2 * Copyright (c) 2009, 2010, 2011, 2012, 2013 Nicira, Inc.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at:
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
19 #include "ofproto/ofproto-provider.h"
26 #include "byte-order.h"
31 #include "dynamic-string.h"
32 #include "fail-open.h"
36 #include "mac-learning.h"
37 #include "meta-flow.h"
38 #include "multipath.h"
39 #include "netdev-vport.h"
46 #include "ofp-actions.h"
47 #include "ofp-parse.h"
48 #include "ofp-print.h"
49 #include "ofproto-dpif-governor.h"
50 #include "ofproto-dpif-ipfix.h"
51 #include "ofproto-dpif-sflow.h"
52 #include "poll-loop.h"
57 #include "unaligned.h"
59 #include "vlan-bitmap.h"
62 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
64 COVERAGE_DEFINE(ofproto_dpif_expired);
65 COVERAGE_DEFINE(ofproto_dpif_xlate);
66 COVERAGE_DEFINE(facet_changed_rule);
67 COVERAGE_DEFINE(facet_revalidate);
68 COVERAGE_DEFINE(facet_unexpected);
69 COVERAGE_DEFINE(facet_suppress);
71 /* Maximum depth of flow table recursion (due to resubmit actions) in a
72 * flow translation. */
73 #define MAX_RESUBMIT_RECURSION 64
75 /* Number of implemented OpenFlow tables. */
76 enum { N_TABLES = 255 };
77 enum { TBL_INTERNAL = N_TABLES - 1 }; /* Used for internal hidden rules. */
78 BUILD_ASSERT_DECL(N_TABLES >= 2 && N_TABLES <= 255);
90 * - Do include packets and bytes from facets that have been deleted or
91 * whose own statistics have been folded into the rule.
93 * - Do include packets and bytes sent "by hand" that were accounted to
94 * the rule without any facet being involved (this is a rare corner
95 * case in rule_execute()).
97 * - Do not include packet or bytes that can be obtained from any facet's
98 * packet_count or byte_count member or that can be obtained from the
99 * datapath by, e.g., dpif_flow_get() for any subfacet.
101 uint64_t packet_count; /* Number of packets received. */
102 uint64_t byte_count; /* Number of bytes received. */
104 tag_type tag; /* Caches rule_calculate_tag() result. */
106 struct list facets; /* List of "struct facet"s. */
109 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
111 return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
114 static struct rule_dpif *rule_dpif_lookup(struct ofproto_dpif *,
115 const struct flow *);
116 static struct rule_dpif *rule_dpif_lookup__(struct ofproto_dpif *,
119 static struct rule_dpif *rule_dpif_miss_rule(struct ofproto_dpif *ofproto,
120 const struct flow *flow);
122 static void rule_get_stats(struct rule *, uint64_t *packets, uint64_t *bytes);
123 static void rule_credit_stats(struct rule_dpif *,
124 const struct dpif_flow_stats *);
125 static tag_type rule_calculate_tag(const struct flow *,
126 const struct minimask *, uint32_t basis);
127 static void rule_invalidate(const struct rule_dpif *);
129 #define MAX_MIRRORS 32
130 typedef uint32_t mirror_mask_t;
131 #define MIRROR_MASK_C(X) UINT32_C(X)
132 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
134 struct ofproto_dpif *ofproto; /* Owning ofproto. */
135 size_t idx; /* In ofproto's "mirrors" array. */
136 void *aux; /* Key supplied by ofproto's client. */
137 char *name; /* Identifier for log messages. */
139 /* Selection criteria. */
140 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
141 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
142 unsigned long *vlans; /* Bitmap of chosen VLANs, NULL selects all. */
144 /* Output (exactly one of out == NULL and out_vlan == -1 is true). */
145 struct ofbundle *out; /* Output port or NULL. */
146 int out_vlan; /* Output VLAN or -1. */
147 mirror_mask_t dup_mirrors; /* Bitmap of mirrors with the same output. */
150 int64_t packet_count; /* Number of packets sent. */
151 int64_t byte_count; /* Number of bytes sent. */
154 static void mirror_destroy(struct ofmirror *);
155 static void update_mirror_stats(struct ofproto_dpif *ofproto,
156 mirror_mask_t mirrors,
157 uint64_t packets, uint64_t bytes);
160 struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
161 struct ofproto_dpif *ofproto; /* Owning ofproto. */
162 void *aux; /* Key supplied by ofproto's client. */
163 char *name; /* Identifier for log messages. */
166 struct list ports; /* Contains "struct ofport"s. */
167 enum port_vlan_mode vlan_mode; /* VLAN mode */
168 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
169 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
170 * NULL if all VLANs are trunked. */
171 struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */
172 struct bond *bond; /* Nonnull iff more than one port. */
173 bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */
176 bool floodable; /* True if no port has OFPUTIL_PC_NO_FLOOD set. */
178 /* Port mirroring info. */
179 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
180 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
181 mirror_mask_t mirror_out; /* Mirrors that output to this bundle. */
184 static void bundle_remove(struct ofport *);
185 static void bundle_update(struct ofbundle *);
186 static void bundle_destroy(struct ofbundle *);
187 static void bundle_del_port(struct ofport_dpif *);
188 static void bundle_run(struct ofbundle *);
189 static void bundle_wait(struct ofbundle *);
190 static struct ofbundle *lookup_input_bundle(const struct ofproto_dpif *,
191 uint16_t in_port, bool warn,
192 struct ofport_dpif **in_ofportp);
194 /* A controller may use OFPP_NONE as the ingress port to indicate that
195 * it did not arrive on a "real" port. 'ofpp_none_bundle' exists for
196 * when an input bundle is needed for validation (e.g., mirroring or
197 * OFPP_NORMAL processing). It is not connected to an 'ofproto' or have
198 * any 'port' structs, so care must be taken when dealing with it. */
199 static struct ofbundle ofpp_none_bundle = {
201 .vlan_mode = PORT_VLAN_TRUNK
204 static void stp_run(struct ofproto_dpif *ofproto);
205 static void stp_wait(struct ofproto_dpif *ofproto);
206 static int set_stp_port(struct ofport *,
207 const struct ofproto_port_stp_settings *);
209 static bool ofbundle_includes_vlan(const struct ofbundle *, uint16_t vlan);
213 /* Initial values of fields of the packet that may be changed during
214 * flow processing and needed later. */
215 struct initial_vals {
216 /* This is the value of vlan_tci in the packet as actually received from
217 * dpif. This is the same as the facet's flow.vlan_tci unless the packet
218 * was received via a VLAN splinter. In that case, this value is 0
219 * (because the packet as actually received from the dpif had no 802.1Q
220 * tag) but the facet's flow.vlan_tci is set to the VLAN that the splinter
223 * This member should be removed when the VLAN splinters feature is no
229 tag_type tags; /* Tags associated with actions. */
230 enum slow_path_reason slow; /* 0 if fast path may be used. */
231 bool has_learn; /* Actions include NXAST_LEARN? */
232 bool has_normal; /* Actions output to OFPP_NORMAL? */
233 bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */
234 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
235 mirror_mask_t mirrors; /* Bitmap of associated mirrors. */
237 uint64_t odp_actions_stub[256 / 8];
238 struct ofpbuf odp_actions;
242 struct ofproto_dpif *ofproto;
244 /* Flow to which the OpenFlow actions apply. xlate_actions() will modify
245 * this flow when actions change header fields. */
248 struct initial_vals initial_vals;
250 /* The packet corresponding to 'flow', or a null pointer if we are
251 * revalidating without a packet to refer to. */
252 const struct ofpbuf *packet;
254 /* Should OFPP_NORMAL update the MAC learning table? Should "learn"
255 * actions update the flow table?
257 * We want to update these tables if we are actually processing a packet,
258 * or if we are accounting for packets that the datapath has processed, but
259 * not if we are just revalidating. */
262 /* The rule initiating translation or NULL. */
263 struct rule_dpif *rule;
265 /* The actions to translate. If 'rule' is not NULL, these may be NULL. */
266 const struct ofpact *ofpacts;
269 /* Union of the set of TCP flags seen so far in this flow. (Used only by
270 * NXAST_FIN_TIMEOUT. Set to zero to avoid updating updating rules'
274 /* If nonnull, flow translation calls this function just before executing a
275 * resubmit or OFPP_TABLE action. In addition, disables logging of traces
276 * when the recursion depth is exceeded.
278 * 'rule' is the rule being submitted into. It will be null if the
279 * resubmit or OFPP_TABLE action didn't find a matching rule.
281 * This is normally null so the client has to set it manually after
282 * calling xlate_in_init(). */
283 void (*resubmit_hook)(struct xlate_ctx *, struct rule_dpif *rule);
285 /* If nonnull, flow translation calls this function to report some
286 * significant decision, e.g. to explain why OFPP_NORMAL translation
287 * dropped a packet. */
288 void (*report_hook)(struct xlate_ctx *, const char *s);
290 /* If nonnull, flow translation credits the specified statistics to each
291 * rule reached through a resubmit or OFPP_TABLE action.
293 * This is normally null so the client has to set it manually after
294 * calling xlate_in_init(). */
295 const struct dpif_flow_stats *resubmit_stats;
298 /* Context used by xlate_actions() and its callees. */
300 struct xlate_in *xin;
301 struct xlate_out *xout;
303 struct ofproto_dpif *ofproto;
305 /* Flow at the last commit. */
306 struct flow base_flow;
308 /* Tunnel IP destination address as received. This is stored separately
309 * as the base_flow.tunnel is cleared on init to reflect the datapath
310 * behavior. Used to make sure not to send tunneled output to ourselves,
311 * which might lead to an infinite loop. This could happen easily
312 * if a tunnel is marked as 'ip_remote=flow', and the flow does not
313 * actually set the tun_dst field. */
314 ovs_be32 orig_tunnel_ip_dst;
316 /* Stack for the push and pop actions. Each stack element is of type
317 * "union mf_subvalue". */
318 union mf_subvalue init_stack[1024 / sizeof(union mf_subvalue)];
321 /* The rule that we are currently translating, or NULL. */
322 struct rule_dpif *rule;
324 int recurse; /* Recursion level, via xlate_table_action. */
325 bool max_resubmit_trigger; /* Recursed too deeply during translation. */
326 uint32_t orig_skb_priority; /* Priority when packet arrived. */
327 uint8_t table_id; /* OpenFlow table ID where flow was found. */
328 uint32_t sflow_n_outputs; /* Number of output ports. */
329 uint32_t sflow_odp_port; /* Output port for composing sFlow action. */
330 uint16_t user_cookie_offset;/* Used for user_action_cookie fixup. */
331 bool exit; /* No further actions should be processed. */
334 static void xlate_in_init(struct xlate_in *, struct ofproto_dpif *,
335 const struct flow *, const struct initial_vals *,
336 struct rule_dpif *, uint8_t tcp_flags,
337 const struct ofpbuf *);
339 static void xlate_out_uninit(struct xlate_out *);
341 static void xlate_actions(struct xlate_in *, struct xlate_out *);
343 static void xlate_actions_for_side_effects(struct xlate_in *);
345 static void xlate_table_action(struct xlate_ctx *, uint16_t in_port,
346 uint8_t table_id, bool may_packet_in);
348 static size_t put_userspace_action(const struct ofproto_dpif *,
349 struct ofpbuf *odp_actions,
351 const union user_action_cookie *,
354 static void compose_slow_path(const struct ofproto_dpif *, const struct flow *,
355 enum slow_path_reason,
356 uint64_t *stub, size_t stub_size,
357 const struct nlattr **actionsp,
358 size_t *actions_lenp);
360 static void xlate_report(struct xlate_ctx *ctx, const char *s);
362 /* A subfacet (see "struct subfacet" below) has three possible installation
365 * - SF_NOT_INSTALLED: Not installed in the datapath. This will only be the
366 * case just after the subfacet is created, just before the subfacet is
367 * destroyed, or if the datapath returns an error when we try to install a
370 * - SF_FAST_PATH: The subfacet's actions are installed in the datapath.
372 * - SF_SLOW_PATH: An action that sends every packet for the subfacet through
373 * ofproto_dpif is installed in the datapath.
376 SF_NOT_INSTALLED, /* No datapath flow for this subfacet. */
377 SF_FAST_PATH, /* Full actions are installed. */
378 SF_SLOW_PATH, /* Send-to-userspace action is installed. */
381 /* A dpif flow and actions associated with a facet.
383 * See also the large comment on struct facet. */
386 struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */
387 struct list list_node; /* In struct facet's 'facets' list. */
388 struct facet *facet; /* Owning facet. */
390 enum odp_key_fitness key_fitness;
394 long long int used; /* Time last used; time created if not used. */
395 long long int created; /* Time created. */
397 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
398 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
400 enum subfacet_path path; /* Installed in datapath? */
403 #define SUBFACET_DESTROY_MAX_BATCH 50
405 static struct subfacet *subfacet_create(struct facet *, struct flow_miss *miss,
407 static struct subfacet *subfacet_find(struct ofproto_dpif *,
408 const struct nlattr *key, size_t key_len,
410 static void subfacet_destroy(struct subfacet *);
411 static void subfacet_destroy__(struct subfacet *);
412 static void subfacet_destroy_batch(struct ofproto_dpif *,
413 struct subfacet **, int n);
414 static void subfacet_reset_dp_stats(struct subfacet *,
415 struct dpif_flow_stats *);
416 static void subfacet_update_stats(struct subfacet *,
417 const struct dpif_flow_stats *);
418 static int subfacet_install(struct subfacet *,
419 const struct ofpbuf *odp_actions,
420 struct dpif_flow_stats *);
421 static void subfacet_uninstall(struct subfacet *);
423 /* An exact-match instantiation of an OpenFlow flow.
425 * A facet associates a "struct flow", which represents the Open vSwitch
426 * userspace idea of an exact-match flow, with one or more subfacets. Each
427 * subfacet tracks the datapath's idea of the exact-match flow equivalent to
428 * the facet. When the kernel module (or other dpif implementation) and Open
429 * vSwitch userspace agree on the definition of a flow key, there is exactly
430 * one subfacet per facet. If the dpif implementation supports more-specific
431 * flow matching than userspace, however, a facet can have more than one
432 * subfacet, each of which corresponds to some distinction in flow that
433 * userspace simply doesn't understand.
435 * Flow expiration works in terms of subfacets, so a facet must have at least
436 * one subfacet or it will never expire, leaking memory. */
439 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
440 struct list list_node; /* In owning rule's 'facets' list. */
441 struct rule_dpif *rule; /* Owning rule. */
444 struct list subfacets;
445 long long int used; /* Time last used; time created if not used. */
452 * - Do include packets and bytes sent "by hand", e.g. with
455 * - Do include packets and bytes that were obtained from the datapath
456 * when a subfacet's statistics were reset (e.g. dpif_flow_put() with
457 * DPIF_FP_ZERO_STATS).
459 * - Do not include packets or bytes that can be obtained from the
460 * datapath for any existing subfacet.
462 uint64_t packet_count; /* Number of packets received. */
463 uint64_t byte_count; /* Number of bytes received. */
465 /* Resubmit statistics. */
466 uint64_t prev_packet_count; /* Number of packets from last stats push. */
467 uint64_t prev_byte_count; /* Number of bytes from last stats push. */
468 long long int prev_used; /* Used time from last stats push. */
471 uint64_t accounted_bytes; /* Bytes processed by facet_account(). */
472 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
473 uint8_t tcp_flags; /* TCP flags seen for this 'rule'. */
475 struct xlate_out xout;
477 /* Initial values of the packet that may be needed later. */
478 struct initial_vals initial_vals;
480 /* Storage for a single subfacet, to reduce malloc() time and space
481 * overhead. (A facet always has at least one subfacet and in the common
482 * case has exactly one subfacet. However, 'one_subfacet' may not
483 * always be valid, since it could have been removed after newer
484 * subfacets were pushed onto the 'subfacets' list.) */
485 struct subfacet one_subfacet;
487 long long int learn_rl; /* Rate limiter for facet_learn(). */
490 static struct facet *facet_create(const struct flow_miss *, uint32_t hash);
491 static void facet_remove(struct facet *);
492 static void facet_free(struct facet *);
494 static struct facet *facet_find(struct ofproto_dpif *,
495 const struct flow *, uint32_t hash);
496 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
497 const struct flow *, uint32_t hash);
498 static bool facet_revalidate(struct facet *);
499 static bool facet_check_consistency(struct facet *);
501 static void facet_flush_stats(struct facet *);
503 static void facet_reset_counters(struct facet *);
504 static void facet_push_stats(struct facet *, bool may_learn);
505 static void facet_learn(struct facet *);
506 static void facet_account(struct facet *);
507 static void push_all_stats(void);
509 static bool facet_is_controller_flow(struct facet *);
512 struct hmap_node odp_port_node; /* In dpif_backer's "odp_to_ofport_map". */
516 struct ofbundle *bundle; /* Bundle that contains this port, if any. */
517 struct list bundle_node; /* In struct ofbundle's "ports" list. */
518 struct cfm *cfm; /* Connectivity Fault Management, if any. */
519 struct bfd *bfd; /* BFD, if any. */
520 tag_type tag; /* Tag associated with this port. */
521 bool may_enable; /* May be enabled in bonds. */
522 long long int carrier_seq; /* Carrier status changes. */
523 struct tnl_port *tnl_port; /* Tunnel handle, or null. */
526 struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */
527 enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */
528 long long int stp_state_entered;
530 struct hmap priorities; /* Map of attached 'priority_to_dscp's. */
532 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
534 * This is deprecated. It is only for compatibility with broken device
535 * drivers in old versions of Linux that do not properly support VLANs when
536 * VLAN devices are not used. When broken device drivers are no longer in
537 * widespread use, we will delete these interfaces. */
538 uint16_t realdev_ofp_port;
542 /* Node in 'ofport_dpif''s 'priorities' map. Used to maintain a map from
543 * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
544 * traffic egressing the 'ofport' with that priority should be marked with. */
545 struct priority_to_dscp {
546 struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'priorities' map. */
547 uint32_t priority; /* Priority of this queue (see struct flow). */
549 uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */
552 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
554 * This is deprecated. It is only for compatibility with broken device drivers
555 * in old versions of Linux that do not properly support VLANs when VLAN
556 * devices are not used. When broken device drivers are no longer in
557 * widespread use, we will delete these interfaces. */
558 struct vlan_splinter {
559 struct hmap_node realdev_vid_node;
560 struct hmap_node vlandev_node;
561 uint16_t realdev_ofp_port;
562 uint16_t vlandev_ofp_port;
566 static uint16_t vsp_realdev_to_vlandev(const struct ofproto_dpif *,
567 uint16_t realdev_ofp_port,
569 static bool vsp_adjust_flow(const struct ofproto_dpif *, struct flow *);
570 static void vsp_remove(struct ofport_dpif *);
571 static void vsp_add(struct ofport_dpif *, uint16_t realdev_ofp_port, int vid);
573 static uint32_t ofp_port_to_odp_port(const struct ofproto_dpif *,
575 static uint16_t odp_port_to_ofp_port(const struct ofproto_dpif *,
578 static struct ofport_dpif *
579 ofport_dpif_cast(const struct ofport *ofport)
581 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
584 static void port_run(struct ofport_dpif *);
585 static void port_run_fast(struct ofport_dpif *);
586 static void port_wait(struct ofport_dpif *);
587 static int set_bfd(struct ofport *, const struct smap *);
588 static int set_cfm(struct ofport *, const struct cfm_settings *);
589 static void ofport_clear_priorities(struct ofport_dpif *);
590 static void run_fast_rl(void);
592 struct dpif_completion {
593 struct list list_node;
594 struct ofoperation *op;
597 /* Extra information about a classifier table.
598 * Currently used just for optimized flow revalidation. */
600 /* If either of these is nonnull, then this table has a form that allows
601 * flows to be tagged to avoid revalidating most flows for the most common
602 * kinds of flow table changes. */
603 struct cls_table *catchall_table; /* Table that wildcards all fields. */
604 struct cls_table *other_table; /* Table with any other wildcard set. */
605 uint32_t basis; /* Keeps each table's tags separate. */
608 /* Reasons that we might need to revalidate every facet, and corresponding
611 * A value of 0 means that there is no need to revalidate.
613 * It would be nice to have some cleaner way to integrate with coverage
614 * counters, but with only a few reasons I guess this is good enough for
616 enum revalidate_reason {
617 REV_RECONFIGURE = 1, /* Switch configuration changed. */
618 REV_STP, /* Spanning tree protocol port status change. */
619 REV_PORT_TOGGLED, /* Port enabled or disabled by CFM, LACP, ...*/
620 REV_FLOW_TABLE, /* Flow table changed. */
621 REV_INCONSISTENCY /* Facet self-check failed. */
623 COVERAGE_DEFINE(rev_reconfigure);
624 COVERAGE_DEFINE(rev_stp);
625 COVERAGE_DEFINE(rev_port_toggled);
626 COVERAGE_DEFINE(rev_flow_table);
627 COVERAGE_DEFINE(rev_inconsistency);
629 /* Drop keys are odp flow keys which have drop flows installed in the kernel.
630 * These are datapath flows which have no associated ofproto, if they did we
631 * would use facets. */
633 struct hmap_node hmap_node;
638 /* All datapaths of a given type share a single dpif backer instance. */
643 struct timer next_expiration;
644 struct hmap odp_to_ofport_map; /* ODP port to ofport mapping. */
646 struct simap tnl_backers; /* Set of dpif ports backing tunnels. */
648 /* Facet revalidation flags applying to facets which use this backer. */
649 enum revalidate_reason need_revalidate; /* Revalidate every facet. */
650 struct tag_set revalidate_set; /* Revalidate only matching facets. */
652 struct hmap drop_keys; /* Set of dropped odp keys. */
653 bool recv_set_enable; /* Enables or disables receiving packets. */
656 /* All existing ofproto_backer instances, indexed by ofproto->up.type. */
657 static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers);
659 static void drop_key_clear(struct dpif_backer *);
660 static struct ofport_dpif *
661 odp_port_to_ofport(const struct dpif_backer *, uint32_t odp_port);
663 struct avg_subfacet_rates {
664 double add_rate; /* Moving average of new flows created per minute. */
665 double del_rate; /* Moving average of flows deleted per minute. */
667 static void show_dp_rates(struct ds *ds, const char *heading,
668 const struct avg_subfacet_rates *rates);
669 static void exp_mavg(double *avg, int base, double new);
671 struct ofproto_dpif {
672 struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
674 struct dpif_backer *backer;
676 /* Special OpenFlow rules. */
677 struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */
678 struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */
681 struct netflow *netflow;
682 struct dpif_sflow *sflow;
683 struct dpif_ipfix *ipfix;
684 struct hmap bundles; /* Contains "struct ofbundle"s. */
685 struct mac_learning *ml;
686 struct ofmirror *mirrors[MAX_MIRRORS];
688 bool has_bonded_bundles;
692 struct hmap subfacets;
693 struct governor *governor;
694 long long int consistency_rl;
697 struct table_dpif tables[N_TABLES];
699 /* Support for debugging async flow mods. */
700 struct list completions;
702 bool has_bundle_action; /* True when the first bundle action appears. */
703 struct netdev_stats stats; /* To account packets generated and consumed in
708 long long int stp_last_tick;
710 /* VLAN splinters. */
711 struct hmap realdev_vid_map; /* (realdev,vid) -> vlandev. */
712 struct hmap vlandev_map; /* vlandev -> (realdev,vid). */
715 struct sset ports; /* Set of standard port names. */
716 struct sset ghost_ports; /* Ports with no datapath port. */
717 struct sset port_poll_set; /* Queued names for port_poll() reply. */
718 int port_poll_errno; /* Last errno for port_poll() reply. */
720 /* Per ofproto's dpif stats. */
724 /* Subfacet statistics.
726 * These keep track of the total number of subfacets added and deleted and
727 * flow life span. They are useful for computing the flow rates stats
728 * exposed via "ovs-appctl dpif/show". The goal is to learn about
729 * traffic patterns in ways that we can use later to improve Open vSwitch
730 * performance in new situations. */
731 long long int created; /* Time when it is created. */
732 unsigned int max_n_subfacet; /* Maximum number of flows */
734 /* The average number of subfacets... */
735 struct avg_subfacet_rates hourly; /* ...over the last hour. */
736 struct avg_subfacet_rates daily; /* ...over the last day. */
737 long long int last_minute; /* Last time 'hourly' was updated. */
739 /* Number of subfacets added or deleted since 'last_minute'. */
740 unsigned int subfacet_add_count;
741 unsigned int subfacet_del_count;
743 /* Number of subfacets added or deleted from 'created' to 'last_minute.' */
744 unsigned long long int total_subfacet_add_count;
745 unsigned long long int total_subfacet_del_count;
747 /* Sum of the number of milliseconds that each subfacet existed,
748 * over the subfacets that have been added and then later deleted. */
749 unsigned long long int total_subfacet_life_span;
751 /* Incremented by the number of currently existing subfacets, each
752 * time we pull statistics from the kernel. */
753 unsigned long long int total_subfacet_count;
755 /* Number of times we pull statistics from the kernel. */
756 unsigned long long int n_update_stats;
758 static unsigned long long int avg_subfacet_life_span(
759 const struct ofproto_dpif *);
760 static double avg_subfacet_count(const struct ofproto_dpif *ofproto);
761 static void update_moving_averages(struct ofproto_dpif *ofproto);
762 static void update_max_subfacet_count(struct ofproto_dpif *ofproto);
764 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
765 * for debugging the asynchronous flow_mod implementation.) */
768 /* All existing ofproto_dpif instances, indexed by ->up.name. */
769 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
771 static void ofproto_dpif_unixctl_init(void);
773 static struct ofproto_dpif *
774 ofproto_dpif_cast(const struct ofproto *ofproto)
776 ovs_assert(ofproto->ofproto_class == &ofproto_dpif_class);
777 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
780 static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *,
782 static struct ofport_dpif *get_odp_port(const struct ofproto_dpif *,
784 static void ofproto_trace(struct ofproto_dpif *, const struct flow *,
785 const struct ofpbuf *,
786 const struct initial_vals *, struct ds *);
788 /* Packet processing. */
789 static void update_learning_table(struct ofproto_dpif *,
790 const struct flow *, int vlan,
793 #define FLOW_MISS_MAX_BATCH 50
794 static int handle_upcalls(struct dpif_backer *, unsigned int max_batch);
796 /* Flow expiration. */
797 static int expire(struct dpif_backer *);
800 static void send_netflow_active_timeouts(struct ofproto_dpif *);
803 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
804 static size_t compose_sflow_action(const struct ofproto_dpif *,
805 struct ofpbuf *odp_actions,
806 const struct flow *, uint32_t odp_port);
807 static void compose_ipfix_action(const struct ofproto_dpif *,
808 struct ofpbuf *odp_actions,
809 const struct flow *);
810 static void add_mirror_actions(struct xlate_ctx *ctx,
811 const struct flow *flow);
812 /* Global variables. */
813 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
815 /* Initial mappings of port to bridge mappings. */
816 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
818 /* Factory functions. */
821 init(const struct shash *iface_hints)
823 struct shash_node *node;
825 /* Make a local copy, since we don't own 'iface_hints' elements. */
826 SHASH_FOR_EACH(node, iface_hints) {
827 const struct iface_hint *orig_hint = node->data;
828 struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
830 new_hint->br_name = xstrdup(orig_hint->br_name);
831 new_hint->br_type = xstrdup(orig_hint->br_type);
832 new_hint->ofp_port = orig_hint->ofp_port;
834 shash_add(&init_ofp_ports, node->name, new_hint);
839 enumerate_types(struct sset *types)
841 dp_enumerate_types(types);
845 enumerate_names(const char *type, struct sset *names)
847 struct ofproto_dpif *ofproto;
850 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
851 if (strcmp(type, ofproto->up.type)) {
854 sset_add(names, ofproto->up.name);
861 del(const char *type, const char *name)
866 error = dpif_open(name, type, &dpif);
868 error = dpif_delete(dpif);
875 port_open_type(const char *datapath_type, const char *port_type)
877 return dpif_port_open_type(datapath_type, port_type);
880 /* Type functions. */
882 static struct ofproto_dpif *
883 lookup_ofproto_dpif_by_port_name(const char *name)
885 struct ofproto_dpif *ofproto;
887 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
888 if (sset_contains(&ofproto->ports, name)) {
897 type_run(const char *type)
899 static long long int push_timer = LLONG_MIN;
900 struct dpif_backer *backer;
904 backer = shash_find_data(&all_dpif_backers, type);
906 /* This is not necessarily a problem, since backers are only
907 * created on demand. */
911 dpif_run(backer->dpif);
913 /* The most natural place to push facet statistics is when they're pulled
914 * from the datapath. However, when there are many flows in the datapath,
915 * this expensive operation can occur so frequently, that it reduces our
916 * ability to quickly set up flows. To reduce the cost, we push statistics
918 if (time_msec() > push_timer) {
919 push_timer = time_msec() + 2000;
923 /* If vswitchd started with other_config:flow_restore_wait set as "true",
924 * and the configuration has now changed to "false", enable receiving
925 * packets from the datapath. */
926 if (!backer->recv_set_enable && !ofproto_get_flow_restore_wait()) {
927 backer->recv_set_enable = true;
929 error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
931 VLOG_ERR("Failed to enable receiving packets in dpif.");
934 dpif_flow_flush(backer->dpif);
935 backer->need_revalidate = REV_RECONFIGURE;
938 if (backer->need_revalidate
939 || !tag_set_is_empty(&backer->revalidate_set)) {
940 struct tag_set revalidate_set = backer->revalidate_set;
941 bool need_revalidate = backer->need_revalidate;
942 struct ofproto_dpif *ofproto;
943 struct simap_node *node;
944 struct simap tmp_backers;
946 /* Handle tunnel garbage collection. */
947 simap_init(&tmp_backers);
948 simap_swap(&backer->tnl_backers, &tmp_backers);
950 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
951 struct ofport_dpif *iter;
953 if (backer != ofproto->backer) {
957 HMAP_FOR_EACH (iter, up.hmap_node, &ofproto->up.ports) {
960 if (!iter->tnl_port) {
964 dp_port = netdev_vport_get_dpif_port(iter->up.netdev);
965 node = simap_find(&tmp_backers, dp_port);
967 simap_put(&backer->tnl_backers, dp_port, node->data);
968 simap_delete(&tmp_backers, node);
969 node = simap_find(&backer->tnl_backers, dp_port);
971 node = simap_find(&backer->tnl_backers, dp_port);
973 uint32_t odp_port = UINT32_MAX;
975 if (!dpif_port_add(backer->dpif, iter->up.netdev,
977 simap_put(&backer->tnl_backers, dp_port, odp_port);
978 node = simap_find(&backer->tnl_backers, dp_port);
983 iter->odp_port = node ? node->data : OVSP_NONE;
984 if (tnl_port_reconfigure(&iter->up, iter->odp_port,
986 backer->need_revalidate = REV_RECONFIGURE;
991 SIMAP_FOR_EACH (node, &tmp_backers) {
992 dpif_port_del(backer->dpif, node->data);
994 simap_destroy(&tmp_backers);
996 switch (backer->need_revalidate) {
997 case REV_RECONFIGURE: COVERAGE_INC(rev_reconfigure); break;
998 case REV_STP: COVERAGE_INC(rev_stp); break;
999 case REV_PORT_TOGGLED: COVERAGE_INC(rev_port_toggled); break;
1000 case REV_FLOW_TABLE: COVERAGE_INC(rev_flow_table); break;
1001 case REV_INCONSISTENCY: COVERAGE_INC(rev_inconsistency); break;
1004 if (backer->need_revalidate) {
1005 /* Clear the drop_keys in case we should now be accepting some
1006 * formerly dropped flows. */
1007 drop_key_clear(backer);
1010 /* Clear the revalidation flags. */
1011 tag_set_init(&backer->revalidate_set);
1012 backer->need_revalidate = 0;
1014 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1015 struct facet *facet, *next;
1017 if (ofproto->backer != backer) {
1021 HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &ofproto->facets) {
1023 || tag_set_intersects(&revalidate_set, facet->xout.tags)) {
1024 facet_revalidate(facet);
1031 if (!backer->recv_set_enable) {
1032 /* Wake up before a max of 1000ms. */
1033 timer_set_duration(&backer->next_expiration, 1000);
1034 } else if (timer_expired(&backer->next_expiration)) {
1035 int delay = expire(backer);
1036 timer_set_duration(&backer->next_expiration, delay);
1039 /* Check for port changes in the dpif. */
1040 while ((error = dpif_port_poll(backer->dpif, &devname)) == 0) {
1041 struct ofproto_dpif *ofproto;
1042 struct dpif_port port;
1044 /* Don't report on the datapath's device. */
1045 if (!strcmp(devname, dpif_base_name(backer->dpif))) {
1049 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
1050 &all_ofproto_dpifs) {
1051 if (simap_contains(&ofproto->backer->tnl_backers, devname)) {
1056 ofproto = lookup_ofproto_dpif_by_port_name(devname);
1057 if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
1058 /* The port was removed. If we know the datapath,
1059 * report it through poll_set(). If we don't, it may be
1060 * notifying us of a removal we initiated, so ignore it.
