0;
1147 if (time_msec() >= port_rl) {
1148 struct ofproto_dpif *ofproto;
1149 struct ofport_dpif *ofport;
1151 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1153 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1154 port_run_fast(ofport);
1157 port_rl = time_msec() + 200;
1160 /* XXX: We have to be careful not to do too much work in this function. If
1161 * we call dpif_backer_run_fast() too often, or with too large a batch,
1162 * performance improves signifcantly, but at a cost. It's possible for the
1163 * number of flows in the datapath to increase without bound, and for poll
1164 * loops to take 10s of seconds. The correct solution to this problem,
1165 * long term, is to separate flow miss handling into it's own thread so it
1166 * isn't affected by revalidations, and expirations. Until then, this is
1167 * the best we can do. */
1168 if (++backer_rl >= 10) {
1169 struct shash_node *node;
1172 SHASH_FOR_EACH (node, &all_dpif_backers) {
1173 dpif_backer_run_fast(node->data, 1);
1179 type_wait(const char *type)
1181 struct dpif_backer *backer;
1183 backer = shash_find_data(&all_dpif_backers, type);
1185 /* This is not necessarily a problem, since backers are only
1186 * created on demand. */
1190 timer_wait(&backer->next_expiration);
1193 /* Basic life-cycle. */
1195 static int add_internal_flows(struct ofproto_dpif *);
1197 static struct ofproto *
1200 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
1201 return &ofproto->up;
1205 dealloc(struct ofproto *ofproto_)
1207 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1212 close_dpif_backer(struct dpif_backer *backer)
1214 struct shash_node *node;
1216 ovs_assert(backer->refcount > 0);
1218 if (--backer->refcount) {
1222 drop_key_clear(backer);
1223 hmap_destroy(&backer->drop_keys);
1225 simap_destroy(&backer->tnl_backers);
1226 hmap_destroy(&backer->odp_to_ofport_map);
1227 node = shash_find(&all_dpif_backers, backer->type);
1229 shash_delete(&all_dpif_backers, node);
1230 dpif_close(backer->dpif);
1235 /* Datapath port slated for removal from datapath. */
1236 struct odp_garbage {
1237 struct list list_node;
1242 open_dpif_backer(const char *type, struct dpif_backer **backerp)
1244 struct dpif_backer *backer;
1245 struct dpif_port_dump port_dump;
1246 struct dpif_port port;
1247 struct shash_node *node;
1248 struct list garbage_list;
1249 struct odp_garbage *garbage, *next;
1255 backer = shash_find_data(&all_dpif_backers, type);
1262 backer_name = xasprintf("ovs-%s", type);
1264 /* Remove any existing datapaths, since we assume we're the only
1265 * userspace controlling the datapath. */
1267 dp_enumerate_names(type, &names);
1268 SSET_FOR_EACH(name, &names) {
1269 struct dpif *old_dpif;
1271 /* Don't remove our backer if it exists. */
1272 if (!strcmp(name, backer_name)) {
1276 if (dpif_open(name, type, &old_dpif)) {
1277 VLOG_WARN("couldn't open old datapath %s to remove it", name);
1279 dpif_delete(old_dpif);
1280 dpif_close(old_dpif);
1283 sset_destroy(&names);
1285 backer = xmalloc(sizeof *backer);
1287 error = dpif_create_and_open(backer_name, type, &backer->dpif);
1290 VLOG_ERR("failed to open datapath of type %s: %s", type,
1296 backer->type = xstrdup(type);
1297 backer->refcount = 1;
1298 hmap_init(&backer->odp_to_ofport_map);
1299 hmap_init(&backer->drop_keys);
1300 timer_set_duration(&backer->next_expiration, 1000);
1301 backer->need_revalidate = 0;
1302 simap_init(&backer->tnl_backers);
1303 tag_set_init(&backer->revalidate_set);
1304 backer->recv_set_enable = !ofproto_get_flow_restore_wait();
1307 if (backer->recv_set_enable) {
1308 dpif_flow_flush(backer->dpif);
1311 /* Loop through the ports already on the datapath and remove any
1312 * that we don't need anymore. */
1313 list_init(&garbage_list);
1314 dpif_port_dump_start(&port_dump, backer->dpif);
1315 while (dpif_port_dump_next(&port_dump, &port)) {
1316 node = shash_find(&init_ofp_ports, port.name);
1317 if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
1318 garbage = xmalloc(sizeof *garbage);
1319 garbage->odp_port = port.port_no;
1320 list_push_front(&garbage_list, &garbage->list_node);
1323 dpif_port_dump_done(&port_dump);
1325 LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) {
1326 dpif_port_del(backer->dpif, garbage->odp_port);
1327 list_remove(&garbage->list_node);
1331 shash_add(&all_dpif_backers, type, backer);
1333 error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
1335 VLOG_ERR("failed to listen on datapath of type %s: %s",
1336 type, strerror(error));
1337 close_dpif_backer(backer);
1345 construct(struct ofproto *ofproto_)
1347 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1348 struct shash_node *node, *next;
1353 error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1358 max_ports = dpif_get_max_ports(ofproto->backer->dpif);
1359 ofproto_init_max_ports(ofproto_, MIN(max_ports, OFPP_MAX));
1361 ofproto->netflow = NULL;
1362 ofproto->sflow = NULL;
1363 ofproto->ipfix = NULL;
1364 ofproto->stp = NULL;
1365 hmap_init(&ofproto->bundles);
1366 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1367 for (i = 0; i < MAX_MIRRORS; i++) {
1368 ofproto->mirrors[i] = NULL;
1370 ofproto->has_bonded_bundles = false;
1372 hmap_init(&ofproto->facets);
1373 hmap_init(&ofproto->subfacets);
1374 ofproto->governor = NULL;
1375 ofproto->consistency_rl = LLONG_MIN;
1377 for (i = 0; i < N_TABLES; i++) {
1378 struct table_dpif *table = &ofproto->tables[i];
1380 table->catchall_table = NULL;
1381 table->other_table = NULL;
1382 table->basis = random_uint32();
1385 list_init(&ofproto->completions);
1387 ofproto_dpif_unixctl_init();
1389 ofproto->has_mirrors = false;
1390 ofproto->has_bundle_action = false;
1392 hmap_init(&ofproto->vlandev_map);
1393 hmap_init(&ofproto->realdev_vid_map);
1395 sset_init(&ofproto->ports);
1396 sset_init(&ofproto->ghost_ports);
1397 sset_init(&ofproto->port_poll_set);
1398 ofproto->port_poll_errno = 0;
1400 SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1401 struct iface_hint *iface_hint = node->data;
1403 if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1404 /* Check if the datapath already has this port. */
1405 if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1406 sset_add(&ofproto->ports, node->name);
1409 free(iface_hint->br_name);
1410 free(iface_hint->br_type);
1412 shash_delete(&init_ofp_ports, node);
1416 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1417 hash_string(ofproto->up.name, 0));
1418 memset(&ofproto->stats, 0, sizeof ofproto->stats);
1420 ofproto_init_tables(ofproto_, N_TABLES);
1421 error = add_internal_flows(ofproto);
1422 ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1425 ofproto->n_missed = 0;
1427 ofproto->max_n_subfacet = 0;
1428 ofproto->created = time_msec();
1429 ofproto->last_minute = ofproto->created;
1430 memset(&ofproto->hourly, 0, sizeof ofproto->hourly);
1431 memset(&ofproto->daily, 0, sizeof ofproto->daily);
1432 ofproto->subfacet_add_count = 0;
1433 ofproto->subfacet_del_count = 0;
1434 ofproto->total_subfacet_add_count = 0;
1435 ofproto->total_subfacet_del_count = 0;
1436 ofproto->total_subfacet_life_span = 0;
1437 ofproto->total_subfacet_count = 0;
1438 ofproto->n_update_stats = 0;
1444 add_internal_flow(struct ofproto_dpif *ofproto, int id,
1445 const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1447 struct ofputil_flow_mod fm;
1450 match_init_catchall(&fm.match);
1452 match_set_reg(&fm.match, 0, id);
1453 fm.new_cookie = htonll(0);
1454 fm.cookie = htonll(0);
1455 fm.cookie_mask = htonll(0);
1456 fm.table_id = TBL_INTERNAL;
1457 fm.command = OFPFC_ADD;
1458 fm.idle_timeout = 0;
1459 fm.hard_timeout = 0;
1463 fm.ofpacts = ofpacts->data;
1464 fm.ofpacts_len = ofpacts->size;
1466 error = ofproto_flow_mod(&ofproto->up, &fm);
1468 VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
1469 id, ofperr_to_string(error));
1473 *rulep = rule_dpif_lookup__(ofproto, &fm.match.flow, TBL_INTERNAL);
1474 ovs_assert(*rulep != NULL);
1480 add_internal_flows(struct ofproto_dpif *ofproto)
1482 struct ofpact_controller *controller;
1483 uint64_t ofpacts_stub[128 / 8];
1484 struct ofpbuf ofpacts;
1488 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1491 controller = ofpact_put_CONTROLLER(&ofpacts);
1492 controller->max_len = UINT16_MAX;
1493 controller->controller_id = 0;
1494 controller->reason = OFPR_NO_MATCH;
1495 ofpact_pad(&ofpacts);
1497 error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule);
1502 ofpbuf_clear(&ofpacts);
1503 error = add_internal_flow(ofproto, id++, &ofpacts,
1504 &ofproto->no_packet_in_rule);
1509 complete_operations(struct ofproto_dpif *ofproto)
1511 struct dpif_completion *c, *next;
1513 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
1514 ofoperation_complete(c->op, 0);
1515 list_remove(&c->list_node);
1521 destruct(struct ofproto *ofproto_)
1523 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1524 struct rule_dpif *rule, *next_rule;
1525 struct oftable *table;
1528 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1529 complete_operations(ofproto);
1531 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1532 struct cls_cursor cursor;
1534 cls_cursor_init(&cursor, &table->cls, NULL);
1535 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1536 ofproto_rule_destroy(&rule->up);
1540 for (i = 0; i < MAX_MIRRORS; i++) {
1541 mirror_destroy(ofproto->mirrors[i]);
1544 netflow_destroy(ofproto->netflow);
1545 dpif_sflow_destroy(ofproto->sflow);
1546 hmap_destroy(&ofproto->bundles);
1547 mac_learning_destroy(ofproto->ml);
1549 hmap_destroy(&ofproto->facets);
1550 hmap_destroy(&ofproto->subfacets);
1551 governor_destroy(ofproto->governor);
1553 hmap_destroy(&ofproto->vlandev_map);
1554 hmap_destroy(&ofproto->realdev_vid_map);
1556 sset_destroy(&ofproto->ports);
1557 sset_destroy(&ofproto->ghost_ports);
1558 sset_destroy(&ofproto->port_poll_set);
1560 close_dpif_backer(ofproto->backer);
1564 run_fast(struct ofproto *ofproto_)
1566 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1567 struct ofport_dpif *ofport;
1569 /* Do not perform any periodic activity required by 'ofproto' while
1570 * waiting for flow restore to complete. */
1571 if (ofproto_get_flow_restore_wait()) {
1575 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1576 port_run_fast(ofport);
1583 run(struct ofproto *ofproto_)
1585 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1586 struct ofport_dpif *ofport;
1587 struct ofbundle *bundle;
1591 complete_operations(ofproto);
1594 /* Do not perform any periodic activity below required by 'ofproto' while
1595 * waiting for flow restore to complete. */
1596 if (ofproto_get_flow_restore_wait()) {
1600 error = run_fast(ofproto_);
1605 if (ofproto->netflow) {
1606 if (netflow_run(ofproto->netflow)) {
1607 send_netflow_active_timeouts(ofproto);
1610 if (ofproto->sflow) {
1611 dpif_sflow_run(ofproto->sflow);
1614 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1617 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1622 mac_learning_run(ofproto->ml, &ofproto->backer->revalidate_set);
1624 /* Check the consistency of a random facet, to aid debugging. */
1625 if (time_msec() >= ofproto->consistency_rl
1626 && !hmap_is_empty(&ofproto->facets)
1627 && !ofproto->backer->need_revalidate) {
1628 struct facet *facet;
1630 ofproto->consistency_rl = time_msec() + 250;
1632 facet = CONTAINER_OF(hmap_random_node(&ofproto->facets),
1633 struct facet, hmap_node);
1634 if (!tag_set_intersects(&ofproto->backer->revalidate_set,
1635 facet->xout.tags)) {
1636 if (!facet_check_consistency(facet)) {
1637 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
1642 if (ofproto->governor) {
1645 governor_run(ofproto->governor);
1647 /* If the governor has shrunk to its minimum size and the number of
1648 * subfacets has dwindled, then drop the governor entirely.
1650 * For hysteresis, the number of subfacets to drop the governor is
1651 * smaller than the number needed to trigger its creation. */
1652 n_subfacets = hmap_count(&ofproto->subfacets);
1653 if (n_subfacets * 4 < ofproto->up.flow_eviction_threshold
1654 && governor_is_idle(ofproto->governor)) {
1655 governor_destroy(ofproto->governor);
1656 ofproto->governor = NULL;
1664 wait(struct ofproto *ofproto_)
1666 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1667 struct ofport_dpif *ofport;
1668 struct ofbundle *bundle;
1670 if (!clogged && !list_is_empty(&ofproto->completions)) {
1671 poll_immediate_wake();
1674 if (ofproto_get_flow_restore_wait()) {
1678 dpif_wait(ofproto->backer->dpif);
1679 dpif_recv_wait(ofproto->backer->dpif);
1680 if (ofproto->sflow) {
1681 dpif_sflow_wait(ofproto->sflow);
1683 if (!tag_set_is_empty(&ofproto->backer->revalidate_set)) {
1684 poll_immediate_wake();
1686 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1689 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1690 bundle_wait(bundle);
1692 if (ofproto->netflow) {
1693 netflow_wait(ofproto->netflow);
1695 mac_learning_wait(ofproto->ml);
1697 if (ofproto->backer->need_revalidate) {
1698 /* Shouldn't happen, but if it does just go around again. */
1699 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1700 poll_immediate_wake();
1702 if (ofproto->governor) {
1703 governor_wait(ofproto->governor);
1708 get_memory_usage(const struct ofproto *ofproto_, struct simap *usage)
1710 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1712 simap_increase(usage, "facets", hmap_count(&ofproto->facets));
1713 simap_increase(usage, "subfacets", hmap_count(&ofproto->subfacets));
1717 flush(struct ofproto *ofproto_)
1719 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1720 struct subfacet *subfacet, *next_subfacet;
1721 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
1725 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
1726 &ofproto->subfacets) {
1727 if (subfacet->path != SF_NOT_INSTALLED) {
1728 batch[n_batch++] = subfacet;
1729 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
1730 subfacet_destroy_batch(ofproto, batch, n_batch);
1734 subfacet_destroy(subfacet);
1739 subfacet_destroy_batch(ofproto, batch, n_batch);
1744 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1745 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1747 *arp_match_ip = true;
1748 *actions = (OFPUTIL_A_OUTPUT |
1749 OFPUTIL_A_SET_VLAN_VID |
1750 OFPUTIL_A_SET_VLAN_PCP |
1751 OFPUTIL_A_STRIP_VLAN |
1752 OFPUTIL_A_SET_DL_SRC |
1753 OFPUTIL_A_SET_DL_DST |
1754 OFPUTIL_A_SET_NW_SRC |
1755 OFPUTIL_A_SET_NW_DST |
1756 OFPUTIL_A_SET_NW_TOS |
1757 OFPUTIL_A_SET_TP_SRC |
1758 OFPUTIL_A_SET_TP_DST |
1763 get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots)
1765 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1766 struct dpif_dp_stats s;
1767 uint64_t n_miss, n_no_pkt_in, n_bytes;
1770 strcpy(ots->name, "classifier");
1772 dpif_get_dp_stats(ofproto->backer->dpif, &s);
1773 rule_get_stats(&ofproto->miss_rule->up, &n_miss, &n_bytes);
1774 rule_get_stats(&ofproto->no_packet_in_rule->up, &n_no_pkt_in, &n_bytes);
1776 n_lookup = s.n_hit + s.n_missed;
1777 ots->lookup_count = htonll(n_lookup);
1778 ots->matched_count = htonll(n_lookup - n_miss - n_no_pkt_in);
1781 static struct ofport *
1784 struct ofport_dpif *port = xmalloc(sizeof *port);
1789 port_dealloc(struct ofport *port_)
1791 struct ofport_dpif *port = ofport_dpif_cast(port_);
1796 port_construct(struct ofport *port_)
1798 struct ofport_dpif *port = ofport_dpif_cast(port_);
1799 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1800 const struct netdev *netdev = port->up.netdev;
1801 struct dpif_port dpif_port;
1804 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1805 port->bundle = NULL;
1808 port->tag = tag_create_random();
1809 port->may_enable = true;
1810 port->stp_port = NULL;
1811 port->stp_state = STP_DISABLED;
1812 port->tnl_port = NULL;
1813 hmap_init(&port->priorities);
1814 port->realdev_ofp_port = 0;
1815 port->vlandev_vid = 0;
1816 port->carrier_seq = netdev_get_carrier_resets(netdev);
1818 if (netdev_vport_is_patch(netdev)) {
1819 /* By bailing out here, we don't submit the port to the sFlow module
1820 * to be considered for counter polling export. This is correct
1821 * because the patch port represents an interface that sFlow considers
1822 * to be "internal" to the switch as a whole, and therefore not an
1823 * candidate for counter polling. */
1824 port->odp_port = OVSP_NONE;
1828 error = dpif_port_query_by_name(ofproto->backer->dpif,
1829 netdev_vport_get_dpif_port(netdev),
1835 port->odp_port = dpif_port.port_no;
1837 if (netdev_get_tunnel_config(netdev)) {
1838 port->tnl_port = tnl_port_add(&port->up, port->odp_port);
1840 /* Sanity-check that a mapping doesn't already exist. This
1841 * shouldn't happen for non-tunnel ports. */
1842 if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1843 VLOG_ERR("port %s already has an OpenFlow port number",
1845 dpif_port_destroy(&dpif_port);
1849 hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1850 hash_int(port->odp_port, 0));
1852 dpif_port_destroy(&dpif_port);
1854 if (ofproto->sflow) {
1855 dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1862 port_destruct(struct ofport *port_)
1864 struct ofport_dpif *port = ofport_dpif_cast(port_);
1865 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1866 const char *dp_port_name = netdev_vport_get_dpif_port(port->up.netdev);
1867 const char *devname = netdev_get_name(port->up.netdev);
1869 if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1870 /* The underlying device is still there, so delete it. This
1871 * happens when the ofproto is being destroyed, since the caller
1872 * assumes that removal of attached ports will happen as part of
1874 if (!port->tnl_port) {
1875 dpif_port_del(ofproto->backer->dpif, port->odp_port);
1877 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1880 if (port->odp_port != OVSP_NONE && !port->tnl_port) {
1881 hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1884 tnl_port_del(port->tnl_port);
1885 sset_find_and_delete(&ofproto->ports, devname);
1886 sset_find_and_delete(&ofproto->ghost_ports, devname);
1887 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1888 bundle_remove(port_);
1889 set_cfm(port_, NULL);
1890 set_bfd(port_, NULL);
1891 if (ofproto->sflow) {
1892 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1895 ofport_clear_priorities(port);
1896 hmap_destroy(&port->priorities);
1900 port_modified(struct ofport *port_)
1902 struct ofport_dpif *port = ofport_dpif_cast(port_);
1904 if (port->bundle && port->bundle->bond) {
1905 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1909 cfm_set_netdev(port->cfm, port->up.netdev);
1914 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1916 struct ofport_dpif *port = ofport_dpif_cast(port_);
1917 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1918 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1920 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1921 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1922 OFPUTIL_PC_NO_PACKET_IN)) {
1923 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1925 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1926 bundle_update(port->bundle);
1932 set_sflow(struct ofproto *ofproto_,
1933 const struct ofproto_sflow_options *sflow_options)
1935 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1936 struct dpif_sflow *ds = ofproto->sflow;
1938 if (sflow_options) {
1940 struct ofport_dpif *ofport;
1942 ds = ofproto->sflow = dpif_sflow_create();
1943 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1944 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1946 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1948 dpif_sflow_set_options(ds, sflow_options);
1951 dpif_sflow_destroy(ds);
1952 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1953 ofproto->sflow = NULL;
1961 struct ofproto *ofproto_,
1962 const struct ofproto_ipfix_bridge_exporter_options *bridge_exporter_options,
1963 const struct ofproto_ipfix_flow_exporter_options *flow_exporters_options,
1964 size_t n_flow_exporters_options)
1966 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1967 struct dpif_ipfix *di = ofproto->ipfix;
1969 if (bridge_exporter_options || flow_exporters_options) {
1971 di = ofproto->ipfix = dpif_ipfix_create();
1973 dpif_ipfix_set_options(
1974 di, bridge_exporter_options, flow_exporters_options,
1975 n_flow_exporters_options);
1978 dpif_ipfix_destroy(di);
1979 ofproto->ipfix = NULL;
1986 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1988 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1995 struct ofproto_dpif *ofproto;
1997 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1998 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1999 ofport->cfm = cfm_create(ofport->up.netdev);
2002 if (cfm_configure(ofport->cfm, s)) {
2008 cfm_destroy(ofport->cfm);
2014 get_cfm_status(const struct ofport *ofport_,
2015 struct ofproto_cfm_status *status)
2017 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2020 status->faults = cfm_get_fault(ofport->cfm);
2021 status->remote_opstate = cfm_get_opup(ofport->cfm);
2022 status->health = cfm_get_health(ofport->cfm);
2023 cfm_get_remote_mpids(ofport->cfm, &status->rmps, &status->n_rmps);
2031 set_bfd(struct ofport *ofport_, const struct smap *cfg)
2033 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
2034 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2038 ofport->bfd = bfd_configure(old, netdev_get_name(ofport->up.netdev), cfg);
2039 if (ofport->bfd != old) {
2040 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2047 get_bfd_status(struct ofport *ofport_, struct smap *smap)
2049 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2052 bfd_get_status(ofport->bfd, smap);
2059 /* Spanning Tree. */
2062 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
2064 struct ofproto_dpif *ofproto = ofproto_;
2065 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
2066 struct ofport_dpif *ofport;
2068 ofport = stp_port_get_aux(sp);
2070 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
2071 ofproto->up.name, port_num);
2073 struct eth_header *eth = pkt->l2;
2075 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
2076 if (eth_addr_is_zero(eth->eth_src)) {
2077 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
2078 "with unknown MAC", ofproto->up.name, port_num);
2080 send_packet(ofport, pkt);
2086 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
2088 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
2090 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2092 /* Only revalidate flows if the configuration changed. */
2093 if (!s != !ofproto->stp) {
2094 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2098 if (!ofproto->stp) {
2099 ofproto->stp = stp_create(ofproto_->name, s->system_id,
2100 send_bpdu_cb, ofproto);
2101 ofproto->stp_last_tick = time_msec();
2104 stp_set_bridge_id(ofproto->stp, s->system_id);
2105 stp_set_bridge_priority(ofproto->stp, s->priority);
2106 stp_set_hello_time(ofproto->stp, s->hello_time);
2107 stp_set_max_age(ofproto->stp, s->max_age);
2108 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
2110 struct ofport *ofport;
2112 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
2113 set_stp_port(ofport, NULL);
2116 stp_destroy(ofproto->stp);
2117 ofproto->stp = NULL;
2124 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
2126 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2130 s->bridge_id = stp_get_bridge_id(ofproto->stp);
2131 s->designated_root = stp_get_designated_root(ofproto->stp);
2132 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
2141 update_stp_port_state(struct ofport_dpif *ofport)
2143 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2144 enum stp_state state;
2146 /* Figure out new state. */
2147 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
2151 if (ofport->stp_state != state) {
2152 enum ofputil_port_state of_state;
2155 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
2156 netdev_get_name(ofport->up.netdev),
2157 stp_state_name(ofport->stp_state),
2158 stp_state_name(state));
2159 if (stp_learn_in_state(ofport->stp_state)
2160 != stp_learn_in_state(state)) {
2161 /* xxx Learning action flows should also be flushed. */
2162 mac_learning_flush(ofproto->ml,
2163 &ofproto->backer->revalidate_set);
2165 fwd_change = stp_forward_in_state(ofport->stp_state)
2166 != stp_forward_in_state(state);
2168 ofproto->backer->need_revalidate = REV_STP;
2169 ofport->stp_state = state;
2170 ofport->stp_state_entered = time_msec();
2172 if (fwd_change && ofport->bundle) {
2173 bundle_update(ofport->bundle);
2176 /* Update the STP state bits in the OpenFlow port description. */
2177 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
2178 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
2179 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
2180 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
2181 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
2183 ofproto_port_set_state(&ofport->up, of_state);
2187 /* Configures STP on 'ofport_' using the settings defined in 's'. The
2188 * caller is responsible for assigning STP port numbers and ensuring
2189 * there are no duplicates. */
2191 set_stp_port(struct ofport *ofport_,
2192 const struct ofproto_port_stp_settings *s)
2194 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2195 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2196 struct stp_port *sp = ofport->stp_port;
2198 if (!s || !s->enable) {
2200 ofport->stp_port = NULL;
2201 stp_port_disable(sp);
2202 update_stp_port_state(ofport);
2205 } else if (sp && stp_port_no(sp) != s->port_num
2206 && ofport == stp_port_get_aux(sp)) {
2207 /* The port-id changed, so disable the old one if it's not
2208 * already in use by another port. */
2209 stp_port_disable(sp);
2212 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
2213 stp_port_enable(sp);
2215 stp_port_set_aux(sp, ofport);
2216 stp_port_set_priority(sp, s->priority);
2217 stp_port_set_path_cost(sp, s->path_cost);
2219 update_stp_port_state(ofport);
2225 get_stp_port_status(struct ofport *ofport_,
2226 struct ofproto_port_stp_status *s)
2228 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2229 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2230 struct stp_port *sp = ofport->stp_port;
2232 if (!ofproto->stp || !sp) {
2238 s->port_id = stp_port_get_id(sp);
2239 s->state = stp_port_get_state(sp);
2240 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
2241 s->role = stp_port_get_role(sp);
2242 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
2248 stp_run(struct ofproto_dpif *ofproto)
2251 long long int now = time_msec();
2252 long long int elapsed = now - ofproto->stp_last_tick;
2253 struct stp_port *sp;
2256 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
2257 ofproto->stp_last_tick = now;
2259 while (stp_get_changed_port(ofproto->stp, &sp)) {
2260 struct ofport_dpif *ofport = stp_port_get_aux(sp);
2263 update_stp_port_state(ofport);
2267 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
2268 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
2274 stp_wait(struct ofproto_dpif *ofproto)
2277 poll_timer_wait(1000);
2281 /* Returns true if STP should process 'flow'. */
2283 stp_should_process_flow(const struct flow *flow)
2285 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
2289 stp_process_packet(const struct ofport_dpif *ofport,
2290 const struct ofpbuf *packet)
2292 struct ofpbuf payload = *packet;
2293 struct eth_header *eth = payload.data;
2294 struct stp_port *sp = ofport->stp_port;
2296 /* Sink packets on ports that have STP disabled when the bridge has
2298 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
2302 /* Trim off padding on payload. */
2303 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
2304 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
2307 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
2308 stp_received_bpdu(sp, payload.data, payload.size);
2312 static struct priority_to_dscp *
2313 get_priority(const struct ofport_dpif *ofport, uint32_t priority)
2315 struct priority_to_dscp *pdscp;
2318 hash = hash_int(priority, 0);
2319 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) {
2320 if (pdscp->priority == priority) {
2328 ofport_clear_priorities(struct ofport_dpif *ofport)
2330 struct priority_to_dscp *pdscp, *next;
2332 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) {
2333 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
2339 set_queues(struct ofport *ofport_,
2340 const struct ofproto_port_queue *qdscp_list,
2343 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2344 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2345 struct hmap new = HMAP_INITIALIZER(&new);
2348 for (i = 0; i < n_qdscp; i++) {
2349 struct priority_to_dscp *pdscp;
2353 dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
2354 if (dpif_queue_to_priority(ofproto->backer->dpif, qdscp_list[i].queue,
2359 pdscp = get_priority(ofport, priority);
2361 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
2363 pdscp = xmalloc(sizeof *pdscp);
2364 pdscp->priority = priority;
2366 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2369 if (pdscp->dscp != dscp) {
2371 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2374 hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0));
2377 if (!hmap_is_empty(&ofport->priorities)) {
2378 ofport_clear_priorities(ofport);
2379 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2382 hmap_swap(&new, &ofport->priorities);
2390 /* Expires all MAC learning entries associated with 'bundle' and forces its
2391 * ofproto to revalidate every flow.
