2 * Copyright (c) 2009, 2010, 2011, 2012, 2013 Nicira, Inc.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at:
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
19 #include "ofproto/ofproto-provider.h"
26 #include "byte-order.h"
31 #include "dynamic-string.h"
32 #include "fail-open.h"
36 #include "mac-learning.h"
37 #include "meta-flow.h"
38 #include "multipath.h"
39 #include "netdev-vport.h"
46 #include "ofp-actions.h"
47 #include "ofp-parse.h"
48 #include "ofp-print.h"
49 #include "ofproto-dpif-governor.h"
50 #include "ofproto-dpif-ipfix.h"
51 #include "ofproto-dpif-sflow.h"
52 #include "poll-loop.h"
57 #include "unaligned.h"
59 #include "vlan-bitmap.h"
62 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
64 COVERAGE_DEFINE(ofproto_dpif_expired);
65 COVERAGE_DEFINE(ofproto_dpif_xlate);
66 COVERAGE_DEFINE(facet_changed_rule);
67 COVERAGE_DEFINE(facet_revalidate);
68 COVERAGE_DEFINE(facet_unexpected);
69 COVERAGE_DEFINE(facet_suppress);
71 /* Maximum depth of flow table recursion (due to resubmit actions) in a
72 * flow translation. */
73 #define MAX_RESUBMIT_RECURSION 64
75 /* Number of implemented OpenFlow tables. */
76 enum { N_TABLES = 255 };
77 enum { TBL_INTERNAL = N_TABLES - 1 }; /* Used for internal hidden rules. */
78 BUILD_ASSERT_DECL(N_TABLES >= 2 && N_TABLES <= 255);
90 * - Do include packets and bytes from facets that have been deleted or
91 * whose own statistics have been folded into the rule.
93 * - Do include packets and bytes sent "by hand" that were accounted to
94 * the rule without any facet being involved (this is a rare corner
95 * case in rule_execute()).
97 * - Do not include packet or bytes that can be obtained from any facet's
98 * packet_count or byte_count member or that can be obtained from the
99 * datapath by, e.g., dpif_flow_get() for any subfacet.
101 uint64_t packet_count; /* Number of packets received. */
102 uint64_t byte_count; /* Number of bytes received. */
104 tag_type tag; /* Caches rule_calculate_tag() result. */
106 struct list facets; /* List of "struct facet"s. */
109 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
111 return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
114 static struct rule_dpif *rule_dpif_lookup(struct ofproto_dpif *,
115 const struct flow *);
116 static struct rule_dpif *rule_dpif_lookup__(struct ofproto_dpif *,
119 static struct rule_dpif *rule_dpif_miss_rule(struct ofproto_dpif *ofproto,
120 const struct flow *flow);
122 static void rule_get_stats(struct rule *, uint64_t *packets, uint64_t *bytes);
123 static void rule_credit_stats(struct rule_dpif *,
124 const struct dpif_flow_stats *);
125 static tag_type rule_calculate_tag(const struct flow *,
126 const struct minimask *, uint32_t basis);
127 static void rule_invalidate(const struct rule_dpif *);
129 #define MAX_MIRRORS 32
130 typedef uint32_t mirror_mask_t;
131 #define MIRROR_MASK_C(X) UINT32_C(X)
132 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
134 struct ofproto_dpif *ofproto; /* Owning ofproto. */
135 size_t idx; /* In ofproto's "mirrors" array. */
136 void *aux; /* Key supplied by ofproto's client. */
137 char *name; /* Identifier for log messages. */
139 /* Selection criteria. */
140 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
141 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
142 unsigned long *vlans; /* Bitmap of chosen VLANs, NULL selects all. */
144 /* Output (exactly one of out == NULL and out_vlan == -1 is true). */
145 struct ofbundle *out; /* Output port or NULL. */
146 int out_vlan; /* Output VLAN or -1. */
147 mirror_mask_t dup_mirrors; /* Bitmap of mirrors with the same output. */
150 int64_t packet_count; /* Number of packets sent. */
151 int64_t byte_count; /* Number of bytes sent. */
154 static void mirror_destroy(struct ofmirror *);
155 static void update_mirror_stats(struct ofproto_dpif *ofproto,
156 mirror_mask_t mirrors,
157 uint64_t packets, uint64_t bytes);
160 struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
161 struct ofproto_dpif *ofproto; /* Owning ofproto. */
162 void *aux; /* Key supplied by ofproto's client. */
163 char *name; /* Identifier for log messages. */
166 struct list ports; /* Contains "struct ofport"s. */
167 enum port_vlan_mode vlan_mode; /* VLAN mode */
168 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
169 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
170 * NULL if all VLANs are trunked. */
171 struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */
172 struct bond *bond; /* Nonnull iff more than one port. */
173 bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */
176 bool floodable; /* True if no port has OFPUTIL_PC_NO_FLOOD set. */
178 /* Port mirroring info. */
179 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
180 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
181 mirror_mask_t mirror_out; /* Mirrors that output to this bundle. */
184 static void bundle_remove(struct ofport *);
185 static void bundle_update(struct ofbundle *);
186 static void bundle_destroy(struct ofbundle *);
187 static void bundle_del_port(struct ofport_dpif *);
188 static void bundle_run(struct ofbundle *);
189 static void bundle_wait(struct ofbundle *);
190 static struct ofbundle *lookup_input_bundle(const struct ofproto_dpif *,
191 uint16_t in_port, bool warn,
192 struct ofport_dpif **in_ofportp);
194 /* A controller may use OFPP_NONE as the ingress port to indicate that
195 * it did not arrive on a "real" port. 'ofpp_none_bundle' exists for
196 * when an input bundle is needed for validation (e.g., mirroring or
197 * OFPP_NORMAL processing). It is not connected to an 'ofproto' or have
198 * any 'port' structs, so care must be taken when dealing with it. */
199 static struct ofbundle ofpp_none_bundle = {
201 .vlan_mode = PORT_VLAN_TRUNK
204 static void stp_run(struct ofproto_dpif *ofproto);
205 static void stp_wait(struct ofproto_dpif *ofproto);
206 static int set_stp_port(struct ofport *,
207 const struct ofproto_port_stp_settings *);
209 static bool ofbundle_includes_vlan(const struct ofbundle *, uint16_t vlan);
213 /* Initial values of fields of the packet that may be changed during
214 * flow processing and needed later. */
215 struct initial_vals {
216 /* This is the value of vlan_tci in the packet as actually received from
217 * dpif. This is the same as the facet's flow.vlan_tci unless the packet
218 * was received via a VLAN splinter. In that case, this value is 0
219 * (because the packet as actually received from the dpif had no 802.1Q
220 * tag) but the facet's flow.vlan_tci is set to the VLAN that the splinter
223 * This member should be removed when the VLAN splinters feature is no
229 tag_type tags; /* Tags associated with actions. */
230 enum slow_path_reason slow; /* 0 if fast path may be used. */
231 bool has_learn; /* Actions include NXAST_LEARN? */
232 bool has_normal; /* Actions output to OFPP_NORMAL? */
233 bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */
234 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
235 mirror_mask_t mirrors; /* Bitmap of associated mirrors. */
237 uint64_t odp_actions_stub[256 / 8];
238 struct ofpbuf odp_actions;
242 struct ofproto_dpif *ofproto;
244 /* Flow to which the OpenFlow actions apply. xlate_actions() will modify
245 * this flow when actions change header fields. */
248 struct initial_vals initial_vals;
250 /* The packet corresponding to 'flow', or a null pointer if we are
251 * revalidating without a packet to refer to. */
252 const struct ofpbuf *packet;
254 /* Should OFPP_NORMAL update the MAC learning table? Should "learn"
255 * actions update the flow table?
257 * We want to update these tables if we are actually processing a packet,
258 * or if we are accounting for packets that the datapath has processed, but
259 * not if we are just revalidating. */
262 /* The rule initiating translation or NULL. */
263 struct rule_dpif *rule;
265 /* The actions to translate. If 'rule' is not NULL, these may be NULL. */
266 const struct ofpact *ofpacts;
269 /* Union of the set of TCP flags seen so far in this flow. (Used only by
270 * NXAST_FIN_TIMEOUT. Set to zero to avoid updating updating rules'
274 /* If nonnull, flow translation calls this function just before executing a
275 * resubmit or OFPP_TABLE action. In addition, disables logging of traces
276 * when the recursion depth is exceeded.
278 * 'rule' is the rule being submitted into. It will be null if the
279 * resubmit or OFPP_TABLE action didn't find a matching rule.
281 * This is normally null so the client has to set it manually after
282 * calling xlate_in_init(). */
283 void (*resubmit_hook)(struct xlate_ctx *, struct rule_dpif *rule);
285 /* If nonnull, flow translation calls this function to report some
286 * significant decision, e.g. to explain why OFPP_NORMAL translation
287 * dropped a packet. */
288 void (*report_hook)(struct xlate_ctx *, const char *s);
290 /* If nonnull, flow translation credits the specified statistics to each
291 * rule reached through a resubmit or OFPP_TABLE action.
293 * This is normally null so the client has to set it manually after
294 * calling xlate_in_init(). */
295 const struct dpif_flow_stats *resubmit_stats;
298 /* Context used by xlate_actions() and its callees. */
300 struct xlate_in *xin;
301 struct xlate_out *xout;
303 struct ofproto_dpif *ofproto;
305 /* Flow at the last commit. */
306 struct flow base_flow;
308 /* Tunnel IP destination address as received. This is stored separately
309 * as the base_flow.tunnel is cleared on init to reflect the datapath
310 * behavior. Used to make sure not to send tunneled output to ourselves,
311 * which might lead to an infinite loop. This could happen easily
312 * if a tunnel is marked as 'ip_remote=flow', and the flow does not
313 * actually set the tun_dst field. */
314 ovs_be32 orig_tunnel_ip_dst;
316 /* Stack for the push and pop actions. Each stack element is of type
317 * "union mf_subvalue". */
318 union mf_subvalue init_stack[1024 / sizeof(union mf_subvalue)];
321 /* The rule that we are currently translating, or NULL. */
322 struct rule_dpif *rule;
324 int recurse; /* Recursion level, via xlate_table_action. */
325 bool max_resubmit_trigger; /* Recursed too deeply during translation. */
326 uint32_t orig_skb_priority; /* Priority when packet arrived. */
327 uint8_t table_id; /* OpenFlow table ID where flow was found. */
328 uint32_t sflow_n_outputs; /* Number of output ports. */
329 uint32_t sflow_odp_port; /* Output port for composing sFlow action. */
330 uint16_t user_cookie_offset;/* Used for user_action_cookie fixup. */
331 bool exit; /* No further actions should be processed. */
334 static void xlate_in_init(struct xlate_in *, struct ofproto_dpif *,
335 const struct flow *, const struct initial_vals *,
336 struct rule_dpif *, uint8_t tcp_flags,
337 const struct ofpbuf *);
339 static void xlate_out_uninit(struct xlate_out *);
341 static void xlate_actions(struct xlate_in *, struct xlate_out *);
343 static void xlate_actions_for_side_effects(struct xlate_in *);
345 static void xlate_table_action(struct xlate_ctx *, uint16_t in_port,
346 uint8_t table_id, bool may_packet_in);
348 static size_t put_userspace_action(const struct ofproto_dpif *,
349 struct ofpbuf *odp_actions,
351 const union user_action_cookie *,
354 static void compose_slow_path(const struct ofproto_dpif *, const struct flow *,
355 enum slow_path_reason,
356 uint64_t *stub, size_t stub_size,
357 const struct nlattr **actionsp,
358 size_t *actions_lenp);
360 static void xlate_report(struct xlate_ctx *ctx, const char *s);
362 /* A subfacet (see "struct subfacet" below) has three possible installation
365 * - SF_NOT_INSTALLED: Not installed in the datapath. This will only be the
366 * case just after the subfacet is created, just before the subfacet is
367 * destroyed, or if the datapath returns an error when we try to install a
370 * - SF_FAST_PATH: The subfacet's actions are installed in the datapath.
372 * - SF_SLOW_PATH: An action that sends every packet for the subfacet through
373 * ofproto_dpif is installed in the datapath.
376 SF_NOT_INSTALLED, /* No datapath flow for this subfacet. */
377 SF_FAST_PATH, /* Full actions are installed. */
378 SF_SLOW_PATH, /* Send-to-userspace action is installed. */
381 /* A dpif flow and actions associated with a facet.
383 * See also the large comment on struct facet. */
386 struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */
387 struct list list_node; /* In struct facet's 'facets' list. */
388 struct facet *facet; /* Owning facet. */
390 enum odp_key_fitness key_fitness;
394 long long int used; /* Time last used; time created if not used. */
395 long long int created; /* Time created. */
397 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
398 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
400 enum subfacet_path path; /* Installed in datapath? */
403 #define SUBFACET_DESTROY_MAX_BATCH 50
405 static struct subfacet *subfacet_create(struct facet *, struct flow_miss *miss,
407 static struct subfacet *subfacet_find(struct ofproto_dpif *,
408 const struct nlattr *key, size_t key_len,
410 static void subfacet_destroy(struct subfacet *);
411 static void subfacet_destroy__(struct subfacet *);
412 static void subfacet_destroy_batch(struct ofproto_dpif *,
413 struct subfacet **, int n);
414 static void subfacet_reset_dp_stats(struct subfacet *,
415 struct dpif_flow_stats *);
416 static void subfacet_update_stats(struct subfacet *,
417 const struct dpif_flow_stats *);
418 static int subfacet_install(struct subfacet *,
419 const struct ofpbuf *odp_actions,
420 struct dpif_flow_stats *);
421 static void subfacet_uninstall(struct subfacet *);
423 /* An exact-match instantiation of an OpenFlow flow.
425 * A facet associates a "struct flow", which represents the Open vSwitch
426 * userspace idea of an exact-match flow, with one or more subfacets. Each
427 * subfacet tracks the datapath's idea of the exact-match flow equivalent to
428 * the facet. When the kernel module (or other dpif implementation) and Open
429 * vSwitch userspace agree on the definition of a flow key, there is exactly
430 * one subfacet per facet. If the dpif implementation supports more-specific
431 * flow matching than userspace, however, a facet can have more than one
432 * subfacet, each of which corresponds to some distinction in flow that
433 * userspace simply doesn't understand.
435 * Flow expiration works in terms of subfacets, so a facet must have at least
436 * one subfacet or it will never expire, leaking memory. */
439 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
440 struct list list_node; /* In owning rule's 'facets' list. */
441 struct rule_dpif *rule; /* Owning rule. */
444 struct list subfacets;
445 long long int used; /* Time last used; time created if not used. */
452 * - Do include packets and bytes sent "by hand", e.g. with
455 * - Do include packets and bytes that were obtained from the datapath
456 * when a subfacet's statistics were reset (e.g. dpif_flow_put() with
457 * DPIF_FP_ZERO_STATS).
459 * - Do not include packets or bytes that can be obtained from the
460 * datapath for any existing subfacet.
462 uint64_t packet_count; /* Number of packets received. */
463 uint64_t byte_count; /* Number of bytes received. */
465 /* Resubmit statistics. */
466 uint64_t prev_packet_count; /* Number of packets from last stats push. */
467 uint64_t prev_byte_count; /* Number of bytes from last stats push. */
468 long long int prev_used; /* Used time from last stats push. */
471 uint64_t accounted_bytes; /* Bytes processed by facet_account(). */
472 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
473 uint8_t tcp_flags; /* TCP flags seen for this 'rule'. */
475 struct xlate_out xout;
477 /* Initial values of the packet that may be needed later. */
478 struct initial_vals initial_vals;
480 /* Storage for a single subfacet, to reduce malloc() time and space
481 * overhead. (A facet always has at least one subfacet and in the common
482 * case has exactly one subfacet. However, 'one_subfacet' may not
483 * always be valid, since it could have been removed after newer
484 * subfacets were pushed onto the 'subfacets' list.) */
485 struct subfacet one_subfacet;
487 long long int learn_rl; /* Rate limiter for facet_learn(). */
490 static struct facet *facet_create(const struct flow_miss *, uint32_t hash);
491 static void facet_remove(struct facet *);
492 static void facet_free(struct facet *);
494 static struct facet *facet_find(struct ofproto_dpif *,
495 const struct flow *, uint32_t hash);
496 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
497 const struct flow *, uint32_t hash);
498 static bool facet_revalidate(struct facet *);
499 static bool facet_check_consistency(struct facet *);
501 static void facet_flush_stats(struct facet *);
503 static void facet_reset_counters(struct facet *);
504 static void facet_push_stats(struct facet *, bool may_learn);
505 static void facet_learn(struct facet *);
506 static void facet_account(struct facet *);
507 static void push_all_stats(void);
509 static bool facet_is_controller_flow(struct facet *);
512 struct hmap_node odp_port_node; /* In dpif_backer's "odp_to_ofport_map". */
516 struct ofbundle *bundle; /* Bundle that contains this port, if any. */
517 struct list bundle_node; /* In struct ofbundle's "ports" list. */
518 struct cfm *cfm; /* Connectivity Fault Management, if any. */
519 struct bfd *bfd; /* BFD, if any. */
520 tag_type tag; /* Tag associated with this port. */
521 bool may_enable; /* May be enabled in bonds. */
522 long long int carrier_seq; /* Carrier status changes. */
523 struct tnl_port *tnl_port; /* Tunnel handle, or null. */
526 struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */
527 enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */
528 long long int stp_state_entered;
530 struct hmap priorities; /* Map of attached 'priority_to_dscp's. */
532 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
534 * This is deprecated. It is only for compatibility with broken device
535 * drivers in old versions of Linux that do not properly support VLANs when
536 * VLAN devices are not used. When broken device drivers are no longer in
537 * widespread use, we will delete these interfaces. */
538 uint16_t realdev_ofp_port;
542 /* Node in 'ofport_dpif''s 'priorities' map. Used to maintain a map from
543 * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
544 * traffic egressing the 'ofport' with that priority should be marked with. */
545 struct priority_to_dscp {
546 struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'priorities' map. */
547 uint32_t priority; /* Priority of this queue (see struct flow). */
549 uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */
552 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
554 * This is deprecated. It is only for compatibility with broken device drivers
555 * in old versions of Linux that do not properly support VLANs when VLAN
556 * devices are not used. When broken device drivers are no longer in
557 * widespread use, we will delete these interfaces. */
558 struct vlan_splinter {
559 struct hmap_node realdev_vid_node;
560 struct hmap_node vlandev_node;
561 uint16_t realdev_ofp_port;
562 uint16_t vlandev_ofp_port;
566 static uint32_t vsp_realdev_to_vlandev(const struct ofproto_dpif *,
567 uint32_t realdev, ovs_be16 vlan_tci);
568 static bool vsp_adjust_flow(const struct ofproto_dpif *, struct flow *);
569 static void vsp_remove(struct ofport_dpif *);
570 static void vsp_add(struct ofport_dpif *, uint16_t realdev_ofp_port, int vid);
572 static uint32_t ofp_port_to_odp_port(const struct ofproto_dpif *,
574 static uint16_t odp_port_to_ofp_port(const struct ofproto_dpif *,
577 static struct ofport_dpif *
578 ofport_dpif_cast(const struct ofport *ofport)
580 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
583 static void port_run(struct ofport_dpif *);
584 static void port_run_fast(struct ofport_dpif *);
585 static void port_wait(struct ofport_dpif *);
586 static int set_bfd(struct ofport *, const struct smap *);
587 static int set_cfm(struct ofport *, const struct cfm_settings *);
588 static void ofport_clear_priorities(struct ofport_dpif *);
589 static void run_fast_rl(void);
591 struct dpif_completion {
592 struct list list_node;
593 struct ofoperation *op;
596 /* Extra information about a classifier table.
597 * Currently used just for optimized flow revalidation. */
599 /* If either of these is nonnull, then this table has a form that allows
600 * flows to be tagged to avoid revalidating most flows for the most common
601 * kinds of flow table changes. */
602 struct cls_table *catchall_table; /* Table that wildcards all fields. */
603 struct cls_table *other_table; /* Table with any other wildcard set. */
604 uint32_t basis; /* Keeps each table's tags separate. */
607 /* Reasons that we might need to revalidate every facet, and corresponding
610 * A value of 0 means that there is no need to revalidate.
612 * It would be nice to have some cleaner way to integrate with coverage
613 * counters, but with only a few reasons I guess this is good enough for
615 enum revalidate_reason {
616 REV_RECONFIGURE = 1, /* Switch configuration changed. */
617 REV_STP, /* Spanning tree protocol port status change. */
618 REV_PORT_TOGGLED, /* Port enabled or disabled by CFM, LACP, ...*/
619 REV_FLOW_TABLE, /* Flow table changed. */
620 REV_INCONSISTENCY /* Facet self-check failed. */
622 COVERAGE_DEFINE(rev_reconfigure);
623 COVERAGE_DEFINE(rev_stp);
624 COVERAGE_DEFINE(rev_port_toggled);
625 COVERAGE_DEFINE(rev_flow_table);
626 COVERAGE_DEFINE(rev_inconsistency);
628 /* Drop keys are odp flow keys which have drop flows installed in the kernel.
629 * These are datapath flows which have no associated ofproto, if they did we
630 * would use facets. */
632 struct hmap_node hmap_node;
637 /* All datapaths of a given type share a single dpif backer instance. */
642 struct timer next_expiration;
643 struct hmap odp_to_ofport_map; /* ODP port to ofport mapping. */
645 struct simap tnl_backers; /* Set of dpif ports backing tunnels. */
647 /* Facet revalidation flags applying to facets which use this backer. */
648 enum revalidate_reason need_revalidate; /* Revalidate every facet. */
649 struct tag_set revalidate_set; /* Revalidate only matching facets. */
651 struct hmap drop_keys; /* Set of dropped odp keys. */
654 /* All existing ofproto_backer instances, indexed by ofproto->up.type. */
655 static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers);
657 static void drop_key_clear(struct dpif_backer *);
658 static struct ofport_dpif *
659 odp_port_to_ofport(const struct dpif_backer *, uint32_t odp_port);
661 struct avg_subfacet_rates {
662 double add_rate; /* Moving average of new flows created per minute. */
663 double del_rate; /* Moving average of flows deleted per minute. */
665 static void show_dp_rates(struct ds *ds, const char *heading,
666 const struct avg_subfacet_rates *rates);
667 static void exp_mavg(double *avg, int base, double new);
669 struct ofproto_dpif {
670 struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
672 struct dpif_backer *backer;
674 /* Special OpenFlow rules. */
675 struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */
676 struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */
679 struct netflow *netflow;
680 struct dpif_sflow *sflow;
681 struct dpif_ipfix *ipfix;
682 struct hmap bundles; /* Contains "struct ofbundle"s. */
683 struct mac_learning *ml;
684 struct ofmirror *mirrors[MAX_MIRRORS];
686 bool has_bonded_bundles;
690 struct hmap subfacets;
691 struct governor *governor;
692 long long int consistency_rl;
695 struct table_dpif tables[N_TABLES];
697 /* Support for debugging async flow mods. */
698 struct list completions;
700 bool has_bundle_action; /* True when the first bundle action appears. */
701 struct netdev_stats stats; /* To account packets generated and consumed in
706 long long int stp_last_tick;
708 /* VLAN splinters. */
709 struct hmap realdev_vid_map; /* (realdev,vid) -> vlandev. */
710 struct hmap vlandev_map; /* vlandev -> (realdev,vid). */
713 struct sset ports; /* Set of standard port names. */
714 struct sset ghost_ports; /* Ports with no datapath port. */
715 struct sset port_poll_set; /* Queued names for port_poll() reply. */
716 int port_poll_errno; /* Last errno for port_poll() reply. */
718 /* Per ofproto's dpif stats. */
722 /* Subfacet statistics.
724 * These keep track of the total number of subfacets added and deleted and
725 * flow life span. They are useful for computing the flow rates stats
726 * exposed via "ovs-appctl dpif/show". The goal is to learn about
727 * traffic patterns in ways that we can use later to improve Open vSwitch
728 * performance in new situations. */
729 long long int created; /* Time when it is created. */
730 unsigned int max_n_subfacet; /* Maximum number of flows */
732 /* The average number of subfacets... */
733 struct avg_subfacet_rates hourly; /* ...over the last hour. */
734 struct avg_subfacet_rates daily; /* ...over the last day. */
735 long long int last_minute; /* Last time 'hourly' was updated. */
737 /* Number of subfacets added or deleted since 'last_minute'. */
738 unsigned int subfacet_add_count;
739 unsigned int subfacet_del_count;
741 /* Number of subfacets added or deleted from 'created' to 'last_minute.' */
742 unsigned long long int total_subfacet_add_count;
743 unsigned long long int total_subfacet_del_count;
745 /* Sum of the number of milliseconds that each subfacet existed,
746 * over the subfacets that have been added and then later deleted. */
747 unsigned long long int total_subfacet_life_span;
749 /* Incremented by the number of currently existing subfacets, each
750 * time we pull statistics from the kernel. */
751 unsigned long long int total_subfacet_count;
753 /* Number of times we pull statistics from the kernel. */
754 unsigned long long int n_update_stats;
756 static unsigned long long int avg_subfacet_life_span(
757 const struct ofproto_dpif *);
758 static double avg_subfacet_count(const struct ofproto_dpif *ofproto);
759 static void update_moving_averages(struct ofproto_dpif *ofproto);
760 static void update_max_subfacet_count(struct ofproto_dpif *ofproto);
762 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
763 * for debugging the asynchronous flow_mod implementation.) */
766 /* All existing ofproto_dpif instances, indexed by ->up.name. */
767 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
769 static void ofproto_dpif_unixctl_init(void);
771 static struct ofproto_dpif *
772 ofproto_dpif_cast(const struct ofproto *ofproto)
774 ovs_assert(ofproto->ofproto_class == &ofproto_dpif_class);
775 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
778 static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *,
780 static struct ofport_dpif *get_odp_port(const struct ofproto_dpif *,
782 static void ofproto_trace(struct ofproto_dpif *, const struct flow *,
783 const struct ofpbuf *,
784 const struct initial_vals *, struct ds *);
786 /* Packet processing. */
787 static void update_learning_table(struct ofproto_dpif *,
788 const struct flow *, int vlan,
791 #define FLOW_MISS_MAX_BATCH 50
792 static int handle_upcalls(struct dpif_backer *, unsigned int max_batch);
794 /* Flow expiration. */
795 static int expire(struct dpif_backer *);
798 static void send_netflow_active_timeouts(struct ofproto_dpif *);
801 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
802 static size_t compose_sflow_action(const struct ofproto_dpif *,
803 struct ofpbuf *odp_actions,
804 const struct flow *, uint32_t odp_port);
805 static void compose_ipfix_action(const struct ofproto_dpif *,
806 struct ofpbuf *odp_actions,
807 const struct flow *);
808 static void add_mirror_actions(struct xlate_ctx *ctx,
809 const struct flow *flow);
810 /* Global variables. */
811 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
813 /* Initial mappings of port to bridge mappings. */
814 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
816 /* Factory functions. */
819 init(const struct shash *iface_hints)
821 struct shash_node *node;
823 /* Make a local copy, since we don't own 'iface_hints' elements. */
824 SHASH_FOR_EACH(node, iface_hints) {
825 const struct iface_hint *orig_hint = node->data;
826 struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
828 new_hint->br_name = xstrdup(orig_hint->br_name);
829 new_hint->br_type = xstrdup(orig_hint->br_type);
830 new_hint->ofp_port = orig_hint->ofp_port;
832 shash_add(&init_ofp_ports, node->name, new_hint);
837 enumerate_types(struct sset *types)
839 dp_enumerate_types(types);
843 enumerate_names(const char *type, struct sset *names)
845 struct ofproto_dpif *ofproto;
848 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
849 if (strcmp(type, ofproto->up.type)) {
852 sset_add(names, ofproto->up.name);
859 del(const char *type, const char *name)
864 error = dpif_open(name, type, &dpif);
866 error = dpif_delete(dpif);
873 port_open_type(const char *datapath_type, const char *port_type)
875 return dpif_port_open_type(datapath_type, port_type);
878 /* Type functions. */
880 static struct ofproto_dpif *
881 lookup_ofproto_dpif_by_port_name(const char *name)
883 struct ofproto_dpif *ofproto;
885 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
886 if (sset_contains(&ofproto->ports, name)) {
895 type_run(const char *type)
897 static long long int push_timer = LLONG_MIN;
898 struct dpif_backer *backer;
902 backer = shash_find_data(&all_dpif_backers, type);
904 /* This is not necessarily a problem, since backers are only
905 * created on demand. */
909 dpif_run(backer->dpif);
911 /* The most natural place to push facet statistics is when they're pulled
912 * from the datapath. However, when there are many flows in the datapath,
913 * this expensive operation can occur so frequently, that it reduces our
914 * ability to quickly set up flows. To reduce the cost, we push statistics
916 if (time_msec() > push_timer) {
917 push_timer = time_msec() + 2000;
921 if (backer->need_revalidate
922 || !tag_set_is_empty(&backer->revalidate_set)) {
923 struct tag_set revalidate_set = backer->revalidate_set;
924 bool need_revalidate = backer->need_revalidate;
925 struct ofproto_dpif *ofproto;
926 struct simap_node *node;
927 struct simap tmp_backers;
929 /* Handle tunnel garbage collection. */
930 simap_init(&tmp_backers);
931 simap_swap(&backer->tnl_backers, &tmp_backers);
933 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
934 struct ofport_dpif *iter;
936 if (backer != ofproto->backer) {
940 HMAP_FOR_EACH (iter, up.hmap_node, &ofproto->up.ports) {
943 if (!iter->tnl_port) {
947 dp_port = netdev_vport_get_dpif_port(iter->up.netdev);
948 node = simap_find(&tmp_backers, dp_port);
950 simap_put(&backer->tnl_backers, dp_port, node->data);
951 simap_delete(&tmp_backers, node);
952 node = simap_find(&backer->tnl_backers, dp_port);
954 node = simap_find(&backer->tnl_backers, dp_port);
956 uint32_t odp_port = UINT32_MAX;
958 if (!dpif_port_add(backer->dpif, iter->up.netdev,
960 simap_put(&backer->tnl_backers, dp_port, odp_port);
961 node = simap_find(&backer->tnl_backers, dp_port);
966 iter->odp_port = node ? node->data : OVSP_NONE;
967 if (tnl_port_reconfigure(&iter->up, iter->odp_port,
969 backer->need_revalidate = REV_RECONFIGURE;
974 SIMAP_FOR_EACH (node, &tmp_backers) {
975 dpif_port_del(backer->dpif, node->data);
977 simap_destroy(&tmp_backers);
979 switch (backer->need_revalidate) {
980 case REV_RECONFIGURE: COVERAGE_INC(rev_reconfigure); break;
981 case REV_STP: COVERAGE_INC(rev_stp); break;
982 case REV_PORT_TOGGLED: COVERAGE_INC(rev_port_toggled); break;
983 case REV_FLOW_TABLE: COVERAGE_INC(rev_flow_table); break;
984 case REV_INCONSISTENCY: COVERAGE_INC(rev_inconsistency); break;
987 if (backer->need_revalidate) {
988 /* Clear the drop_keys in case we should now be accepting some
989 * formerly dropped flows. */
990 drop_key_clear(backer);
993 /* Clear the revalidation flags. */
994 tag_set_init(&backer->revalidate_set);
995 backer->need_revalidate = 0;
997 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
998 struct facet *facet, *next;
1000 if (ofproto->backer != backer) {
1004 HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &ofproto->facets) {
1006 || tag_set_intersects(&revalidate_set, facet->xout.tags)) {
1007 facet_revalidate(facet);
1014 if (timer_expired(&backer->next_expiration)) {
1015 int delay = expire(backer);
1016 timer_set_duration(&backer->next_expiration, delay);
1019 /* Check for port changes in the dpif. */
1020 while ((error = dpif_port_poll(backer->dpif, &devname)) == 0) {
1021 struct ofproto_dpif *ofproto;
1022 struct dpif_port port;
1024 /* Don't report on the datapath's device. */
1025 if (!strcmp(devname, dpif_base_name(backer->dpif))) {
1029 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
1030 &all_ofproto_dpifs) {
1031 if (simap_contains(&ofproto->backer->tnl_backers, devname)) {
1036 ofproto = lookup_ofproto_dpif_by_port_name(devname);
1037 if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
1038 /* The port was removed. If we know the datapath,
1039 * report it through poll_set(). If we don't, it may be
1040 * notifying us of a removal we initiated, so ignore it.