1061 * If there's a pending ENOBUFS, let it stand, since
1062 * everything will be reevaluated. */
1063 if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
1064 sset_add(&ofproto->port_poll_set, devname);
1065 ofproto->port_poll_errno = 0;
1067 } else if (!ofproto) {
1068 /* The port was added, but we don't know with which
1069 * ofproto we should associate it. Delete it. */
1070 dpif_port_del(backer->dpif, port.port_no);
1072 dpif_port_destroy(&port);
1078 if (error != EAGAIN) {
1079 struct ofproto_dpif *ofproto;
1081 /* There was some sort of error, so propagate it to all
1082 * ofprotos that use this backer. */
1083 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
1084 &all_ofproto_dpifs) {
1085 if (ofproto->backer == backer) {
1086 sset_clear(&ofproto->port_poll_set);
1087 ofproto->port_poll_errno = error;
1096 dpif_backer_run_fast(struct dpif_backer *backer, int max_batch)
1100 /* If recv_set_enable is false, we should not handle upcalls. */
1101 if (!backer->recv_set_enable) {
1105 /* Handle one or more batches of upcalls, until there's nothing left to do
1106 * or until we do a fixed total amount of work.
1108 * We do work in batches because it can be much cheaper to set up a number
1109 * of flows and fire off their patches all at once. We do multiple batches
1110 * because in some cases handling a packet can cause another packet to be
1111 * queued almost immediately as part of the return flow. Both
1112 * optimizations can make major improvements on some benchmarks and
1113 * presumably for real traffic as well. */
1115 while (work < max_batch) {
1116 int retval = handle_upcalls(backer, max_batch - work);
1127 type_run_fast(const char *type)
1129 struct dpif_backer *backer;
1131 backer = shash_find_data(&all_dpif_backers, type);
1133 /* This is not necessarily a problem, since backers are only
1134 * created on demand. */
1138 return dpif_backer_run_fast(backer, FLOW_MISS_MAX_BATCH);
1144 static long long int port_rl = LLONG_MIN;
1145 static unsigned int backer_rl = 0;
1147 if (time_msec() >= port_rl) {
1148 struct ofproto_dpif *ofproto;
1149 struct ofport_dpif *ofport;
1151 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1153 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1154 port_run_fast(ofport);
1157 port_rl = time_msec() + 200;
1160 /* XXX: We have to be careful not to do too much work in this function. If
1161 * we call dpif_backer_run_fast() too often, or with too large a batch,
1162 * performance improves signifcantly, but at a cost. It's possible for the
1163 * number of flows in the datapath to increase without bound, and for poll
1164 * loops to take 10s of seconds. The correct solution to this problem,
1165 * long term, is to separate flow miss handling into it's own thread so it
1166 * isn't affected by revalidations, and expirations. Until then, this is
1167 * the best we can do. */
1168 if (++backer_rl >= 10) {
1169 struct shash_node *node;
1172 SHASH_FOR_EACH (node, &all_dpif_backers) {
1173 dpif_backer_run_fast(node->data, 1);
1179 type_wait(const char *type)
1181 struct dpif_backer *backer;
1183 backer = shash_find_data(&all_dpif_backers, type);
1185 /* This is not necessarily a problem, since backers are only
1186 * created on demand. */
1190 timer_wait(&backer->next_expiration);
1193 /* Basic life-cycle. */
1195 static int add_internal_flows(struct ofproto_dpif *);
1197 static struct ofproto *
1200 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
1201 return &ofproto->up;
1205 dealloc(struct ofproto *ofproto_)
1207 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1212 close_dpif_backer(struct dpif_backer *backer)
1214 struct shash_node *node;
1216 ovs_assert(backer->refcount > 0);
1218 if (--backer->refcount) {
1222 drop_key_clear(backer);
1223 hmap_destroy(&backer->drop_keys);
1225 simap_destroy(&backer->tnl_backers);
1226 hmap_destroy(&backer->odp_to_ofport_map);
1227 node = shash_find(&all_dpif_backers, backer->type);
1229 shash_delete(&all_dpif_backers, node);
1230 dpif_close(backer->dpif);
1235 /* Datapath port slated for removal from datapath. */
1236 struct odp_garbage {
1237 struct list list_node;
1242 open_dpif_backer(const char *type, struct dpif_backer **backerp)
1244 struct dpif_backer *backer;
1245 struct dpif_port_dump port_dump;
1246 struct dpif_port port;
1247 struct shash_node *node;
1248 struct list garbage_list;
1249 struct odp_garbage *garbage, *next;
1255 backer = shash_find_data(&all_dpif_backers, type);
1262 backer_name = xasprintf("ovs-%s", type);
1264 /* Remove any existing datapaths, since we assume we're the only
1265 * userspace controlling the datapath. */
1267 dp_enumerate_names(type, &names);
1268 SSET_FOR_EACH(name, &names) {
1269 struct dpif *old_dpif;
1271 /* Don't remove our backer if it exists. */
1272 if (!strcmp(name, backer_name)) {
1276 if (dpif_open(name, type, &old_dpif)) {
1277 VLOG_WARN("couldn't open old datapath %s to remove it", name);
1279 dpif_delete(old_dpif);
1280 dpif_close(old_dpif);
1283 sset_destroy(&names);
1285 backer = xmalloc(sizeof *backer);
1287 error = dpif_create_and_open(backer_name, type, &backer->dpif);
1290 VLOG_ERR("failed to open datapath of type %s: %s", type,
1296 backer->type = xstrdup(type);
1297 backer->refcount = 1;
1298 hmap_init(&backer->odp_to_ofport_map);
1299 hmap_init(&backer->drop_keys);
1300 timer_set_duration(&backer->next_expiration, 1000);
1301 backer->need_revalidate = 0;
1302 simap_init(&backer->tnl_backers);
1303 tag_set_init(&backer->revalidate_set);
1304 backer->recv_set_enable = !ofproto_get_flow_restore_wait();
1307 if (backer->recv_set_enable) {
1308 dpif_flow_flush(backer->dpif);
1311 /* Loop through the ports already on the datapath and remove any
1312 * that we don't need anymore. */
1313 list_init(&garbage_list);
1314 dpif_port_dump_start(&port_dump, backer->dpif);
1315 while (dpif_port_dump_next(&port_dump, &port)) {
1316 node = shash_find(&init_ofp_ports, port.name);
1317 if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
1318 garbage = xmalloc(sizeof *garbage);
1319 garbage->odp_port = port.port_no;
1320 list_push_front(&garbage_list, &garbage->list_node);
1323 dpif_port_dump_done(&port_dump);
1325 LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) {
1326 dpif_port_del(backer->dpif, garbage->odp_port);
1327 list_remove(&garbage->list_node);
1331 shash_add(&all_dpif_backers, type, backer);
1333 error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
1335 VLOG_ERR("failed to listen on datapath of type %s: %s",
1336 type, strerror(error));
1337 close_dpif_backer(backer);
1345 construct(struct ofproto *ofproto_)
1347 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1348 struct shash_node *node, *next;
1353 error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1358 max_ports = dpif_get_max_ports(ofproto->backer->dpif);
1359 ofproto_init_max_ports(ofproto_, MIN(max_ports, OFPP_MAX));
1361 ofproto->netflow = NULL;
1362 ofproto->sflow = NULL;
1363 ofproto->ipfix = NULL;
1364 ofproto->stp = NULL;
1365 hmap_init(&ofproto->bundles);
1366 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1367 for (i = 0; i < MAX_MIRRORS; i++) {
1368 ofproto->mirrors[i] = NULL;
1370 ofproto->has_bonded_bundles = false;
1372 hmap_init(&ofproto->facets);
1373 hmap_init(&ofproto->subfacets);
1374 ofproto->governor = NULL;
1375 ofproto->consistency_rl = LLONG_MIN;
1377 for (i = 0; i < N_TABLES; i++) {
1378 struct table_dpif *table = &ofproto->tables[i];
1380 table->catchall_table = NULL;
1381 table->other_table = NULL;
1382 table->basis = random_uint32();
1385 list_init(&ofproto->completions);
1387 ofproto_dpif_unixctl_init();
1389 ofproto->has_mirrors = false;
1390 ofproto->has_bundle_action = false;
1392 hmap_init(&ofproto->vlandev_map);
1393 hmap_init(&ofproto->realdev_vid_map);
1395 sset_init(&ofproto->ports);
1396 sset_init(&ofproto->ghost_ports);
1397 sset_init(&ofproto->port_poll_set);
1398 ofproto->port_poll_errno = 0;
1400 SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1401 struct iface_hint *iface_hint = node->data;
1403 if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1404 /* Check if the datapath already has this port. */
1405 if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1406 sset_add(&ofproto->ports, node->name);
1409 free(iface_hint->br_name);
1410 free(iface_hint->br_type);
1412 shash_delete(&init_ofp_ports, node);
1416 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1417 hash_string(ofproto->up.name, 0));
1418 memset(&ofproto->stats, 0, sizeof ofproto->stats);
1420 ofproto_init_tables(ofproto_, N_TABLES);
1421 error = add_internal_flows(ofproto);
1422 ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1425 ofproto->n_missed = 0;
1427 ofproto->max_n_subfacet = 0;
1428 ofproto->created = time_msec();
1429 ofproto->last_minute = ofproto->created;
1430 memset(&ofproto->hourly, 0, sizeof ofproto->hourly);
1431 memset(&ofproto->daily, 0, sizeof ofproto->daily);
1432 ofproto->subfacet_add_count = 0;
1433 ofproto->subfacet_del_count = 0;
1434 ofproto->total_subfacet_add_count = 0;
1435 ofproto->total_subfacet_del_count = 0;
1436 ofproto->total_subfacet_life_span = 0;
1437 ofproto->total_subfacet_count = 0;
1438 ofproto->n_update_stats = 0;
1444 add_internal_flow(struct ofproto_dpif *ofproto, int id,
1445 const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1447 struct ofputil_flow_mod fm;
1450 match_init_catchall(&fm.match);
1452 match_set_reg(&fm.match, 0, id);
1453 fm.new_cookie = htonll(0);
1454 fm.cookie = htonll(0);
1455 fm.cookie_mask = htonll(0);
1456 fm.table_id = TBL_INTERNAL;
1457 fm.command = OFPFC_ADD;
1458 fm.idle_timeout = 0;
1459 fm.hard_timeout = 0;
1463 fm.ofpacts = ofpacts->data;
1464 fm.ofpacts_len = ofpacts->size;
1466 error = ofproto_flow_mod(&ofproto->up, &fm);
1468 VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
1469 id, ofperr_to_string(error));
1473 *rulep = rule_dpif_lookup__(ofproto, &fm.match.flow, TBL_INTERNAL);
1474 ovs_assert(*rulep != NULL);
1480 add_internal_flows(struct ofproto_dpif *ofproto)
1482 struct ofpact_controller *controller;
1483 uint64_t ofpacts_stub[128 / 8];
1484 struct ofpbuf ofpacts;
1488 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1491 controller = ofpact_put_CONTROLLER(&ofpacts);
1492 controller->max_len = UINT16_MAX;
1493 controller->controller_id = 0;
1494 controller->reason = OFPR_NO_MATCH;
1495 ofpact_pad(&ofpacts);
1497 error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule);
1502 ofpbuf_clear(&ofpacts);
1503 error = add_internal_flow(ofproto, id++, &ofpacts,
1504 &ofproto->no_packet_in_rule);
1509 complete_operations(struct ofproto_dpif *ofproto)
1511 struct dpif_completion *c, *next;
1513 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
1514 ofoperation_complete(c->op, 0);
1515 list_remove(&c->list_node);
1521 destruct(struct ofproto *ofproto_)
1523 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1524 struct rule_dpif *rule, *next_rule;
1525 struct oftable *table;
1528 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1529 complete_operations(ofproto);
1531 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1532 struct cls_cursor cursor;
1534 cls_cursor_init(&cursor, &table->cls, NULL);
1535 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1536 ofproto_rule_destroy(&rule->up);
1540 for (i = 0; i < MAX_MIRRORS; i++) {
1541 mirror_destroy(ofproto->mirrors[i]);
1544 netflow_destroy(ofproto->netflow);
1545 dpif_sflow_destroy(ofproto->sflow);
1546 hmap_destroy(&ofproto->bundles);
1547 mac_learning_destroy(ofproto->ml);
1549 hmap_destroy(&ofproto->facets);
1550 hmap_destroy(&ofproto->subfacets);
1551 governor_destroy(ofproto->governor);
1553 hmap_destroy(&ofproto->vlandev_map);
1554 hmap_destroy(&ofproto->realdev_vid_map);
1556 sset_destroy(&ofproto->ports);
1557 sset_destroy(&ofproto->ghost_ports);
1558 sset_destroy(&ofproto->port_poll_set);
1560 close_dpif_backer(ofproto->backer);
1564 run_fast(struct ofproto *ofproto_)
1566 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1567 struct ofport_dpif *ofport;
1569 /* Do not perform any periodic activity required by 'ofproto' while
1570 * waiting for flow restore to complete. */
1571 if (ofproto_get_flow_restore_wait()) {
1575 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1576 port_run_fast(ofport);
1583 run(struct ofproto *ofproto_)
1585 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1586 struct ofport_dpif *ofport;
1587 struct ofbundle *bundle;
1591 complete_operations(ofproto);
1594 /* Do not perform any periodic activity below required by 'ofproto' while
1595 * waiting for flow restore to complete. */
1596 if (ofproto_get_flow_restore_wait()) {
1600 error = run_fast(ofproto_);
1605 if (ofproto->netflow) {
1606 if (netflow_run(ofproto->netflow)) {
1607 send_netflow_active_timeouts(ofproto);
1610 if (ofproto->sflow) {
1611 dpif_sflow_run(ofproto->sflow);
1614 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1617 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1622 mac_learning_run(ofproto->ml, &ofproto->backer->revalidate_set);
1624 /* Check the consistency of a random facet, to aid debugging. */
1625 if (time_msec() >= ofproto->consistency_rl
1626 && !hmap_is_empty(&ofproto->facets)
1627 && !ofproto->backer->need_revalidate) {
1628 struct facet *facet;
1630 ofproto->consistency_rl = time_msec() + 250;
1632 facet = CONTAINER_OF(hmap_random_node(&ofproto->facets),
1633 struct facet, hmap_node);
1634 if (!tag_set_intersects(&ofproto->backer->revalidate_set,
1635 facet->xout.tags)) {
1636 if (!facet_check_consistency(facet)) {
1637 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
1642 if (ofproto->governor) {
1645 governor_run(ofproto->governor);
1647 /* If the governor has shrunk to its minimum size and the number of
1648 * subfacets has dwindled, then drop the governor entirely.
1650 * For hysteresis, the number of subfacets to drop the governor is
1651 * smaller than the number needed to trigger its creation. */
1652 n_subfacets = hmap_count(&ofproto->subfacets);
1653 if (n_subfacets * 4 < ofproto->up.flow_eviction_threshold
1654 && governor_is_idle(ofproto->governor)) {
1655 governor_destroy(ofproto->governor);
1656 ofproto->governor = NULL;
1664 wait(struct ofproto *ofproto_)
1666 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1667 struct ofport_dpif *ofport;
1668 struct ofbundle *bundle;
1670 if (!clogged && !list_is_empty(&ofproto->completions)) {
1671 poll_immediate_wake();
1674 if (ofproto_get_flow_restore_wait()) {
1678 dpif_wait(ofproto->backer->dpif);
1679 dpif_recv_wait(ofproto->backer->dpif);
1680 if (ofproto->sflow) {
1681 dpif_sflow_wait(ofproto->sflow);
1683 if (!tag_set_is_empty(&ofproto->backer->revalidate_set)) {
1684 poll_immediate_wake();
1686 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1689 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1690 bundle_wait(bundle);
1692 if (ofproto->netflow) {
1693 netflow_wait(ofproto->netflow);
1695 mac_learning_wait(ofproto->ml);
1697 if (ofproto->backer->need_revalidate) {
1698 /* Shouldn't happen, but if it does just go around again. */
1699 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1700 poll_immediate_wake();
1702 if (ofproto->governor) {
1703 governor_wait(ofproto->governor);
1708 get_memory_usage(const struct ofproto *ofproto_, struct simap *usage)
1710 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1712 simap_increase(usage, "facets", hmap_count(&ofproto->facets));
1713 simap_increase(usage, "subfacets", hmap_count(&ofproto->subfacets));
1717 flush(struct ofproto *ofproto_)
1719 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1720 struct subfacet *subfacet, *next_subfacet;
1721 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
1725 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
1726 &ofproto->subfacets) {
1727 if (subfacet->path != SF_NOT_INSTALLED) {
1728 batch[n_batch++] = subfacet;
1729 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
1730 subfacet_destroy_batch(ofproto, batch, n_batch);
1734 subfacet_destroy(subfacet);
1739 subfacet_destroy_batch(ofproto, batch, n_batch);
1744 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1745 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1747 *arp_match_ip = true;
1748 *actions = (OFPUTIL_A_OUTPUT |
1749 OFPUTIL_A_SET_VLAN_VID |
1750 OFPUTIL_A_SET_VLAN_PCP |
1751 OFPUTIL_A_STRIP_VLAN |
1752 OFPUTIL_A_SET_DL_SRC |
1753 OFPUTIL_A_SET_DL_DST |
1754 OFPUTIL_A_SET_NW_SRC |
1755 OFPUTIL_A_SET_NW_DST |
1756 OFPUTIL_A_SET_NW_TOS |
1757 OFPUTIL_A_SET_TP_SRC |
1758 OFPUTIL_A_SET_TP_DST |
1763 get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots)
1765 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1766 struct dpif_dp_stats s;
1767 uint64_t n_miss, n_no_pkt_in, n_bytes;
1770 strcpy(ots->name, "classifier");
1772 dpif_get_dp_stats(ofproto->backer->dpif, &s);
1773 rule_get_stats(&ofproto->miss_rule->up, &n_miss, &n_bytes);
1774 rule_get_stats(&ofproto->no_packet_in_rule->up, &n_no_pkt_in, &n_bytes);
1776 n_lookup = s.n_hit + s.n_missed;
1777 ots->lookup_count = htonll(n_lookup);
1778 ots->matched_count = htonll(n_lookup - n_miss - n_no_pkt_in);
1781 static struct ofport *
1784 struct ofport_dpif *port = xmalloc(sizeof *port);
1789 port_dealloc(struct ofport *port_)
1791 struct ofport_dpif *port = ofport_dpif_cast(port_);
1796 port_construct(struct ofport *port_)
1798 struct ofport_dpif *port = ofport_dpif_cast(port_);
1799 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1800 const struct netdev *netdev = port->up.netdev;
1801 struct dpif_port dpif_port;
1804 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1805 port->bundle = NULL;
1808 port->tag = tag_create_random();
1809 port->may_enable = true;
1810 port->stp_port = NULL;
1811 port->stp_state = STP_DISABLED;
1812 port->tnl_port = NULL;
1813 hmap_init(&port->priorities);
1814 port->realdev_ofp_port = 0;
1815 port->vlandev_vid = 0;
1816 port->carrier_seq = netdev_get_carrier_resets(netdev);
1818 if (netdev_vport_is_patch(netdev)) {
1819 /* By bailing out here, we don't submit the port to the sFlow module
1820 * to be considered for counter polling export. This is correct
1821 * because the patch port represents an interface that sFlow considers
1822 * to be "internal" to the switch as a whole, and therefore not an
1823 * candidate for counter polling. */
1824 port->odp_port = OVSP_NONE;
1828 error = dpif_port_query_by_name(ofproto->backer->dpif,
1829 netdev_vport_get_dpif_port(netdev),
1835 port->odp_port = dpif_port.port_no;
1837 if (netdev_get_tunnel_config(netdev)) {
1838 port->tnl_port = tnl_port_add(&port->up, port->odp_port);
1840 /* Sanity-check that a mapping doesn't already exist. This
1841 * shouldn't happen for non-tunnel ports. */
1842 if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1843 VLOG_ERR("port %s already has an OpenFlow port number",
1845 dpif_port_destroy(&dpif_port);
1849 hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1850 hash_int(port->odp_port, 0));
1852 dpif_port_destroy(&dpif_port);
1854 if (ofproto->sflow) {
1855 dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1862 port_destruct(struct ofport *port_)
1864 struct ofport_dpif *port = ofport_dpif_cast(port_);
1865 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1866 const char *dp_port_name = netdev_vport_get_dpif_port(port->up.netdev);
1867 const char *devname = netdev_get_name(port->up.netdev);
1869 if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1870 /* The underlying device is still there, so delete it. This
1871 * happens when the ofproto is being destroyed, since the caller
1872 * assumes that removal of attached ports will happen as part of
1874 if (!port->tnl_port) {
1875 dpif_port_del(ofproto->backer->dpif, port->odp_port);
1877 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1880 if (port->odp_port != OVSP_NONE && !port->tnl_port) {
1881 hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1884 tnl_port_del(port->tnl_port);
1885 sset_find_and_delete(&ofproto->ports, devname);
1886 sset_find_and_delete(&ofproto->ghost_ports, devname);
1887 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1888 bundle_remove(port_);
1889 set_cfm(port_, NULL);
1890 set_bfd(port_, NULL);
1891 if (ofproto->sflow) {
1892 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1895 ofport_clear_priorities(port);
1896 hmap_destroy(&port->priorities);
1900 port_modified(struct ofport *port_)
1902 struct ofport_dpif *port = ofport_dpif_cast(port_);
1904 if (port->bundle && port->bundle->bond) {
1905 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1910 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1912 struct ofport_dpif *port = ofport_dpif_cast(port_);
1913 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1914 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1916 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1917 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1918 OFPUTIL_PC_NO_PACKET_IN)) {
1919 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1921 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1922 bundle_update(port->bundle);
1928 set_sflow(struct ofproto *ofproto_,
1929 const struct ofproto_sflow_options *sflow_options)
1931 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1932 struct dpif_sflow *ds = ofproto->sflow;
1934 if (sflow_options) {
1936 struct ofport_dpif *ofport;
1938 ds = ofproto->sflow = dpif_sflow_create();
1939 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1940 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1942 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1944 dpif_sflow_set_options(ds, sflow_options);
1947 dpif_sflow_destroy(ds);
1948 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1949 ofproto->sflow = NULL;
1957 struct ofproto *ofproto_,
1958 const struct ofproto_ipfix_bridge_exporter_options *bridge_exporter_options,
1959 const struct ofproto_ipfix_flow_exporter_options *flow_exporters_options,
1960 size_t n_flow_exporters_options)
1962 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1963 struct dpif_ipfix *di = ofproto->ipfix;
1965 if (bridge_exporter_options || flow_exporters_options) {
1967 di = ofproto->ipfix = dpif_ipfix_create();
1969 dpif_ipfix_set_options(
1970 di, bridge_exporter_options, flow_exporters_options,
1971 n_flow_exporters_options);
1974 dpif_ipfix_destroy(di);
1975 ofproto->ipfix = NULL;
1982 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1984 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1991 struct ofproto_dpif *ofproto;
1993 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1994 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1995 ofport->cfm = cfm_create(ofport->up.netdev);
1998 if (cfm_configure(ofport->cfm, s)) {
2004 cfm_destroy(ofport->cfm);
2010 get_cfm_status(const struct ofport *ofport_,
2011 struct ofproto_cfm_status *status)
2013 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2016 status->faults = cfm_get_fault(ofport->cfm);
2017 status->remote_opstate = cfm_get_opup(ofport->cfm);
2018 status->health = cfm_get_health(ofport->cfm);
2019 cfm_get_remote_mpids(ofport->cfm, &status->rmps, &status->n_rmps);
2027 set_bfd(struct ofport *ofport_, const struct smap *cfg)
2029 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
2030 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2034 ofport->bfd = bfd_configure(old, netdev_get_name(ofport->up.netdev), cfg);
2035 if (ofport->bfd != old) {
2036 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2043 get_bfd_status(struct ofport *ofport_, struct smap *smap)
2045 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2048 bfd_get_status(ofport->bfd, smap);
2055 /* Spanning Tree. */
2058 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
2060 struct ofproto_dpif *ofproto = ofproto_;
2061 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
2062 struct ofport_dpif *ofport;
2064 ofport = stp_port_get_aux(sp);
2066 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
2067 ofproto->up.name, port_num);
2069 struct eth_header *eth = pkt->l2;
2071 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
2072 if (eth_addr_is_zero(eth->eth_src)) {
2073 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
2074 "with unknown MAC", ofproto->up.name, port_num);
2076 send_packet(ofport, pkt);
2082 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
2084 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
2086 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2088 /* Only revalidate flows if the configuration changed. */
2089 if (!s != !ofproto->stp) {
2090 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2094 if (!ofproto->stp) {
2095 ofproto->stp = stp_create(ofproto_->name, s->system_id,
2096 send_bpdu_cb, ofproto);
2097 ofproto->stp_last_tick = time_msec();
2100 stp_set_bridge_id(ofproto->stp, s->system_id);
2101 stp_set_bridge_priority(ofproto->stp, s->priority);
2102 stp_set_hello_time(ofproto->stp, s->hello_time);
2103 stp_set_max_age(ofproto->stp, s->max_age);
2104 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
2106 struct ofport *ofport;
2108 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
2109 set_stp_port(ofport, NULL);
2112 stp_destroy(ofproto->stp);
2113 ofproto->stp = NULL;
2120 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
2122 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2126 s->bridge_id = stp_get_bridge_id(ofproto->stp);
2127 s->designated_root = stp_get_designated_root(ofproto->stp);
2128 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
2137 update_stp_port_state(struct ofport_dpif *ofport)
2139 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2140 enum stp_state state;
2142 /* Figure out new state. */
2143 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
2147 if (ofport->stp_state != state) {
2148 enum ofputil_port_state of_state;
2151 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
2152 netdev_get_name(ofport->up.netdev),
2153 stp_state_name(ofport->stp_state),
2154 stp_state_name(state));
2155 if (stp_learn_in_state(ofport->stp_state)
2156 != stp_learn_in_state(state)) {
2157 /* xxx Learning action flows should also be flushed. */
2158 mac_learning_flush(ofproto->ml,
2159 &ofproto->backer->revalidate_set);
2161 fwd_change = stp_forward_in_state(ofport->stp_state)
2162 != stp_forward_in_state(state);
2164 ofproto->backer->need_revalidate = REV_STP;
2165 ofport->stp_state = state;
2166 ofport->stp_state_entered = time_msec();
2168 if (fwd_change && ofport->bundle) {
2169 bundle_update(ofport->bundle);
2172 /* Update the STP state bits in the OpenFlow port description. */
2173 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
2174 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
2175 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
2176 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
2177 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
2179 ofproto_port_set_state(&ofport->up, of_state);
2183 /* Configures STP on 'ofport_' using the settings defined in 's'. The
2184 * caller is responsible for assigning STP port numbers and ensuring
2185 * there are no duplicates. */
2187 set_stp_port(struct ofport *ofport_,
2188 const struct ofproto_port_stp_settings *s)
2190 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2191 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2192 struct stp_port *sp = ofport->stp_port;
2194 if (!s || !s->enable) {
2196 ofport->stp_port = NULL;
2197 stp_port_disable(sp);
2198 update_stp_port_state(ofport);
2201 } else if (sp && stp_port_no(sp) != s->port_num
2202 && ofport == stp_port_get_aux(sp)) {
2203 /* The port-id changed, so disable the old one if it's not
2204 * already in use by another port. */
2205 stp_port_disable(sp);
2208 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
2209 stp_port_enable(sp);
2211 stp_port_set_aux(sp, ofport);
2212 stp_port_set_priority(sp, s->priority);
2213 stp_port_set_path_cost(sp, s->path_cost);
2215 update_stp_port_state(ofport);
2221 get_stp_port_status(struct ofport *ofport_,
2222 struct ofproto_port_stp_status *s)
2224 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2225 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2226 struct stp_port *sp = ofport->stp_port;
2228 if (!ofproto->stp || !sp) {
2234 s->port_id = stp_port_get_id(sp);
2235 s->state = stp_port_get_state(sp);
2236 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
2237 s->role = stp_port_get_role(sp);
2238 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
2244 stp_run(struct ofproto_dpif *ofproto)
2247 long long int now = time_msec();
2248 long long int elapsed = now - ofproto->stp_last_tick;
2249 struct stp_port *sp;
2252 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
2253 ofproto->stp_last_tick = now;
2255 while (stp_get_changed_port(ofproto->stp, &sp)) {
2256 struct ofport_dpif *ofport = stp_port_get_aux(sp);
2259 update_stp_port_state(ofport);
2263 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
2264 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
2270 stp_wait(struct ofproto_dpif *ofproto)
2273 poll_timer_wait(1000);
2277 /* Returns true if STP should process 'flow'. */
2279 stp_should_process_flow(const struct flow *flow)
2281 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
2285 stp_process_packet(const struct ofport_dpif *ofport,
2286 const struct ofpbuf *packet)
2288 struct ofpbuf payload = *packet;
2289 struct eth_header *eth = payload.data;
2290 struct stp_port *sp = ofport->stp_port;
2292 /* Sink packets on ports that have STP disabled when the bridge has
2294 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
2298 /* Trim off padding on payload. */
2299 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
2300 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
2303 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
2304 stp_received_bpdu(sp, payload.data, payload.size);
2308 static struct priority_to_dscp *
2309 get_priority(const struct ofport_dpif *ofport, uint32_t priority)
2311 struct priority_to_dscp *pdscp;
2314 hash = hash_int(priority, 0);
2315 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) {
2316 if (pdscp->priority == priority) {
2324 ofport_clear_priorities(struct ofport_dpif *ofport)
2326 struct priority_to_dscp *pdscp, *next;
2328 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) {
2329 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
2335 set_queues(struct ofport *ofport_,
2336 const struct ofproto_port_queue *qdscp_list,
2339 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2340 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2341 struct hmap new = HMAP_INITIALIZER(&new);
2344 for (i = 0; i < n_qdscp; i++) {
2345 struct priority_to_dscp *pdscp;
2349 dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
2350 if (dpif_queue_to_priority(ofproto->backer->dpif, qdscp_list[i].queue,
2355 pdscp = get_priority(ofport, priority);
2357 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
2359 pdscp = xmalloc(sizeof *pdscp);
2360 pdscp->priority = priority;
2362 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2365 if (pdscp->dscp != dscp) {
2367 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2370 hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0));
2373 if (!hmap_is_empty(&ofport->priorities)) {
2374 ofport_clear_priorities(ofport);
2375 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2378 hmap_swap(&new, &ofport->priorities);
2386 /* Expires all MAC learning entries associated with 'bundle' and forces its
2387 * ofproto to revalidate every flow.