2393 * Normally MAC learning entries are removed only from the ofproto associated
2394 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2395 * are removed from every ofproto. When patch ports and SLB bonds are in use
2396 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2397 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2398 * with the host from which it migrated. */
2400 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2402 struct ofproto_dpif *ofproto = bundle->ofproto;
2403 struct mac_learning *ml = ofproto->ml;
2404 struct mac_entry *mac, *next_mac;
2406 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2407 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2408 if (mac->port.p == bundle) {
2410 struct ofproto_dpif *o;
2412 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2414 struct mac_entry *e;
2416 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan,
2419 mac_learning_expire(o->ml, e);
2425 mac_learning_expire(ml, mac);
2430 static struct ofbundle *
2431 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2433 struct ofbundle *bundle;
2435 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2436 &ofproto->bundles) {
2437 if (bundle->aux == aux) {
2444 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
2445 * ones that are found to 'bundles'. */
2447 bundle_lookup_multiple(struct ofproto_dpif *ofproto,
2448 void **auxes, size_t n_auxes,
2449 struct hmapx *bundles)
2453 hmapx_init(bundles);
2454 for (i = 0; i < n_auxes; i++) {
2455 struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
2457 hmapx_add(bundles, bundle);
2463 bundle_update(struct ofbundle *bundle)
2465 struct ofport_dpif *port;
2467 bundle->floodable = true;
2468 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2469 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2470 || !stp_forward_in_state(port->stp_state)) {
2471 bundle->floodable = false;
2478 bundle_del_port(struct ofport_dpif *port)
2480 struct ofbundle *bundle = port->bundle;
2482 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2484 list_remove(&port->bundle_node);
2485 port->bundle = NULL;
2488 lacp_slave_unregister(bundle->lacp, port);
2491 bond_slave_unregister(bundle->bond, port);
2494 bundle_update(bundle);
2498 bundle_add_port(struct ofbundle *bundle, uint16_t ofp_port,
2499 struct lacp_slave_settings *lacp)
2501 struct ofport_dpif *port;
2503 port = get_ofp_port(bundle->ofproto, ofp_port);
2508 if (port->bundle != bundle) {
2509 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2511 bundle_del_port(port);
2514 port->bundle = bundle;
2515 list_push_back(&bundle->ports, &port->bundle_node);
2516 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2517 || !stp_forward_in_state(port->stp_state)) {
2518 bundle->floodable = false;
2522 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2523 lacp_slave_register(bundle->lacp, port, lacp);
2530 bundle_destroy(struct ofbundle *bundle)
2532 struct ofproto_dpif *ofproto;
2533 struct ofport_dpif *port, *next_port;
2540 ofproto = bundle->ofproto;
2541 for (i = 0; i < MAX_MIRRORS; i++) {
2542 struct ofmirror *m = ofproto->mirrors[i];
2544 if (m->out == bundle) {
2546 } else if (hmapx_find_and_delete(&m->srcs, bundle)
2547 || hmapx_find_and_delete(&m->dsts, bundle)) {
2548 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2553 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2554 bundle_del_port(port);
2557 bundle_flush_macs(bundle, true);
2558 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2560 free(bundle->trunks);
2561 lacp_destroy(bundle->lacp);
2562 bond_destroy(bundle->bond);
2567 bundle_set(struct ofproto *ofproto_, void *aux,
2568 const struct ofproto_bundle_settings *s)
2570 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2571 bool need_flush = false;
2572 struct ofport_dpif *port;
2573 struct ofbundle *bundle;
2574 unsigned long *trunks;
2580 bundle_destroy(bundle_lookup(ofproto, aux));
2584 ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2585 ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2587 bundle = bundle_lookup(ofproto, aux);
2589 bundle = xmalloc(sizeof *bundle);
2591 bundle->ofproto = ofproto;
2592 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2593 hash_pointer(aux, 0));
2595 bundle->name = NULL;
2597 list_init(&bundle->ports);
2598 bundle->vlan_mode = PORT_VLAN_TRUNK;
2600 bundle->trunks = NULL;
2601 bundle->use_priority_tags = s->use_priority_tags;
2602 bundle->lacp = NULL;
2603 bundle->bond = NULL;
2605 bundle->floodable = true;
2607 bundle->src_mirrors = 0;
2608 bundle->dst_mirrors = 0;
2609 bundle->mirror_out = 0;
2612 if (!bundle->name || strcmp(s->name, bundle->name)) {
2614 bundle->name = xstrdup(s->name);
2619 if (!bundle->lacp) {
2620 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2621 bundle->lacp = lacp_create();
2623 lacp_configure(bundle->lacp, s->lacp);
2625 lacp_destroy(bundle->lacp);
2626 bundle->lacp = NULL;
2629 /* Update set of ports. */
2631 for (i = 0; i < s->n_slaves; i++) {
2632 if (!bundle_add_port(bundle, s->slaves[i],
2633 s->lacp ? &s->lacp_slaves[i] : NULL)) {
2637 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2638 struct ofport_dpif *next_port;
2640 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2641 for (i = 0; i < s->n_slaves; i++) {
2642 if (s->slaves[i] == port->up.ofp_port) {
2647 bundle_del_port(port);
2651 ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2653 if (list_is_empty(&bundle->ports)) {
2654 bundle_destroy(bundle);
2658 /* Set VLAN tagging mode */
2659 if (s->vlan_mode != bundle->vlan_mode
2660 || s->use_priority_tags != bundle->use_priority_tags) {
2661 bundle->vlan_mode = s->vlan_mode;
2662 bundle->use_priority_tags = s->use_priority_tags;
2667 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2668 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2670 if (vlan != bundle->vlan) {
2671 bundle->vlan = vlan;
2675 /* Get trunked VLANs. */
2676 switch (s->vlan_mode) {
2677 case PORT_VLAN_ACCESS:
2681 case PORT_VLAN_TRUNK:
2682 trunks = CONST_CAST(unsigned long *, s->trunks);
2685 case PORT_VLAN_NATIVE_UNTAGGED:
2686 case PORT_VLAN_NATIVE_TAGGED:
2687 if (vlan != 0 && (!s->trunks
2688 || !bitmap_is_set(s->trunks, vlan)
2689 || bitmap_is_set(s->trunks, 0))) {
2690 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2692 trunks = bitmap_clone(s->trunks, 4096);
2694 trunks = bitmap_allocate1(4096);
2696 bitmap_set1(trunks, vlan);
2697 bitmap_set0(trunks, 0);
2699 trunks = CONST_CAST(unsigned long *, s->trunks);
2706 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2707 free(bundle->trunks);
2708 if (trunks == s->trunks) {
2709 bundle->trunks = vlan_bitmap_clone(trunks);
2711 bundle->trunks = trunks;
2716 if (trunks != s->trunks) {
2721 if (!list_is_short(&bundle->ports)) {
2722 bundle->ofproto->has_bonded_bundles = true;
2724 if (bond_reconfigure(bundle->bond, s->bond)) {
2725 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2728 bundle->bond = bond_create(s->bond);
2729 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2732 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2733 bond_slave_register(bundle->bond, port, port->up.netdev);
2736 bond_destroy(bundle->bond);
2737 bundle->bond = NULL;
2740 /* If we changed something that would affect MAC learning, un-learn
2741 * everything on this port and force flow revalidation. */
2743 bundle_flush_macs(bundle, false);
2750 bundle_remove(struct ofport *port_)
2752 struct ofport_dpif *port = ofport_dpif_cast(port_);
2753 struct ofbundle *bundle = port->bundle;
2756 bundle_del_port(port);
2757 if (list_is_empty(&bundle->ports)) {
2758 bundle_destroy(bundle);
2759 } else if (list_is_short(&bundle->ports)) {
2760 bond_destroy(bundle->bond);
2761 bundle->bond = NULL;
2767 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2769 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2770 struct ofport_dpif *port = port_;
2771 uint8_t ea[ETH_ADDR_LEN];
2774 error = netdev_get_etheraddr(port->up.netdev, ea);
2776 struct ofpbuf packet;
2779 ofpbuf_init(&packet, 0);
2780 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2782 memcpy(packet_pdu, pdu, pdu_size);
2784 send_packet(port, &packet);
2785 ofpbuf_uninit(&packet);
2787 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2788 "%s (%s)", port->bundle->name,
2789 netdev_get_name(port->up.netdev), strerror(error));
2794 bundle_send_learning_packets(struct ofbundle *bundle)
2796 struct ofproto_dpif *ofproto = bundle->ofproto;
2797 int error, n_packets, n_errors;
2798 struct mac_entry *e;
2800 error = n_packets = n_errors = 0;
2801 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2802 if (e->port.p != bundle) {
2803 struct ofpbuf *learning_packet;
2804 struct ofport_dpif *port;
2808 /* The assignment to "port" is unnecessary but makes "grep"ing for
2809 * struct ofport_dpif more effective. */
2810 learning_packet = bond_compose_learning_packet(bundle->bond,
2814 ret = send_packet(port, learning_packet);
2815 ofpbuf_delete(learning_packet);
2825 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2826 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2827 "packets, last error was: %s",
2828 bundle->name, n_errors, n_packets, strerror(error));
2830 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2831 bundle->name, n_packets);
2836 bundle_run(struct ofbundle *bundle)
2839 lacp_run(bundle->lacp, send_pdu_cb);
2842 struct ofport_dpif *port;
2844 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2845 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2848 bond_run(bundle->bond, &bundle->ofproto->backer->revalidate_set,
2849 lacp_status(bundle->lacp));
2850 if (bond_should_send_learning_packets(bundle->bond)) {
2851 bundle_send_learning_packets(bundle);
2857 bundle_wait(struct ofbundle *bundle)
2860 lacp_wait(bundle->lacp);
2863 bond_wait(bundle->bond);
2870 mirror_scan(struct ofproto_dpif *ofproto)
2874 for (idx = 0; idx < MAX_MIRRORS; idx++) {
2875 if (!ofproto->mirrors[idx]) {
2882 static struct ofmirror *
2883 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
2887 for (i = 0; i < MAX_MIRRORS; i++) {
2888 struct ofmirror *mirror = ofproto->mirrors[i];
2889 if (mirror && mirror->aux == aux) {
2897 /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */
2899 mirror_update_dups(struct ofproto_dpif *ofproto)
2903 for (i = 0; i < MAX_MIRRORS; i++) {
2904 struct ofmirror *m = ofproto->mirrors[i];
2907 m->dup_mirrors = MIRROR_MASK_C(1) << i;
2911 for (i = 0; i < MAX_MIRRORS; i++) {
2912 struct ofmirror *m1 = ofproto->mirrors[i];
2919 for (j = i + 1; j < MAX_MIRRORS; j++) {
2920 struct ofmirror *m2 = ofproto->mirrors[j];
2922 if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) {
2923 m1->dup_mirrors |= MIRROR_MASK_C(1) << j;
2924 m2->dup_mirrors |= m1->dup_mirrors;
2931 mirror_set(struct ofproto *ofproto_, void *aux,
2932 const struct ofproto_mirror_settings *s)
2934 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2935 mirror_mask_t mirror_bit;
2936 struct ofbundle *bundle;
2937 struct ofmirror *mirror;
2938 struct ofbundle *out;
2939 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
2940 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
2943 mirror = mirror_lookup(ofproto, aux);
2945 mirror_destroy(mirror);
2951 idx = mirror_scan(ofproto);
2953 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
2955 ofproto->up.name, MAX_MIRRORS, s->name);
2959 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
2960 mirror->ofproto = ofproto;
2963 mirror->out_vlan = -1;
2964 mirror->name = NULL;
2967 if (!mirror->name || strcmp(s->name, mirror->name)) {
2969 mirror->name = xstrdup(s->name);
2972 /* Get the new configuration. */
2973 if (s->out_bundle) {
2974 out = bundle_lookup(ofproto, s->out_bundle);
2976 mirror_destroy(mirror);
2982 out_vlan = s->out_vlan;
2984 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
2985 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
2987 /* If the configuration has not changed, do nothing. */
2988 if (hmapx_equals(&srcs, &mirror->srcs)
2989 && hmapx_equals(&dsts, &mirror->dsts)
2990 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
2991 && mirror->out == out
2992 && mirror->out_vlan == out_vlan)
2994 hmapx_destroy(&srcs);
2995 hmapx_destroy(&dsts);
2999 hmapx_swap(&srcs, &mirror->srcs);
3000 hmapx_destroy(&srcs);
3002 hmapx_swap(&dsts, &mirror->dsts);
3003 hmapx_destroy(&dsts);
3005 free(mirror->vlans);
3006 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
3009 mirror->out_vlan = out_vlan;
3011 /* Update bundles. */
3012 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
3013 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
3014 if (hmapx_contains(&mirror->srcs, bundle)) {
3015 bundle->src_mirrors |= mirror_bit;
3017 bundle->src_mirrors &= ~mirror_bit;
3020 if (hmapx_contains(&mirror->dsts, bundle)) {
3021 bundle->dst_mirrors |= mirror_bit;
3023 bundle->dst_mirrors &= ~mirror_bit;
3026 if (mirror->out == bundle) {
3027 bundle->mirror_out |= mirror_bit;
3029 bundle->mirror_out &= ~mirror_bit;
3033 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3034 ofproto->has_mirrors = true;
3035 mac_learning_flush(ofproto->ml,
3036 &ofproto->backer->revalidate_set);
3037 mirror_update_dups(ofproto);
3043 mirror_destroy(struct ofmirror *mirror)
3045 struct ofproto_dpif *ofproto;
3046 mirror_mask_t mirror_bit;
3047 struct ofbundle *bundle;
3054 ofproto = mirror->ofproto;
3055 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3056 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
3058 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
3059 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
3060 bundle->src_mirrors &= ~mirror_bit;
3061 bundle->dst_mirrors &= ~mirror_bit;
3062 bundle->mirror_out &= ~mirror_bit;
3065 hmapx_destroy(&mirror->srcs);
3066 hmapx_destroy(&mirror->dsts);
3067 free(mirror->vlans);
3069 ofproto->mirrors[mirror->idx] = NULL;
3073 mirror_update_dups(ofproto);
3075 ofproto->has_mirrors = false;
3076 for (i = 0; i < MAX_MIRRORS; i++) {
3077 if (ofproto->mirrors[i]) {
3078 ofproto->has_mirrors = true;
3085 mirror_get_stats(struct ofproto *ofproto_, void *aux,
3086 uint64_t *packets, uint64_t *bytes)
3088 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3089 struct ofmirror *mirror = mirror_lookup(ofproto, aux);
3092 *packets = *bytes = UINT64_MAX;
3098 *packets = mirror->packet_count;
3099 *bytes = mirror->byte_count;
3105 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
3107 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3108 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
3109 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
3115 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
3117 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3118 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
3119 return bundle && bundle->mirror_out != 0;
3123 forward_bpdu_changed(struct ofproto *ofproto_)
3125 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3126 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3130 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
3133 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3134 mac_learning_set_idle_time(ofproto->ml, idle_time);
3135 mac_learning_set_max_entries(ofproto->ml, max_entries);
3140 static struct ofport_dpif *
3141 get_ofp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
3143 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
3144 return ofport ? ofport_dpif_cast(ofport) : NULL;
3147 static struct ofport_dpif *
3148 get_odp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
3150 struct ofport_dpif *port = odp_port_to_ofport(ofproto->backer, odp_port);
3151 return port && &ofproto->up == port->up.ofproto ? port : NULL;
3155 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
3156 struct ofproto_port *ofproto_port,
3157 struct dpif_port *dpif_port)
3159 ofproto_port->name = dpif_port->name;
3160 ofproto_port->type = dpif_port->type;
3161 ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
3164 static struct ofport_dpif *
3165 ofport_get_peer(const struct ofport_dpif *ofport_dpif)
3167 const struct ofproto_dpif *ofproto;
3170 peer = netdev_vport_patch_peer(ofport_dpif->up.netdev);
3175 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
3176 struct ofport *ofport;
3178 ofport = shash_find_data(&ofproto->up.port_by_name, peer);
3179 if (ofport && ofport->ofproto->ofproto_class == &ofproto_dpif_class) {
3180 return ofport_dpif_cast(ofport);
3187 port_run_fast(struct ofport_dpif *ofport)
3189 if (ofport->cfm && cfm_should_send_ccm(ofport->cfm)) {
3190 struct ofpbuf packet;
3192 ofpbuf_init(&packet, 0);
3193 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
3194 send_packet(ofport, &packet);
3195 ofpbuf_uninit(&packet);
3198 if (ofport->bfd && bfd_should_send_packet(ofport->bfd)) {
3199 struct ofpbuf packet;
3201 ofpbuf_init(&packet, 0);
3202 bfd_put_packet(ofport->bfd, &packet, ofport->up.pp.hw_addr);
3203 send_packet(ofport, &packet);
3204 ofpbuf_uninit(&packet);
3209 port_run(struct ofport_dpif *ofport)
3211 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
3212 bool carrier_changed = carrier_seq != ofport->carrier_seq;
3213 bool enable = netdev_get_carrier(ofport->up.netdev);
3215 ofport->carrier_seq = carrier_seq;
3217 port_run_fast(ofport);
3219 if (ofport->tnl_port
3220 && tnl_port_reconfigure(&ofport->up, ofport->odp_port,
3221 &ofport->tnl_port)) {
3222 ofproto_dpif_cast(ofport->up.ofproto)->backer->need_revalidate = true;
3226 int cfm_opup = cfm_get_opup(ofport->cfm);
3228 cfm_run(ofport->cfm);
3229 enable = enable && !cfm_get_fault(ofport->cfm);
3231 if (cfm_opup >= 0) {
3232 enable = enable && cfm_opup;
3237 bfd_run(ofport->bfd);
3238 enable = enable && bfd_forwarding(ofport->bfd);
3241 if (ofport->bundle) {
3242 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
3243 if (carrier_changed) {
3244 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
3248 if (ofport->may_enable != enable) {
3249 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3251 if (ofproto->has_bundle_action) {
3252 ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
3256 ofport->may_enable = enable;
3260 port_wait(struct ofport_dpif *ofport)
3263 cfm_wait(ofport->cfm);
3267 bfd_wait(ofport->bfd);
3272 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
3273 struct ofproto_port *ofproto_port)
3275 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3276 struct dpif_port dpif_port;
3279 if (sset_contains(&ofproto->ghost_ports, devname)) {
3280 const char *type = netdev_get_type_from_name(devname);
3282 /* We may be called before ofproto->up.port_by_name is populated with
3283 * the appropriate ofport. For this reason, we must get the name and
3284 * type from the netdev layer directly. */
3286 const struct ofport *ofport;
3288 ofport = shash_find_data(&ofproto->up.port_by_name, devname);
3289 ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
3290 ofproto_port->name = xstrdup(devname);
3291 ofproto_port->type = xstrdup(type);
3297 if (!sset_contains(&ofproto->ports, devname)) {
3300 error = dpif_port_query_by_name(ofproto->backer->dpif,
3301 devname, &dpif_port);
3303 ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
3309 port_add(struct ofproto *ofproto_, struct netdev *netdev)
3311 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3312 const char *dp_port_name = netdev_vport_get_dpif_port(netdev);
3313 const char *devname = netdev_get_name(netdev);
3315 if (netdev_vport_is_patch(netdev)) {
3316 sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
3320 if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
3321 uint32_t port_no = UINT32_MAX;
3324 error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
3328 if (netdev_get_tunnel_config(netdev)) {
3329 simap_put(&ofproto->backer->tnl_backers, dp_port_name, port_no);
3333 if (netdev_get_tunnel_config(netdev)) {
3334 sset_add(&ofproto->ghost_ports, devname);
3336 sset_add(&ofproto->ports, devname);
3342 port_del(struct ofproto *ofproto_, uint16_t ofp_port)
3344 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3345 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
3352 sset_find_and_delete(&ofproto->ghost_ports,
3353 netdev_get_name(ofport->up.netdev));
3354 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3355 if (!ofport->tnl_port) {
3356 error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
3358 /* The caller is going to close ofport->up.netdev. If this is a
3359 * bonded port, then the bond is using that netdev, so remove it
3360 * from the bond. The client will need to reconfigure everything
3361 * after deleting ports, so then the slave will get re-added. */
3362 bundle_remove(&ofport->up);
3369 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
3371 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3376 error = netdev_get_stats(ofport->up.netdev, stats);
3378 if (!error && ofport_->ofp_port == OFPP_LOCAL) {
3379 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3381 /* ofproto->stats.tx_packets represents packets that we created
3382 * internally and sent to some port (e.g. packets sent with
3383 * send_packet()). Account for them as if they had come from
3384 * OFPP_LOCAL and got forwarded. */
3386 if (stats->rx_packets != UINT64_MAX) {
3387 stats->rx_packets += ofproto->stats.tx_packets;
3390 if (stats->rx_bytes != UINT64_MAX) {
3391 stats->rx_bytes += ofproto->stats.tx_bytes;
3394 /* ofproto->stats.rx_packets represents packets that were received on
3395 * some port and we processed internally and dropped (e.g. STP).
3396 * Account for them as if they had been forwarded to OFPP_LOCAL. */
3398 if (stats->tx_packets != UINT64_MAX) {
3399 stats->tx_packets += ofproto->stats.rx_packets;
3402 if (stats->tx_bytes != UINT64_MAX) {
3403 stats->tx_bytes += ofproto->stats.rx_bytes;
3410 struct port_dump_state {
3415 struct ofproto_port port;
3420 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
3422 *statep = xzalloc(sizeof(struct port_dump_state));
3427 port_dump_next(const struct ofproto *ofproto_, void *state_,
3428 struct ofproto_port *port)
3430 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3431 struct port_dump_state *state = state_;
3432 const struct sset *sset;
3433 struct sset_node *node;
3435 if (state->has_port) {
3436 ofproto_port_destroy(&state->port);
3437 state->has_port = false;
3439 sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
3440 while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
3443 error = port_query_by_name(ofproto_, node->name, &state->port);
3445 *port = state->port;
3446 state->has_port = true;
3448 } else if (error != ENODEV) {
3453 if (!state->ghost) {
3454 state->ghost = true;
3457 return port_dump_next(ofproto_, state_, port);
3464 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3466 struct port_dump_state *state = state_;
3468 if (state->has_port) {
3469 ofproto_port_destroy(&state->port);
3476 port_poll(const struct ofproto *ofproto_, char **devnamep)
3478 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3480 if (ofproto->port_poll_errno) {
3481 int error = ofproto->port_poll_errno;
3482 ofproto->port_poll_errno = 0;
3486 if (sset_is_empty(&ofproto->port_poll_set)) {
3490 *devnamep = sset_pop(&ofproto->port_poll_set);
3495 port_poll_wait(const struct ofproto *ofproto_)
3497 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3498 dpif_port_poll_wait(ofproto->backer->dpif);
3502 port_is_lacp_current(const struct ofport *ofport_)
3504 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3505 return (ofport->bundle && ofport->bundle->lacp
3506 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3510 /* Upcall handling. */
3512 /* Flow miss batching.
3514 * Some dpifs implement operations faster when you hand them off in a batch.
3515 * To allow batching, "struct flow_miss" queues the dpif-related work needed
3516 * for a given flow. Each "struct flow_miss" corresponds to sending one or
3517 * more packets, plus possibly installing the flow in the dpif.
3519 * So far we only batch the operations that affect flow setup time the most.