1041 * If there's a pending ENOBUFS, let it stand, since
1042 * everything will be reevaluated. */
1043 if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
1044 sset_add(&ofproto->port_poll_set, devname);
1045 ofproto->port_poll_errno = 0;
1047 } else if (!ofproto) {
1048 /* The port was added, but we don't know with which
1049 * ofproto we should associate it. Delete it. */
1050 dpif_port_del(backer->dpif, port.port_no);
1052 dpif_port_destroy(&port);
1058 if (error != EAGAIN) {
1059 struct ofproto_dpif *ofproto;
1061 /* There was some sort of error, so propagate it to all
1062 * ofprotos that use this backer. */
1063 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
1064 &all_ofproto_dpifs) {
1065 if (ofproto->backer == backer) {
1066 sset_clear(&ofproto->port_poll_set);
1067 ofproto->port_poll_errno = error;
1076 dpif_backer_run_fast(struct dpif_backer *backer, int max_batch)
1080 /* Handle one or more batches of upcalls, until there's nothing left to do
1081 * or until we do a fixed total amount of work.
1083 * We do work in batches because it can be much cheaper to set up a number
1084 * of flows and fire off their patches all at once. We do multiple batches
1085 * because in some cases handling a packet can cause another packet to be
1086 * queued almost immediately as part of the return flow. Both
1087 * optimizations can make major improvements on some benchmarks and
1088 * presumably for real traffic as well. */
1090 while (work < max_batch) {
1091 int retval = handle_upcalls(backer, max_batch - work);
1102 type_run_fast(const char *type)
1104 struct dpif_backer *backer;
1106 backer = shash_find_data(&all_dpif_backers, type);
1108 /* This is not necessarily a problem, since backers are only
1109 * created on demand. */
1113 return dpif_backer_run_fast(backer, FLOW_MISS_MAX_BATCH);
1119 static long long int port_rl = LLONG_MIN;
1120 static unsigned int backer_rl = 0;
1122 if (time_msec() >= port_rl) {
1123 struct ofproto_dpif *ofproto;
1124 struct ofport_dpif *ofport;
1126 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1128 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1129 port_run_fast(ofport);
1132 port_rl = time_msec() + 200;
1135 /* XXX: We have to be careful not to do too much work in this function. If
1136 * we call dpif_backer_run_fast() too often, or with too large a batch,
1137 * performance improves signifcantly, but at a cost. It's possible for the
1138 * number of flows in the datapath to increase without bound, and for poll
1139 * loops to take 10s of seconds. The correct solution to this problem,
1140 * long term, is to separate flow miss handling into it's own thread so it
1141 * isn't affected by revalidations, and expirations. Until then, this is
1142 * the best we can do. */
1143 if (++backer_rl >= 10) {
1144 struct shash_node *node;
1147 SHASH_FOR_EACH (node, &all_dpif_backers) {
1148 dpif_backer_run_fast(node->data, 1);
1154 type_wait(const char *type)
1156 struct dpif_backer *backer;
1158 backer = shash_find_data(&all_dpif_backers, type);
1160 /* This is not necessarily a problem, since backers are only
1161 * created on demand. */
1165 timer_wait(&backer->next_expiration);
1168 /* Basic life-cycle. */
1170 static int add_internal_flows(struct ofproto_dpif *);
1172 static struct ofproto *
1175 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
1176 return &ofproto->up;
1180 dealloc(struct ofproto *ofproto_)
1182 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1187 close_dpif_backer(struct dpif_backer *backer)
1189 struct shash_node *node;
1191 ovs_assert(backer->refcount > 0);
1193 if (--backer->refcount) {
1197 drop_key_clear(backer);
1198 hmap_destroy(&backer->drop_keys);
1200 simap_destroy(&backer->tnl_backers);
1201 hmap_destroy(&backer->odp_to_ofport_map);
1202 node = shash_find(&all_dpif_backers, backer->type);
1204 shash_delete(&all_dpif_backers, node);
1205 dpif_close(backer->dpif);
1210 /* Datapath port slated for removal from datapath. */
1211 struct odp_garbage {
1212 struct list list_node;
1217 open_dpif_backer(const char *type, struct dpif_backer **backerp)
1219 struct dpif_backer *backer;
1220 struct dpif_port_dump port_dump;
1221 struct dpif_port port;
1222 struct shash_node *node;
1223 struct list garbage_list;
1224 struct odp_garbage *garbage, *next;
1230 backer = shash_find_data(&all_dpif_backers, type);
1237 backer_name = xasprintf("ovs-%s", type);
1239 /* Remove any existing datapaths, since we assume we're the only
1240 * userspace controlling the datapath. */
1242 dp_enumerate_names(type, &names);
1243 SSET_FOR_EACH(name, &names) {
1244 struct dpif *old_dpif;
1246 /* Don't remove our backer if it exists. */
1247 if (!strcmp(name, backer_name)) {
1251 if (dpif_open(name, type, &old_dpif)) {
1252 VLOG_WARN("couldn't open old datapath %s to remove it", name);
1254 dpif_delete(old_dpif);
1255 dpif_close(old_dpif);
1258 sset_destroy(&names);
1260 backer = xmalloc(sizeof *backer);
1262 error = dpif_create_and_open(backer_name, type, &backer->dpif);
1265 VLOG_ERR("failed to open datapath of type %s: %s", type,
1271 backer->type = xstrdup(type);
1272 backer->refcount = 1;
1273 hmap_init(&backer->odp_to_ofport_map);
1274 hmap_init(&backer->drop_keys);
1275 timer_set_duration(&backer->next_expiration, 1000);
1276 backer->need_revalidate = 0;
1277 simap_init(&backer->tnl_backers);
1278 tag_set_init(&backer->revalidate_set);
1281 dpif_flow_flush(backer->dpif);
1283 /* Loop through the ports already on the datapath and remove any
1284 * that we don't need anymore. */
1285 list_init(&garbage_list);
1286 dpif_port_dump_start(&port_dump, backer->dpif);
1287 while (dpif_port_dump_next(&port_dump, &port)) {
1288 node = shash_find(&init_ofp_ports, port.name);
1289 if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
1290 garbage = xmalloc(sizeof *garbage);
1291 garbage->odp_port = port.port_no;
1292 list_push_front(&garbage_list, &garbage->list_node);
1295 dpif_port_dump_done(&port_dump);
1297 LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) {
1298 dpif_port_del(backer->dpif, garbage->odp_port);
1299 list_remove(&garbage->list_node);
1303 shash_add(&all_dpif_backers, type, backer);
1305 error = dpif_recv_set(backer->dpif, true);
1307 VLOG_ERR("failed to listen on datapath of type %s: %s",
1308 type, strerror(error));
1309 close_dpif_backer(backer);
1317 construct(struct ofproto *ofproto_)
1319 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1320 struct shash_node *node, *next;
1325 error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1330 max_ports = dpif_get_max_ports(ofproto->backer->dpif);
1331 ofproto_init_max_ports(ofproto_, MIN(max_ports, OFPP_MAX));
1333 ofproto->netflow = NULL;
1334 ofproto->sflow = NULL;
1335 ofproto->ipfix = NULL;
1336 ofproto->stp = NULL;
1337 hmap_init(&ofproto->bundles);
1338 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1339 for (i = 0; i < MAX_MIRRORS; i++) {
1340 ofproto->mirrors[i] = NULL;
1342 ofproto->has_bonded_bundles = false;
1344 hmap_init(&ofproto->facets);
1345 hmap_init(&ofproto->subfacets);
1346 ofproto->governor = NULL;
1347 ofproto->consistency_rl = LLONG_MIN;
1349 for (i = 0; i < N_TABLES; i++) {
1350 struct table_dpif *table = &ofproto->tables[i];
1352 table->catchall_table = NULL;
1353 table->other_table = NULL;
1354 table->basis = random_uint32();
1357 list_init(&ofproto->completions);
1359 ofproto_dpif_unixctl_init();
1361 ofproto->has_mirrors = false;
1362 ofproto->has_bundle_action = false;
1364 hmap_init(&ofproto->vlandev_map);
1365 hmap_init(&ofproto->realdev_vid_map);
1367 sset_init(&ofproto->ports);
1368 sset_init(&ofproto->ghost_ports);
1369 sset_init(&ofproto->port_poll_set);
1370 ofproto->port_poll_errno = 0;
1372 SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1373 struct iface_hint *iface_hint = node->data;
1375 if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1376 /* Check if the datapath already has this port. */
1377 if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1378 sset_add(&ofproto->ports, node->name);
1381 free(iface_hint->br_name);
1382 free(iface_hint->br_type);
1384 shash_delete(&init_ofp_ports, node);
1388 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1389 hash_string(ofproto->up.name, 0));
1390 memset(&ofproto->stats, 0, sizeof ofproto->stats);
1392 ofproto_init_tables(ofproto_, N_TABLES);
1393 error = add_internal_flows(ofproto);
1394 ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1397 ofproto->n_missed = 0;
1399 ofproto->max_n_subfacet = 0;
1400 ofproto->created = time_msec();
1401 ofproto->last_minute = ofproto->created;
1402 memset(&ofproto->hourly, 0, sizeof ofproto->hourly);
1403 memset(&ofproto->daily, 0, sizeof ofproto->daily);
1404 ofproto->subfacet_add_count = 0;
1405 ofproto->subfacet_del_count = 0;
1406 ofproto->total_subfacet_add_count = 0;
1407 ofproto->total_subfacet_del_count = 0;
1408 ofproto->total_subfacet_life_span = 0;
1409 ofproto->total_subfacet_count = 0;
1410 ofproto->n_update_stats = 0;
1416 add_internal_flow(struct ofproto_dpif *ofproto, int id,
1417 const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1419 struct ofputil_flow_mod fm;
1422 match_init_catchall(&fm.match);
1424 match_set_reg(&fm.match, 0, id);
1425 fm.new_cookie = htonll(0);
1426 fm.cookie = htonll(0);
1427 fm.cookie_mask = htonll(0);
1428 fm.table_id = TBL_INTERNAL;
1429 fm.command = OFPFC_ADD;
1430 fm.idle_timeout = 0;
1431 fm.hard_timeout = 0;
1435 fm.ofpacts = ofpacts->data;
1436 fm.ofpacts_len = ofpacts->size;
1438 error = ofproto_flow_mod(&ofproto->up, &fm);
1440 VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
1441 id, ofperr_to_string(error));
1445 *rulep = rule_dpif_lookup__(ofproto, &fm.match.flow, TBL_INTERNAL);
1446 ovs_assert(*rulep != NULL);
1452 add_internal_flows(struct ofproto_dpif *ofproto)
1454 struct ofpact_controller *controller;
1455 uint64_t ofpacts_stub[128 / 8];
1456 struct ofpbuf ofpacts;
1460 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1463 controller = ofpact_put_CONTROLLER(&ofpacts);
1464 controller->max_len = UINT16_MAX;
1465 controller->controller_id = 0;
1466 controller->reason = OFPR_NO_MATCH;
1467 ofpact_pad(&ofpacts);
1469 error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule);
1474 ofpbuf_clear(&ofpacts);
1475 error = add_internal_flow(ofproto, id++, &ofpacts,
1476 &ofproto->no_packet_in_rule);
1481 complete_operations(struct ofproto_dpif *ofproto)
1483 struct dpif_completion *c, *next;
1485 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
1486 ofoperation_complete(c->op, 0);
1487 list_remove(&c->list_node);
1493 destruct(struct ofproto *ofproto_)
1495 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1496 struct rule_dpif *rule, *next_rule;
1497 struct oftable *table;
1500 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1501 complete_operations(ofproto);
1503 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1504 struct cls_cursor cursor;
1506 cls_cursor_init(&cursor, &table->cls, NULL);
1507 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1508 ofproto_rule_destroy(&rule->up);
1512 for (i = 0; i < MAX_MIRRORS; i++) {
1513 mirror_destroy(ofproto->mirrors[i]);
1516 netflow_destroy(ofproto->netflow);
1517 dpif_sflow_destroy(ofproto->sflow);
1518 hmap_destroy(&ofproto->bundles);
1519 mac_learning_destroy(ofproto->ml);
1521 hmap_destroy(&ofproto->facets);
1522 hmap_destroy(&ofproto->subfacets);
1523 governor_destroy(ofproto->governor);
1525 hmap_destroy(&ofproto->vlandev_map);
1526 hmap_destroy(&ofproto->realdev_vid_map);
1528 sset_destroy(&ofproto->ports);
1529 sset_destroy(&ofproto->ghost_ports);
1530 sset_destroy(&ofproto->port_poll_set);
1532 close_dpif_backer(ofproto->backer);
1536 run_fast(struct ofproto *ofproto_)
1538 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1539 struct ofport_dpif *ofport;
1541 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1542 port_run_fast(ofport);
1549 run(struct ofproto *ofproto_)
1551 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1552 struct ofport_dpif *ofport;
1553 struct ofbundle *bundle;
1557 complete_operations(ofproto);
1560 error = run_fast(ofproto_);
1565 if (ofproto->netflow) {
1566 if (netflow_run(ofproto->netflow)) {
1567 send_netflow_active_timeouts(ofproto);
1570 if (ofproto->sflow) {
1571 dpif_sflow_run(ofproto->sflow);
1574 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1577 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1582 mac_learning_run(ofproto->ml, &ofproto->backer->revalidate_set);
1584 /* Check the consistency of a random facet, to aid debugging. */
1585 if (time_msec() >= ofproto->consistency_rl
1586 && !hmap_is_empty(&ofproto->facets)
1587 && !ofproto->backer->need_revalidate) {
1588 struct facet *facet;
1590 ofproto->consistency_rl = time_msec() + 250;
1592 facet = CONTAINER_OF(hmap_random_node(&ofproto->facets),
1593 struct facet, hmap_node);
1594 if (!tag_set_intersects(&ofproto->backer->revalidate_set,
1595 facet->xout.tags)) {
1596 if (!facet_check_consistency(facet)) {
1597 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
1602 if (ofproto->governor) {
1605 governor_run(ofproto->governor);
1607 /* If the governor has shrunk to its minimum size and the number of
1608 * subfacets has dwindled, then drop the governor entirely.
1610 * For hysteresis, the number of subfacets to drop the governor is
1611 * smaller than the number needed to trigger its creation. */
1612 n_subfacets = hmap_count(&ofproto->subfacets);
1613 if (n_subfacets * 4 < ofproto->up.flow_eviction_threshold
1614 && governor_is_idle(ofproto->governor)) {
1615 governor_destroy(ofproto->governor);
1616 ofproto->governor = NULL;
1624 wait(struct ofproto *ofproto_)
1626 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1627 struct ofport_dpif *ofport;
1628 struct ofbundle *bundle;
1630 if (!clogged && !list_is_empty(&ofproto->completions)) {
1631 poll_immediate_wake();
1634 dpif_wait(ofproto->backer->dpif);
1635 dpif_recv_wait(ofproto->backer->dpif);
1636 if (ofproto->sflow) {
1637 dpif_sflow_wait(ofproto->sflow);
1639 if (!tag_set_is_empty(&ofproto->backer->revalidate_set)) {
1640 poll_immediate_wake();
1642 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1645 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1646 bundle_wait(bundle);
1648 if (ofproto->netflow) {
1649 netflow_wait(ofproto->netflow);
1651 mac_learning_wait(ofproto->ml);
1653 if (ofproto->backer->need_revalidate) {
1654 /* Shouldn't happen, but if it does just go around again. */
1655 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1656 poll_immediate_wake();
1658 if (ofproto->governor) {
1659 governor_wait(ofproto->governor);
1664 get_memory_usage(const struct ofproto *ofproto_, struct simap *usage)
1666 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1668 simap_increase(usage, "facets", hmap_count(&ofproto->facets));
1669 simap_increase(usage, "subfacets", hmap_count(&ofproto->subfacets));
1673 flush(struct ofproto *ofproto_)
1675 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1676 struct subfacet *subfacet, *next_subfacet;
1677 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
1681 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
1682 &ofproto->subfacets) {
1683 if (subfacet->path != SF_NOT_INSTALLED) {
1684 batch[n_batch++] = subfacet;
1685 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
1686 subfacet_destroy_batch(ofproto, batch, n_batch);
1690 subfacet_destroy(subfacet);
1695 subfacet_destroy_batch(ofproto, batch, n_batch);
1700 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1701 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1703 *arp_match_ip = true;
1704 *actions = (OFPUTIL_A_OUTPUT |
1705 OFPUTIL_A_SET_VLAN_VID |
1706 OFPUTIL_A_SET_VLAN_PCP |
1707 OFPUTIL_A_STRIP_VLAN |
1708 OFPUTIL_A_SET_DL_SRC |
1709 OFPUTIL_A_SET_DL_DST |
1710 OFPUTIL_A_SET_NW_SRC |
1711 OFPUTIL_A_SET_NW_DST |
1712 OFPUTIL_A_SET_NW_TOS |
1713 OFPUTIL_A_SET_TP_SRC |
1714 OFPUTIL_A_SET_TP_DST |
1719 get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots)
1721 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1722 struct dpif_dp_stats s;
1723 uint64_t n_miss, n_no_pkt_in, n_bytes;
1726 strcpy(ots->name, "classifier");
1728 dpif_get_dp_stats(ofproto->backer->dpif, &s);
1729 rule_get_stats(&ofproto->miss_rule->up, &n_miss, &n_bytes);
1730 rule_get_stats(&ofproto->no_packet_in_rule->up, &n_no_pkt_in, &n_bytes);
1732 n_lookup = s.n_hit + s.n_missed;
1733 ots->lookup_count = htonll(n_lookup);
1734 ots->matched_count = htonll(n_lookup - n_miss - n_no_pkt_in);
1737 static struct ofport *
1740 struct ofport_dpif *port = xmalloc(sizeof *port);
1745 port_dealloc(struct ofport *port_)
1747 struct ofport_dpif *port = ofport_dpif_cast(port_);
1752 port_construct(struct ofport *port_)
1754 struct ofport_dpif *port = ofport_dpif_cast(port_);
1755 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1756 const struct netdev *netdev = port->up.netdev;
1757 struct dpif_port dpif_port;
1760 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1761 port->bundle = NULL;
1764 port->tag = tag_create_random();
1765 port->may_enable = true;
1766 port->stp_port = NULL;
1767 port->stp_state = STP_DISABLED;
1768 port->tnl_port = NULL;
1769 hmap_init(&port->priorities);
1770 port->realdev_ofp_port = 0;
1771 port->vlandev_vid = 0;
1772 port->carrier_seq = netdev_get_carrier_resets(netdev);
1774 if (netdev_vport_is_patch(netdev)) {
1775 /* By bailing out here, we don't submit the port to the sFlow module
1776 * to be considered for counter polling export. This is correct
1777 * because the patch port represents an interface that sFlow considers
1778 * to be "internal" to the switch as a whole, and therefore not an
1779 * candidate for counter polling. */
1780 port->odp_port = OVSP_NONE;
1784 error = dpif_port_query_by_name(ofproto->backer->dpif,
1785 netdev_vport_get_dpif_port(netdev),
1791 port->odp_port = dpif_port.port_no;
1793 if (netdev_get_tunnel_config(netdev)) {
1794 port->tnl_port = tnl_port_add(&port->up, port->odp_port);
1796 /* Sanity-check that a mapping doesn't already exist. This
1797 * shouldn't happen for non-tunnel ports. */
1798 if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1799 VLOG_ERR("port %s already has an OpenFlow port number",
1801 dpif_port_destroy(&dpif_port);
1805 hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1806 hash_int(port->odp_port, 0));
1808 dpif_port_destroy(&dpif_port);
1810 if (ofproto->sflow) {
1811 dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1818 port_destruct(struct ofport *port_)
1820 struct ofport_dpif *port = ofport_dpif_cast(port_);
1821 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1822 const char *dp_port_name = netdev_vport_get_dpif_port(port->up.netdev);
1823 const char *devname = netdev_get_name(port->up.netdev);
1825 if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1826 /* The underlying device is still there, so delete it. This
1827 * happens when the ofproto is being destroyed, since the caller
1828 * assumes that removal of attached ports will happen as part of
1830 if (!port->tnl_port) {
1831 dpif_port_del(ofproto->backer->dpif, port->odp_port);
1833 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1836 if (port->odp_port != OVSP_NONE && !port->tnl_port) {
1837 hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1840 tnl_port_del(port->tnl_port);
1841 sset_find_and_delete(&ofproto->ports, devname);
1842 sset_find_and_delete(&ofproto->ghost_ports, devname);
1843 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1844 bundle_remove(port_);
1845 set_cfm(port_, NULL);
1846 set_bfd(port_, NULL);
1847 if (ofproto->sflow) {
1848 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1851 ofport_clear_priorities(port);
1852 hmap_destroy(&port->priorities);
1856 port_modified(struct ofport *port_)
1858 struct ofport_dpif *port = ofport_dpif_cast(port_);
1860 if (port->bundle && port->bundle->bond) {
1861 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1866 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1868 struct ofport_dpif *port = ofport_dpif_cast(port_);
1869 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1870 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1872 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1873 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1874 OFPUTIL_PC_NO_PACKET_IN)) {
1875 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1877 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1878 bundle_update(port->bundle);
1884 set_sflow(struct ofproto *ofproto_,
1885 const struct ofproto_sflow_options *sflow_options)
1887 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1888 struct dpif_sflow *ds = ofproto->sflow;
1890 if (sflow_options) {
1892 struct ofport_dpif *ofport;
1894 ds = ofproto->sflow = dpif_sflow_create();
1895 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1896 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1898 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1900 dpif_sflow_set_options(ds, sflow_options);
1903 dpif_sflow_destroy(ds);
1904 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1905 ofproto->sflow = NULL;
1913 struct ofproto *ofproto_,
1914 const struct ofproto_ipfix_bridge_exporter_options *bridge_exporter_options,
1915 const struct ofproto_ipfix_flow_exporter_options *flow_exporters_options,
1916 size_t n_flow_exporters_options)
1918 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1919 struct dpif_ipfix *di = ofproto->ipfix;
1921 if (bridge_exporter_options || flow_exporters_options) {
1923 di = ofproto->ipfix = dpif_ipfix_create();
1925 dpif_ipfix_set_options(
1926 di, bridge_exporter_options, flow_exporters_options,
1927 n_flow_exporters_options);
1930 dpif_ipfix_destroy(di);
1931 ofproto->ipfix = NULL;
1938 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1940 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1947 struct ofproto_dpif *ofproto;
1949 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1950 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1951 ofport->cfm = cfm_create(ofport->up.netdev);
1954 if (cfm_configure(ofport->cfm, s)) {
1960 cfm_destroy(ofport->cfm);
1966 get_cfm_status(const struct ofport *ofport_,
1967 struct ofproto_cfm_status *status)
1969 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1972 status->faults = cfm_get_fault(ofport->cfm);
1973 status->remote_opstate = cfm_get_opup(ofport->cfm);
1974 status->health = cfm_get_health(ofport->cfm);
1975 cfm_get_remote_mpids(ofport->cfm, &status->rmps, &status->n_rmps);
1983 set_bfd(struct ofport *ofport_, const struct smap *cfg)
1985 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
1986 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1990 ofport->bfd = bfd_configure(old, netdev_get_name(ofport->up.netdev), cfg);
1991 if (ofport->bfd != old) {
1992 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1999 get_bfd_status(struct ofport *ofport_, struct smap *smap)
2001 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2004 bfd_get_status(ofport->bfd, smap);
2011 /* Spanning Tree. */
2014 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
2016 struct ofproto_dpif *ofproto = ofproto_;
2017 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
2018 struct ofport_dpif *ofport;
2020 ofport = stp_port_get_aux(sp);
2022 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
2023 ofproto->up.name, port_num);
2025 struct eth_header *eth = pkt->l2;
2027 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
2028 if (eth_addr_is_zero(eth->eth_src)) {
2029 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
2030 "with unknown MAC", ofproto->up.name, port_num);
2032 send_packet(ofport, pkt);
2038 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
2040 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
2042 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2044 /* Only revalidate flows if the configuration changed. */
2045 if (!s != !ofproto->stp) {
2046 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2050 if (!ofproto->stp) {
2051 ofproto->stp = stp_create(ofproto_->name, s->system_id,
2052 send_bpdu_cb, ofproto);
2053 ofproto->stp_last_tick = time_msec();
2056 stp_set_bridge_id(ofproto->stp, s->system_id);
2057 stp_set_bridge_priority(ofproto->stp, s->priority);
2058 stp_set_hello_time(ofproto->stp, s->hello_time);
2059 stp_set_max_age(ofproto->stp, s->max_age);
2060 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
2062 struct ofport *ofport;
2064 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
2065 set_stp_port(ofport, NULL);
2068 stp_destroy(ofproto->stp);
2069 ofproto->stp = NULL;
2076 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
2078 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2082 s->bridge_id = stp_get_bridge_id(ofproto->stp);
2083 s->designated_root = stp_get_designated_root(ofproto->stp);
2084 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
2093 update_stp_port_state(struct ofport_dpif *ofport)
2095 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2096 enum stp_state state;
2098 /* Figure out new state. */
2099 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
2103 if (ofport->stp_state != state) {
2104 enum ofputil_port_state of_state;
2107 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
2108 netdev_get_name(ofport->up.netdev),
2109 stp_state_name(ofport->stp_state),
2110 stp_state_name(state));
2111 if (stp_learn_in_state(ofport->stp_state)
2112 != stp_learn_in_state(state)) {
2113 /* xxx Learning action flows should also be flushed. */
2114 mac_learning_flush(ofproto->ml,
2115 &ofproto->backer->revalidate_set);
2117 fwd_change = stp_forward_in_state(ofport->stp_state)
2118 != stp_forward_in_state(state);
2120 ofproto->backer->need_revalidate = REV_STP;
2121 ofport->stp_state = state;
2122 ofport->stp_state_entered = time_msec();
2124 if (fwd_change && ofport->bundle) {
2125 bundle_update(ofport->bundle);
2128 /* Update the STP state bits in the OpenFlow port description. */
2129 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
2130 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
2131 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
2132 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
2133 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
2135 ofproto_port_set_state(&ofport->up, of_state);
2139 /* Configures STP on 'ofport_' using the settings defined in 's'. The
2140 * caller is responsible for assigning STP port numbers and ensuring
2141 * there are no duplicates. */
2143 set_stp_port(struct ofport *ofport_,
2144 const struct ofproto_port_stp_settings *s)
2146 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2147 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2148 struct stp_port *sp = ofport->stp_port;
2150 if (!s || !s->enable) {
2152 ofport->stp_port = NULL;
2153 stp_port_disable(sp);
2154 update_stp_port_state(ofport);
2157 } else if (sp && stp_port_no(sp) != s->port_num
2158 && ofport == stp_port_get_aux(sp)) {
2159 /* The port-id changed, so disable the old one if it's not
2160 * already in use by another port. */
2161 stp_port_disable(sp);
2164 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
2165 stp_port_enable(sp);
2167 stp_port_set_aux(sp, ofport);
2168 stp_port_set_priority(sp, s->priority);
2169 stp_port_set_path_cost(sp, s->path_cost);
2171 update_stp_port_state(ofport);
2177 get_stp_port_status(struct ofport *ofport_,
2178 struct ofproto_port_stp_status *s)
2180 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2181 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2182 struct stp_port *sp = ofport->stp_port;
2184 if (!ofproto->stp || !sp) {
2190 s->port_id = stp_port_get_id(sp);
2191 s->state = stp_port_get_state(sp);
2192 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
2193 s->role = stp_port_get_role(sp);
2194 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
2200 stp_run(struct ofproto_dpif *ofproto)
2203 long long int now = time_msec();
2204 long long int elapsed = now - ofproto->stp_last_tick;
2205 struct stp_port *sp;
2208 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
2209 ofproto->stp_last_tick = now;
2211 while (stp_get_changed_port(ofproto->stp, &sp)) {
2212 struct ofport_dpif *ofport = stp_port_get_aux(sp);
2215 update_stp_port_state(ofport);
2219 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
2220 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
2226 stp_wait(struct ofproto_dpif *ofproto)
2229 poll_timer_wait(1000);
2233 /* Returns true if STP should process 'flow'. */
2235 stp_should_process_flow(const struct flow *flow)
2237 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
2241 stp_process_packet(const struct ofport_dpif *ofport,
2242 const struct ofpbuf *packet)
2244 struct ofpbuf payload = *packet;
2245 struct eth_header *eth = payload.data;
2246 struct stp_port *sp = ofport->stp_port;
2248 /* Sink packets on ports that have STP disabled when the bridge has
2250 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
2254 /* Trim off padding on payload. */
2255 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
2256 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
2259 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
2260 stp_received_bpdu(sp, payload.data, payload.size);
2264 static struct priority_to_dscp *
2265 get_priority(const struct ofport_dpif *ofport, uint32_t priority)
2267 struct priority_to_dscp *pdscp;
2270 hash = hash_int(priority, 0);
2271 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) {
2272 if (pdscp->priority == priority) {
2280 ofport_clear_priorities(struct ofport_dpif *ofport)
2282 struct priority_to_dscp *pdscp, *next;
2284 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) {
2285 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
2291 set_queues(struct ofport *ofport_,
2292 const struct ofproto_port_queue *qdscp_list,
2295 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2296 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2297 struct hmap new = HMAP_INITIALIZER(&new);
2300 for (i = 0; i < n_qdscp; i++) {
2301 struct priority_to_dscp *pdscp;
2305 dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
2306 if (dpif_queue_to_priority(ofproto->backer->dpif, qdscp_list[i].queue,
2311 pdscp = get_priority(ofport, priority);
2313 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
2315 pdscp = xmalloc(sizeof *pdscp);
2316 pdscp->priority = priority;
2318 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2321 if (pdscp->dscp != dscp) {
2323 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2326 hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0));
2329 if (!hmap_is_empty(&ofport->priorities)) {
2330 ofport_clear_priorities(ofport);
2331 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2334 hmap_swap(&new, &ofport->priorities);
2342 /* Expires all MAC learning entries associated with 'bundle' and forces its
2343 * ofproto to revalidate every flow.