2389 * Normally MAC learning entries are removed only from the ofproto associated
2390 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2391 * are removed from every ofproto. When patch ports and SLB bonds are in use
2392 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2393 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2394 * with the host from which it migrated. */
2396 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2398 struct ofproto_dpif *ofproto = bundle->ofproto;
2399 struct mac_learning *ml = ofproto->ml;
2400 struct mac_entry *mac, *next_mac;
2402 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2403 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2404 if (mac->port.p == bundle) {
2406 struct ofproto_dpif *o;
2408 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2410 struct mac_entry *e;
2412 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan,
2415 mac_learning_expire(o->ml, e);
2421 mac_learning_expire(ml, mac);
2426 static struct ofbundle *
2427 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2429 struct ofbundle *bundle;
2431 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2432 &ofproto->bundles) {
2433 if (bundle->aux == aux) {
2440 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
2441 * ones that are found to 'bundles'. */
2443 bundle_lookup_multiple(struct ofproto_dpif *ofproto,
2444 void **auxes, size_t n_auxes,
2445 struct hmapx *bundles)
2449 hmapx_init(bundles);
2450 for (i = 0; i < n_auxes; i++) {
2451 struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
2453 hmapx_add(bundles, bundle);
2459 bundle_update(struct ofbundle *bundle)
2461 struct ofport_dpif *port;
2463 bundle->floodable = true;
2464 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2465 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2466 || !stp_forward_in_state(port->stp_state)) {
2467 bundle->floodable = false;
2474 bundle_del_port(struct ofport_dpif *port)
2476 struct ofbundle *bundle = port->bundle;
2478 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2480 list_remove(&port->bundle_node);
2481 port->bundle = NULL;
2484 lacp_slave_unregister(bundle->lacp, port);
2487 bond_slave_unregister(bundle->bond, port);
2490 bundle_update(bundle);
2494 bundle_add_port(struct ofbundle *bundle, uint16_t ofp_port,
2495 struct lacp_slave_settings *lacp)
2497 struct ofport_dpif *port;
2499 port = get_ofp_port(bundle->ofproto, ofp_port);
2504 if (port->bundle != bundle) {
2505 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2507 bundle_del_port(port);
2510 port->bundle = bundle;
2511 list_push_back(&bundle->ports, &port->bundle_node);
2512 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2513 || !stp_forward_in_state(port->stp_state)) {
2514 bundle->floodable = false;
2518 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2519 lacp_slave_register(bundle->lacp, port, lacp);
2526 bundle_destroy(struct ofbundle *bundle)
2528 struct ofproto_dpif *ofproto;
2529 struct ofport_dpif *port, *next_port;
2536 ofproto = bundle->ofproto;
2537 for (i = 0; i < MAX_MIRRORS; i++) {
2538 struct ofmirror *m = ofproto->mirrors[i];
2540 if (m->out == bundle) {
2542 } else if (hmapx_find_and_delete(&m->srcs, bundle)
2543 || hmapx_find_and_delete(&m->dsts, bundle)) {
2544 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2549 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2550 bundle_del_port(port);
2553 bundle_flush_macs(bundle, true);
2554 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2556 free(bundle->trunks);
2557 lacp_destroy(bundle->lacp);
2558 bond_destroy(bundle->bond);
2563 bundle_set(struct ofproto *ofproto_, void *aux,
2564 const struct ofproto_bundle_settings *s)
2566 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2567 bool need_flush = false;
2568 struct ofport_dpif *port;
2569 struct ofbundle *bundle;
2570 unsigned long *trunks;
2576 bundle_destroy(bundle_lookup(ofproto, aux));
2580 ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2581 ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2583 bundle = bundle_lookup(ofproto, aux);
2585 bundle = xmalloc(sizeof *bundle);
2587 bundle->ofproto = ofproto;
2588 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2589 hash_pointer(aux, 0));
2591 bundle->name = NULL;
2593 list_init(&bundle->ports);
2594 bundle->vlan_mode = PORT_VLAN_TRUNK;
2596 bundle->trunks = NULL;
2597 bundle->use_priority_tags = s->use_priority_tags;
2598 bundle->lacp = NULL;
2599 bundle->bond = NULL;
2601 bundle->floodable = true;
2603 bundle->src_mirrors = 0;
2604 bundle->dst_mirrors = 0;
2605 bundle->mirror_out = 0;
2608 if (!bundle->name || strcmp(s->name, bundle->name)) {
2610 bundle->name = xstrdup(s->name);
2615 if (!bundle->lacp) {
2616 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2617 bundle->lacp = lacp_create();
2619 lacp_configure(bundle->lacp, s->lacp);
2621 lacp_destroy(bundle->lacp);
2622 bundle->lacp = NULL;
2625 /* Update set of ports. */
2627 for (i = 0; i < s->n_slaves; i++) {
2628 if (!bundle_add_port(bundle, s->slaves[i],
2629 s->lacp ? &s->lacp_slaves[i] : NULL)) {
2633 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2634 struct ofport_dpif *next_port;
2636 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2637 for (i = 0; i < s->n_slaves; i++) {
2638 if (s->slaves[i] == port->up.ofp_port) {
2643 bundle_del_port(port);
2647 ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2649 if (list_is_empty(&bundle->ports)) {
2650 bundle_destroy(bundle);
2654 /* Set VLAN tagging mode */
2655 if (s->vlan_mode != bundle->vlan_mode
2656 || s->use_priority_tags != bundle->use_priority_tags) {
2657 bundle->vlan_mode = s->vlan_mode;
2658 bundle->use_priority_tags = s->use_priority_tags;
2663 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2664 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2666 if (vlan != bundle->vlan) {
2667 bundle->vlan = vlan;
2671 /* Get trunked VLANs. */
2672 switch (s->vlan_mode) {
2673 case PORT_VLAN_ACCESS:
2677 case PORT_VLAN_TRUNK:
2678 trunks = CONST_CAST(unsigned long *, s->trunks);
2681 case PORT_VLAN_NATIVE_UNTAGGED:
2682 case PORT_VLAN_NATIVE_TAGGED:
2683 if (vlan != 0 && (!s->trunks
2684 || !bitmap_is_set(s->trunks, vlan)
2685 || bitmap_is_set(s->trunks, 0))) {
2686 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2688 trunks = bitmap_clone(s->trunks, 4096);
2690 trunks = bitmap_allocate1(4096);
2692 bitmap_set1(trunks, vlan);
2693 bitmap_set0(trunks, 0);
2695 trunks = CONST_CAST(unsigned long *, s->trunks);
2702 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2703 free(bundle->trunks);
2704 if (trunks == s->trunks) {
2705 bundle->trunks = vlan_bitmap_clone(trunks);
2707 bundle->trunks = trunks;
2712 if (trunks != s->trunks) {
2717 if (!list_is_short(&bundle->ports)) {
2718 bundle->ofproto->has_bonded_bundles = true;
2720 if (bond_reconfigure(bundle->bond, s->bond)) {
2721 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2724 bundle->bond = bond_create(s->bond);
2725 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2728 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2729 bond_slave_register(bundle->bond, port, port->up.netdev);
2732 bond_destroy(bundle->bond);
2733 bundle->bond = NULL;
2736 /* If we changed something that would affect MAC learning, un-learn
2737 * everything on this port and force flow revalidation. */
2739 bundle_flush_macs(bundle, false);
2746 bundle_remove(struct ofport *port_)
2748 struct ofport_dpif *port = ofport_dpif_cast(port_);
2749 struct ofbundle *bundle = port->bundle;
2752 bundle_del_port(port);
2753 if (list_is_empty(&bundle->ports)) {
2754 bundle_destroy(bundle);
2755 } else if (list_is_short(&bundle->ports)) {
2756 bond_destroy(bundle->bond);
2757 bundle->bond = NULL;
2763 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2765 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2766 struct ofport_dpif *port = port_;
2767 uint8_t ea[ETH_ADDR_LEN];
2770 error = netdev_get_etheraddr(port->up.netdev, ea);
2772 struct ofpbuf packet;
2775 ofpbuf_init(&packet, 0);
2776 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2778 memcpy(packet_pdu, pdu, pdu_size);
2780 send_packet(port, &packet);
2781 ofpbuf_uninit(&packet);
2783 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2784 "%s (%s)", port->bundle->name,
2785 netdev_get_name(port->up.netdev), strerror(error));
2790 bundle_send_learning_packets(struct ofbundle *bundle)
2792 struct ofproto_dpif *ofproto = bundle->ofproto;
2793 int error, n_packets, n_errors;
2794 struct mac_entry *e;
2796 error = n_packets = n_errors = 0;
2797 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2798 if (e->port.p != bundle) {
2799 struct ofpbuf *learning_packet;
2800 struct ofport_dpif *port;
2804 /* The assignment to "port" is unnecessary but makes "grep"ing for
2805 * struct ofport_dpif more effective. */
2806 learning_packet = bond_compose_learning_packet(bundle->bond,
2810 ret = send_packet(port, learning_packet);
2811 ofpbuf_delete(learning_packet);
2821 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2822 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2823 "packets, last error was: %s",
2824 bundle->name, n_errors, n_packets, strerror(error));
2826 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2827 bundle->name, n_packets);
2832 bundle_run(struct ofbundle *bundle)
2835 lacp_run(bundle->lacp, send_pdu_cb);
2838 struct ofport_dpif *port;
2840 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2841 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2844 bond_run(bundle->bond, &bundle->ofproto->backer->revalidate_set,
2845 lacp_status(bundle->lacp));
2846 if (bond_should_send_learning_packets(bundle->bond)) {
2847 bundle_send_learning_packets(bundle);
2853 bundle_wait(struct ofbundle *bundle)
2856 lacp_wait(bundle->lacp);
2859 bond_wait(bundle->bond);
2866 mirror_scan(struct ofproto_dpif *ofproto)
2870 for (idx = 0; idx < MAX_MIRRORS; idx++) {
2871 if (!ofproto->mirrors[idx]) {
2878 static struct ofmirror *
2879 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
2883 for (i = 0; i < MAX_MIRRORS; i++) {
2884 struct ofmirror *mirror = ofproto->mirrors[i];
2885 if (mirror && mirror->aux == aux) {
2893 /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */
2895 mirror_update_dups(struct ofproto_dpif *ofproto)
2899 for (i = 0; i < MAX_MIRRORS; i++) {
2900 struct ofmirror *m = ofproto->mirrors[i];
2903 m->dup_mirrors = MIRROR_MASK_C(1) << i;
2907 for (i = 0; i < MAX_MIRRORS; i++) {
2908 struct ofmirror *m1 = ofproto->mirrors[i];
2915 for (j = i + 1; j < MAX_MIRRORS; j++) {
2916 struct ofmirror *m2 = ofproto->mirrors[j];
2918 if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) {
2919 m1->dup_mirrors |= MIRROR_MASK_C(1) << j;
2920 m2->dup_mirrors |= m1->dup_mirrors;
2927 mirror_set(struct ofproto *ofproto_, void *aux,
2928 const struct ofproto_mirror_settings *s)
2930 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2931 mirror_mask_t mirror_bit;
2932 struct ofbundle *bundle;
2933 struct ofmirror *mirror;
2934 struct ofbundle *out;
2935 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
2936 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
2939 mirror = mirror_lookup(ofproto, aux);
2941 mirror_destroy(mirror);
2947 idx = mirror_scan(ofproto);
2949 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
2951 ofproto->up.name, MAX_MIRRORS, s->name);
2955 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
2956 mirror->ofproto = ofproto;
2959 mirror->out_vlan = -1;
2960 mirror->name = NULL;
2963 if (!mirror->name || strcmp(s->name, mirror->name)) {
2965 mirror->name = xstrdup(s->name);
2968 /* Get the new configuration. */
2969 if (s->out_bundle) {
2970 out = bundle_lookup(ofproto, s->out_bundle);
2972 mirror_destroy(mirror);
2978 out_vlan = s->out_vlan;
2980 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
2981 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
2983 /* If the configuration has not changed, do nothing. */
2984 if (hmapx_equals(&srcs, &mirror->srcs)
2985 && hmapx_equals(&dsts, &mirror->dsts)
2986 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
2987 && mirror->out == out
2988 && mirror->out_vlan == out_vlan)
2990 hmapx_destroy(&srcs);
2991 hmapx_destroy(&dsts);
2995 hmapx_swap(&srcs, &mirror->srcs);
2996 hmapx_destroy(&srcs);
2998 hmapx_swap(&dsts, &mirror->dsts);
2999 hmapx_destroy(&dsts);
3001 free(mirror->vlans);
3002 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
3005 mirror->out_vlan = out_vlan;
3007 /* Update bundles. */
3008 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
3009 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
3010 if (hmapx_contains(&mirror->srcs, bundle)) {
3011 bundle->src_mirrors |= mirror_bit;
3013 bundle->src_mirrors &= ~mirror_bit;
3016 if (hmapx_contains(&mirror->dsts, bundle)) {
3017 bundle->dst_mirrors |= mirror_bit;
3019 bundle->dst_mirrors &= ~mirror_bit;
3022 if (mirror->out == bundle) {
3023 bundle->mirror_out |= mirror_bit;
3025 bundle->mirror_out &= ~mirror_bit;
3029 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3030 ofproto->has_mirrors = true;
3031 mac_learning_flush(ofproto->ml,
3032 &ofproto->backer->revalidate_set);
3033 mirror_update_dups(ofproto);
3039 mirror_destroy(struct ofmirror *mirror)
3041 struct ofproto_dpif *ofproto;
3042 mirror_mask_t mirror_bit;
3043 struct ofbundle *bundle;
3050 ofproto = mirror->ofproto;
3051 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3052 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
3054 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
3055 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
3056 bundle->src_mirrors &= ~mirror_bit;
3057 bundle->dst_mirrors &= ~mirror_bit;
3058 bundle->mirror_out &= ~mirror_bit;
3061 hmapx_destroy(&mirror->srcs);
3062 hmapx_destroy(&mirror->dsts);
3063 free(mirror->vlans);
3065 ofproto->mirrors[mirror->idx] = NULL;
3069 mirror_update_dups(ofproto);
3071 ofproto->has_mirrors = false;
3072 for (i = 0; i < MAX_MIRRORS; i++) {
3073 if (ofproto->mirrors[i]) {
3074 ofproto->has_mirrors = true;
3081 mirror_get_stats(struct ofproto *ofproto_, void *aux,
3082 uint64_t *packets, uint64_t *bytes)
3084 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3085 struct ofmirror *mirror = mirror_lookup(ofproto, aux);
3088 *packets = *bytes = UINT64_MAX;
3094 *packets = mirror->packet_count;
3095 *bytes = mirror->byte_count;
3101 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
3103 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3104 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
3105 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
3111 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
3113 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3114 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
3115 return bundle && bundle->mirror_out != 0;
3119 forward_bpdu_changed(struct ofproto *ofproto_)
3121 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3122 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3126 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
3129 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3130 mac_learning_set_idle_time(ofproto->ml, idle_time);
3131 mac_learning_set_max_entries(ofproto->ml, max_entries);
3136 static struct ofport_dpif *
3137 get_ofp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
3139 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
3140 return ofport ? ofport_dpif_cast(ofport) : NULL;
3143 static struct ofport_dpif *
3144 get_odp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
3146 struct ofport_dpif *port = odp_port_to_ofport(ofproto->backer, odp_port);
3147 return port && &ofproto->up == port->up.ofproto ? port : NULL;
3151 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
3152 struct ofproto_port *ofproto_port,
3153 struct dpif_port *dpif_port)
3155 ofproto_port->name = dpif_port->name;
3156 ofproto_port->type = dpif_port->type;
3157 ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
3160 static struct ofport_dpif *
3161 ofport_get_peer(const struct ofport_dpif *ofport_dpif)
3163 const struct ofproto_dpif *ofproto;
3166 peer = netdev_vport_patch_peer(ofport_dpif->up.netdev);
3171 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
3172 struct ofport *ofport;
3174 ofport = shash_find_data(&ofproto->up.port_by_name, peer);
3175 if (ofport && ofport->ofproto->ofproto_class == &ofproto_dpif_class) {
3176 return ofport_dpif_cast(ofport);
3183 port_run_fast(struct ofport_dpif *ofport)
3185 if (ofport->cfm && cfm_should_send_ccm(ofport->cfm)) {
3186 struct ofpbuf packet;
3188 ofpbuf_init(&packet, 0);
3189 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
3190 send_packet(ofport, &packet);
3191 ofpbuf_uninit(&packet);
3194 if (ofport->bfd && bfd_should_send_packet(ofport->bfd)) {
3195 struct ofpbuf packet;
3197 ofpbuf_init(&packet, 0);
3198 bfd_put_packet(ofport->bfd, &packet, ofport->up.pp.hw_addr);
3199 send_packet(ofport, &packet);
3200 ofpbuf_uninit(&packet);
3205 port_run(struct ofport_dpif *ofport)
3207 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
3208 bool carrier_changed = carrier_seq != ofport->carrier_seq;
3209 bool enable = netdev_get_carrier(ofport->up.netdev);
3211 ofport->carrier_seq = carrier_seq;
3213 port_run_fast(ofport);
3215 if (ofport->tnl_port
3216 && tnl_port_reconfigure(&ofport->up, ofport->odp_port,
3217 &ofport->tnl_port)) {
3218 ofproto_dpif_cast(ofport->up.ofproto)->backer->need_revalidate = true;
3222 int cfm_opup = cfm_get_opup(ofport->cfm);
3224 cfm_run(ofport->cfm);
3225 enable = enable && !cfm_get_fault(ofport->cfm);
3227 if (cfm_opup >= 0) {
3228 enable = enable && cfm_opup;
3233 bfd_run(ofport->bfd);
3234 enable = enable && bfd_forwarding(ofport->bfd);
3237 if (ofport->bundle) {
3238 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
3239 if (carrier_changed) {
3240 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
3244 if (ofport->may_enable != enable) {
3245 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3247 if (ofproto->has_bundle_action) {
3248 ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
3252 ofport->may_enable = enable;
3256 port_wait(struct ofport_dpif *ofport)
3259 cfm_wait(ofport->cfm);
3263 bfd_wait(ofport->bfd);
3268 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
3269 struct ofproto_port *ofproto_port)
3271 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3272 struct dpif_port dpif_port;
3275 if (sset_contains(&ofproto->ghost_ports, devname)) {
3276 const char *type = netdev_get_type_from_name(devname);
3278 /* We may be called before ofproto->up.port_by_name is populated with
3279 * the appropriate ofport. For this reason, we must get the name and
3280 * type from the netdev layer directly. */
3282 const struct ofport *ofport;
3284 ofport = shash_find_data(&ofproto->up.port_by_name, devname);
3285 ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
3286 ofproto_port->name = xstrdup(devname);
3287 ofproto_port->type = xstrdup(type);
3293 if (!sset_contains(&ofproto->ports, devname)) {
3296 error = dpif_port_query_by_name(ofproto->backer->dpif,
3297 devname, &dpif_port);
3299 ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
3305 port_add(struct ofproto *ofproto_, struct netdev *netdev)
3307 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3308 const char *dp_port_name = netdev_vport_get_dpif_port(netdev);
3309 const char *devname = netdev_get_name(netdev);
3311 if (netdev_vport_is_patch(netdev)) {
3312 sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
3316 if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
3317 uint32_t port_no = UINT32_MAX;
3320 error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
3324 if (netdev_get_tunnel_config(netdev)) {
3325 simap_put(&ofproto->backer->tnl_backers, dp_port_name, port_no);
3329 if (netdev_get_tunnel_config(netdev)) {
3330 sset_add(&ofproto->ghost_ports, devname);
3332 sset_add(&ofproto->ports, devname);
3338 port_del(struct ofproto *ofproto_, uint16_t ofp_port)
3340 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3341 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
3348 sset_find_and_delete(&ofproto->ghost_ports,
3349 netdev_get_name(ofport->up.netdev));
3350 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3351 if (!ofport->tnl_port) {
3352 error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
3354 /* The caller is going to close ofport->up.netdev. If this is a
3355 * bonded port, then the bond is using that netdev, so remove it
3356 * from the bond. The client will need to reconfigure everything
3357 * after deleting ports, so then the slave will get re-added. */
3358 bundle_remove(&ofport->up);
3365 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
3367 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3372 error = netdev_get_stats(ofport->up.netdev, stats);
3374 if (!error && ofport_->ofp_port == OFPP_LOCAL) {
3375 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3377 /* ofproto->stats.tx_packets represents packets that we created
3378 * internally and sent to some port (e.g. packets sent with
3379 * send_packet()). Account for them as if they had come from
3380 * OFPP_LOCAL and got forwarded. */
3382 if (stats->rx_packets != UINT64_MAX) {
3383 stats->rx_packets += ofproto->stats.tx_packets;
3386 if (stats->rx_bytes != UINT64_MAX) {
3387 stats->rx_bytes += ofproto->stats.tx_bytes;
3390 /* ofproto->stats.rx_packets represents packets that were received on
3391 * some port and we processed internally and dropped (e.g. STP).
3392 * Account for them as if they had been forwarded to OFPP_LOCAL. */
3394 if (stats->tx_packets != UINT64_MAX) {
3395 stats->tx_packets += ofproto->stats.rx_packets;
3398 if (stats->tx_bytes != UINT64_MAX) {
3399 stats->tx_bytes += ofproto->stats.rx_bytes;
3406 struct port_dump_state {
3411 struct ofproto_port port;
3416 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
3418 *statep = xzalloc(sizeof(struct port_dump_state));
3423 port_dump_next(const struct ofproto *ofproto_, void *state_,
3424 struct ofproto_port *port)
3426 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3427 struct port_dump_state *state = state_;
3428 const struct sset *sset;
3429 struct sset_node *node;
3431 if (state->has_port) {
3432 ofproto_port_destroy(&state->port);
3433 state->has_port = false;
3435 sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
3436 while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
3439 error = port_query_by_name(ofproto_, node->name, &state->port);
3441 *port = state->port;
3442 state->has_port = true;
3444 } else if (error != ENODEV) {
3449 if (!state->ghost) {
3450 state->ghost = true;
3453 return port_dump_next(ofproto_, state_, port);
3460 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3462 struct port_dump_state *state = state_;
3464 if (state->has_port) {
3465 ofproto_port_destroy(&state->port);
3472 port_poll(const struct ofproto *ofproto_, char **devnamep)
3474 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3476 if (ofproto->port_poll_errno) {
3477 int error = ofproto->port_poll_errno;
3478 ofproto->port_poll_errno = 0;
3482 if (sset_is_empty(&ofproto->port_poll_set)) {
3486 *devnamep = sset_pop(&ofproto->port_poll_set);
3491 port_poll_wait(const struct ofproto *ofproto_)
3493 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3494 dpif_port_poll_wait(ofproto->backer->dpif);
3498 port_is_lacp_current(const struct ofport *ofport_)
3500 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3501 return (ofport->bundle && ofport->bundle->lacp
3502 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3506 /* Upcall handling. */
3508 /* Flow miss batching.
3510 * Some dpifs implement operations faster when you hand them off in a batch.
3511 * To allow batching, "struct flow_miss" queues the dpif-related work needed
3512 * for a given flow. Each "struct flow_miss" corresponds to sending one or
3513 * more packets, plus possibly installing the flow in the dpif.
3515 * So far we only batch the operations that affect flow setup time the most.
3516 * It's possible to batch more than that, but the benefit might be minimal. */
3518 struct hmap_node hmap_node;
3519 struct ofproto_dpif *ofproto;
3521 enum odp_key_fitness key_fitness;
3522 const struct nlattr *key;
3524 struct initial_vals initial_vals;
3525 struct list packets;
3526 enum dpif_upcall_type upcall_type;
3529 struct flow_miss_op {
3530 struct dpif_op dpif_op;
3532 uint64_t slow_stub[128 / 8]; /* Buffer for compose_slow_path() */
3533 struct xlate_out xout;
3534 bool xout_garbage; /* 'xout' needs to be uninitialized? */
3537 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
3538 * OpenFlow controller as necessary according to their individual
3539 * configurations. */
3541 send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet,
3542 const struct flow *flow)
3544 struct ofputil_packet_in pin;
3546 pin.packet = packet->data;
3547 pin.packet_len = packet->size;
3548 pin.reason = OFPR_NO_MATCH;
3549 pin.controller_id = 0;
3554 pin.send_len = 0; /* not used for flow table misses */
3556 flow_get_metadata(flow, &pin.fmd);
3558 connmgr_send_packet_in(ofproto->up.connmgr, &pin);
3561 static enum slow_path_reason
3562 process_special(struct ofproto_dpif *ofproto, const struct flow *flow,
3563 const struct ofport_dpif *ofport, const struct ofpbuf *packet)
3567 } else if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) {
3569 cfm_process_heartbeat(ofport->cfm, packet);
3572 } else if (ofport->bfd && bfd_should_process_flow(flow)) {
3574 bfd_process_packet(ofport->bfd, flow, packet);
3577 } else if (ofport->bundle && ofport->bundle->lacp
3578 && flow->dl_type == htons(ETH_TYPE_LACP)) {
3580 lacp_process_packet(ofport->bundle->lacp, ofport, packet);
3583 } else if (ofproto->stp && stp_should_process_flow(flow)) {
3585 stp_process_packet(ofport, packet);
3593 static struct flow_miss *
3594 flow_miss_find(struct hmap *todo, const struct ofproto_dpif *ofproto,
3595 const struct flow *flow, uint32_t hash)
3597 struct flow_miss *miss;
3599 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
3600 if (miss->ofproto == ofproto && flow_equal(&miss->flow, flow)) {
3608 /* Partially Initializes 'op' as an "execute" operation for 'miss' and
3609 * 'packet'. The caller must initialize op->actions and op->actions_len. If
3610 * 'miss' is associated with a subfacet the caller must also initialize the
3611 * returned op->subfacet, and if anything needs to be freed after processing
3612 * the op, the caller must initialize op->garbage also. */
3614 init_flow_miss_execute_op(struct flow_miss *miss, struct ofpbuf *packet,
3615 struct flow_miss_op *op)
3617 if (miss->flow.vlan_tci != miss->initial_vals.vlan_tci) {
3618 /* This packet was received on a VLAN splinter port. We
3619 * added a VLAN to the packet to make the packet resemble
3620 * the flow, but the actions were composed assuming that
3621 * the packet contained no VLAN. So, we must remove the
3622 * VLAN header from the packet before trying to execute the
3624 eth_pop_vlan(packet);
3627 op->xout_garbage = false;
3628 op->dpif_op.type = DPIF_OP_EXECUTE;
3629 op->dpif_op.u.execute.key = miss->key;
3630 op->dpif_op.u.execute.key_len = miss->key_len;
3631 op->dpif_op.u.execute.packet = packet;
3634 /* Helper for handle_flow_miss_without_facet() and
3635 * handle_flow_miss_with_facet(). */
3637 handle_flow_miss_common(struct rule_dpif *rule,
3638 struct ofpbuf *packet, const struct flow *flow)
3640 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3642 if (rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
3644 * Extra-special case for fail-open mode.
3646 * We are in fail-open mode and the packet matched the fail-open
3647 * rule, but we are connected to a controller too. We should send
3648 * the packet up to the controller in the hope that it will try to
3649 * set up a flow and thereby allow us to exit fail-open.
3651 * See the top-level comment in fail-open.c for more information.