3520 * It's possible to batch more than that, but the benefit might be minimal. */
3522 struct hmap_node hmap_node;
3523 struct ofproto_dpif *ofproto;
3525 enum odp_key_fitness key_fitness;
3526 const struct nlattr *key;
3528 struct initial_vals initial_vals;
3529 struct list packets;
3530 enum dpif_upcall_type upcall_type;
3533 struct flow_miss_op {
3534 struct dpif_op dpif_op;
3536 uint64_t slow_stub[128 / 8]; /* Buffer for compose_slow_path() */
3537 struct xlate_out xout;
3538 bool xout_garbage; /* 'xout' needs to be uninitialized? */
3541 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
3542 * OpenFlow controller as necessary according to their individual
3543 * configurations. */
3545 send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet,
3546 const struct flow *flow)
3548 struct ofputil_packet_in pin;
3550 pin.packet = packet->data;
3551 pin.packet_len = packet->size;
3552 pin.reason = OFPR_NO_MATCH;
3553 pin.controller_id = 0;
3558 pin.send_len = 0; /* not used for flow table misses */
3560 flow_get_metadata(flow, &pin.fmd);
3562 connmgr_send_packet_in(ofproto->up.connmgr, &pin);
3565 static enum slow_path_reason
3566 process_special(struct ofproto_dpif *ofproto, const struct flow *flow,
3567 const struct ofport_dpif *ofport, const struct ofpbuf *packet)
3571 } else if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) {
3573 cfm_process_heartbeat(ofport->cfm, packet);
3576 } else if (ofport->bfd && bfd_should_process_flow(flow)) {
3578 bfd_process_packet(ofport->bfd, flow, packet);
3581 } else if (ofport->bundle && ofport->bundle->lacp
3582 && flow->dl_type == htons(ETH_TYPE_LACP)) {
3584 lacp_process_packet(ofport->bundle->lacp, ofport, packet);
3587 } else if (ofproto->stp && stp_should_process_flow(flow)) {
3589 stp_process_packet(ofport, packet);
3597 static struct flow_miss *
3598 flow_miss_find(struct hmap *todo, const struct ofproto_dpif *ofproto,
3599 const struct flow *flow, uint32_t hash)
3601 struct flow_miss *miss;
3603 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
3604 if (miss->ofproto == ofproto && flow_equal(&miss->flow, flow)) {
3612 /* Partially Initializes 'op' as an "execute" operation for 'miss' and
3613 * 'packet'. The caller must initialize op->actions and op->actions_len. If
3614 * 'miss' is associated with a subfacet the caller must also initialize the
3615 * returned op->subfacet, and if anything needs to be freed after processing
3616 * the op, the caller must initialize op->garbage also. */
3618 init_flow_miss_execute_op(struct flow_miss *miss, struct ofpbuf *packet,
3619 struct flow_miss_op *op)
3621 if (miss->flow.vlan_tci != miss->initial_vals.vlan_tci) {
3622 /* This packet was received on a VLAN splinter port. We
3623 * added a VLAN to the packet to make the packet resemble
3624 * the flow, but the actions were composed assuming that
3625 * the packet contained no VLAN. So, we must remove the
3626 * VLAN header from the packet before trying to execute the
3628 eth_pop_vlan(packet);
3631 op->xout_garbage = false;
3632 op->dpif_op.type = DPIF_OP_EXECUTE;
3633 op->dpif_op.u.execute.key = miss->key;
3634 op->dpif_op.u.execute.key_len = miss->key_len;
3635 op->dpif_op.u.execute.packet = packet;
3638 /* Helper for handle_flow_miss_without_facet() and
3639 * handle_flow_miss_with_facet(). */
3641 handle_flow_miss_common(struct rule_dpif *rule,
3642 struct ofpbuf *packet, const struct flow *flow)
3644 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3646 if (rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
3648 * Extra-special case for fail-open mode.
3650 * We are in fail-open mode and the packet matched the fail-open
3651 * rule, but we are connected to a controller too. We should send
3652 * the packet up to the controller in the hope that it will try to
3653 * set up a flow and thereby allow us to exit fail-open.
3655 * See the top-level comment in fail-open.c for more information.
3657 send_packet_in_miss(ofproto, packet, flow);
3661 /* Figures out whether a flow that missed in 'ofproto', whose details are in
3662 * 'miss', is likely to be worth tracking in detail in userspace and (usually)
3663 * installing a datapath flow. The answer is usually "yes" (a return value of
3664 * true). However, for short flows the cost of bookkeeping is much higher than
3665 * the benefits, so when the datapath holds a large number of flows we impose
3666 * some heuristics to decide which flows are likely to be worth tracking. */
3668 flow_miss_should_make_facet(struct ofproto_dpif *ofproto,
3669 struct flow_miss *miss, uint32_t hash)
3671 if (!ofproto->governor) {
3674 n_subfacets = hmap_count(&ofproto->subfacets);
3675 if (n_subfacets * 2 <= ofproto->up.flow_eviction_threshold) {
3679 ofproto->governor = governor_create(ofproto->up.name);
3682 return governor_should_install_flow(ofproto->governor, hash,
3683 list_size(&miss->packets));
3686 /* Handles 'miss', which matches 'rule', without creating a facet or subfacet
3687 * or creating any datapath flow. May add an "execute" operation to 'ops' and
3688 * increment '*n_ops'. */
3690 handle_flow_miss_without_facet(struct flow_miss *miss,
3691 struct flow_miss_op *ops, size_t *n_ops)
3693 struct rule_dpif *rule = rule_dpif_lookup(miss->ofproto, &miss->flow);
3694 long long int now = time_msec();
3695 struct ofpbuf *packet;
3696 struct xlate_in xin;
3698 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3699 struct flow_miss_op *op = &ops[*n_ops];
3700 struct dpif_flow_stats stats;
3702 COVERAGE_INC(facet_suppress);
3704 handle_flow_miss_common(rule, packet, &miss->flow);
3706 dpif_flow_stats_extract(&miss->flow, packet, now, &stats);
3707 rule_credit_stats(rule, &stats);
3709 xlate_in_init(&xin, miss->ofproto, &miss->flow, &miss->initial_vals,
3710 rule, stats.tcp_flags, packet);
3711 xin.resubmit_stats = &stats;
3712 xlate_actions(&xin, &op->xout);
3714 if (op->xout.odp_actions.size) {
3715 struct dpif_execute *execute = &op->dpif_op.u.execute;
3717 init_flow_miss_execute_op(miss, packet, op);
3718 execute->actions = op->xout.odp_actions.data;
3719 execute->actions_len = op->xout.odp_actions.size;
3720 op->xout_garbage = true;
3724 xlate_out_uninit(&op->xout);
3729 /* Handles 'miss', which matches 'facet'. May add any required datapath
3730 * operations to 'ops', incrementing '*n_ops' for each new op.
3732 * All of the packets in 'miss' are considered to have arrived at time 'now'.
3733 * This is really important only for new facets: if we just called time_msec()
3734 * here, then the new subfacet or its packets could look (occasionally) as
3735 * though it was used some time after the facet was used. That can make a
3736 * one-packet flow look like it has a nonzero duration, which looks odd in
3737 * e.g. NetFlow statistics. */
3739 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
3741 struct flow_miss_op *ops, size_t *n_ops)
3743 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3744 enum subfacet_path want_path;
3745 struct subfacet *subfacet;
3746 struct ofpbuf *packet;
3748 subfacet = subfacet_create(facet, miss, now);
3749 want_path = subfacet->facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
3751 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3752 struct flow_miss_op *op = &ops[*n_ops];
3753 struct dpif_flow_stats stats;
3755 handle_flow_miss_common(facet->rule, packet, &miss->flow);
3757 if (want_path != SF_FAST_PATH) {
3758 struct xlate_in xin;
3760 xlate_in_init(&xin, ofproto, &facet->flow, &facet->initial_vals,
3761 facet->rule, 0, packet);
3762 xlate_actions_for_side_effects(&xin);
3765 dpif_flow_stats_extract(&facet->flow, packet, now, &stats);
3766 subfacet_update_stats(subfacet, &stats);
3768 if (facet->xout.odp_actions.size) {
3769 struct dpif_execute *execute = &op->dpif_op.u.execute;
3771 init_flow_miss_execute_op(miss, packet, op);
3772 execute->actions = facet->xout.odp_actions.data,
3773 execute->actions_len = facet->xout.odp_actions.size;
3778 if (miss->upcall_type == DPIF_UC_MISS || subfacet->path != want_path) {
3779 struct flow_miss_op *op = &ops[(*n_ops)++];
3780 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
3782 subfacet->path = want_path;
3784 op->xout_garbage = false;
3785 op->dpif_op.type = DPIF_OP_FLOW_PUT;
3786 put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
3787 put->key = miss->key;
3788 put->key_len = miss->key_len;
3789 if (want_path == SF_FAST_PATH) {
3790 put->actions = facet->xout.odp_actions.data;
3791 put->actions_len = facet->xout.odp_actions.size;
3793 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
3794 op->slow_stub, sizeof op->slow_stub,
3795 &put->actions, &put->actions_len);
3801 /* Handles flow miss 'miss'. May add any required datapath operations
3802 * to 'ops', incrementing '*n_ops' for each new op. */
3804 handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops,
3807 struct ofproto_dpif *ofproto = miss->ofproto;
3808 struct facet *facet;
3812 /* The caller must ensure that miss->hmap_node.hash contains
3813 * flow_hash(miss->flow, 0). */
3814 hash = miss->hmap_node.hash;
3816 facet = facet_lookup_valid(ofproto, &miss->flow, hash);
3818 /* There does not exist a bijection between 'struct flow' and datapath
3819 * flow keys with fitness ODP_FIT_TO_LITTLE. This breaks a fundamental
3820 * assumption used throughout the facet and subfacet handling code.
3821 * Since we have to handle these misses in userspace anyway, we simply
3822 * skip facet creation, avoiding the problem alltogether. */
3823 if (miss->key_fitness == ODP_FIT_TOO_LITTLE
3824 || !flow_miss_should_make_facet(ofproto, miss, hash)) {
3825 handle_flow_miss_without_facet(miss, ops, n_ops);
3829 facet = facet_create(miss, hash);
3834 handle_flow_miss_with_facet(miss, facet, now, ops, n_ops);
3837 static struct drop_key *
3838 drop_key_lookup(const struct dpif_backer *backer, const struct nlattr *key,
3841 struct drop_key *drop_key;
3843 HMAP_FOR_EACH_WITH_HASH (drop_key, hmap_node, hash_bytes(key, key_len, 0),
3844 &backer->drop_keys) {
3845 if (drop_key->key_len == key_len
3846 && !memcmp(drop_key->key, key, key_len)) {
3854 drop_key_clear(struct dpif_backer *backer)
3856 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3857 struct drop_key *drop_key, *next;
3859 HMAP_FOR_EACH_SAFE (drop_key, next, hmap_node, &backer->drop_keys) {
3862 error = dpif_flow_del(backer->dpif, drop_key->key, drop_key->key_len,
3864 if (error && !VLOG_DROP_WARN(&rl)) {
3865 struct ds ds = DS_EMPTY_INITIALIZER;
3866 odp_flow_key_format(drop_key->key, drop_key->key_len, &ds);
3867 VLOG_WARN("Failed to delete drop key (%s) (%s)", strerror(error),
3872 hmap_remove(&backer->drop_keys, &drop_key->hmap_node);
3873 free(drop_key->key);
3878 /* Given a datpath, packet, and flow metadata ('backer', 'packet', and 'key'
3879 * respectively), populates 'flow' with the result of odp_flow_key_to_flow().
3880 * Optionally, if nonnull, populates 'fitnessp' with the fitness of 'flow' as
3881 * returned by odp_flow_key_to_flow(). Also, optionally populates 'ofproto'
3882 * with the ofproto_dpif, and 'odp_in_port' with the datapath in_port, that
3883 * 'packet' ingressed.
3885 * If 'ofproto' is nonnull, requires 'flow''s in_port to exist. Otherwise sets
3886 * 'flow''s in_port to OFPP_NONE.
3888 * This function does post-processing on data returned from
3889 * odp_flow_key_to_flow() to help make VLAN splinters transparent to the rest
3890 * of the upcall processing logic. In particular, if the extracted in_port is
3891 * a VLAN splinter port, it replaces flow->in_port by the "real" port, sets
3892 * flow->vlan_tci correctly for the VLAN of the VLAN splinter port, and pushes
3893 * a VLAN header onto 'packet' (if it is nonnull).
3895 * Optionally, if 'initial_vals' is nonnull, sets 'initial_vals->vlan_tci'
3896 * to the VLAN TCI with which the packet was really received, that is, the
3897 * actual VLAN TCI extracted by odp_flow_key_to_flow(). (This differs from
3898 * the value returned in flow->vlan_tci only for packets received on
3901 * Similarly, this function also includes some logic to help with tunnels. It
3902 * may modify 'flow' as necessary to make the tunneling implementation
3903 * transparent to the upcall processing logic.
3905 * Returns 0 if successful, ENODEV if the parsed flow has no associated ofport,
3906 * or some other positive errno if there are other problems. */
3908 ofproto_receive(const struct dpif_backer *backer, struct ofpbuf *packet,
3909 const struct nlattr *key, size_t key_len,
3910 struct flow *flow, enum odp_key_fitness *fitnessp,
3911 struct ofproto_dpif **ofproto, uint32_t *odp_in_port,
3912 struct initial_vals *initial_vals)
3914 const struct ofport_dpif *port;
3915 enum odp_key_fitness fitness;
3918 fitness = odp_flow_key_to_flow(key, key_len, flow);
3919 if (fitness == ODP_FIT_ERROR) {
3925 initial_vals->vlan_tci = flow->vlan_tci;
3929 *odp_in_port = flow->in_port;
3932 port = (tnl_port_should_receive(flow)
3933 ? ofport_dpif_cast(tnl_port_receive(flow))
3934 : odp_port_to_ofport(backer, flow->in_port));
3935 flow->in_port = port ? port->up.ofp_port : OFPP_NONE;
3940 /* XXX: Since the tunnel module is not scoped per backer, for a tunnel port
3941 * it's theoretically possible that we'll receive an ofport belonging to an
3942 * entirely different datapath. In practice, this can't happen because no
3943 * platforms has two separate datapaths which each support tunneling. */
3944 ovs_assert(ofproto_dpif_cast(port->up.ofproto)->backer == backer);
3946 if (vsp_adjust_flow(ofproto_dpif_cast(port->up.ofproto), flow)) {
3948 /* Make the packet resemble the flow, so that it gets sent to
3949 * an OpenFlow controller properly, so that it looks correct
3950 * for sFlow, and so that flow_extract() will get the correct
3951 * vlan_tci if it is called on 'packet'.
3953 * The allocated space inside 'packet' probably also contains
3954 * 'key', that is, both 'packet' and 'key' are probably part of
3955 * a struct dpif_upcall (see the large comment on that
3956 * structure definition), so pushing data on 'packet' is in
3957 * general not a good idea since it could overwrite 'key' or
3958 * free it as a side effect. However, it's OK in this special
3959 * case because we know that 'packet' is inside a Netlink
3960 * attribute: pushing 4 bytes will just overwrite the 4-byte
3961 * "struct nlattr", which is fine since we don't need that
3962 * header anymore. */
3963 eth_push_vlan(packet, flow->vlan_tci);
3965 /* We can't reproduce 'key' from 'flow'. */
3966 fitness = fitness == ODP_FIT_PERFECT ? ODP_FIT_TOO_MUCH : fitness;
3971 *ofproto = ofproto_dpif_cast(port->up.ofproto);
3976 *fitnessp = fitness;
3982 handle_miss_upcalls(struct dpif_backer *backer, struct dpif_upcall *upcalls,
3985 struct dpif_upcall *upcall;
3986 struct flow_miss *miss;
3987 struct flow_miss misses[FLOW_MISS_MAX_BATCH];
3988 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
3989 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
3999 /* Construct the to-do list.
4001 * This just amounts to extracting the flow from each packet and sticking
4002 * the packets that have the same flow in the same "flow_miss" structure so
4003 * that we can process them together. */
4006 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
4007 struct flow_miss *miss = &misses[n_misses];
4008 struct flow_miss *existing_miss;
4009 struct ofproto_dpif *ofproto;
4010 uint32_t odp_in_port;
4015 error = ofproto_receive(backer, upcall->packet, upcall->key,
4016 upcall->key_len, &flow, &miss->key_fitness,
4017 &ofproto, &odp_in_port, &miss->initial_vals);
4018 if (error == ENODEV) {
4019 struct drop_key *drop_key;
4021 /* Received packet on datapath port for which we couldn't
4022 * associate an ofproto. This can happen if a port is removed
4023 * while traffic is being received. Print a rate-limited message
4024 * in case it happens frequently. Install a drop flow so
4025 * that future packets of the flow are inexpensively dropped
4027 VLOG_INFO_RL(&rl, "received packet on unassociated datapath port "
4028 "%"PRIu32, odp_in_port);
4030 drop_key = drop_key_lookup(backer, upcall->key, upcall->key_len);
4032 drop_key = xmalloc(sizeof *drop_key);
4033 drop_key->key = xmemdup(upcall->key, upcall->key_len);
4034 drop_key->key_len = upcall->key_len;
4036 hmap_insert(&backer->drop_keys, &drop_key->hmap_node,
4037 hash_bytes(drop_key->key, drop_key->key_len, 0));
4038 dpif_flow_put(backer->dpif, DPIF_FP_CREATE | DPIF_FP_MODIFY,
4039 drop_key->key, drop_key->key_len, NULL, 0, NULL);
4047 ofproto->n_missed++;
4048 flow_extract(upcall->packet, flow.skb_priority, flow.skb_mark,
4049 &flow.tunnel, flow.in_port, &miss->flow);
4051 /* Add other packets to a to-do list. */
4052 hash = flow_hash(&miss->flow, 0);
4053 existing_miss = flow_miss_find(&todo, ofproto, &miss->flow, hash);
4054 if (!existing_miss) {
4055 hmap_insert(&todo, &miss->hmap_node, hash);
4056 miss->ofproto = ofproto;
4057 miss->key = upcall->key;
4058 miss->key_len = upcall->key_len;
4059 miss->upcall_type = upcall->type;
4060 list_init(&miss->packets);
4064 miss = existing_miss;
4066 list_push_back(&miss->packets, &upcall->packet->list_node);
4069 /* Process each element in the to-do list, constructing the set of
4070 * operations to batch. */
4072 HMAP_FOR_EACH (miss, hmap_node, &todo) {
4073 handle_flow_miss(miss, flow_miss_ops, &n_ops);
4075 ovs_assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
4077 /* Execute batch. */
4078 for (i = 0; i < n_ops; i++) {
4079 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
4081 dpif_operate(backer->dpif, dpif_ops, n_ops);
4084 for (i = 0; i < n_ops; i++) {
4085 if (flow_miss_ops[i].xout_garbage) {
4086 xlate_out_uninit(&flow_miss_ops[i].xout);
4089 hmap_destroy(&todo);
4092 static enum { SFLOW_UPCALL, MISS_UPCALL, BAD_UPCALL, FLOW_SAMPLE_UPCALL,
4094 classify_upcall(const struct dpif_upcall *upcall)
4096 size_t userdata_len;
4097 union user_action_cookie cookie;
4099 /* First look at the upcall type. */
4100 switch (upcall->type) {
4101 case DPIF_UC_ACTION:
4107 case DPIF_N_UC_TYPES:
4109 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
4113 /* "action" upcalls need a closer look. */
4114 if (!upcall->userdata) {
4115 VLOG_WARN_RL(&rl, "action upcall missing cookie");
4118 userdata_len = nl_attr_get_size(upcall->userdata);
4119 if (userdata_len < sizeof cookie.type
4120 || userdata_len > sizeof cookie) {
4121 VLOG_WARN_RL(&rl, "action upcall cookie has unexpected size %zu",
4125 memset(&cookie, 0, sizeof cookie);
4126 memcpy(&cookie, nl_attr_get(upcall->userdata), userdata_len);
4127 if (userdata_len == sizeof cookie.sflow
4128 && cookie.type == USER_ACTION_COOKIE_SFLOW) {
4129 return SFLOW_UPCALL;
4130 } else if (userdata_len == sizeof cookie.slow_path
4131 && cookie.type == USER_ACTION_COOKIE_SLOW_PATH) {
4133 } else if (userdata_len == sizeof cookie.flow_sample
4134 && cookie.type == USER_ACTION_COOKIE_FLOW_SAMPLE) {
4135 return FLOW_SAMPLE_UPCALL;
4136 } else if (userdata_len == sizeof cookie.ipfix
4137 && cookie.type == USER_ACTION_COOKIE_IPFIX) {
4138 return IPFIX_UPCALL;
4140 VLOG_WARN_RL(&rl, "invalid user cookie of type %"PRIu16
4141 " and size %zu", cookie.type, userdata_len);
4147 handle_sflow_upcall(struct dpif_backer *backer,
4148 const struct dpif_upcall *upcall)
4150 struct ofproto_dpif *ofproto;
4151 union user_action_cookie cookie;
4153 uint32_t odp_in_port;
4155 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
4156 &flow, NULL, &ofproto, &odp_in_port, NULL)
4157 || !ofproto->sflow) {
4161 memset(&cookie, 0, sizeof cookie);
4162 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.sflow);
4163 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
4164 odp_in_port, &cookie);
4168 handle_flow_sample_upcall(struct dpif_backer *backer,
4169 const struct dpif_upcall *upcall)
4171 struct ofproto_dpif *ofproto;
4172 union user_action_cookie cookie;
4175 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
4176 &flow, NULL, &ofproto, NULL, NULL)
4177 || !ofproto->ipfix) {
4181 memset(&cookie, 0, sizeof cookie);
4182 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.flow_sample);
4184 /* The flow reflects exactly the contents of the packet. Sample
4185 * the packet using it. */
4186 dpif_ipfix_flow_sample(ofproto->ipfix, upcall->packet, &flow,
4187 cookie.flow_sample.collector_set_id,
4188 cookie.flow_sample.probability,
4189 cookie.flow_sample.obs_domain_id,
4190 cookie.flow_sample.obs_point_id);
4194 handle_ipfix_upcall(struct dpif_backer *backer,
4195 const struct dpif_upcall *upcall)
4197 struct ofproto_dpif *ofproto;
4200 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
4201 &flow, NULL, &ofproto, NULL, NULL)
4202 || !ofproto->ipfix) {
4206 /* The flow reflects exactly the contents of the packet. Sample
4207 * the packet using it. */
4208 dpif_ipfix_bridge_sample(ofproto->ipfix, upcall->packet, &flow);
4212 handle_upcalls(struct dpif_backer *backer, unsigned int max_batch)
4214 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
4215 struct ofpbuf miss_bufs[FLOW_MISS_MAX_BATCH];
4216 uint64_t miss_buf_stubs[FLOW_MISS_MAX_BATCH][4096 / 8];
4221 ovs_assert(max_batch <= FLOW_MISS_MAX_BATCH);
4224 for (n_processed = 0; n_processed < max_batch; n_processed++) {
4225 struct dpif_upcall *upcall = &misses[n_misses];
4226 struct ofpbuf *buf = &miss_bufs[n_misses];
4229 ofpbuf_use_stub(buf, miss_buf_stubs[n_misses],
4230 sizeof miss_buf_stubs[n_misses]);
4231 error = dpif_recv(backer->dpif, upcall, buf);
4237 switch (classify_upcall(upcall)) {
4239 /* Handle it later. */
4244 handle_sflow_upcall(backer, upcall);
4248 case FLOW_SAMPLE_UPCALL:
4249 handle_flow_sample_upcall(backer, upcall);
4254 handle_ipfix_upcall(backer, upcall);
4264 /* Handle deferred MISS_UPCALL processing. */
4265 handle_miss_upcalls(backer, misses, n_misses);
4266 for (i = 0; i < n_misses; i++) {
4267 ofpbuf_uninit(&miss_bufs[i]);
4273 /* Flow expiration. */
4275 static int subfacet_max_idle(const struct ofproto_dpif *);
4276 static void update_stats(struct dpif_backer *);
4277 static void rule_expire(struct rule_dpif *);
4278 static void expire_subfacets(struct ofproto_dpif *, int dp_max_idle);
4280 /* This function is called periodically by run(). Its job is to collect
4281 * updates for the flows that have been installed into the datapath, most
4282 * importantly when they last were used, and then use that information to
4283 * expire flows that have not been used recently.