2345 * Normally MAC learning entries are removed only from the ofproto associated
2346 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2347 * are removed from every ofproto. When patch ports and SLB bonds are in use
2348 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2349 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2350 * with the host from which it migrated. */
2352 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2354 struct ofproto_dpif *ofproto = bundle->ofproto;
2355 struct mac_learning *ml = ofproto->ml;
2356 struct mac_entry *mac, *next_mac;
2358 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2359 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2360 if (mac->port.p == bundle) {
2362 struct ofproto_dpif *o;
2364 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2366 struct mac_entry *e;
2368 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan,
2371 mac_learning_expire(o->ml, e);
2377 mac_learning_expire(ml, mac);
2382 static struct ofbundle *
2383 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2385 struct ofbundle *bundle;
2387 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2388 &ofproto->bundles) {
2389 if (bundle->aux == aux) {
2396 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
2397 * ones that are found to 'bundles'. */
2399 bundle_lookup_multiple(struct ofproto_dpif *ofproto,
2400 void **auxes, size_t n_auxes,
2401 struct hmapx *bundles)
2405 hmapx_init(bundles);
2406 for (i = 0; i < n_auxes; i++) {
2407 struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
2409 hmapx_add(bundles, bundle);
2415 bundle_update(struct ofbundle *bundle)
2417 struct ofport_dpif *port;
2419 bundle->floodable = true;
2420 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2421 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2422 || !stp_forward_in_state(port->stp_state)) {
2423 bundle->floodable = false;
2430 bundle_del_port(struct ofport_dpif *port)
2432 struct ofbundle *bundle = port->bundle;
2434 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2436 list_remove(&port->bundle_node);
2437 port->bundle = NULL;
2440 lacp_slave_unregister(bundle->lacp, port);
2443 bond_slave_unregister(bundle->bond, port);
2446 bundle_update(bundle);
2450 bundle_add_port(struct ofbundle *bundle, uint16_t ofp_port,
2451 struct lacp_slave_settings *lacp)
2453 struct ofport_dpif *port;
2455 port = get_ofp_port(bundle->ofproto, ofp_port);
2460 if (port->bundle != bundle) {
2461 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2463 bundle_del_port(port);
2466 port->bundle = bundle;
2467 list_push_back(&bundle->ports, &port->bundle_node);
2468 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2469 || !stp_forward_in_state(port->stp_state)) {
2470 bundle->floodable = false;
2474 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2475 lacp_slave_register(bundle->lacp, port, lacp);
2482 bundle_destroy(struct ofbundle *bundle)
2484 struct ofproto_dpif *ofproto;
2485 struct ofport_dpif *port, *next_port;
2492 ofproto = bundle->ofproto;
2493 for (i = 0; i < MAX_MIRRORS; i++) {
2494 struct ofmirror *m = ofproto->mirrors[i];
2496 if (m->out == bundle) {
2498 } else if (hmapx_find_and_delete(&m->srcs, bundle)
2499 || hmapx_find_and_delete(&m->dsts, bundle)) {
2500 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2505 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2506 bundle_del_port(port);
2509 bundle_flush_macs(bundle, true);
2510 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2512 free(bundle->trunks);
2513 lacp_destroy(bundle->lacp);
2514 bond_destroy(bundle->bond);
2519 bundle_set(struct ofproto *ofproto_, void *aux,
2520 const struct ofproto_bundle_settings *s)
2522 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2523 bool need_flush = false;
2524 struct ofport_dpif *port;
2525 struct ofbundle *bundle;
2526 unsigned long *trunks;
2532 bundle_destroy(bundle_lookup(ofproto, aux));
2536 ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2537 ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2539 bundle = bundle_lookup(ofproto, aux);
2541 bundle = xmalloc(sizeof *bundle);
2543 bundle->ofproto = ofproto;
2544 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2545 hash_pointer(aux, 0));
2547 bundle->name = NULL;
2549 list_init(&bundle->ports);
2550 bundle->vlan_mode = PORT_VLAN_TRUNK;
2552 bundle->trunks = NULL;
2553 bundle->use_priority_tags = s->use_priority_tags;
2554 bundle->lacp = NULL;
2555 bundle->bond = NULL;
2557 bundle->floodable = true;
2559 bundle->src_mirrors = 0;
2560 bundle->dst_mirrors = 0;
2561 bundle->mirror_out = 0;
2564 if (!bundle->name || strcmp(s->name, bundle->name)) {
2566 bundle->name = xstrdup(s->name);
2571 if (!bundle->lacp) {
2572 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2573 bundle->lacp = lacp_create();
2575 lacp_configure(bundle->lacp, s->lacp);
2577 lacp_destroy(bundle->lacp);
2578 bundle->lacp = NULL;
2581 /* Update set of ports. */
2583 for (i = 0; i < s->n_slaves; i++) {
2584 if (!bundle_add_port(bundle, s->slaves[i],
2585 s->lacp ? &s->lacp_slaves[i] : NULL)) {
2589 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2590 struct ofport_dpif *next_port;
2592 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2593 for (i = 0; i < s->n_slaves; i++) {
2594 if (s->slaves[i] == port->up.ofp_port) {
2599 bundle_del_port(port);
2603 ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2605 if (list_is_empty(&bundle->ports)) {
2606 bundle_destroy(bundle);
2610 /* Set VLAN tagging mode */
2611 if (s->vlan_mode != bundle->vlan_mode
2612 || s->use_priority_tags != bundle->use_priority_tags) {
2613 bundle->vlan_mode = s->vlan_mode;
2614 bundle->use_priority_tags = s->use_priority_tags;
2619 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2620 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2622 if (vlan != bundle->vlan) {
2623 bundle->vlan = vlan;
2627 /* Get trunked VLANs. */
2628 switch (s->vlan_mode) {
2629 case PORT_VLAN_ACCESS:
2633 case PORT_VLAN_TRUNK:
2634 trunks = CONST_CAST(unsigned long *, s->trunks);
2637 case PORT_VLAN_NATIVE_UNTAGGED:
2638 case PORT_VLAN_NATIVE_TAGGED:
2639 if (vlan != 0 && (!s->trunks
2640 || !bitmap_is_set(s->trunks, vlan)
2641 || bitmap_is_set(s->trunks, 0))) {
2642 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2644 trunks = bitmap_clone(s->trunks, 4096);
2646 trunks = bitmap_allocate1(4096);
2648 bitmap_set1(trunks, vlan);
2649 bitmap_set0(trunks, 0);
2651 trunks = CONST_CAST(unsigned long *, s->trunks);
2658 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2659 free(bundle->trunks);
2660 if (trunks == s->trunks) {
2661 bundle->trunks = vlan_bitmap_clone(trunks);
2663 bundle->trunks = trunks;
2668 if (trunks != s->trunks) {
2673 if (!list_is_short(&bundle->ports)) {
2674 bundle->ofproto->has_bonded_bundles = true;
2676 if (bond_reconfigure(bundle->bond, s->bond)) {
2677 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2680 bundle->bond = bond_create(s->bond);
2681 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2684 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2685 bond_slave_register(bundle->bond, port, port->up.netdev);
2688 bond_destroy(bundle->bond);
2689 bundle->bond = NULL;
2692 /* If we changed something that would affect MAC learning, un-learn
2693 * everything on this port and force flow revalidation. */
2695 bundle_flush_macs(bundle, false);
2702 bundle_remove(struct ofport *port_)
2704 struct ofport_dpif *port = ofport_dpif_cast(port_);
2705 struct ofbundle *bundle = port->bundle;
2708 bundle_del_port(port);
2709 if (list_is_empty(&bundle->ports)) {
2710 bundle_destroy(bundle);
2711 } else if (list_is_short(&bundle->ports)) {
2712 bond_destroy(bundle->bond);
2713 bundle->bond = NULL;
2719 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2721 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2722 struct ofport_dpif *port = port_;
2723 uint8_t ea[ETH_ADDR_LEN];
2726 error = netdev_get_etheraddr(port->up.netdev, ea);
2728 struct ofpbuf packet;
2731 ofpbuf_init(&packet, 0);
2732 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2734 memcpy(packet_pdu, pdu, pdu_size);
2736 send_packet(port, &packet);
2737 ofpbuf_uninit(&packet);
2739 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2740 "%s (%s)", port->bundle->name,
2741 netdev_get_name(port->up.netdev), strerror(error));
2746 bundle_send_learning_packets(struct ofbundle *bundle)
2748 struct ofproto_dpif *ofproto = bundle->ofproto;
2749 int error, n_packets, n_errors;
2750 struct mac_entry *e;
2752 error = n_packets = n_errors = 0;
2753 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2754 if (e->port.p != bundle) {
2755 struct ofpbuf *learning_packet;
2756 struct ofport_dpif *port;
2760 /* The assignment to "port" is unnecessary but makes "grep"ing for
2761 * struct ofport_dpif more effective. */
2762 learning_packet = bond_compose_learning_packet(bundle->bond,
2766 ret = send_packet(port, learning_packet);
2767 ofpbuf_delete(learning_packet);
2777 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2778 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2779 "packets, last error was: %s",
2780 bundle->name, n_errors, n_packets, strerror(error));
2782 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2783 bundle->name, n_packets);
2788 bundle_run(struct ofbundle *bundle)
2791 lacp_run(bundle->lacp, send_pdu_cb);
2794 struct ofport_dpif *port;
2796 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2797 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2800 bond_run(bundle->bond, &bundle->ofproto->backer->revalidate_set,
2801 lacp_status(bundle->lacp));
2802 if (bond_should_send_learning_packets(bundle->bond)) {
2803 bundle_send_learning_packets(bundle);
2809 bundle_wait(struct ofbundle *bundle)
2812 lacp_wait(bundle->lacp);
2815 bond_wait(bundle->bond);
2822 mirror_scan(struct ofproto_dpif *ofproto)
2826 for (idx = 0; idx < MAX_MIRRORS; idx++) {
2827 if (!ofproto->mirrors[idx]) {
2834 static struct ofmirror *
2835 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
2839 for (i = 0; i < MAX_MIRRORS; i++) {
2840 struct ofmirror *mirror = ofproto->mirrors[i];
2841 if (mirror && mirror->aux == aux) {
2849 /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */
2851 mirror_update_dups(struct ofproto_dpif *ofproto)
2855 for (i = 0; i < MAX_MIRRORS; i++) {
2856 struct ofmirror *m = ofproto->mirrors[i];
2859 m->dup_mirrors = MIRROR_MASK_C(1) << i;
2863 for (i = 0; i < MAX_MIRRORS; i++) {
2864 struct ofmirror *m1 = ofproto->mirrors[i];
2871 for (j = i + 1; j < MAX_MIRRORS; j++) {
2872 struct ofmirror *m2 = ofproto->mirrors[j];
2874 if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) {
2875 m1->dup_mirrors |= MIRROR_MASK_C(1) << j;
2876 m2->dup_mirrors |= m1->dup_mirrors;
2883 mirror_set(struct ofproto *ofproto_, void *aux,
2884 const struct ofproto_mirror_settings *s)
2886 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2887 mirror_mask_t mirror_bit;
2888 struct ofbundle *bundle;
2889 struct ofmirror *mirror;
2890 struct ofbundle *out;
2891 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
2892 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
2895 mirror = mirror_lookup(ofproto, aux);
2897 mirror_destroy(mirror);
2903 idx = mirror_scan(ofproto);
2905 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
2907 ofproto->up.name, MAX_MIRRORS, s->name);
2911 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
2912 mirror->ofproto = ofproto;
2915 mirror->out_vlan = -1;
2916 mirror->name = NULL;
2919 if (!mirror->name || strcmp(s->name, mirror->name)) {
2921 mirror->name = xstrdup(s->name);
2924 /* Get the new configuration. */
2925 if (s->out_bundle) {
2926 out = bundle_lookup(ofproto, s->out_bundle);
2928 mirror_destroy(mirror);
2934 out_vlan = s->out_vlan;
2936 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
2937 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
2939 /* If the configuration has not changed, do nothing. */
2940 if (hmapx_equals(&srcs, &mirror->srcs)
2941 && hmapx_equals(&dsts, &mirror->dsts)
2942 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
2943 && mirror->out == out
2944 && mirror->out_vlan == out_vlan)
2946 hmapx_destroy(&srcs);
2947 hmapx_destroy(&dsts);
2951 hmapx_swap(&srcs, &mirror->srcs);
2952 hmapx_destroy(&srcs);
2954 hmapx_swap(&dsts, &mirror->dsts);
2955 hmapx_destroy(&dsts);
2957 free(mirror->vlans);
2958 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
2961 mirror->out_vlan = out_vlan;
2963 /* Update bundles. */
2964 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2965 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
2966 if (hmapx_contains(&mirror->srcs, bundle)) {
2967 bundle->src_mirrors |= mirror_bit;
2969 bundle->src_mirrors &= ~mirror_bit;
2972 if (hmapx_contains(&mirror->dsts, bundle)) {
2973 bundle->dst_mirrors |= mirror_bit;
2975 bundle->dst_mirrors &= ~mirror_bit;
2978 if (mirror->out == bundle) {
2979 bundle->mirror_out |= mirror_bit;
2981 bundle->mirror_out &= ~mirror_bit;
2985 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2986 ofproto->has_mirrors = true;
2987 mac_learning_flush(ofproto->ml,
2988 &ofproto->backer->revalidate_set);
2989 mirror_update_dups(ofproto);
2995 mirror_destroy(struct ofmirror *mirror)
2997 struct ofproto_dpif *ofproto;
2998 mirror_mask_t mirror_bit;
2999 struct ofbundle *bundle;
3006 ofproto = mirror->ofproto;
3007 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3008 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
3010 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
3011 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
3012 bundle->src_mirrors &= ~mirror_bit;
3013 bundle->dst_mirrors &= ~mirror_bit;
3014 bundle->mirror_out &= ~mirror_bit;
3017 hmapx_destroy(&mirror->srcs);
3018 hmapx_destroy(&mirror->dsts);
3019 free(mirror->vlans);
3021 ofproto->mirrors[mirror->idx] = NULL;
3025 mirror_update_dups(ofproto);
3027 ofproto->has_mirrors = false;
3028 for (i = 0; i < MAX_MIRRORS; i++) {
3029 if (ofproto->mirrors[i]) {
3030 ofproto->has_mirrors = true;
3037 mirror_get_stats(struct ofproto *ofproto_, void *aux,
3038 uint64_t *packets, uint64_t *bytes)
3040 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3041 struct ofmirror *mirror = mirror_lookup(ofproto, aux);
3044 *packets = *bytes = UINT64_MAX;
3050 *packets = mirror->packet_count;
3051 *bytes = mirror->byte_count;
3057 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
3059 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3060 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
3061 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
3067 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
3069 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3070 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
3071 return bundle && bundle->mirror_out != 0;
3075 forward_bpdu_changed(struct ofproto *ofproto_)
3077 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3078 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3082 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
3085 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3086 mac_learning_set_idle_time(ofproto->ml, idle_time);
3087 mac_learning_set_max_entries(ofproto->ml, max_entries);
3092 static struct ofport_dpif *
3093 get_ofp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
3095 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
3096 return ofport ? ofport_dpif_cast(ofport) : NULL;
3099 static struct ofport_dpif *
3100 get_odp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
3102 struct ofport_dpif *port = odp_port_to_ofport(ofproto->backer, odp_port);
3103 return port && &ofproto->up == port->up.ofproto ? port : NULL;
3107 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
3108 struct ofproto_port *ofproto_port,
3109 struct dpif_port *dpif_port)
3111 ofproto_port->name = dpif_port->name;
3112 ofproto_port->type = dpif_port->type;
3113 ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
3116 static struct ofport_dpif *
3117 ofport_get_peer(const struct ofport_dpif *ofport_dpif)
3119 const struct ofproto_dpif *ofproto;
3122 peer = netdev_vport_patch_peer(ofport_dpif->up.netdev);
3127 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
3128 struct ofport *ofport;
3130 ofport = shash_find_data(&ofproto->up.port_by_name, peer);
3131 if (ofport && ofport->ofproto->ofproto_class == &ofproto_dpif_class) {
3132 return ofport_dpif_cast(ofport);
3139 port_run_fast(struct ofport_dpif *ofport)
3141 if (ofport->cfm && cfm_should_send_ccm(ofport->cfm)) {
3142 struct ofpbuf packet;
3144 ofpbuf_init(&packet, 0);
3145 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
3146 send_packet(ofport, &packet);
3147 ofpbuf_uninit(&packet);
3150 if (ofport->bfd && bfd_should_send_packet(ofport->bfd)) {
3151 struct ofpbuf packet;
3153 ofpbuf_init(&packet, 0);
3154 bfd_put_packet(ofport->bfd, &packet, ofport->up.pp.hw_addr);
3155 send_packet(ofport, &packet);
3156 ofpbuf_uninit(&packet);
3161 port_run(struct ofport_dpif *ofport)
3163 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
3164 bool carrier_changed = carrier_seq != ofport->carrier_seq;
3165 bool enable = netdev_get_carrier(ofport->up.netdev);
3167 ofport->carrier_seq = carrier_seq;
3169 port_run_fast(ofport);
3171 if (ofport->tnl_port
3172 && tnl_port_reconfigure(&ofport->up, ofport->odp_port,
3173 &ofport->tnl_port)) {
3174 ofproto_dpif_cast(ofport->up.ofproto)->backer->need_revalidate = true;
3178 int cfm_opup = cfm_get_opup(ofport->cfm);
3180 cfm_run(ofport->cfm);
3181 enable = enable && !cfm_get_fault(ofport->cfm);
3183 if (cfm_opup >= 0) {
3184 enable = enable && cfm_opup;
3189 bfd_run(ofport->bfd);
3190 enable = enable && bfd_forwarding(ofport->bfd);
3193 if (ofport->bundle) {
3194 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
3195 if (carrier_changed) {
3196 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
3200 if (ofport->may_enable != enable) {
3201 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3203 if (ofproto->has_bundle_action) {
3204 ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
3208 ofport->may_enable = enable;
3212 port_wait(struct ofport_dpif *ofport)
3215 cfm_wait(ofport->cfm);
3219 bfd_wait(ofport->bfd);
3224 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
3225 struct ofproto_port *ofproto_port)
3227 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3228 struct dpif_port dpif_port;
3231 if (sset_contains(&ofproto->ghost_ports, devname)) {
3232 const char *type = netdev_get_type_from_name(devname);
3234 /* We may be called before ofproto->up.port_by_name is populated with
3235 * the appropriate ofport. For this reason, we must get the name and
3236 * type from the netdev layer directly. */
3238 const struct ofport *ofport;
3240 ofport = shash_find_data(&ofproto->up.port_by_name, devname);
3241 ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
3242 ofproto_port->name = xstrdup(devname);
3243 ofproto_port->type = xstrdup(type);
3249 if (!sset_contains(&ofproto->ports, devname)) {
3252 error = dpif_port_query_by_name(ofproto->backer->dpif,
3253 devname, &dpif_port);
3255 ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
3261 port_add(struct ofproto *ofproto_, struct netdev *netdev)
3263 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3264 const char *dp_port_name = netdev_vport_get_dpif_port(netdev);
3265 const char *devname = netdev_get_name(netdev);
3267 if (netdev_vport_is_patch(netdev)) {
3268 sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
3272 if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
3273 uint32_t port_no = UINT32_MAX;
3276 error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
3280 if (netdev_get_tunnel_config(netdev)) {
3281 simap_put(&ofproto->backer->tnl_backers, dp_port_name, port_no);
3285 if (netdev_get_tunnel_config(netdev)) {
3286 sset_add(&ofproto->ghost_ports, devname);
3288 sset_add(&ofproto->ports, devname);
3294 port_del(struct ofproto *ofproto_, uint16_t ofp_port)
3296 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3297 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
3304 sset_find_and_delete(&ofproto->ghost_ports,
3305 netdev_get_name(ofport->up.netdev));
3306 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3307 if (!ofport->tnl_port) {
3308 error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
3310 /* The caller is going to close ofport->up.netdev. If this is a
3311 * bonded port, then the bond is using that netdev, so remove it
3312 * from the bond. The client will need to reconfigure everything
3313 * after deleting ports, so then the slave will get re-added. */
3314 bundle_remove(&ofport->up);
3321 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
3323 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3328 error = netdev_get_stats(ofport->up.netdev, stats);
3330 if (!error && ofport_->ofp_port == OFPP_LOCAL) {
3331 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3333 /* ofproto->stats.tx_packets represents packets that we created
3334 * internally and sent to some port (e.g. packets sent with
3335 * send_packet()). Account for them as if they had come from
3336 * OFPP_LOCAL and got forwarded. */
3338 if (stats->rx_packets != UINT64_MAX) {
3339 stats->rx_packets += ofproto->stats.tx_packets;
3342 if (stats->rx_bytes != UINT64_MAX) {
3343 stats->rx_bytes += ofproto->stats.tx_bytes;
3346 /* ofproto->stats.rx_packets represents packets that were received on
3347 * some port and we processed internally and dropped (e.g. STP).
3348 * Account for them as if they had been forwarded to OFPP_LOCAL. */
3350 if (stats->tx_packets != UINT64_MAX) {
3351 stats->tx_packets += ofproto->stats.rx_packets;
3354 if (stats->tx_bytes != UINT64_MAX) {
3355 stats->tx_bytes += ofproto->stats.rx_bytes;
3362 struct port_dump_state {
3367 struct ofproto_port port;
3372 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
3374 *statep = xzalloc(sizeof(struct port_dump_state));
3379 port_dump_next(const struct ofproto *ofproto_, void *state_,
3380 struct ofproto_port *port)
3382 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3383 struct port_dump_state *state = state_;
3384 const struct sset *sset;
3385 struct sset_node *node;
3387 if (state->has_port) {
3388 ofproto_port_destroy(&state->port);
3389 state->has_port = false;
3391 sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
3392 while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
3395 error = port_query_by_name(ofproto_, node->name, &state->port);
3397 *port = state->port;
3398 state->has_port = true;
3400 } else if (error != ENODEV) {
3405 if (!state->ghost) {
3406 state->ghost = true;
3409 return port_dump_next(ofproto_, state_, port);
3416 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3418 struct port_dump_state *state = state_;
3420 if (state->has_port) {
3421 ofproto_port_destroy(&state->port);
3428 port_poll(const struct ofproto *ofproto_, char **devnamep)
3430 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3432 if (ofproto->port_poll_errno) {
3433 int error = ofproto->port_poll_errno;
3434 ofproto->port_poll_errno = 0;
3438 if (sset_is_empty(&ofproto->port_poll_set)) {
3442 *devnamep = sset_pop(&ofproto->port_poll_set);
3447 port_poll_wait(const struct ofproto *ofproto_)
3449 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3450 dpif_port_poll_wait(ofproto->backer->dpif);
3454 port_is_lacp_current(const struct ofport *ofport_)
3456 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3457 return (ofport->bundle && ofport->bundle->lacp
3458 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3462 /* Upcall handling. */
3464 /* Flow miss batching.
3466 * Some dpifs implement operations faster when you hand them off in a batch.
3467 * To allow batching, "struct flow_miss" queues the dpif-related work needed
3468 * for a given flow. Each "struct flow_miss" corresponds to sending one or
3469 * more packets, plus possibly installing the flow in the dpif.
3471 * So far we only batch the operations that affect flow setup time the most.
3472 * It's possible to batch more than that, but the benefit might be minimal. */
3474 struct hmap_node hmap_node;
3475 struct ofproto_dpif *ofproto;
3477 enum odp_key_fitness key_fitness;
3478 const struct nlattr *key;
3480 struct initial_vals initial_vals;
3481 struct list packets;
3482 enum dpif_upcall_type upcall_type;
3485 struct flow_miss_op {
3486 struct dpif_op dpif_op;
3488 uint64_t slow_stub[128 / 8]; /* Buffer for compose_slow_path() */
3489 struct xlate_out xout;
3490 bool xout_garbage; /* 'xout' needs to be uninitialized? */
3493 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
3494 * OpenFlow controller as necessary according to their individual
3495 * configurations. */
3497 send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet,
3498 const struct flow *flow)
3500 struct ofputil_packet_in pin;
3502 pin.packet = packet->data;
3503 pin.packet_len = packet->size;
3504 pin.reason = OFPR_NO_MATCH;
3505 pin.controller_id = 0;
3510 pin.send_len = 0; /* not used for flow table misses */
3512 flow_get_metadata(flow, &pin.fmd);
3514 connmgr_send_packet_in(ofproto->up.connmgr, &pin);
3517 static enum slow_path_reason
3518 process_special(struct ofproto_dpif *ofproto, const struct flow *flow,
3519 const struct ofport_dpif *ofport, const struct ofpbuf *packet)
3523 } else if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) {
3525 cfm_process_heartbeat(ofport->cfm, packet);
3528 } else if (ofport->bfd && bfd_should_process_flow(flow)) {
3530 bfd_process_packet(ofport->bfd, flow, packet);
3533 } else if (ofport->bundle && ofport->bundle->lacp
3534 && flow->dl_type == htons(ETH_TYPE_LACP)) {
3536 lacp_process_packet(ofport->bundle->lacp, ofport, packet);
3539 } else if (ofproto->stp && stp_should_process_flow(flow)) {
3541 stp_process_packet(ofport, packet);
3549 static struct flow_miss *
3550 flow_miss_find(struct hmap *todo, const struct ofproto_dpif *ofproto,
3551 const struct flow *flow, uint32_t hash)
3553 struct flow_miss *miss;
3555 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
3556 if (miss->ofproto == ofproto && flow_equal(&miss->flow, flow)) {
3564 /* Partially Initializes 'op' as an "execute" operation for 'miss' and
3565 * 'packet'. The caller must initialize op->actions and op->actions_len. If
3566 * 'miss' is associated with a subfacet the caller must also initialize the
3567 * returned op->subfacet, and if anything needs to be freed after processing
3568 * the op, the caller must initialize op->garbage also. */
3570 init_flow_miss_execute_op(struct flow_miss *miss, struct ofpbuf *packet,
3571 struct flow_miss_op *op)
3573 if (miss->flow.vlan_tci != miss->initial_vals.vlan_tci) {
3574 /* This packet was received on a VLAN splinter port. We
3575 * added a VLAN to the packet to make the packet resemble
3576 * the flow, but the actions were composed assuming that
3577 * the packet contained no VLAN. So, we must remove the
3578 * VLAN header from the packet before trying to execute the
3580 eth_pop_vlan(packet);
3583 op->xout_garbage = false;
3584 op->dpif_op.type = DPIF_OP_EXECUTE;
3585 op->dpif_op.u.execute.key = miss->key;
3586 op->dpif_op.u.execute.key_len = miss->key_len;
3587 op->dpif_op.u.execute.packet = packet;
3590 /* Helper for handle_flow_miss_without_facet() and
3591 * handle_flow_miss_with_facet(). */
3593 handle_flow_miss_common(struct rule_dpif *rule,
3594 struct ofpbuf *packet, const struct flow *flow)
3596 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3598 if (rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
3600 * Extra-special case for fail-open mode.
3602 * We are in fail-open mode and the packet matched the fail-open
3603 * rule, but we are connected to a controller too. We should send
3604 * the packet up to the controller in the hope that it will try to
3605 * set up a flow and thereby allow us to exit fail-open.
3607 * See the top-level comment in fail-open.c for more information.