3653 send_packet_in_miss(ofproto, packet, flow);
3657 /* Figures out whether a flow that missed in 'ofproto', whose details are in
3658 * 'miss', is likely to be worth tracking in detail in userspace and (usually)
3659 * installing a datapath flow. The answer is usually "yes" (a return value of
3660 * true). However, for short flows the cost of bookkeeping is much higher than
3661 * the benefits, so when the datapath holds a large number of flows we impose
3662 * some heuristics to decide which flows are likely to be worth tracking. */
3664 flow_miss_should_make_facet(struct ofproto_dpif *ofproto,
3665 struct flow_miss *miss, uint32_t hash)
3667 if (!ofproto->governor) {
3670 n_subfacets = hmap_count(&ofproto->subfacets);
3671 if (n_subfacets * 2 <= ofproto->up.flow_eviction_threshold) {
3675 ofproto->governor = governor_create(ofproto->up.name);
3678 return governor_should_install_flow(ofproto->governor, hash,
3679 list_size(&miss->packets));
3682 /* Handles 'miss', which matches 'rule', without creating a facet or subfacet
3683 * or creating any datapath flow. May add an "execute" operation to 'ops' and
3684 * increment '*n_ops'. */
3686 handle_flow_miss_without_facet(struct flow_miss *miss,
3687 struct flow_miss_op *ops, size_t *n_ops)
3689 struct rule_dpif *rule = rule_dpif_lookup(miss->ofproto, &miss->flow);
3690 long long int now = time_msec();
3691 struct ofpbuf *packet;
3692 struct xlate_in xin;
3694 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3695 struct flow_miss_op *op = &ops[*n_ops];
3696 struct dpif_flow_stats stats;
3698 COVERAGE_INC(facet_suppress);
3700 handle_flow_miss_common(rule, packet, &miss->flow);
3702 dpif_flow_stats_extract(&miss->flow, packet, now, &stats);
3703 rule_credit_stats(rule, &stats);
3705 xlate_in_init(&xin, miss->ofproto, &miss->flow, &miss->initial_vals,
3706 rule, stats.tcp_flags, packet);
3707 xin.resubmit_stats = &stats;
3708 xlate_actions(&xin, &op->xout);
3710 if (op->xout.odp_actions.size) {
3711 struct dpif_execute *execute = &op->dpif_op.u.execute;
3713 init_flow_miss_execute_op(miss, packet, op);
3714 execute->actions = op->xout.odp_actions.data;
3715 execute->actions_len = op->xout.odp_actions.size;
3716 op->xout_garbage = true;
3720 xlate_out_uninit(&op->xout);
3725 /* Handles 'miss', which matches 'facet'. May add any required datapath
3726 * operations to 'ops', incrementing '*n_ops' for each new op.
3728 * All of the packets in 'miss' are considered to have arrived at time 'now'.
3729 * This is really important only for new facets: if we just called time_msec()
3730 * here, then the new subfacet or its packets could look (occasionally) as
3731 * though it was used some time after the facet was used. That can make a
3732 * one-packet flow look like it has a nonzero duration, which looks odd in
3733 * e.g. NetFlow statistics. */
3735 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
3737 struct flow_miss_op *ops, size_t *n_ops)
3739 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3740 enum subfacet_path want_path;
3741 struct subfacet *subfacet;
3742 struct ofpbuf *packet;
3744 subfacet = subfacet_create(facet, miss, now);
3745 want_path = subfacet->facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
3747 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3748 struct flow_miss_op *op = &ops[*n_ops];
3749 struct dpif_flow_stats stats;
3751 handle_flow_miss_common(facet->rule, packet, &miss->flow);
3753 if (want_path != SF_FAST_PATH) {
3754 struct xlate_in xin;
3756 xlate_in_init(&xin, ofproto, &facet->flow, &facet->initial_vals,
3757 facet->rule, 0, packet);
3758 xlate_actions_for_side_effects(&xin);
3761 dpif_flow_stats_extract(&facet->flow, packet, now, &stats);
3762 subfacet_update_stats(subfacet, &stats);
3764 if (facet->xout.odp_actions.size) {
3765 struct dpif_execute *execute = &op->dpif_op.u.execute;
3767 init_flow_miss_execute_op(miss, packet, op);
3768 execute->actions = facet->xout.odp_actions.data,
3769 execute->actions_len = facet->xout.odp_actions.size;
3774 if (miss->upcall_type == DPIF_UC_MISS || subfacet->path != want_path) {
3775 struct flow_miss_op *op = &ops[(*n_ops)++];
3776 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
3778 subfacet->path = want_path;
3780 op->xout_garbage = false;
3781 op->dpif_op.type = DPIF_OP_FLOW_PUT;
3782 put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
3783 put->key = miss->key;
3784 put->key_len = miss->key_len;
3785 if (want_path == SF_FAST_PATH) {
3786 put->actions = facet->xout.odp_actions.data;
3787 put->actions_len = facet->xout.odp_actions.size;
3789 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
3790 op->slow_stub, sizeof op->slow_stub,
3791 &put->actions, &put->actions_len);
3797 /* Handles flow miss 'miss'. May add any required datapath operations
3798 * to 'ops', incrementing '*n_ops' for each new op. */
3800 handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops,
3803 struct ofproto_dpif *ofproto = miss->ofproto;
3804 struct facet *facet;
3808 /* The caller must ensure that miss->hmap_node.hash contains
3809 * flow_hash(miss->flow, 0). */
3810 hash = miss->hmap_node.hash;
3812 facet = facet_lookup_valid(ofproto, &miss->flow, hash);
3814 /* There does not exist a bijection between 'struct flow' and datapath
3815 * flow keys with fitness ODP_FIT_TO_LITTLE. This breaks a fundamental
3816 * assumption used throughout the facet and subfacet handling code.
3817 * Since we have to handle these misses in userspace anyway, we simply
3818 * skip facet creation, avoiding the problem alltogether. */
3819 if (miss->key_fitness == ODP_FIT_TOO_LITTLE
3820 || !flow_miss_should_make_facet(ofproto, miss, hash)) {
3821 handle_flow_miss_without_facet(miss, ops, n_ops);
3825 facet = facet_create(miss, hash);
3830 handle_flow_miss_with_facet(miss, facet, now, ops, n_ops);
3833 static struct drop_key *
3834 drop_key_lookup(const struct dpif_backer *backer, const struct nlattr *key,
3837 struct drop_key *drop_key;
3839 HMAP_FOR_EACH_WITH_HASH (drop_key, hmap_node, hash_bytes(key, key_len, 0),
3840 &backer->drop_keys) {
3841 if (drop_key->key_len == key_len
3842 && !memcmp(drop_key->key, key, key_len)) {
3850 drop_key_clear(struct dpif_backer *backer)
3852 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3853 struct drop_key *drop_key, *next;
3855 HMAP_FOR_EACH_SAFE (drop_key, next, hmap_node, &backer->drop_keys) {
3858 error = dpif_flow_del(backer->dpif, drop_key->key, drop_key->key_len,
3860 if (error && !VLOG_DROP_WARN(&rl)) {
3861 struct ds ds = DS_EMPTY_INITIALIZER;
3862 odp_flow_key_format(drop_key->key, drop_key->key_len, &ds);
3863 VLOG_WARN("Failed to delete drop key (%s) (%s)", strerror(error),
3868 hmap_remove(&backer->drop_keys, &drop_key->hmap_node);
3869 free(drop_key->key);
3874 /* Given a datpath, packet, and flow metadata ('backer', 'packet', and 'key'
3875 * respectively), populates 'flow' with the result of odp_flow_key_to_flow().
3876 * Optionally, if nonnull, populates 'fitnessp' with the fitness of 'flow' as
3877 * returned by odp_flow_key_to_flow(). Also, optionally populates 'ofproto'
3878 * with the ofproto_dpif, and 'odp_in_port' with the datapath in_port, that
3879 * 'packet' ingressed.
3881 * If 'ofproto' is nonnull, requires 'flow''s in_port to exist. Otherwise sets
3882 * 'flow''s in_port to OFPP_NONE.
3884 * This function does post-processing on data returned from
3885 * odp_flow_key_to_flow() to help make VLAN splinters transparent to the rest
3886 * of the upcall processing logic. In particular, if the extracted in_port is
3887 * a VLAN splinter port, it replaces flow->in_port by the "real" port, sets
3888 * flow->vlan_tci correctly for the VLAN of the VLAN splinter port, and pushes
3889 * a VLAN header onto 'packet' (if it is nonnull).
3891 * Optionally, if 'initial_vals' is nonnull, sets 'initial_vals->vlan_tci'
3892 * to the VLAN TCI with which the packet was really received, that is, the
3893 * actual VLAN TCI extracted by odp_flow_key_to_flow(). (This differs from
3894 * the value returned in flow->vlan_tci only for packets received on
3897 * Similarly, this function also includes some logic to help with tunnels. It
3898 * may modify 'flow' as necessary to make the tunneling implementation
3899 * transparent to the upcall processing logic.
3901 * Returns 0 if successful, ENODEV if the parsed flow has no associated ofport,
3902 * or some other positive errno if there are other problems. */
3904 ofproto_receive(const struct dpif_backer *backer, struct ofpbuf *packet,
3905 const struct nlattr *key, size_t key_len,
3906 struct flow *flow, enum odp_key_fitness *fitnessp,
3907 struct ofproto_dpif **ofproto, uint32_t *odp_in_port,
3908 struct initial_vals *initial_vals)
3910 const struct ofport_dpif *port;
3911 enum odp_key_fitness fitness;
3914 fitness = odp_flow_key_to_flow(key, key_len, flow);
3915 if (fitness == ODP_FIT_ERROR) {
3921 initial_vals->vlan_tci = flow->vlan_tci;
3925 *odp_in_port = flow->in_port;
3928 port = (tnl_port_should_receive(flow)
3929 ? ofport_dpif_cast(tnl_port_receive(flow))
3930 : odp_port_to_ofport(backer, flow->in_port));
3931 flow->in_port = port ? port->up.ofp_port : OFPP_NONE;
3936 /* XXX: Since the tunnel module is not scoped per backer, for a tunnel port
3937 * it's theoretically possible that we'll receive an ofport belonging to an
3938 * entirely different datapath. In practice, this can't happen because no
3939 * platforms has two separate datapaths which each support tunneling. */
3940 ovs_assert(ofproto_dpif_cast(port->up.ofproto)->backer == backer);
3942 if (vsp_adjust_flow(ofproto_dpif_cast(port->up.ofproto), flow)) {
3944 /* Make the packet resemble the flow, so that it gets sent to
3945 * an OpenFlow controller properly, so that it looks correct
3946 * for sFlow, and so that flow_extract() will get the correct
3947 * vlan_tci if it is called on 'packet'.
3949 * The allocated space inside 'packet' probably also contains
3950 * 'key', that is, both 'packet' and 'key' are probably part of
3951 * a struct dpif_upcall (see the large comment on that
3952 * structure definition), so pushing data on 'packet' is in
3953 * general not a good idea since it could overwrite 'key' or
3954 * free it as a side effect. However, it's OK in this special
3955 * case because we know that 'packet' is inside a Netlink
3956 * attribute: pushing 4 bytes will just overwrite the 4-byte
3957 * "struct nlattr", which is fine since we don't need that
3958 * header anymore. */
3959 eth_push_vlan(packet, flow->vlan_tci);
3961 /* We can't reproduce 'key' from 'flow'. */
3962 fitness = fitness == ODP_FIT_PERFECT ? ODP_FIT_TOO_MUCH : fitness;
3967 *ofproto = ofproto_dpif_cast(port->up.ofproto);
3972 *fitnessp = fitness;
3978 handle_miss_upcalls(struct dpif_backer *backer, struct dpif_upcall *upcalls,
3981 struct dpif_upcall *upcall;
3982 struct flow_miss *miss;
3983 struct flow_miss misses[FLOW_MISS_MAX_BATCH];
3984 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
3985 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
3995 /* Construct the to-do list.
3997 * This just amounts to extracting the flow from each packet and sticking
3998 * the packets that have the same flow in the same "flow_miss" structure so
3999 * that we can process them together. */
4002 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
4003 struct flow_miss *miss = &misses[n_misses];
4004 struct flow_miss *existing_miss;
4005 struct ofproto_dpif *ofproto;
4006 uint32_t odp_in_port;
4011 error = ofproto_receive(backer, upcall->packet, upcall->key,
4012 upcall->key_len, &flow, &miss->key_fitness,
4013 &ofproto, &odp_in_port, &miss->initial_vals);
4014 if (error == ENODEV) {
4015 struct drop_key *drop_key;
4017 /* Received packet on datapath port for which we couldn't
4018 * associate an ofproto. This can happen if a port is removed
4019 * while traffic is being received. Print a rate-limited message
4020 * in case it happens frequently. Install a drop flow so
4021 * that future packets of the flow are inexpensively dropped
4023 VLOG_INFO_RL(&rl, "received packet on unassociated datapath port "
4024 "%"PRIu32, odp_in_port);
4026 drop_key = drop_key_lookup(backer, upcall->key, upcall->key_len);
4028 drop_key = xmalloc(sizeof *drop_key);
4029 drop_key->key = xmemdup(upcall->key, upcall->key_len);
4030 drop_key->key_len = upcall->key_len;
4032 hmap_insert(&backer->drop_keys, &drop_key->hmap_node,
4033 hash_bytes(drop_key->key, drop_key->key_len, 0));
4034 dpif_flow_put(backer->dpif, DPIF_FP_CREATE | DPIF_FP_MODIFY,
4035 drop_key->key, drop_key->key_len, NULL, 0, NULL);
4043 ofproto->n_missed++;
4044 flow_extract(upcall->packet, flow.skb_priority, flow.skb_mark,
4045 &flow.tunnel, flow.in_port, &miss->flow);
4047 /* Add other packets to a to-do list. */
4048 hash = flow_hash(&miss->flow, 0);
4049 existing_miss = flow_miss_find(&todo, ofproto, &miss->flow, hash);
4050 if (!existing_miss) {
4051 hmap_insert(&todo, &miss->hmap_node, hash);
4052 miss->ofproto = ofproto;
4053 miss->key = upcall->key;
4054 miss->key_len = upcall->key_len;
4055 miss->upcall_type = upcall->type;
4056 list_init(&miss->packets);
4060 miss = existing_miss;
4062 list_push_back(&miss->packets, &upcall->packet->list_node);
4065 /* Process each element in the to-do list, constructing the set of
4066 * operations to batch. */
4068 HMAP_FOR_EACH (miss, hmap_node, &todo) {
4069 handle_flow_miss(miss, flow_miss_ops, &n_ops);
4071 ovs_assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
4073 /* Execute batch. */
4074 for (i = 0; i < n_ops; i++) {
4075 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
4077 dpif_operate(backer->dpif, dpif_ops, n_ops);
4080 for (i = 0; i < n_ops; i++) {
4081 if (flow_miss_ops[i].xout_garbage) {
4082 xlate_out_uninit(&flow_miss_ops[i].xout);
4085 hmap_destroy(&todo);
4088 static enum { SFLOW_UPCALL, MISS_UPCALL, BAD_UPCALL, FLOW_SAMPLE_UPCALL,
4090 classify_upcall(const struct dpif_upcall *upcall)
4092 size_t userdata_len;
4093 union user_action_cookie cookie;
4095 /* First look at the upcall type. */
4096 switch (upcall->type) {
4097 case DPIF_UC_ACTION:
4103 case DPIF_N_UC_TYPES:
4105 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
4109 /* "action" upcalls need a closer look. */
4110 if (!upcall->userdata) {
4111 VLOG_WARN_RL(&rl, "action upcall missing cookie");
4114 userdata_len = nl_attr_get_size(upcall->userdata);
4115 if (userdata_len < sizeof cookie.type
4116 || userdata_len > sizeof cookie) {
4117 VLOG_WARN_RL(&rl, "action upcall cookie has unexpected size %zu",
4121 memset(&cookie, 0, sizeof cookie);
4122 memcpy(&cookie, nl_attr_get(upcall->userdata), userdata_len);
4123 if (userdata_len == sizeof cookie.sflow
4124 && cookie.type == USER_ACTION_COOKIE_SFLOW) {
4125 return SFLOW_UPCALL;
4126 } else if (userdata_len == sizeof cookie.slow_path
4127 && cookie.type == USER_ACTION_COOKIE_SLOW_PATH) {
4129 } else if (userdata_len == sizeof cookie.flow_sample
4130 && cookie.type == USER_ACTION_COOKIE_FLOW_SAMPLE) {
4131 return FLOW_SAMPLE_UPCALL;
4132 } else if (userdata_len == sizeof cookie.ipfix
4133 && cookie.type == USER_ACTION_COOKIE_IPFIX) {
4134 return IPFIX_UPCALL;
4136 VLOG_WARN_RL(&rl, "invalid user cookie of type %"PRIu16
4137 " and size %zu", cookie.type, userdata_len);
4143 handle_sflow_upcall(struct dpif_backer *backer,
4144 const struct dpif_upcall *upcall)
4146 struct ofproto_dpif *ofproto;
4147 union user_action_cookie cookie;
4149 uint32_t odp_in_port;
4151 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
4152 &flow, NULL, &ofproto, &odp_in_port, NULL)
4153 || !ofproto->sflow) {
4157 memset(&cookie, 0, sizeof cookie);
4158 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.sflow);
4159 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
4160 odp_in_port, &cookie);
4164 handle_flow_sample_upcall(struct dpif_backer *backer,
4165 const struct dpif_upcall *upcall)
4167 struct ofproto_dpif *ofproto;
4168 union user_action_cookie cookie;
4171 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
4172 &flow, NULL, &ofproto, NULL, NULL)
4173 || !ofproto->ipfix) {
4177 memset(&cookie, 0, sizeof cookie);
4178 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.flow_sample);
4180 /* The flow reflects exactly the contents of the packet. Sample
4181 * the packet using it. */
4182 dpif_ipfix_flow_sample(ofproto->ipfix, upcall->packet, &flow,
4183 cookie.flow_sample.collector_set_id,
4184 cookie.flow_sample.probability,
4185 cookie.flow_sample.obs_domain_id,
4186 cookie.flow_sample.obs_point_id);
4190 handle_ipfix_upcall(struct dpif_backer *backer,
4191 const struct dpif_upcall *upcall)
4193 struct ofproto_dpif *ofproto;
4196 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
4197 &flow, NULL, &ofproto, NULL, NULL)
4198 || !ofproto->ipfix) {
4202 /* The flow reflects exactly the contents of the packet. Sample
4203 * the packet using it. */
4204 dpif_ipfix_bridge_sample(ofproto->ipfix, upcall->packet, &flow);
4208 handle_upcalls(struct dpif_backer *backer, unsigned int max_batch)
4210 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
4211 struct ofpbuf miss_bufs[FLOW_MISS_MAX_BATCH];
4212 uint64_t miss_buf_stubs[FLOW_MISS_MAX_BATCH][4096 / 8];
4217 ovs_assert(max_batch <= FLOW_MISS_MAX_BATCH);
4220 for (n_processed = 0; n_processed < max_batch; n_processed++) {
4221 struct dpif_upcall *upcall = &misses[n_misses];
4222 struct ofpbuf *buf = &miss_bufs[n_misses];
4225 ofpbuf_use_stub(buf, miss_buf_stubs[n_misses],
4226 sizeof miss_buf_stubs[n_misses]);
4227 error = dpif_recv(backer->dpif, upcall, buf);
4233 switch (classify_upcall(upcall)) {
4235 /* Handle it later. */
4240 handle_sflow_upcall(backer, upcall);
4244 case FLOW_SAMPLE_UPCALL:
4245 handle_flow_sample_upcall(backer, upcall);
4250 handle_ipfix_upcall(backer, upcall);
4260 /* Handle deferred MISS_UPCALL processing. */
4261 handle_miss_upcalls(backer, misses, n_misses);
4262 for (i = 0; i < n_misses; i++) {
4263 ofpbuf_uninit(&miss_bufs[i]);
4269 /* Flow expiration. */
4271 static int subfacet_max_idle(const struct ofproto_dpif *);
4272 static void update_stats(struct dpif_backer *);
4273 static void rule_expire(struct rule_dpif *);
4274 static void expire_subfacets(struct ofproto_dpif *, int dp_max_idle);
4276 /* This function is called periodically by run(). Its job is to collect
4277 * updates for the flows that have been installed into the datapath, most
4278 * importantly when they last were used, and then use that information to
4279 * expire flows that have not been used recently.
4281 * Returns the number of milliseconds after which it should be called again. */
4283 expire(struct dpif_backer *backer)
4285 struct ofproto_dpif *ofproto;
4286 int max_idle = INT32_MAX;
4288 /* Periodically clear out the drop keys in an effort to keep them
4289 * relatively few. */
4290 drop_key_clear(backer);
4292 /* Update stats for each flow in the backer. */
4293 update_stats(backer);
4295 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4296 struct rule *rule, *next_rule;
4299 if (ofproto->backer != backer) {
4303 /* Keep track of the max number of flows per ofproto_dpif. */
4304 update_max_subfacet_count(ofproto);
4306 /* Expire subfacets that have been idle too long. */
4307 dp_max_idle = subfacet_max_idle(ofproto);
4308 expire_subfacets(ofproto, dp_max_idle);
4310 max_idle = MIN(max_idle, dp_max_idle);
4312 /* Expire OpenFlow flows whose idle_timeout or hard_timeout
4314 LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
4315 &ofproto->up.expirable) {
4316 rule_expire(rule_dpif_cast(rule));
4319 /* All outstanding data in existing flows has been accounted, so it's a
4320 * good time to do bond rebalancing. */
4321 if (ofproto->has_bonded_bundles) {
4322 struct ofbundle *bundle;
4324 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
4326 bond_rebalance(bundle->bond, &backer->revalidate_set);
4332 return MIN(max_idle, 1000);
4335 /* Updates flow table statistics given that the datapath just reported 'stats'
4336 * as 'subfacet''s statistics. */
4338 update_subfacet_stats(struct subfacet *subfacet,
4339 const struct dpif_flow_stats *stats)
4341 struct facet *facet = subfacet->facet;
4342 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4343 struct dpif_flow_stats diff;
4345 diff.tcp_flags = stats->tcp_flags;
4346 diff.used = stats->used;
4348 if (stats->n_packets >= subfacet->dp_packet_count) {
4349 diff.n_packets = stats->n_packets - subfacet->dp_packet_count;
4351 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
4355 if (stats->n_bytes >= subfacet->dp_byte_count) {
4356 diff.n_bytes = stats->n_bytes - subfacet->dp_byte_count;
4358 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
4362 ofproto->n_hit += diff.n_packets;
4363 subfacet->dp_packet_count = stats->n_packets;
4364 subfacet->dp_byte_count = stats->n_bytes;
4365 subfacet_update_stats(subfacet, &diff);
4367 if (facet->accounted_bytes < facet->byte_count) {
4369 facet_account(facet);
4370 facet->accounted_bytes = facet->byte_count;
4374 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
4375 * about, or a flow that shouldn't be installed but was anyway. Delete it. */
4377 delete_unexpected_flow(struct ofproto_dpif *ofproto,
4378 const struct nlattr *key, size_t key_len)
4380 if (!VLOG_DROP_WARN(&rl)) {
4384 odp_flow_key_format(key, key_len, &s);
4385 VLOG_WARN("unexpected flow on %s: %s", ofproto->up.name, ds_cstr(&s));
4389 COVERAGE_INC(facet_unexpected);
4390 dpif_flow_del(ofproto->backer->dpif, key, key_len, NULL);
4393 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
4395 * This function also pushes statistics updates to rules which each facet
4396 * resubmits into. Generally these statistics will be accurate. However, if a
4397 * facet changes the rule it resubmits into at some time in between
4398 * update_stats() runs, it is possible that statistics accrued to the
4399 * old rule will be incorrectly attributed to the new rule. This could be
4400 * avoided by calling update_stats() whenever rules are created or
4401 * deleted. However, the performance impact of making so many calls to the
4402 * datapath do not justify the benefit of having perfectly accurate statistics.
4404 * In addition, this function maintains per ofproto flow hit counts. The patch
4405 * port is not treated specially. e.g. A packet ingress from br0 patched into
4406 * br1 will increase the hit count of br0 by 1, however, does not affect
4407 * the hit or miss counts of br1.
4410 update_stats(struct dpif_backer *backer)
4412 const struct dpif_flow_stats *stats;
4413 struct dpif_flow_dump dump;
4414 const struct nlattr *key;
4415 struct ofproto_dpif *ofproto;
4418 dpif_flow_dump_start(&dump, backer->dpif);
4419 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
4421 struct subfacet *subfacet;
4424 if (ofproto_receive(backer, NULL, key, key_len, &flow, NULL, &ofproto,
4429 ofproto->total_subfacet_count += hmap_count(&ofproto->subfacets);
4430 ofproto->n_update_stats++;
4432 key_hash = odp_flow_key_hash(key, key_len);
4433 subfacet = subfacet_find(ofproto, key, key_len, key_hash);
4434 switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
4436 update_subfacet_stats(subfacet, stats);
4440 /* Stats are updated per-packet. */
4443 case SF_NOT_INSTALLED:
4445 delete_unexpected_flow(ofproto, key, key_len);
4450 dpif_flow_dump_done(&dump);
4452 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4453 update_moving_averages(ofproto);
4458 /* Calculates and returns the number of milliseconds of idle time after which
4459 * subfacets should expire from the datapath. When a subfacet expires, we fold
4460 * its statistics into its facet, and when a facet's last subfacet expires, we
4461 * fold its statistic into its rule. */
4463 subfacet_max_idle(const struct ofproto_dpif *ofproto)
4466 * Idle time histogram.
4468 * Most of the time a switch has a relatively small number of subfacets.
4469 * When this is the case we might as well keep statistics for all of them
4470 * in userspace and to cache them in the kernel datapath for performance as
4473 * As the number of subfacets increases, the memory required to maintain
4474 * statistics about them in userspace and in the kernel becomes
4475 * significant. However, with a large number of subfacets it is likely
4476 * that only a few of them are "heavy hitters" that consume a large amount
4477 * of bandwidth. At this point, only heavy hitters are worth caching in
4478 * the kernel and maintaining in userspaces; other subfacets we can
4481 * The technique used to compute the idle time is to build a histogram with
4482 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
4483 * that is installed in the kernel gets dropped in the appropriate bucket.
4484 * After the histogram has been built, we compute the cutoff so that only
4485 * the most-recently-used 1% of subfacets (but at least
4486 * ofproto->up.flow_eviction_threshold flows) are kept cached. At least
4487 * the most-recently-used bucket of subfacets is kept, so actually an
4488 * arbitrary number of subfacets can be kept in any given expiration run
4489 * (though the next run will delete most of those unless they receive
4492 * This requires a second pass through the subfacets, in addition to the
4493 * pass made by update_stats(), because the former function never looks at
4494 * uninstallable subfacets.
4496 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
4497 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
4498 int buckets[N_BUCKETS] = { 0 };
4499 int total, subtotal, bucket;
4500 struct subfacet *subfacet;
4504 total = hmap_count(&ofproto->subfacets);
4505 if (total <= ofproto->up.flow_eviction_threshold) {
4506 return N_BUCKETS * BUCKET_WIDTH;
4509 /* Build histogram. */
4511 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
4512 long long int idle = now - subfacet->used;
4513 int bucket = (idle <= 0 ? 0
4514 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
4515 : (unsigned int) idle / BUCKET_WIDTH);
4519 /* Find the first bucket whose flows should be expired. */
4520 subtotal = bucket = 0;
4522 subtotal += buckets[bucket++];
4523 } while (bucket < N_BUCKETS &&
4524 subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100));
4526 if (VLOG_IS_DBG_ENABLED()) {
4530 ds_put_cstr(&s, "keep");
4531 for (i = 0; i < N_BUCKETS; i++) {
4533 ds_put_cstr(&s, ", drop");
4536 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
4539 VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s));
4543 return bucket * BUCKET_WIDTH;
4547 expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle)
4549 /* Cutoff time for most flows. */
4550 long long int normal_cutoff = time_msec() - dp_max_idle;
4552 /* We really want to keep flows for special protocols around, so use a more
4553 * conservative cutoff. */
4554 long long int special_cutoff = time_msec() - 10000;
4556 struct subfacet *subfacet, *next_subfacet;
4557 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
4561 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
4562 &ofproto->subfacets) {
4563 long long int cutoff;
4565 cutoff = (subfacet->facet->xout.slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP
4569 if (subfacet->used < cutoff) {
4570 if (subfacet->path != SF_NOT_INSTALLED) {
4571 batch[n_batch++] = subfacet;
4572 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
4573 subfacet_destroy_batch(ofproto, batch, n_batch);
4577 subfacet_destroy(subfacet);
4583 subfacet_destroy_batch(ofproto, batch, n_batch);
4587 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
4588 * then delete it entirely. */
4590 rule_expire(struct rule_dpif *rule)
4592 struct facet *facet, *next_facet;
4596 if (rule->up.pending) {
4597 /* We'll have to expire it later. */
4601 /* Has 'rule' expired? */
4603 if (rule->up.hard_timeout
4604 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
4605 reason = OFPRR_HARD_TIMEOUT;
4606 } else if (rule->up.idle_timeout
4607 && now > rule->up.used + rule->up.idle_timeout * 1000) {
4608 reason = OFPRR_IDLE_TIMEOUT;
4613 COVERAGE_INC(ofproto_dpif_expired);
4615 /* Update stats. (This is a no-op if the rule expired due to an idle
4616 * timeout, because that only happens when the rule has no facets left.) */
4617 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4618 facet_remove(facet);
4621 /* Get rid of the rule. */
4622 ofproto_rule_expire(&rule->up, reason);
4627 /* Creates and returns a new facet based on 'miss'.
4629 * The caller must already have determined that no facet with an identical
4630 * 'miss->flow' exists in 'miss->ofproto'.
4632 * 'hash' must be the return value of flow_hash(miss->flow, 0).