4285 * Returns the number of milliseconds after which it should be called again. */
4287 expire(struct dpif_backer *backer)
4289 struct ofproto_dpif *ofproto;
4290 int max_idle = INT32_MAX;
4292 /* Periodically clear out the drop keys in an effort to keep them
4293 * relatively few. */
4294 drop_key_clear(backer);
4296 /* Update stats for each flow in the backer. */
4297 update_stats(backer);
4299 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4300 struct rule *rule, *next_rule;
4303 if (ofproto->backer != backer) {
4307 /* Keep track of the max number of flows per ofproto_dpif. */
4308 update_max_subfacet_count(ofproto);
4310 /* Expire subfacets that have been idle too long. */
4311 dp_max_idle = subfacet_max_idle(ofproto);
4312 expire_subfacets(ofproto, dp_max_idle);
4314 max_idle = MIN(max_idle, dp_max_idle);
4316 /* Expire OpenFlow flows whose idle_timeout or hard_timeout
4318 LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
4319 &ofproto->up.expirable) {
4320 rule_expire(rule_dpif_cast(rule));
4323 /* All outstanding data in existing flows has been accounted, so it's a
4324 * good time to do bond rebalancing. */
4325 if (ofproto->has_bonded_bundles) {
4326 struct ofbundle *bundle;
4328 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
4330 bond_rebalance(bundle->bond, &backer->revalidate_set);
4336 return MIN(max_idle, 1000);
4339 /* Updates flow table statistics given that the datapath just reported 'stats'
4340 * as 'subfacet''s statistics. */
4342 update_subfacet_stats(struct subfacet *subfacet,
4343 const struct dpif_flow_stats *stats)
4345 struct facet *facet = subfacet->facet;
4346 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4347 struct dpif_flow_stats diff;
4349 diff.tcp_flags = stats->tcp_flags;
4350 diff.used = stats->used;
4352 if (stats->n_packets >= subfacet->dp_packet_count) {
4353 diff.n_packets = stats->n_packets - subfacet->dp_packet_count;
4355 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
4359 if (stats->n_bytes >= subfacet->dp_byte_count) {
4360 diff.n_bytes = stats->n_bytes - subfacet->dp_byte_count;
4362 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
4366 ofproto->n_hit += diff.n_packets;
4367 subfacet->dp_packet_count = stats->n_packets;
4368 subfacet->dp_byte_count = stats->n_bytes;
4369 subfacet_update_stats(subfacet, &diff);
4371 if (facet->accounted_bytes < facet->byte_count) {
4373 facet_account(facet);
4374 facet->accounted_bytes = facet->byte_count;
4378 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
4379 * about, or a flow that shouldn't be installed but was anyway. Delete it. */
4381 delete_unexpected_flow(struct ofproto_dpif *ofproto,
4382 const struct nlattr *key, size_t key_len)
4384 if (!VLOG_DROP_WARN(&rl)) {
4388 odp_flow_key_format(key, key_len, &s);
4389 VLOG_WARN("unexpected flow on %s: %s", ofproto->up.name, ds_cstr(&s));
4393 COVERAGE_INC(facet_unexpected);
4394 dpif_flow_del(ofproto->backer->dpif, key, key_len, NULL);
4397 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
4399 * This function also pushes statistics updates to rules which each facet
4400 * resubmits into. Generally these statistics will be accurate. However, if a
4401 * facet changes the rule it resubmits into at some time in between
4402 * update_stats() runs, it is possible that statistics accrued to the
4403 * old rule will be incorrectly attributed to the new rule. This could be
4404 * avoided by calling update_stats() whenever rules are created or
4405 * deleted. However, the performance impact of making so many calls to the
4406 * datapath do not justify the benefit of having perfectly accurate statistics.
4408 * In addition, this function maintains per ofproto flow hit counts. The patch
4409 * port is not treated specially. e.g. A packet ingress from br0 patched into
4410 * br1 will increase the hit count of br0 by 1, however, does not affect
4411 * the hit or miss counts of br1.
4414 update_stats(struct dpif_backer *backer)
4416 const struct dpif_flow_stats *stats;
4417 struct dpif_flow_dump dump;
4418 const struct nlattr *key;
4419 struct ofproto_dpif *ofproto;
4422 dpif_flow_dump_start(&dump, backer->dpif);
4423 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
4425 struct subfacet *subfacet;
4428 if (ofproto_receive(backer, NULL, key, key_len, &flow, NULL, &ofproto,
4433 ofproto->total_subfacet_count += hmap_count(&ofproto->subfacets);
4434 ofproto->n_update_stats++;
4436 key_hash = odp_flow_key_hash(key, key_len);
4437 subfacet = subfacet_find(ofproto, key, key_len, key_hash);
4438 switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
4440 update_subfacet_stats(subfacet, stats);
4444 /* Stats are updated per-packet. */
4447 case SF_NOT_INSTALLED:
4449 delete_unexpected_flow(ofproto, key, key_len);
4454 dpif_flow_dump_done(&dump);
4456 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4457 update_moving_averages(ofproto);
4462 /* Calculates and returns the number of milliseconds of idle time after which
4463 * subfacets should expire from the datapath. When a subfacet expires, we fold
4464 * its statistics into its facet, and when a facet's last subfacet expires, we
4465 * fold its statistic into its rule. */
4467 subfacet_max_idle(const struct ofproto_dpif *ofproto)
4470 * Idle time histogram.
4472 * Most of the time a switch has a relatively small number of subfacets.
4473 * When this is the case we might as well keep statistics for all of them
4474 * in userspace and to cache them in the kernel datapath for performance as
4477 * As the number of subfacets increases, the memory required to maintain
4478 * statistics about them in userspace and in the kernel becomes
4479 * significant. However, with a large number of subfacets it is likely
4480 * that only a few of them are "heavy hitters" that consume a large amount
4481 * of bandwidth. At this point, only heavy hitters are worth caching in
4482 * the kernel and maintaining in userspaces; other subfacets we can
4485 * The technique used to compute the idle time is to build a histogram with
4486 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
4487 * that is installed in the kernel gets dropped in the appropriate bucket.
4488 * After the histogram has been built, we compute the cutoff so that only
4489 * the most-recently-used 1% of subfacets (but at least
4490 * ofproto->up.flow_eviction_threshold flows) are kept cached. At least
4491 * the most-recently-used bucket of subfacets is kept, so actually an
4492 * arbitrary number of subfacets can be kept in any given expiration run
4493 * (though the next run will delete most of those unless they receive
4496 * This requires a second pass through the subfacets, in addition to the
4497 * pass made by update_stats(), because the former function never looks at
4498 * uninstallable subfacets.
4500 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
4501 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
4502 int buckets[N_BUCKETS] = { 0 };
4503 int total, subtotal, bucket;
4504 struct subfacet *subfacet;
4508 total = hmap_count(&ofproto->subfacets);
4509 if (total <= ofproto->up.flow_eviction_threshold) {
4510 return N_BUCKETS * BUCKET_WIDTH;
4513 /* Build histogram. */
4515 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
4516 long long int idle = now - subfacet->used;
4517 int bucket = (idle <= 0 ? 0
4518 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
4519 : (unsigned int) idle / BUCKET_WIDTH);
4523 /* Find the first bucket whose flows should be expired. */
4524 subtotal = bucket = 0;
4526 subtotal += buckets[bucket++];
4527 } while (bucket < N_BUCKETS &&
4528 subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100));
4530 if (VLOG_IS_DBG_ENABLED()) {
4534 ds_put_cstr(&s, "keep");
4535 for (i = 0; i < N_BUCKETS; i++) {
4537 ds_put_cstr(&s, ", drop");
4540 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
4543 VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s));
4547 return bucket * BUCKET_WIDTH;
4551 expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle)
4553 /* Cutoff time for most flows. */
4554 long long int normal_cutoff = time_msec() - dp_max_idle;
4556 /* We really want to keep flows for special protocols around, so use a more
4557 * conservative cutoff. */
4558 long long int special_cutoff = time_msec() - 10000;
4560 struct subfacet *subfacet, *next_subfacet;
4561 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
4565 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
4566 &ofproto->subfacets) {
4567 long long int cutoff;
4569 cutoff = (subfacet->facet->xout.slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP
4573 if (subfacet->used < cutoff) {
4574 if (subfacet->path != SF_NOT_INSTALLED) {
4575 batch[n_batch++] = subfacet;
4576 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
4577 subfacet_destroy_batch(ofproto, batch, n_batch);
4581 subfacet_destroy(subfacet);
4587 subfacet_destroy_batch(ofproto, batch, n_batch);
4591 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
4592 * then delete it entirely. */
4594 rule_expire(struct rule_dpif *rule)
4596 struct facet *facet, *next_facet;
4600 if (rule->up.pending) {
4601 /* We'll have to expire it later. */
4605 /* Has 'rule' expired? */
4607 if (rule->up.hard_timeout
4608 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
4609 reason = OFPRR_HARD_TIMEOUT;
4610 } else if (rule->up.idle_timeout
4611 && now > rule->up.used + rule->up.idle_timeout * 1000) {
4612 reason = OFPRR_IDLE_TIMEOUT;
4617 COVERAGE_INC(ofproto_dpif_expired);
4619 /* Update stats. (This is a no-op if the rule expired due to an idle
4620 * timeout, because that only happens when the rule has no facets left.) */
4621 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4622 facet_remove(facet);
4625 /* Get rid of the rule. */
4626 ofproto_rule_expire(&rule->up, reason);
4631 /* Creates and returns a new facet based on 'miss'.
4633 * The caller must already have determined that no facet with an identical
4634 * 'miss->flow' exists in 'miss->ofproto'.
4636 * 'hash' must be the return value of flow_hash(miss->flow, 0).
4638 * The facet will initially have no subfacets. The caller should create (at
4639 * least) one subfacet with subfacet_create(). */
4640 static struct facet *
4641 facet_create(const struct flow_miss *miss, uint32_t hash)
4643 struct ofproto_dpif *ofproto = miss->ofproto;
4644 struct xlate_in xin;
4645 struct facet *facet;
4647 facet = xzalloc(sizeof *facet);
4648 facet->used = time_msec();
4649 facet->flow = miss->flow;
4650 facet->initial_vals = miss->initial_vals;
4651 facet->rule = rule_dpif_lookup(ofproto, &facet->flow);
4652 facet->learn_rl = time_msec() + 500;
4654 hmap_insert(&ofproto->facets, &facet->hmap_node, hash);
4655 list_push_back(&facet->rule->facets, &facet->list_node);
4656 list_init(&facet->subfacets);
4657 netflow_flow_init(&facet->nf_flow);
4658 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
4660 xlate_in_init(&xin, ofproto, &facet->flow, &facet->initial_vals,
4661 facet->rule, 0, NULL);
4662 xin.may_learn = true;
4663 xlate_actions(&xin, &facet->xout);
4664 facet->nf_flow.output_iface = facet->xout.nf_output_iface;
4670 facet_free(struct facet *facet)
4673 xlate_out_uninit(&facet->xout);
4678 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
4679 * 'packet', which arrived on 'in_port'. */
4681 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
4682 const struct nlattr *odp_actions, size_t actions_len,
4683 struct ofpbuf *packet)
4685 struct odputil_keybuf keybuf;
4689 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4690 odp_flow_key_from_flow(&key, flow,
4691 ofp_port_to_odp_port(ofproto, flow->in_port));
4693 error = dpif_execute(ofproto->backer->dpif, key.data, key.size,
4694 odp_actions, actions_len, packet);
4698 /* Remove 'facet' from 'ofproto' and free up the associated memory:
4700 * - If 'facet' was installed in the datapath, uninstalls it and updates its
4701 * rule's statistics, via subfacet_uninstall().
4703 * - Removes 'facet' from its rule and from ofproto->facets.
4706 facet_remove(struct facet *facet)
4708 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4709 struct subfacet *subfacet, *next_subfacet;
4711 ovs_assert(!list_is_empty(&facet->subfacets));
4713 /* First uninstall all of the subfacets to get final statistics. */
4714 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4715 subfacet_uninstall(subfacet);
4718 /* Flush the final stats to the rule.
4720 * This might require us to have at least one subfacet around so that we
4721 * can use its actions for accounting in facet_account(), which is why we
4722 * have uninstalled but not yet destroyed the subfacets. */
4723 facet_flush_stats(facet);
4725 /* Now we're really all done so destroy everything. */
4726 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
4727 &facet->subfacets) {
4728 subfacet_destroy__(subfacet);
4730 hmap_remove(&ofproto->facets, &facet->hmap_node);
4731 list_remove(&facet->list_node);
4735 /* Feed information from 'facet' back into the learning table to keep it in
4736 * sync with what is actually flowing through the datapath. */
4738 facet_learn(struct facet *facet)
4740 long long int now = time_msec();
4742 if (!facet->xout.has_fin_timeout && now < facet->learn_rl) {
4746 facet->learn_rl = now + 500;
4748 if (!facet->xout.has_learn
4749 && !facet->xout.has_normal
4750 && (!facet->xout.has_fin_timeout
4751 || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
4755 facet_push_stats(facet, true);
4759 facet_account(struct facet *facet)
4761 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4762 const struct nlattr *a;
4767 if (!facet->xout.has_normal || !ofproto->has_bonded_bundles) {
4770 n_bytes = facet->byte_count - facet->accounted_bytes;
4772 /* This loop feeds byte counters to bond_account() for rebalancing to use
4773 * as a basis. We also need to track the actual VLAN on which the packet
4774 * is going to be sent to ensure that it matches the one passed to
4775 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
4778 * We use the actions from an arbitrary subfacet because they should all
4779 * be equally valid for our purpose. */
4780 vlan_tci = facet->flow.vlan_tci;
4781 NL_ATTR_FOR_EACH_UNSAFE (a, left, facet->xout.odp_actions.data,
4782 facet->xout.odp_actions.size) {
4783 const struct ovs_action_push_vlan *vlan;
4784 struct ofport_dpif *port;
4786 switch (nl_attr_type(a)) {
4787 case OVS_ACTION_ATTR_OUTPUT:
4788 port = get_odp_port(ofproto, nl_attr_get_u32(a));
4789 if (port && port->bundle && port->bundle->bond) {
4790 bond_account(port->bundle->bond, &facet->flow,
4791 vlan_tci_to_vid(vlan_tci), n_bytes);
4795 case OVS_ACTION_ATTR_POP_VLAN:
4796 vlan_tci = htons(0);
4799 case OVS_ACTION_ATTR_PUSH_VLAN:
4800 vlan = nl_attr_get(a);
4801 vlan_tci = vlan->vlan_tci;
4807 /* Returns true if the only action for 'facet' is to send to the controller.
4808 * (We don't report NetFlow expiration messages for such facets because they
4809 * are just part of the control logic for the network, not real traffic). */
4811 facet_is_controller_flow(struct facet *facet)
4814 const struct rule *rule = &facet->rule->up;
4815 const struct ofpact *ofpacts = rule->ofpacts;
4816 size_t ofpacts_len = rule->ofpacts_len;
4818 if (ofpacts_len > 0 &&
4819 ofpacts->type == OFPACT_CONTROLLER &&
4820 ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len)) {
4827 /* Folds all of 'facet''s statistics into its rule. Also updates the
4828 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
4829 * 'facet''s statistics in the datapath should have been zeroed and folded into
4830 * its packet and byte counts before this function is called. */
4832 facet_flush_stats(struct facet *facet)
4834 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4835 struct subfacet *subfacet;
4837 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4838 ovs_assert(!subfacet->dp_byte_count);
4839 ovs_assert(!subfacet->dp_packet_count);
4842 facet_push_stats(facet, false);
4843 if (facet->accounted_bytes < facet->byte_count) {
4844 facet_account(facet);
4845 facet->accounted_bytes = facet->byte_count;
4848 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
4849 struct ofexpired expired;
4850 expired.flow = facet->flow;
4851 expired.packet_count = facet->packet_count;
4852 expired.byte_count = facet->byte_count;
4853 expired.used = facet->used;
4854 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4857 /* Reset counters to prevent double counting if 'facet' ever gets
4859 facet_reset_counters(facet);
4861 netflow_flow_clear(&facet->nf_flow);
4862 facet->tcp_flags = 0;
4865 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
4866 * Returns it if found, otherwise a null pointer.
4868 * 'hash' must be the return value of flow_hash(flow, 0).
4870 * The returned facet might need revalidation; use facet_lookup_valid()
4871 * instead if that is important. */
4872 static struct facet *
4873 facet_find(struct ofproto_dpif *ofproto,
4874 const struct flow *flow, uint32_t hash)
4876 struct facet *facet;
4878 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, hash, &ofproto->facets) {
4879 if (flow_equal(flow, &facet->flow)) {
4887 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
4888 * Returns it if found, otherwise a null pointer.
4890 * 'hash' must be the return value of flow_hash(flow, 0).
4892 * The returned facet is guaranteed to be valid. */
4893 static struct facet *
4894 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow,
4897 struct facet *facet;
4899 facet = facet_find(ofproto, flow, hash);
4901 && (ofproto->backer->need_revalidate
4902 || tag_set_intersects(&ofproto->backer->revalidate_set,
4904 && !facet_revalidate(facet)) {
4912 facet_check_consistency(struct facet *facet)
4914 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
4916 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4918 struct xlate_out xout;
4919 struct xlate_in xin;
4921 struct rule_dpif *rule;
4924 /* Check the rule for consistency. */
4925 rule = rule_dpif_lookup(ofproto, &facet->flow);
4926 if (rule != facet->rule) {
4927 if (!VLOG_DROP_WARN(&rl)) {
4928 struct ds s = DS_EMPTY_INITIALIZER;
4930 flow_format(&s, &facet->flow);
4931 ds_put_format(&s, ": facet associated with wrong rule (was "
4932 "table=%"PRIu8",", facet->rule->up.table_id);
4933 cls_rule_format(&facet->rule->up.cr, &s);
4934 ds_put_format(&s, ") (should have been table=%"PRIu8",",
4936 cls_rule_format(&rule->up.cr, &s);
4937 ds_put_cstr(&s, ")\n");
4944 /* Check the datapath actions for consistency. */
4945 xlate_in_init(&xin, ofproto, &facet->flow, &facet->initial_vals, rule,
4947 xlate_actions(&xin, &xout);
4949 ok = ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)
4950 && facet->xout.slow == xout.slow;
4951 if (!ok && !VLOG_DROP_WARN(&rl)) {
4952 struct ds s = DS_EMPTY_INITIALIZER;
4954 flow_format(&s, &facet->flow);
4955 ds_put_cstr(&s, ": inconsistency in facet");
4957 if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4958 ds_put_cstr(&s, " (actions were: ");
4959 format_odp_actions(&s, facet->xout.odp_actions.data,
4960 facet->xout.odp_actions.size);
4961 ds_put_cstr(&s, ") (correct actions: ");
4962 format_odp_actions(&s, xout.odp_actions.data,
4963 xout.odp_actions.size);
4964 ds_put_cstr(&s, ")");
4967 if (facet->xout.slow != xout.slow) {
4968 ds_put_format(&s, " slow path incorrect. should be %d", xout.slow);
4973 xlate_out_uninit(&xout);
4978 /* Re-searches the classifier for 'facet':
4980 * - If the rule found is different from 'facet''s current rule, moves
4981 * 'facet' to the new rule and recompiles its actions.
4983 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
4984 * where it is and recompiles its actions anyway.
4986 * - If any of 'facet''s subfacets correspond to a new flow according to
4987 * ofproto_receive(), 'facet' is removed.
4989 * Returns true if 'facet' is still valid. False if 'facet' was removed. */
4991 facet_revalidate(struct facet *facet)
4993 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4994 struct rule_dpif *new_rule;
4995 struct subfacet *subfacet;
4996 struct xlate_out xout;
4997 struct xlate_in xin;
4999 COVERAGE_INC(facet_revalidate);
5001 /* Check that child subfacets still correspond to this facet. Tunnel
5002 * configuration changes could cause a subfacet's OpenFlow in_port to
5004 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
5005 struct ofproto_dpif *recv_ofproto;
5006 struct flow recv_flow;
5009 error = ofproto_receive(ofproto->backer, NULL, subfacet->key,
5010 subfacet->key_len, &recv_flow, NULL,
5011 &recv_ofproto, NULL, NULL);
5013 || recv_ofproto != ofproto
5014 || memcmp(&recv_flow, &facet->flow, sizeof recv_flow)) {
5015 facet_remove(facet);
5020 new_rule = rule_dpif_lookup(ofproto, &facet->flow);
5022 /* Calculate new datapath actions.
5024 * We do not modify any 'facet' state yet, because we might need to, e.g.,
5025 * emit a NetFlow expiration and, if so, we need to have the old state
5026 * around to properly compose it. */
5027 xlate_in_init(&xin, ofproto, &facet->flow, &facet->initial_vals, new_rule,
5029 xlate_actions(&xin, &xout);
5031 /* A facet's slow path reason should only change under dramatic
5032 * circumstances. Rather than try to update everything, it's simpler to
5033 * remove the facet and start over. */
5034 if (facet->xout.slow != xout.slow) {
5035 facet_remove(facet);
5036 xlate_out_uninit(&xout);
5040 if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
5041 LIST_FOR_EACH(subfacet, list_node, &facet->subfacets) {
5042 if (subfacet->path == SF_FAST_PATH) {
5043 struct dpif_flow_stats stats;
5045 subfacet_install(subfacet, &xout.odp_actions, &stats);
5046 subfacet_update_stats(subfacet, &stats);
5050 facet_flush_stats(facet);
5052 ofpbuf_clear(&facet->xout.odp_actions);
5053 ofpbuf_put(&facet->xout.odp_actions, xout.odp_actions.data,
5054 xout.odp_actions.size);
5057 /* Update 'facet' now that we've taken care of all the old state. */
5058 facet->xout.tags = xout.tags;
5059 facet->xout.slow = xout.slow;
5060 facet->xout.has_learn = xout.has_learn;
5061 facet->xout.has_normal = xout.has_normal;
5062 facet->xout.has_fin_timeout = xout.has_fin_timeout;
5063 facet->xout.nf_output_iface = xout.nf_output_iface;
5064 facet->xout.mirrors = xout.mirrors;
5065 facet->nf_flow.output_iface = facet->xout.nf_output_iface;
5067 if (facet->rule != new_rule) {
5068 COVERAGE_INC(facet_changed_rule);
5069 list_remove(&facet->list_node);
5070 list_push_back(&new_rule->facets, &facet->list_node);
5071 facet->rule = new_rule;
5072 facet->used = new_rule->up.created;
5073 facet->prev_used = facet->used;
5076 xlate_out_uninit(&xout);
5081 facet_reset_counters(struct facet *facet)
5083 facet->packet_count = 0;
5084 facet->byte_count = 0;
5085 facet->prev_packet_count = 0;
5086 facet->prev_byte_count = 0;
5087 facet->accounted_bytes = 0;
5091 facet_push_stats(struct facet *facet, bool may_learn)
5093 struct dpif_flow_stats stats;
5095 ovs_assert(facet->packet_count >= facet->prev_packet_count);
5096 ovs_assert(facet->byte_count >= facet->prev_byte_count);
5097 ovs_assert(facet->used >= facet->prev_used);
5099 stats.n_packets = facet->packet_count - facet->prev_packet_count;
5100 stats.n_bytes = facet->byte_count - facet->prev_byte_count;
5101 stats.used = facet->used;
5102 stats.tcp_flags = facet->tcp_flags;
5104 if (may_learn || stats.n_packets || facet->used > facet->prev_used) {
5105 struct ofproto_dpif *ofproto =
5106 ofproto_dpif_cast(facet->rule->up.ofproto);
5108 struct ofport_dpif *in_port;
5109 struct xlate_in xin;
5111 facet->prev_packet_count = facet->packet_count;
5112 facet->prev_byte_count = facet->byte_count;
5113 facet->prev_used = facet->used;
5115 in_port = get_ofp_port(ofproto, facet->flow.in_port);
5116 if (in_port && in_port->tnl_port) {
5117 netdev_vport_inc_rx(in_port->up.netdev, &stats);
5120 rule_credit_stats(facet->rule, &stats);
5121 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow,
5123 netflow_flow_update_flags(&facet->nf_flow, facet->tcp_flags);
5124 update_mirror_stats(ofproto, facet->xout.mirrors, stats.n_packets,
5127 xlate_in_init(&xin, ofproto, &facet->flow, &facet->initial_vals,
5128 facet->rule, stats.tcp_flags, NULL);
5129 xin.resubmit_stats = &stats;
5130 xin.may_learn = may_learn;
5131 xlate_actions_for_side_effects(&xin);
5136 push_all_stats__(bool run_fast)
5138 static long long int rl = LLONG_MIN;
5139 struct ofproto_dpif *ofproto;
5141 if (time_msec() < rl) {
5145 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5146 struct facet *facet;
5148 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
5149 facet_push_stats(facet, false);
5156 rl = time_msec() + 100;
5160 push_all_stats(void)
5162 push_all_stats__(true);
5166 rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats)
5168 rule->packet_count += stats->n_packets;
5169 rule->byte_count += stats->n_bytes;
5170 ofproto_rule_update_used(&rule->up, stats->used);
5175 static struct subfacet *
5176 subfacet_find(struct ofproto_dpif *ofproto,
5177 const struct nlattr *key, size_t key_len, uint32_t key_hash)
5179 struct subfacet *subfacet;
5181 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
5182 &ofproto->subfacets) {
5183 if (subfacet->key_len == key_len
5184 && !memcmp(key, subfacet->key, key_len)) {
5192 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
5193 * 'key_fitness', 'key', and 'key_len' members in 'miss'. Returns the
5194 * existing subfacet if there is one, otherwise creates and returns a
5196 static struct subfacet *
5197 subfacet_create(struct facet *facet, struct flow_miss *miss,
5200 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5201 enum odp_key_fitness key_fitness = miss->key_fitness;
5202 const struct nlattr *key = miss->key;
5203 size_t key_len = miss->key_len;
5205 struct subfacet *subfacet;
5207 key_hash = odp_flow_key_hash(key, key_len);
5209 if (list_is_empty(&facet->subfacets)) {
5210 subfacet = &facet->one_subfacet;
5212 subfacet = subfacet_find(ofproto, key, key_len, key_hash);
5214 if (subfacet->facet == facet) {
5218 /* This shouldn't happen. */
5219 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
5220 subfacet_destroy(subfacet);
5223 subfacet = xmalloc(sizeof *subfacet);
5226 hmap_insert(&ofproto->subfacets, &subfacet->hmap_node, key_hash);
5227 list_push_back(&facet->subfacets, &subfacet->list_node);
5228 subfacet->facet = facet;
5229 subfacet->key_fitness = key_fitness;
5230 subfacet->key = xmemdup(key, key_len);
5231 subfacet->key_len = key_len;
5232 subfacet->used = now;
5233 subfacet->created = now;
5234 subfacet->dp_packet_count = 0;
5235 subfacet->dp_byte_count = 0;
5236 subfacet->path = SF_NOT_INSTALLED;
5238 ofproto->subfacet_add_count++;
5242 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
5243 * its facet within 'ofproto', and frees it. */
5245 subfacet_destroy__(struct subfacet *subfacet)
5247 struct facet *facet = subfacet->facet;
5248 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5250 /* Update ofproto stats before uninstall the subfacet. */
5251 ofproto->subfacet_del_count++;
5252 ofproto->total_subfacet_life_span += (time_msec() - subfacet->created);
5254 subfacet_uninstall(subfacet);
5255 hmap_remove(&ofproto->subfacets, &subfacet->hmap_node);
5256 list_remove(&subfacet->list_node);
5257 free(subfacet->key);
5258 if (subfacet != &facet->one_subfacet) {
5263 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
5264 * last remaining subfacet in its facet destroys the facet too. */
5266 subfacet_destroy(struct subfacet *subfacet)
5268 struct facet *facet = subfacet->facet;
5270 if (list_is_singleton(&facet->subfacets)) {
5271 /* facet_remove() needs at least one subfacet (it will remove it). */
5272 facet_remove(facet);
5274 subfacet_destroy__(subfacet);
5279 subfacet_destroy_batch(struct ofproto_dpif *ofproto,
5280 struct subfacet **subfacets, int n)
5282 struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH];
5283 struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH];
5284 struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH];
5287 for (i = 0; i < n; i++) {
5288 ops[i].type = DPIF_OP_FLOW_DEL;
5289 ops[i].u.flow_del.key = subfacets[i]->key;
5290 ops[i].u.flow_del.key_len = subfacets[i]->key_len;
5291 ops[i].u.flow_del.stats = &stats[i];
5295 dpif_operate(ofproto->backer->dpif, opsp, n);
5296 for (i = 0; i < n; i++) {
5297 subfacet_reset_dp_stats(subfacets[i], &stats[i]);
5298 subfacets[i]->path = SF_NOT_INSTALLED;
5299 subfacet_destroy(subfacets[i]);
5304 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
5305 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
5306 * in the datapath will be zeroed and 'stats' will be updated with traffic new
5307 * since 'subfacet' was last updated.