3609 send_packet_in_miss(ofproto, packet, flow);
3613 /* Figures out whether a flow that missed in 'ofproto', whose details are in
3614 * 'miss', is likely to be worth tracking in detail in userspace and (usually)
3615 * installing a datapath flow. The answer is usually "yes" (a return value of
3616 * true). However, for short flows the cost of bookkeeping is much higher than
3617 * the benefits, so when the datapath holds a large number of flows we impose
3618 * some heuristics to decide which flows are likely to be worth tracking. */
3620 flow_miss_should_make_facet(struct ofproto_dpif *ofproto,
3621 struct flow_miss *miss, uint32_t hash)
3623 if (!ofproto->governor) {
3626 n_subfacets = hmap_count(&ofproto->subfacets);
3627 if (n_subfacets * 2 <= ofproto->up.flow_eviction_threshold) {
3631 ofproto->governor = governor_create(ofproto->up.name);
3634 return governor_should_install_flow(ofproto->governor, hash,
3635 list_size(&miss->packets));
3638 /* Handles 'miss', which matches 'rule', without creating a facet or subfacet
3639 * or creating any datapath flow. May add an "execute" operation to 'ops' and
3640 * increment '*n_ops'. */
3642 handle_flow_miss_without_facet(struct flow_miss *miss,
3643 struct flow_miss_op *ops, size_t *n_ops)
3645 struct rule_dpif *rule = rule_dpif_lookup(miss->ofproto, &miss->flow);
3646 long long int now = time_msec();
3647 struct ofpbuf *packet;
3648 struct xlate_in xin;
3650 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3651 struct flow_miss_op *op = &ops[*n_ops];
3652 struct dpif_flow_stats stats;
3654 COVERAGE_INC(facet_suppress);
3656 handle_flow_miss_common(rule, packet, &miss->flow);
3658 dpif_flow_stats_extract(&miss->flow, packet, now, &stats);
3659 rule_credit_stats(rule, &stats);
3661 xlate_in_init(&xin, miss->ofproto, &miss->flow, &miss->initial_vals,
3662 rule, stats.tcp_flags, packet);
3663 xin.resubmit_stats = &stats;
3664 xlate_actions(&xin, &op->xout);
3666 if (op->xout.odp_actions.size) {
3667 struct dpif_execute *execute = &op->dpif_op.u.execute;
3669 init_flow_miss_execute_op(miss, packet, op);
3670 execute->actions = op->xout.odp_actions.data;
3671 execute->actions_len = op->xout.odp_actions.size;
3672 op->xout_garbage = true;
3676 xlate_out_uninit(&op->xout);
3681 /* Handles 'miss', which matches 'facet'. May add any required datapath
3682 * operations to 'ops', incrementing '*n_ops' for each new op.
3684 * All of the packets in 'miss' are considered to have arrived at time 'now'.
3685 * This is really important only for new facets: if we just called time_msec()
3686 * here, then the new subfacet or its packets could look (occasionally) as
3687 * though it was used some time after the facet was used. That can make a
3688 * one-packet flow look like it has a nonzero duration, which looks odd in
3689 * e.g. NetFlow statistics. */
3691 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
3693 struct flow_miss_op *ops, size_t *n_ops)
3695 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3696 enum subfacet_path want_path;
3697 struct subfacet *subfacet;
3698 struct ofpbuf *packet;
3700 subfacet = subfacet_create(facet, miss, now);
3701 want_path = subfacet->facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
3703 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3704 struct flow_miss_op *op = &ops[*n_ops];
3705 struct dpif_flow_stats stats;
3707 handle_flow_miss_common(facet->rule, packet, &miss->flow);
3709 if (want_path != SF_FAST_PATH) {
3710 struct xlate_in xin;
3712 xlate_in_init(&xin, ofproto, &facet->flow, &facet->initial_vals,
3713 facet->rule, 0, packet);
3714 xlate_actions_for_side_effects(&xin);
3717 dpif_flow_stats_extract(&facet->flow, packet, now, &stats);
3718 subfacet_update_stats(subfacet, &stats);
3720 if (facet->xout.odp_actions.size) {
3721 struct dpif_execute *execute = &op->dpif_op.u.execute;
3723 init_flow_miss_execute_op(miss, packet, op);
3724 execute->actions = facet->xout.odp_actions.data,
3725 execute->actions_len = facet->xout.odp_actions.size;
3730 if (miss->upcall_type == DPIF_UC_MISS || subfacet->path != want_path) {
3731 struct flow_miss_op *op = &ops[(*n_ops)++];
3732 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
3734 subfacet->path = want_path;
3736 op->xout_garbage = false;
3737 op->dpif_op.type = DPIF_OP_FLOW_PUT;
3738 put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
3739 put->key = miss->key;
3740 put->key_len = miss->key_len;
3741 if (want_path == SF_FAST_PATH) {
3742 put->actions = facet->xout.odp_actions.data;
3743 put->actions_len = facet->xout.odp_actions.size;
3745 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
3746 op->slow_stub, sizeof op->slow_stub,
3747 &put->actions, &put->actions_len);
3753 /* Handles flow miss 'miss'. May add any required datapath operations
3754 * to 'ops', incrementing '*n_ops' for each new op. */
3756 handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops,
3759 struct ofproto_dpif *ofproto = miss->ofproto;
3760 struct facet *facet;
3764 /* The caller must ensure that miss->hmap_node.hash contains
3765 * flow_hash(miss->flow, 0). */
3766 hash = miss->hmap_node.hash;
3768 facet = facet_lookup_valid(ofproto, &miss->flow, hash);
3770 /* There does not exist a bijection between 'struct flow' and datapath
3771 * flow keys with fitness ODP_FIT_TO_LITTLE. This breaks a fundamental
3772 * assumption used throughout the facet and subfacet handling code.
3773 * Since we have to handle these misses in userspace anyway, we simply
3774 * skip facet creation, avoiding the problem alltogether. */
3775 if (miss->key_fitness == ODP_FIT_TOO_LITTLE
3776 || !flow_miss_should_make_facet(ofproto, miss, hash)) {
3777 handle_flow_miss_without_facet(miss, ops, n_ops);
3781 facet = facet_create(miss, hash);
3786 handle_flow_miss_with_facet(miss, facet, now, ops, n_ops);
3789 static struct drop_key *
3790 drop_key_lookup(const struct dpif_backer *backer, const struct nlattr *key,
3793 struct drop_key *drop_key;
3795 HMAP_FOR_EACH_WITH_HASH (drop_key, hmap_node, hash_bytes(key, key_len, 0),
3796 &backer->drop_keys) {
3797 if (drop_key->key_len == key_len
3798 && !memcmp(drop_key->key, key, key_len)) {
3806 drop_key_clear(struct dpif_backer *backer)
3808 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3809 struct drop_key *drop_key, *next;
3811 HMAP_FOR_EACH_SAFE (drop_key, next, hmap_node, &backer->drop_keys) {
3814 error = dpif_flow_del(backer->dpif, drop_key->key, drop_key->key_len,
3816 if (error && !VLOG_DROP_WARN(&rl)) {
3817 struct ds ds = DS_EMPTY_INITIALIZER;
3818 odp_flow_key_format(drop_key->key, drop_key->key_len, &ds);
3819 VLOG_WARN("Failed to delete drop key (%s) (%s)", strerror(error),
3824 hmap_remove(&backer->drop_keys, &drop_key->hmap_node);
3825 free(drop_key->key);
3830 /* Given a datpath, packet, and flow metadata ('backer', 'packet', and 'key'
3831 * respectively), populates 'flow' with the result of odp_flow_key_to_flow().
3832 * Optionally, if nonnull, populates 'fitnessp' with the fitness of 'flow' as
3833 * returned by odp_flow_key_to_flow(). Also, optionally populates 'ofproto'
3834 * with the ofproto_dpif, and 'odp_in_port' with the datapath in_port, that
3835 * 'packet' ingressed.
3837 * If 'ofproto' is nonnull, requires 'flow''s in_port to exist. Otherwise sets
3838 * 'flow''s in_port to OFPP_NONE.
3840 * This function does post-processing on data returned from
3841 * odp_flow_key_to_flow() to help make VLAN splinters transparent to the rest
3842 * of the upcall processing logic. In particular, if the extracted in_port is
3843 * a VLAN splinter port, it replaces flow->in_port by the "real" port, sets
3844 * flow->vlan_tci correctly for the VLAN of the VLAN splinter port, and pushes
3845 * a VLAN header onto 'packet' (if it is nonnull).
3847 * Optionally, if 'initial_vals' is nonnull, sets 'initial_vals->vlan_tci'
3848 * to the VLAN TCI with which the packet was really received, that is, the
3849 * actual VLAN TCI extracted by odp_flow_key_to_flow(). (This differs from
3850 * the value returned in flow->vlan_tci only for packets received on
3853 * Similarly, this function also includes some logic to help with tunnels. It
3854 * may modify 'flow' as necessary to make the tunneling implementation
3855 * transparent to the upcall processing logic.
3857 * Returns 0 if successful, ENODEV if the parsed flow has no associated ofport,
3858 * or some other positive errno if there are other problems. */
3860 ofproto_receive(const struct dpif_backer *backer, struct ofpbuf *packet,
3861 const struct nlattr *key, size_t key_len,
3862 struct flow *flow, enum odp_key_fitness *fitnessp,
3863 struct ofproto_dpif **ofproto, uint32_t *odp_in_port,
3864 struct initial_vals *initial_vals)
3866 const struct ofport_dpif *port;
3867 enum odp_key_fitness fitness;
3870 fitness = odp_flow_key_to_flow(key, key_len, flow);
3871 if (fitness == ODP_FIT_ERROR) {
3877 initial_vals->vlan_tci = flow->vlan_tci;
3881 *odp_in_port = flow->in_port;
3884 port = (tnl_port_should_receive(flow)
3885 ? ofport_dpif_cast(tnl_port_receive(flow))
3886 : odp_port_to_ofport(backer, flow->in_port));
3887 flow->in_port = port ? port->up.ofp_port : OFPP_NONE;
3892 /* XXX: Since the tunnel module is not scoped per backer, for a tunnel port
3893 * it's theoretically possible that we'll receive an ofport belonging to an
3894 * entirely different datapath. In practice, this can't happen because no
3895 * platforms has two separate datapaths which each support tunneling. */
3896 ovs_assert(ofproto_dpif_cast(port->up.ofproto)->backer == backer);
3898 if (vsp_adjust_flow(ofproto_dpif_cast(port->up.ofproto), flow)) {
3900 /* Make the packet resemble the flow, so that it gets sent to
3901 * an OpenFlow controller properly, so that it looks correct
3902 * for sFlow, and so that flow_extract() will get the correct
3903 * vlan_tci if it is called on 'packet'.
3905 * The allocated space inside 'packet' probably also contains
3906 * 'key', that is, both 'packet' and 'key' are probably part of
3907 * a struct dpif_upcall (see the large comment on that
3908 * structure definition), so pushing data on 'packet' is in
3909 * general not a good idea since it could overwrite 'key' or
3910 * free it as a side effect. However, it's OK in this special
3911 * case because we know that 'packet' is inside a Netlink
3912 * attribute: pushing 4 bytes will just overwrite the 4-byte
3913 * "struct nlattr", which is fine since we don't need that
3914 * header anymore. */
3915 eth_push_vlan(packet, flow->vlan_tci);
3917 /* We can't reproduce 'key' from 'flow'. */
3918 fitness = fitness == ODP_FIT_PERFECT ? ODP_FIT_TOO_MUCH : fitness;
3923 *ofproto = ofproto_dpif_cast(port->up.ofproto);
3928 *fitnessp = fitness;
3934 handle_miss_upcalls(struct dpif_backer *backer, struct dpif_upcall *upcalls,
3937 struct dpif_upcall *upcall;
3938 struct flow_miss *miss;
3939 struct flow_miss misses[FLOW_MISS_MAX_BATCH];
3940 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
3941 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
3951 /* Construct the to-do list.
3953 * This just amounts to extracting the flow from each packet and sticking
3954 * the packets that have the same flow in the same "flow_miss" structure so
3955 * that we can process them together. */
3958 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
3959 struct flow_miss *miss = &misses[n_misses];
3960 struct flow_miss *existing_miss;
3961 struct ofproto_dpif *ofproto;
3962 uint32_t odp_in_port;
3967 error = ofproto_receive(backer, upcall->packet, upcall->key,
3968 upcall->key_len, &flow, &miss->key_fitness,
3969 &ofproto, &odp_in_port, &miss->initial_vals);
3970 if (error == ENODEV) {
3971 struct drop_key *drop_key;
3973 /* Received packet on datapath port for which we couldn't
3974 * associate an ofproto. This can happen if a port is removed
3975 * while traffic is being received. Print a rate-limited message
3976 * in case it happens frequently. Install a drop flow so
3977 * that future packets of the flow are inexpensively dropped
3979 VLOG_INFO_RL(&rl, "received packet on unassociated datapath port "
3980 "%"PRIu32, odp_in_port);
3982 drop_key = drop_key_lookup(backer, upcall->key, upcall->key_len);
3984 drop_key = xmalloc(sizeof *drop_key);
3985 drop_key->key = xmemdup(upcall->key, upcall->key_len);
3986 drop_key->key_len = upcall->key_len;
3988 hmap_insert(&backer->drop_keys, &drop_key->hmap_node,
3989 hash_bytes(drop_key->key, drop_key->key_len, 0));
3990 dpif_flow_put(backer->dpif, DPIF_FP_CREATE | DPIF_FP_MODIFY,
3991 drop_key->key, drop_key->key_len, NULL, 0, NULL);
3999 ofproto->n_missed++;
4000 flow_extract(upcall->packet, flow.skb_priority, flow.skb_mark,
4001 &flow.tunnel, flow.in_port, &miss->flow);
4003 /* Add other packets to a to-do list. */
4004 hash = flow_hash(&miss->flow, 0);
4005 existing_miss = flow_miss_find(&todo, ofproto, &miss->flow, hash);
4006 if (!existing_miss) {
4007 hmap_insert(&todo, &miss->hmap_node, hash);
4008 miss->ofproto = ofproto;
4009 miss->key = upcall->key;
4010 miss->key_len = upcall->key_len;
4011 miss->upcall_type = upcall->type;
4012 list_init(&miss->packets);
4016 miss = existing_miss;
4018 list_push_back(&miss->packets, &upcall->packet->list_node);
4021 /* Process each element in the to-do list, constructing the set of
4022 * operations to batch. */
4024 HMAP_FOR_EACH (miss, hmap_node, &todo) {
4025 handle_flow_miss(miss, flow_miss_ops, &n_ops);
4027 ovs_assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
4029 /* Execute batch. */
4030 for (i = 0; i < n_ops; i++) {
4031 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
4033 dpif_operate(backer->dpif, dpif_ops, n_ops);
4036 for (i = 0; i < n_ops; i++) {
4037 if (flow_miss_ops[i].xout_garbage) {
4038 xlate_out_uninit(&flow_miss_ops[i].xout);
4041 hmap_destroy(&todo);
4044 static enum { SFLOW_UPCALL, MISS_UPCALL, BAD_UPCALL, FLOW_SAMPLE_UPCALL,
4046 classify_upcall(const struct dpif_upcall *upcall)
4048 size_t userdata_len;
4049 union user_action_cookie cookie;
4051 /* First look at the upcall type. */
4052 switch (upcall->type) {
4053 case DPIF_UC_ACTION:
4059 case DPIF_N_UC_TYPES:
4061 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
4065 /* "action" upcalls need a closer look. */
4066 if (!upcall->userdata) {
4067 VLOG_WARN_RL(&rl, "action upcall missing cookie");
4070 userdata_len = nl_attr_get_size(upcall->userdata);
4071 if (userdata_len < sizeof cookie.type
4072 || userdata_len > sizeof cookie) {
4073 VLOG_WARN_RL(&rl, "action upcall cookie has unexpected size %zu",
4077 memset(&cookie, 0, sizeof cookie);
4078 memcpy(&cookie, nl_attr_get(upcall->userdata), userdata_len);
4079 if (userdata_len == sizeof cookie.sflow
4080 && cookie.type == USER_ACTION_COOKIE_SFLOW) {
4081 return SFLOW_UPCALL;
4082 } else if (userdata_len == sizeof cookie.slow_path
4083 && cookie.type == USER_ACTION_COOKIE_SLOW_PATH) {
4085 } else if (userdata_len == sizeof cookie.flow_sample
4086 && cookie.type == USER_ACTION_COOKIE_FLOW_SAMPLE) {
4087 return FLOW_SAMPLE_UPCALL;
4088 } else if (userdata_len == sizeof cookie.ipfix
4089 && cookie.type == USER_ACTION_COOKIE_IPFIX) {
4090 return IPFIX_UPCALL;
4092 VLOG_WARN_RL(&rl, "invalid user cookie of type %"PRIu16
4093 " and size %zu", cookie.type, userdata_len);
4099 handle_sflow_upcall(struct dpif_backer *backer,
4100 const struct dpif_upcall *upcall)
4102 struct ofproto_dpif *ofproto;
4103 union user_action_cookie cookie;
4105 uint32_t odp_in_port;
4107 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
4108 &flow, NULL, &ofproto, &odp_in_port, NULL)
4109 || !ofproto->sflow) {
4113 memset(&cookie, 0, sizeof cookie);
4114 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.sflow);
4115 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
4116 odp_in_port, &cookie);
4120 handle_flow_sample_upcall(struct dpif_backer *backer,
4121 const struct dpif_upcall *upcall)
4123 struct ofproto_dpif *ofproto;
4124 union user_action_cookie cookie;
4127 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
4128 &flow, NULL, &ofproto, NULL, NULL)
4129 || !ofproto->ipfix) {
4133 memset(&cookie, 0, sizeof cookie);
4134 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.flow_sample);
4136 /* The flow reflects exactly the contents of the packet. Sample
4137 * the packet using it. */
4138 dpif_ipfix_flow_sample(ofproto->ipfix, upcall->packet, &flow,
4139 cookie.flow_sample.collector_set_id,
4140 cookie.flow_sample.probability,
4141 cookie.flow_sample.obs_domain_id,
4142 cookie.flow_sample.obs_point_id);
4146 handle_ipfix_upcall(struct dpif_backer *backer,
4147 const struct dpif_upcall *upcall)
4149 struct ofproto_dpif *ofproto;
4152 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
4153 &flow, NULL, &ofproto, NULL, NULL)
4154 || !ofproto->ipfix) {
4158 /* The flow reflects exactly the contents of the packet. Sample
4159 * the packet using it. */
4160 dpif_ipfix_bridge_sample(ofproto->ipfix, upcall->packet, &flow);
4164 handle_upcalls(struct dpif_backer *backer, unsigned int max_batch)
4166 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
4167 struct ofpbuf miss_bufs[FLOW_MISS_MAX_BATCH];
4168 uint64_t miss_buf_stubs[FLOW_MISS_MAX_BATCH][4096 / 8];
4173 ovs_assert(max_batch <= FLOW_MISS_MAX_BATCH);
4176 for (n_processed = 0; n_processed < max_batch; n_processed++) {
4177 struct dpif_upcall *upcall = &misses[n_misses];
4178 struct ofpbuf *buf = &miss_bufs[n_misses];
4181 ofpbuf_use_stub(buf, miss_buf_stubs[n_misses],
4182 sizeof miss_buf_stubs[n_misses]);
4183 error = dpif_recv(backer->dpif, upcall, buf);
4189 switch (classify_upcall(upcall)) {
4191 /* Handle it later. */
4196 handle_sflow_upcall(backer, upcall);
4200 case FLOW_SAMPLE_UPCALL:
4201 handle_flow_sample_upcall(backer, upcall);
4206 handle_ipfix_upcall(backer, upcall);
4216 /* Handle deferred MISS_UPCALL processing. */
4217 handle_miss_upcalls(backer, misses, n_misses);
4218 for (i = 0; i < n_misses; i++) {
4219 ofpbuf_uninit(&miss_bufs[i]);
4225 /* Flow expiration. */
4227 static int subfacet_max_idle(const struct ofproto_dpif *);
4228 static void update_stats(struct dpif_backer *);
4229 static void rule_expire(struct rule_dpif *);
4230 static void expire_subfacets(struct ofproto_dpif *, int dp_max_idle);
4232 /* This function is called periodically by run(). Its job is to collect
4233 * updates for the flows that have been installed into the datapath, most
4234 * importantly when they last were used, and then use that information to
4235 * expire flows that have not been used recently.
4237 * Returns the number of milliseconds after which it should be called again. */
4239 expire(struct dpif_backer *backer)
4241 struct ofproto_dpif *ofproto;
4242 int max_idle = INT32_MAX;
4244 /* Periodically clear out the drop keys in an effort to keep them
4245 * relatively few. */
4246 drop_key_clear(backer);
4248 /* Update stats for each flow in the backer. */
4249 update_stats(backer);
4251 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4252 struct rule *rule, *next_rule;
4255 if (ofproto->backer != backer) {
4259 /* Keep track of the max number of flows per ofproto_dpif. */
4260 update_max_subfacet_count(ofproto);
4262 /* Expire subfacets that have been idle too long. */
4263 dp_max_idle = subfacet_max_idle(ofproto);
4264 expire_subfacets(ofproto, dp_max_idle);
4266 max_idle = MIN(max_idle, dp_max_idle);
4268 /* Expire OpenFlow flows whose idle_timeout or hard_timeout
4270 LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
4271 &ofproto->up.expirable) {
4272 rule_expire(rule_dpif_cast(rule));
4275 /* All outstanding data in existing flows has been accounted, so it's a
4276 * good time to do bond rebalancing. */
4277 if (ofproto->has_bonded_bundles) {
4278 struct ofbundle *bundle;
4280 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
4282 bond_rebalance(bundle->bond, &backer->revalidate_set);
4288 return MIN(max_idle, 1000);
4291 /* Updates flow table statistics given that the datapath just reported 'stats'
4292 * as 'subfacet''s statistics. */
4294 update_subfacet_stats(struct subfacet *subfacet,
4295 const struct dpif_flow_stats *stats)
4297 struct facet *facet = subfacet->facet;
4298 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4299 struct dpif_flow_stats diff;
4301 diff.tcp_flags = stats->tcp_flags;
4302 diff.used = stats->used;
4304 if (stats->n_packets >= subfacet->dp_packet_count) {
4305 diff.n_packets = stats->n_packets - subfacet->dp_packet_count;
4307 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
4311 if (stats->n_bytes >= subfacet->dp_byte_count) {
4312 diff.n_bytes = stats->n_bytes - subfacet->dp_byte_count;
4314 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
4318 ofproto->n_hit += diff.n_packets;
4319 subfacet->dp_packet_count = stats->n_packets;
4320 subfacet->dp_byte_count = stats->n_bytes;
4321 subfacet_update_stats(subfacet, &diff);
4323 if (facet->accounted_bytes < facet->byte_count) {
4325 facet_account(facet);
4326 facet->accounted_bytes = facet->byte_count;
4330 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
4331 * about, or a flow that shouldn't be installed but was anyway. Delete it. */
4333 delete_unexpected_flow(struct ofproto_dpif *ofproto,
4334 const struct nlattr *key, size_t key_len)
4336 if (!VLOG_DROP_WARN(&rl)) {
4340 odp_flow_key_format(key, key_len, &s);
4341 VLOG_WARN("unexpected flow on %s: %s", ofproto->up.name, ds_cstr(&s));
4345 COVERAGE_INC(facet_unexpected);
4346 dpif_flow_del(ofproto->backer->dpif, key, key_len, NULL);
4349 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
4351 * This function also pushes statistics updates to rules which each facet
4352 * resubmits into. Generally these statistics will be accurate. However, if a
4353 * facet changes the rule it resubmits into at some time in between
4354 * update_stats() runs, it is possible that statistics accrued to the
4355 * old rule will be incorrectly attributed to the new rule. This could be
4356 * avoided by calling update_stats() whenever rules are created or
4357 * deleted. However, the performance impact of making so many calls to the
4358 * datapath do not justify the benefit of having perfectly accurate statistics.
4360 * In addition, this function maintains per ofproto flow hit counts. The patch
4361 * port is not treated specially. e.g. A packet ingress from br0 patched into
4362 * br1 will increase the hit count of br0 by 1, however, does not affect
4363 * the hit or miss counts of br1.
4366 update_stats(struct dpif_backer *backer)
4368 const struct dpif_flow_stats *stats;
4369 struct dpif_flow_dump dump;
4370 const struct nlattr *key;
4371 struct ofproto_dpif *ofproto;
4374 dpif_flow_dump_start(&dump, backer->dpif);
4375 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
4377 struct subfacet *subfacet;
4380 if (ofproto_receive(backer, NULL, key, key_len, &flow, NULL, &ofproto,
4385 ofproto->total_subfacet_count += hmap_count(&ofproto->subfacets);
4386 ofproto->n_update_stats++;
4388 key_hash = odp_flow_key_hash(key, key_len);
4389 subfacet = subfacet_find(ofproto, key, key_len, key_hash);
4390 switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
4392 update_subfacet_stats(subfacet, stats);
4396 /* Stats are updated per-packet. */
4399 case SF_NOT_INSTALLED:
4401 delete_unexpected_flow(ofproto, key, key_len);
4406 dpif_flow_dump_done(&dump);
4408 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4409 update_moving_averages(ofproto);
4414 /* Calculates and returns the number of milliseconds of idle time after which
4415 * subfacets should expire from the datapath. When a subfacet expires, we fold
4416 * its statistics into its facet, and when a facet's last subfacet expires, we
4417 * fold its statistic into its rule. */
4419 subfacet_max_idle(const struct ofproto_dpif *ofproto)
4422 * Idle time histogram.
4424 * Most of the time a switch has a relatively small number of subfacets.
4425 * When this is the case we might as well keep statistics for all of them
4426 * in userspace and to cache them in the kernel datapath for performance as
4429 * As the number of subfacets increases, the memory required to maintain
4430 * statistics about them in userspace and in the kernel becomes
4431 * significant. However, with a large number of subfacets it is likely
4432 * that only a few of them are "heavy hitters" that consume a large amount
4433 * of bandwidth. At this point, only heavy hitters are worth caching in
4434 * the kernel and maintaining in userspaces; other subfacets we can
4437 * The technique used to compute the idle time is to build a histogram with
4438 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
4439 * that is installed in the kernel gets dropped in the appropriate bucket.
4440 * After the histogram has been built, we compute the cutoff so that only
4441 * the most-recently-used 1% of subfacets (but at least
4442 * ofproto->up.flow_eviction_threshold flows) are kept cached. At least
4443 * the most-recently-used bucket of subfacets is kept, so actually an
4444 * arbitrary number of subfacets can be kept in any given expiration run
4445 * (though the next run will delete most of those unless they receive
4448 * This requires a second pass through the subfacets, in addition to the
4449 * pass made by update_stats(), because the former function never looks at
4450 * uninstallable subfacets.
4452 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
4453 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
4454 int buckets[N_BUCKETS] = { 0 };
4455 int total, subtotal, bucket;
4456 struct subfacet *subfacet;
4460 total = hmap_count(&ofproto->subfacets);
4461 if (total <= ofproto->up.flow_eviction_threshold) {
4462 return N_BUCKETS * BUCKET_WIDTH;
4465 /* Build histogram. */
4467 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
4468 long long int idle = now - subfacet->used;
4469 int bucket = (idle <= 0 ? 0
4470 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
4471 : (unsigned int) idle / BUCKET_WIDTH);
4475 /* Find the first bucket whose flows should be expired. */
4476 subtotal = bucket = 0;
4478 subtotal += buckets[bucket++];
4479 } while (bucket < N_BUCKETS &&
4480 subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100));
4482 if (VLOG_IS_DBG_ENABLED()) {
4486 ds_put_cstr(&s, "keep");
4487 for (i = 0; i < N_BUCKETS; i++) {
4489 ds_put_cstr(&s, ", drop");
4492 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
4495 VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s));
4499 return bucket * BUCKET_WIDTH;
4503 expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle)
4505 /* Cutoff time for most flows. */
4506 long long int normal_cutoff = time_msec() - dp_max_idle;
4508 /* We really want to keep flows for special protocols around, so use a more
4509 * conservative cutoff. */
4510 long long int special_cutoff = time_msec() - 10000;
4512 struct subfacet *subfacet, *next_subfacet;
4513 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
4517 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
4518 &ofproto->subfacets) {
4519 long long int cutoff;
4521 cutoff = (subfacet->facet->xout.slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP
4525 if (subfacet->used < cutoff) {
4526 if (subfacet->path != SF_NOT_INSTALLED) {
4527 batch[n_batch++] = subfacet;
4528 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
4529 subfacet_destroy_batch(ofproto, batch, n_batch);
4533 subfacet_destroy(subfacet);
4539 subfacet_destroy_batch(ofproto, batch, n_batch);
4543 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
4544 * then delete it entirely. */
4546 rule_expire(struct rule_dpif *rule)
4548 struct facet *facet, *next_facet;
4552 if (rule->up.pending) {
4553 /* We'll have to expire it later. */
4557 /* Has 'rule' expired? */
4559 if (rule->up.hard_timeout
4560 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
4561 reason = OFPRR_HARD_TIMEOUT;
4562 } else if (rule->up.idle_timeout
4563 && now > rule->up.used + rule->up.idle_timeout * 1000) {
4564 reason = OFPRR_IDLE_TIMEOUT;
4569 COVERAGE_INC(ofproto_dpif_expired);
4571 /* Update stats. (This is a no-op if the rule expired due to an idle
4572 * timeout, because that only happens when the rule has no facets left.) */
4573 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4574 facet_remove(facet);
4577 /* Get rid of the rule. */
4578 ofproto_rule_expire(&rule->up, reason);
4583 /* Creates and returns a new facet based on 'miss'.
4585 * The caller must already have determined that no facet with an identical
4586 * 'miss->flow' exists in 'miss->ofproto'.
4588 * 'hash' must be the return value of flow_hash(miss->flow, 0).
4590 * The facet will initially have no subfacets. The caller should create (at
4591 * least) one subfacet with subfacet_create(). */
4592 static struct facet *
4593 facet_create(const struct flow_miss *miss, uint32_t hash)
4595 struct ofproto_dpif *ofproto = miss->ofproto;
4596 struct xlate_in xin;
4597 struct facet *facet;
4599 facet = xzalloc(sizeof *facet);
4600 facet->used = time_msec();
4601 facet->flow = miss->flow;
4602 facet->initial_vals = miss->initial_vals;
4603 facet->rule = rule_dpif_lookup(ofproto, &facet->flow);
4604 facet->learn_rl = time_msec() + 500;
4606 hmap_insert(&ofproto->facets, &facet->hmap_node, hash);
4607 list_push_back(&facet->rule->facets, &facet->list_node);
4608 list_init(&facet->subfacets);
4609 netflow_flow_init(&facet->nf_flow);
4610 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
4612 xlate_in_init(&xin, ofproto, &facet->flow, &facet->initial_vals,
4613 facet->rule, 0, NULL);
4614 xin.may_learn = true;
4615 xlate_actions(&xin, &facet->xout);
4616 facet->nf_flow.output_iface = facet->xout.nf_output_iface;
4622 facet_free(struct facet *facet)
4625 xlate_out_uninit(&facet->xout);
4630 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
4631 * 'packet', which arrived on 'in_port'. */
4633 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
4634 const struct nlattr *odp_actions, size_t actions_len,
4635 struct ofpbuf *packet)
4637 struct odputil_keybuf keybuf;
4641 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4642 odp_flow_key_from_flow(&key, flow,
4643 ofp_port_to_odp_port(ofproto, flow->in_port));
4645 error = dpif_execute(ofproto->backer->dpif, key.data, key.size,
4646 odp_actions, actions_len, packet);
4650 /* Remove 'facet' from 'ofproto' and free up the associated memory:
4652 * - If 'facet' was installed in the datapath, uninstalls it and updates its
4653 * rule's statistics, via subfacet_uninstall().