4634 * The facet will initially have no subfacets. The caller should create (at
4635 * least) one subfacet with subfacet_create(). */
4636 static struct facet *
4637 facet_create(const struct flow_miss *miss, uint32_t hash)
4639 struct ofproto_dpif *ofproto = miss->ofproto;
4640 struct xlate_in xin;
4641 struct facet *facet;
4643 facet = xzalloc(sizeof *facet);
4644 facet->used = time_msec();
4645 facet->flow = miss->flow;
4646 facet->initial_vals = miss->initial_vals;
4647 facet->rule = rule_dpif_lookup(ofproto, &facet->flow);
4648 facet->learn_rl = time_msec() + 500;
4650 hmap_insert(&ofproto->facets, &facet->hmap_node, hash);
4651 list_push_back(&facet->rule->facets, &facet->list_node);
4652 list_init(&facet->subfacets);
4653 netflow_flow_init(&facet->nf_flow);
4654 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
4656 xlate_in_init(&xin, ofproto, &facet->flow, &facet->initial_vals,
4657 facet->rule, 0, NULL);
4658 xin.may_learn = true;
4659 xlate_actions(&xin, &facet->xout);
4660 facet->nf_flow.output_iface = facet->xout.nf_output_iface;
4666 facet_free(struct facet *facet)
4669 xlate_out_uninit(&facet->xout);
4674 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
4675 * 'packet', which arrived on 'in_port'. */
4677 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
4678 const struct nlattr *odp_actions, size_t actions_len,
4679 struct ofpbuf *packet)
4681 struct odputil_keybuf keybuf;
4685 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4686 odp_flow_key_from_flow(&key, flow,
4687 ofp_port_to_odp_port(ofproto, flow->in_port));
4689 error = dpif_execute(ofproto->backer->dpif, key.data, key.size,
4690 odp_actions, actions_len, packet);
4694 /* Remove 'facet' from 'ofproto' and free up the associated memory:
4696 * - If 'facet' was installed in the datapath, uninstalls it and updates its
4697 * rule's statistics, via subfacet_uninstall().
4699 * - Removes 'facet' from its rule and from ofproto->facets.
4702 facet_remove(struct facet *facet)
4704 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4705 struct subfacet *subfacet, *next_subfacet;
4707 ovs_assert(!list_is_empty(&facet->subfacets));
4709 /* First uninstall all of the subfacets to get final statistics. */
4710 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4711 subfacet_uninstall(subfacet);
4714 /* Flush the final stats to the rule.
4716 * This might require us to have at least one subfacet around so that we
4717 * can use its actions for accounting in facet_account(), which is why we
4718 * have uninstalled but not yet destroyed the subfacets. */
4719 facet_flush_stats(facet);
4721 /* Now we're really all done so destroy everything. */
4722 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
4723 &facet->subfacets) {
4724 subfacet_destroy__(subfacet);
4726 hmap_remove(&ofproto->facets, &facet->hmap_node);
4727 list_remove(&facet->list_node);
4731 /* Feed information from 'facet' back into the learning table to keep it in
4732 * sync with what is actually flowing through the datapath. */
4734 facet_learn(struct facet *facet)
4736 long long int now = time_msec();
4738 if (!facet->xout.has_fin_timeout && now < facet->learn_rl) {
4742 facet->learn_rl = now + 500;
4744 if (!facet->xout.has_learn
4745 && !facet->xout.has_normal
4746 && (!facet->xout.has_fin_timeout
4747 || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
4751 facet_push_stats(facet, true);
4755 facet_account(struct facet *facet)
4757 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4758 const struct nlattr *a;
4763 if (!facet->xout.has_normal || !ofproto->has_bonded_bundles) {
4766 n_bytes = facet->byte_count - facet->accounted_bytes;
4768 /* This loop feeds byte counters to bond_account() for rebalancing to use
4769 * as a basis. We also need to track the actual VLAN on which the packet
4770 * is going to be sent to ensure that it matches the one passed to
4771 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
4774 * We use the actions from an arbitrary subfacet because they should all
4775 * be equally valid for our purpose. */
4776 vlan_tci = facet->flow.vlan_tci;
4777 NL_ATTR_FOR_EACH_UNSAFE (a, left, facet->xout.odp_actions.data,
4778 facet->xout.odp_actions.size) {
4779 const struct ovs_action_push_vlan *vlan;
4780 struct ofport_dpif *port;
4782 switch (nl_attr_type(a)) {
4783 case OVS_ACTION_ATTR_OUTPUT:
4784 port = get_odp_port(ofproto, nl_attr_get_u32(a));
4785 if (port && port->bundle && port->bundle->bond) {
4786 bond_account(port->bundle->bond, &facet->flow,
4787 vlan_tci_to_vid(vlan_tci), n_bytes);
4791 case OVS_ACTION_ATTR_POP_VLAN:
4792 vlan_tci = htons(0);
4795 case OVS_ACTION_ATTR_PUSH_VLAN:
4796 vlan = nl_attr_get(a);
4797 vlan_tci = vlan->vlan_tci;
4803 /* Returns true if the only action for 'facet' is to send to the controller.
4804 * (We don't report NetFlow expiration messages for such facets because they
4805 * are just part of the control logic for the network, not real traffic). */
4807 facet_is_controller_flow(struct facet *facet)
4810 const struct rule *rule = &facet->rule->up;
4811 const struct ofpact *ofpacts = rule->ofpacts;
4812 size_t ofpacts_len = rule->ofpacts_len;
4814 if (ofpacts_len > 0 &&
4815 ofpacts->type == OFPACT_CONTROLLER &&
4816 ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len)) {
4823 /* Folds all of 'facet''s statistics into its rule. Also updates the
4824 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
4825 * 'facet''s statistics in the datapath should have been zeroed and folded into
4826 * its packet and byte counts before this function is called. */
4828 facet_flush_stats(struct facet *facet)
4830 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4831 struct subfacet *subfacet;
4833 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4834 ovs_assert(!subfacet->dp_byte_count);
4835 ovs_assert(!subfacet->dp_packet_count);
4838 facet_push_stats(facet, false);
4839 if (facet->accounted_bytes < facet->byte_count) {
4840 facet_account(facet);
4841 facet->accounted_bytes = facet->byte_count;
4844 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
4845 struct ofexpired expired;
4846 expired.flow = facet->flow;
4847 expired.packet_count = facet->packet_count;
4848 expired.byte_count = facet->byte_count;
4849 expired.used = facet->used;
4850 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4853 /* Reset counters to prevent double counting if 'facet' ever gets
4855 facet_reset_counters(facet);
4857 netflow_flow_clear(&facet->nf_flow);
4858 facet->tcp_flags = 0;
4861 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
4862 * Returns it if found, otherwise a null pointer.
4864 * 'hash' must be the return value of flow_hash(flow, 0).
4866 * The returned facet might need revalidation; use facet_lookup_valid()
4867 * instead if that is important. */
4868 static struct facet *
4869 facet_find(struct ofproto_dpif *ofproto,
4870 const struct flow *flow, uint32_t hash)
4872 struct facet *facet;
4874 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, hash, &ofproto->facets) {
4875 if (flow_equal(flow, &facet->flow)) {
4883 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
4884 * Returns it if found, otherwise a null pointer.
4886 * 'hash' must be the return value of flow_hash(flow, 0).
4888 * The returned facet is guaranteed to be valid. */
4889 static struct facet *
4890 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow,
4893 struct facet *facet;
4895 facet = facet_find(ofproto, flow, hash);
4897 && (ofproto->backer->need_revalidate
4898 || tag_set_intersects(&ofproto->backer->revalidate_set,
4900 && !facet_revalidate(facet)) {
4908 facet_check_consistency(struct facet *facet)
4910 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
4912 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4914 struct xlate_out xout;
4915 struct xlate_in xin;
4917 struct rule_dpif *rule;
4920 /* Check the rule for consistency. */
4921 rule = rule_dpif_lookup(ofproto, &facet->flow);
4922 if (rule != facet->rule) {
4923 if (!VLOG_DROP_WARN(&rl)) {
4924 struct ds s = DS_EMPTY_INITIALIZER;
4926 flow_format(&s, &facet->flow);
4927 ds_put_format(&s, ": facet associated with wrong rule (was "
4928 "table=%"PRIu8",", facet->rule->up.table_id);
4929 cls_rule_format(&facet->rule->up.cr, &s);
4930 ds_put_format(&s, ") (should have been table=%"PRIu8",",
4932 cls_rule_format(&rule->up.cr, &s);
4933 ds_put_cstr(&s, ")\n");
4940 /* Check the datapath actions for consistency. */
4941 xlate_in_init(&xin, ofproto, &facet->flow, &facet->initial_vals, rule,
4943 xlate_actions(&xin, &xout);
4945 ok = ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)
4946 && facet->xout.slow == xout.slow;
4947 if (!ok && !VLOG_DROP_WARN(&rl)) {
4948 struct ds s = DS_EMPTY_INITIALIZER;
4950 flow_format(&s, &facet->flow);
4951 ds_put_cstr(&s, ": inconsistency in facet");
4953 if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4954 ds_put_cstr(&s, " (actions were: ");
4955 format_odp_actions(&s, facet->xout.odp_actions.data,
4956 facet->xout.odp_actions.size);
4957 ds_put_cstr(&s, ") (correct actions: ");
4958 format_odp_actions(&s, xout.odp_actions.data,
4959 xout.odp_actions.size);
4960 ds_put_cstr(&s, ")");
4963 if (facet->xout.slow != xout.slow) {
4964 ds_put_format(&s, " slow path incorrect. should be %d", xout.slow);
4969 xlate_out_uninit(&xout);
4974 /* Re-searches the classifier for 'facet':
4976 * - If the rule found is different from 'facet''s current rule, moves
4977 * 'facet' to the new rule and recompiles its actions.
4979 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
4980 * where it is and recompiles its actions anyway.
4982 * - If any of 'facet''s subfacets correspond to a new flow according to
4983 * ofproto_receive(), 'facet' is removed.
4985 * Returns true if 'facet' is still valid. False if 'facet' was removed. */
4987 facet_revalidate(struct facet *facet)
4989 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4990 struct rule_dpif *new_rule;
4991 struct subfacet *subfacet;
4992 struct xlate_out xout;
4993 struct xlate_in xin;
4995 COVERAGE_INC(facet_revalidate);
4997 /* Check that child subfacets still correspond to this facet. Tunnel
4998 * configuration changes could cause a subfacet's OpenFlow in_port to
5000 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
5001 struct ofproto_dpif *recv_ofproto;
5002 struct flow recv_flow;
5005 error = ofproto_receive(ofproto->backer, NULL, subfacet->key,
5006 subfacet->key_len, &recv_flow, NULL,
5007 &recv_ofproto, NULL, NULL);
5009 || recv_ofproto != ofproto
5010 || memcmp(&recv_flow, &facet->flow, sizeof recv_flow)) {
5011 facet_remove(facet);
5016 new_rule = rule_dpif_lookup(ofproto, &facet->flow);
5018 /* Calculate new datapath actions.
5020 * We do not modify any 'facet' state yet, because we might need to, e.g.,
5021 * emit a NetFlow expiration and, if so, we need to have the old state
5022 * around to properly compose it. */
5023 xlate_in_init(&xin, ofproto, &facet->flow, &facet->initial_vals, new_rule,
5025 xlate_actions(&xin, &xout);
5027 /* A facet's slow path reason should only change under dramatic
5028 * circumstances. Rather than try to update everything, it's simpler to
5029 * remove the facet and start over. */
5030 if (facet->xout.slow != xout.slow) {
5031 facet_remove(facet);
5032 xlate_out_uninit(&xout);
5036 if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
5037 LIST_FOR_EACH(subfacet, list_node, &facet->subfacets) {
5038 if (subfacet->path == SF_FAST_PATH) {
5039 struct dpif_flow_stats stats;
5041 subfacet_install(subfacet, &xout.odp_actions, &stats);
5042 subfacet_update_stats(subfacet, &stats);
5046 facet_flush_stats(facet);
5048 ofpbuf_clear(&facet->xout.odp_actions);
5049 ofpbuf_put(&facet->xout.odp_actions, xout.odp_actions.data,
5050 xout.odp_actions.size);
5053 /* Update 'facet' now that we've taken care of all the old state. */
5054 facet->xout.tags = xout.tags;
5055 facet->xout.slow = xout.slow;
5056 facet->xout.has_learn = xout.has_learn;
5057 facet->xout.has_normal = xout.has_normal;
5058 facet->xout.has_fin_timeout = xout.has_fin_timeout;
5059 facet->xout.nf_output_iface = xout.nf_output_iface;
5060 facet->xout.mirrors = xout.mirrors;
5061 facet->nf_flow.output_iface = facet->xout.nf_output_iface;
5063 if (facet->rule != new_rule) {
5064 COVERAGE_INC(facet_changed_rule);
5065 list_remove(&facet->list_node);
5066 list_push_back(&new_rule->facets, &facet->list_node);
5067 facet->rule = new_rule;
5068 facet->used = new_rule->up.created;
5069 facet->prev_used = facet->used;
5072 xlate_out_uninit(&xout);
5077 facet_reset_counters(struct facet *facet)
5079 facet->packet_count = 0;
5080 facet->byte_count = 0;
5081 facet->prev_packet_count = 0;
5082 facet->prev_byte_count = 0;
5083 facet->accounted_bytes = 0;
5087 facet_push_stats(struct facet *facet, bool may_learn)
5089 struct dpif_flow_stats stats;
5091 ovs_assert(facet->packet_count >= facet->prev_packet_count);
5092 ovs_assert(facet->byte_count >= facet->prev_byte_count);
5093 ovs_assert(facet->used >= facet->prev_used);
5095 stats.n_packets = facet->packet_count - facet->prev_packet_count;
5096 stats.n_bytes = facet->byte_count - facet->prev_byte_count;
5097 stats.used = facet->used;
5098 stats.tcp_flags = facet->tcp_flags;
5100 if (may_learn || stats.n_packets || facet->used > facet->prev_used) {
5101 struct ofproto_dpif *ofproto =
5102 ofproto_dpif_cast(facet->rule->up.ofproto);
5104 struct ofport_dpif *in_port;
5105 struct xlate_in xin;
5107 facet->prev_packet_count = facet->packet_count;
5108 facet->prev_byte_count = facet->byte_count;
5109 facet->prev_used = facet->used;
5111 in_port = get_ofp_port(ofproto, facet->flow.in_port);
5112 if (in_port && in_port->tnl_port) {
5113 netdev_vport_inc_rx(in_port->up.netdev, &stats);
5116 rule_credit_stats(facet->rule, &stats);
5117 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow,
5119 netflow_flow_update_flags(&facet->nf_flow, facet->tcp_flags);
5120 update_mirror_stats(ofproto, facet->xout.mirrors, stats.n_packets,
5123 xlate_in_init(&xin, ofproto, &facet->flow, &facet->initial_vals,
5124 facet->rule, stats.tcp_flags, NULL);
5125 xin.resubmit_stats = &stats;
5126 xin.may_learn = may_learn;
5127 xlate_actions_for_side_effects(&xin);
5132 push_all_stats__(bool run_fast)
5134 static long long int rl = LLONG_MIN;
5135 struct ofproto_dpif *ofproto;
5137 if (time_msec() < rl) {
5141 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5142 struct facet *facet;
5144 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
5145 facet_push_stats(facet, false);
5152 rl = time_msec() + 100;
5156 push_all_stats(void)
5158 push_all_stats__(true);
5162 rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats)
5164 rule->packet_count += stats->n_packets;
5165 rule->byte_count += stats->n_bytes;
5166 ofproto_rule_update_used(&rule->up, stats->used);
5171 static struct subfacet *
5172 subfacet_find(struct ofproto_dpif *ofproto,
5173 const struct nlattr *key, size_t key_len, uint32_t key_hash)
5175 struct subfacet *subfacet;
5177 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
5178 &ofproto->subfacets) {
5179 if (subfacet->key_len == key_len
5180 && !memcmp(key, subfacet->key, key_len)) {
5188 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
5189 * 'key_fitness', 'key', and 'key_len' members in 'miss'. Returns the
5190 * existing subfacet if there is one, otherwise creates and returns a
5192 static struct subfacet *
5193 subfacet_create(struct facet *facet, struct flow_miss *miss,
5196 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5197 enum odp_key_fitness key_fitness = miss->key_fitness;
5198 const struct nlattr *key = miss->key;
5199 size_t key_len = miss->key_len;
5201 struct subfacet *subfacet;
5203 key_hash = odp_flow_key_hash(key, key_len);
5205 if (list_is_empty(&facet->subfacets)) {
5206 subfacet = &facet->one_subfacet;
5208 subfacet = subfacet_find(ofproto, key, key_len, key_hash);
5210 if (subfacet->facet == facet) {
5214 /* This shouldn't happen. */
5215 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
5216 subfacet_destroy(subfacet);
5219 subfacet = xmalloc(sizeof *subfacet);
5222 hmap_insert(&ofproto->subfacets, &subfacet->hmap_node, key_hash);
5223 list_push_back(&facet->subfacets, &subfacet->list_node);
5224 subfacet->facet = facet;
5225 subfacet->key_fitness = key_fitness;
5226 subfacet->key = xmemdup(key, key_len);
5227 subfacet->key_len = key_len;
5228 subfacet->used = now;
5229 subfacet->created = now;
5230 subfacet->dp_packet_count = 0;
5231 subfacet->dp_byte_count = 0;
5232 subfacet->path = SF_NOT_INSTALLED;
5234 ofproto->subfacet_add_count++;
5238 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
5239 * its facet within 'ofproto', and frees it. */
5241 subfacet_destroy__(struct subfacet *subfacet)
5243 struct facet *facet = subfacet->facet;
5244 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5246 /* Update ofproto stats before uninstall the subfacet. */
5247 ofproto->subfacet_del_count++;
5248 ofproto->total_subfacet_life_span += (time_msec() - subfacet->created);
5250 subfacet_uninstall(subfacet);
5251 hmap_remove(&ofproto->subfacets, &subfacet->hmap_node);
5252 list_remove(&subfacet->list_node);
5253 free(subfacet->key);
5254 if (subfacet != &facet->one_subfacet) {
5259 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
5260 * last remaining subfacet in its facet destroys the facet too. */
5262 subfacet_destroy(struct subfacet *subfacet)
5264 struct facet *facet = subfacet->facet;
5266 if (list_is_singleton(&facet->subfacets)) {
5267 /* facet_remove() needs at least one subfacet (it will remove it). */
5268 facet_remove(facet);
5270 subfacet_destroy__(subfacet);
5275 subfacet_destroy_batch(struct ofproto_dpif *ofproto,
5276 struct subfacet **subfacets, int n)
5278 struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH];
5279 struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH];
5280 struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH];
5283 for (i = 0; i < n; i++) {
5284 ops[i].type = DPIF_OP_FLOW_DEL;
5285 ops[i].u.flow_del.key = subfacets[i]->key;
5286 ops[i].u.flow_del.key_len = subfacets[i]->key_len;
5287 ops[i].u.flow_del.stats = &stats[i];
5291 dpif_operate(ofproto->backer->dpif, opsp, n);
5292 for (i = 0; i < n; i++) {
5293 subfacet_reset_dp_stats(subfacets[i], &stats[i]);
5294 subfacets[i]->path = SF_NOT_INSTALLED;
5295 subfacet_destroy(subfacets[i]);
5300 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
5301 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
5302 * in the datapath will be zeroed and 'stats' will be updated with traffic new
5303 * since 'subfacet' was last updated.
5305 * Returns 0 if successful, otherwise a positive errno value. */
5307 subfacet_install(struct subfacet *subfacet, const struct ofpbuf *odp_actions,
5308 struct dpif_flow_stats *stats)
5310 struct facet *facet = subfacet->facet;
5311 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5312 enum subfacet_path path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
5313 const struct nlattr *actions = odp_actions->data;
5314 size_t actions_len = odp_actions->size;
5316 uint64_t slow_path_stub[128 / 8];
5317 enum dpif_flow_put_flags flags;
5320 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
5322 flags |= DPIF_FP_ZERO_STATS;
5325 if (path == SF_SLOW_PATH) {
5326 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
5327 slow_path_stub, sizeof slow_path_stub,
5328 &actions, &actions_len);
5331 ret = dpif_flow_put(ofproto->backer->dpif, flags, subfacet->key,
5332 subfacet->key_len, actions, actions_len, stats);
5335 subfacet_reset_dp_stats(subfacet, stats);
5339 subfacet->path = path;
5344 /* If 'subfacet' is installed in the datapath, uninstalls it. */
5346 subfacet_uninstall(struct subfacet *subfacet)
5348 if (subfacet->path != SF_NOT_INSTALLED) {
5349 struct rule_dpif *rule = subfacet->facet->rule;
5350 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5351 struct dpif_flow_stats stats;
5354 error = dpif_flow_del(ofproto->backer->dpif, subfacet->key,
5355 subfacet->key_len, &stats);
5356 subfacet_reset_dp_stats(subfacet, &stats);
5358 subfacet_update_stats(subfacet, &stats);
5360 subfacet->path = SF_NOT_INSTALLED;
5362 ovs_assert(subfacet->dp_packet_count == 0);
5363 ovs_assert(subfacet->dp_byte_count == 0);
5367 /* Resets 'subfacet''s datapath statistics counters. This should be called
5368 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
5369 * non-null, it should contain the statistics returned by dpif when 'subfacet'
5370 * was reset in the datapath. 'stats' will be modified to include only
5371 * statistics new since 'subfacet' was last updated. */
5373 subfacet_reset_dp_stats(struct subfacet *subfacet,
5374 struct dpif_flow_stats *stats)
5377 && subfacet->dp_packet_count <= stats->n_packets
5378 && subfacet->dp_byte_count <= stats->n_bytes) {
5379 stats->n_packets -= subfacet->dp_packet_count;
5380 stats->n_bytes -= subfacet->dp_byte_count;
5383 subfacet->dp_packet_count = 0;
5384 subfacet->dp_byte_count = 0;
5387 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
5389 * Because of the meaning of a subfacet's counters, it only makes sense to do
5390 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
5391 * represents a packet that was sent by hand or if it represents statistics
5392 * that have been cleared out of the datapath. */
5394 subfacet_update_stats(struct subfacet *subfacet,
5395 const struct dpif_flow_stats *stats)
5397 if (stats->n_packets || stats->used > subfacet->used) {
5398 struct facet *facet = subfacet->facet;
5400 subfacet->used = MAX(subfacet->used, stats->used);
5401 facet->used = MAX(facet->used, stats->used);
5402 facet->packet_count += stats->n_packets;
5403 facet->byte_count += stats->n_bytes;
5404 facet->tcp_flags |= stats->tcp_flags;
5410 static struct rule_dpif *
5411 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow)
5413 struct rule_dpif *rule;
5415 rule = rule_dpif_lookup__(ofproto, flow, 0);
5420 return rule_dpif_miss_rule(ofproto, flow);
5423 static struct rule_dpif *
5424 rule_dpif_lookup__(struct ofproto_dpif *ofproto, const struct flow *flow,
5427 struct cls_rule *cls_rule;
5428 struct classifier *cls;
5430 if (table_id >= N_TABLES) {
5434 cls = &ofproto->up.tables[table_id].cls;
5435 if (flow->nw_frag & FLOW_NW_FRAG_ANY
5436 && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
5437 /* For OFPC_NORMAL frag_handling, we must pretend that transport ports
5438 * are unavailable. */
5439 struct flow ofpc_normal_flow = *flow;
5440 ofpc_normal_flow.tp_src = htons(0);
5441 ofpc_normal_flow.tp_dst = htons(0);
5442 cls_rule = classifier_lookup(cls, &ofpc_normal_flow);
5444 cls_rule = classifier_lookup(cls, flow);
5446 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
5449 static struct rule_dpif *
5450 rule_dpif_miss_rule(struct ofproto_dpif *ofproto, const struct flow *flow)
5452 struct ofport_dpif *port;
5454 port = get_ofp_port(ofproto, flow->in_port);
5456 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, flow->in_port);
5457 return ofproto->miss_rule;
5460 if (port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN) {
5461 return ofproto->no_packet_in_rule;
5463 return ofproto->miss_rule;
5467 complete_operation(struct rule_dpif *rule)
5469 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5471 rule_invalidate(rule);
5473 struct dpif_completion *c = xmalloc(sizeof *c);
5474 c->op = rule->up.pending;
5475 list_push_back(&ofproto->completions, &c->list_node);
5477 ofoperation_complete(rule->up.pending, 0);
5481 static struct rule *
5484 struct rule_dpif *rule = xmalloc(sizeof *rule);
5489 rule_dealloc(struct rule *rule_)
5491 struct rule_dpif *rule = rule_dpif_cast(rule_);
5496 rule_construct(struct rule *rule_)
5498 struct rule_dpif *rule = rule_dpif_cast(rule_);
5499 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5500 struct rule_dpif *victim;
5503 rule->packet_count = 0;
5504 rule->byte_count = 0;
5506 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
5507 if (victim && !list_is_empty(&victim->facets)) {
5508 struct facet *facet;
5510 rule->facets = victim->facets;
5511 list_moved(&rule->facets);
5512 LIST_FOR_EACH (facet, list_node, &rule->facets) {
5513 /* XXX: We're only clearing our local counters here. It's possible
5514 * that quite a few packets are unaccounted for in the datapath
5515 * statistics. These will be accounted to the new rule instead of
5516 * cleared as required. This could be fixed by clearing out the
5517 * datapath statistics for this facet, but currently it doesn't
5519 facet_reset_counters(facet);
5523 /* Must avoid list_moved() in this case. */
5524 list_init(&rule->facets);
5527 table_id = rule->up.table_id;
5529 rule->tag = victim->tag;
5530 } else if (table_id == 0) {
5535 miniflow_expand(&rule->up.cr.match.flow, &flow);
5536 rule->tag = rule_calculate_tag(&flow, &rule->up.cr.match.mask,
5537 ofproto->tables[table_id].basis);
5540 complete_operation(rule);
5545 rule_destruct(struct rule *rule_)
5547 struct rule_dpif *rule = rule_dpif_cast(rule_);
5548 struct facet *facet, *next_facet;
5550 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
5551 facet_revalidate(facet);
5554 complete_operation(rule);
5558 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
5560 struct rule_dpif *rule = rule_dpif_cast(rule_);
5562 /* push_all_stats() can handle flow misses which, when using the learn
5563 * action, can cause rules to be added and deleted. This can corrupt our
5564 * caller's datastructures which assume that rule_get_stats() doesn't have
5565 * an impact on the flow table. To be safe, we disable miss handling. */
5566 push_all_stats__(false);
5568 /* Start from historical data for 'rule' itself that are no longer tracked
5569 * in facets. This counts, for example, facets that have expired. */
5570 *packets = rule->packet_count;
5571 *bytes = rule->byte_count;
5575 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
5576 struct ofpbuf *packet)
5578 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5579 struct initial_vals initial_vals;
5580 struct dpif_flow_stats stats;
5581 struct xlate_out xout;
5582 struct xlate_in xin;
5584 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
5585 rule_credit_stats(rule, &stats);
5587 initial_vals.vlan_tci = flow->vlan_tci;
5588 xlate_in_init(&xin, ofproto, flow, &initial_vals, rule, stats.tcp_flags,
5590 xin.resubmit_stats = &stats;
5591 xlate_actions(&xin, &xout);
5593 execute_odp_actions(ofproto, flow, xout.odp_actions.data,
5594 xout.odp_actions.size, packet);
5596 xlate_out_uninit(&xout);
5600 rule_execute(struct rule *rule, const struct flow *flow,
5601 struct ofpbuf *packet)
5603 rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
5604 ofpbuf_delete(packet);
5609 rule_modify_actions(struct rule *rule_)
5611 struct rule_dpif *rule = rule_dpif_cast(rule_);
5613 complete_operation(rule);
5616 /* Sends 'packet' out 'ofport'.
5617 * May modify 'packet'.
5618 * Returns 0 if successful, otherwise a positive errno value. */
5620 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
5622 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
5623 uint64_t odp_actions_stub[1024 / 8];
5624 struct ofpbuf key, odp_actions;
5625 struct dpif_flow_stats stats;
5626 struct odputil_keybuf keybuf;
5627 struct ofpact_output output;
5628 struct xlate_out xout;
5629 struct xlate_in xin;
5633 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5634 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5636 /* Use OFPP_NONE as the in_port to avoid special packet processing. */
5637 flow_extract(packet, 0, 0, NULL, OFPP_NONE, &flow);
5638 odp_flow_key_from_flow(&key, &flow, ofp_port_to_odp_port(ofproto,
5640 dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
5642 ofpact_init(&output.ofpact, OFPACT_OUTPUT, sizeof output);
5643 output.port = ofport->up.ofp_port;
5646 xlate_in_init(&xin, ofproto, &flow, NULL, NULL, 0, packet);
5647 xin.ofpacts_len = sizeof output;
5648 xin.ofpacts = &output.ofpact;
5649 xin.resubmit_stats = &stats;
5650 xlate_actions(&xin, &xout);
5652 error = dpif_execute(ofproto->backer->dpif,
5654 xout.odp_actions.data, xout.odp_actions.size,
5656 xlate_out_uninit(&xout);
5659 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %s (%s)",
5660 ofproto->up.name, netdev_get_name(ofport->up.netdev),
5664 ofproto->stats.tx_packets++;
5665 ofproto->stats.tx_bytes += packet->size;
5669 /* OpenFlow to datapath action translation. */
5671 static bool may_receive(const struct ofport_dpif *, struct xlate_ctx *);
5672 static void do_xlate_actions(const struct ofpact *, size_t ofpacts_len,
5673 struct xlate_ctx *);
5674 static void xlate_normal(struct xlate_ctx *);
5676 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
5677 * The action will state 'slow' as the reason that the action is in the slow
5678 * path. (This is purely informational: it allows a human viewing "ovs-dpctl
5679 * dump-flows" output to see why a flow is in the slow path.)
5681 * The 'stub_size' bytes in 'stub' will be used to store the action.
5682 * 'stub_size' must be large enough for the action.
5684 * The action and its size will be stored in '*actionsp' and '*actions_lenp',
5687 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
5688 enum slow_path_reason slow,
5689 uint64_t *stub, size_t stub_size,
5690 const struct nlattr **actionsp, size_t *actions_lenp)
5692 union user_action_cookie cookie;
5695 cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
5696 cookie.slow_path.unused = 0;
5697 cookie.slow_path.reason = slow;
5699 ofpbuf_use_stack(&buf, stub, stub_size);
5700 if (slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP | SLOW_STP)) {
5701 uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif, UINT32_MAX);
5702 odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
5704 put_userspace_action(ofproto, &buf, flow, &cookie,
5705 sizeof cookie.slow_path);
5707 *actionsp = buf.data;
5708 *actions_lenp = buf.size;
5712 put_userspace_action(const struct ofproto_dpif *ofproto,
5713 struct ofpbuf *odp_actions,
5714 const struct flow *flow,
5715 const union user_action_cookie *cookie,
5716 const size_t cookie_size)
5720 pid = dpif_port_get_pid(ofproto->backer->dpif,
5721 ofp_port_to_odp_port(ofproto, flow->in_port));
5723 return odp_put_userspace_action(pid, cookie, cookie_size, odp_actions);
5726 /* Compose SAMPLE action for sFlow or IPFIX. The given probability is
5727 * the number of packets out of UINT32_MAX to sample. The given
5728 * cookie is passed back in the callback for each sampled packet.