5309 * Returns 0 if successful, otherwise a positive errno value. */
5311 subfacet_install(struct subfacet *subfacet, const struct ofpbuf *odp_actions,
5312 struct dpif_flow_stats *stats)
5314 struct facet *facet = subfacet->facet;
5315 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5316 enum subfacet_path path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
5317 const struct nlattr *actions = odp_actions->data;
5318 size_t actions_len = odp_actions->size;
5320 uint64_t slow_path_stub[128 / 8];
5321 enum dpif_flow_put_flags flags;
5324 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
5326 flags |= DPIF_FP_ZERO_STATS;
5329 if (path == SF_SLOW_PATH) {
5330 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
5331 slow_path_stub, sizeof slow_path_stub,
5332 &actions, &actions_len);
5335 ret = dpif_flow_put(ofproto->backer->dpif, flags, subfacet->key,
5336 subfacet->key_len, actions, actions_len, stats);
5339 subfacet_reset_dp_stats(subfacet, stats);
5343 subfacet->path = path;
5348 /* If 'subfacet' is installed in the datapath, uninstalls it. */
5350 subfacet_uninstall(struct subfacet *subfacet)
5352 if (subfacet->path != SF_NOT_INSTALLED) {
5353 struct rule_dpif *rule = subfacet->facet->rule;
5354 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5355 struct dpif_flow_stats stats;
5358 error = dpif_flow_del(ofproto->backer->dpif, subfacet->key,
5359 subfacet->key_len, &stats);
5360 subfacet_reset_dp_stats(subfacet, &stats);
5362 subfacet_update_stats(subfacet, &stats);
5364 subfacet->path = SF_NOT_INSTALLED;
5366 ovs_assert(subfacet->dp_packet_count == 0);
5367 ovs_assert(subfacet->dp_byte_count == 0);
5371 /* Resets 'subfacet''s datapath statistics counters. This should be called
5372 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
5373 * non-null, it should contain the statistics returned by dpif when 'subfacet'
5374 * was reset in the datapath. 'stats' will be modified to include only
5375 * statistics new since 'subfacet' was last updated. */
5377 subfacet_reset_dp_stats(struct subfacet *subfacet,
5378 struct dpif_flow_stats *stats)
5381 && subfacet->dp_packet_count <= stats->n_packets
5382 && subfacet->dp_byte_count <= stats->n_bytes) {
5383 stats->n_packets -= subfacet->dp_packet_count;
5384 stats->n_bytes -= subfacet->dp_byte_count;
5387 subfacet->dp_packet_count = 0;
5388 subfacet->dp_byte_count = 0;
5391 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
5393 * Because of the meaning of a subfacet's counters, it only makes sense to do
5394 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
5395 * represents a packet that was sent by hand or if it represents statistics
5396 * that have been cleared out of the datapath. */
5398 subfacet_update_stats(struct subfacet *subfacet,
5399 const struct dpif_flow_stats *stats)
5401 if (stats->n_packets || stats->used > subfacet->used) {
5402 struct facet *facet = subfacet->facet;
5404 subfacet->used = MAX(subfacet->used, stats->used);
5405 facet->used = MAX(facet->used, stats->used);
5406 facet->packet_count += stats->n_packets;
5407 facet->byte_count += stats->n_bytes;
5408 facet->tcp_flags |= stats->tcp_flags;
5414 static struct rule_dpif *
5415 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow)
5417 struct rule_dpif *rule;
5419 rule = rule_dpif_lookup__(ofproto, flow, 0);
5424 return rule_dpif_miss_rule(ofproto, flow);
5427 static struct rule_dpif *
5428 rule_dpif_lookup__(struct ofproto_dpif *ofproto, const struct flow *flow,
5431 struct cls_rule *cls_rule;
5432 struct classifier *cls;
5434 if (table_id >= N_TABLES) {
5438 cls = &ofproto->up.tables[table_id].cls;
5439 if (flow->nw_frag & FLOW_NW_FRAG_ANY
5440 && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
5441 /* For OFPC_NORMAL frag_handling, we must pretend that transport ports
5442 * are unavailable. */
5443 struct flow ofpc_normal_flow = *flow;
5444 ofpc_normal_flow.tp_src = htons(0);
5445 ofpc_normal_flow.tp_dst = htons(0);
5446 cls_rule = classifier_lookup(cls, &ofpc_normal_flow);
5448 cls_rule = classifier_lookup(cls, flow);
5450 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
5453 static struct rule_dpif *
5454 rule_dpif_miss_rule(struct ofproto_dpif *ofproto, const struct flow *flow)
5456 struct ofport_dpif *port;
5458 port = get_ofp_port(ofproto, flow->in_port);
5460 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, flow->in_port);
5461 return ofproto->miss_rule;
5464 if (port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN) {
5465 return ofproto->no_packet_in_rule;
5467 return ofproto->miss_rule;
5471 complete_operation(struct rule_dpif *rule)
5473 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5475 rule_invalidate(rule);
5477 struct dpif_completion *c = xmalloc(sizeof *c);
5478 c->op = rule->up.pending;
5479 list_push_back(&ofproto->completions, &c->list_node);
5481 ofoperation_complete(rule->up.pending, 0);
5485 static struct rule *
5488 struct rule_dpif *rule = xmalloc(sizeof *rule);
5493 rule_dealloc(struct rule *rule_)
5495 struct rule_dpif *rule = rule_dpif_cast(rule_);
5500 rule_construct(struct rule *rule_)
5502 struct rule_dpif *rule = rule_dpif_cast(rule_);
5503 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5504 struct rule_dpif *victim;
5507 rule->packet_count = 0;
5508 rule->byte_count = 0;
5510 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
5511 if (victim && !list_is_empty(&victim->facets)) {
5512 struct facet *facet;
5514 rule->facets = victim->facets;
5515 list_moved(&rule->facets);
5516 LIST_FOR_EACH (facet, list_node, &rule->facets) {
5517 /* XXX: We're only clearing our local counters here. It's possible
5518 * that quite a few packets are unaccounted for in the datapath
5519 * statistics. These will be accounted to the new rule instead of
5520 * cleared as required. This could be fixed by clearing out the
5521 * datapath statistics for this facet, but currently it doesn't
5523 facet_reset_counters(facet);
5527 /* Must avoid list_moved() in this case. */
5528 list_init(&rule->facets);
5531 table_id = rule->up.table_id;
5533 rule->tag = victim->tag;
5534 } else if (table_id == 0) {
5539 miniflow_expand(&rule->up.cr.match.flow, &flow);
5540 rule->tag = rule_calculate_tag(&flow, &rule->up.cr.match.mask,
5541 ofproto->tables[table_id].basis);
5544 complete_operation(rule);
5549 rule_destruct(struct rule *rule_)
5551 struct rule_dpif *rule = rule_dpif_cast(rule_);
5552 struct facet *facet, *next_facet;
5554 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
5555 facet_revalidate(facet);
5558 complete_operation(rule);
5562 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
5564 struct rule_dpif *rule = rule_dpif_cast(rule_);
5566 /* push_all_stats() can handle flow misses which, when using the learn
5567 * action, can cause rules to be added and deleted. This can corrupt our
5568 * caller's datastructures which assume that rule_get_stats() doesn't have
5569 * an impact on the flow table. To be safe, we disable miss handling. */
5570 push_all_stats__(false);
5572 /* Start from historical data for 'rule' itself that are no longer tracked
5573 * in facets. This counts, for example, facets that have expired. */
5574 *packets = rule->packet_count;
5575 *bytes = rule->byte_count;
5579 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
5580 struct ofpbuf *packet)
5582 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5583 struct initial_vals initial_vals;
5584 struct dpif_flow_stats stats;
5585 struct xlate_out xout;
5586 struct xlate_in xin;
5588 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
5589 rule_credit_stats(rule, &stats);
5591 initial_vals.vlan_tci = flow->vlan_tci;
5592 xlate_in_init(&xin, ofproto, flow, &initial_vals, rule, stats.tcp_flags,
5594 xin.resubmit_stats = &stats;
5595 xlate_actions(&xin, &xout);
5597 execute_odp_actions(ofproto, flow, xout.odp_actions.data,
5598 xout.odp_actions.size, packet);
5600 xlate_out_uninit(&xout);
5604 rule_execute(struct rule *rule, const struct flow *flow,
5605 struct ofpbuf *packet)
5607 rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
5608 ofpbuf_delete(packet);
5613 rule_modify_actions(struct rule *rule_)
5615 struct rule_dpif *rule = rule_dpif_cast(rule_);
5617 complete_operation(rule);
5620 /* Sends 'packet' out 'ofport'.
5621 * May modify 'packet'.
5622 * Returns 0 if successful, otherwise a positive errno value. */
5624 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
5626 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
5627 uint64_t odp_actions_stub[1024 / 8];
5628 struct ofpbuf key, odp_actions;
5629 struct dpif_flow_stats stats;
5630 struct odputil_keybuf keybuf;
5631 struct ofpact_output output;
5632 struct xlate_out xout;
5633 struct xlate_in xin;
5637 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5638 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5640 /* Use OFPP_NONE as the in_port to avoid special packet processing. */
5641 flow_extract(packet, 0, 0, NULL, OFPP_NONE, &flow);
5642 odp_flow_key_from_flow(&key, &flow, ofp_port_to_odp_port(ofproto,
5644 dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
5646 ofpact_init(&output.ofpact, OFPACT_OUTPUT, sizeof output);
5647 output.port = ofport->up.ofp_port;
5650 xlate_in_init(&xin, ofproto, &flow, NULL, NULL, 0, packet);
5651 xin.ofpacts_len = sizeof output;
5652 xin.ofpacts = &output.ofpact;
5653 xin.resubmit_stats = &stats;
5654 xlate_actions(&xin, &xout);
5656 error = dpif_execute(ofproto->backer->dpif,
5658 xout.odp_actions.data, xout.odp_actions.size,
5660 xlate_out_uninit(&xout);
5663 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %s (%s)",
5664 ofproto->up.name, netdev_get_name(ofport->up.netdev),
5668 ofproto->stats.tx_packets++;
5669 ofproto->stats.tx_bytes += packet->size;
5673 /* OpenFlow to datapath action translation. */
5675 static bool may_receive(const struct ofport_dpif *, struct xlate_ctx *);
5676 static void do_xlate_actions(const struct ofpact *, size_t ofpacts_len,
5677 struct xlate_ctx *);
5678 static void xlate_normal(struct xlate_ctx *);
5680 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
5681 * The action will state 'slow' as the reason that the action is in the slow
5682 * path. (This is purely informational: it allows a human viewing "ovs-dpctl
5683 * dump-flows" output to see why a flow is in the slow path.)
5685 * The 'stub_size' bytes in 'stub' will be used to store the action.
5686 * 'stub_size' must be large enough for the action.
5688 * The action and its size will be stored in '*actionsp' and '*actions_lenp',
5691 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
5692 enum slow_path_reason slow,
5693 uint64_t *stub, size_t stub_size,
5694 const struct nlattr **actionsp, size_t *actions_lenp)
5696 union user_action_cookie cookie;
5699 cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
5700 cookie.slow_path.unused = 0;
5701 cookie.slow_path.reason = slow;
5703 ofpbuf_use_stack(&buf, stub, stub_size);
5704 if (slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP | SLOW_STP)) {
5705 uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif, UINT32_MAX);
5706 odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
5708 put_userspace_action(ofproto, &buf, flow, &cookie,
5709 sizeof cookie.slow_path);
5711 *actionsp = buf.data;
5712 *actions_lenp = buf.size;
5716 put_userspace_action(const struct ofproto_dpif *ofproto,
5717 struct ofpbuf *odp_actions,
5718 const struct flow *flow,
5719 const union user_action_cookie *cookie,
5720 const size_t cookie_size)
5724 pid = dpif_port_get_pid(ofproto->backer->dpif,
5725 ofp_port_to_odp_port(ofproto, flow->in_port));
5727 return odp_put_userspace_action(pid, cookie, cookie_size, odp_actions);
5730 /* Compose SAMPLE action for sFlow or IPFIX. The given probability is
5731 * the number of packets out of UINT32_MAX to sample. The given
5732 * cookie is passed back in the callback for each sampled packet.
5735 compose_sample_action(const struct ofproto_dpif *ofproto,
5736 struct ofpbuf *odp_actions,
5737 const struct flow *flow,
5738 const uint32_t probability,
5739 const union user_action_cookie *cookie,
5740 const size_t cookie_size)
5742 size_t sample_offset, actions_offset;
5745 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
5747 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
5749 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
5750 cookie_offset = put_userspace_action(ofproto, odp_actions, flow, cookie,
5753 nl_msg_end_nested(odp_actions, actions_offset);
5754 nl_msg_end_nested(odp_actions, sample_offset);
5755 return cookie_offset;
5759 compose_sflow_cookie(const struct ofproto_dpif *ofproto,
5760 ovs_be16 vlan_tci, uint32_t odp_port,
5761 unsigned int n_outputs, union user_action_cookie *cookie)
5765 cookie->type = USER_ACTION_COOKIE_SFLOW;
5766 cookie->sflow.vlan_tci = vlan_tci;
5768 /* See http://www.sflow.org/sflow_version_5.txt (search for "Input/output
5769 * port information") for the interpretation of cookie->output. */
5770 switch (n_outputs) {
5772 /* 0x40000000 | 256 means "packet dropped for unknown reason". */
5773 cookie->sflow.output = 0x40000000 | 256;
5777 ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port);
5779 cookie->sflow.output = ifindex;
5784 /* 0x80000000 means "multiple output ports. */
5785 cookie->sflow.output = 0x80000000 | n_outputs;
5790 /* Compose SAMPLE action for sFlow bridge sampling. */
5792 compose_sflow_action(const struct ofproto_dpif *ofproto,
5793 struct ofpbuf *odp_actions,
5794 const struct flow *flow,
5797 uint32_t probability;
5798 union user_action_cookie cookie;
5800 if (!ofproto->sflow || flow->in_port == OFPP_NONE) {
5804 probability = dpif_sflow_get_probability(ofproto->sflow);
5805 compose_sflow_cookie(ofproto, htons(0), odp_port,
5806 odp_port == OVSP_NONE ? 0 : 1, &cookie);
5808 return compose_sample_action(ofproto, odp_actions, flow, probability,
5809 &cookie, sizeof cookie.sflow);
5813 compose_flow_sample_cookie(uint16_t probability, uint32_t collector_set_id,
5814 uint32_t obs_domain_id, uint32_t obs_point_id,
5815 union user_action_cookie *cookie)
5817 cookie->type = USER_ACTION_COOKIE_FLOW_SAMPLE;
5818 cookie->flow_sample.probability = probability;
5819 cookie->flow_sample.collector_set_id = collector_set_id;
5820 cookie->flow_sample.obs_domain_id = obs_domain_id;
5821 cookie->flow_sample.obs_point_id = obs_point_id;
5825 compose_ipfix_cookie(union user_action_cookie *cookie)
5827 cookie->type = USER_ACTION_COOKIE_IPFIX;
5830 /* Compose SAMPLE action for IPFIX bridge sampling. */
5832 compose_ipfix_action(const struct ofproto_dpif *ofproto,
5833 struct ofpbuf *odp_actions,
5834 const struct flow *flow)
5836 uint32_t probability;
5837 union user_action_cookie cookie;
5839 if (!ofproto->ipfix || flow->in_port == OFPP_NONE) {
5843 probability = dpif_ipfix_get_bridge_exporter_probability(ofproto->ipfix);
5844 compose_ipfix_cookie(&cookie);
5846 compose_sample_action(ofproto, odp_actions, flow, probability,
5847 &cookie, sizeof cookie.ipfix);
5850 /* SAMPLE action for sFlow must be first action in any given list of
5851 * actions. At this point we do not have all information required to
5852 * build it. So try to build sample action as complete as possible. */
5854 add_sflow_action(struct xlate_ctx *ctx)
5856 ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto,
5857 &ctx->xout->odp_actions,
5858 &ctx->xin->flow, OVSP_NONE);
5859 ctx->sflow_odp_port = 0;
5860 ctx->sflow_n_outputs = 0;
5863 /* SAMPLE action for IPFIX must be 1st or 2nd action in any given list
5864 * of actions, eventually after the SAMPLE action for sFlow. */
5866 add_ipfix_action(struct xlate_ctx *ctx)
5868 compose_ipfix_action(ctx->ofproto, &ctx->xout->odp_actions,
5872 /* Fix SAMPLE action according to data collected while composing ODP actions.
5873 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
5874 * USERSPACE action's user-cookie which is required for sflow. */
5876 fix_sflow_action(struct xlate_ctx *ctx)
5878 const struct flow *base = &ctx->base_flow;
5879 union user_action_cookie *cookie;
5881 if (!ctx->user_cookie_offset) {
5885 cookie = ofpbuf_at(&ctx->xout->odp_actions, ctx->user_cookie_offset,
5886 sizeof cookie->sflow);
5887 ovs_assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
5889 compose_sflow_cookie(ctx->ofproto, base->vlan_tci,
5890 ctx->sflow_odp_port, ctx->sflow_n_outputs, cookie);
5894 compose_output_action__(struct xlate_ctx *ctx, uint16_t ofp_port,
5897 const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
5898 ovs_be16 flow_vlan_tci;
5899 uint32_t flow_skb_mark;
5900 uint8_t flow_nw_tos;
5901 struct priority_to_dscp *pdscp;
5902 uint32_t out_port, odp_port;
5904 /* If 'struct flow' gets additional metadata, we'll need to zero it out
5905 * before traversing a patch port. */
5906 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 20);
5909 xlate_report(ctx, "Nonexistent output port");
5911 } else if (ofport->up.pp.config & OFPUTIL_PC_NO_FWD) {
5912 xlate_report(ctx, "OFPPC_NO_FWD set, skipping output");
5914 } else if (check_stp && !stp_forward_in_state(ofport->stp_state)) {
5915 xlate_report(ctx, "STP not in forwarding state, skipping output");
5919 if (netdev_vport_is_patch(ofport->up.netdev)) {
5920 struct ofport_dpif *peer = ofport_get_peer(ofport);
5921 struct flow old_flow = ctx->xin->flow;
5922 const struct ofproto_dpif *peer_ofproto;
5923 enum slow_path_reason special;
5924 struct ofport_dpif *in_port;
5927 xlate_report(ctx, "Nonexistent patch port peer");
5931 peer_ofproto = ofproto_dpif_cast(peer->up.ofproto);
5932 if (peer_ofproto->backer != ctx->ofproto->backer) {
5933 xlate_report(ctx, "Patch port peer on a different datapath");
5937 ctx->ofproto = ofproto_dpif_cast(peer->up.ofproto);
5938 ctx->xin->flow.in_port = peer->up.ofp_port;
5939 ctx->xin->flow.metadata = htonll(0);
5940 memset(&ctx->xin->flow.tunnel, 0, sizeof ctx->xin->flow.tunnel);
5941 memset(ctx->xin->flow.regs, 0, sizeof ctx->xin->flow.regs);
5943 in_port = get_ofp_port(ctx->ofproto, ctx->xin->flow.in_port);
5944 special = process_special(ctx->ofproto, &ctx->xin->flow, in_port,
5947 ctx->xout->slow = special;
5948 } else if (!in_port || may_receive(in_port, ctx)) {
5949 if (!in_port || stp_forward_in_state(in_port->stp_state)) {
5950 xlate_table_action(ctx, ctx->xin->flow.in_port, 0, true);
5952 /* Forwarding is disabled by STP. Let OFPP_NORMAL and the
5953 * learning action look at the packet, then drop it. */
5954 struct flow old_base_flow = ctx->base_flow;
5955 size_t old_size = ctx->xout->odp_actions.size;
5956 xlate_table_action(ctx, ctx->xin->flow.in_port, 0, true);
5957 ctx->base_flow = old_base_flow;
5958 ctx->xout->odp_actions.size = old_size;
5962 ctx->xin->flow = old_flow;
5963 ctx->ofproto = ofproto_dpif_cast(ofport->up.ofproto);
5965 if (ctx->xin->resubmit_stats) {
5966 netdev_vport_inc_tx(ofport->up.netdev, ctx->xin->resubmit_stats);
5967 netdev_vport_inc_rx(peer->up.netdev, ctx->xin->resubmit_stats);
5973 flow_vlan_tci = ctx->xin->flow.vlan_tci;
5974 flow_skb_mark = ctx->xin->flow.skb_mark;
5975 flow_nw_tos = ctx->xin->flow.nw_tos;
5977 pdscp = get_priority(ofport, ctx->xin->flow.skb_priority);
5979 ctx->xin->flow.nw_tos &= ~IP_DSCP_MASK;
5980 ctx->xin->flow.nw_tos |= pdscp->dscp;
5983 if (ofport->tnl_port) {
5984 /* Save tunnel metadata so that changes made due to
5985 * the Logical (tunnel) Port are not visible for any further
5986 * matches, while explicit set actions on tunnel metadata are.
5988 struct flow_tnl flow_tnl = ctx->xin->flow.tunnel;
5989 odp_port = tnl_port_send(ofport->tnl_port, &ctx->xin->flow);
5990 if (odp_port == OVSP_NONE) {
5991 xlate_report(ctx, "Tunneling decided against output");
5992 goto out; /* restore flow_nw_tos */
5994 if (ctx->xin->flow.tunnel.ip_dst == ctx->orig_tunnel_ip_dst) {
5995 xlate_report(ctx, "Not tunneling to our own address");
5996 goto out; /* restore flow_nw_tos */
5998 if (ctx->xin->resubmit_stats) {
5999 netdev_vport_inc_tx(ofport->up.netdev, ctx->xin->resubmit_stats);
6001 out_port = odp_port;
6002 commit_odp_tunnel_action(&ctx->xin->flow, &ctx->base_flow,
6003 &ctx->xout->odp_actions);
6004 ctx->xin->flow.tunnel = flow_tnl; /* Restore tunnel metadata */
6006 uint16_t vlandev_port;
6007 odp_port = ofport->odp_port;
6008 vlandev_port = vsp_realdev_to_vlandev(ctx->ofproto, ofp_port,
6009 ctx->xin->flow.vlan_tci);
6010 if (vlandev_port == ofp_port) {
6011 out_port = odp_port;
6013 out_port = ofp_port_to_odp_port(ctx->ofproto, vlandev_port);
6014 ctx->xin->flow.vlan_tci = htons(0);
6016 ctx->xin->flow.skb_mark &= ~IPSEC_MARK;
6018 commit_odp_actions(&ctx->xin->flow, &ctx->base_flow,
6019 &ctx->xout->odp_actions);
6020 nl_msg_put_u32(&ctx->xout->odp_actions, OVS_ACTION_ATTR_OUTPUT, out_port);
6022 ctx->sflow_odp_port = odp_port;
6023 ctx->sflow_n_outputs++;
6024 ctx->xout->nf_output_iface = ofp_port;
6027 ctx->xin->flow.vlan_tci = flow_vlan_tci;
6028 ctx->xin->flow.skb_mark = flow_skb_mark;
6030 ctx->xin->flow.nw_tos = flow_nw_tos;
6034 compose_output_action(struct xlate_ctx *ctx, uint16_t ofp_port)
6036 compose_output_action__(ctx, ofp_port, true);
6040 tag_the_flow(struct xlate_ctx *ctx, struct rule_dpif *rule)
6042 struct ofproto_dpif *ofproto = ctx->ofproto;
6043 uint8_t table_id = ctx->table_id;
6045 if (table_id > 0 && table_id < N_TABLES) {
6046 struct table_dpif *table = &ofproto->tables[table_id];
6047 if (table->other_table) {
6048 ctx->xout->tags |= (rule && rule->tag
6050 : rule_calculate_tag(&ctx->xin->flow,
6051 &table->other_table->mask,
6057 /* Common rule processing in one place to avoid duplicating code. */
6058 static struct rule_dpif *
6059 ctx_rule_hooks(struct xlate_ctx *ctx, struct rule_dpif *rule,
6062 if (ctx->xin->resubmit_hook) {
6063 ctx->xin->resubmit_hook(ctx, rule);
6065 if (rule == NULL && may_packet_in) {
6067 * check if table configuration flags
6068 * OFPTC_TABLE_MISS_CONTROLLER, default.
6069 * OFPTC_TABLE_MISS_CONTINUE,
6070 * OFPTC_TABLE_MISS_DROP
6071 * When OF1.0, OFPTC_TABLE_MISS_CONTINUE is used. What to do?