4655 * - Removes 'facet' from its rule and from ofproto->facets.
4658 facet_remove(struct facet *facet)
4660 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4661 struct subfacet *subfacet, *next_subfacet;
4663 ovs_assert(!list_is_empty(&facet->subfacets));
4665 /* First uninstall all of the subfacets to get final statistics. */
4666 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4667 subfacet_uninstall(subfacet);
4670 /* Flush the final stats to the rule.
4672 * This might require us to have at least one subfacet around so that we
4673 * can use its actions for accounting in facet_account(), which is why we
4674 * have uninstalled but not yet destroyed the subfacets. */
4675 facet_flush_stats(facet);
4677 /* Now we're really all done so destroy everything. */
4678 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
4679 &facet->subfacets) {
4680 subfacet_destroy__(subfacet);
4682 hmap_remove(&ofproto->facets, &facet->hmap_node);
4683 list_remove(&facet->list_node);
4687 /* Feed information from 'facet' back into the learning table to keep it in
4688 * sync with what is actually flowing through the datapath. */
4690 facet_learn(struct facet *facet)
4692 long long int now = time_msec();
4694 if (!facet->xout.has_fin_timeout && now < facet->learn_rl) {
4698 facet->learn_rl = now + 500;
4700 if (!facet->xout.has_learn
4701 && !facet->xout.has_normal
4702 && (!facet->xout.has_fin_timeout
4703 || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
4707 facet_push_stats(facet, true);
4711 facet_account(struct facet *facet)
4713 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4714 const struct nlattr *a;
4719 if (!facet->xout.has_normal || !ofproto->has_bonded_bundles) {
4722 n_bytes = facet->byte_count - facet->accounted_bytes;
4724 /* This loop feeds byte counters to bond_account() for rebalancing to use
4725 * as a basis. We also need to track the actual VLAN on which the packet
4726 * is going to be sent to ensure that it matches the one passed to
4727 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
4730 * We use the actions from an arbitrary subfacet because they should all
4731 * be equally valid for our purpose. */
4732 vlan_tci = facet->flow.vlan_tci;
4733 NL_ATTR_FOR_EACH_UNSAFE (a, left, facet->xout.odp_actions.data,
4734 facet->xout.odp_actions.size) {
4735 const struct ovs_action_push_vlan *vlan;
4736 struct ofport_dpif *port;
4738 switch (nl_attr_type(a)) {
4739 case OVS_ACTION_ATTR_OUTPUT:
4740 port = get_odp_port(ofproto, nl_attr_get_u32(a));
4741 if (port && port->bundle && port->bundle->bond) {
4742 bond_account(port->bundle->bond, &facet->flow,
4743 vlan_tci_to_vid(vlan_tci), n_bytes);
4747 case OVS_ACTION_ATTR_POP_VLAN:
4748 vlan_tci = htons(0);
4751 case OVS_ACTION_ATTR_PUSH_VLAN:
4752 vlan = nl_attr_get(a);
4753 vlan_tci = vlan->vlan_tci;
4759 /* Returns true if the only action for 'facet' is to send to the controller.
4760 * (We don't report NetFlow expiration messages for such facets because they
4761 * are just part of the control logic for the network, not real traffic). */
4763 facet_is_controller_flow(struct facet *facet)
4766 const struct rule *rule = &facet->rule->up;
4767 const struct ofpact *ofpacts = rule->ofpacts;
4768 size_t ofpacts_len = rule->ofpacts_len;
4770 if (ofpacts_len > 0 &&
4771 ofpacts->type == OFPACT_CONTROLLER &&
4772 ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len)) {
4779 /* Folds all of 'facet''s statistics into its rule. Also updates the
4780 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
4781 * 'facet''s statistics in the datapath should have been zeroed and folded into
4782 * its packet and byte counts before this function is called. */
4784 facet_flush_stats(struct facet *facet)
4786 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4787 struct subfacet *subfacet;
4789 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4790 ovs_assert(!subfacet->dp_byte_count);
4791 ovs_assert(!subfacet->dp_packet_count);
4794 facet_push_stats(facet, false);
4795 if (facet->accounted_bytes < facet->byte_count) {
4796 facet_account(facet);
4797 facet->accounted_bytes = facet->byte_count;
4800 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
4801 struct ofexpired expired;
4802 expired.flow = facet->flow;
4803 expired.packet_count = facet->packet_count;
4804 expired.byte_count = facet->byte_count;
4805 expired.used = facet->used;
4806 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4809 /* Reset counters to prevent double counting if 'facet' ever gets
4811 facet_reset_counters(facet);
4813 netflow_flow_clear(&facet->nf_flow);
4814 facet->tcp_flags = 0;
4817 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
4818 * Returns it if found, otherwise a null pointer.
4820 * 'hash' must be the return value of flow_hash(flow, 0).
4822 * The returned facet might need revalidation; use facet_lookup_valid()
4823 * instead if that is important. */
4824 static struct facet *
4825 facet_find(struct ofproto_dpif *ofproto,
4826 const struct flow *flow, uint32_t hash)
4828 struct facet *facet;
4830 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, hash, &ofproto->facets) {
4831 if (flow_equal(flow, &facet->flow)) {
4839 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
4840 * Returns it if found, otherwise a null pointer.
4842 * 'hash' must be the return value of flow_hash(flow, 0).
4844 * The returned facet is guaranteed to be valid. */
4845 static struct facet *
4846 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow,
4849 struct facet *facet;
4851 facet = facet_find(ofproto, flow, hash);
4853 && (ofproto->backer->need_revalidate
4854 || tag_set_intersects(&ofproto->backer->revalidate_set,
4856 && !facet_revalidate(facet)) {
4864 facet_check_consistency(struct facet *facet)
4866 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
4868 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4870 struct xlate_out xout;
4871 struct xlate_in xin;
4873 struct rule_dpif *rule;
4876 /* Check the rule for consistency. */
4877 rule = rule_dpif_lookup(ofproto, &facet->flow);
4878 if (rule != facet->rule) {
4879 if (!VLOG_DROP_WARN(&rl)) {
4880 struct ds s = DS_EMPTY_INITIALIZER;
4882 flow_format(&s, &facet->flow);
4883 ds_put_format(&s, ": facet associated with wrong rule (was "
4884 "table=%"PRIu8",", facet->rule->up.table_id);
4885 cls_rule_format(&facet->rule->up.cr, &s);
4886 ds_put_format(&s, ") (should have been table=%"PRIu8",",
4888 cls_rule_format(&rule->up.cr, &s);
4889 ds_put_cstr(&s, ")\n");
4896 /* Check the datapath actions for consistency. */
4897 xlate_in_init(&xin, ofproto, &facet->flow, &facet->initial_vals, rule,
4899 xlate_actions(&xin, &xout);
4901 ok = ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)
4902 && facet->xout.slow == xout.slow;
4903 if (!ok && !VLOG_DROP_WARN(&rl)) {
4904 struct ds s = DS_EMPTY_INITIALIZER;
4906 flow_format(&s, &facet->flow);
4907 ds_put_cstr(&s, ": inconsistency in facet");
4909 if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4910 ds_put_cstr(&s, " (actions were: ");
4911 format_odp_actions(&s, facet->xout.odp_actions.data,
4912 facet->xout.odp_actions.size);
4913 ds_put_cstr(&s, ") (correct actions: ");
4914 format_odp_actions(&s, xout.odp_actions.data,
4915 xout.odp_actions.size);
4916 ds_put_cstr(&s, ")");
4919 if (facet->xout.slow != xout.slow) {
4920 ds_put_format(&s, " slow path incorrect. should be %d", xout.slow);
4925 xlate_out_uninit(&xout);
4930 /* Re-searches the classifier for 'facet':
4932 * - If the rule found is different from 'facet''s current rule, moves
4933 * 'facet' to the new rule and recompiles its actions.
4935 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
4936 * where it is and recompiles its actions anyway.
4938 * - If any of 'facet''s subfacets correspond to a new flow according to
4939 * ofproto_receive(), 'facet' is removed.
4941 * Returns true if 'facet' is still valid. False if 'facet' was removed. */
4943 facet_revalidate(struct facet *facet)
4945 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4946 struct rule_dpif *new_rule;
4947 struct subfacet *subfacet;
4948 struct xlate_out xout;
4949 struct xlate_in xin;
4951 COVERAGE_INC(facet_revalidate);
4953 /* Check that child subfacets still correspond to this facet. Tunnel
4954 * configuration changes could cause a subfacet's OpenFlow in_port to
4956 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4957 struct ofproto_dpif *recv_ofproto;
4958 struct flow recv_flow;
4961 error = ofproto_receive(ofproto->backer, NULL, subfacet->key,
4962 subfacet->key_len, &recv_flow, NULL,
4963 &recv_ofproto, NULL, NULL);
4965 || recv_ofproto != ofproto
4966 || memcmp(&recv_flow, &facet->flow, sizeof recv_flow)) {
4967 facet_remove(facet);
4972 new_rule = rule_dpif_lookup(ofproto, &facet->flow);
4974 /* Calculate new datapath actions.
4976 * We do not modify any 'facet' state yet, because we might need to, e.g.,
4977 * emit a NetFlow expiration and, if so, we need to have the old state
4978 * around to properly compose it. */
4979 xlate_in_init(&xin, ofproto, &facet->flow, &facet->initial_vals, new_rule,
4981 xlate_actions(&xin, &xout);
4983 /* A facet's slow path reason should only change under dramatic
4984 * circumstances. Rather than try to update everything, it's simpler to
4985 * remove the facet and start over. */
4986 if (facet->xout.slow != xout.slow) {
4987 facet_remove(facet);
4988 xlate_out_uninit(&xout);
4992 if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4993 LIST_FOR_EACH(subfacet, list_node, &facet->subfacets) {
4994 if (subfacet->path == SF_FAST_PATH) {
4995 struct dpif_flow_stats stats;
4997 subfacet_install(subfacet, &xout.odp_actions, &stats);
4998 subfacet_update_stats(subfacet, &stats);
5002 facet_flush_stats(facet);
5004 ofpbuf_clear(&facet->xout.odp_actions);
5005 ofpbuf_put(&facet->xout.odp_actions, xout.odp_actions.data,
5006 xout.odp_actions.size);
5009 /* Update 'facet' now that we've taken care of all the old state. */
5010 facet->xout.tags = xout.tags;
5011 facet->xout.slow = xout.slow;
5012 facet->xout.has_learn = xout.has_learn;
5013 facet->xout.has_normal = xout.has_normal;
5014 facet->xout.has_fin_timeout = xout.has_fin_timeout;
5015 facet->xout.nf_output_iface = xout.nf_output_iface;
5016 facet->xout.mirrors = xout.mirrors;
5017 facet->nf_flow.output_iface = facet->xout.nf_output_iface;
5019 if (facet->rule != new_rule) {
5020 COVERAGE_INC(facet_changed_rule);
5021 list_remove(&facet->list_node);
5022 list_push_back(&new_rule->facets, &facet->list_node);
5023 facet->rule = new_rule;
5024 facet->used = new_rule->up.created;
5025 facet->prev_used = facet->used;
5028 xlate_out_uninit(&xout);
5033 facet_reset_counters(struct facet *facet)
5035 facet->packet_count = 0;
5036 facet->byte_count = 0;
5037 facet->prev_packet_count = 0;
5038 facet->prev_byte_count = 0;
5039 facet->accounted_bytes = 0;
5043 facet_push_stats(struct facet *facet, bool may_learn)
5045 struct dpif_flow_stats stats;
5047 ovs_assert(facet->packet_count >= facet->prev_packet_count);
5048 ovs_assert(facet->byte_count >= facet->prev_byte_count);
5049 ovs_assert(facet->used >= facet->prev_used);
5051 stats.n_packets = facet->packet_count - facet->prev_packet_count;
5052 stats.n_bytes = facet->byte_count - facet->prev_byte_count;
5053 stats.used = facet->used;
5054 stats.tcp_flags = facet->tcp_flags;
5056 if (may_learn || stats.n_packets || facet->used > facet->prev_used) {
5057 struct ofproto_dpif *ofproto =
5058 ofproto_dpif_cast(facet->rule->up.ofproto);
5060 struct ofport_dpif *in_port;
5061 struct xlate_in xin;
5063 facet->prev_packet_count = facet->packet_count;
5064 facet->prev_byte_count = facet->byte_count;
5065 facet->prev_used = facet->used;
5067 in_port = get_ofp_port(ofproto, facet->flow.in_port);
5068 if (in_port && in_port->tnl_port) {
5069 netdev_vport_inc_rx(in_port->up.netdev, &stats);
5072 rule_credit_stats(facet->rule, &stats);
5073 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow,
5075 netflow_flow_update_flags(&facet->nf_flow, facet->tcp_flags);
5076 update_mirror_stats(ofproto, facet->xout.mirrors, stats.n_packets,
5079 xlate_in_init(&xin, ofproto, &facet->flow, &facet->initial_vals,
5080 facet->rule, stats.tcp_flags, NULL);
5081 xin.resubmit_stats = &stats;
5082 xin.may_learn = may_learn;
5083 xlate_actions_for_side_effects(&xin);
5088 push_all_stats__(bool run_fast)
5090 static long long int rl = LLONG_MIN;
5091 struct ofproto_dpif *ofproto;
5093 if (time_msec() < rl) {
5097 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5098 struct facet *facet;
5100 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
5101 facet_push_stats(facet, false);
5108 rl = time_msec() + 100;
5112 push_all_stats(void)
5114 push_all_stats__(true);
5118 rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats)
5120 rule->packet_count += stats->n_packets;
5121 rule->byte_count += stats->n_bytes;
5122 ofproto_rule_update_used(&rule->up, stats->used);
5127 static struct subfacet *
5128 subfacet_find(struct ofproto_dpif *ofproto,
5129 const struct nlattr *key, size_t key_len, uint32_t key_hash)
5131 struct subfacet *subfacet;
5133 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
5134 &ofproto->subfacets) {
5135 if (subfacet->key_len == key_len
5136 && !memcmp(key, subfacet->key, key_len)) {
5144 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
5145 * 'key_fitness', 'key', and 'key_len' members in 'miss'. Returns the
5146 * existing subfacet if there is one, otherwise creates and returns a
5148 static struct subfacet *
5149 subfacet_create(struct facet *facet, struct flow_miss *miss,
5152 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5153 enum odp_key_fitness key_fitness = miss->key_fitness;
5154 const struct nlattr *key = miss->key;
5155 size_t key_len = miss->key_len;
5157 struct subfacet *subfacet;
5159 key_hash = odp_flow_key_hash(key, key_len);
5161 if (list_is_empty(&facet->subfacets)) {
5162 subfacet = &facet->one_subfacet;
5164 subfacet = subfacet_find(ofproto, key, key_len, key_hash);
5166 if (subfacet->facet == facet) {
5170 /* This shouldn't happen. */
5171 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
5172 subfacet_destroy(subfacet);
5175 subfacet = xmalloc(sizeof *subfacet);
5178 hmap_insert(&ofproto->subfacets, &subfacet->hmap_node, key_hash);
5179 list_push_back(&facet->subfacets, &subfacet->list_node);
5180 subfacet->facet = facet;
5181 subfacet->key_fitness = key_fitness;
5182 subfacet->key = xmemdup(key, key_len);
5183 subfacet->key_len = key_len;
5184 subfacet->used = now;
5185 subfacet->created = now;
5186 subfacet->dp_packet_count = 0;
5187 subfacet->dp_byte_count = 0;
5188 subfacet->path = SF_NOT_INSTALLED;
5190 ofproto->subfacet_add_count++;
5194 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
5195 * its facet within 'ofproto', and frees it. */
5197 subfacet_destroy__(struct subfacet *subfacet)
5199 struct facet *facet = subfacet->facet;
5200 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5202 /* Update ofproto stats before uninstall the subfacet. */
5203 ofproto->subfacet_del_count++;
5204 ofproto->total_subfacet_life_span += (time_msec() - subfacet->created);
5206 subfacet_uninstall(subfacet);
5207 hmap_remove(&ofproto->subfacets, &subfacet->hmap_node);
5208 list_remove(&subfacet->list_node);
5209 free(subfacet->key);
5210 if (subfacet != &facet->one_subfacet) {
5215 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
5216 * last remaining subfacet in its facet destroys the facet too. */
5218 subfacet_destroy(struct subfacet *subfacet)
5220 struct facet *facet = subfacet->facet;
5222 if (list_is_singleton(&facet->subfacets)) {
5223 /* facet_remove() needs at least one subfacet (it will remove it). */
5224 facet_remove(facet);
5226 subfacet_destroy__(subfacet);
5231 subfacet_destroy_batch(struct ofproto_dpif *ofproto,
5232 struct subfacet **subfacets, int n)
5234 struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH];
5235 struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH];
5236 struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH];
5239 for (i = 0; i < n; i++) {
5240 ops[i].type = DPIF_OP_FLOW_DEL;
5241 ops[i].u.flow_del.key = subfacets[i]->key;
5242 ops[i].u.flow_del.key_len = subfacets[i]->key_len;
5243 ops[i].u.flow_del.stats = &stats[i];
5247 dpif_operate(ofproto->backer->dpif, opsp, n);
5248 for (i = 0; i < n; i++) {
5249 subfacet_reset_dp_stats(subfacets[i], &stats[i]);
5250 subfacets[i]->path = SF_NOT_INSTALLED;
5251 subfacet_destroy(subfacets[i]);
5256 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
5257 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
5258 * in the datapath will be zeroed and 'stats' will be updated with traffic new
5259 * since 'subfacet' was last updated.
5261 * Returns 0 if successful, otherwise a positive errno value. */
5263 subfacet_install(struct subfacet *subfacet, const struct ofpbuf *odp_actions,
5264 struct dpif_flow_stats *stats)
5266 struct facet *facet = subfacet->facet;
5267 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5268 enum subfacet_path path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
5269 const struct nlattr *actions = odp_actions->data;
5270 size_t actions_len = odp_actions->size;
5272 uint64_t slow_path_stub[128 / 8];
5273 enum dpif_flow_put_flags flags;
5276 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
5278 flags |= DPIF_FP_ZERO_STATS;
5281 if (path == SF_SLOW_PATH) {
5282 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
5283 slow_path_stub, sizeof slow_path_stub,
5284 &actions, &actions_len);
5287 ret = dpif_flow_put(ofproto->backer->dpif, flags, subfacet->key,
5288 subfacet->key_len, actions, actions_len, stats);
5291 subfacet_reset_dp_stats(subfacet, stats);
5295 subfacet->path = path;
5300 /* If 'subfacet' is installed in the datapath, uninstalls it. */
5302 subfacet_uninstall(struct subfacet *subfacet)
5304 if (subfacet->path != SF_NOT_INSTALLED) {
5305 struct rule_dpif *rule = subfacet->facet->rule;
5306 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5307 struct dpif_flow_stats stats;
5310 error = dpif_flow_del(ofproto->backer->dpif, subfacet->key,
5311 subfacet->key_len, &stats);
5312 subfacet_reset_dp_stats(subfacet, &stats);
5314 subfacet_update_stats(subfacet, &stats);
5316 subfacet->path = SF_NOT_INSTALLED;
5318 ovs_assert(subfacet->dp_packet_count == 0);
5319 ovs_assert(subfacet->dp_byte_count == 0);
5323 /* Resets 'subfacet''s datapath statistics counters. This should be called
5324 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
5325 * non-null, it should contain the statistics returned by dpif when 'subfacet'
5326 * was reset in the datapath. 'stats' will be modified to include only
5327 * statistics new since 'subfacet' was last updated. */
5329 subfacet_reset_dp_stats(struct subfacet *subfacet,
5330 struct dpif_flow_stats *stats)
5333 && subfacet->dp_packet_count <= stats->n_packets
5334 && subfacet->dp_byte_count <= stats->n_bytes) {
5335 stats->n_packets -= subfacet->dp_packet_count;
5336 stats->n_bytes -= subfacet->dp_byte_count;
5339 subfacet->dp_packet_count = 0;
5340 subfacet->dp_byte_count = 0;
5343 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
5345 * Because of the meaning of a subfacet's counters, it only makes sense to do
5346 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
5347 * represents a packet that was sent by hand or if it represents statistics
5348 * that have been cleared out of the datapath. */
5350 subfacet_update_stats(struct subfacet *subfacet,
5351 const struct dpif_flow_stats *stats)
5353 if (stats->n_packets || stats->used > subfacet->used) {
5354 struct facet *facet = subfacet->facet;
5356 subfacet->used = MAX(subfacet->used, stats->used);
5357 facet->used = MAX(facet->used, stats->used);
5358 facet->packet_count += stats->n_packets;
5359 facet->byte_count += stats->n_bytes;
5360 facet->tcp_flags |= stats->tcp_flags;
5366 static struct rule_dpif *
5367 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow)
5369 struct rule_dpif *rule;
5371 rule = rule_dpif_lookup__(ofproto, flow, 0);
5376 return rule_dpif_miss_rule(ofproto, flow);
5379 static struct rule_dpif *
5380 rule_dpif_lookup__(struct ofproto_dpif *ofproto, const struct flow *flow,
5383 struct cls_rule *cls_rule;
5384 struct classifier *cls;
5386 if (table_id >= N_TABLES) {
5390 cls = &ofproto->up.tables[table_id].cls;
5391 if (flow->nw_frag & FLOW_NW_FRAG_ANY
5392 && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
5393 /* For OFPC_NORMAL frag_handling, we must pretend that transport ports
5394 * are unavailable. */
5395 struct flow ofpc_normal_flow = *flow;
5396 ofpc_normal_flow.tp_src = htons(0);
5397 ofpc_normal_flow.tp_dst = htons(0);
5398 cls_rule = classifier_lookup(cls, &ofpc_normal_flow);
5400 cls_rule = classifier_lookup(cls, flow);
5402 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
5405 static struct rule_dpif *
5406 rule_dpif_miss_rule(struct ofproto_dpif *ofproto, const struct flow *flow)
5408 struct ofport_dpif *port;
5410 port = get_ofp_port(ofproto, flow->in_port);
5412 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, flow->in_port);
5413 return ofproto->miss_rule;
5416 if (port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN) {
5417 return ofproto->no_packet_in_rule;
5419 return ofproto->miss_rule;
5423 complete_operation(struct rule_dpif *rule)
5425 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5427 rule_invalidate(rule);
5429 struct dpif_completion *c = xmalloc(sizeof *c);
5430 c->op = rule->up.pending;
5431 list_push_back(&ofproto->completions, &c->list_node);
5433 ofoperation_complete(rule->up.pending, 0);
5437 static struct rule *
5440 struct rule_dpif *rule = xmalloc(sizeof *rule);
5445 rule_dealloc(struct rule *rule_)
5447 struct rule_dpif *rule = rule_dpif_cast(rule_);
5452 rule_construct(struct rule *rule_)
5454 struct rule_dpif *rule = rule_dpif_cast(rule_);
5455 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5456 struct rule_dpif *victim;
5459 rule->packet_count = 0;
5460 rule->byte_count = 0;
5462 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
5463 if (victim && !list_is_empty(&victim->facets)) {
5464 struct facet *facet;
5466 rule->facets = victim->facets;
5467 list_moved(&rule->facets);
5468 LIST_FOR_EACH (facet, list_node, &rule->facets) {
5469 /* XXX: We're only clearing our local counters here. It's possible
5470 * that quite a few packets are unaccounted for in the datapath
5471 * statistics. These will be accounted to the new rule instead of
5472 * cleared as required. This could be fixed by clearing out the
5473 * datapath statistics for this facet, but currently it doesn't
5475 facet_reset_counters(facet);
5479 /* Must avoid list_moved() in this case. */
5480 list_init(&rule->facets);
5483 table_id = rule->up.table_id;
5485 rule->tag = victim->tag;
5486 } else if (table_id == 0) {
5491 miniflow_expand(&rule->up.cr.match.flow, &flow);
5492 rule->tag = rule_calculate_tag(&flow, &rule->up.cr.match.mask,
5493 ofproto->tables[table_id].basis);
5496 complete_operation(rule);
5501 rule_destruct(struct rule *rule_)
5503 struct rule_dpif *rule = rule_dpif_cast(rule_);
5504 struct facet *facet, *next_facet;
5506 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
5507 facet_revalidate(facet);
5510 complete_operation(rule);
5514 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
5516 struct rule_dpif *rule = rule_dpif_cast(rule_);
5518 /* push_all_stats() can handle flow misses which, when using the learn
5519 * action, can cause rules to be added and deleted. This can corrupt our
5520 * caller's datastructures which assume that rule_get_stats() doesn't have
5521 * an impact on the flow table. To be safe, we disable miss handling. */
5522 push_all_stats__(false);
5524 /* Start from historical data for 'rule' itself that are no longer tracked
5525 * in facets. This counts, for example, facets that have expired. */
5526 *packets = rule->packet_count;
5527 *bytes = rule->byte_count;
5531 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
5532 struct ofpbuf *packet)
5534 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5535 struct initial_vals initial_vals;
5536 struct dpif_flow_stats stats;
5537 struct xlate_out xout;
5538 struct xlate_in xin;
5540 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
5541 rule_credit_stats(rule, &stats);
5543 initial_vals.vlan_tci = flow->vlan_tci;
5544 xlate_in_init(&xin, ofproto, flow, &initial_vals, rule, stats.tcp_flags,
5546 xin.resubmit_stats = &stats;
5547 xlate_actions(&xin, &xout);
5549 execute_odp_actions(ofproto, flow, xout.odp_actions.data,
5550 xout.odp_actions.size, packet);
5552 xlate_out_uninit(&xout);
5556 rule_execute(struct rule *rule, const struct flow *flow,
5557 struct ofpbuf *packet)
5559 rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
5560 ofpbuf_delete(packet);
5565 rule_modify_actions(struct rule *rule_)
5567 struct rule_dpif *rule = rule_dpif_cast(rule_);
5569 complete_operation(rule);
5572 /* Sends 'packet' out 'ofport'.
5573 * May modify 'packet'.
5574 * Returns 0 if successful, otherwise a positive errno value. */
5576 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
5578 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
5579 uint64_t odp_actions_stub[1024 / 8];
5580 struct ofpbuf key, odp_actions;
5581 struct dpif_flow_stats stats;
5582 struct odputil_keybuf keybuf;
5583 struct ofpact_output output;
5584 struct xlate_out xout;
5585 struct xlate_in xin;
5589 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5590 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5592 /* Use OFPP_NONE as the in_port to avoid special packet processing. */
5593 flow_extract(packet, 0, 0, NULL, OFPP_NONE, &flow);
5594 odp_flow_key_from_flow(&key, &flow, ofp_port_to_odp_port(ofproto,
5596 dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
5598 ofpact_init(&output.ofpact, OFPACT_OUTPUT, sizeof output);
5599 output.port = ofport->up.ofp_port;
5602 xlate_in_init(&xin, ofproto, &flow, NULL, NULL, 0, packet);
5603 xin.ofpacts_len = sizeof output;
5604 xin.ofpacts = &output.ofpact;
5605 xin.resubmit_stats = &stats;
5606 xlate_actions(&xin, &xout);
5608 error = dpif_execute(ofproto->backer->dpif,
5610 xout.odp_actions.data, xout.odp_actions.size,
5612 xlate_out_uninit(&xout);
5615 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %s (%s)",
5616 ofproto->up.name, netdev_get_name(ofport->up.netdev),
5620 ofproto->stats.tx_packets++;
5621 ofproto->stats.tx_bytes += packet->size;
5625 /* OpenFlow to datapath action translation. */
5627 static bool may_receive(const struct ofport_dpif *, struct xlate_ctx *);
5628 static void do_xlate_actions(const struct ofpact *, size_t ofpacts_len,
5629 struct xlate_ctx *);
5630 static void xlate_normal(struct xlate_ctx *);
5632 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
5633 * The action will state 'slow' as the reason that the action is in the slow
5634 * path. (This is purely informational: it allows a human viewing "ovs-dpctl
5635 * dump-flows" output to see why a flow is in the slow path.)
5637 * The 'stub_size' bytes in 'stub' will be used to store the action.
5638 * 'stub_size' must be large enough for the action.
5640 * The action and its size will be stored in '*actionsp' and '*actions_lenp',
5643 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
5644 enum slow_path_reason slow,
5645 uint64_t *stub, size_t stub_size,
5646 const struct nlattr **actionsp, size_t *actions_lenp)
5648 union user_action_cookie cookie;
5651 cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
5652 cookie.slow_path.unused = 0;
5653 cookie.slow_path.reason = slow;
5655 ofpbuf_use_stack(&buf, stub, stub_size);
5656 if (slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP | SLOW_STP)) {
5657 uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif, UINT32_MAX);
5658 odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
5660 put_userspace_action(ofproto, &buf, flow, &cookie,
5661 sizeof cookie.slow_path);
5663 *actionsp = buf.data;
5664 *actions_lenp = buf.size;
5668 put_userspace_action(const struct ofproto_dpif *ofproto,
5669 struct ofpbuf *odp_actions,
5670 const struct flow *flow,
5671 const union user_action_cookie *cookie,
5672 const size_t cookie_size)
5676 pid = dpif_port_get_pid(ofproto->backer->dpif,
5677 ofp_port_to_odp_port(ofproto, flow->in_port));
5679 return odp_put_userspace_action(pid, cookie, cookie_size, odp_actions);
5682 /* Compose SAMPLE action for sFlow or IPFIX. The given probability is
5683 * the number of packets out of UINT32_MAX to sample. The given
5684 * cookie is passed back in the callback for each sampled packet.