5731 compose_sample_action(const struct ofproto_dpif *ofproto,
5732 struct ofpbuf *odp_actions,
5733 const struct flow *flow,
5734 const uint32_t probability,
5735 const union user_action_cookie *cookie,
5736 const size_t cookie_size)
5738 size_t sample_offset, actions_offset;
5741 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
5743 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
5745 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
5746 cookie_offset = put_userspace_action(ofproto, odp_actions, flow, cookie,
5749 nl_msg_end_nested(odp_actions, actions_offset);
5750 nl_msg_end_nested(odp_actions, sample_offset);
5751 return cookie_offset;
5755 compose_sflow_cookie(const struct ofproto_dpif *ofproto,
5756 ovs_be16 vlan_tci, uint32_t odp_port,
5757 unsigned int n_outputs, union user_action_cookie *cookie)
5761 cookie->type = USER_ACTION_COOKIE_SFLOW;
5762 cookie->sflow.vlan_tci = vlan_tci;
5764 /* See http://www.sflow.org/sflow_version_5.txt (search for "Input/output
5765 * port information") for the interpretation of cookie->output. */
5766 switch (n_outputs) {
5768 /* 0x40000000 | 256 means "packet dropped for unknown reason". */
5769 cookie->sflow.output = 0x40000000 | 256;
5773 ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port);
5775 cookie->sflow.output = ifindex;
5780 /* 0x80000000 means "multiple output ports. */
5781 cookie->sflow.output = 0x80000000 | n_outputs;
5786 /* Compose SAMPLE action for sFlow bridge sampling. */
5788 compose_sflow_action(const struct ofproto_dpif *ofproto,
5789 struct ofpbuf *odp_actions,
5790 const struct flow *flow,
5793 uint32_t probability;
5794 union user_action_cookie cookie;
5796 if (!ofproto->sflow || flow->in_port == OFPP_NONE) {
5800 probability = dpif_sflow_get_probability(ofproto->sflow);
5801 compose_sflow_cookie(ofproto, htons(0), odp_port,
5802 odp_port == OVSP_NONE ? 0 : 1, &cookie);
5804 return compose_sample_action(ofproto, odp_actions, flow, probability,
5805 &cookie, sizeof cookie.sflow);
5809 compose_flow_sample_cookie(uint16_t probability, uint32_t collector_set_id,
5810 uint32_t obs_domain_id, uint32_t obs_point_id,
5811 union user_action_cookie *cookie)
5813 cookie->type = USER_ACTION_COOKIE_FLOW_SAMPLE;
5814 cookie->flow_sample.probability = probability;
5815 cookie->flow_sample.collector_set_id = collector_set_id;
5816 cookie->flow_sample.obs_domain_id = obs_domain_id;
5817 cookie->flow_sample.obs_point_id = obs_point_id;
5821 compose_ipfix_cookie(union user_action_cookie *cookie)
5823 cookie->type = USER_ACTION_COOKIE_IPFIX;
5826 /* Compose SAMPLE action for IPFIX bridge sampling. */
5828 compose_ipfix_action(const struct ofproto_dpif *ofproto,
5829 struct ofpbuf *odp_actions,
5830 const struct flow *flow)
5832 uint32_t probability;
5833 union user_action_cookie cookie;
5835 if (!ofproto->ipfix || flow->in_port == OFPP_NONE) {
5839 probability = dpif_ipfix_get_bridge_exporter_probability(ofproto->ipfix);
5840 compose_ipfix_cookie(&cookie);
5842 compose_sample_action(ofproto, odp_actions, flow, probability,
5843 &cookie, sizeof cookie.ipfix);
5846 /* SAMPLE action for sFlow must be first action in any given list of
5847 * actions. At this point we do not have all information required to
5848 * build it. So try to build sample action as complete as possible. */
5850 add_sflow_action(struct xlate_ctx *ctx)
5852 ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto,
5853 &ctx->xout->odp_actions,
5854 &ctx->xin->flow, OVSP_NONE);
5855 ctx->sflow_odp_port = 0;
5856 ctx->sflow_n_outputs = 0;
5859 /* SAMPLE action for IPFIX must be 1st or 2nd action in any given list
5860 * of actions, eventually after the SAMPLE action for sFlow. */
5862 add_ipfix_action(struct xlate_ctx *ctx)
5864 compose_ipfix_action(ctx->ofproto, &ctx->xout->odp_actions,
5868 /* Fix SAMPLE action according to data collected while composing ODP actions.
5869 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
5870 * USERSPACE action's user-cookie which is required for sflow. */
5872 fix_sflow_action(struct xlate_ctx *ctx)
5874 const struct flow *base = &ctx->base_flow;
5875 union user_action_cookie *cookie;
5877 if (!ctx->user_cookie_offset) {
5881 cookie = ofpbuf_at(&ctx->xout->odp_actions, ctx->user_cookie_offset,
5882 sizeof cookie->sflow);
5883 ovs_assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
5885 compose_sflow_cookie(ctx->ofproto, base->vlan_tci,
5886 ctx->sflow_odp_port, ctx->sflow_n_outputs, cookie);
5890 compose_output_action__(struct xlate_ctx *ctx, uint16_t ofp_port,
5893 const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
5894 ovs_be16 flow_vlan_tci;
5895 uint32_t flow_skb_mark;
5896 uint8_t flow_nw_tos;
5897 struct priority_to_dscp *pdscp;
5898 uint32_t out_port, odp_port;
5900 /* If 'struct flow' gets additional metadata, we'll need to zero it out
5901 * before traversing a patch port. */
5902 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 20);
5905 xlate_report(ctx, "Nonexistent output port");
5907 } else if (ofport->up.pp.config & OFPUTIL_PC_NO_FWD) {
5908 xlate_report(ctx, "OFPPC_NO_FWD set, skipping output");
5910 } else if (check_stp && !stp_forward_in_state(ofport->stp_state)) {
5911 xlate_report(ctx, "STP not in forwarding state, skipping output");
5915 if (netdev_vport_is_patch(ofport->up.netdev)) {
5916 struct ofport_dpif *peer = ofport_get_peer(ofport);
5917 struct flow old_flow = ctx->xin->flow;
5918 const struct ofproto_dpif *peer_ofproto;
5919 enum slow_path_reason special;
5920 struct ofport_dpif *in_port;
5923 xlate_report(ctx, "Nonexistent patch port peer");
5927 peer_ofproto = ofproto_dpif_cast(peer->up.ofproto);
5928 if (peer_ofproto->backer != ctx->ofproto->backer) {
5929 xlate_report(ctx, "Patch port peer on a different datapath");
5933 ctx->ofproto = ofproto_dpif_cast(peer->up.ofproto);
5934 ctx->xin->flow.in_port = peer->up.ofp_port;
5935 ctx->xin->flow.metadata = htonll(0);
5936 memset(&ctx->xin->flow.tunnel, 0, sizeof ctx->xin->flow.tunnel);
5937 memset(ctx->xin->flow.regs, 0, sizeof ctx->xin->flow.regs);
5939 in_port = get_ofp_port(ctx->ofproto, ctx->xin->flow.in_port);
5940 special = process_special(ctx->ofproto, &ctx->xin->flow, in_port,
5943 ctx->xout->slow = special;
5944 } else if (!in_port || may_receive(in_port, ctx)) {
5945 if (!in_port || stp_forward_in_state(in_port->stp_state)) {
5946 xlate_table_action(ctx, ctx->xin->flow.in_port, 0, true);
5948 /* Forwarding is disabled by STP. Let OFPP_NORMAL and the
5949 * learning action look at the packet, then drop it. */
5950 struct flow old_base_flow = ctx->base_flow;
5951 size_t old_size = ctx->xout->odp_actions.size;
5952 xlate_table_action(ctx, ctx->xin->flow.in_port, 0, true);
5953 ctx->base_flow = old_base_flow;
5954 ctx->xout->odp_actions.size = old_size;
5958 ctx->xin->flow = old_flow;
5959 ctx->ofproto = ofproto_dpif_cast(ofport->up.ofproto);
5961 if (ctx->xin->resubmit_stats) {
5962 netdev_vport_inc_tx(ofport->up.netdev, ctx->xin->resubmit_stats);
5963 netdev_vport_inc_rx(peer->up.netdev, ctx->xin->resubmit_stats);
5969 flow_vlan_tci = ctx->xin->flow.vlan_tci;
5970 flow_skb_mark = ctx->xin->flow.skb_mark;
5971 flow_nw_tos = ctx->xin->flow.nw_tos;
5973 pdscp = get_priority(ofport, ctx->xin->flow.skb_priority);
5975 ctx->xin->flow.nw_tos &= ~IP_DSCP_MASK;
5976 ctx->xin->flow.nw_tos |= pdscp->dscp;
5979 if (ofport->tnl_port) {
5980 /* Save tunnel metadata so that changes made due to
5981 * the Logical (tunnel) Port are not visible for any further
5982 * matches, while explicit set actions on tunnel metadata are.
5984 struct flow_tnl flow_tnl = ctx->xin->flow.tunnel;
5985 odp_port = tnl_port_send(ofport->tnl_port, &ctx->xin->flow);
5986 if (odp_port == OVSP_NONE) {
5987 xlate_report(ctx, "Tunneling decided against output");
5988 goto out; /* restore flow_nw_tos */
5990 if (ctx->xin->flow.tunnel.ip_dst == ctx->orig_tunnel_ip_dst) {
5991 xlate_report(ctx, "Not tunneling to our own address");
5992 goto out; /* restore flow_nw_tos */
5994 if (ctx->xin->resubmit_stats) {
5995 netdev_vport_inc_tx(ofport->up.netdev, ctx->xin->resubmit_stats);
5997 out_port = odp_port;
5998 commit_odp_tunnel_action(&ctx->xin->flow, &ctx->base_flow,
5999 &ctx->xout->odp_actions);
6000 ctx->xin->flow.tunnel = flow_tnl; /* Restore tunnel metadata */
6002 uint16_t vlandev_port;
6003 odp_port = ofport->odp_port;
6004 vlandev_port = vsp_realdev_to_vlandev(ctx->ofproto, ofp_port,
6005 ctx->xin->flow.vlan_tci);
6006 if (vlandev_port == ofp_port) {
6007 out_port = odp_port;
6009 out_port = ofp_port_to_odp_port(ctx->ofproto, vlandev_port);
6010 ctx->xin->flow.vlan_tci = htons(0);
6012 ctx->xin->flow.skb_mark &= ~IPSEC_MARK;
6014 commit_odp_actions(&ctx->xin->flow, &ctx->base_flow,
6015 &ctx->xout->odp_actions);
6016 nl_msg_put_u32(&ctx->xout->odp_actions, OVS_ACTION_ATTR_OUTPUT, out_port);
6018 ctx->sflow_odp_port = odp_port;
6019 ctx->sflow_n_outputs++;
6020 ctx->xout->nf_output_iface = ofp_port;
6023 ctx->xin->flow.vlan_tci = flow_vlan_tci;
6024 ctx->xin->flow.skb_mark = flow_skb_mark;
6026 ctx->xin->flow.nw_tos = flow_nw_tos;
6030 compose_output_action(struct xlate_ctx *ctx, uint16_t ofp_port)
6032 compose_output_action__(ctx, ofp_port, true);
6036 tag_the_flow(struct xlate_ctx *ctx, struct rule_dpif *rule)
6038 struct ofproto_dpif *ofproto = ctx->ofproto;
6039 uint8_t table_id = ctx->table_id;
6041 if (table_id > 0 && table_id < N_TABLES) {
6042 struct table_dpif *table = &ofproto->tables[table_id];
6043 if (table->other_table) {
6044 ctx->xout->tags |= (rule && rule->tag
6046 : rule_calculate_tag(&ctx->xin->flow,
6047 &table->other_table->mask,
6053 /* Common rule processing in one place to avoid duplicating code. */
6054 static struct rule_dpif *
6055 ctx_rule_hooks(struct xlate_ctx *ctx, struct rule_dpif *rule,
6058 if (ctx->xin->resubmit_hook) {
6059 ctx->xin->resubmit_hook(ctx, rule);
6061 if (rule == NULL && may_packet_in) {
6063 * check if table configuration flags
6064 * OFPTC_TABLE_MISS_CONTROLLER, default.
6065 * OFPTC_TABLE_MISS_CONTINUE,
6066 * OFPTC_TABLE_MISS_DROP
6067 * When OF1.0, OFPTC_TABLE_MISS_CONTINUE is used. What to do?
6069 rule = rule_dpif_miss_rule(ctx->ofproto, &ctx->xin->flow);
6071 if (rule && ctx->xin->resubmit_stats) {
6072 rule_credit_stats(rule, ctx->xin->resubmit_stats);
6078 xlate_table_action(struct xlate_ctx *ctx,
6079 uint16_t in_port, uint8_t table_id, bool may_packet_in)
6081 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
6082 struct rule_dpif *rule;
6083 uint16_t old_in_port = ctx->xin->flow.in_port;
6084 uint8_t old_table_id = ctx->table_id;
6086 ctx->table_id = table_id;
6088 /* Look up a flow with 'in_port' as the input port. */
6089 ctx->xin->flow.in_port = in_port;
6090 rule = rule_dpif_lookup__(ctx->ofproto, &ctx->xin->flow, table_id);
6092 tag_the_flow(ctx, rule);
6094 /* Restore the original input port. Otherwise OFPP_NORMAL and
6095 * OFPP_IN_PORT will have surprising behavior. */
6096 ctx->xin->flow.in_port = old_in_port;
6098 rule = ctx_rule_hooks(ctx, rule, may_packet_in);
6101 struct rule_dpif *old_rule = ctx->rule;
6105 do_xlate_actions(rule->up.ofpacts, rule->up.ofpacts_len, ctx);
6106 ctx->rule = old_rule;
6110 ctx->table_id = old_table_id;
6112 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
6114 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
6115 MAX_RESUBMIT_RECURSION);
6116 ctx->max_resubmit_trigger = true;
6121 xlate_ofpact_resubmit(struct xlate_ctx *ctx,
6122 const struct ofpact_resubmit *resubmit)
6127 in_port = resubmit->in_port;
6128 if (in_port == OFPP_IN_PORT) {
6129 in_port = ctx->xin->flow.in_port;
6132 table_id = resubmit->table_id;
6133 if (table_id == 255) {
6134 table_id = ctx->table_id;
6137 xlate_table_action(ctx, in_port, table_id, false);
6141 flood_packets(struct xlate_ctx *ctx, bool all)
6143 struct ofport_dpif *ofport;
6145 HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) {
6146 uint16_t ofp_port = ofport->up.ofp_port;
6148 if (ofp_port == ctx->xin->flow.in_port) {
6153 compose_output_action__(ctx, ofp_port, false);
6154 } else if (!(ofport->up.pp.config & OFPUTIL_PC_NO_FLOOD)) {
6155 compose_output_action(ctx, ofp_port);
6159 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
6163 execute_controller_action(struct xlate_ctx *ctx, int len,
6164 enum ofp_packet_in_reason reason,
6165 uint16_t controller_id)
6167 struct ofputil_packet_in pin;
6168 struct ofpbuf *packet;
6170 ovs_assert(!ctx->xout->slow || ctx->xout->slow == SLOW_CONTROLLER);
6171 ctx->xout->slow = SLOW_CONTROLLER;
6172 if (!ctx->xin->packet) {
6176 packet = ofpbuf_clone(ctx->xin->packet);
6178 if (packet->l2 && packet->l3) {
6179 struct eth_header *eh;
6180 uint16_t mpls_depth;
6182 eth_pop_vlan(packet);
6185 memcpy(eh->eth_src, ctx->xin->flow.dl_src, sizeof eh->eth_src);
6186 memcpy(eh->eth_dst, ctx->xin->flow.dl_dst, sizeof eh->eth_dst);
6188 if (ctx->xin->flow.vlan_tci & htons(VLAN_CFI)) {
6189 eth_push_vlan(packet, ctx->xin->flow.vlan_tci);
6192 mpls_depth = eth_mpls_depth(packet);
6194 if (mpls_depth < ctx->xin->flow.mpls_depth) {
6195 push_mpls(packet, ctx->xin->flow.dl_type, ctx->xin->flow.mpls_lse);
6196 } else if (mpls_depth > ctx->xin->flow.mpls_depth) {
6197 pop_mpls(packet, ctx->xin->flow.dl_type);
6198 } else if (mpls_depth) {
6199 set_mpls_lse(packet, ctx->xin->flow.mpls_lse);
6203 if (ctx->xin->flow.dl_type == htons(ETH_TYPE_IP)) {
6204 packet_set_ipv4(packet, ctx->xin->flow.nw_src,
6205 ctx->xin->flow.nw_dst, ctx->xin->flow.nw_tos,
6206 ctx->xin->flow.nw_ttl);
6210 if (ctx->xin->flow.nw_proto == IPPROTO_TCP) {
6211 packet_set_tcp_port(packet, ctx->xin->flow.tp_src,
6212 ctx->xin->flow.tp_dst);
6213 } else if (ctx->xin->flow.nw_proto == IPPROTO_UDP) {
6214 packet_set_udp_port(packet, ctx->xin->flow.tp_src,
6215 ctx->xin->flow.tp_dst);
6221 pin.packet = packet->data;
6222 pin.packet_len = packet->size;
6223 pin.reason = reason;
6224 pin.controller_id = controller_id;
6225 pin.table_id = ctx->table_id;
6226 pin.cookie = ctx->rule ? ctx->rule->up.flow_cookie : 0;
6229 flow_get_metadata(&ctx->xin->flow, &pin.fmd);
6231 connmgr_send_packet_in(ctx->ofproto->up.connmgr, &pin);
6232 ofpbuf_delete(packet);
6236 execute_mpls_push_action(struct xlate_ctx *ctx, ovs_be16 eth_type)
6238 ovs_assert(eth_type_mpls(eth_type));
6240 if (ctx->base_flow.mpls_depth) {
6241 ctx->xin->flow.mpls_lse &= ~htonl(MPLS_BOS_MASK);
6242 ctx->xin->flow.mpls_depth++;
6247 if (ctx->xin->flow.dl_type == htons(ETH_TYPE_IPV6)) {
6248 label = htonl(0x2); /* IPV6 Explicit Null. */
6250 label = htonl(0x0); /* IPV4 Explicit Null. */
6252 tc = (ctx->xin->flow.nw_tos & IP_DSCP_MASK) >> 2;
6253 ttl = ctx->xin->flow.nw_ttl ? ctx->xin->flow.nw_ttl : 0x40;
6254 ctx->xin->flow.mpls_lse = set_mpls_lse_values(ttl, tc, 1, label);
6255 ctx->xin->flow.mpls_depth = 1;
6257 ctx->xin->flow.dl_type = eth_type;
6261 execute_mpls_pop_action(struct xlate_ctx *ctx, ovs_be16 eth_type)
6263 ovs_assert(eth_type_mpls(ctx->xin->flow.dl_type));
6264 ovs_assert(!eth_type_mpls(eth_type));
6266 if (ctx->xin->flow.mpls_depth) {
6267 ctx->xin->flow.mpls_depth--;
6268 ctx->xin->flow.mpls_lse = htonl(0);
6269 if (!ctx->xin->flow.mpls_depth) {
6270 ctx->xin->flow.dl_type = eth_type;
6276 compose_dec_ttl(struct xlate_ctx *ctx, struct ofpact_cnt_ids *ids)
6278 if (ctx->xin->flow.dl_type != htons(ETH_TYPE_IP) &&
6279 ctx->xin->flow.dl_type != htons(ETH_TYPE_IPV6)) {
6283 if (ctx->xin->flow.nw_ttl > 1) {
6284 ctx->xin->flow.nw_ttl--;
6289 for (i = 0; i < ids->n_controllers; i++) {
6290 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL,
6294 /* Stop processing for current table. */
6300 execute_set_mpls_ttl_action(struct xlate_ctx *ctx, uint8_t ttl)
6302 if (!eth_type_mpls(ctx->xin->flow.dl_type)) {
6306 set_mpls_lse_ttl(&ctx->xin->flow.mpls_lse, ttl);
6311 execute_dec_mpls_ttl_action(struct xlate_ctx *ctx)
6313 uint8_t ttl = mpls_lse_to_ttl(ctx->xin->flow.mpls_lse);
6315 if (!eth_type_mpls(ctx->xin->flow.dl_type)) {
6321 set_mpls_lse_ttl(&ctx->xin->flow.mpls_lse, ttl);
6324 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL, 0);
6326 /* Stop processing for current table. */
6332 xlate_output_action(struct xlate_ctx *ctx,
6333 uint16_t port, uint16_t max_len, bool may_packet_in)
6335 uint16_t prev_nf_output_iface = ctx->xout->nf_output_iface;
6337 ctx->xout->nf_output_iface = NF_OUT_DROP;
6341 compose_output_action(ctx, ctx->xin->flow.in_port);
6344 xlate_table_action(ctx, ctx->xin->flow.in_port, 0, may_packet_in);
6350 flood_packets(ctx, false);
6353 flood_packets(ctx, true);
6355 case OFPP_CONTROLLER:
6356 execute_controller_action(ctx, max_len, OFPR_ACTION, 0);
6362 if (port != ctx->xin->flow.in_port) {
6363 compose_output_action(ctx, port);
6365 xlate_report(ctx, "skipping output to input port");
6370 if (prev_nf_output_iface == NF_OUT_FLOOD) {
6371 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
6372 } else if (ctx->xout->nf_output_iface == NF_OUT_DROP) {
6373 ctx->xout->nf_output_iface = prev_nf_output_iface;
6374 } else if (prev_nf_output_iface != NF_OUT_DROP &&
6375 ctx->xout->nf_output_iface != NF_OUT_FLOOD) {
6376 ctx->xout->nf_output_iface = NF_OUT_MULTI;
6381 xlate_output_reg_action(struct xlate_ctx *ctx,
6382 const struct ofpact_output_reg *or)
6384 uint64_t port = mf_get_subfield(&or->src, &ctx->xin->flow);
6385 if (port <= UINT16_MAX) {
6386 xlate_output_action(ctx, port, or->max_len, false);
6391 xlate_enqueue_action(struct xlate_ctx *ctx,
6392 const struct ofpact_enqueue *enqueue)
6394 uint16_t ofp_port = enqueue->port;
6395 uint32_t queue_id = enqueue->queue;
6396 uint32_t flow_priority, priority;
6399 /* Translate queue to priority. */
6400 error = dpif_queue_to_priority(ctx->ofproto->backer->dpif,
6401 queue_id, &priority);
6403 /* Fall back to ordinary output action. */
6404 xlate_output_action(ctx, enqueue->port, 0, false);
6408 /* Check output port. */
6409 if (ofp_port == OFPP_IN_PORT) {
6410 ofp_port = ctx->xin->flow.in_port;
6411 } else if (ofp_port == ctx->xin->flow.in_port) {
6415 /* Add datapath actions. */
6416 flow_priority = ctx->xin->flow.skb_priority;
6417 ctx->xin->flow.skb_priority = priority;
6418 compose_output_action(ctx, ofp_port);
6419 ctx->xin->flow.skb_priority = flow_priority;
6421 /* Update NetFlow output port. */
6422 if (ctx->xout->nf_output_iface == NF_OUT_DROP) {
6423 ctx->xout->nf_output_iface = ofp_port;
6424 } else if (ctx->xout->nf_output_iface != NF_OUT_FLOOD) {
6425 ctx->xout->nf_output_iface = NF_OUT_MULTI;
6430 xlate_set_queue_action(struct xlate_ctx *ctx, uint32_t queue_id)
6432 uint32_t skb_priority;
6434 if (!dpif_queue_to_priority(ctx->ofproto->backer->dpif,
6435 queue_id, &skb_priority)) {
6436 ctx->xin->flow.skb_priority = skb_priority;
6438 /* Couldn't translate queue to a priority. Nothing to do. A warning
6439 * has already been logged. */
6444 slave_enabled_cb(uint16_t ofp_port, void *ofproto_)
6446 struct ofproto_dpif *ofproto = ofproto_;
6447 struct ofport_dpif *port;
6457 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
6460 port = get_ofp_port(ofproto, ofp_port);
6461 return port ? port->may_enable : false;
6466 xlate_bundle_action(struct xlate_ctx *ctx,
6467 const struct ofpact_bundle *bundle)
6471 port = bundle_execute(bundle, &ctx->xin->flow, slave_enabled_cb,
6473 if (bundle->dst.field) {
6474 nxm_reg_load(&bundle->dst, port, &ctx->xin->flow);
6476 xlate_output_action(ctx, port, 0, false);
6481 xlate_learn_action(struct xlate_ctx *ctx,
6482 const struct ofpact_learn *learn)
6484 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
6485 struct ofputil_flow_mod fm;
6486 uint64_t ofpacts_stub[1024 / 8];
6487 struct ofpbuf ofpacts;
6490 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
6491 learn_execute(learn, &ctx->xin->flow, &fm, &ofpacts);
6493 error = ofproto_flow_mod(&ctx->ofproto->up, &fm);
6494 if (error && !VLOG_DROP_WARN(&rl)) {
6495 VLOG_WARN("learning action failed to modify flow table (%s)",
6496 ofperr_get_name(error));
6499 ofpbuf_uninit(&ofpacts);
6502 /* Reduces '*timeout' to no more than 'max'. A value of zero in either case
6503 * means "infinite". */
6505 reduce_timeout(uint16_t max, uint16_t *timeout)
6507 if (max && (!*timeout || *timeout > max)) {
6513 xlate_fin_timeout(struct xlate_ctx *ctx,
6514 const struct ofpact_fin_timeout *oft)
6516 if (ctx->xin->tcp_flags & (TCP_FIN | TCP_RST) && ctx->rule) {
6517 struct rule_dpif *rule = ctx->rule;
6519 reduce_timeout(oft->fin_idle_timeout, &rule->up.idle_timeout);
6520 reduce_timeout(oft->fin_hard_timeout, &rule->up.hard_timeout);
6525 xlate_sample_action(struct xlate_ctx *ctx,
6526 const struct ofpact_sample *os)
6528 union user_action_cookie cookie;
6529 /* Scale the probability from 16-bit to 32-bit while representing
6530 * the same percentage. */
6531 uint32_t probability = (os->probability << 16) | os->probability;
6533 commit_odp_actions(&ctx->xin->flow, &ctx->base_flow,
6534 &ctx->xout->odp_actions);
6536 compose_flow_sample_cookie(os->probability, os->collector_set_id,
6537 os->obs_domain_id, os->obs_point_id, &cookie);
6538 compose_sample_action(ctx->ofproto, &ctx->xout->odp_actions, &ctx->xin->flow,
6539 probability, &cookie, sizeof cookie.flow_sample);
6543 may_receive(const struct ofport_dpif *port, struct xlate_ctx *ctx)
6545 if (port->up.pp.config & (eth_addr_equals(ctx->xin->flow.dl_dst,
6547 ? OFPUTIL_PC_NO_RECV_STP
6548 : OFPUTIL_PC_NO_RECV)) {
6552 /* Only drop packets here if both forwarding and learning are
6553 * disabled. If just learning is enabled, we need to have
6554 * OFPP_NORMAL and the learning action have a look at the packet
6555 * before we can drop it. */
6556 if (!stp_forward_in_state(port->stp_state)
6557 && !stp_learn_in_state(port->stp_state)) {
6565 tunnel_ecn_ok(struct xlate_ctx *ctx)
6567 if (is_ip_any(&ctx->base_flow)
6568 && (ctx->xin->flow.tunnel.ip_tos & IP_ECN_MASK) == IP_ECN_CE) {
6569 if ((ctx->base_flow.nw_tos & IP_ECN_MASK) == IP_ECN_NOT_ECT) {
6570 VLOG_WARN_RL(&rl, "dropping tunnel packet marked ECN CE"
6571 " but is not ECN capable");
6574 /* Set the ECN CE value in the tunneled packet. */
6575 ctx->xin->flow.nw_tos |= IP_ECN_CE;
6583 do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len,
6584 struct xlate_ctx *ctx)
6586 bool was_evictable = true;
6587 const struct ofpact *a;
6590 /* Don't let the rule we're working on get evicted underneath us. */
6591 was_evictable = ctx->rule->up.evictable;
6592 ctx->rule->up.evictable = false;
6595 do_xlate_actions_again:
6596 OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) {
6597 struct ofpact_controller *controller;
6598 const struct ofpact_metadata *metadata;
6606 xlate_output_action(ctx, ofpact_get_OUTPUT(a)->port,
6607 ofpact_get_OUTPUT(a)->max_len, true);
6610 case OFPACT_CONTROLLER:
6611 controller = ofpact_get_CONTROLLER(a);
6612 execute_controller_action(ctx, controller->max_len,
6614 controller->controller_id);
6617 case OFPACT_ENQUEUE:
6618 xlate_enqueue_action(ctx, ofpact_get_ENQUEUE(a));
6621 case OFPACT_SET_VLAN_VID:
6622 ctx->xin->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
6623 ctx->xin->flow.vlan_tci |=
6624 (htons(ofpact_get_SET_VLAN_VID(a)->vlan_vid)
6628 case OFPACT_SET_VLAN_PCP:
6629 ctx->xin->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
6630 ctx->xin->flow.vlan_tci |=
6631 htons((ofpact_get_SET_VLAN_PCP(a)->vlan_pcp << VLAN_PCP_SHIFT)
6635 case OFPACT_STRIP_VLAN:
6636 ctx->xin->flow.vlan_tci = htons(0);
6639 case OFPACT_PUSH_VLAN:
6640 /* XXX 802.1AD(QinQ) */
6641 ctx->xin->flow.vlan_tci = htons(VLAN_CFI);
6644 case OFPACT_SET_ETH_SRC:
6645 memcpy(ctx->xin->flow.dl_src, ofpact_get_SET_ETH_SRC(a)->mac,
6649 case OFPACT_SET_ETH_DST:
6650 memcpy(ctx->xin->flow.dl_dst, ofpact_get_SET_ETH_DST(a)->mac,
6654 case OFPACT_SET_IPV4_SRC:
6655 if (ctx->xin->flow.dl_type == htons(ETH_TYPE_IP)) {
6656 ctx->xin->flow.nw_src = ofpact_get_SET_IPV4_SRC(a)->ipv4;
6660 case OFPACT_SET_IPV4_DST:
6661 if (ctx->xin->flow.dl_type == htons(ETH_TYPE_IP)) {
6662 ctx->xin->flow.nw_dst = ofpact_get_SET_IPV4_DST(a)->ipv4;
6666 case OFPACT_SET_IPV4_DSCP:
6667 /* OpenFlow 1.0 only supports IPv4. */
6668 if (ctx->xin->flow.dl_type == htons(ETH_TYPE_IP)) {
6669 ctx->xin->flow.nw_tos &= ~IP_DSCP_MASK;
6670 ctx->xin->flow.nw_tos |= ofpact_get_SET_IPV4_DSCP(a)->dscp;
6674 case OFPACT_SET_L4_SRC_PORT:
6675 if (is_ip_any(&ctx->xin->flow)) {
6676 ctx->xin->flow.tp_src =
6677 htons(ofpact_get_SET_L4_SRC_PORT(a)->port);
6681 case OFPACT_SET_L4_DST_PORT:
6682 if (is_ip_any(&ctx->xin->flow)) {
6683 ctx->xin->flow.tp_dst =
6684 htons(ofpact_get_SET_L4_DST_PORT(a)->port);
6688 case OFPACT_RESUBMIT:
6689 xlate_ofpact_resubmit(ctx, ofpact_get_RESUBMIT(a));
6692 case OFPACT_SET_TUNNEL:
6693 ctx->xin->flow.tunnel.tun_id =
6694 htonll(ofpact_get_SET_TUNNEL(a)->tun_id);
6697 case OFPACT_SET_QUEUE:
6698 xlate_set_queue_action(ctx, ofpact_get_SET_QUEUE(a)->queue_id);
6701 case OFPACT_POP_QUEUE:
6702 ctx->xin->flow.skb_priority = ctx->orig_skb_priority;
6705 case OFPACT_REG_MOVE:
6706 nxm_execute_reg_move(ofpact_get_REG_MOVE(a), &ctx->xin->flow);
6709 case OFPACT_REG_LOAD:
6710 nxm_execute_reg_load(ofpact_get_REG_LOAD(a), &ctx->xin->flow);
6713 case OFPACT_STACK_PUSH:
6714 nxm_execute_stack_push(ofpact_get_STACK_PUSH(a), &ctx->xin->flow,
6718 case OFPACT_STACK_POP:
6719 nxm_execute_stack_pop(ofpact_get_STACK_POP(a), &ctx->xin->flow,
6723 case OFPACT_PUSH_MPLS:
6724 execute_mpls_push_action(ctx, ofpact_get_PUSH_MPLS(a)->ethertype);
6727 case OFPACT_POP_MPLS:
6728 execute_mpls_pop_action(ctx, ofpact_get_POP_MPLS(a)->ethertype);
6731 case OFPACT_SET_MPLS_TTL:
6732 if (execute_set_mpls_ttl_action(ctx,
6733 ofpact_get_SET_MPLS_TTL(a)->ttl)) {
6738 case OFPACT_DEC_MPLS_TTL:
6739 if (execute_dec_mpls_ttl_action(ctx)) {
6744 case OFPACT_DEC_TTL:
6745 if (compose_dec_ttl(ctx, ofpact_get_DEC_TTL(a))) {
6751 /* Nothing to do. */
6754 case OFPACT_MULTIPATH:
6755 multipath_execute(ofpact_get_MULTIPATH(a), &ctx->xin->flow);
6759 ctx->ofproto->has_bundle_action = true;
6760 xlate_bundle_action(ctx, ofpact_get_BUNDLE(a));
6763 case OFPACT_OUTPUT_REG:
6764 xlate_output_reg_action(ctx, ofpact_get_OUTPUT_REG(a));
6768 ctx->xout->has_learn = true;
6769 if (ctx->xin->may_learn) {
6770 xlate_learn_action(ctx, ofpact_get_LEARN(a));
6778 case OFPACT_FIN_TIMEOUT:
6779 ctx->xout->has_fin_timeout = true;
6780 xlate_fin_timeout(ctx, ofpact_get_FIN_TIMEOUT(a));
6783 case OFPACT_CLEAR_ACTIONS:
6785 * Nothing to do because writa-actions is not supported for now.