6073 rule = rule_dpif_miss_rule(ctx->ofproto, &ctx->xin->flow);
6075 if (rule && ctx->xin->resubmit_stats) {
6076 rule_credit_stats(rule, ctx->xin->resubmit_stats);
6082 xlate_table_action(struct xlate_ctx *ctx,
6083 uint16_t in_port, uint8_t table_id, bool may_packet_in)
6085 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
6086 struct rule_dpif *rule;
6087 uint16_t old_in_port = ctx->xin->flow.in_port;
6088 uint8_t old_table_id = ctx->table_id;
6090 ctx->table_id = table_id;
6092 /* Look up a flow with 'in_port' as the input port. */
6093 ctx->xin->flow.in_port = in_port;
6094 rule = rule_dpif_lookup__(ctx->ofproto, &ctx->xin->flow, table_id);
6096 tag_the_flow(ctx, rule);
6098 /* Restore the original input port. Otherwise OFPP_NORMAL and
6099 * OFPP_IN_PORT will have surprising behavior. */
6100 ctx->xin->flow.in_port = old_in_port;
6102 rule = ctx_rule_hooks(ctx, rule, may_packet_in);
6105 struct rule_dpif *old_rule = ctx->rule;
6109 do_xlate_actions(rule->up.ofpacts, rule->up.ofpacts_len, ctx);
6110 ctx->rule = old_rule;
6114 ctx->table_id = old_table_id;
6116 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
6118 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
6119 MAX_RESUBMIT_RECURSION);
6120 ctx->max_resubmit_trigger = true;
6125 xlate_ofpact_resubmit(struct xlate_ctx *ctx,
6126 const struct ofpact_resubmit *resubmit)
6131 in_port = resubmit->in_port;
6132 if (in_port == OFPP_IN_PORT) {
6133 in_port = ctx->xin->flow.in_port;
6136 table_id = resubmit->table_id;
6137 if (table_id == 255) {
6138 table_id = ctx->table_id;
6141 xlate_table_action(ctx, in_port, table_id, false);
6145 flood_packets(struct xlate_ctx *ctx, bool all)
6147 struct ofport_dpif *ofport;
6149 HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) {
6150 uint16_t ofp_port = ofport->up.ofp_port;
6152 if (ofp_port == ctx->xin->flow.in_port) {
6157 compose_output_action__(ctx, ofp_port, false);
6158 } else if (!(ofport->up.pp.config & OFPUTIL_PC_NO_FLOOD)) {
6159 compose_output_action(ctx, ofp_port);
6163 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
6167 execute_controller_action(struct xlate_ctx *ctx, int len,
6168 enum ofp_packet_in_reason reason,
6169 uint16_t controller_id)
6171 struct ofputil_packet_in pin;
6172 struct ofpbuf *packet;
6174 ovs_assert(!ctx->xout->slow || ctx->xout->slow == SLOW_CONTROLLER);
6175 ctx->xout->slow = SLOW_CONTROLLER;
6176 if (!ctx->xin->packet) {
6180 packet = ofpbuf_clone(ctx->xin->packet);
6182 if (packet->l2 && packet->l3) {
6183 struct eth_header *eh;
6184 uint16_t mpls_depth;
6186 eth_pop_vlan(packet);
6189 memcpy(eh->eth_src, ctx->xin->flow.dl_src, sizeof eh->eth_src);
6190 memcpy(eh->eth_dst, ctx->xin->flow.dl_dst, sizeof eh->eth_dst);
6192 if (ctx->xin->flow.vlan_tci & htons(VLAN_CFI)) {
6193 eth_push_vlan(packet, ctx->xin->flow.vlan_tci);
6196 mpls_depth = eth_mpls_depth(packet);
6198 if (mpls_depth < ctx->xin->flow.mpls_depth) {
6199 push_mpls(packet, ctx->xin->flow.dl_type, ctx->xin->flow.mpls_lse);
6200 } else if (mpls_depth > ctx->xin->flow.mpls_depth) {
6201 pop_mpls(packet, ctx->xin->flow.dl_type);
6202 } else if (mpls_depth) {
6203 set_mpls_lse(packet, ctx->xin->flow.mpls_lse);
6207 if (ctx->xin->flow.dl_type == htons(ETH_TYPE_IP)) {
6208 packet_set_ipv4(packet, ctx->xin->flow.nw_src,
6209 ctx->xin->flow.nw_dst, ctx->xin->flow.nw_tos,
6210 ctx->xin->flow.nw_ttl);
6214 if (ctx->xin->flow.nw_proto == IPPROTO_TCP) {
6215 packet_set_tcp_port(packet, ctx->xin->flow.tp_src,
6216 ctx->xin->flow.tp_dst);
6217 } else if (ctx->xin->flow.nw_proto == IPPROTO_UDP) {
6218 packet_set_udp_port(packet, ctx->xin->flow.tp_src,
6219 ctx->xin->flow.tp_dst);
6225 pin.packet = packet->data;
6226 pin.packet_len = packet->size;
6227 pin.reason = reason;
6228 pin.controller_id = controller_id;
6229 pin.table_id = ctx->table_id;
6230 pin.cookie = ctx->rule ? ctx->rule->up.flow_cookie : 0;
6233 flow_get_metadata(&ctx->xin->flow, &pin.fmd);
6235 connmgr_send_packet_in(ctx->ofproto->up.connmgr, &pin);
6236 ofpbuf_delete(packet);
6240 execute_mpls_push_action(struct xlate_ctx *ctx, ovs_be16 eth_type)
6242 ovs_assert(eth_type_mpls(eth_type));
6244 if (ctx->base_flow.mpls_depth) {
6245 ctx->xin->flow.mpls_lse &= ~htonl(MPLS_BOS_MASK);
6246 ctx->xin->flow.mpls_depth++;
6251 if (ctx->xin->flow.dl_type == htons(ETH_TYPE_IPV6)) {
6252 label = htonl(0x2); /* IPV6 Explicit Null. */
6254 label = htonl(0x0); /* IPV4 Explicit Null. */
6256 tc = (ctx->xin->flow.nw_tos & IP_DSCP_MASK) >> 2;
6257 ttl = ctx->xin->flow.nw_ttl ? ctx->xin->flow.nw_ttl : 0x40;
6258 ctx->xin->flow.mpls_lse = set_mpls_lse_values(ttl, tc, 1, label);
6259 ctx->xin->flow.mpls_depth = 1;
6261 ctx->xin->flow.dl_type = eth_type;
6265 execute_mpls_pop_action(struct xlate_ctx *ctx, ovs_be16 eth_type)
6267 ovs_assert(eth_type_mpls(ctx->xin->flow.dl_type));
6268 ovs_assert(!eth_type_mpls(eth_type));
6270 if (ctx->xin->flow.mpls_depth) {
6271 ctx->xin->flow.mpls_depth--;
6272 ctx->xin->flow.mpls_lse = htonl(0);
6273 if (!ctx->xin->flow.mpls_depth) {
6274 ctx->xin->flow.dl_type = eth_type;
6280 compose_dec_ttl(struct xlate_ctx *ctx, struct ofpact_cnt_ids *ids)
6282 if (ctx->xin->flow.dl_type != htons(ETH_TYPE_IP) &&
6283 ctx->xin->flow.dl_type != htons(ETH_TYPE_IPV6)) {
6287 if (ctx->xin->flow.nw_ttl > 1) {
6288 ctx->xin->flow.nw_ttl--;
6293 for (i = 0; i < ids->n_controllers; i++) {
6294 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL,
6298 /* Stop processing for current table. */
6304 execute_set_mpls_ttl_action(struct xlate_ctx *ctx, uint8_t ttl)
6306 if (!eth_type_mpls(ctx->xin->flow.dl_type)) {
6310 set_mpls_lse_ttl(&ctx->xin->flow.mpls_lse, ttl);
6315 execute_dec_mpls_ttl_action(struct xlate_ctx *ctx)
6317 uint8_t ttl = mpls_lse_to_ttl(ctx->xin->flow.mpls_lse);
6319 if (!eth_type_mpls(ctx->xin->flow.dl_type)) {
6325 set_mpls_lse_ttl(&ctx->xin->flow.mpls_lse, ttl);
6328 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL, 0);
6330 /* Stop processing for current table. */
6336 xlate_output_action(struct xlate_ctx *ctx,
6337 uint16_t port, uint16_t max_len, bool may_packet_in)
6339 uint16_t prev_nf_output_iface = ctx->xout->nf_output_iface;
6341 ctx->xout->nf_output_iface = NF_OUT_DROP;
6345 compose_output_action(ctx, ctx->xin->flow.in_port);
6348 xlate_table_action(ctx, ctx->xin->flow.in_port, 0, may_packet_in);
6354 flood_packets(ctx, false);
6357 flood_packets(ctx, true);
6359 case OFPP_CONTROLLER:
6360 execute_controller_action(ctx, max_len, OFPR_ACTION, 0);
6366 if (port != ctx->xin->flow.in_port) {
6367 compose_output_action(ctx, port);
6369 xlate_report(ctx, "skipping output to input port");
6374 if (prev_nf_output_iface == NF_OUT_FLOOD) {
6375 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
6376 } else if (ctx->xout->nf_output_iface == NF_OUT_DROP) {
6377 ctx->xout->nf_output_iface = prev_nf_output_iface;
6378 } else if (prev_nf_output_iface != NF_OUT_DROP &&
6379 ctx->xout->nf_output_iface != NF_OUT_FLOOD) {
6380 ctx->xout->nf_output_iface = NF_OUT_MULTI;
6385 xlate_output_reg_action(struct xlate_ctx *ctx,
6386 const struct ofpact_output_reg *or)
6388 uint64_t port = mf_get_subfield(&or->src, &ctx->xin->flow);
6389 if (port <= UINT16_MAX) {
6390 xlate_output_action(ctx, port, or->max_len, false);
6395 xlate_enqueue_action(struct xlate_ctx *ctx,
6396 const struct ofpact_enqueue *enqueue)
6398 uint16_t ofp_port = enqueue->port;
6399 uint32_t queue_id = enqueue->queue;
6400 uint32_t flow_priority, priority;
6403 /* Translate queue to priority. */
6404 error = dpif_queue_to_priority(ctx->ofproto->backer->dpif,
6405 queue_id, &priority);
6407 /* Fall back to ordinary output action. */
6408 xlate_output_action(ctx, enqueue->port, 0, false);
6412 /* Check output port. */
6413 if (ofp_port == OFPP_IN_PORT) {
6414 ofp_port = ctx->xin->flow.in_port;
6415 } else if (ofp_port == ctx->xin->flow.in_port) {
6419 /* Add datapath actions. */
6420 flow_priority = ctx->xin->flow.skb_priority;
6421 ctx->xin->flow.skb_priority = priority;
6422 compose_output_action(ctx, ofp_port);
6423 ctx->xin->flow.skb_priority = flow_priority;
6425 /* Update NetFlow output port. */
6426 if (ctx->xout->nf_output_iface == NF_OUT_DROP) {
6427 ctx->xout->nf_output_iface = ofp_port;
6428 } else if (ctx->xout->nf_output_iface != NF_OUT_FLOOD) {
6429 ctx->xout->nf_output_iface = NF_OUT_MULTI;
6434 xlate_set_queue_action(struct xlate_ctx *ctx, uint32_t queue_id)
6436 uint32_t skb_priority;
6438 if (!dpif_queue_to_priority(ctx->ofproto->backer->dpif,
6439 queue_id, &skb_priority)) {
6440 ctx->xin->flow.skb_priority = skb_priority;
6442 /* Couldn't translate queue to a priority. Nothing to do. A warning
6443 * has already been logged. */
6448 slave_enabled_cb(uint16_t ofp_port, void *ofproto_)
6450 struct ofproto_dpif *ofproto = ofproto_;
6451 struct ofport_dpif *port;
6461 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
6464 port = get_ofp_port(ofproto, ofp_port);
6465 return port ? port->may_enable : false;
6470 xlate_bundle_action(struct xlate_ctx *ctx,
6471 const struct ofpact_bundle *bundle)
6475 port = bundle_execute(bundle, &ctx->xin->flow, slave_enabled_cb,
6477 if (bundle->dst.field) {
6478 nxm_reg_load(&bundle->dst, port, &ctx->xin->flow);
6480 xlate_output_action(ctx, port, 0, false);
6485 xlate_learn_action(struct xlate_ctx *ctx,
6486 const struct ofpact_learn *learn)
6488 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
6489 struct ofputil_flow_mod fm;
6490 uint64_t ofpacts_stub[1024 / 8];
6491 struct ofpbuf ofpacts;
6494 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
6495 learn_execute(learn, &ctx->xin->flow, &fm, &ofpacts);
6497 error = ofproto_flow_mod(&ctx->ofproto->up, &fm);
6498 if (error && !VLOG_DROP_WARN(&rl)) {
6499 VLOG_WARN("learning action failed to modify flow table (%s)",
6500 ofperr_get_name(error));
6503 ofpbuf_uninit(&ofpacts);
6506 /* Reduces '*timeout' to no more than 'max'. A value of zero in either case
6507 * means "infinite". */
6509 reduce_timeout(uint16_t max, uint16_t *timeout)
6511 if (max && (!*timeout || *timeout > max)) {
6517 xlate_fin_timeout(struct xlate_ctx *ctx,
6518 const struct ofpact_fin_timeout *oft)
6520 if (ctx->xin->tcp_flags & (TCP_FIN | TCP_RST) && ctx->rule) {
6521 struct rule_dpif *rule = ctx->rule;
6523 reduce_timeout(oft->fin_idle_timeout, &rule->up.idle_timeout);
6524 reduce_timeout(oft->fin_hard_timeout, &rule->up.hard_timeout);
6529 xlate_sample_action(struct xlate_ctx *ctx,
6530 const struct ofpact_sample *os)
6532 union user_action_cookie cookie;
6533 /* Scale the probability from 16-bit to 32-bit while representing
6534 * the same percentage. */
6535 uint32_t probability = (os->probability << 16) | os->probability;
6537 commit_odp_actions(&ctx->xin->flow, &ctx->base_flow,
6538 &ctx->xout->odp_actions);
6540 compose_flow_sample_cookie(os->probability, os->collector_set_id,
6541 os->obs_domain_id, os->obs_point_id, &cookie);
6542 compose_sample_action(ctx->ofproto, &ctx->xout->odp_actions, &ctx->xin->flow,
6543 probability, &cookie, sizeof cookie.flow_sample);
6547 may_receive(const struct ofport_dpif *port, struct xlate_ctx *ctx)
6549 if (port->up.pp.config & (eth_addr_equals(ctx->xin->flow.dl_dst,
6551 ? OFPUTIL_PC_NO_RECV_STP
6552 : OFPUTIL_PC_NO_RECV)) {
6556 /* Only drop packets here if both forwarding and learning are
6557 * disabled. If just learning is enabled, we need to have
6558 * OFPP_NORMAL and the learning action have a look at the packet
6559 * before we can drop it. */
6560 if (!stp_forward_in_state(port->stp_state)
6561 && !stp_learn_in_state(port->stp_state)) {
6569 tunnel_ecn_ok(struct xlate_ctx *ctx)
6571 if (is_ip_any(&ctx->base_flow)
6572 && (ctx->xin->flow.tunnel.ip_tos & IP_ECN_MASK) == IP_ECN_CE) {
6573 if ((ctx->base_flow.nw_tos & IP_ECN_MASK) == IP_ECN_NOT_ECT) {
6574 VLOG_WARN_RL(&rl, "dropping tunnel packet marked ECN CE"
6575 " but is not ECN capable");
6578 /* Set the ECN CE value in the tunneled packet. */
6579 ctx->xin->flow.nw_tos |= IP_ECN_CE;
6587 do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len,
6588 struct xlate_ctx *ctx)
6590 bool was_evictable = true;
6591 const struct ofpact *a;
6594 /* Don't let the rule we're working on get evicted underneath us. */
6595 was_evictable = ctx->rule->up.evictable;
6596 ctx->rule->up.evictable = false;
6599 do_xlate_actions_again:
6600 OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) {
6601 struct ofpact_controller *controller;
6602 const struct ofpact_metadata *metadata;
6610 xlate_output_action(ctx, ofpact_get_OUTPUT(a)->port,
6611 ofpact_get_OUTPUT(a)->max_len, true);
6614 case OFPACT_CONTROLLER:
6615 controller = ofpact_get_CONTROLLER(a);
6616 execute_controller_action(ctx, controller->max_len,
6618 controller->controller_id);
6621 case OFPACT_ENQUEUE:
6622 xlate_enqueue_action(ctx, ofpact_get_ENQUEUE(a));
6625 case OFPACT_SET_VLAN_VID:
6626 ctx->xin->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
6627 ctx->xin->flow.vlan_tci |=
6628 (htons(ofpact_get_SET_VLAN_VID(a)->vlan_vid)
6632 case OFPACT_SET_VLAN_PCP:
6633 ctx->xin->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
6634 ctx->xin->flow.vlan_tci |=
6635 htons((ofpact_get_SET_VLAN_PCP(a)->vlan_pcp << VLAN_PCP_SHIFT)
6639 case OFPACT_STRIP_VLAN:
6640 ctx->xin->flow.vlan_tci = htons(0);
6643 case OFPACT_PUSH_VLAN:
6644 /* XXX 802.1AD(QinQ) */
6645 ctx->xin->flow.vlan_tci = htons(VLAN_CFI);
6648 case OFPACT_SET_ETH_SRC:
6649 memcpy(ctx->xin->flow.dl_src, ofpact_get_SET_ETH_SRC(a)->mac,
6653 case OFPACT_SET_ETH_DST:
6654 memcpy(ctx->xin->flow.dl_dst, ofpact_get_SET_ETH_DST(a)->mac,
6658 case OFPACT_SET_IPV4_SRC:
6659 if (ctx->xin->flow.dl_type == htons(ETH_TYPE_IP)) {
6660 ctx->xin->flow.nw_src = ofpact_get_SET_IPV4_SRC(a)->ipv4;
6664 case OFPACT_SET_IPV4_DST:
6665 if (ctx->xin->flow.dl_type == htons(ETH_TYPE_IP)) {
6666 ctx->xin->flow.nw_dst = ofpact_get_SET_IPV4_DST(a)->ipv4;
6670 case OFPACT_SET_IPV4_DSCP:
6671 /* OpenFlow 1.0 only supports IPv4. */
6672 if (ctx->xin->flow.dl_type == htons(ETH_TYPE_IP)) {
6673 ctx->xin->flow.nw_tos &= ~IP_DSCP_MASK;
6674 ctx->xin->flow.nw_tos |= ofpact_get_SET_IPV4_DSCP(a)->dscp;
6678 case OFPACT_SET_L4_SRC_PORT:
6679 if (is_ip_any(&ctx->xin->flow)) {
6680 ctx->xin->flow.tp_src =
6681 htons(ofpact_get_SET_L4_SRC_PORT(a)->port);
6685 case OFPACT_SET_L4_DST_PORT:
6686 if (is_ip_any(&ctx->xin->flow)) {
6687 ctx->xin->flow.tp_dst =
6688 htons(ofpact_get_SET_L4_DST_PORT(a)->port);
6692 case OFPACT_RESUBMIT:
6693 xlate_ofpact_resubmit(ctx, ofpact_get_RESUBMIT(a));
6696 case OFPACT_SET_TUNNEL:
6697 ctx->xin->flow.tunnel.tun_id =
6698 htonll(ofpact_get_SET_TUNNEL(a)->tun_id);
6701 case OFPACT_SET_QUEUE:
6702 xlate_set_queue_action(ctx, ofpact_get_SET_QUEUE(a)->queue_id);
6705 case OFPACT_POP_QUEUE:
6706 ctx->xin->flow.skb_priority = ctx->orig_skb_priority;
6709 case OFPACT_REG_MOVE:
6710 nxm_execute_reg_move(ofpact_get_REG_MOVE(a), &ctx->xin->flow);
6713 case OFPACT_REG_LOAD:
6714 nxm_execute_reg_load(ofpact_get_REG_LOAD(a), &ctx->xin->flow);
6717 case OFPACT_STACK_PUSH:
6718 nxm_execute_stack_push(ofpact_get_STACK_PUSH(a), &ctx->xin->flow,
6722 case OFPACT_STACK_POP:
6723 nxm_execute_stack_pop(ofpact_get_STACK_POP(a), &ctx->xin->flow,
6727 case OFPACT_PUSH_MPLS:
6728 execute_mpls_push_action(ctx, ofpact_get_PUSH_MPLS(a)->ethertype);
6731 case OFPACT_POP_MPLS:
6732 execute_mpls_pop_action(ctx, ofpact_get_POP_MPLS(a)->ethertype);
6735 case OFPACT_SET_MPLS_TTL:
6736 if (execute_set_mpls_ttl_action(ctx,
6737 ofpact_get_SET_MPLS_TTL(a)->ttl)) {
6742 case OFPACT_DEC_MPLS_TTL:
6743 if (execute_dec_mpls_ttl_action(ctx)) {
6748 case OFPACT_DEC_TTL:
6749 if (compose_dec_ttl(ctx, ofpact_get_DEC_TTL(a))) {
6755 /* Nothing to do. */
6758 case OFPACT_MULTIPATH:
6759 multipath_execute(ofpact_get_MULTIPATH(a), &ctx->xin->flow);
6763 ctx->ofproto->has_bundle_action = true;
6764 xlate_bundle_action(ctx, ofpact_get_BUNDLE(a));
6767 case OFPACT_OUTPUT_REG:
6768 xlate_output_reg_action(ctx, ofpact_get_OUTPUT_REG(a));
6772 ctx->xout->has_learn = true;
6773 if (ctx->xin->may_learn) {
6774 xlate_learn_action(ctx, ofpact_get_LEARN(a));
6782 case OFPACT_FIN_TIMEOUT:
6783 ctx->xout->has_fin_timeout = true;
6784 xlate_fin_timeout(ctx, ofpact_get_FIN_TIMEOUT(a));
6787 case OFPACT_CLEAR_ACTIONS:
6789 * Nothing to do because writa-actions is not supported for now.
6790 * When writa-actions is supported, clear-actions also must
6791 * be supported at the same time.
6795 case OFPACT_WRITE_METADATA:
6796 metadata = ofpact_get_WRITE_METADATA(a);
6797 ctx->xin->flow.metadata &= ~metadata->mask;
6798 ctx->xin->flow.metadata |= metadata->metadata & metadata->mask;
6801 case OFPACT_GOTO_TABLE: {
6802 /* It is assumed that goto-table is the last action. */
6803 struct ofpact_goto_table *ogt = ofpact_get_GOTO_TABLE(a);
6804 struct rule_dpif *rule;
6806 ovs_assert(ctx->table_id < ogt->table_id);
6808 ctx->table_id = ogt->table_id;
6810 /* Look up a flow from the new table. */
6811 rule = rule_dpif_lookup__(ctx->ofproto, &ctx->xin->flow, ctx->table_id);
6813 tag_the_flow(ctx, rule);
6815 rule = ctx_rule_hooks(ctx, rule, true);
6819 ctx->rule->up.evictable = was_evictable;
6822 was_evictable = rule->up.evictable;
6823 rule->up.evictable = false;
6825 /* Tail recursion removal. */
6826 ofpacts = rule->up.ofpacts;
6827 ofpacts_len = rule->up.ofpacts_len;
6828 goto do_xlate_actions_again;
6834 xlate_sample_action(ctx, ofpact_get_SAMPLE(a));
6841 ctx->rule->up.evictable = was_evictable;
6846 xlate_in_init(struct xlate_in *xin, struct ofproto_dpif *ofproto,
6847 const struct flow *flow,
6848 const struct initial_vals *initial_vals,
6849 struct rule_dpif *rule, uint8_t tcp_flags,
6850 const struct ofpbuf *packet)
6852 xin->ofproto = ofproto;
6854 xin->packet = packet;
6855 xin->may_learn = packet != NULL;
6857 xin->ofpacts = NULL;
6858 xin->ofpacts_len = 0;
6859 xin->tcp_flags = tcp_flags;
6860 xin->resubmit_hook = NULL;
6861 xin->report_hook = NULL;
6862 xin->resubmit_stats = NULL;
6865 xin->initial_vals = *initial_vals;
6867 xin->initial_vals.vlan_tci = xin->flow.vlan_tci;
6872 xlate_out_uninit(struct xlate_out *xout)
6875 ofpbuf_uninit(&xout->odp_actions);
6879 /* Translates the 'ofpacts_len' bytes of "struct ofpacts" starting at 'ofpacts'
6880 * into datapath actions in 'odp_actions', using 'ctx'. */
6882 xlate_actions(struct xlate_in *xin, struct xlate_out *xout)
6884 /* Normally false. Set to true if we ever hit MAX_RESUBMIT_RECURSION, so
6885 * that in the future we always keep a copy of the original flow for
6886 * tracing purposes. */
6887 static bool hit_resubmit_limit;
6889 enum slow_path_reason special;
6890 const struct ofpact *ofpacts;
6891 struct ofport_dpif *in_port;
6892 struct flow orig_flow;
6893 struct xlate_ctx ctx;
6896 COVERAGE_INC(ofproto_dpif_xlate);
6898 /* Flow initialization rules:
6899 * - 'base_flow' must match the kernel's view of the packet at the
6900 * time that action processing starts. 'flow' represents any
6901 * transformations we wish to make through actions.
6902 * - By default 'base_flow' and 'flow' are the same since the input
6903 * packet matches the output before any actions are applied.
6904 * - When using VLAN splinters, 'base_flow''s VLAN is set to the value
6905 * of the received packet as seen by the kernel. If we later output
6906 * to another device without any modifications this will cause us to
6907 * insert a new tag since the original one was stripped off by the
6909 * - Tunnel metadata as received is retained in 'flow'. This allows
6910 * tunnel metadata matching also in later tables.
6911 * Since a kernel action for setting the tunnel metadata will only be
6912 * generated with actual tunnel output, changing the tunnel metadata
6913 * values in 'flow' (such as tun_id) will only have effect with a later
6914 * tunnel output action.