5687 compose_sample_action(const struct ofproto_dpif *ofproto,
5688 struct ofpbuf *odp_actions,
5689 const struct flow *flow,
5690 const uint32_t probability,
5691 const union user_action_cookie *cookie,
5692 const size_t cookie_size)
5694 size_t sample_offset, actions_offset;
5697 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
5699 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
5701 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
5702 cookie_offset = put_userspace_action(ofproto, odp_actions, flow, cookie,
5705 nl_msg_end_nested(odp_actions, actions_offset);
5706 nl_msg_end_nested(odp_actions, sample_offset);
5707 return cookie_offset;
5711 compose_sflow_cookie(const struct ofproto_dpif *ofproto,
5712 ovs_be16 vlan_tci, uint32_t odp_port,
5713 unsigned int n_outputs, union user_action_cookie *cookie)
5717 cookie->type = USER_ACTION_COOKIE_SFLOW;
5718 cookie->sflow.vlan_tci = vlan_tci;
5720 /* See http://www.sflow.org/sflow_version_5.txt (search for "Input/output
5721 * port information") for the interpretation of cookie->output. */
5722 switch (n_outputs) {
5724 /* 0x40000000 | 256 means "packet dropped for unknown reason". */
5725 cookie->sflow.output = 0x40000000 | 256;
5729 ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port);
5731 cookie->sflow.output = ifindex;
5736 /* 0x80000000 means "multiple output ports. */
5737 cookie->sflow.output = 0x80000000 | n_outputs;
5742 /* Compose SAMPLE action for sFlow bridge sampling. */
5744 compose_sflow_action(const struct ofproto_dpif *ofproto,
5745 struct ofpbuf *odp_actions,
5746 const struct flow *flow,
5749 uint32_t probability;
5750 union user_action_cookie cookie;
5752 if (!ofproto->sflow || flow->in_port == OFPP_NONE) {
5756 probability = dpif_sflow_get_probability(ofproto->sflow);
5757 compose_sflow_cookie(ofproto, htons(0), odp_port,
5758 odp_port == OVSP_NONE ? 0 : 1, &cookie);
5760 return compose_sample_action(ofproto, odp_actions, flow, probability,
5761 &cookie, sizeof cookie.sflow);
5765 compose_flow_sample_cookie(uint16_t probability, uint32_t collector_set_id,
5766 uint32_t obs_domain_id, uint32_t obs_point_id,
5767 union user_action_cookie *cookie)
5769 cookie->type = USER_ACTION_COOKIE_FLOW_SAMPLE;
5770 cookie->flow_sample.probability = probability;
5771 cookie->flow_sample.collector_set_id = collector_set_id;
5772 cookie->flow_sample.obs_domain_id = obs_domain_id;
5773 cookie->flow_sample.obs_point_id = obs_point_id;
5777 compose_ipfix_cookie(union user_action_cookie *cookie)
5779 cookie->type = USER_ACTION_COOKIE_IPFIX;
5782 /* Compose SAMPLE action for IPFIX bridge sampling. */
5784 compose_ipfix_action(const struct ofproto_dpif *ofproto,
5785 struct ofpbuf *odp_actions,
5786 const struct flow *flow)
5788 uint32_t probability;
5789 union user_action_cookie cookie;
5791 if (!ofproto->ipfix || flow->in_port == OFPP_NONE) {
5795 probability = dpif_ipfix_get_bridge_exporter_probability(ofproto->ipfix);
5796 compose_ipfix_cookie(&cookie);
5798 compose_sample_action(ofproto, odp_actions, flow, probability,
5799 &cookie, sizeof cookie.ipfix);
5802 /* SAMPLE action for sFlow must be first action in any given list of
5803 * actions. At this point we do not have all information required to
5804 * build it. So try to build sample action as complete as possible. */
5806 add_sflow_action(struct xlate_ctx *ctx)
5808 ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto,
5809 &ctx->xout->odp_actions,
5810 &ctx->xin->flow, OVSP_NONE);
5811 ctx->sflow_odp_port = 0;
5812 ctx->sflow_n_outputs = 0;
5815 /* SAMPLE action for IPFIX must be 1st or 2nd action in any given list
5816 * of actions, eventually after the SAMPLE action for sFlow. */
5818 add_ipfix_action(struct xlate_ctx *ctx)
5820 compose_ipfix_action(ctx->ofproto, &ctx->xout->odp_actions,
5824 /* Fix SAMPLE action according to data collected while composing ODP actions.
5825 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
5826 * USERSPACE action's user-cookie which is required for sflow. */
5828 fix_sflow_action(struct xlate_ctx *ctx)
5830 const struct flow *base = &ctx->base_flow;
5831 union user_action_cookie *cookie;
5833 if (!ctx->user_cookie_offset) {
5837 cookie = ofpbuf_at(&ctx->xout->odp_actions, ctx->user_cookie_offset,
5838 sizeof cookie->sflow);
5839 ovs_assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
5841 compose_sflow_cookie(ctx->ofproto, base->vlan_tci,
5842 ctx->sflow_odp_port, ctx->sflow_n_outputs, cookie);
5846 compose_output_action__(struct xlate_ctx *ctx, uint16_t ofp_port,
5849 const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
5850 ovs_be16 flow_vlan_tci;
5851 uint32_t flow_skb_mark;
5852 uint8_t flow_nw_tos;
5853 struct priority_to_dscp *pdscp;
5854 uint32_t out_port, odp_port;
5856 /* If 'struct flow' gets additional metadata, we'll need to zero it out
5857 * before traversing a patch port. */
5858 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 20);
5861 xlate_report(ctx, "Nonexistent output port");
5863 } else if (ofport->up.pp.config & OFPUTIL_PC_NO_FWD) {
5864 xlate_report(ctx, "OFPPC_NO_FWD set, skipping output");
5866 } else if (check_stp && !stp_forward_in_state(ofport->stp_state)) {
5867 xlate_report(ctx, "STP not in forwarding state, skipping output");
5871 if (netdev_vport_is_patch(ofport->up.netdev)) {
5872 struct ofport_dpif *peer = ofport_get_peer(ofport);
5873 struct flow old_flow = ctx->xin->flow;
5874 const struct ofproto_dpif *peer_ofproto;
5875 enum slow_path_reason special;
5876 struct ofport_dpif *in_port;
5879 xlate_report(ctx, "Nonexistent patch port peer");
5883 peer_ofproto = ofproto_dpif_cast(peer->up.ofproto);
5884 if (peer_ofproto->backer != ctx->ofproto->backer) {
5885 xlate_report(ctx, "Patch port peer on a different datapath");
5889 ctx->ofproto = ofproto_dpif_cast(peer->up.ofproto);
5890 ctx->xin->flow.in_port = peer->up.ofp_port;
5891 ctx->xin->flow.metadata = htonll(0);
5892 memset(&ctx->xin->flow.tunnel, 0, sizeof ctx->xin->flow.tunnel);
5893 memset(ctx->xin->flow.regs, 0, sizeof ctx->xin->flow.regs);
5895 in_port = get_ofp_port(ctx->ofproto, ctx->xin->flow.in_port);
5896 special = process_special(ctx->ofproto, &ctx->xin->flow, in_port,
5899 ctx->xout->slow = special;
5900 } else if (!in_port || may_receive(in_port, ctx)) {
5901 if (!in_port || stp_forward_in_state(in_port->stp_state)) {
5902 xlate_table_action(ctx, ctx->xin->flow.in_port, 0, true);
5904 /* Forwarding is disabled by STP. Let OFPP_NORMAL and the
5905 * learning action look at the packet, then drop it. */
5906 struct flow old_base_flow = ctx->base_flow;
5907 size_t old_size = ctx->xout->odp_actions.size;
5908 xlate_table_action(ctx, ctx->xin->flow.in_port, 0, true);
5909 ctx->base_flow = old_base_flow;
5910 ctx->xout->odp_actions.size = old_size;
5914 ctx->xin->flow = old_flow;
5915 ctx->ofproto = ofproto_dpif_cast(ofport->up.ofproto);
5917 if (ctx->xin->resubmit_stats) {
5918 netdev_vport_inc_tx(ofport->up.netdev, ctx->xin->resubmit_stats);
5919 netdev_vport_inc_rx(peer->up.netdev, ctx->xin->resubmit_stats);
5925 flow_vlan_tci = ctx->xin->flow.vlan_tci;
5926 flow_skb_mark = ctx->xin->flow.skb_mark;
5927 flow_nw_tos = ctx->xin->flow.nw_tos;
5929 pdscp = get_priority(ofport, ctx->xin->flow.skb_priority);
5931 ctx->xin->flow.nw_tos &= ~IP_DSCP_MASK;
5932 ctx->xin->flow.nw_tos |= pdscp->dscp;
5935 if (ofport->tnl_port) {
5936 /* Save tunnel metadata so that changes made due to
5937 * the Logical (tunnel) Port are not visible for any further
5938 * matches, while explicit set actions on tunnel metadata are.
5940 struct flow_tnl flow_tnl = ctx->xin->flow.tunnel;
5941 odp_port = tnl_port_send(ofport->tnl_port, &ctx->xin->flow);
5942 if (odp_port == OVSP_NONE) {
5943 xlate_report(ctx, "Tunneling decided against output");
5944 goto out; /* restore flow_nw_tos */
5946 if (ctx->xin->flow.tunnel.ip_dst == ctx->orig_tunnel_ip_dst) {
5947 xlate_report(ctx, "Not tunneling to our own address");
5948 goto out; /* restore flow_nw_tos */
5950 if (ctx->xin->resubmit_stats) {
5951 netdev_vport_inc_tx(ofport->up.netdev, ctx->xin->resubmit_stats);
5953 out_port = odp_port;
5954 commit_odp_tunnel_action(&ctx->xin->flow, &ctx->base_flow,
5955 &ctx->xout->odp_actions);
5956 ctx->xin->flow.tunnel = flow_tnl; /* Restore tunnel metadata */
5958 odp_port = ofport->odp_port;
5959 out_port = vsp_realdev_to_vlandev(ctx->ofproto, odp_port,
5960 ctx->xin->flow.vlan_tci);
5961 if (out_port != odp_port) {
5962 ctx->xin->flow.vlan_tci = htons(0);
5964 ctx->xin->flow.skb_mark &= ~IPSEC_MARK;
5966 commit_odp_actions(&ctx->xin->flow, &ctx->base_flow,
5967 &ctx->xout->odp_actions);
5968 nl_msg_put_u32(&ctx->xout->odp_actions, OVS_ACTION_ATTR_OUTPUT, out_port);
5970 ctx->sflow_odp_port = odp_port;
5971 ctx->sflow_n_outputs++;
5972 ctx->xout->nf_output_iface = ofp_port;
5975 ctx->xin->flow.vlan_tci = flow_vlan_tci;
5976 ctx->xin->flow.skb_mark = flow_skb_mark;
5978 ctx->xin->flow.nw_tos = flow_nw_tos;
5982 compose_output_action(struct xlate_ctx *ctx, uint16_t ofp_port)
5984 compose_output_action__(ctx, ofp_port, true);
5988 tag_the_flow(struct xlate_ctx *ctx, struct rule_dpif *rule)
5990 struct ofproto_dpif *ofproto = ctx->ofproto;
5991 uint8_t table_id = ctx->table_id;
5993 if (table_id > 0 && table_id < N_TABLES) {
5994 struct table_dpif *table = &ofproto->tables[table_id];
5995 if (table->other_table) {
5996 ctx->xout->tags |= (rule && rule->tag
5998 : rule_calculate_tag(&ctx->xin->flow,
5999 &table->other_table->mask,
6005 /* Common rule processing in one place to avoid duplicating code. */
6006 static struct rule_dpif *
6007 ctx_rule_hooks(struct xlate_ctx *ctx, struct rule_dpif *rule,
6010 if (ctx->xin->resubmit_hook) {
6011 ctx->xin->resubmit_hook(ctx, rule);
6013 if (rule == NULL && may_packet_in) {
6015 * check if table configuration flags
6016 * OFPTC_TABLE_MISS_CONTROLLER, default.
6017 * OFPTC_TABLE_MISS_CONTINUE,
6018 * OFPTC_TABLE_MISS_DROP
6019 * When OF1.0, OFPTC_TABLE_MISS_CONTINUE is used. What to do?
6021 rule = rule_dpif_miss_rule(ctx->ofproto, &ctx->xin->flow);
6023 if (rule && ctx->xin->resubmit_stats) {
6024 rule_credit_stats(rule, ctx->xin->resubmit_stats);
6030 xlate_table_action(struct xlate_ctx *ctx,
6031 uint16_t in_port, uint8_t table_id, bool may_packet_in)
6033 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
6034 struct rule_dpif *rule;
6035 uint16_t old_in_port = ctx->xin->flow.in_port;
6036 uint8_t old_table_id = ctx->table_id;
6038 ctx->table_id = table_id;
6040 /* Look up a flow with 'in_port' as the input port. */
6041 ctx->xin->flow.in_port = in_port;
6042 rule = rule_dpif_lookup__(ctx->ofproto, &ctx->xin->flow, table_id);
6044 tag_the_flow(ctx, rule);
6046 /* Restore the original input port. Otherwise OFPP_NORMAL and
6047 * OFPP_IN_PORT will have surprising behavior. */
6048 ctx->xin->flow.in_port = old_in_port;
6050 rule = ctx_rule_hooks(ctx, rule, may_packet_in);
6053 struct rule_dpif *old_rule = ctx->rule;
6057 do_xlate_actions(rule->up.ofpacts, rule->up.ofpacts_len, ctx);
6058 ctx->rule = old_rule;
6062 ctx->table_id = old_table_id;
6064 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
6066 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
6067 MAX_RESUBMIT_RECURSION);
6068 ctx->max_resubmit_trigger = true;
6073 xlate_ofpact_resubmit(struct xlate_ctx *ctx,
6074 const struct ofpact_resubmit *resubmit)
6079 in_port = resubmit->in_port;
6080 if (in_port == OFPP_IN_PORT) {
6081 in_port = ctx->xin->flow.in_port;
6084 table_id = resubmit->table_id;
6085 if (table_id == 255) {
6086 table_id = ctx->table_id;
6089 xlate_table_action(ctx, in_port, table_id, false);
6093 flood_packets(struct xlate_ctx *ctx, bool all)
6095 struct ofport_dpif *ofport;
6097 HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) {
6098 uint16_t ofp_port = ofport->up.ofp_port;
6100 if (ofp_port == ctx->xin->flow.in_port) {
6105 compose_output_action__(ctx, ofp_port, false);
6106 } else if (!(ofport->up.pp.config & OFPUTIL_PC_NO_FLOOD)) {
6107 compose_output_action(ctx, ofp_port);
6111 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
6115 execute_controller_action(struct xlate_ctx *ctx, int len,
6116 enum ofp_packet_in_reason reason,
6117 uint16_t controller_id)
6119 struct ofputil_packet_in pin;
6120 struct ofpbuf *packet;
6122 ovs_assert(!ctx->xout->slow || ctx->xout->slow == SLOW_CONTROLLER);
6123 ctx->xout->slow = SLOW_CONTROLLER;
6124 if (!ctx->xin->packet) {
6128 packet = ofpbuf_clone(ctx->xin->packet);
6130 if (packet->l2 && packet->l3) {
6131 struct eth_header *eh;
6132 uint16_t mpls_depth;
6134 eth_pop_vlan(packet);
6137 memcpy(eh->eth_src, ctx->xin->flow.dl_src, sizeof eh->eth_src);
6138 memcpy(eh->eth_dst, ctx->xin->flow.dl_dst, sizeof eh->eth_dst);
6140 if (ctx->xin->flow.vlan_tci & htons(VLAN_CFI)) {
6141 eth_push_vlan(packet, ctx->xin->flow.vlan_tci);
6144 mpls_depth = eth_mpls_depth(packet);
6146 if (mpls_depth < ctx->xin->flow.mpls_depth) {
6147 push_mpls(packet, ctx->xin->flow.dl_type, ctx->xin->flow.mpls_lse);
6148 } else if (mpls_depth > ctx->xin->flow.mpls_depth) {
6149 pop_mpls(packet, ctx->xin->flow.dl_type);
6150 } else if (mpls_depth) {
6151 set_mpls_lse(packet, ctx->xin->flow.mpls_lse);
6155 if (ctx->xin->flow.dl_type == htons(ETH_TYPE_IP)) {
6156 packet_set_ipv4(packet, ctx->xin->flow.nw_src,
6157 ctx->xin->flow.nw_dst, ctx->xin->flow.nw_tos,
6158 ctx->xin->flow.nw_ttl);
6162 if (ctx->xin->flow.nw_proto == IPPROTO_TCP) {
6163 packet_set_tcp_port(packet, ctx->xin->flow.tp_src,
6164 ctx->xin->flow.tp_dst);
6165 } else if (ctx->xin->flow.nw_proto == IPPROTO_UDP) {
6166 packet_set_udp_port(packet, ctx->xin->flow.tp_src,
6167 ctx->xin->flow.tp_dst);
6173 pin.packet = packet->data;
6174 pin.packet_len = packet->size;
6175 pin.reason = reason;
6176 pin.controller_id = controller_id;
6177 pin.table_id = ctx->table_id;
6178 pin.cookie = ctx->rule ? ctx->rule->up.flow_cookie : 0;
6181 flow_get_metadata(&ctx->xin->flow, &pin.fmd);
6183 connmgr_send_packet_in(ctx->ofproto->up.connmgr, &pin);
6184 ofpbuf_delete(packet);
6188 execute_mpls_push_action(struct xlate_ctx *ctx, ovs_be16 eth_type)
6190 ovs_assert(eth_type_mpls(eth_type));
6192 if (ctx->base_flow.mpls_depth) {
6193 ctx->xin->flow.mpls_lse &= ~htonl(MPLS_BOS_MASK);
6194 ctx->xin->flow.mpls_depth++;
6199 if (ctx->xin->flow.dl_type == htons(ETH_TYPE_IPV6)) {
6200 label = htonl(0x2); /* IPV6 Explicit Null. */
6202 label = htonl(0x0); /* IPV4 Explicit Null. */
6204 tc = (ctx->xin->flow.nw_tos & IP_DSCP_MASK) >> 2;
6205 ttl = ctx->xin->flow.nw_ttl ? ctx->xin->flow.nw_ttl : 0x40;
6206 ctx->xin->flow.mpls_lse = set_mpls_lse_values(ttl, tc, 1, label);
6207 ctx->xin->flow.mpls_depth = 1;
6209 ctx->xin->flow.dl_type = eth_type;
6213 execute_mpls_pop_action(struct xlate_ctx *ctx, ovs_be16 eth_type)
6215 ovs_assert(eth_type_mpls(ctx->xin->flow.dl_type));
6216 ovs_assert(!eth_type_mpls(eth_type));
6218 if (ctx->xin->flow.mpls_depth) {
6219 ctx->xin->flow.mpls_depth--;
6220 ctx->xin->flow.mpls_lse = htonl(0);
6221 if (!ctx->xin->flow.mpls_depth) {
6222 ctx->xin->flow.dl_type = eth_type;
6228 compose_dec_ttl(struct xlate_ctx *ctx, struct ofpact_cnt_ids *ids)
6230 if (ctx->xin->flow.dl_type != htons(ETH_TYPE_IP) &&
6231 ctx->xin->flow.dl_type != htons(ETH_TYPE_IPV6)) {
6235 if (ctx->xin->flow.nw_ttl > 1) {
6236 ctx->xin->flow.nw_ttl--;
6241 for (i = 0; i < ids->n_controllers; i++) {
6242 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL,
6246 /* Stop processing for current table. */
6252 execute_set_mpls_ttl_action(struct xlate_ctx *ctx, uint8_t ttl)
6254 if (!eth_type_mpls(ctx->xin->flow.dl_type)) {
6258 set_mpls_lse_ttl(&ctx->xin->flow.mpls_lse, ttl);
6263 execute_dec_mpls_ttl_action(struct xlate_ctx *ctx)
6265 uint8_t ttl = mpls_lse_to_ttl(ctx->xin->flow.mpls_lse);
6267 if (!eth_type_mpls(ctx->xin->flow.dl_type)) {
6273 set_mpls_lse_ttl(&ctx->xin->flow.mpls_lse, ttl);
6276 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL, 0);
6278 /* Stop processing for current table. */
6284 xlate_output_action(struct xlate_ctx *ctx,
6285 uint16_t port, uint16_t max_len, bool may_packet_in)
6287 uint16_t prev_nf_output_iface = ctx->xout->nf_output_iface;
6289 ctx->xout->nf_output_iface = NF_OUT_DROP;
6293 compose_output_action(ctx, ctx->xin->flow.in_port);
6296 xlate_table_action(ctx, ctx->xin->flow.in_port, 0, may_packet_in);
6302 flood_packets(ctx, false);
6305 flood_packets(ctx, true);
6307 case OFPP_CONTROLLER:
6308 execute_controller_action(ctx, max_len, OFPR_ACTION, 0);
6314 if (port != ctx->xin->flow.in_port) {
6315 compose_output_action(ctx, port);
6317 xlate_report(ctx, "skipping output to input port");
6322 if (prev_nf_output_iface == NF_OUT_FLOOD) {
6323 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
6324 } else if (ctx->xout->nf_output_iface == NF_OUT_DROP) {
6325 ctx->xout->nf_output_iface = prev_nf_output_iface;
6326 } else if (prev_nf_output_iface != NF_OUT_DROP &&
6327 ctx->xout->nf_output_iface != NF_OUT_FLOOD) {
6328 ctx->xout->nf_output_iface = NF_OUT_MULTI;
6333 xlate_output_reg_action(struct xlate_ctx *ctx,
6334 const struct ofpact_output_reg *or)
6336 uint64_t port = mf_get_subfield(&or->src, &ctx->xin->flow);
6337 if (port <= UINT16_MAX) {
6338 xlate_output_action(ctx, port, or->max_len, false);
6343 xlate_enqueue_action(struct xlate_ctx *ctx,
6344 const struct ofpact_enqueue *enqueue)
6346 uint16_t ofp_port = enqueue->port;
6347 uint32_t queue_id = enqueue->queue;
6348 uint32_t flow_priority, priority;
6351 /* Translate queue to priority. */
6352 error = dpif_queue_to_priority(ctx->ofproto->backer->dpif,
6353 queue_id, &priority);
6355 /* Fall back to ordinary output action. */
6356 xlate_output_action(ctx, enqueue->port, 0, false);
6360 /* Check output port. */
6361 if (ofp_port == OFPP_IN_PORT) {
6362 ofp_port = ctx->xin->flow.in_port;
6363 } else if (ofp_port == ctx->xin->flow.in_port) {
6367 /* Add datapath actions. */
6368 flow_priority = ctx->xin->flow.skb_priority;
6369 ctx->xin->flow.skb_priority = priority;
6370 compose_output_action(ctx, ofp_port);
6371 ctx->xin->flow.skb_priority = flow_priority;
6373 /* Update NetFlow output port. */
6374 if (ctx->xout->nf_output_iface == NF_OUT_DROP) {
6375 ctx->xout->nf_output_iface = ofp_port;
6376 } else if (ctx->xout->nf_output_iface != NF_OUT_FLOOD) {
6377 ctx->xout->nf_output_iface = NF_OUT_MULTI;
6382 xlate_set_queue_action(struct xlate_ctx *ctx, uint32_t queue_id)
6384 uint32_t skb_priority;
6386 if (!dpif_queue_to_priority(ctx->ofproto->backer->dpif,
6387 queue_id, &skb_priority)) {
6388 ctx->xin->flow.skb_priority = skb_priority;
6390 /* Couldn't translate queue to a priority. Nothing to do. A warning
6391 * has already been logged. */
6396 slave_enabled_cb(uint16_t ofp_port, void *ofproto_)
6398 struct ofproto_dpif *ofproto = ofproto_;
6399 struct ofport_dpif *port;
6409 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
6412 port = get_ofp_port(ofproto, ofp_port);
6413 return port ? port->may_enable : false;
6418 xlate_bundle_action(struct xlate_ctx *ctx,
6419 const struct ofpact_bundle *bundle)
6423 port = bundle_execute(bundle, &ctx->xin->flow, slave_enabled_cb,
6425 if (bundle->dst.field) {
6426 nxm_reg_load(&bundle->dst, port, &ctx->xin->flow);
6428 xlate_output_action(ctx, port, 0, false);
6433 xlate_learn_action(struct xlate_ctx *ctx,
6434 const struct ofpact_learn *learn)
6436 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
6437 struct ofputil_flow_mod fm;
6438 uint64_t ofpacts_stub[1024 / 8];
6439 struct ofpbuf ofpacts;
6442 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
6443 learn_execute(learn, &ctx->xin->flow, &fm, &ofpacts);
6445 error = ofproto_flow_mod(&ctx->ofproto->up, &fm);
6446 if (error && !VLOG_DROP_WARN(&rl)) {
6447 VLOG_WARN("learning action failed to modify flow table (%s)",
6448 ofperr_get_name(error));
6451 ofpbuf_uninit(&ofpacts);
6454 /* Reduces '*timeout' to no more than 'max'. A value of zero in either case
6455 * means "infinite". */
6457 reduce_timeout(uint16_t max, uint16_t *timeout)
6459 if (max && (!*timeout || *timeout > max)) {
6465 xlate_fin_timeout(struct xlate_ctx *ctx,
6466 const struct ofpact_fin_timeout *oft)
6468 if (ctx->xin->tcp_flags & (TCP_FIN | TCP_RST) && ctx->rule) {
6469 struct rule_dpif *rule = ctx->rule;
6471 reduce_timeout(oft->fin_idle_timeout, &rule->up.idle_timeout);
6472 reduce_timeout(oft->fin_hard_timeout, &rule->up.hard_timeout);
6477 xlate_sample_action(struct xlate_ctx *ctx,
6478 const struct ofpact_sample *os)
6480 union user_action_cookie cookie;
6481 /* Scale the probability from 16-bit to 32-bit while representing
6482 * the same percentage. */
6483 uint32_t probability = (os->probability << 16) | os->probability;
6485 commit_odp_actions(&ctx->xin->flow, &ctx->base_flow,
6486 &ctx->xout->odp_actions);
6488 compose_flow_sample_cookie(os->probability, os->collector_set_id,
6489 os->obs_domain_id, os->obs_point_id, &cookie);
6490 compose_sample_action(ctx->ofproto, &ctx->xout->odp_actions, &ctx->xin->flow,
6491 probability, &cookie, sizeof cookie.flow_sample);
6495 may_receive(const struct ofport_dpif *port, struct xlate_ctx *ctx)
6497 if (port->up.pp.config & (eth_addr_equals(ctx->xin->flow.dl_dst,
6499 ? OFPUTIL_PC_NO_RECV_STP
6500 : OFPUTIL_PC_NO_RECV)) {
6504 /* Only drop packets here if both forwarding and learning are
6505 * disabled. If just learning is enabled, we need to have
6506 * OFPP_NORMAL and the learning action have a look at the packet
6507 * before we can drop it. */
6508 if (!stp_forward_in_state(port->stp_state)
6509 && !stp_learn_in_state(port->stp_state)) {
6517 tunnel_ecn_ok(struct xlate_ctx *ctx)
6519 if (is_ip_any(&ctx->base_flow)
6520 && (ctx->xin->flow.tunnel.ip_tos & IP_ECN_MASK) == IP_ECN_CE) {
6521 if ((ctx->base_flow.nw_tos & IP_ECN_MASK) == IP_ECN_NOT_ECT) {
6522 VLOG_WARN_RL(&rl, "dropping tunnel packet marked ECN CE"
6523 " but is not ECN capable");
6526 /* Set the ECN CE value in the tunneled packet. */
6527 ctx->xin->flow.nw_tos |= IP_ECN_CE;
6535 do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len,
6536 struct xlate_ctx *ctx)
6538 bool was_evictable = true;
6539 const struct ofpact *a;
6542 /* Don't let the rule we're working on get evicted underneath us. */
6543 was_evictable = ctx->rule->up.evictable;
6544 ctx->rule->up.evictable = false;
6547 do_xlate_actions_again:
6548 OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) {
6549 struct ofpact_controller *controller;
6550 const struct ofpact_metadata *metadata;
6558 xlate_output_action(ctx, ofpact_get_OUTPUT(a)->port,
6559 ofpact_get_OUTPUT(a)->max_len, true);
6562 case OFPACT_CONTROLLER:
6563 controller = ofpact_get_CONTROLLER(a);
6564 execute_controller_action(ctx, controller->max_len,
6566 controller->controller_id);
6569 case OFPACT_ENQUEUE:
6570 xlate_enqueue_action(ctx, ofpact_get_ENQUEUE(a));
6573 case OFPACT_SET_VLAN_VID:
6574 ctx->xin->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
6575 ctx->xin->flow.vlan_tci |=
6576 (htons(ofpact_get_SET_VLAN_VID(a)->vlan_vid)
6580 case OFPACT_SET_VLAN_PCP:
6581 ctx->xin->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
6582 ctx->xin->flow.vlan_tci |=
6583 htons((ofpact_get_SET_VLAN_PCP(a)->vlan_pcp << VLAN_PCP_SHIFT)
6587 case OFPACT_STRIP_VLAN:
6588 ctx->xin->flow.vlan_tci = htons(0);
6591 case OFPACT_PUSH_VLAN:
6592 /* XXX 802.1AD(QinQ) */
6593 ctx->xin->flow.vlan_tci = htons(VLAN_CFI);
6596 case OFPACT_SET_ETH_SRC:
6597 memcpy(ctx->xin->flow.dl_src, ofpact_get_SET_ETH_SRC(a)->mac,
6601 case OFPACT_SET_ETH_DST:
6602 memcpy(ctx->xin->flow.dl_dst, ofpact_get_SET_ETH_DST(a)->mac,
6606 case OFPACT_SET_IPV4_SRC:
6607 if (ctx->xin->flow.dl_type == htons(ETH_TYPE_IP)) {
6608 ctx->xin->flow.nw_src = ofpact_get_SET_IPV4_SRC(a)->ipv4;
6612 case OFPACT_SET_IPV4_DST:
6613 if (ctx->xin->flow.dl_type == htons(ETH_TYPE_IP)) {
6614 ctx->xin->flow.nw_dst = ofpact_get_SET_IPV4_DST(a)->ipv4;
6618 case OFPACT_SET_IPV4_DSCP:
6619 /* OpenFlow 1.0 only supports IPv4. */
6620 if (ctx->xin->flow.dl_type == htons(ETH_TYPE_IP)) {
6621 ctx->xin->flow.nw_tos &= ~IP_DSCP_MASK;
6622 ctx->xin->flow.nw_tos |= ofpact_get_SET_IPV4_DSCP(a)->dscp;
6626 case OFPACT_SET_L4_SRC_PORT:
6627 if (is_ip_any(&ctx->xin->flow)) {
6628 ctx->xin->flow.tp_src =
6629 htons(ofpact_get_SET_L4_SRC_PORT(a)->port);
6633 case OFPACT_SET_L4_DST_PORT:
6634 if (is_ip_any(&ctx->xin->flow)) {
6635 ctx->xin->flow.tp_dst =
6636 htons(ofpact_get_SET_L4_DST_PORT(a)->port);
6640 case OFPACT_RESUBMIT:
6641 xlate_ofpact_resubmit(ctx, ofpact_get_RESUBMIT(a));
6644 case OFPACT_SET_TUNNEL:
6645 ctx->xin->flow.tunnel.tun_id =
6646 htonll(ofpact_get_SET_TUNNEL(a)->tun_id);
6649 case OFPACT_SET_QUEUE:
6650 xlate_set_queue_action(ctx, ofpact_get_SET_QUEUE(a)->queue_id);
6653 case OFPACT_POP_QUEUE:
6654 ctx->xin->flow.skb_priority = ctx->orig_skb_priority;
6657 case OFPACT_REG_MOVE:
6658 nxm_execute_reg_move(ofpact_get_REG_MOVE(a), &ctx->xin->flow);
6661 case OFPACT_REG_LOAD:
6662 nxm_execute_reg_load(ofpact_get_REG_LOAD(a), &ctx->xin->flow);
6665 case OFPACT_STACK_PUSH:
6666 nxm_execute_stack_push(ofpact_get_STACK_PUSH(a), &ctx->xin->flow,
6670 case OFPACT_STACK_POP:
6671 nxm_execute_stack_pop(ofpact_get_STACK_POP(a), &ctx->xin->flow,
6675 case OFPACT_PUSH_MPLS:
6676 execute_mpls_push_action(ctx, ofpact_get_PUSH_MPLS(a)->ethertype);
6679 case OFPACT_POP_MPLS:
6680 execute_mpls_pop_action(ctx, ofpact_get_POP_MPLS(a)->ethertype);
6683 case OFPACT_SET_MPLS_TTL:
6684 if (execute_set_mpls_ttl_action(ctx,
6685 ofpact_get_SET_MPLS_TTL(a)->ttl)) {
6690 case OFPACT_DEC_MPLS_TTL:
6691 if (execute_dec_mpls_ttl_action(ctx)) {
6696 case OFPACT_DEC_TTL:
6697 if (compose_dec_ttl(ctx, ofpact_get_DEC_TTL(a))) {
6703 /* Nothing to do. */
6706 case OFPACT_MULTIPATH:
6707 multipath_execute(ofpact_get_MULTIPATH(a), &ctx->xin->flow);
6711 ctx->ofproto->has_bundle_action = true;
6712 xlate_bundle_action(ctx, ofpact_get_BUNDLE(a));
6715 case OFPACT_OUTPUT_REG:
6716 xlate_output_reg_action(ctx, ofpact_get_OUTPUT_REG(a));
6720 ctx->xout->has_learn = true;
6721 if (ctx->xin->may_learn) {
6722 xlate_learn_action(ctx, ofpact_get_LEARN(a));
6730 case OFPACT_FIN_TIMEOUT:
6731 ctx->xout->has_fin_timeout = true;
6732 xlate_fin_timeout(ctx, ofpact_get_FIN_TIMEOUT(a));
6735 case OFPACT_CLEAR_ACTIONS:
6737 * Nothing to do because writa-actions is not supported for now.