6786 * When writa-actions is supported, clear-actions also must
6787 * be supported at the same time.
6791 case OFPACT_WRITE_METADATA:
6792 metadata = ofpact_get_WRITE_METADATA(a);
6793 ctx->xin->flow.metadata &= ~metadata->mask;
6794 ctx->xin->flow.metadata |= metadata->metadata & metadata->mask;
6797 case OFPACT_GOTO_TABLE: {
6798 /* It is assumed that goto-table is the last action. */
6799 struct ofpact_goto_table *ogt = ofpact_get_GOTO_TABLE(a);
6800 struct rule_dpif *rule;
6802 ovs_assert(ctx->table_id < ogt->table_id);
6804 ctx->table_id = ogt->table_id;
6806 /* Look up a flow from the new table. */
6807 rule = rule_dpif_lookup__(ctx->ofproto, &ctx->xin->flow, ctx->table_id);
6809 tag_the_flow(ctx, rule);
6811 rule = ctx_rule_hooks(ctx, rule, true);
6815 ctx->rule->up.evictable = was_evictable;
6818 was_evictable = rule->up.evictable;
6819 rule->up.evictable = false;
6821 /* Tail recursion removal. */
6822 ofpacts = rule->up.ofpacts;
6823 ofpacts_len = rule->up.ofpacts_len;
6824 goto do_xlate_actions_again;
6830 xlate_sample_action(ctx, ofpact_get_SAMPLE(a));
6837 ctx->rule->up.evictable = was_evictable;
6842 xlate_in_init(struct xlate_in *xin, struct ofproto_dpif *ofproto,
6843 const struct flow *flow,
6844 const struct initial_vals *initial_vals,
6845 struct rule_dpif *rule, uint8_t tcp_flags,
6846 const struct ofpbuf *packet)
6848 xin->ofproto = ofproto;
6850 xin->packet = packet;
6851 xin->may_learn = packet != NULL;
6853 xin->ofpacts = NULL;
6854 xin->ofpacts_len = 0;
6855 xin->tcp_flags = tcp_flags;
6856 xin->resubmit_hook = NULL;
6857 xin->report_hook = NULL;
6858 xin->resubmit_stats = NULL;
6861 xin->initial_vals = *initial_vals;
6863 xin->initial_vals.vlan_tci = xin->flow.vlan_tci;
6868 xlate_out_uninit(struct xlate_out *xout)
6871 ofpbuf_uninit(&xout->odp_actions);
6875 /* Translates the 'ofpacts_len' bytes of "struct ofpacts" starting at 'ofpacts'
6876 * into datapath actions in 'odp_actions', using 'ctx'. */
6878 xlate_actions(struct xlate_in *xin, struct xlate_out *xout)
6880 /* Normally false. Set to true if we ever hit MAX_RESUBMIT_RECURSION, so
6881 * that in the future we always keep a copy of the original flow for
6882 * tracing purposes. */
6883 static bool hit_resubmit_limit;
6885 enum slow_path_reason special;
6886 const struct ofpact *ofpacts;
6887 struct ofport_dpif *in_port;
6888 struct flow orig_flow;
6889 struct xlate_ctx ctx;
6892 COVERAGE_INC(ofproto_dpif_xlate);
6894 /* Flow initialization rules:
6895 * - 'base_flow' must match the kernel's view of the packet at the
6896 * time that action processing starts. 'flow' represents any
6897 * transformations we wish to make through actions.
6898 * - By default 'base_flow' and 'flow' are the same since the input
6899 * packet matches the output before any actions are applied.
6900 * - When using VLAN splinters, 'base_flow''s VLAN is set to the value
6901 * of the received packet as seen by the kernel. If we later output
6902 * to another device without any modifications this will cause us to
6903 * insert a new tag since the original one was stripped off by the
6905 * - Tunnel metadata as received is retained in 'flow'. This allows
6906 * tunnel metadata matching also in later tables.
6907 * Since a kernel action for setting the tunnel metadata will only be
6908 * generated with actual tunnel output, changing the tunnel metadata
6909 * values in 'flow' (such as tun_id) will only have effect with a later
6910 * tunnel output action.
6911 * - Tunnel 'base_flow' is completely cleared since that is what the
6912 * kernel does. If we wish to maintain the original values an action
6913 * needs to be generated. */
6918 ctx.ofproto = xin->ofproto;
6919 ctx.rule = xin->rule;
6921 ctx.base_flow = ctx.xin->flow;
6922 ctx.base_flow.vlan_tci = xin->initial_vals.vlan_tci;
6923 memset(&ctx.base_flow.tunnel, 0, sizeof ctx.base_flow.tunnel);
6924 ctx.orig_tunnel_ip_dst = ctx.xin->flow.tunnel.ip_dst;
6928 ctx.xout->has_learn = false;
6929 ctx.xout->has_normal = false;
6930 ctx.xout->has_fin_timeout = false;
6931 ctx.xout->nf_output_iface = NF_OUT_DROP;
6932 ctx.xout->mirrors = 0;
6934 ofpbuf_use_stub(&ctx.xout->odp_actions, ctx.xout->odp_actions_stub,
6935 sizeof ctx.xout->odp_actions_stub);
6936 ofpbuf_reserve(&ctx.xout->odp_actions, NL_A_U32_SIZE);
6939 ctx.max_resubmit_trigger = false;
6940 ctx.orig_skb_priority = ctx.xin->flow.skb_priority;
6945 ofpacts = xin->ofpacts;
6946 ofpacts_len = xin->ofpacts_len;
6947 } else if (xin->rule) {
6948 ofpacts = xin->rule->up.ofpacts;
6949 ofpacts_len = xin->rule->up.ofpacts_len;
6954 ofpbuf_use_stub(&ctx.stack, ctx.init_stack, sizeof ctx.init_stack);
6956 if (ctx.ofproto->has_mirrors || hit_resubmit_limit) {
6957 /* Do this conditionally because the copy is expensive enough that it
6958 * shows up in profiles. */
6959 orig_flow = ctx.xin->flow;
6962 if (ctx.xin->flow.nw_frag & FLOW_NW_FRAG_ANY) {
6963 switch (ctx.ofproto->up.frag_handling) {
6964 case OFPC_FRAG_NORMAL:
6965 /* We must pretend that transport ports are unavailable. */
6966 ctx.xin->flow.tp_src = ctx.base_flow.tp_src = htons(0);
6967 ctx.xin->flow.tp_dst = ctx.base_flow.tp_dst = htons(0);
6970 case OFPC_FRAG_DROP:
6973 case OFPC_FRAG_REASM:
6976 case OFPC_FRAG_NX_MATCH:
6977 /* Nothing to do. */
6980 case OFPC_INVALID_TTL_TO_CONTROLLER:
6985 in_port = get_ofp_port(ctx.ofproto, ctx.xin->flow.in_port);
6986 special = process_special(ctx.ofproto, &ctx.xin->flow, in_port,
6989 ctx.xout->slow = special;
6991 static struct vlog_rate_limit trace_rl = VLOG_RATE_LIMIT_INIT(1, 1);
6992 struct initial_vals initial_vals;
6993 size_t sample_actions_len;
6994 uint32_t local_odp_port;
6996 initial_vals.vlan_tci = ctx.base_flow.vlan_tci;
6998 add_sflow_action(&ctx);
6999 add_ipfix_action(&ctx);
7000 sample_actions_len = ctx.xout->odp_actions.size;
7002 if (tunnel_ecn_ok(&ctx) && (!in_port || may_receive(in_port, &ctx))) {
7003 do_xlate_actions(ofpacts, ofpacts_len, &ctx);
7005 /* We've let OFPP_NORMAL and the learning action look at the
7006 * packet, so drop it now if forwarding is disabled. */
7007 if (in_port && !stp_forward_in_state(in_port->stp_state)) {
7008 ctx.xout->odp_actions.size = sample_actions_len;
7012 if (ctx.max_resubmit_trigger && !ctx.xin->resubmit_hook) {
7013 if (!hit_resubmit_limit) {
7014 /* We didn't record the original flow. Make sure we do from
7016 hit_resubmit_limit = true;
7017 } else if (!VLOG_DROP_ERR(&trace_rl)) {
7018 struct ds ds = DS_EMPTY_INITIALIZER;
7020 ofproto_trace(ctx.ofproto, &orig_flow, ctx.xin->packet,
7021 &initial_vals, &ds);
7022 VLOG_ERR("Trace triggered by excessive resubmit "
7023 "recursion:\n%s", ds_cstr(&ds));
7028 local_odp_port = ofp_port_to_odp_port(ctx.ofproto, OFPP_LOCAL);
7029 if (!connmgr_must_output_local(ctx.ofproto->up.connmgr, &ctx.xin->flow,
7031 ctx.xout->odp_actions.data,
7032 ctx.xout->odp_actions.size)) {
7033 compose_output_action(&ctx, OFPP_LOCAL);
7035 if (ctx.ofproto->has_mirrors) {
7036 add_mirror_actions(&ctx, &orig_flow);
7038 fix_sflow_action(&ctx);
7041 ofpbuf_uninit(&ctx.stack);
7044 /* Translates the 'ofpacts_len' bytes of "struct ofpact"s starting at 'ofpacts'
7045 * into datapath actions, using 'ctx', and discards the datapath actions. */
7047 xlate_actions_for_side_effects(struct xlate_in *xin)
7049 struct xlate_out xout;
7051 xlate_actions(xin, &xout);
7052 xlate_out_uninit(&xout);
7056 xlate_report(struct xlate_ctx *ctx, const char *s)
7058 if (ctx->xin->report_hook) {
7059 ctx->xin->report_hook(ctx, s);
7063 /* OFPP_NORMAL implementation. */
7065 static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *);
7067 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
7068 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle',
7069 * the bundle on which the packet was received, returns the VLAN to which the
7072 * Both 'vid' and the return value are in the range 0...4095. */
7074 input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid)
7076 switch (in_bundle->vlan_mode) {
7077 case PORT_VLAN_ACCESS:
7078 return in_bundle->vlan;
7081 case PORT_VLAN_TRUNK:
7084 case PORT_VLAN_NATIVE_UNTAGGED:
7085 case PORT_VLAN_NATIVE_TAGGED:
7086 return vid ? vid : in_bundle->vlan;
7093 /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'.
7094 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
7097 * 'vid' should be the VID obtained from the 802.1Q header that was received as
7098 * part of a packet (specify 0 if there was no 802.1Q header), in the range
7101 input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn)
7103 /* Allow any VID on the OFPP_NONE port. */
7104 if (in_bundle == &ofpp_none_bundle) {
7108 switch (in_bundle->vlan_mode) {
7109 case PORT_VLAN_ACCESS:
7112 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7113 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
7114 "packet received on port %s configured as VLAN "
7115 "%"PRIu16" access port",
7116 in_bundle->ofproto->up.name, vid,
7117 in_bundle->name, in_bundle->vlan);
7123 case PORT_VLAN_NATIVE_UNTAGGED:
7124 case PORT_VLAN_NATIVE_TAGGED:
7126 /* Port must always carry its native VLAN. */
7130 case PORT_VLAN_TRUNK:
7131 if (!ofbundle_includes_vlan(in_bundle, vid)) {
7133 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7134 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet "
7135 "received on port %s not configured for trunking "
7137 in_bundle->ofproto->up.name, vid,
7138 in_bundle->name, vid);
7150 /* Given 'vlan', the VLAN that a packet belongs to, and
7151 * 'out_bundle', a bundle on which the packet is to be output, returns the VID
7152 * that should be included in the 802.1Q header. (If the return value is 0,
7153 * then the 802.1Q header should only be included in the packet if there is a
7156 * Both 'vlan' and the return value are in the range 0...4095. */
7158 output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan)
7160 switch (out_bundle->vlan_mode) {
7161 case PORT_VLAN_ACCESS:
7164 case PORT_VLAN_TRUNK:
7165 case PORT_VLAN_NATIVE_TAGGED:
7168 case PORT_VLAN_NATIVE_UNTAGGED:
7169 return vlan == out_bundle->vlan ? 0 : vlan;
7177 output_normal(struct xlate_ctx *ctx, const struct ofbundle *out_bundle,
7180 struct ofport_dpif *port;
7182 ovs_be16 tci, old_tci;
7184 vid = output_vlan_to_vid(out_bundle, vlan);
7185 if (!out_bundle->bond) {
7186 port = ofbundle_get_a_port(out_bundle);
7188 port = bond_choose_output_slave(out_bundle->bond, &ctx->xin->flow,
7189 vid, &ctx->xout->tags);
7191 /* No slaves enabled, so drop packet. */
7196 old_tci = ctx->xin->flow.vlan_tci;
7198 if (tci || out_bundle->use_priority_tags) {
7199 tci |= ctx->xin->flow.vlan_tci & htons(VLAN_PCP_MASK);
7201 tci |= htons(VLAN_CFI);
7204 ctx->xin->flow.vlan_tci = tci;
7206 compose_output_action(ctx, port->up.ofp_port);
7207 ctx->xin->flow.vlan_tci = old_tci;
7211 mirror_mask_ffs(mirror_mask_t mask)
7213 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
7218 ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
7220 return (bundle->vlan_mode != PORT_VLAN_ACCESS
7221 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
7225 ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan)
7227 return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan);
7230 /* Returns an arbitrary interface within 'bundle'. */
7231 static struct ofport_dpif *
7232 ofbundle_get_a_port(const struct ofbundle *bundle)
7234 return CONTAINER_OF(list_front(&bundle->ports),
7235 struct ofport_dpif, bundle_node);
7239 vlan_is_mirrored(const struct ofmirror *m, int vlan)
7241 return !m->vlans || bitmap_is_set(m->vlans, vlan);
7245 add_mirror_actions(struct xlate_ctx *ctx, const struct flow *orig_flow)
7247 struct ofproto_dpif *ofproto = ctx->ofproto;
7248 mirror_mask_t mirrors;
7249 struct ofbundle *in_bundle;
7252 const struct nlattr *a;
7255 in_bundle = lookup_input_bundle(ctx->ofproto, orig_flow->in_port,
7256 ctx->xin->packet != NULL, NULL);
7260 mirrors = in_bundle->src_mirrors;
7262 /* Drop frames on bundles reserved for mirroring. */
7263 if (in_bundle->mirror_out) {
7264 if (ctx->xin->packet != NULL) {
7265 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7266 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
7267 "%s, which is reserved exclusively for mirroring",
7268 ctx->ofproto->up.name, in_bundle->name);
7274 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
7275 if (!input_vid_is_valid(vid, in_bundle, ctx->xin->packet != NULL)) {
7278 vlan = input_vid_to_vlan(in_bundle, vid);
7280 /* Look at the output ports to check for destination selections. */
7282 NL_ATTR_FOR_EACH (a, left, ctx->xout->odp_actions.data,
7283 ctx->xout->odp_actions.size) {
7284 enum ovs_action_attr type = nl_attr_type(a);
7285 struct ofport_dpif *ofport;
7287 if (type != OVS_ACTION_ATTR_OUTPUT) {
7291 ofport = get_odp_port(ofproto, nl_attr_get_u32(a));
7292 if (ofport && ofport->bundle) {
7293 mirrors |= ofport->bundle->dst_mirrors;
7301 /* Restore the original packet before adding the mirror actions. */
7302 ctx->xin->flow = *orig_flow;
7307 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
7309 if (!vlan_is_mirrored(m, vlan)) {
7310 mirrors = zero_rightmost_1bit(mirrors);
7314 mirrors &= ~m->dup_mirrors;
7315 ctx->xout->mirrors |= m->dup_mirrors;
7317 output_normal(ctx, m->out, vlan);
7318 } else if (vlan != m->out_vlan
7319 && !eth_addr_is_reserved(orig_flow->dl_dst)) {
7320 struct ofbundle *bundle;
7322 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
7323 if (ofbundle_includes_vlan(bundle, m->out_vlan)
7324 && !bundle->mirror_out) {
7325 output_normal(ctx, bundle, m->out_vlan);
7333 update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors,
7334 uint64_t packets, uint64_t bytes)
7340 for (; mirrors; mirrors = zero_rightmost_1bit(mirrors)) {
7343 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
7346 /* In normal circumstances 'm' will not be NULL. However,
7347 * if mirrors are reconfigured, we can temporarily get out
7348 * of sync in facet_revalidate(). We could "correct" the
7349 * mirror list before reaching here, but doing that would
7350 * not properly account the traffic stats we've currently
7351 * accumulated for previous mirror configuration. */
7355 m->packet_count += packets;
7356 m->byte_count += bytes;
7360 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
7361 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
7362 * indicate this; newer upstream kernels use gratuitous ARP requests. */
7364 is_gratuitous_arp(const struct flow *flow)
7366 return (flow->dl_type == htons(ETH_TYPE_ARP)
7367 && eth_addr_is_broadcast(flow->dl_dst)
7368 && (flow->nw_proto == ARP_OP_REPLY
7369 || (flow->nw_proto == ARP_OP_REQUEST
7370 && flow->nw_src == flow->nw_dst)));
7374 update_learning_table(struct ofproto_dpif *ofproto,
7375 const struct flow *flow, int vlan,
7376 struct ofbundle *in_bundle)
7378 struct mac_entry *mac;
7380 /* Don't learn the OFPP_NONE port. */
7381 if (in_bundle == &ofpp_none_bundle) {
7385 if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
7389 mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan);
7390 if (is_gratuitous_arp(flow)) {
7391 /* We don't want to learn from gratuitous ARP packets that are
7392 * reflected back over bond slaves so we lock the learning table. */
7393 if (!in_bundle->bond) {
7394 mac_entry_set_grat_arp_lock(mac);
7395 } else if (mac_entry_is_grat_arp_locked(mac)) {
7400 if (mac_entry_is_new(mac) || mac->port.p != in_bundle) {
7401 /* The log messages here could actually be useful in debugging,
7402 * so keep the rate limit relatively high. */
7403 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
7404 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
7405 "on port %s in VLAN %d",
7406 ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src),
7407 in_bundle->name, vlan);
7409 mac->port.p = in_bundle;
7410 tag_set_add(&ofproto->backer->revalidate_set,
7411 mac_learning_changed(ofproto->ml, mac));
7415 static struct ofbundle *
7416 lookup_input_bundle(const struct ofproto_dpif *ofproto, uint16_t in_port,
7417 bool warn, struct ofport_dpif **in_ofportp)
7419 struct ofport_dpif *ofport;
7421 /* Find the port and bundle for the received packet. */
7422 ofport = get_ofp_port(ofproto, in_port);
7424 *in_ofportp = ofport;
7426 if (ofport && ofport->bundle) {
7427 return ofport->bundle;
7430 /* Special-case OFPP_NONE, which a controller may use as the ingress
7431 * port for traffic that it is sourcing. */
7432 if (in_port == OFPP_NONE) {
7433 return &ofpp_none_bundle;
7436 /* Odd. A few possible reasons here:
7438 * - We deleted a port but there are still a few packets queued up
7441 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
7442 * we don't know about.
7444 * - The ofproto client didn't configure the port as part of a bundle.
7445 * This is particularly likely to happen if a packet was received on the
7446 * port after it was created, but before the client had a chance to
7447 * configure its bundle.
7450 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7452 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
7453 "port %"PRIu16, ofproto->up.name, in_port);
7458 /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or
7459 * dropped. Returns true if they may be forwarded, false if they should be
7462 * 'in_port' must be the ofport_dpif that corresponds to flow->in_port.
7463 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
7465 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
7466 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
7467 * checked by input_vid_is_valid().
7469 * May also add tags to '*tags', although the current implementation only does
7470 * so in one special case.
7473 is_admissible(struct xlate_ctx *ctx, struct ofport_dpif *in_port,
7476 struct ofproto_dpif *ofproto = ctx->ofproto;
7477 struct flow *flow = &ctx->xin->flow;
7478 struct ofbundle *in_bundle = in_port->bundle;
7480 /* Drop frames for reserved multicast addresses
7481 * only if forward_bpdu option is absent. */
7482 if (!ofproto->up.forward_bpdu && eth_addr_is_reserved(flow->dl_dst)) {
7483 xlate_report(ctx, "packet has reserved destination MAC, dropping");
7487 if (in_bundle->bond) {
7488 struct mac_entry *mac;
7490 switch (bond_check_admissibility(in_bundle->bond, in_port,
7491 flow->dl_dst, &ctx->xout->tags)) {
7496 xlate_report(ctx, "bonding refused admissibility, dropping");
7499 case BV_DROP_IF_MOVED:
7500 mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL);
7501 if (mac && mac->port.p != in_bundle &&
7502 (!is_gratuitous_arp(flow)
7503 || mac_entry_is_grat_arp_locked(mac))) {
7504 xlate_report(ctx, "SLB bond thinks this packet looped back, "
7516 xlate_normal(struct xlate_ctx *ctx)
7518 struct ofport_dpif *in_port;
7519 struct ofbundle *in_bundle;
7520 struct mac_entry *mac;
7524 ctx->xout->has_normal = true;
7526 in_bundle = lookup_input_bundle(ctx->ofproto, ctx->xin->flow.in_port,
7527 ctx->xin->packet != NULL, &in_port);
7529 xlate_report(ctx, "no input bundle, dropping");
7533 /* Drop malformed frames. */
7534 if (ctx->xin->flow.dl_type == htons(ETH_TYPE_VLAN) &&
7535 !(ctx->xin->flow.vlan_tci & htons(VLAN_CFI))) {
7536 if (ctx->xin->packet != NULL) {
7537 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7538 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
7539 "VLAN tag received on port %s",
7540 ctx->ofproto->up.name, in_bundle->name);
7542 xlate_report(ctx, "partial VLAN tag, dropping");
7546 /* Drop frames on bundles reserved for mirroring. */
7547 if (in_bundle->mirror_out) {
7548 if (ctx->xin->packet != NULL) {
7549 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7550 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
7551 "%s, which is reserved exclusively for mirroring",
7552 ctx->ofproto->up.name, in_bundle->name);
7554 xlate_report(ctx, "input port is mirror output port, dropping");
7559 vid = vlan_tci_to_vid(ctx->xin->flow.vlan_tci);
7560 if (!input_vid_is_valid(vid, in_bundle, ctx->xin->packet != NULL)) {
7561 xlate_report(ctx, "disallowed VLAN VID for this input port, dropping");
7564 vlan = input_vid_to_vlan(in_bundle, vid);
7566 /* Check other admissibility requirements. */
7567 if (in_port && !is_admissible(ctx, in_port, vlan)) {
7571 /* Learn source MAC. */
7572 if (ctx->xin->may_learn) {
7573 update_learning_table(ctx->ofproto, &ctx->xin->flow, vlan, in_bundle);
7576 /* Determine output bundle. */
7577 mac = mac_learning_lookup(ctx->ofproto->ml, ctx->xin->flow.dl_dst, vlan,
7580 if (mac->port.p != in_bundle) {
7581 xlate_report(ctx, "forwarding to learned port");
7582 output_normal(ctx, mac->port.p, vlan);
7584 xlate_report(ctx, "learned port is input port, dropping");
7587 struct ofbundle *bundle;
7589 xlate_report(ctx, "no learned MAC for destination, flooding");
7590 HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
7591 if (bundle != in_bundle
7592 && ofbundle_includes_vlan(bundle, vlan)
7593 && bundle->floodable
7594 && !bundle->mirror_out) {
7595 output_normal(ctx, bundle, vlan);
7598 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
7602 /* Optimized flow revalidation.
7604 * It's a difficult problem, in general, to tell which facets need to have
7605 * their actions recalculated whenever the OpenFlow flow table changes. We
7606 * don't try to solve that general problem: for most kinds of OpenFlow flow
7607 * table changes, we recalculate the actions for every facet. This is
7608 * relatively expensive, but it's good enough if the OpenFlow flow table
7609 * doesn't change very often.
7611 * However, we can expect one particular kind of OpenFlow flow table change to
7612 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
7613 * of CPU on revalidating every facet whenever MAC learning modifies the flow
7614 * table, we add a special case that applies to flow tables in which every rule
7615 * has the same form (that is, the same wildcards), except that the table is
7616 * also allowed to have a single "catch-all" flow that matches all packets. We
7617 * optimize this case by tagging all of the facets that resubmit into the table
7618 * and invalidating the same tag whenever a flow changes in that table. The
7619 * end result is that we revalidate just the facets that need it (and sometimes
7620 * a few more, but not all of the facets or even all of the facets that
7621 * resubmit to the table modified by MAC learning). */
7623 /* Calculates the tag to use for 'flow' and mask 'mask' when it is inserted
7624 * into an OpenFlow table with the given 'basis'. */
7626 rule_calculate_tag(const struct flow *flow, const struct minimask *mask,
7629 if (minimask_is_catchall(mask)) {
7632 uint32_t hash = flow_hash_in_minimask(flow, mask, secret);
7633 return tag_create_deterministic(hash);
7637 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
7638 * taggability of that table.
7640 * This function must be called after *each* change to a flow table. If you
7641 * skip calling it on some changes then the pointer comparisons at the end can
7642 * be invalid if you get unlucky. For example, if a flow removal causes a
7643 * cls_table to be destroyed and then a flow insertion causes a cls_table with
7644 * different wildcards to be created with the same address, then this function
7645 * will incorrectly skip revalidation. */
7647 table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
7649 struct table_dpif *table = &ofproto->tables[table_id];
7650 const struct oftable *oftable = &ofproto->up.tables[table_id];
7651 struct cls_table *catchall, *other;
7652 struct cls_table *t;
7654 catchall = other = NULL;
7656 switch (hmap_count(&oftable->cls.tables)) {
7658 /* We could tag this OpenFlow table but it would make the logic a
7659 * little harder and it's a corner case that doesn't seem worth it
7665 HMAP_FOR_EACH (t, hmap_node, &oftable->cls.tables) {
7666 if (cls_table_is_catchall(t)) {
7668 } else if (!other) {
7671 /* Indicate that we can't tag this by setting both tables to
7672 * NULL. (We know that 'catchall' is already NULL.) */
7679 /* Can't tag this table. */
7683 if (table->catchall_table != catchall || table->other_table != other) {
7684 table->catchall_table = catchall;
7685 table->other_table = other;
7686 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
7690 /* Given 'rule' that has changed in some way (either it is a rule being
7691 * inserted, a rule being deleted, or a rule whose actions are being
7692 * modified), marks facets for revalidation to ensure that packets will be
7693 * forwarded correctly according to the new state of the flow table.