6915 * - Tunnel 'base_flow' is completely cleared since that is what the
6916 * kernel does. If we wish to maintain the original values an action
6917 * needs to be generated. */
6922 ctx.ofproto = xin->ofproto;
6923 ctx.rule = xin->rule;
6925 ctx.base_flow = ctx.xin->flow;
6926 ctx.base_flow.vlan_tci = xin->initial_vals.vlan_tci;
6927 memset(&ctx.base_flow.tunnel, 0, sizeof ctx.base_flow.tunnel);
6928 ctx.orig_tunnel_ip_dst = ctx.xin->flow.tunnel.ip_dst;
6932 ctx.xout->has_learn = false;
6933 ctx.xout->has_normal = false;
6934 ctx.xout->has_fin_timeout = false;
6935 ctx.xout->nf_output_iface = NF_OUT_DROP;
6936 ctx.xout->mirrors = 0;
6938 ofpbuf_use_stub(&ctx.xout->odp_actions, ctx.xout->odp_actions_stub,
6939 sizeof ctx.xout->odp_actions_stub);
6940 ofpbuf_reserve(&ctx.xout->odp_actions, NL_A_U32_SIZE);
6943 ctx.max_resubmit_trigger = false;
6944 ctx.orig_skb_priority = ctx.xin->flow.skb_priority;
6949 ofpacts = xin->ofpacts;
6950 ofpacts_len = xin->ofpacts_len;
6951 } else if (xin->rule) {
6952 ofpacts = xin->rule->up.ofpacts;
6953 ofpacts_len = xin->rule->up.ofpacts_len;
6958 ofpbuf_use_stub(&ctx.stack, ctx.init_stack, sizeof ctx.init_stack);
6960 if (ctx.ofproto->has_mirrors || hit_resubmit_limit) {
6961 /* Do this conditionally because the copy is expensive enough that it
6962 * shows up in profiles. */
6963 orig_flow = ctx.xin->flow;
6966 if (ctx.xin->flow.nw_frag & FLOW_NW_FRAG_ANY) {
6967 switch (ctx.ofproto->up.frag_handling) {
6968 case OFPC_FRAG_NORMAL:
6969 /* We must pretend that transport ports are unavailable. */
6970 ctx.xin->flow.tp_src = ctx.base_flow.tp_src = htons(0);
6971 ctx.xin->flow.tp_dst = ctx.base_flow.tp_dst = htons(0);
6974 case OFPC_FRAG_DROP:
6977 case OFPC_FRAG_REASM:
6980 case OFPC_FRAG_NX_MATCH:
6981 /* Nothing to do. */
6984 case OFPC_INVALID_TTL_TO_CONTROLLER:
6989 in_port = get_ofp_port(ctx.ofproto, ctx.xin->flow.in_port);
6990 special = process_special(ctx.ofproto, &ctx.xin->flow, in_port,
6993 ctx.xout->slow = special;
6995 static struct vlog_rate_limit trace_rl = VLOG_RATE_LIMIT_INIT(1, 1);
6996 struct initial_vals initial_vals;
6997 size_t sample_actions_len;
6998 uint32_t local_odp_port;
7000 initial_vals.vlan_tci = ctx.base_flow.vlan_tci;
7002 add_sflow_action(&ctx);
7003 add_ipfix_action(&ctx);
7004 sample_actions_len = ctx.xout->odp_actions.size;
7006 if (tunnel_ecn_ok(&ctx) && (!in_port || may_receive(in_port, &ctx))) {
7007 do_xlate_actions(ofpacts, ofpacts_len, &ctx);
7009 /* We've let OFPP_NORMAL and the learning action look at the
7010 * packet, so drop it now if forwarding is disabled. */
7011 if (in_port && !stp_forward_in_state(in_port->stp_state)) {
7012 ctx.xout->odp_actions.size = sample_actions_len;
7016 if (ctx.max_resubmit_trigger && !ctx.xin->resubmit_hook) {
7017 if (!hit_resubmit_limit) {
7018 /* We didn't record the original flow. Make sure we do from
7020 hit_resubmit_limit = true;
7021 } else if (!VLOG_DROP_ERR(&trace_rl)) {
7022 struct ds ds = DS_EMPTY_INITIALIZER;
7024 ofproto_trace(ctx.ofproto, &orig_flow, ctx.xin->packet,
7025 &initial_vals, &ds);
7026 VLOG_ERR("Trace triggered by excessive resubmit "
7027 "recursion:\n%s", ds_cstr(&ds));
7032 local_odp_port = ofp_port_to_odp_port(ctx.ofproto, OFPP_LOCAL);
7033 if (!connmgr_must_output_local(ctx.ofproto->up.connmgr, &ctx.xin->flow,
7035 ctx.xout->odp_actions.data,
7036 ctx.xout->odp_actions.size)) {
7037 compose_output_action(&ctx, OFPP_LOCAL);
7039 if (ctx.ofproto->has_mirrors) {
7040 add_mirror_actions(&ctx, &orig_flow);
7042 fix_sflow_action(&ctx);
7045 ofpbuf_uninit(&ctx.stack);
7048 /* Translates the 'ofpacts_len' bytes of "struct ofpact"s starting at 'ofpacts'
7049 * into datapath actions, using 'ctx', and discards the datapath actions. */
7051 xlate_actions_for_side_effects(struct xlate_in *xin)
7053 struct xlate_out xout;
7055 xlate_actions(xin, &xout);
7056 xlate_out_uninit(&xout);
7060 xlate_report(struct xlate_ctx *ctx, const char *s)
7062 if (ctx->xin->report_hook) {
7063 ctx->xin->report_hook(ctx, s);
7067 /* OFPP_NORMAL implementation. */
7069 static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *);
7071 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
7072 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle',
7073 * the bundle on which the packet was received, returns the VLAN to which the
7076 * Both 'vid' and the return value are in the range 0...4095. */
7078 input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid)
7080 switch (in_bundle->vlan_mode) {
7081 case PORT_VLAN_ACCESS:
7082 return in_bundle->vlan;
7085 case PORT_VLAN_TRUNK:
7088 case PORT_VLAN_NATIVE_UNTAGGED:
7089 case PORT_VLAN_NATIVE_TAGGED:
7090 return vid ? vid : in_bundle->vlan;
7097 /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'.
7098 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
7101 * 'vid' should be the VID obtained from the 802.1Q header that was received as
7102 * part of a packet (specify 0 if there was no 802.1Q header), in the range
7105 input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn)
7107 /* Allow any VID on the OFPP_NONE port. */
7108 if (in_bundle == &ofpp_none_bundle) {
7112 switch (in_bundle->vlan_mode) {
7113 case PORT_VLAN_ACCESS:
7116 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7117 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
7118 "packet received on port %s configured as VLAN "
7119 "%"PRIu16" access port",
7120 in_bundle->ofproto->up.name, vid,
7121 in_bundle->name, in_bundle->vlan);
7127 case PORT_VLAN_NATIVE_UNTAGGED:
7128 case PORT_VLAN_NATIVE_TAGGED:
7130 /* Port must always carry its native VLAN. */
7134 case PORT_VLAN_TRUNK:
7135 if (!ofbundle_includes_vlan(in_bundle, vid)) {
7137 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7138 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet "
7139 "received on port %s not configured for trunking "
7141 in_bundle->ofproto->up.name, vid,
7142 in_bundle->name, vid);
7154 /* Given 'vlan', the VLAN that a packet belongs to, and
7155 * 'out_bundle', a bundle on which the packet is to be output, returns the VID
7156 * that should be included in the 802.1Q header. (If the return value is 0,
7157 * then the 802.1Q header should only be included in the packet if there is a
7160 * Both 'vlan' and the return value are in the range 0...4095. */
7162 output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan)
7164 switch (out_bundle->vlan_mode) {
7165 case PORT_VLAN_ACCESS:
7168 case PORT_VLAN_TRUNK:
7169 case PORT_VLAN_NATIVE_TAGGED:
7172 case PORT_VLAN_NATIVE_UNTAGGED:
7173 return vlan == out_bundle->vlan ? 0 : vlan;
7181 output_normal(struct xlate_ctx *ctx, const struct ofbundle *out_bundle,
7184 struct ofport_dpif *port;
7186 ovs_be16 tci, old_tci;
7188 vid = output_vlan_to_vid(out_bundle, vlan);
7189 if (!out_bundle->bond) {
7190 port = ofbundle_get_a_port(out_bundle);
7192 port = bond_choose_output_slave(out_bundle->bond, &ctx->xin->flow,
7193 vid, &ctx->xout->tags);
7195 /* No slaves enabled, so drop packet. */
7200 old_tci = ctx->xin->flow.vlan_tci;
7202 if (tci || out_bundle->use_priority_tags) {
7203 tci |= ctx->xin->flow.vlan_tci & htons(VLAN_PCP_MASK);
7205 tci |= htons(VLAN_CFI);
7208 ctx->xin->flow.vlan_tci = tci;
7210 compose_output_action(ctx, port->up.ofp_port);
7211 ctx->xin->flow.vlan_tci = old_tci;
7215 mirror_mask_ffs(mirror_mask_t mask)
7217 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
7222 ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
7224 return (bundle->vlan_mode != PORT_VLAN_ACCESS
7225 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
7229 ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan)
7231 return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan);
7234 /* Returns an arbitrary interface within 'bundle'. */
7235 static struct ofport_dpif *
7236 ofbundle_get_a_port(const struct ofbundle *bundle)
7238 return CONTAINER_OF(list_front(&bundle->ports),
7239 struct ofport_dpif, bundle_node);
7243 vlan_is_mirrored(const struct ofmirror *m, int vlan)
7245 return !m->vlans || bitmap_is_set(m->vlans, vlan);
7249 add_mirror_actions(struct xlate_ctx *ctx, const struct flow *orig_flow)
7251 struct ofproto_dpif *ofproto = ctx->ofproto;
7252 mirror_mask_t mirrors;
7253 struct ofbundle *in_bundle;
7256 const struct nlattr *a;
7259 in_bundle = lookup_input_bundle(ctx->ofproto, orig_flow->in_port,
7260 ctx->xin->packet != NULL, NULL);
7264 mirrors = in_bundle->src_mirrors;
7266 /* Drop frames on bundles reserved for mirroring. */
7267 if (in_bundle->mirror_out) {
7268 if (ctx->xin->packet != NULL) {
7269 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7270 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
7271 "%s, which is reserved exclusively for mirroring",
7272 ctx->ofproto->up.name, in_bundle->name);
7278 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
7279 if (!input_vid_is_valid(vid, in_bundle, ctx->xin->packet != NULL)) {
7282 vlan = input_vid_to_vlan(in_bundle, vid);
7284 /* Look at the output ports to check for destination selections. */
7286 NL_ATTR_FOR_EACH (a, left, ctx->xout->odp_actions.data,
7287 ctx->xout->odp_actions.size) {
7288 enum ovs_action_attr type = nl_attr_type(a);
7289 struct ofport_dpif *ofport;
7291 if (type != OVS_ACTION_ATTR_OUTPUT) {
7295 ofport = get_odp_port(ofproto, nl_attr_get_u32(a));
7296 if (ofport && ofport->bundle) {
7297 mirrors |= ofport->bundle->dst_mirrors;
7305 /* Restore the original packet before adding the mirror actions. */
7306 ctx->xin->flow = *orig_flow;
7311 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
7313 if (!vlan_is_mirrored(m, vlan)) {
7314 mirrors = zero_rightmost_1bit(mirrors);
7318 mirrors &= ~m->dup_mirrors;
7319 ctx->xout->mirrors |= m->dup_mirrors;
7321 output_normal(ctx, m->out, vlan);
7322 } else if (vlan != m->out_vlan
7323 && !eth_addr_is_reserved(orig_flow->dl_dst)) {
7324 struct ofbundle *bundle;
7326 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
7327 if (ofbundle_includes_vlan(bundle, m->out_vlan)
7328 && !bundle->mirror_out) {
7329 output_normal(ctx, bundle, m->out_vlan);
7337 update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors,
7338 uint64_t packets, uint64_t bytes)
7344 for (; mirrors; mirrors = zero_rightmost_1bit(mirrors)) {
7347 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
7350 /* In normal circumstances 'm' will not be NULL. However,
7351 * if mirrors are reconfigured, we can temporarily get out
7352 * of sync in facet_revalidate(). We could "correct" the
7353 * mirror list before reaching here, but doing that would
7354 * not properly account the traffic stats we've currently
7355 * accumulated for previous mirror configuration. */
7359 m->packet_count += packets;
7360 m->byte_count += bytes;
7364 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
7365 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
7366 * indicate this; newer upstream kernels use gratuitous ARP requests. */
7368 is_gratuitous_arp(const struct flow *flow)
7370 return (flow->dl_type == htons(ETH_TYPE_ARP)
7371 && eth_addr_is_broadcast(flow->dl_dst)
7372 && (flow->nw_proto == ARP_OP_REPLY
7373 || (flow->nw_proto == ARP_OP_REQUEST
7374 && flow->nw_src == flow->nw_dst)));
7378 update_learning_table(struct ofproto_dpif *ofproto,
7379 const struct flow *flow, int vlan,
7380 struct ofbundle *in_bundle)
7382 struct mac_entry *mac;
7384 /* Don't learn the OFPP_NONE port. */
7385 if (in_bundle == &ofpp_none_bundle) {
7389 if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
7393 mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan);
7394 if (is_gratuitous_arp(flow)) {
7395 /* We don't want to learn from gratuitous ARP packets that are
7396 * reflected back over bond slaves so we lock the learning table. */
7397 if (!in_bundle->bond) {
7398 mac_entry_set_grat_arp_lock(mac);
7399 } else if (mac_entry_is_grat_arp_locked(mac)) {
7404 if (mac_entry_is_new(mac) || mac->port.p != in_bundle) {
7405 /* The log messages here could actually be useful in debugging,
7406 * so keep the rate limit relatively high. */
7407 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
7408 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
7409 "on port %s in VLAN %d",
7410 ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src),
7411 in_bundle->name, vlan);
7413 mac->port.p = in_bundle;
7414 tag_set_add(&ofproto->backer->revalidate_set,
7415 mac_learning_changed(ofproto->ml, mac));
7419 static struct ofbundle *
7420 lookup_input_bundle(const struct ofproto_dpif *ofproto, uint16_t in_port,
7421 bool warn, struct ofport_dpif **in_ofportp)
7423 struct ofport_dpif *ofport;
7425 /* Find the port and bundle for the received packet. */
7426 ofport = get_ofp_port(ofproto, in_port);
7428 *in_ofportp = ofport;
7430 if (ofport && ofport->bundle) {
7431 return ofport->bundle;
7434 /* Special-case OFPP_NONE, which a controller may use as the ingress
7435 * port for traffic that it is sourcing. */
7436 if (in_port == OFPP_NONE) {
7437 return &ofpp_none_bundle;
7440 /* Odd. A few possible reasons here:
7442 * - We deleted a port but there are still a few packets queued up
7445 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
7446 * we don't know about.
7448 * - The ofproto client didn't configure the port as part of a bundle.
7449 * This is particularly likely to happen if a packet was received on the
7450 * port after it was created, but before the client had a chance to
7451 * configure its bundle.
7454 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7456 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
7457 "port %"PRIu16, ofproto->up.name, in_port);
7462 /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or
7463 * dropped. Returns true if they may be forwarded, false if they should be
7466 * 'in_port' must be the ofport_dpif that corresponds to flow->in_port.
7467 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
7469 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
7470 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
7471 * checked by input_vid_is_valid().
7473 * May also add tags to '*tags', although the current implementation only does
7474 * so in one special case.
7477 is_admissible(struct xlate_ctx *ctx, struct ofport_dpif *in_port,
7480 struct ofproto_dpif *ofproto = ctx->ofproto;
7481 struct flow *flow = &ctx->xin->flow;
7482 struct ofbundle *in_bundle = in_port->bundle;
7484 /* Drop frames for reserved multicast addresses
7485 * only if forward_bpdu option is absent. */
7486 if (!ofproto->up.forward_bpdu && eth_addr_is_reserved(flow->dl_dst)) {
7487 xlate_report(ctx, "packet has reserved destination MAC, dropping");
7491 if (in_bundle->bond) {
7492 struct mac_entry *mac;
7494 switch (bond_check_admissibility(in_bundle->bond, in_port,
7495 flow->dl_dst, &ctx->xout->tags)) {
7500 xlate_report(ctx, "bonding refused admissibility, dropping");
7503 case BV_DROP_IF_MOVED:
7504 mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL);
7505 if (mac && mac->port.p != in_bundle &&
7506 (!is_gratuitous_arp(flow)
7507 || mac_entry_is_grat_arp_locked(mac))) {
7508 xlate_report(ctx, "SLB bond thinks this packet looped back, "
7520 xlate_normal(struct xlate_ctx *ctx)
7522 struct ofport_dpif *in_port;
7523 struct ofbundle *in_bundle;
7524 struct mac_entry *mac;
7528 ctx->xout->has_normal = true;
7530 in_bundle = lookup_input_bundle(ctx->ofproto, ctx->xin->flow.in_port,
7531 ctx->xin->packet != NULL, &in_port);
7533 xlate_report(ctx, "no input bundle, dropping");
7537 /* Drop malformed frames. */
7538 if (ctx->xin->flow.dl_type == htons(ETH_TYPE_VLAN) &&
7539 !(ctx->xin->flow.vlan_tci & htons(VLAN_CFI))) {
7540 if (ctx->xin->packet != NULL) {
7541 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7542 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
7543 "VLAN tag received on port %s",
7544 ctx->ofproto->up.name, in_bundle->name);
7546 xlate_report(ctx, "partial VLAN tag, dropping");
7550 /* Drop frames on bundles reserved for mirroring. */
7551 if (in_bundle->mirror_out) {
7552 if (ctx->xin->packet != NULL) {
7553 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7554 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
7555 "%s, which is reserved exclusively for mirroring",
7556 ctx->ofproto->up.name, in_bundle->name);
7558 xlate_report(ctx, "input port is mirror output port, dropping");
7563 vid = vlan_tci_to_vid(ctx->xin->flow.vlan_tci);
7564 if (!input_vid_is_valid(vid, in_bundle, ctx->xin->packet != NULL)) {
7565 xlate_report(ctx, "disallowed VLAN VID for this input port, dropping");
7568 vlan = input_vid_to_vlan(in_bundle, vid);
7570 /* Check other admissibility requirements. */
7571 if (in_port && !is_admissible(ctx, in_port, vlan)) {
7575 /* Learn source MAC. */
7576 if (ctx->xin->may_learn) {
7577 update_learning_table(ctx->ofproto, &ctx->xin->flow, vlan, in_bundle);
7580 /* Determine output bundle. */
7581 mac = mac_learning_lookup(ctx->ofproto->ml, ctx->xin->flow.dl_dst, vlan,
7584 if (mac->port.p != in_bundle) {
7585 xlate_report(ctx, "forwarding to learned port");
7586 output_normal(ctx, mac->port.p, vlan);
7588 xlate_report(ctx, "learned port is input port, dropping");
7591 struct ofbundle *bundle;
7593 xlate_report(ctx, "no learned MAC for destination, flooding");
7594 HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
7595 if (bundle != in_bundle
7596 && ofbundle_includes_vlan(bundle, vlan)
7597 && bundle->floodable
7598 && !bundle->mirror_out) {
7599 output_normal(ctx, bundle, vlan);
7602 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
7606 /* Optimized flow revalidation.
7608 * It's a difficult problem, in general, to tell which facets need to have
7609 * their actions recalculated whenever the OpenFlow flow table changes. We
7610 * don't try to solve that general problem: for most kinds of OpenFlow flow
7611 * table changes, we recalculate the actions for every facet. This is
7612 * relatively expensive, but it's good enough if the OpenFlow flow table
7613 * doesn't change very often.
7615 * However, we can expect one particular kind of OpenFlow flow table change to
7616 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
7617 * of CPU on revalidating every facet whenever MAC learning modifies the flow
7618 * table, we add a special case that applies to flow tables in which every rule
7619 * has the same form (that is, the same wildcards), except that the table is
7620 * also allowed to have a single "catch-all" flow that matches all packets. We
7621 * optimize this case by tagging all of the facets that resubmit into the table
7622 * and invalidating the same tag whenever a flow changes in that table. The
7623 * end result is that we revalidate just the facets that need it (and sometimes
7624 * a few more, but not all of the facets or even all of the facets that
7625 * resubmit to the table modified by MAC learning). */
7627 /* Calculates the tag to use for 'flow' and mask 'mask' when it is inserted
7628 * into an OpenFlow table with the given 'basis'. */
7630 rule_calculate_tag(const struct flow *flow, const struct minimask *mask,
7633 if (minimask_is_catchall(mask)) {
7636 uint32_t hash = flow_hash_in_minimask(flow, mask, secret);
7637 return tag_create_deterministic(hash);
7641 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
7642 * taggability of that table.
7644 * This function must be called after *each* change to a flow table. If you
7645 * skip calling it on some changes then the pointer comparisons at the end can
7646 * be invalid if you get unlucky. For example, if a flow removal causes a
7647 * cls_table to be destroyed and then a flow insertion causes a cls_table with
7648 * different wildcards to be created with the same address, then this function
7649 * will incorrectly skip revalidation. */
7651 table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
7653 struct table_dpif *table = &ofproto->tables[table_id];
7654 const struct oftable *oftable = &ofproto->up.tables[table_id];
7655 struct cls_table *catchall, *other;
7656 struct cls_table *t;
7658 catchall = other = NULL;
7660 switch (hmap_count(&oftable->cls.tables)) {
7662 /* We could tag this OpenFlow table but it would make the logic a
7663 * little harder and it's a corner case that doesn't seem worth it
7669 HMAP_FOR_EACH (t, hmap_node, &oftable->cls.tables) {
7670 if (cls_table_is_catchall(t)) {
7672 } else if (!other) {
7675 /* Indicate that we can't tag this by setting both tables to
7676 * NULL. (We know that 'catchall' is already NULL.) */
7683 /* Can't tag this table. */
7687 if (table->catchall_table != catchall || table->other_table != other) {
7688 table->catchall_table = catchall;
7689 table->other_table = other;
7690 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
7694 /* Given 'rule' that has changed in some way (either it is a rule being
7695 * inserted, a rule being deleted, or a rule whose actions are being
7696 * modified), marks facets for revalidation to ensure that packets will be
7697 * forwarded correctly according to the new state of the flow table.