6738 * When writa-actions is supported, clear-actions also must
6739 * be supported at the same time.
6743 case OFPACT_WRITE_METADATA:
6744 metadata = ofpact_get_WRITE_METADATA(a);
6745 ctx->xin->flow.metadata &= ~metadata->mask;
6746 ctx->xin->flow.metadata |= metadata->metadata & metadata->mask;
6749 case OFPACT_GOTO_TABLE: {
6750 /* It is assumed that goto-table is the last action. */
6751 struct ofpact_goto_table *ogt = ofpact_get_GOTO_TABLE(a);
6752 struct rule_dpif *rule;
6754 ovs_assert(ctx->table_id < ogt->table_id);
6756 ctx->table_id = ogt->table_id;
6758 /* Look up a flow from the new table. */
6759 rule = rule_dpif_lookup__(ctx->ofproto, &ctx->xin->flow, ctx->table_id);
6761 tag_the_flow(ctx, rule);
6763 rule = ctx_rule_hooks(ctx, rule, true);
6767 ctx->rule->up.evictable = was_evictable;
6770 was_evictable = rule->up.evictable;
6771 rule->up.evictable = false;
6773 /* Tail recursion removal. */
6774 ofpacts = rule->up.ofpacts;
6775 ofpacts_len = rule->up.ofpacts_len;
6776 goto do_xlate_actions_again;
6782 xlate_sample_action(ctx, ofpact_get_SAMPLE(a));
6789 ctx->rule->up.evictable = was_evictable;
6794 xlate_in_init(struct xlate_in *xin, struct ofproto_dpif *ofproto,
6795 const struct flow *flow,
6796 const struct initial_vals *initial_vals,
6797 struct rule_dpif *rule, uint8_t tcp_flags,
6798 const struct ofpbuf *packet)
6800 xin->ofproto = ofproto;
6802 xin->packet = packet;
6803 xin->may_learn = packet != NULL;
6805 xin->ofpacts = NULL;
6806 xin->ofpacts_len = 0;
6807 xin->tcp_flags = tcp_flags;
6808 xin->resubmit_hook = NULL;
6809 xin->report_hook = NULL;
6810 xin->resubmit_stats = NULL;
6813 xin->initial_vals = *initial_vals;
6815 xin->initial_vals.vlan_tci = xin->flow.vlan_tci;
6820 xlate_out_uninit(struct xlate_out *xout)
6823 ofpbuf_uninit(&xout->odp_actions);
6827 /* Translates the 'ofpacts_len' bytes of "struct ofpacts" starting at 'ofpacts'
6828 * into datapath actions in 'odp_actions', using 'ctx'. */
6830 xlate_actions(struct xlate_in *xin, struct xlate_out *xout)
6832 /* Normally false. Set to true if we ever hit MAX_RESUBMIT_RECURSION, so
6833 * that in the future we always keep a copy of the original flow for
6834 * tracing purposes. */
6835 static bool hit_resubmit_limit;
6837 enum slow_path_reason special;
6838 const struct ofpact *ofpacts;
6839 struct ofport_dpif *in_port;
6840 struct flow orig_flow;
6841 struct xlate_ctx ctx;
6844 COVERAGE_INC(ofproto_dpif_xlate);
6846 /* Flow initialization rules:
6847 * - 'base_flow' must match the kernel's view of the packet at the
6848 * time that action processing starts. 'flow' represents any
6849 * transformations we wish to make through actions.
6850 * - By default 'base_flow' and 'flow' are the same since the input
6851 * packet matches the output before any actions are applied.
6852 * - When using VLAN splinters, 'base_flow''s VLAN is set to the value
6853 * of the received packet as seen by the kernel. If we later output
6854 * to another device without any modifications this will cause us to
6855 * insert a new tag since the original one was stripped off by the
6857 * - Tunnel metadata as received is retained in 'flow'. This allows
6858 * tunnel metadata matching also in later tables.
6859 * Since a kernel action for setting the tunnel metadata will only be
6860 * generated with actual tunnel output, changing the tunnel metadata
6861 * values in 'flow' (such as tun_id) will only have effect with a later
6862 * tunnel output action.
6863 * - Tunnel 'base_flow' is completely cleared since that is what the
6864 * kernel does. If we wish to maintain the original values an action
6865 * needs to be generated. */
6870 ctx.ofproto = xin->ofproto;
6871 ctx.rule = xin->rule;
6873 ctx.base_flow = ctx.xin->flow;
6874 ctx.base_flow.vlan_tci = xin->initial_vals.vlan_tci;
6875 memset(&ctx.base_flow.tunnel, 0, sizeof ctx.base_flow.tunnel);
6876 ctx.orig_tunnel_ip_dst = ctx.xin->flow.tunnel.ip_dst;
6880 ctx.xout->has_learn = false;
6881 ctx.xout->has_normal = false;
6882 ctx.xout->has_fin_timeout = false;
6883 ctx.xout->nf_output_iface = NF_OUT_DROP;
6884 ctx.xout->mirrors = 0;
6886 ofpbuf_use_stub(&ctx.xout->odp_actions, ctx.xout->odp_actions_stub,
6887 sizeof ctx.xout->odp_actions_stub);
6888 ofpbuf_reserve(&ctx.xout->odp_actions, NL_A_U32_SIZE);
6891 ctx.max_resubmit_trigger = false;
6892 ctx.orig_skb_priority = ctx.xin->flow.skb_priority;
6897 ofpacts = xin->ofpacts;
6898 ofpacts_len = xin->ofpacts_len;
6899 } else if (xin->rule) {
6900 ofpacts = xin->rule->up.ofpacts;
6901 ofpacts_len = xin->rule->up.ofpacts_len;
6906 ofpbuf_use_stub(&ctx.stack, ctx.init_stack, sizeof ctx.init_stack);
6908 if (ctx.ofproto->has_mirrors || hit_resubmit_limit) {
6909 /* Do this conditionally because the copy is expensive enough that it
6910 * shows up in profiles. */
6911 orig_flow = ctx.xin->flow;
6914 if (ctx.xin->flow.nw_frag & FLOW_NW_FRAG_ANY) {
6915 switch (ctx.ofproto->up.frag_handling) {
6916 case OFPC_FRAG_NORMAL:
6917 /* We must pretend that transport ports are unavailable. */
6918 ctx.xin->flow.tp_src = ctx.base_flow.tp_src = htons(0);
6919 ctx.xin->flow.tp_dst = ctx.base_flow.tp_dst = htons(0);
6922 case OFPC_FRAG_DROP:
6925 case OFPC_FRAG_REASM:
6928 case OFPC_FRAG_NX_MATCH:
6929 /* Nothing to do. */
6932 case OFPC_INVALID_TTL_TO_CONTROLLER:
6937 in_port = get_ofp_port(ctx.ofproto, ctx.xin->flow.in_port);
6938 special = process_special(ctx.ofproto, &ctx.xin->flow, in_port,
6941 ctx.xout->slow = special;
6943 static struct vlog_rate_limit trace_rl = VLOG_RATE_LIMIT_INIT(1, 1);
6944 struct initial_vals initial_vals;
6945 size_t sample_actions_len;
6946 uint32_t local_odp_port;
6948 initial_vals.vlan_tci = ctx.base_flow.vlan_tci;
6950 add_sflow_action(&ctx);
6951 add_ipfix_action(&ctx);
6952 sample_actions_len = ctx.xout->odp_actions.size;
6954 if (tunnel_ecn_ok(&ctx) && (!in_port || may_receive(in_port, &ctx))) {
6955 do_xlate_actions(ofpacts, ofpacts_len, &ctx);
6957 /* We've let OFPP_NORMAL and the learning action look at the
6958 * packet, so drop it now if forwarding is disabled. */
6959 if (in_port && !stp_forward_in_state(in_port->stp_state)) {
6960 ctx.xout->odp_actions.size = sample_actions_len;
6964 if (ctx.max_resubmit_trigger && !ctx.xin->resubmit_hook) {
6965 if (!hit_resubmit_limit) {
6966 /* We didn't record the original flow. Make sure we do from
6968 hit_resubmit_limit = true;
6969 } else if (!VLOG_DROP_ERR(&trace_rl)) {
6970 struct ds ds = DS_EMPTY_INITIALIZER;
6972 ofproto_trace(ctx.ofproto, &orig_flow, ctx.xin->packet,
6973 &initial_vals, &ds);
6974 VLOG_ERR("Trace triggered by excessive resubmit "
6975 "recursion:\n%s", ds_cstr(&ds));
6980 local_odp_port = ofp_port_to_odp_port(ctx.ofproto, OFPP_LOCAL);
6981 if (!connmgr_must_output_local(ctx.ofproto->up.connmgr, &ctx.xin->flow,
6983 ctx.xout->odp_actions.data,
6984 ctx.xout->odp_actions.size)) {
6985 compose_output_action(&ctx, OFPP_LOCAL);
6987 if (ctx.ofproto->has_mirrors) {
6988 add_mirror_actions(&ctx, &orig_flow);
6990 fix_sflow_action(&ctx);
6993 ofpbuf_uninit(&ctx.stack);
6996 /* Translates the 'ofpacts_len' bytes of "struct ofpact"s starting at 'ofpacts'
6997 * into datapath actions, using 'ctx', and discards the datapath actions. */
6999 xlate_actions_for_side_effects(struct xlate_in *xin)
7001 struct xlate_out xout;
7003 xlate_actions(xin, &xout);
7004 xlate_out_uninit(&xout);
7008 xlate_report(struct xlate_ctx *ctx, const char *s)
7010 if (ctx->xin->report_hook) {
7011 ctx->xin->report_hook(ctx, s);
7015 /* OFPP_NORMAL implementation. */
7017 static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *);
7019 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
7020 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle',
7021 * the bundle on which the packet was received, returns the VLAN to which the
7024 * Both 'vid' and the return value are in the range 0...4095. */
7026 input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid)
7028 switch (in_bundle->vlan_mode) {
7029 case PORT_VLAN_ACCESS:
7030 return in_bundle->vlan;
7033 case PORT_VLAN_TRUNK:
7036 case PORT_VLAN_NATIVE_UNTAGGED:
7037 case PORT_VLAN_NATIVE_TAGGED:
7038 return vid ? vid : in_bundle->vlan;
7045 /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'.
7046 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
7049 * 'vid' should be the VID obtained from the 802.1Q header that was received as
7050 * part of a packet (specify 0 if there was no 802.1Q header), in the range
7053 input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn)
7055 /* Allow any VID on the OFPP_NONE port. */
7056 if (in_bundle == &ofpp_none_bundle) {
7060 switch (in_bundle->vlan_mode) {
7061 case PORT_VLAN_ACCESS:
7064 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7065 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
7066 "packet received on port %s configured as VLAN "
7067 "%"PRIu16" access port",
7068 in_bundle->ofproto->up.name, vid,
7069 in_bundle->name, in_bundle->vlan);
7075 case PORT_VLAN_NATIVE_UNTAGGED:
7076 case PORT_VLAN_NATIVE_TAGGED:
7078 /* Port must always carry its native VLAN. */
7082 case PORT_VLAN_TRUNK:
7083 if (!ofbundle_includes_vlan(in_bundle, vid)) {
7085 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7086 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet "
7087 "received on port %s not configured for trunking "
7089 in_bundle->ofproto->up.name, vid,
7090 in_bundle->name, vid);
7102 /* Given 'vlan', the VLAN that a packet belongs to, and
7103 * 'out_bundle', a bundle on which the packet is to be output, returns the VID
7104 * that should be included in the 802.1Q header. (If the return value is 0,
7105 * then the 802.1Q header should only be included in the packet if there is a
7108 * Both 'vlan' and the return value are in the range 0...4095. */
7110 output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan)
7112 switch (out_bundle->vlan_mode) {
7113 case PORT_VLAN_ACCESS:
7116 case PORT_VLAN_TRUNK:
7117 case PORT_VLAN_NATIVE_TAGGED:
7120 case PORT_VLAN_NATIVE_UNTAGGED:
7121 return vlan == out_bundle->vlan ? 0 : vlan;
7129 output_normal(struct xlate_ctx *ctx, const struct ofbundle *out_bundle,
7132 struct ofport_dpif *port;
7134 ovs_be16 tci, old_tci;
7136 vid = output_vlan_to_vid(out_bundle, vlan);
7137 if (!out_bundle->bond) {
7138 port = ofbundle_get_a_port(out_bundle);
7140 port = bond_choose_output_slave(out_bundle->bond, &ctx->xin->flow,
7141 vid, &ctx->xout->tags);
7143 /* No slaves enabled, so drop packet. */
7148 old_tci = ctx->xin->flow.vlan_tci;
7150 if (tci || out_bundle->use_priority_tags) {
7151 tci |= ctx->xin->flow.vlan_tci & htons(VLAN_PCP_MASK);
7153 tci |= htons(VLAN_CFI);
7156 ctx->xin->flow.vlan_tci = tci;
7158 compose_output_action(ctx, port->up.ofp_port);
7159 ctx->xin->flow.vlan_tci = old_tci;
7163 mirror_mask_ffs(mirror_mask_t mask)
7165 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
7170 ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
7172 return (bundle->vlan_mode != PORT_VLAN_ACCESS
7173 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
7177 ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan)
7179 return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan);
7182 /* Returns an arbitrary interface within 'bundle'. */
7183 static struct ofport_dpif *
7184 ofbundle_get_a_port(const struct ofbundle *bundle)
7186 return CONTAINER_OF(list_front(&bundle->ports),
7187 struct ofport_dpif, bundle_node);
7191 vlan_is_mirrored(const struct ofmirror *m, int vlan)
7193 return !m->vlans || bitmap_is_set(m->vlans, vlan);
7197 add_mirror_actions(struct xlate_ctx *ctx, const struct flow *orig_flow)
7199 struct ofproto_dpif *ofproto = ctx->ofproto;
7200 mirror_mask_t mirrors;
7201 struct ofbundle *in_bundle;
7204 const struct nlattr *a;
7207 in_bundle = lookup_input_bundle(ctx->ofproto, orig_flow->in_port,
7208 ctx->xin->packet != NULL, NULL);
7212 mirrors = in_bundle->src_mirrors;
7214 /* Drop frames on bundles reserved for mirroring. */
7215 if (in_bundle->mirror_out) {
7216 if (ctx->xin->packet != NULL) {
7217 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7218 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
7219 "%s, which is reserved exclusively for mirroring",
7220 ctx->ofproto->up.name, in_bundle->name);
7226 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
7227 if (!input_vid_is_valid(vid, in_bundle, ctx->xin->packet != NULL)) {
7230 vlan = input_vid_to_vlan(in_bundle, vid);
7232 /* Look at the output ports to check for destination selections. */
7234 NL_ATTR_FOR_EACH (a, left, ctx->xout->odp_actions.data,
7235 ctx->xout->odp_actions.size) {
7236 enum ovs_action_attr type = nl_attr_type(a);
7237 struct ofport_dpif *ofport;
7239 if (type != OVS_ACTION_ATTR_OUTPUT) {
7243 ofport = get_odp_port(ofproto, nl_attr_get_u32(a));
7244 if (ofport && ofport->bundle) {
7245 mirrors |= ofport->bundle->dst_mirrors;
7253 /* Restore the original packet before adding the mirror actions. */
7254 ctx->xin->flow = *orig_flow;
7259 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
7261 if (!vlan_is_mirrored(m, vlan)) {
7262 mirrors = zero_rightmost_1bit(mirrors);
7266 mirrors &= ~m->dup_mirrors;
7267 ctx->xout->mirrors |= m->dup_mirrors;
7269 output_normal(ctx, m->out, vlan);
7270 } else if (vlan != m->out_vlan
7271 && !eth_addr_is_reserved(orig_flow->dl_dst)) {
7272 struct ofbundle *bundle;
7274 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
7275 if (ofbundle_includes_vlan(bundle, m->out_vlan)
7276 && !bundle->mirror_out) {
7277 output_normal(ctx, bundle, m->out_vlan);
7285 update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors,
7286 uint64_t packets, uint64_t bytes)
7292 for (; mirrors; mirrors = zero_rightmost_1bit(mirrors)) {
7295 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
7298 /* In normal circumstances 'm' will not be NULL. However,
7299 * if mirrors are reconfigured, we can temporarily get out
7300 * of sync in facet_revalidate(). We could "correct" the
7301 * mirror list before reaching here, but doing that would
7302 * not properly account the traffic stats we've currently
7303 * accumulated for previous mirror configuration. */
7307 m->packet_count += packets;
7308 m->byte_count += bytes;
7312 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
7313 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
7314 * indicate this; newer upstream kernels use gratuitous ARP requests. */
7316 is_gratuitous_arp(const struct flow *flow)
7318 return (flow->dl_type == htons(ETH_TYPE_ARP)
7319 && eth_addr_is_broadcast(flow->dl_dst)
7320 && (flow->nw_proto == ARP_OP_REPLY
7321 || (flow->nw_proto == ARP_OP_REQUEST
7322 && flow->nw_src == flow->nw_dst)));
7326 update_learning_table(struct ofproto_dpif *ofproto,
7327 const struct flow *flow, int vlan,
7328 struct ofbundle *in_bundle)
7330 struct mac_entry *mac;
7332 /* Don't learn the OFPP_NONE port. */
7333 if (in_bundle == &ofpp_none_bundle) {
7337 if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
7341 mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan);
7342 if (is_gratuitous_arp(flow)) {
7343 /* We don't want to learn from gratuitous ARP packets that are
7344 * reflected back over bond slaves so we lock the learning table. */
7345 if (!in_bundle->bond) {
7346 mac_entry_set_grat_arp_lock(mac);
7347 } else if (mac_entry_is_grat_arp_locked(mac)) {
7352 if (mac_entry_is_new(mac) || mac->port.p != in_bundle) {
7353 /* The log messages here could actually be useful in debugging,
7354 * so keep the rate limit relatively high. */
7355 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
7356 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
7357 "on port %s in VLAN %d",
7358 ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src),
7359 in_bundle->name, vlan);
7361 mac->port.p = in_bundle;
7362 tag_set_add(&ofproto->backer->revalidate_set,
7363 mac_learning_changed(ofproto->ml, mac));
7367 static struct ofbundle *
7368 lookup_input_bundle(const struct ofproto_dpif *ofproto, uint16_t in_port,
7369 bool warn, struct ofport_dpif **in_ofportp)
7371 struct ofport_dpif *ofport;
7373 /* Find the port and bundle for the received packet. */
7374 ofport = get_ofp_port(ofproto, in_port);
7376 *in_ofportp = ofport;
7378 if (ofport && ofport->bundle) {
7379 return ofport->bundle;
7382 /* Special-case OFPP_NONE, which a controller may use as the ingress
7383 * port for traffic that it is sourcing. */
7384 if (in_port == OFPP_NONE) {
7385 return &ofpp_none_bundle;
7388 /* Odd. A few possible reasons here:
7390 * - We deleted a port but there are still a few packets queued up
7393 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
7394 * we don't know about.
7396 * - The ofproto client didn't configure the port as part of a bundle.
7397 * This is particularly likely to happen if a packet was received on the
7398 * port after it was created, but before the client had a chance to
7399 * configure its bundle.
7402 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7404 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
7405 "port %"PRIu16, ofproto->up.name, in_port);
7410 /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or
7411 * dropped. Returns true if they may be forwarded, false if they should be
7414 * 'in_port' must be the ofport_dpif that corresponds to flow->in_port.
7415 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
7417 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
7418 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
7419 * checked by input_vid_is_valid().
7421 * May also add tags to '*tags', although the current implementation only does
7422 * so in one special case.
7425 is_admissible(struct xlate_ctx *ctx, struct ofport_dpif *in_port,
7428 struct ofproto_dpif *ofproto = ctx->ofproto;
7429 struct flow *flow = &ctx->xin->flow;
7430 struct ofbundle *in_bundle = in_port->bundle;
7432 /* Drop frames for reserved multicast addresses
7433 * only if forward_bpdu option is absent. */
7434 if (!ofproto->up.forward_bpdu && eth_addr_is_reserved(flow->dl_dst)) {
7435 xlate_report(ctx, "packet has reserved destination MAC, dropping");
7439 if (in_bundle->bond) {
7440 struct mac_entry *mac;
7442 switch (bond_check_admissibility(in_bundle->bond, in_port,
7443 flow->dl_dst, &ctx->xout->tags)) {
7448 xlate_report(ctx, "bonding refused admissibility, dropping");
7451 case BV_DROP_IF_MOVED:
7452 mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL);
7453 if (mac && mac->port.p != in_bundle &&
7454 (!is_gratuitous_arp(flow)
7455 || mac_entry_is_grat_arp_locked(mac))) {
7456 xlate_report(ctx, "SLB bond thinks this packet looped back, "
7468 xlate_normal(struct xlate_ctx *ctx)
7470 struct ofport_dpif *in_port;
7471 struct ofbundle *in_bundle;
7472 struct mac_entry *mac;
7476 ctx->xout->has_normal = true;
7478 in_bundle = lookup_input_bundle(ctx->ofproto, ctx->xin->flow.in_port,
7479 ctx->xin->packet != NULL, &in_port);
7481 xlate_report(ctx, "no input bundle, dropping");
7485 /* Drop malformed frames. */
7486 if (ctx->xin->flow.dl_type == htons(ETH_TYPE_VLAN) &&
7487 !(ctx->xin->flow.vlan_tci & htons(VLAN_CFI))) {
7488 if (ctx->xin->packet != NULL) {
7489 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7490 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
7491 "VLAN tag received on port %s",
7492 ctx->ofproto->up.name, in_bundle->name);
7494 xlate_report(ctx, "partial VLAN tag, dropping");
7498 /* Drop frames on bundles reserved for mirroring. */
7499 if (in_bundle->mirror_out) {
7500 if (ctx->xin->packet != NULL) {
7501 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7502 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
7503 "%s, which is reserved exclusively for mirroring",
7504 ctx->ofproto->up.name, in_bundle->name);
7506 xlate_report(ctx, "input port is mirror output port, dropping");
7511 vid = vlan_tci_to_vid(ctx->xin->flow.vlan_tci);
7512 if (!input_vid_is_valid(vid, in_bundle, ctx->xin->packet != NULL)) {
7513 xlate_report(ctx, "disallowed VLAN VID for this input port, dropping");
7516 vlan = input_vid_to_vlan(in_bundle, vid);
7518 /* Check other admissibility requirements. */
7519 if (in_port && !is_admissible(ctx, in_port, vlan)) {
7523 /* Learn source MAC. */
7524 if (ctx->xin->may_learn) {
7525 update_learning_table(ctx->ofproto, &ctx->xin->flow, vlan, in_bundle);
7528 /* Determine output bundle. */
7529 mac = mac_learning_lookup(ctx->ofproto->ml, ctx->xin->flow.dl_dst, vlan,
7532 if (mac->port.p != in_bundle) {
7533 xlate_report(ctx, "forwarding to learned port");
7534 output_normal(ctx, mac->port.p, vlan);
7536 xlate_report(ctx, "learned port is input port, dropping");
7539 struct ofbundle *bundle;
7541 xlate_report(ctx, "no learned MAC for destination, flooding");
7542 HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
7543 if (bundle != in_bundle
7544 && ofbundle_includes_vlan(bundle, vlan)
7545 && bundle->floodable
7546 && !bundle->mirror_out) {
7547 output_normal(ctx, bundle, vlan);
7550 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
7554 /* Optimized flow revalidation.
7556 * It's a difficult problem, in general, to tell which facets need to have
7557 * their actions recalculated whenever the OpenFlow flow table changes. We
7558 * don't try to solve that general problem: for most kinds of OpenFlow flow
7559 * table changes, we recalculate the actions for every facet. This is
7560 * relatively expensive, but it's good enough if the OpenFlow flow table
7561 * doesn't change very often.
7563 * However, we can expect one particular kind of OpenFlow flow table change to
7564 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
7565 * of CPU on revalidating every facet whenever MAC learning modifies the flow
7566 * table, we add a special case that applies to flow tables in which every rule
7567 * has the same form (that is, the same wildcards), except that the table is
7568 * also allowed to have a single "catch-all" flow that matches all packets. We
7569 * optimize this case by tagging all of the facets that resubmit into the table
7570 * and invalidating the same tag whenever a flow changes in that table. The
7571 * end result is that we revalidate just the facets that need it (and sometimes
7572 * a few more, but not all of the facets or even all of the facets that
7573 * resubmit to the table modified by MAC learning). */
7575 /* Calculates the tag to use for 'flow' and mask 'mask' when it is inserted
7576 * into an OpenFlow table with the given 'basis'. */
7578 rule_calculate_tag(const struct flow *flow, const struct minimask *mask,
7581 if (minimask_is_catchall(mask)) {
7584 uint32_t hash = flow_hash_in_minimask(flow, mask, secret);
7585 return tag_create_deterministic(hash);
7589 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
7590 * taggability of that table.
7592 * This function must be called after *each* change to a flow table. If you
7593 * skip calling it on some changes then the pointer comparisons at the end can
7594 * be invalid if you get unlucky. For example, if a flow removal causes a
7595 * cls_table to be destroyed and then a flow insertion causes a cls_table with
7596 * different wildcards to be created with the same address, then this function
7597 * will incorrectly skip revalidation. */
7599 table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
7601 struct table_dpif *table = &ofproto->tables[table_id];
7602 const struct oftable *oftable = &ofproto->up.tables[table_id];
7603 struct cls_table *catchall, *other;
7604 struct cls_table *t;
7606 catchall = other = NULL;
7608 switch (hmap_count(&oftable->cls.tables)) {
7610 /* We could tag this OpenFlow table but it would make the logic a
7611 * little harder and it's a corner case that doesn't seem worth it
7617 HMAP_FOR_EACH (t, hmap_node, &oftable->cls.tables) {
7618 if (cls_table_is_catchall(t)) {
7620 } else if (!other) {
7623 /* Indicate that we can't tag this by setting both tables to
7624 * NULL. (We know that 'catchall' is already NULL.) */
7631 /* Can't tag this table. */
7635 if (table->catchall_table != catchall || table->other_table != other) {
7636 table->catchall_table = catchall;
7637 table->other_table = other;
7638 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
7642 /* Given 'rule' that has changed in some way (either it is a rule being
7643 * inserted, a rule being deleted, or a rule whose actions are being
7644 * modified), marks facets for revalidation to ensure that packets will be
7645 * forwarded correctly according to the new state of the flow table.