7695 * This function must be called after *each* change to a flow table. See
7696 * the comment on table_update_taggable() for more information. */
7698 rule_invalidate(const struct rule_dpif *rule)
7700 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
7702 table_update_taggable(ofproto, rule->up.table_id);
7704 if (!ofproto->backer->need_revalidate) {
7705 struct table_dpif *table = &ofproto->tables[rule->up.table_id];
7707 if (table->other_table && rule->tag) {
7708 tag_set_add(&ofproto->backer->revalidate_set, rule->tag);
7710 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
7716 set_frag_handling(struct ofproto *ofproto_,
7717 enum ofp_config_flags frag_handling)
7719 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7720 if (frag_handling != OFPC_FRAG_REASM) {
7721 ofproto->backer->need_revalidate = REV_RECONFIGURE;
7729 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
7730 const struct flow *flow,
7731 const struct ofpact *ofpacts, size_t ofpacts_len)
7733 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7734 struct initial_vals initial_vals;
7735 struct odputil_keybuf keybuf;
7736 struct dpif_flow_stats stats;
7737 struct xlate_out xout;
7738 struct xlate_in xin;
7742 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
7743 odp_flow_key_from_flow(&key, flow,
7744 ofp_port_to_odp_port(ofproto, flow->in_port));
7746 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
7748 initial_vals.vlan_tci = flow->vlan_tci;
7749 xlate_in_init(&xin, ofproto, flow, &initial_vals, NULL, stats.tcp_flags,
7751 xin.resubmit_stats = &stats;
7752 xin.ofpacts_len = ofpacts_len;
7753 xin.ofpacts = ofpacts;
7755 xlate_actions(&xin, &xout);
7756 dpif_execute(ofproto->backer->dpif, key.data, key.size,
7757 xout.odp_actions.data, xout.odp_actions.size, packet);
7758 xlate_out_uninit(&xout);
7766 set_netflow(struct ofproto *ofproto_,
7767 const struct netflow_options *netflow_options)
7769 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7771 if (netflow_options) {
7772 if (!ofproto->netflow) {
7773 ofproto->netflow = netflow_create();
7775 return netflow_set_options(ofproto->netflow, netflow_options);
7777 netflow_destroy(ofproto->netflow);
7778 ofproto->netflow = NULL;
7784 get_netflow_ids(const struct ofproto *ofproto_,
7785 uint8_t *engine_type, uint8_t *engine_id)
7787 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7789 dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
7793 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
7795 if (!facet_is_controller_flow(facet) &&
7796 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
7797 struct subfacet *subfacet;
7798 struct ofexpired expired;
7800 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
7801 if (subfacet->path == SF_FAST_PATH) {
7802 struct dpif_flow_stats stats;
7804 subfacet_install(subfacet, &facet->xout.odp_actions, &stats);
7805 subfacet_update_stats(subfacet, &stats);
7809 expired.flow = facet->flow;
7810 expired.packet_count = facet->packet_count;
7811 expired.byte_count = facet->byte_count;
7812 expired.used = facet->used;
7813 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
7818 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
7820 struct facet *facet;
7822 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
7823 send_active_timeout(ofproto, facet);
7827 static struct ofproto_dpif *
7828 ofproto_dpif_lookup(const char *name)
7830 struct ofproto_dpif *ofproto;
7832 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
7833 hash_string(name, 0), &all_ofproto_dpifs) {
7834 if (!strcmp(ofproto->up.name, name)) {
7842 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
7843 const char *argv[], void *aux OVS_UNUSED)
7845 struct ofproto_dpif *ofproto;
7848 ofproto = ofproto_dpif_lookup(argv[1]);
7850 unixctl_command_reply_error(conn, "no such bridge");
7853 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
7855 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
7856 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
7860 unixctl_command_reply(conn, "table successfully flushed");
7864 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
7865 const char *argv[], void *aux OVS_UNUSED)
7867 struct ds ds = DS_EMPTY_INITIALIZER;
7868 const struct ofproto_dpif *ofproto;
7869 const struct mac_entry *e;
7871 ofproto = ofproto_dpif_lookup(argv[1]);
7873 unixctl_command_reply_error(conn, "no such bridge");
7877 ds_put_cstr(&ds, " port VLAN MAC Age\n");
7878 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
7879 struct ofbundle *bundle = e->port.p;
7880 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
7881 ofbundle_get_a_port(bundle)->odp_port,
7882 e->vlan, ETH_ADDR_ARGS(e->mac),
7883 mac_entry_age(ofproto->ml, e));
7885 unixctl_command_reply(conn, ds_cstr(&ds));
7890 struct xlate_out xout;
7891 struct xlate_in xin;
7897 trace_format_rule(struct ds *result, uint8_t table_id, int level,
7898 const struct rule_dpif *rule)
7900 ds_put_char_multiple(result, '\t', level);
7902 ds_put_cstr(result, "No match\n");
7906 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
7907 table_id, ntohll(rule->up.flow_cookie));
7908 cls_rule_format(&rule->up.cr, result);
7909 ds_put_char(result, '\n');
7911 ds_put_char_multiple(result, '\t', level);
7912 ds_put_cstr(result, "OpenFlow ");
7913 ofpacts_format(rule->up.ofpacts, rule->up.ofpacts_len, result);
7914 ds_put_char(result, '\n');
7918 trace_format_flow(struct ds *result, int level, const char *title,
7919 struct trace_ctx *trace)
7921 ds_put_char_multiple(result, '\t', level);
7922 ds_put_format(result, "%s: ", title);
7923 if (flow_equal(&trace->xin.flow, &trace->flow)) {
7924 ds_put_cstr(result, "unchanged");
7926 flow_format(result, &trace->xin.flow);
7927 trace->flow = trace->xin.flow;
7929 ds_put_char(result, '\n');
7933 trace_format_regs(struct ds *result, int level, const char *title,
7934 struct trace_ctx *trace)
7938 ds_put_char_multiple(result, '\t', level);
7939 ds_put_format(result, "%s:", title);
7940 for (i = 0; i < FLOW_N_REGS; i++) {
7941 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
7943 ds_put_char(result, '\n');
7947 trace_format_odp(struct ds *result, int level, const char *title,
7948 struct trace_ctx *trace)
7950 struct ofpbuf *odp_actions = &trace->xout.odp_actions;
7952 ds_put_char_multiple(result, '\t', level);
7953 ds_put_format(result, "%s: ", title);
7954 format_odp_actions(result, odp_actions->data, odp_actions->size);
7955 ds_put_char(result, '\n');
7959 trace_resubmit(struct xlate_ctx *ctx, struct rule_dpif *rule)
7961 struct trace_ctx *trace = CONTAINER_OF(ctx->xin, struct trace_ctx, xin);
7962 struct ds *result = trace->result;
7964 ds_put_char(result, '\n');
7965 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
7966 trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace);
7967 trace_format_odp(result, ctx->recurse + 1, "Resubmitted odp", trace);
7968 trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule);
7972 trace_report(struct xlate_ctx *ctx, const char *s)
7974 struct trace_ctx *trace = CONTAINER_OF(ctx->xin, struct trace_ctx, xin);
7975 struct ds *result = trace->result;
7977 ds_put_char_multiple(result, '\t', ctx->recurse);
7978 ds_put_cstr(result, s);
7979 ds_put_char(result, '\n');
7983 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
7984 void *aux OVS_UNUSED)
7986 const struct dpif_backer *backer;
7987 struct ofproto_dpif *ofproto;
7988 struct ofpbuf odp_key;
7989 struct ofpbuf *packet;
7990 struct initial_vals initial_vals;
7998 ofpbuf_init(&odp_key, 0);
8000 /* Handle "-generate" or a hex string as the last argument. */
8001 if (!strcmp(argv[argc - 1], "-generate")) {
8002 packet = ofpbuf_new(0);
8005 const char *error = eth_from_hex(argv[argc - 1], &packet);
8008 } else if (argc == 4) {
8009 /* The 3-argument form must end in "-generate' or a hex string. */
8010 unixctl_command_reply_error(conn, error);
8015 /* Parse the flow and determine whether a datapath or
8016 * bridge is specified. If function odp_flow_key_from_string()
8017 * returns 0, the flow is a odp_flow. If function
8018 * parse_ofp_exact_flow() returns 0, the flow is a br_flow. */
8019 if (!odp_flow_key_from_string(argv[argc - 1], NULL, &odp_key)) {
8020 /* If the odp_flow is the second argument,
8021 * the datapath name is the first argument. */
8023 const char *dp_type;
8024 if (!strncmp(argv[1], "ovs-", 4)) {
8025 dp_type = argv[1] + 4;
8029 backer = shash_find_data(&all_dpif_backers, dp_type);
8031 unixctl_command_reply_error(conn, "Cannot find datapath "
8036 /* No datapath name specified, so there should be only one
8038 struct shash_node *node;
8039 if (shash_count(&all_dpif_backers) != 1) {
8040 unixctl_command_reply_error(conn, "Must specify datapath "
8041 "name, there is more than one type of datapath");
8044 node = shash_first(&all_dpif_backers);
8045 backer = node->data;
8048 /* Extract the ofproto_dpif object from the ofproto_receive()
8050 if (ofproto_receive(backer, NULL, odp_key.data,
8051 odp_key.size, &flow, NULL, &ofproto, NULL,
8053 unixctl_command_reply_error(conn, "Invalid datapath flow");
8056 ds_put_format(&result, "Bridge: %s\n", ofproto->up.name);
8057 } else if (!parse_ofp_exact_flow(&flow, argv[argc - 1])) {
8059 unixctl_command_reply_error(conn, "Must specify bridge name");
8063 ofproto = ofproto_dpif_lookup(argv[1]);
8065 unixctl_command_reply_error(conn, "Unknown bridge name");
8068 initial_vals.vlan_tci = flow.vlan_tci;
8070 unixctl_command_reply_error(conn, "Bad flow syntax");
8074 /* Generate a packet, if requested. */
8076 if (!packet->size) {
8077 flow_compose(packet, &flow);
8079 ds_put_cstr(&result, "Packet: ");
8080 s = ofp_packet_to_string(packet->data, packet->size);
8081 ds_put_cstr(&result, s);
8084 /* Use the metadata from the flow and the packet argument
8085 * to reconstruct the flow. */
8086 flow_extract(packet, flow.skb_priority, flow.skb_mark, NULL,
8087 flow.in_port, &flow);
8088 initial_vals.vlan_tci = flow.vlan_tci;
8092 ofproto_trace(ofproto, &flow, packet, &initial_vals, &result);
8093 unixctl_command_reply(conn, ds_cstr(&result));
8096 ds_destroy(&result);
8097 ofpbuf_delete(packet);
8098 ofpbuf_uninit(&odp_key);
8102 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
8103 const struct ofpbuf *packet,
8104 const struct initial_vals *initial_vals, struct ds *ds)
8106 struct rule_dpif *rule;
8108 ds_put_cstr(ds, "Flow: ");
8109 flow_format(ds, flow);
8110 ds_put_char(ds, '\n');
8112 rule = rule_dpif_lookup(ofproto, flow);
8114 trace_format_rule(ds, 0, 0, rule);
8115 if (rule == ofproto->miss_rule) {
8116 ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
8117 } else if (rule == ofproto->no_packet_in_rule) {
8118 ds_put_cstr(ds, "\nNo match, packets dropped because "
8119 "OFPPC_NO_PACKET_IN is set on in_port.\n");
8123 uint64_t odp_actions_stub[1024 / 8];
8124 struct ofpbuf odp_actions;
8126 struct trace_ctx trace;
8129 tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
8132 ofpbuf_use_stub(&odp_actions,
8133 odp_actions_stub, sizeof odp_actions_stub);
8134 xlate_in_init(&trace.xin, ofproto, flow, initial_vals, rule, tcp_flags,
8136 trace.xin.resubmit_hook = trace_resubmit;
8137 trace.xin.report_hook = trace_report;
8138 xlate_actions(&trace.xin, &trace.xout);
8140 ds_put_char(ds, '\n');
8141 trace_format_flow(ds, 0, "Final flow", &trace);
8142 ds_put_cstr(ds, "Datapath actions: ");
8143 format_odp_actions(ds, trace.xout.odp_actions.data,
8144 trace.xout.odp_actions.size);
8146 if (trace.xout.slow) {
8147 ds_put_cstr(ds, "\nThis flow is handled by the userspace "
8148 "slow path because it:");
8149 switch (trace.xout.slow) {
8151 ds_put_cstr(ds, "\n\t- Consists of CFM packets.");
8154 ds_put_cstr(ds, "\n\t- Consists of LACP packets.");
8157 ds_put_cstr(ds, "\n\t- Consists of STP packets.");
8160 ds_put_cstr(ds, "\n\t- Consists of BFD packets.");
8162 case SLOW_CONTROLLER:
8163 ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages "
8164 "to the OpenFlow controller.");
8171 xlate_out_uninit(&trace.xout);
8176 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
8177 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
8180 unixctl_command_reply(conn, NULL);
8184 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
8185 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
8188 unixctl_command_reply(conn, NULL);
8191 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
8192 * 'reply' describing the results. */
8194 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
8196 struct facet *facet;
8200 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
8201 if (!facet_check_consistency(facet)) {
8206 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
8210 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
8211 ofproto->up.name, errors);
8213 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
8218 ofproto_dpif_self_check(struct unixctl_conn *conn,
8219 int argc, const char *argv[], void *aux OVS_UNUSED)
8221 struct ds reply = DS_EMPTY_INITIALIZER;
8222 struct ofproto_dpif *ofproto;
8225 ofproto = ofproto_dpif_lookup(argv[1]);
8227 unixctl_command_reply_error(conn, "Unknown ofproto (use "
8228 "ofproto/list for help)");
8231 ofproto_dpif_self_check__(ofproto, &reply);
8233 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
8234 ofproto_dpif_self_check__(ofproto, &reply);
8238 unixctl_command_reply(conn, ds_cstr(&reply));
8242 /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list
8243 * of the 'ofproto_shash' nodes. It is the responsibility of the caller
8244 * to destroy 'ofproto_shash' and free the returned value. */
8245 static const struct shash_node **
8246 get_ofprotos(struct shash *ofproto_shash)
8248 const struct ofproto_dpif *ofproto;
8250 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
8251 char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
8252 shash_add_nocopy(ofproto_shash, name, ofproto);
8255 return shash_sort(ofproto_shash);
8259 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
8260 const char *argv[] OVS_UNUSED,
8261 void *aux OVS_UNUSED)
8263 struct ds ds = DS_EMPTY_INITIALIZER;
8264 struct shash ofproto_shash;
8265 const struct shash_node **sorted_ofprotos;
8268 shash_init(&ofproto_shash);
8269 sorted_ofprotos = get_ofprotos(&ofproto_shash);
8270 for (i = 0; i < shash_count(&ofproto_shash); i++) {
8271 const struct shash_node *node = sorted_ofprotos[i];
8272 ds_put_format(&ds, "%s\n", node->name);
8275 shash_destroy(&ofproto_shash);
8276 free(sorted_ofprotos);
8278 unixctl_command_reply(conn, ds_cstr(&ds));
8283 show_dp_format(const struct ofproto_dpif *ofproto, struct ds *ds)
8285 const struct shash_node **ports;
8287 struct avg_subfacet_rates lifetime;
8288 unsigned long long int minutes;
8289 const int min_ms = 60 * 1000; /* milliseconds in one minute. */
8291 minutes = (time_msec() - ofproto->created) / min_ms;
8294 lifetime.add_rate = (double)ofproto->total_subfacet_add_count
8296 lifetime.del_rate = (double)ofproto->total_subfacet_del_count
8299 lifetime.add_rate = 0.0;
8300 lifetime.del_rate = 0.0;
8303 ds_put_format(ds, "%s (%s):\n", ofproto->up.name,
8304 dpif_name(ofproto->backer->dpif));
8306 "\tlookups: hit:%"PRIu64" missed:%"PRIu64"\n",
8307 ofproto->n_hit, ofproto->n_missed);
8308 ds_put_format(ds, "\tflows: cur: %zu, avg: %5.3f, max: %d,"
8309 " life span: %llu(ms)\n",
8310 hmap_count(&ofproto->subfacets),
8311 avg_subfacet_count(ofproto),
8312 ofproto->max_n_subfacet,
8313 avg_subfacet_life_span(ofproto));
8314 if (minutes >= 60) {
8315 show_dp_rates(ds, "\t\thourly avg:", &ofproto->hourly);
8317 if (minutes >= 60 * 24) {
8318 show_dp_rates(ds, "\t\tdaily avg:", &ofproto->daily);
8320 show_dp_rates(ds, "\t\toverall avg:", &lifetime);
8322 ports = shash_sort(&ofproto->up.port_by_name);
8323 for (i = 0; i < shash_count(&ofproto->up.port_by_name); i++) {
8324 const struct shash_node *node = ports[i];
8325 struct ofport *ofport = node->data;
8326 const char *name = netdev_get_name(ofport->netdev);
8327 const char *type = netdev_get_type(ofport->netdev);
8330 ds_put_format(ds, "\t%s %u/", name, ofport->ofp_port);
8332 odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
8333 if (odp_port != OVSP_NONE) {
8334 ds_put_format(ds, "%"PRIu32":", odp_port);
8336 ds_put_cstr(ds, "none:");
8339 if (strcmp(type, "system")) {
8340 struct netdev *netdev;
8343 ds_put_format(ds, " (%s", type);
8345 error = netdev_open(name, type, &netdev);
8350 error = netdev_get_config(netdev, &config);
8352 const struct smap_node **nodes;
8355 nodes = smap_sort(&config);
8356 for (i = 0; i < smap_count(&config); i++) {
8357 const struct smap_node *node = nodes[i];
8358 ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
8359 node->key, node->value);
8363 smap_destroy(&config);
8365 netdev_close(netdev);
8367 ds_put_char(ds, ')');
8369 ds_put_char(ds, '\n');
8375 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc,
8376 const char *argv[], void *aux OVS_UNUSED)
8378 struct ds ds = DS_EMPTY_INITIALIZER;
8379 const struct ofproto_dpif *ofproto;
8383 for (i = 1; i < argc; i++) {
8384 ofproto = ofproto_dpif_lookup(argv[i]);
8386 ds_put_format(&ds, "Unknown bridge %s (use dpif/dump-dps "
8387 "for help)", argv[i]);
8388 unixctl_command_reply_error(conn, ds_cstr(&ds));
8391 show_dp_format(ofproto, &ds);
8394 struct shash ofproto_shash;
8395 const struct shash_node **sorted_ofprotos;
8398 shash_init(&ofproto_shash);
8399 sorted_ofprotos = get_ofprotos(&ofproto_shash);
8400 for (i = 0; i < shash_count(&ofproto_shash); i++) {
8401 const struct shash_node *node = sorted_ofprotos[i];
8402 show_dp_format(node->data, &ds);
8405 shash_destroy(&ofproto_shash);
8406 free(sorted_ofprotos);
8409 unixctl_command_reply(conn, ds_cstr(&ds));
8414 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
8415 int argc OVS_UNUSED, const char *argv[],
8416 void *aux OVS_UNUSED)
8418 struct ds ds = DS_EMPTY_INITIALIZER;
8419 const struct ofproto_dpif *ofproto;
8420 struct subfacet *subfacet;
8422 ofproto = ofproto_dpif_lookup(argv[1]);
8424 unixctl_command_reply_error(conn, "no such bridge");
8428 update_stats(ofproto->backer);
8430 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
8431 struct facet *facet = subfacet->facet;
8433 odp_flow_key_format(subfacet->key, subfacet->key_len, &ds);
8435 ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:",
8436 subfacet->dp_packet_count, subfacet->dp_byte_count);
8437 if (subfacet->used) {
8438 ds_put_format(&ds, "%.3fs",
8439 (time_msec() - subfacet->used) / 1000.0);
8441 ds_put_format(&ds, "never");
8443 if (subfacet->facet->tcp_flags) {
8444 ds_put_cstr(&ds, ", flags:");
8445 packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags);
8448 ds_put_cstr(&ds, ", actions:");
8449 if (facet->xout.slow) {
8450 uint64_t slow_path_stub[128 / 8];
8451 const struct nlattr *actions;
8454 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
8455 slow_path_stub, sizeof slow_path_stub,
8456 &actions, &actions_len);
8457 format_odp_actions(&ds, actions, actions_len);
8459 format_odp_actions(&ds, facet->xout.odp_actions.data,
8460 facet->xout.odp_actions.size);
8462 ds_put_char(&ds, '\n');
8465 unixctl_command_reply(conn, ds_cstr(&ds));
8470 ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn,
8471 int argc OVS_UNUSED, const char *argv[],
8472 void *aux OVS_UNUSED)
8474 struct ds ds = DS_EMPTY_INITIALIZER;
8475 struct ofproto_dpif *ofproto;
8477 ofproto = ofproto_dpif_lookup(argv[1]);
8479 unixctl_command_reply_error(conn, "no such bridge");
8483 flush(&ofproto->up);
8485 unixctl_command_reply(conn, ds_cstr(&ds));
8490 ofproto_dpif_unixctl_init(void)
8492 static bool registered;
8498 unixctl_command_register(
8500 "[dp_name]|bridge odp_flow|br_flow [-generate|packet]",
8501 1, 3, ofproto_unixctl_trace, NULL);
8502 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
8503 ofproto_unixctl_fdb_flush, NULL);
8504 unixctl_command_register("fdb/show", "bridge", 1, 1,
8505 ofproto_unixctl_fdb_show, NULL);
8506 unixctl_command_register("ofproto/clog", "", 0, 0,
8507 ofproto_dpif_clog, NULL);
8508 unixctl_command_register("ofproto/unclog", "", 0, 0,
8509 ofproto_dpif_unclog, NULL);
8510 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
8511 ofproto_dpif_self_check, NULL);
8512 unixctl_command_register("dpif/dump-dps", "", 0, 0,
8513 ofproto_unixctl_dpif_dump_dps, NULL);
8514 unixctl_command_register("dpif/show", "[bridge]", 0, INT_MAX,
8515 ofproto_unixctl_dpif_show, NULL);
8516 unixctl_command_register("dpif/dump-flows", "bridge", 1, 1,
8517 ofproto_unixctl_dpif_dump_flows, NULL);
8518 unixctl_command_register("dpif/del-flows", "bridge", 1, 1,
8519 ofproto_unixctl_dpif_del_flows, NULL);
8522 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
8524 * This is deprecated. It is only for compatibility with broken device drivers
8525 * in old versions of Linux that do not properly support VLANs when VLAN
8526 * devices are not used. When broken device drivers are no longer in
8527 * widespread use, we will delete these interfaces. */
8530 set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid)
8532 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
8533 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
8535 if (realdev_ofp_port == ofport->realdev_ofp_port
8536 && vid == ofport->vlandev_vid) {
8540 ofproto->backer->need_revalidate = REV_RECONFIGURE;
8542 if (ofport->realdev_ofp_port) {
8545 if (realdev_ofp_port && ofport->bundle) {
8546 /* vlandevs are enslaved to their realdevs, so they are not allowed to
8547 * themselves be part of a bundle. */
8548 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
8551 ofport->realdev_ofp_port = realdev_ofp_port;
8552 ofport->vlandev_vid = vid;
8554 if (realdev_ofp_port) {
8555 vsp_add(ofport, realdev_ofp_port, vid);
8562 hash_realdev_vid(uint16_t realdev_ofp_port, int vid)
8564 return hash_2words(realdev_ofp_port, vid);
8567 /* Returns the OFP port number of the Linux VLAN device that corresponds to
8568 * 'vlan_tci' on the network device with port number 'realdev_ofp_port' in
8569 * 'struct ofport_dpif'. For example, given 'realdev_ofp_port' of eth0 and
8570 * 'vlan_tci' 9, it would return the port number of eth0.9.
8572 * Unless VLAN splinters are enabled for port 'realdev_ofp_port', this
8573 * function just returns its 'realdev_ofp_port' argument. */
8575 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
8576 uint16_t realdev_ofp_port, ovs_be16 vlan_tci)
8578 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
8579 int vid = vlan_tci_to_vid(vlan_tci);
8580 const struct vlan_splinter *vsp;
8582 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
8583 hash_realdev_vid(realdev_ofp_port, vid),
8584 &ofproto->realdev_vid_map) {
8585 if (vsp->realdev_ofp_port == realdev_ofp_port
8586 && vsp->vid == vid) {
8587 return vsp->vlandev_ofp_port;
8591 return realdev_ofp_port;
8594 static struct vlan_splinter *
8595 vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port)
8597 struct vlan_splinter *vsp;
8599 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, hash_int(vlandev_ofp_port, 0),
8600 &ofproto->vlandev_map) {
8601 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
8609 /* Returns the OpenFlow port number of the "real" device underlying the Linux
8610 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
8611 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
8612 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
8613 * eth0 and store 9 in '*vid'.
8615 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
8616 * VLAN device. Unless VLAN splinters are enabled, this is what this function
8619 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
8620 uint16_t vlandev_ofp_port, int *vid)
8622 if (!hmap_is_empty(&ofproto->vlandev_map)) {
8623 const struct vlan_splinter *vsp;
8625 vsp = vlandev_find(ofproto, vlandev_ofp_port);
8630 return vsp->realdev_ofp_port;
8636 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
8637 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
8638 * 'flow->in_port' to the "real" device backing the VLAN device, sets
8639 * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is
8640 * always the case unless VLAN splinters are enabled), returns false without
8641 * making any changes. */
8643 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
8648 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port, &vid);
8653 /* Cause the flow to be processed as if it came in on the real device with
8654 * the VLAN device's VLAN ID. */
8655 flow->in_port = realdev;
8656 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
8661 vsp_remove(struct ofport_dpif *port)
8663 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
8664 struct vlan_splinter *vsp;
8666 vsp = vlandev_find(ofproto, port->up.ofp_port);
8668 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
8669 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
8672 port->realdev_ofp_port = 0;
8674 VLOG_ERR("missing vlan device record");
8679 vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid)
8681 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
8683 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
8684 && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid))
8685 == realdev_ofp_port)) {
8686 struct vlan_splinter *vsp;
8688 vsp = xmalloc(sizeof *vsp);
8689 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
8690 hash_int(port->up.ofp_port, 0));
8691 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
8692 hash_realdev_vid(realdev_ofp_port, vid));
8693 vsp->realdev_ofp_port = realdev_ofp_port;
8694 vsp->vlandev_ofp_port = port->up.ofp_port;
8697 port->realdev_ofp_port = realdev_ofp_port;
8699 VLOG_ERR("duplicate vlan device record");
8704 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
8706 const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
8707 return ofport ? ofport->odp_port : OVSP_NONE;
8710 static struct ofport_dpif *
8711 odp_port_to_ofport(const struct dpif_backer *backer, uint32_t odp_port)
8713 struct ofport_dpif *port;
8715 HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node,
8716 hash_int(odp_port, 0),
8717 &backer->odp_to_ofport_map) {
8718 if (port->odp_port == odp_port) {
8727 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
8729 struct ofport_dpif *port;
8731 port = odp_port_to_ofport(ofproto->backer, odp_port);
8732 if (port && &ofproto->up == port->up.ofproto) {
8733 return port->up.ofp_port;
8738 static unsigned long long int
8739 avg_subfacet_life_span(const struct ofproto_dpif *ofproto)
8741 unsigned long long int dc;
8742 unsigned long long int avg;
8744 dc = ofproto->total_subfacet_del_count + ofproto->subfacet_del_count;
8745 avg = dc ? ofproto->total_subfacet_life_span / dc : 0;
8751 avg_subfacet_count(const struct ofproto_dpif *ofproto)
8755 if (ofproto->n_update_stats) {
8756 avg_c = (double)ofproto->total_subfacet_count
8757 / ofproto->n_update_stats;
8764 show_dp_rates(struct ds *ds, const char *heading,
8765 const struct avg_subfacet_rates *rates)
8767 ds_put_format(ds, "%s add rate: %5.3f/min, del rate: %5.3f/min\n",
8768 heading, rates->add_rate, rates->del_rate);
8772 update_max_subfacet_count(struct ofproto_dpif *ofproto)
8774 ofproto->max_n_subfacet = MAX(ofproto->max_n_subfacet,
8775 hmap_count(&ofproto->subfacets));
8778 /* Compute exponentially weighted moving average, adding 'new' as the newest,
8779 * most heavily weighted element. 'base' designates the rate of decay: after
8780 * 'base' further updates, 'new''s weight in the EWMA decays to about 1/e
8783 exp_mavg(double *avg, int base, double new)
8785 *avg = (*avg * (base - 1) + new) / base;
8789 update_moving_averages(struct ofproto_dpif *ofproto)
8791 const int min_ms = 60 * 1000; /* milliseconds in one minute. */
8793 /* Update hourly averages on the minute boundaries. */
8794 if (time_msec() - ofproto->last_minute >= min_ms) {
8795 exp_mavg(&ofproto->hourly.add_rate, 60, ofproto->subfacet_add_count);
8796 exp_mavg(&ofproto->hourly.del_rate, 60, ofproto->subfacet_del_count);
8798 /* Update daily averages on the hour boundaries. */
8799 if ((ofproto->last_minute - ofproto->created) / min_ms % 60 == 59) {
8800 exp_mavg(&ofproto->daily.add_rate, 24, ofproto->hourly.add_rate);
8801 exp_mavg(&ofproto->daily.del_rate, 24, ofproto->hourly.del_rate);
8804 ofproto->total_subfacet_add_count += ofproto->subfacet_add_count;
8805 ofproto->total_subfacet_del_count += ofproto->subfacet_del_count;
8806 ofproto->subfacet_add_count = 0;
8807 ofproto->subfacet_del_count = 0;
8808 ofproto->last_minute += min_ms;
8812 const struct ofproto_class ofproto_dpif_class = {
8847 port_is_lacp_current,
8848 NULL, /* rule_choose_table */
8855 rule_modify_actions,
8869 get_stp_port_status,
8876 is_mirror_output_bundle,
8877 forward_bpdu_changed,
8878 set_mac_table_config,