7699 * This function must be called after *each* change to a flow table. See
7700 * the comment on table_update_taggable() for more information. */
7702 rule_invalidate(const struct rule_dpif *rule)
7704 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
7706 table_update_taggable(ofproto, rule->up.table_id);
7708 if (!ofproto->backer->need_revalidate) {
7709 struct table_dpif *table = &ofproto->tables[rule->up.table_id];
7711 if (table->other_table && rule->tag) {
7712 tag_set_add(&ofproto->backer->revalidate_set, rule->tag);
7714 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
7720 set_frag_handling(struct ofproto *ofproto_,
7721 enum ofp_config_flags frag_handling)
7723 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7724 if (frag_handling != OFPC_FRAG_REASM) {
7725 ofproto->backer->need_revalidate = REV_RECONFIGURE;
7733 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
7734 const struct flow *flow,
7735 const struct ofpact *ofpacts, size_t ofpacts_len)
7737 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7738 struct initial_vals initial_vals;
7739 struct odputil_keybuf keybuf;
7740 struct dpif_flow_stats stats;
7741 struct xlate_out xout;
7742 struct xlate_in xin;
7746 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
7747 odp_flow_key_from_flow(&key, flow,
7748 ofp_port_to_odp_port(ofproto, flow->in_port));
7750 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
7752 initial_vals.vlan_tci = flow->vlan_tci;
7753 xlate_in_init(&xin, ofproto, flow, &initial_vals, NULL, stats.tcp_flags,
7755 xin.resubmit_stats = &stats;
7756 xin.ofpacts_len = ofpacts_len;
7757 xin.ofpacts = ofpacts;
7759 xlate_actions(&xin, &xout);
7760 dpif_execute(ofproto->backer->dpif, key.data, key.size,
7761 xout.odp_actions.data, xout.odp_actions.size, packet);
7762 xlate_out_uninit(&xout);
7770 set_netflow(struct ofproto *ofproto_,
7771 const struct netflow_options *netflow_options)
7773 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7775 if (netflow_options) {
7776 if (!ofproto->netflow) {
7777 ofproto->netflow = netflow_create();
7779 return netflow_set_options(ofproto->netflow, netflow_options);
7781 netflow_destroy(ofproto->netflow);
7782 ofproto->netflow = NULL;
7788 get_netflow_ids(const struct ofproto *ofproto_,
7789 uint8_t *engine_type, uint8_t *engine_id)
7791 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7793 dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
7797 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
7799 if (!facet_is_controller_flow(facet) &&
7800 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
7801 struct subfacet *subfacet;
7802 struct ofexpired expired;
7804 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
7805 if (subfacet->path == SF_FAST_PATH) {
7806 struct dpif_flow_stats stats;
7808 subfacet_install(subfacet, &facet->xout.odp_actions, &stats);
7809 subfacet_update_stats(subfacet, &stats);
7813 expired.flow = facet->flow;
7814 expired.packet_count = facet->packet_count;
7815 expired.byte_count = facet->byte_count;
7816 expired.used = facet->used;
7817 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
7822 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
7824 struct facet *facet;
7826 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
7827 send_active_timeout(ofproto, facet);
7831 static struct ofproto_dpif *
7832 ofproto_dpif_lookup(const char *name)
7834 struct ofproto_dpif *ofproto;
7836 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
7837 hash_string(name, 0), &all_ofproto_dpifs) {
7838 if (!strcmp(ofproto->up.name, name)) {
7846 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
7847 const char *argv[], void *aux OVS_UNUSED)
7849 struct ofproto_dpif *ofproto;
7852 ofproto = ofproto_dpif_lookup(argv[1]);
7854 unixctl_command_reply_error(conn, "no such bridge");
7857 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
7859 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
7860 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
7864 unixctl_command_reply(conn, "table successfully flushed");
7868 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
7869 const char *argv[], void *aux OVS_UNUSED)
7871 struct ds ds = DS_EMPTY_INITIALIZER;
7872 const struct ofproto_dpif *ofproto;
7873 const struct mac_entry *e;
7875 ofproto = ofproto_dpif_lookup(argv[1]);
7877 unixctl_command_reply_error(conn, "no such bridge");
7881 ds_put_cstr(&ds, " port VLAN MAC Age\n");
7882 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
7883 struct ofbundle *bundle = e->port.p;
7884 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
7885 ofbundle_get_a_port(bundle)->odp_port,
7886 e->vlan, ETH_ADDR_ARGS(e->mac),
7887 mac_entry_age(ofproto->ml, e));
7889 unixctl_command_reply(conn, ds_cstr(&ds));
7894 struct xlate_out xout;
7895 struct xlate_in xin;
7901 trace_format_rule(struct ds *result, uint8_t table_id, int level,
7902 const struct rule_dpif *rule)
7904 ds_put_char_multiple(result, '\t', level);
7906 ds_put_cstr(result, "No match\n");
7910 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
7911 table_id, ntohll(rule->up.flow_cookie));
7912 cls_rule_format(&rule->up.cr, result);
7913 ds_put_char(result, '\n');
7915 ds_put_char_multiple(result, '\t', level);
7916 ds_put_cstr(result, "OpenFlow ");
7917 ofpacts_format(rule->up.ofpacts, rule->up.ofpacts_len, result);
7918 ds_put_char(result, '\n');
7922 trace_format_flow(struct ds *result, int level, const char *title,
7923 struct trace_ctx *trace)
7925 ds_put_char_multiple(result, '\t', level);
7926 ds_put_format(result, "%s: ", title);
7927 if (flow_equal(&trace->xin.flow, &trace->flow)) {
7928 ds_put_cstr(result, "unchanged");
7930 flow_format(result, &trace->xin.flow);
7931 trace->flow = trace->xin.flow;
7933 ds_put_char(result, '\n');
7937 trace_format_regs(struct ds *result, int level, const char *title,
7938 struct trace_ctx *trace)
7942 ds_put_char_multiple(result, '\t', level);
7943 ds_put_format(result, "%s:", title);
7944 for (i = 0; i < FLOW_N_REGS; i++) {
7945 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
7947 ds_put_char(result, '\n');
7951 trace_format_odp(struct ds *result, int level, const char *title,
7952 struct trace_ctx *trace)
7954 struct ofpbuf *odp_actions = &trace->xout.odp_actions;
7956 ds_put_char_multiple(result, '\t', level);
7957 ds_put_format(result, "%s: ", title);
7958 format_odp_actions(result, odp_actions->data, odp_actions->size);
7959 ds_put_char(result, '\n');
7963 trace_resubmit(struct xlate_ctx *ctx, struct rule_dpif *rule)
7965 struct trace_ctx *trace = CONTAINER_OF(ctx->xin, struct trace_ctx, xin);
7966 struct ds *result = trace->result;
7968 ds_put_char(result, '\n');
7969 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
7970 trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace);
7971 trace_format_odp(result, ctx->recurse + 1, "Resubmitted odp", trace);
7972 trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule);
7976 trace_report(struct xlate_ctx *ctx, const char *s)
7978 struct trace_ctx *trace = CONTAINER_OF(ctx->xin, struct trace_ctx, xin);
7979 struct ds *result = trace->result;
7981 ds_put_char_multiple(result, '\t', ctx->recurse);
7982 ds_put_cstr(result, s);
7983 ds_put_char(result, '\n');
7987 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
7988 void *aux OVS_UNUSED)
7990 const struct dpif_backer *backer;
7991 struct ofproto_dpif *ofproto;
7992 struct ofpbuf odp_key;
7993 struct ofpbuf *packet;
7994 struct initial_vals initial_vals;
8002 ofpbuf_init(&odp_key, 0);
8004 /* Handle "-generate" or a hex string as the last argument. */
8005 if (!strcmp(argv[argc - 1], "-generate")) {
8006 packet = ofpbuf_new(0);
8009 const char *error = eth_from_hex(argv[argc - 1], &packet);
8012 } else if (argc == 4) {
8013 /* The 3-argument form must end in "-generate' or a hex string. */
8014 unixctl_command_reply_error(conn, error);
8019 /* Parse the flow and determine whether a datapath or
8020 * bridge is specified. If function odp_flow_key_from_string()
8021 * returns 0, the flow is a odp_flow. If function
8022 * parse_ofp_exact_flow() returns 0, the flow is a br_flow. */
8023 if (!odp_flow_key_from_string(argv[argc - 1], NULL, &odp_key)) {
8024 /* If the odp_flow is the second argument,
8025 * the datapath name is the first argument. */
8027 const char *dp_type;
8028 if (!strncmp(argv[1], "ovs-", 4)) {
8029 dp_type = argv[1] + 4;
8033 backer = shash_find_data(&all_dpif_backers, dp_type);
8035 unixctl_command_reply_error(conn, "Cannot find datapath "
8040 /* No datapath name specified, so there should be only one
8042 struct shash_node *node;
8043 if (shash_count(&all_dpif_backers) != 1) {
8044 unixctl_command_reply_error(conn, "Must specify datapath "
8045 "name, there is more than one type of datapath");
8048 node = shash_first(&all_dpif_backers);
8049 backer = node->data;
8052 /* Extract the ofproto_dpif object from the ofproto_receive()
8054 if (ofproto_receive(backer, NULL, odp_key.data,
8055 odp_key.size, &flow, NULL, &ofproto, NULL,
8057 unixctl_command_reply_error(conn, "Invalid datapath flow");
8060 ds_put_format(&result, "Bridge: %s\n", ofproto->up.name);
8061 } else if (!parse_ofp_exact_flow(&flow, argv[argc - 1])) {
8063 unixctl_command_reply_error(conn, "Must specify bridge name");
8067 ofproto = ofproto_dpif_lookup(argv[1]);
8069 unixctl_command_reply_error(conn, "Unknown bridge name");
8072 initial_vals.vlan_tci = flow.vlan_tci;
8074 unixctl_command_reply_error(conn, "Bad flow syntax");
8078 /* Generate a packet, if requested. */
8080 if (!packet->size) {
8081 flow_compose(packet, &flow);
8083 ds_put_cstr(&result, "Packet: ");
8084 s = ofp_packet_to_string(packet->data, packet->size);
8085 ds_put_cstr(&result, s);
8088 /* Use the metadata from the flow and the packet argument
8089 * to reconstruct the flow. */
8090 flow_extract(packet, flow.skb_priority, flow.skb_mark, NULL,
8091 flow.in_port, &flow);
8092 initial_vals.vlan_tci = flow.vlan_tci;
8096 ofproto_trace(ofproto, &flow, packet, &initial_vals, &result);
8097 unixctl_command_reply(conn, ds_cstr(&result));
8100 ds_destroy(&result);
8101 ofpbuf_delete(packet);
8102 ofpbuf_uninit(&odp_key);
8106 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
8107 const struct ofpbuf *packet,
8108 const struct initial_vals *initial_vals, struct ds *ds)
8110 struct rule_dpif *rule;
8112 ds_put_cstr(ds, "Flow: ");
8113 flow_format(ds, flow);
8114 ds_put_char(ds, '\n');
8116 rule = rule_dpif_lookup(ofproto, flow);
8118 trace_format_rule(ds, 0, 0, rule);
8119 if (rule == ofproto->miss_rule) {
8120 ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
8121 } else if (rule == ofproto->no_packet_in_rule) {
8122 ds_put_cstr(ds, "\nNo match, packets dropped because "
8123 "OFPPC_NO_PACKET_IN is set on in_port.\n");
8127 uint64_t odp_actions_stub[1024 / 8];
8128 struct ofpbuf odp_actions;
8130 struct trace_ctx trace;
8133 tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
8136 ofpbuf_use_stub(&odp_actions,
8137 odp_actions_stub, sizeof odp_actions_stub);
8138 xlate_in_init(&trace.xin, ofproto, flow, initial_vals, rule, tcp_flags,
8140 trace.xin.resubmit_hook = trace_resubmit;
8141 trace.xin.report_hook = trace_report;
8142 xlate_actions(&trace.xin, &trace.xout);
8144 ds_put_char(ds, '\n');
8145 trace_format_flow(ds, 0, "Final flow", &trace);
8146 ds_put_cstr(ds, "Datapath actions: ");
8147 format_odp_actions(ds, trace.xout.odp_actions.data,
8148 trace.xout.odp_actions.size);
8150 if (trace.xout.slow) {
8151 ds_put_cstr(ds, "\nThis flow is handled by the userspace "
8152 "slow path because it:");
8153 switch (trace.xout.slow) {
8155 ds_put_cstr(ds, "\n\t- Consists of CFM packets.");
8158 ds_put_cstr(ds, "\n\t- Consists of LACP packets.");
8161 ds_put_cstr(ds, "\n\t- Consists of STP packets.");
8164 ds_put_cstr(ds, "\n\t- Consists of BFD packets.");
8166 case SLOW_CONTROLLER:
8167 ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages "
8168 "to the OpenFlow controller.");
8175 xlate_out_uninit(&trace.xout);
8180 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
8181 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
8184 unixctl_command_reply(conn, NULL);
8188 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
8189 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
8192 unixctl_command_reply(conn, NULL);
8195 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
8196 * 'reply' describing the results. */
8198 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
8200 struct facet *facet;
8204 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
8205 if (!facet_check_consistency(facet)) {
8210 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
8214 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
8215 ofproto->up.name, errors);
8217 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
8222 ofproto_dpif_self_check(struct unixctl_conn *conn,
8223 int argc, const char *argv[], void *aux OVS_UNUSED)
8225 struct ds reply = DS_EMPTY_INITIALIZER;
8226 struct ofproto_dpif *ofproto;
8229 ofproto = ofproto_dpif_lookup(argv[1]);
8231 unixctl_command_reply_error(conn, "Unknown ofproto (use "
8232 "ofproto/list for help)");
8235 ofproto_dpif_self_check__(ofproto, &reply);
8237 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
8238 ofproto_dpif_self_check__(ofproto, &reply);
8242 unixctl_command_reply(conn, ds_cstr(&reply));
8246 /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list
8247 * of the 'ofproto_shash' nodes. It is the responsibility of the caller
8248 * to destroy 'ofproto_shash' and free the returned value. */
8249 static const struct shash_node **
8250 get_ofprotos(struct shash *ofproto_shash)
8252 const struct ofproto_dpif *ofproto;
8254 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
8255 char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
8256 shash_add_nocopy(ofproto_shash, name, ofproto);
8259 return shash_sort(ofproto_shash);
8263 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
8264 const char *argv[] OVS_UNUSED,
8265 void *aux OVS_UNUSED)
8267 struct ds ds = DS_EMPTY_INITIALIZER;
8268 struct shash ofproto_shash;
8269 const struct shash_node **sorted_ofprotos;
8272 shash_init(&ofproto_shash);
8273 sorted_ofprotos = get_ofprotos(&ofproto_shash);
8274 for (i = 0; i < shash_count(&ofproto_shash); i++) {
8275 const struct shash_node *node = sorted_ofprotos[i];
8276 ds_put_format(&ds, "%s\n", node->name);
8279 shash_destroy(&ofproto_shash);
8280 free(sorted_ofprotos);
8282 unixctl_command_reply(conn, ds_cstr(&ds));
8287 show_dp_format(const struct ofproto_dpif *ofproto, struct ds *ds)
8289 const struct shash_node **ports;
8291 struct avg_subfacet_rates lifetime;
8292 unsigned long long int minutes;
8293 const int min_ms = 60 * 1000; /* milliseconds in one minute. */
8295 minutes = (time_msec() - ofproto->created) / min_ms;
8298 lifetime.add_rate = (double)ofproto->total_subfacet_add_count
8300 lifetime.del_rate = (double)ofproto->total_subfacet_del_count
8303 lifetime.add_rate = 0.0;
8304 lifetime.del_rate = 0.0;
8307 ds_put_format(ds, "%s (%s):\n", ofproto->up.name,
8308 dpif_name(ofproto->backer->dpif));
8310 "\tlookups: hit:%"PRIu64" missed:%"PRIu64"\n",
8311 ofproto->n_hit, ofproto->n_missed);
8312 ds_put_format(ds, "\tflows: cur: %zu, avg: %5.3f, max: %d,"
8313 " life span: %llu(ms)\n",
8314 hmap_count(&ofproto->subfacets),
8315 avg_subfacet_count(ofproto),
8316 ofproto->max_n_subfacet,
8317 avg_subfacet_life_span(ofproto));
8318 if (minutes >= 60) {
8319 show_dp_rates(ds, "\t\thourly avg:", &ofproto->hourly);
8321 if (minutes >= 60 * 24) {
8322 show_dp_rates(ds, "\t\tdaily avg:", &ofproto->daily);
8324 show_dp_rates(ds, "\t\toverall avg:", &lifetime);
8326 ports = shash_sort(&ofproto->up.port_by_name);
8327 for (i = 0; i < shash_count(&ofproto->up.port_by_name); i++) {
8328 const struct shash_node *node = ports[i];
8329 struct ofport *ofport = node->data;
8330 const char *name = netdev_get_name(ofport->netdev);
8331 const char *type = netdev_get_type(ofport->netdev);
8334 ds_put_format(ds, "\t%s %u/", name, ofport->ofp_port);
8336 odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
8337 if (odp_port != OVSP_NONE) {
8338 ds_put_format(ds, "%"PRIu32":", odp_port);
8340 ds_put_cstr(ds, "none:");
8343 if (strcmp(type, "system")) {
8344 struct netdev *netdev;
8347 ds_put_format(ds, " (%s", type);
8349 error = netdev_open(name, type, &netdev);
8354 error = netdev_get_config(netdev, &config);
8356 const struct smap_node **nodes;
8359 nodes = smap_sort(&config);
8360 for (i = 0; i < smap_count(&config); i++) {
8361 const struct smap_node *node = nodes[i];
8362 ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
8363 node->key, node->value);
8367 smap_destroy(&config);
8369 netdev_close(netdev);
8371 ds_put_char(ds, ')');
8373 ds_put_char(ds, '\n');
8379 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc,
8380 const char *argv[], void *aux OVS_UNUSED)
8382 struct ds ds = DS_EMPTY_INITIALIZER;
8383 const struct ofproto_dpif *ofproto;
8387 for (i = 1; i < argc; i++) {
8388 ofproto = ofproto_dpif_lookup(argv[i]);
8390 ds_put_format(&ds, "Unknown bridge %s (use dpif/dump-dps "
8391 "for help)", argv[i]);
8392 unixctl_command_reply_error(conn, ds_cstr(&ds));
8395 show_dp_format(ofproto, &ds);
8398 struct shash ofproto_shash;
8399 const struct shash_node **sorted_ofprotos;
8402 shash_init(&ofproto_shash);
8403 sorted_ofprotos = get_ofprotos(&ofproto_shash);
8404 for (i = 0; i < shash_count(&ofproto_shash); i++) {
8405 const struct shash_node *node = sorted_ofprotos[i];
8406 show_dp_format(node->data, &ds);
8409 shash_destroy(&ofproto_shash);
8410 free(sorted_ofprotos);
8413 unixctl_command_reply(conn, ds_cstr(&ds));
8418 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
8419 int argc OVS_UNUSED, const char *argv[],
8420 void *aux OVS_UNUSED)
8422 struct ds ds = DS_EMPTY_INITIALIZER;
8423 const struct ofproto_dpif *ofproto;
8424 struct subfacet *subfacet;
8426 ofproto = ofproto_dpif_lookup(argv[1]);
8428 unixctl_command_reply_error(conn, "no such bridge");
8432 update_stats(ofproto->backer);
8434 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
8435 struct facet *facet = subfacet->facet;
8437 odp_flow_key_format(subfacet->key, subfacet->key_len, &ds);
8439 ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:",
8440 subfacet->dp_packet_count, subfacet->dp_byte_count);
8441 if (subfacet->used) {
8442 ds_put_format(&ds, "%.3fs",
8443 (time_msec() - subfacet->used) / 1000.0);
8445 ds_put_format(&ds, "never");
8447 if (subfacet->facet->tcp_flags) {
8448 ds_put_cstr(&ds, ", flags:");
8449 packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags);
8452 ds_put_cstr(&ds, ", actions:");
8453 if (facet->xout.slow) {
8454 uint64_t slow_path_stub[128 / 8];
8455 const struct nlattr *actions;
8458 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
8459 slow_path_stub, sizeof slow_path_stub,
8460 &actions, &actions_len);
8461 format_odp_actions(&ds, actions, actions_len);
8463 format_odp_actions(&ds, facet->xout.odp_actions.data,
8464 facet->xout.odp_actions.size);
8466 ds_put_char(&ds, '\n');
8469 unixctl_command_reply(conn, ds_cstr(&ds));
8474 ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn,
8475 int argc OVS_UNUSED, const char *argv[],
8476 void *aux OVS_UNUSED)
8478 struct ds ds = DS_EMPTY_INITIALIZER;
8479 struct ofproto_dpif *ofproto;
8481 ofproto = ofproto_dpif_lookup(argv[1]);
8483 unixctl_command_reply_error(conn, "no such bridge");
8487 flush(&ofproto->up);
8489 unixctl_command_reply(conn, ds_cstr(&ds));
8494 ofproto_dpif_unixctl_init(void)
8496 static bool registered;
8502 unixctl_command_register(
8504 "[dp_name]|bridge odp_flow|br_flow [-generate|packet]",
8505 1, 3, ofproto_unixctl_trace, NULL);
8506 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
8507 ofproto_unixctl_fdb_flush, NULL);
8508 unixctl_command_register("fdb/show", "bridge", 1, 1,
8509 ofproto_unixctl_fdb_show, NULL);
8510 unixctl_command_register("ofproto/clog", "", 0, 0,
8511 ofproto_dpif_clog, NULL);
8512 unixctl_command_register("ofproto/unclog", "", 0, 0,
8513 ofproto_dpif_unclog, NULL);
8514 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
8515 ofproto_dpif_self_check, NULL);
8516 unixctl_command_register("dpif/dump-dps", "", 0, 0,
8517 ofproto_unixctl_dpif_dump_dps, NULL);
8518 unixctl_command_register("dpif/show", "[bridge]", 0, INT_MAX,
8519 ofproto_unixctl_dpif_show, NULL);
8520 unixctl_command_register("dpif/dump-flows", "bridge", 1, 1,
8521 ofproto_unixctl_dpif_dump_flows, NULL);
8522 unixctl_command_register("dpif/del-flows", "bridge", 1, 1,
8523 ofproto_unixctl_dpif_del_flows, NULL);
8526 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
8528 * This is deprecated. It is only for compatibility with broken device drivers
8529 * in old versions of Linux that do not properly support VLANs when VLAN
8530 * devices are not used. When broken device drivers are no longer in
8531 * widespread use, we will delete these interfaces. */
8534 set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid)
8536 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
8537 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
8539 if (realdev_ofp_port == ofport->realdev_ofp_port
8540 && vid == ofport->vlandev_vid) {
8544 ofproto->backer->need_revalidate = REV_RECONFIGURE;
8546 if (ofport->realdev_ofp_port) {
8549 if (realdev_ofp_port && ofport->bundle) {
8550 /* vlandevs are enslaved to their realdevs, so they are not allowed to
8551 * themselves be part of a bundle. */
8552 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
8555 ofport->realdev_ofp_port = realdev_ofp_port;
8556 ofport->vlandev_vid = vid;
8558 if (realdev_ofp_port) {
8559 vsp_add(ofport, realdev_ofp_port, vid);
8566 hash_realdev_vid(uint16_t realdev_ofp_port, int vid)
8568 return hash_2words(realdev_ofp_port, vid);
8571 /* Returns the OFP port number of the Linux VLAN device that corresponds to
8572 * 'vlan_tci' on the network device with port number 'realdev_ofp_port' in
8573 * 'struct ofport_dpif'. For example, given 'realdev_ofp_port' of eth0 and
8574 * 'vlan_tci' 9, it would return the port number of eth0.9.
8576 * Unless VLAN splinters are enabled for port 'realdev_ofp_port', this
8577 * function just returns its 'realdev_ofp_port' argument. */
8579 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
8580 uint16_t realdev_ofp_port, ovs_be16 vlan_tci)
8582 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
8583 int vid = vlan_tci_to_vid(vlan_tci);
8584 const struct vlan_splinter *vsp;
8586 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
8587 hash_realdev_vid(realdev_ofp_port, vid),
8588 &ofproto->realdev_vid_map) {
8589 if (vsp->realdev_ofp_port == realdev_ofp_port
8590 && vsp->vid == vid) {
8591 return vsp->vlandev_ofp_port;
8595 return realdev_ofp_port;
8598 static struct vlan_splinter *
8599 vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port)
8601 struct vlan_splinter *vsp;
8603 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, hash_int(vlandev_ofp_port, 0),
8604 &ofproto->vlandev_map) {
8605 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
8613 /* Returns the OpenFlow port number of the "real" device underlying the Linux
8614 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
8615 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
8616 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
8617 * eth0 and store 9 in '*vid'.
8619 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
8620 * VLAN device. Unless VLAN splinters are enabled, this is what this function
8623 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
8624 uint16_t vlandev_ofp_port, int *vid)
8626 if (!hmap_is_empty(&ofproto->vlandev_map)) {
8627 const struct vlan_splinter *vsp;
8629 vsp = vlandev_find(ofproto, vlandev_ofp_port);
8634 return vsp->realdev_ofp_port;
8640 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
8641 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
8642 * 'flow->in_port' to the "real" device backing the VLAN device, sets
8643 * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is
8644 * always the case unless VLAN splinters are enabled), returns false without
8645 * making any changes. */
8647 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
8652 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port, &vid);
8657 /* Cause the flow to be processed as if it came in on the real device with
8658 * the VLAN device's VLAN ID. */
8659 flow->in_port = realdev;
8660 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
8665 vsp_remove(struct ofport_dpif *port)
8667 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
8668 struct vlan_splinter *vsp;
8670 vsp = vlandev_find(ofproto, port->up.ofp_port);
8672 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
8673 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
8676 port->realdev_ofp_port = 0;
8678 VLOG_ERR("missing vlan device record");
8683 vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid)
8685 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
8687 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
8688 && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid))
8689 == realdev_ofp_port)) {
8690 struct vlan_splinter *vsp;
8692 vsp = xmalloc(sizeof *vsp);
8693 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
8694 hash_int(port->up.ofp_port, 0));
8695 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
8696 hash_realdev_vid(realdev_ofp_port, vid));
8697 vsp->realdev_ofp_port = realdev_ofp_port;
8698 vsp->vlandev_ofp_port = port->up.ofp_port;
8701 port->realdev_ofp_port = realdev_ofp_port;
8703 VLOG_ERR("duplicate vlan device record");
8708 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
8710 const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
8711 return ofport ? ofport->odp_port : OVSP_NONE;
8714 static struct ofport_dpif *
8715 odp_port_to_ofport(const struct dpif_backer *backer, uint32_t odp_port)
8717 struct ofport_dpif *port;
8719 HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node,
8720 hash_int(odp_port, 0),
8721 &backer->odp_to_ofport_map) {
8722 if (port->odp_port == odp_port) {
8731 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
8733 struct ofport_dpif *port;
8735 port = odp_port_to_ofport(ofproto->backer, odp_port);
8736 if (port && &ofproto->up == port->up.ofproto) {
8737 return port->up.ofp_port;
8742 static unsigned long long int
8743 avg_subfacet_life_span(const struct ofproto_dpif *ofproto)
8745 unsigned long long int dc;
8746 unsigned long long int avg;
8748 dc = ofproto->total_subfacet_del_count + ofproto->subfacet_del_count;
8749 avg = dc ? ofproto->total_subfacet_life_span / dc : 0;
8755 avg_subfacet_count(const struct ofproto_dpif *ofproto)
8759 if (ofproto->n_update_stats) {
8760 avg_c = (double)ofproto->total_subfacet_count
8761 / ofproto->n_update_stats;
8768 show_dp_rates(struct ds *ds, const char *heading,
8769 const struct avg_subfacet_rates *rates)
8771 ds_put_format(ds, "%s add rate: %5.3f/min, del rate: %5.3f/min\n",
8772 heading, rates->add_rate, rates->del_rate);
8776 update_max_subfacet_count(struct ofproto_dpif *ofproto)
8778 ofproto->max_n_subfacet = MAX(ofproto->max_n_subfacet,
8779 hmap_count(&ofproto->subfacets));
8782 /* Compute exponentially weighted moving average, adding 'new' as the newest,
8783 * most heavily weighted element. 'base' designates the rate of decay: after
8784 * 'base' further updates, 'new''s weight in the EWMA decays to about 1/e
8787 exp_mavg(double *avg, int base, double new)
8789 *avg = (*avg * (base - 1) + new) / base;
8793 update_moving_averages(struct ofproto_dpif *ofproto)
8795 const int min_ms = 60 * 1000; /* milliseconds in one minute. */
8797 /* Update hourly averages on the minute boundaries. */
8798 if (time_msec() - ofproto->last_minute >= min_ms) {
8799 exp_mavg(&ofproto->hourly.add_rate, 60, ofproto->subfacet_add_count);
8800 exp_mavg(&ofproto->hourly.del_rate, 60, ofproto->subfacet_del_count);
8802 /* Update daily averages on the hour boundaries. */
8803 if ((ofproto->last_minute - ofproto->created) / min_ms % 60 == 59) {
8804 exp_mavg(&ofproto->daily.add_rate, 24, ofproto->hourly.add_rate);
8805 exp_mavg(&ofproto->daily.del_rate, 24, ofproto->hourly.del_rate);
8808 ofproto->total_subfacet_add_count += ofproto->subfacet_add_count;
8809 ofproto->total_subfacet_del_count += ofproto->subfacet_del_count;
8810 ofproto->subfacet_add_count = 0;
8811 ofproto->subfacet_del_count = 0;
8812 ofproto->last_minute += min_ms;
8816 const struct ofproto_class ofproto_dpif_class = {
8851 port_is_lacp_current,
8852 NULL, /* rule_choose_table */
8859 rule_modify_actions,
8873 get_stp_port_status,
8880 is_mirror_output_bundle,
8881 forward_bpdu_changed,
8882 set_mac_table_config,