7647 * This function must be called after *each* change to a flow table. See
7648 * the comment on table_update_taggable() for more information. */
7650 rule_invalidate(const struct rule_dpif *rule)
7652 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
7654 table_update_taggable(ofproto, rule->up.table_id);
7656 if (!ofproto->backer->need_revalidate) {
7657 struct table_dpif *table = &ofproto->tables[rule->up.table_id];
7659 if (table->other_table && rule->tag) {
7660 tag_set_add(&ofproto->backer->revalidate_set, rule->tag);
7662 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
7668 set_frag_handling(struct ofproto *ofproto_,
7669 enum ofp_config_flags frag_handling)
7671 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7672 if (frag_handling != OFPC_FRAG_REASM) {
7673 ofproto->backer->need_revalidate = REV_RECONFIGURE;
7681 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
7682 const struct flow *flow,
7683 const struct ofpact *ofpacts, size_t ofpacts_len)
7685 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7686 struct initial_vals initial_vals;
7687 struct odputil_keybuf keybuf;
7688 struct dpif_flow_stats stats;
7689 struct xlate_out xout;
7690 struct xlate_in xin;
7694 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
7695 odp_flow_key_from_flow(&key, flow,
7696 ofp_port_to_odp_port(ofproto, flow->in_port));
7698 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
7700 initial_vals.vlan_tci = flow->vlan_tci;
7701 xlate_in_init(&xin, ofproto, flow, &initial_vals, NULL, stats.tcp_flags,
7703 xin.resubmit_stats = &stats;
7704 xin.ofpacts_len = ofpacts_len;
7705 xin.ofpacts = ofpacts;
7707 xlate_actions(&xin, &xout);
7708 dpif_execute(ofproto->backer->dpif, key.data, key.size,
7709 xout.odp_actions.data, xout.odp_actions.size, packet);
7710 xlate_out_uninit(&xout);
7718 set_netflow(struct ofproto *ofproto_,
7719 const struct netflow_options *netflow_options)
7721 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7723 if (netflow_options) {
7724 if (!ofproto->netflow) {
7725 ofproto->netflow = netflow_create();
7727 return netflow_set_options(ofproto->netflow, netflow_options);
7729 netflow_destroy(ofproto->netflow);
7730 ofproto->netflow = NULL;
7736 get_netflow_ids(const struct ofproto *ofproto_,
7737 uint8_t *engine_type, uint8_t *engine_id)
7739 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7741 dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
7745 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
7747 if (!facet_is_controller_flow(facet) &&
7748 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
7749 struct subfacet *subfacet;
7750 struct ofexpired expired;
7752 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
7753 if (subfacet->path == SF_FAST_PATH) {
7754 struct dpif_flow_stats stats;
7756 subfacet_install(subfacet, &facet->xout.odp_actions, &stats);
7757 subfacet_update_stats(subfacet, &stats);
7761 expired.flow = facet->flow;
7762 expired.packet_count = facet->packet_count;
7763 expired.byte_count = facet->byte_count;
7764 expired.used = facet->used;
7765 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
7770 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
7772 struct facet *facet;
7774 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
7775 send_active_timeout(ofproto, facet);
7779 static struct ofproto_dpif *
7780 ofproto_dpif_lookup(const char *name)
7782 struct ofproto_dpif *ofproto;
7784 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
7785 hash_string(name, 0), &all_ofproto_dpifs) {
7786 if (!strcmp(ofproto->up.name, name)) {
7794 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
7795 const char *argv[], void *aux OVS_UNUSED)
7797 struct ofproto_dpif *ofproto;
7800 ofproto = ofproto_dpif_lookup(argv[1]);
7802 unixctl_command_reply_error(conn, "no such bridge");
7805 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
7807 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
7808 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
7812 unixctl_command_reply(conn, "table successfully flushed");
7816 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
7817 const char *argv[], void *aux OVS_UNUSED)
7819 struct ds ds = DS_EMPTY_INITIALIZER;
7820 const struct ofproto_dpif *ofproto;
7821 const struct mac_entry *e;
7823 ofproto = ofproto_dpif_lookup(argv[1]);
7825 unixctl_command_reply_error(conn, "no such bridge");
7829 ds_put_cstr(&ds, " port VLAN MAC Age\n");
7830 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
7831 struct ofbundle *bundle = e->port.p;
7832 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
7833 ofbundle_get_a_port(bundle)->odp_port,
7834 e->vlan, ETH_ADDR_ARGS(e->mac),
7835 mac_entry_age(ofproto->ml, e));
7837 unixctl_command_reply(conn, ds_cstr(&ds));
7842 struct xlate_out xout;
7843 struct xlate_in xin;
7849 trace_format_rule(struct ds *result, uint8_t table_id, int level,
7850 const struct rule_dpif *rule)
7852 ds_put_char_multiple(result, '\t', level);
7854 ds_put_cstr(result, "No match\n");
7858 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
7859 table_id, ntohll(rule->up.flow_cookie));
7860 cls_rule_format(&rule->up.cr, result);
7861 ds_put_char(result, '\n');
7863 ds_put_char_multiple(result, '\t', level);
7864 ds_put_cstr(result, "OpenFlow ");
7865 ofpacts_format(rule->up.ofpacts, rule->up.ofpacts_len, result);
7866 ds_put_char(result, '\n');
7870 trace_format_flow(struct ds *result, int level, const char *title,
7871 struct trace_ctx *trace)
7873 ds_put_char_multiple(result, '\t', level);
7874 ds_put_format(result, "%s: ", title);
7875 if (flow_equal(&trace->xin.flow, &trace->flow)) {
7876 ds_put_cstr(result, "unchanged");
7878 flow_format(result, &trace->xin.flow);
7879 trace->flow = trace->xin.flow;
7881 ds_put_char(result, '\n');
7885 trace_format_regs(struct ds *result, int level, const char *title,
7886 struct trace_ctx *trace)
7890 ds_put_char_multiple(result, '\t', level);
7891 ds_put_format(result, "%s:", title);
7892 for (i = 0; i < FLOW_N_REGS; i++) {
7893 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
7895 ds_put_char(result, '\n');
7899 trace_format_odp(struct ds *result, int level, const char *title,
7900 struct trace_ctx *trace)
7902 struct ofpbuf *odp_actions = &trace->xout.odp_actions;
7904 ds_put_char_multiple(result, '\t', level);
7905 ds_put_format(result, "%s: ", title);
7906 format_odp_actions(result, odp_actions->data, odp_actions->size);
7907 ds_put_char(result, '\n');
7911 trace_resubmit(struct xlate_ctx *ctx, struct rule_dpif *rule)
7913 struct trace_ctx *trace = CONTAINER_OF(ctx->xin, struct trace_ctx, xin);
7914 struct ds *result = trace->result;
7916 ds_put_char(result, '\n');
7917 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
7918 trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace);
7919 trace_format_odp(result, ctx->recurse + 1, "Resubmitted odp", trace);
7920 trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule);
7924 trace_report(struct xlate_ctx *ctx, const char *s)
7926 struct trace_ctx *trace = CONTAINER_OF(ctx->xin, struct trace_ctx, xin);
7927 struct ds *result = trace->result;
7929 ds_put_char_multiple(result, '\t', ctx->recurse);
7930 ds_put_cstr(result, s);
7931 ds_put_char(result, '\n');
7935 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
7936 void *aux OVS_UNUSED)
7938 const struct dpif_backer *backer;
7939 struct ofproto_dpif *ofproto;
7940 struct ofpbuf odp_key;
7941 struct ofpbuf *packet;
7942 struct initial_vals initial_vals;
7950 ofpbuf_init(&odp_key, 0);
7952 /* Handle "-generate" or a hex string as the last argument. */
7953 if (!strcmp(argv[argc - 1], "-generate")) {
7954 packet = ofpbuf_new(0);
7957 const char *error = eth_from_hex(argv[argc - 1], &packet);
7960 } else if (argc == 4) {
7961 /* The 3-argument form must end in "-generate' or a hex string. */
7962 unixctl_command_reply_error(conn, error);
7967 /* Parse the flow and determine whether a datapath or
7968 * bridge is specified. If function odp_flow_key_from_string()
7969 * returns 0, the flow is a odp_flow. If function
7970 * parse_ofp_exact_flow() returns 0, the flow is a br_flow. */
7971 if (!odp_flow_key_from_string(argv[argc - 1], NULL, &odp_key)) {
7972 /* If the odp_flow is the second argument,
7973 * the datapath name is the first argument. */
7975 const char *dp_type;
7976 if (!strncmp(argv[1], "ovs-", 4)) {
7977 dp_type = argv[1] + 4;
7981 backer = shash_find_data(&all_dpif_backers, dp_type);
7983 unixctl_command_reply_error(conn, "Cannot find datapath "
7988 /* No datapath name specified, so there should be only one
7990 struct shash_node *node;
7991 if (shash_count(&all_dpif_backers) != 1) {
7992 unixctl_command_reply_error(conn, "Must specify datapath "
7993 "name, there is more than one type of datapath");
7996 node = shash_first(&all_dpif_backers);
7997 backer = node->data;
8000 /* Extract the ofproto_dpif object from the ofproto_receive()
8002 if (ofproto_receive(backer, NULL, odp_key.data,
8003 odp_key.size, &flow, NULL, &ofproto, NULL,
8005 unixctl_command_reply_error(conn, "Invalid datapath flow");
8008 ds_put_format(&result, "Bridge: %s\n", ofproto->up.name);
8009 } else if (!parse_ofp_exact_flow(&flow, argv[argc - 1])) {
8011 unixctl_command_reply_error(conn, "Must specify bridge name");
8015 ofproto = ofproto_dpif_lookup(argv[1]);
8017 unixctl_command_reply_error(conn, "Unknown bridge name");
8020 initial_vals.vlan_tci = flow.vlan_tci;
8022 unixctl_command_reply_error(conn, "Bad flow syntax");
8026 /* Generate a packet, if requested. */
8028 if (!packet->size) {
8029 flow_compose(packet, &flow);
8031 ds_put_cstr(&result, "Packet: ");
8032 s = ofp_packet_to_string(packet->data, packet->size);
8033 ds_put_cstr(&result, s);
8036 /* Use the metadata from the flow and the packet argument
8037 * to reconstruct the flow. */
8038 flow_extract(packet, flow.skb_priority, flow.skb_mark, NULL,
8039 flow.in_port, &flow);
8040 initial_vals.vlan_tci = flow.vlan_tci;
8044 ofproto_trace(ofproto, &flow, packet, &initial_vals, &result);
8045 unixctl_command_reply(conn, ds_cstr(&result));
8048 ds_destroy(&result);
8049 ofpbuf_delete(packet);
8050 ofpbuf_uninit(&odp_key);
8054 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
8055 const struct ofpbuf *packet,
8056 const struct initial_vals *initial_vals, struct ds *ds)
8058 struct rule_dpif *rule;
8060 ds_put_cstr(ds, "Flow: ");
8061 flow_format(ds, flow);
8062 ds_put_char(ds, '\n');
8064 rule = rule_dpif_lookup(ofproto, flow);
8066 trace_format_rule(ds, 0, 0, rule);
8067 if (rule == ofproto->miss_rule) {
8068 ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
8069 } else if (rule == ofproto->no_packet_in_rule) {
8070 ds_put_cstr(ds, "\nNo match, packets dropped because "
8071 "OFPPC_NO_PACKET_IN is set on in_port.\n");
8075 uint64_t odp_actions_stub[1024 / 8];
8076 struct ofpbuf odp_actions;
8078 struct trace_ctx trace;
8081 tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
8084 ofpbuf_use_stub(&odp_actions,
8085 odp_actions_stub, sizeof odp_actions_stub);
8086 xlate_in_init(&trace.xin, ofproto, flow, initial_vals, rule, tcp_flags,
8088 trace.xin.resubmit_hook = trace_resubmit;
8089 trace.xin.report_hook = trace_report;
8090 xlate_actions(&trace.xin, &trace.xout);
8092 ds_put_char(ds, '\n');
8093 trace_format_flow(ds, 0, "Final flow", &trace);
8094 ds_put_cstr(ds, "Datapath actions: ");
8095 format_odp_actions(ds, trace.xout.odp_actions.data,
8096 trace.xout.odp_actions.size);
8098 if (trace.xout.slow) {
8099 ds_put_cstr(ds, "\nThis flow is handled by the userspace "
8100 "slow path because it:");
8101 switch (trace.xout.slow) {
8103 ds_put_cstr(ds, "\n\t- Consists of CFM packets.");
8106 ds_put_cstr(ds, "\n\t- Consists of LACP packets.");
8109 ds_put_cstr(ds, "\n\t- Consists of STP packets.");
8112 ds_put_cstr(ds, "\n\t- Consists of BFD packets.");
8114 case SLOW_CONTROLLER:
8115 ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages "
8116 "to the OpenFlow controller.");
8123 xlate_out_uninit(&trace.xout);
8128 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
8129 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
8132 unixctl_command_reply(conn, NULL);
8136 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
8137 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
8140 unixctl_command_reply(conn, NULL);
8143 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
8144 * 'reply' describing the results. */
8146 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
8148 struct facet *facet;
8152 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
8153 if (!facet_check_consistency(facet)) {
8158 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
8162 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
8163 ofproto->up.name, errors);
8165 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
8170 ofproto_dpif_self_check(struct unixctl_conn *conn,
8171 int argc, const char *argv[], void *aux OVS_UNUSED)
8173 struct ds reply = DS_EMPTY_INITIALIZER;
8174 struct ofproto_dpif *ofproto;
8177 ofproto = ofproto_dpif_lookup(argv[1]);
8179 unixctl_command_reply_error(conn, "Unknown ofproto (use "
8180 "ofproto/list for help)");
8183 ofproto_dpif_self_check__(ofproto, &reply);
8185 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
8186 ofproto_dpif_self_check__(ofproto, &reply);
8190 unixctl_command_reply(conn, ds_cstr(&reply));
8194 /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list
8195 * of the 'ofproto_shash' nodes. It is the responsibility of the caller
8196 * to destroy 'ofproto_shash' and free the returned value. */
8197 static const struct shash_node **
8198 get_ofprotos(struct shash *ofproto_shash)
8200 const struct ofproto_dpif *ofproto;
8202 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
8203 char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
8204 shash_add_nocopy(ofproto_shash, name, ofproto);
8207 return shash_sort(ofproto_shash);
8211 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
8212 const char *argv[] OVS_UNUSED,
8213 void *aux OVS_UNUSED)
8215 struct ds ds = DS_EMPTY_INITIALIZER;
8216 struct shash ofproto_shash;
8217 const struct shash_node **sorted_ofprotos;
8220 shash_init(&ofproto_shash);
8221 sorted_ofprotos = get_ofprotos(&ofproto_shash);
8222 for (i = 0; i < shash_count(&ofproto_shash); i++) {
8223 const struct shash_node *node = sorted_ofprotos[i];
8224 ds_put_format(&ds, "%s\n", node->name);
8227 shash_destroy(&ofproto_shash);
8228 free(sorted_ofprotos);
8230 unixctl_command_reply(conn, ds_cstr(&ds));
8235 show_dp_format(const struct ofproto_dpif *ofproto, struct ds *ds)
8237 const struct shash_node **ports;
8239 struct avg_subfacet_rates lifetime;
8240 unsigned long long int minutes;
8241 const int min_ms = 60 * 1000; /* milliseconds in one minute. */
8243 minutes = (time_msec() - ofproto->created) / min_ms;
8246 lifetime.add_rate = (double)ofproto->total_subfacet_add_count
8248 lifetime.del_rate = (double)ofproto->total_subfacet_del_count
8251 lifetime.add_rate = 0.0;
8252 lifetime.del_rate = 0.0;
8255 ds_put_format(ds, "%s (%s):\n", ofproto->up.name,
8256 dpif_name(ofproto->backer->dpif));
8258 "\tlookups: hit:%"PRIu64" missed:%"PRIu64"\n",
8259 ofproto->n_hit, ofproto->n_missed);
8260 ds_put_format(ds, "\tflows: cur: %zu, avg: %5.3f, max: %d,"
8261 " life span: %llu(ms)\n",
8262 hmap_count(&ofproto->subfacets),
8263 avg_subfacet_count(ofproto),
8264 ofproto->max_n_subfacet,
8265 avg_subfacet_life_span(ofproto));
8266 if (minutes >= 60) {
8267 show_dp_rates(ds, "\t\thourly avg:", &ofproto->hourly);
8269 if (minutes >= 60 * 24) {
8270 show_dp_rates(ds, "\t\tdaily avg:", &ofproto->daily);
8272 show_dp_rates(ds, "\t\toverall avg:", &lifetime);
8274 ports = shash_sort(&ofproto->up.port_by_name);
8275 for (i = 0; i < shash_count(&ofproto->up.port_by_name); i++) {
8276 const struct shash_node *node = ports[i];
8277 struct ofport *ofport = node->data;
8278 const char *name = netdev_get_name(ofport->netdev);
8279 const char *type = netdev_get_type(ofport->netdev);
8282 ds_put_format(ds, "\t%s %u/", name, ofport->ofp_port);
8284 odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
8285 if (odp_port != OVSP_NONE) {
8286 ds_put_format(ds, "%"PRIu32":", odp_port);
8288 ds_put_cstr(ds, "none:");
8291 if (strcmp(type, "system")) {
8292 struct netdev *netdev;
8295 ds_put_format(ds, " (%s", type);
8297 error = netdev_open(name, type, &netdev);
8302 error = netdev_get_config(netdev, &config);
8304 const struct smap_node **nodes;
8307 nodes = smap_sort(&config);
8308 for (i = 0; i < smap_count(&config); i++) {
8309 const struct smap_node *node = nodes[i];
8310 ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
8311 node->key, node->value);
8315 smap_destroy(&config);
8317 netdev_close(netdev);
8319 ds_put_char(ds, ')');
8321 ds_put_char(ds, '\n');
8327 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc,
8328 const char *argv[], void *aux OVS_UNUSED)
8330 struct ds ds = DS_EMPTY_INITIALIZER;
8331 const struct ofproto_dpif *ofproto;
8335 for (i = 1; i < argc; i++) {
8336 ofproto = ofproto_dpif_lookup(argv[i]);
8338 ds_put_format(&ds, "Unknown bridge %s (use dpif/dump-dps "
8339 "for help)", argv[i]);
8340 unixctl_command_reply_error(conn, ds_cstr(&ds));
8343 show_dp_format(ofproto, &ds);
8346 struct shash ofproto_shash;
8347 const struct shash_node **sorted_ofprotos;
8350 shash_init(&ofproto_shash);
8351 sorted_ofprotos = get_ofprotos(&ofproto_shash);
8352 for (i = 0; i < shash_count(&ofproto_shash); i++) {
8353 const struct shash_node *node = sorted_ofprotos[i];
8354 show_dp_format(node->data, &ds);
8357 shash_destroy(&ofproto_shash);
8358 free(sorted_ofprotos);
8361 unixctl_command_reply(conn, ds_cstr(&ds));
8366 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
8367 int argc OVS_UNUSED, const char *argv[],
8368 void *aux OVS_UNUSED)
8370 struct ds ds = DS_EMPTY_INITIALIZER;
8371 const struct ofproto_dpif *ofproto;
8372 struct subfacet *subfacet;
8374 ofproto = ofproto_dpif_lookup(argv[1]);
8376 unixctl_command_reply_error(conn, "no such bridge");
8380 update_stats(ofproto->backer);
8382 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
8383 struct facet *facet = subfacet->facet;
8385 odp_flow_key_format(subfacet->key, subfacet->key_len, &ds);
8387 ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:",
8388 subfacet->dp_packet_count, subfacet->dp_byte_count);
8389 if (subfacet->used) {
8390 ds_put_format(&ds, "%.3fs",
8391 (time_msec() - subfacet->used) / 1000.0);
8393 ds_put_format(&ds, "never");
8395 if (subfacet->facet->tcp_flags) {
8396 ds_put_cstr(&ds, ", flags:");
8397 packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags);
8400 ds_put_cstr(&ds, ", actions:");
8401 if (facet->xout.slow) {
8402 uint64_t slow_path_stub[128 / 8];
8403 const struct nlattr *actions;
8406 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
8407 slow_path_stub, sizeof slow_path_stub,
8408 &actions, &actions_len);
8409 format_odp_actions(&ds, actions, actions_len);
8411 format_odp_actions(&ds, facet->xout.odp_actions.data,
8412 facet->xout.odp_actions.size);
8414 ds_put_char(&ds, '\n');
8417 unixctl_command_reply(conn, ds_cstr(&ds));
8422 ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn,
8423 int argc OVS_UNUSED, const char *argv[],
8424 void *aux OVS_UNUSED)
8426 struct ds ds = DS_EMPTY_INITIALIZER;
8427 struct ofproto_dpif *ofproto;
8429 ofproto = ofproto_dpif_lookup(argv[1]);
8431 unixctl_command_reply_error(conn, "no such bridge");
8435 flush(&ofproto->up);
8437 unixctl_command_reply(conn, ds_cstr(&ds));
8442 ofproto_dpif_unixctl_init(void)
8444 static bool registered;
8450 unixctl_command_register(
8452 "[dp_name]|bridge odp_flow|br_flow [-generate|packet]",
8453 1, 3, ofproto_unixctl_trace, NULL);
8454 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
8455 ofproto_unixctl_fdb_flush, NULL);
8456 unixctl_command_register("fdb/show", "bridge", 1, 1,
8457 ofproto_unixctl_fdb_show, NULL);
8458 unixctl_command_register("ofproto/clog", "", 0, 0,
8459 ofproto_dpif_clog, NULL);
8460 unixctl_command_register("ofproto/unclog", "", 0, 0,
8461 ofproto_dpif_unclog, NULL);
8462 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
8463 ofproto_dpif_self_check, NULL);
8464 unixctl_command_register("dpif/dump-dps", "", 0, 0,
8465 ofproto_unixctl_dpif_dump_dps, NULL);
8466 unixctl_command_register("dpif/show", "[bridge]", 0, INT_MAX,
8467 ofproto_unixctl_dpif_show, NULL);
8468 unixctl_command_register("dpif/dump-flows", "bridge", 1, 1,
8469 ofproto_unixctl_dpif_dump_flows, NULL);
8470 unixctl_command_register("dpif/del-flows", "bridge", 1, 1,
8471 ofproto_unixctl_dpif_del_flows, NULL);
8474 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
8476 * This is deprecated. It is only for compatibility with broken device drivers
8477 * in old versions of Linux that do not properly support VLANs when VLAN
8478 * devices are not used. When broken device drivers are no longer in
8479 * widespread use, we will delete these interfaces. */
8482 set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid)
8484 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
8485 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
8487 if (realdev_ofp_port == ofport->realdev_ofp_port
8488 && vid == ofport->vlandev_vid) {
8492 ofproto->backer->need_revalidate = REV_RECONFIGURE;
8494 if (ofport->realdev_ofp_port) {
8497 if (realdev_ofp_port && ofport->bundle) {
8498 /* vlandevs are enslaved to their realdevs, so they are not allowed to
8499 * themselves be part of a bundle. */
8500 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
8503 ofport->realdev_ofp_port = realdev_ofp_port;
8504 ofport->vlandev_vid = vid;
8506 if (realdev_ofp_port) {
8507 vsp_add(ofport, realdev_ofp_port, vid);
8514 hash_realdev_vid(uint16_t realdev_ofp_port, int vid)
8516 return hash_2words(realdev_ofp_port, vid);
8519 /* Returns the ODP port number of the Linux VLAN device that corresponds to
8520 * 'vlan_tci' on the network device with port number 'realdev_odp_port' in
8521 * 'ofproto'. For example, given 'realdev_odp_port' of eth0 and 'vlan_tci' 9,
8522 * it would return the port number of eth0.9.
8524 * Unless VLAN splinters are enabled for port 'realdev_odp_port', this
8525 * function just returns its 'realdev_odp_port' argument. */
8527 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
8528 uint32_t realdev_odp_port, ovs_be16 vlan_tci)
8530 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
8531 uint16_t realdev_ofp_port;
8532 int vid = vlan_tci_to_vid(vlan_tci);
8533 const struct vlan_splinter *vsp;
8535 realdev_ofp_port = odp_port_to_ofp_port(ofproto, realdev_odp_port);
8536 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
8537 hash_realdev_vid(realdev_ofp_port, vid),
8538 &ofproto->realdev_vid_map) {
8539 if (vsp->realdev_ofp_port == realdev_ofp_port
8540 && vsp->vid == vid) {
8541 return ofp_port_to_odp_port(ofproto, vsp->vlandev_ofp_port);
8545 return realdev_odp_port;
8548 static struct vlan_splinter *
8549 vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port)
8551 struct vlan_splinter *vsp;
8553 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, hash_int(vlandev_ofp_port, 0),
8554 &ofproto->vlandev_map) {
8555 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
8563 /* Returns the OpenFlow port number of the "real" device underlying the Linux
8564 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
8565 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
8566 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
8567 * eth0 and store 9 in '*vid'.
8569 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
8570 * VLAN device. Unless VLAN splinters are enabled, this is what this function
8573 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
8574 uint16_t vlandev_ofp_port, int *vid)
8576 if (!hmap_is_empty(&ofproto->vlandev_map)) {
8577 const struct vlan_splinter *vsp;
8579 vsp = vlandev_find(ofproto, vlandev_ofp_port);
8584 return vsp->realdev_ofp_port;
8590 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
8591 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
8592 * 'flow->in_port' to the "real" device backing the VLAN device, sets
8593 * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is
8594 * always the case unless VLAN splinters are enabled), returns false without
8595 * making any changes. */
8597 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
8602 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port, &vid);
8607 /* Cause the flow to be processed as if it came in on the real device with
8608 * the VLAN device's VLAN ID. */
8609 flow->in_port = realdev;
8610 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
8615 vsp_remove(struct ofport_dpif *port)
8617 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
8618 struct vlan_splinter *vsp;
8620 vsp = vlandev_find(ofproto, port->up.ofp_port);
8622 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
8623 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
8626 port->realdev_ofp_port = 0;
8628 VLOG_ERR("missing vlan device record");
8633 vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid)
8635 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
8637 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
8638 && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid))
8639 == realdev_ofp_port)) {
8640 struct vlan_splinter *vsp;
8642 vsp = xmalloc(sizeof *vsp);
8643 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
8644 hash_int(port->up.ofp_port, 0));
8645 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
8646 hash_realdev_vid(realdev_ofp_port, vid));
8647 vsp->realdev_ofp_port = realdev_ofp_port;
8648 vsp->vlandev_ofp_port = port->up.ofp_port;
8651 port->realdev_ofp_port = realdev_ofp_port;
8653 VLOG_ERR("duplicate vlan device record");
8658 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
8660 const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
8661 return ofport ? ofport->odp_port : OVSP_NONE;
8664 static struct ofport_dpif *
8665 odp_port_to_ofport(const struct dpif_backer *backer, uint32_t odp_port)
8667 struct ofport_dpif *port;
8669 HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node,
8670 hash_int(odp_port, 0),
8671 &backer->odp_to_ofport_map) {
8672 if (port->odp_port == odp_port) {
8681 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
8683 struct ofport_dpif *port;
8685 port = odp_port_to_ofport(ofproto->backer, odp_port);
8686 if (port && &ofproto->up == port->up.ofproto) {
8687 return port->up.ofp_port;
8692 static unsigned long long int
8693 avg_subfacet_life_span(const struct ofproto_dpif *ofproto)
8695 unsigned long long int dc;
8696 unsigned long long int avg;
8698 dc = ofproto->total_subfacet_del_count + ofproto->subfacet_del_count;
8699 avg = dc ? ofproto->total_subfacet_life_span / dc : 0;
8705 avg_subfacet_count(const struct ofproto_dpif *ofproto)
8709 if (ofproto->n_update_stats) {
8710 avg_c = (double)ofproto->total_subfacet_count
8711 / ofproto->n_update_stats;
8718 show_dp_rates(struct ds *ds, const char *heading,
8719 const struct avg_subfacet_rates *rates)
8721 ds_put_format(ds, "%s add rate: %5.3f/min, del rate: %5.3f/min\n",
8722 heading, rates->add_rate, rates->del_rate);
8726 update_max_subfacet_count(struct ofproto_dpif *ofproto)
8728 ofproto->max_n_subfacet = MAX(ofproto->max_n_subfacet,
8729 hmap_count(&ofproto->subfacets));
8732 /* Compute exponentially weighted moving average, adding 'new' as the newest,
8733 * most heavily weighted element. 'base' designates the rate of decay: after
8734 * 'base' further updates, 'new''s weight in the EWMA decays to about 1/e
8737 exp_mavg(double *avg, int base, double new)
8739 *avg = (*avg * (base - 1) + new) / base;
8743 update_moving_averages(struct ofproto_dpif *ofproto)
8745 const int min_ms = 60 * 1000; /* milliseconds in one minute. */
8747 /* Update hourly averages on the minute boundaries. */
8748 if (time_msec() - ofproto->last_minute >= min_ms) {
8749 exp_mavg(&ofproto->hourly.add_rate, 60, ofproto->subfacet_add_count);
8750 exp_mavg(&ofproto->hourly.del_rate, 60, ofproto->subfacet_del_count);
8752 /* Update daily averages on the hour boundaries. */
8753 if ((ofproto->last_minute - ofproto->created) / min_ms % 60 == 59) {
8754 exp_mavg(&ofproto->daily.add_rate, 24, ofproto->hourly.add_rate);
8755 exp_mavg(&ofproto->daily.del_rate, 24, ofproto->hourly.del_rate);
8758 ofproto->total_subfacet_add_count += ofproto->subfacet_add_count;
8759 ofproto->total_subfacet_del_count += ofproto->subfacet_del_count;
8760 ofproto->subfacet_add_count = 0;
8761 ofproto->subfacet_del_count = 0;
8762 ofproto->last_minute += min_ms;
8766 const struct ofproto_class ofproto_dpif_class = {
8801 port_is_lacp_current,
8802 NULL, /* rule_choose_table */
8809 rule_modify_actions,
8823 get_stp_port_status,
8830 is_mirror_output_bundle,
8831 forward_bpdu_changed,
8832 set_mac_table_config,