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"
44 #include "odp-execute.h"
47 #include "ofp-actions.h"
48 #include "ofp-parse.h"
49 #include "ofp-print.h"
50 #include "ofproto-dpif-governor.h"
51 #include "ofproto-dpif-ipfix.h"
52 #include "ofproto-dpif-sflow.h"
53 #include "poll-loop.h"
58 #include "unaligned.h"
60 #include "vlan-bitmap.h"
63 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
65 COVERAGE_DEFINE(ofproto_dpif_expired);
66 COVERAGE_DEFINE(ofproto_dpif_xlate);
67 COVERAGE_DEFINE(facet_changed_rule);
68 COVERAGE_DEFINE(facet_revalidate);
69 COVERAGE_DEFINE(facet_unexpected);
70 COVERAGE_DEFINE(facet_suppress);
72 /* Maximum depth of flow table recursion (due to resubmit actions) in a
73 * flow translation. */
74 #define MAX_RESUBMIT_RECURSION 64
76 /* Number of implemented OpenFlow tables. */
77 enum { N_TABLES = 255 };
78 enum { TBL_INTERNAL = N_TABLES - 1 }; /* Used for internal hidden rules. */
79 BUILD_ASSERT_DECL(N_TABLES >= 2 && N_TABLES <= 255);
91 * - Do include packets and bytes from facets that have been deleted or
92 * whose own statistics have been folded into the rule.
94 * - Do include packets and bytes sent "by hand" that were accounted to
95 * the rule without any facet being involved (this is a rare corner
96 * case in rule_execute()).
98 * - Do not include packet or bytes that can be obtained from any facet's
99 * packet_count or byte_count member or that can be obtained from the
100 * datapath by, e.g., dpif_flow_get() for any subfacet.
102 uint64_t packet_count; /* Number of packets received. */
103 uint64_t byte_count; /* Number of bytes received. */
105 tag_type tag; /* Caches rule_calculate_tag() result. */
107 struct list facets; /* List of "struct facet"s. */
110 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
112 return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
115 static struct rule_dpif *rule_dpif_lookup(struct ofproto_dpif *,
116 const struct flow *);
117 static struct rule_dpif *rule_dpif_lookup__(struct ofproto_dpif *,
120 static struct rule_dpif *rule_dpif_miss_rule(struct ofproto_dpif *ofproto,
121 const struct flow *flow);
123 static void rule_get_stats(struct rule *, uint64_t *packets, uint64_t *bytes);
124 static void rule_credit_stats(struct rule_dpif *,
125 const struct dpif_flow_stats *);
126 static tag_type rule_calculate_tag(const struct flow *,
127 const struct minimask *, uint32_t basis);
128 static void rule_invalidate(const struct rule_dpif *);
130 #define MAX_MIRRORS 32
131 typedef uint32_t mirror_mask_t;
132 #define MIRROR_MASK_C(X) UINT32_C(X)
133 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
135 struct ofproto_dpif *ofproto; /* Owning ofproto. */
136 size_t idx; /* In ofproto's "mirrors" array. */
137 void *aux; /* Key supplied by ofproto's client. */
138 char *name; /* Identifier for log messages. */
140 /* Selection criteria. */
141 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
142 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
143 unsigned long *vlans; /* Bitmap of chosen VLANs, NULL selects all. */
145 /* Output (exactly one of out == NULL and out_vlan == -1 is true). */
146 struct ofbundle *out; /* Output port or NULL. */
147 int out_vlan; /* Output VLAN or -1. */
148 mirror_mask_t dup_mirrors; /* Bitmap of mirrors with the same output. */
151 int64_t packet_count; /* Number of packets sent. */
152 int64_t byte_count; /* Number of bytes sent. */
155 static void mirror_destroy(struct ofmirror *);
156 static void update_mirror_stats(struct ofproto_dpif *ofproto,
157 mirror_mask_t mirrors,
158 uint64_t packets, uint64_t bytes);
161 struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
162 struct ofproto_dpif *ofproto; /* Owning ofproto. */
163 void *aux; /* Key supplied by ofproto's client. */
164 char *name; /* Identifier for log messages. */
167 struct list ports; /* Contains "struct ofport"s. */
168 enum port_vlan_mode vlan_mode; /* VLAN mode */
169 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
170 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
171 * NULL if all VLANs are trunked. */
172 struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */
173 struct bond *bond; /* Nonnull iff more than one port. */
174 bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */
177 bool floodable; /* True if no port has OFPUTIL_PC_NO_FLOOD set. */
179 /* Port mirroring info. */
180 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
181 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
182 mirror_mask_t mirror_out; /* Mirrors that output to this bundle. */
185 static void bundle_remove(struct ofport *);
186 static void bundle_update(struct ofbundle *);
187 static void bundle_destroy(struct ofbundle *);
188 static void bundle_del_port(struct ofport_dpif *);
189 static void bundle_run(struct ofbundle *);
190 static void bundle_wait(struct ofbundle *);
191 static struct ofbundle *lookup_input_bundle(const struct ofproto_dpif *,
192 uint16_t in_port, bool warn,
193 struct ofport_dpif **in_ofportp);
195 /* A controller may use OFPP_NONE as the ingress port to indicate that
196 * it did not arrive on a "real" port. 'ofpp_none_bundle' exists for
197 * when an input bundle is needed for validation (e.g., mirroring or
198 * OFPP_NORMAL processing). It is not connected to an 'ofproto' or have
199 * any 'port' structs, so care must be taken when dealing with it. */
200 static struct ofbundle ofpp_none_bundle = {
202 .vlan_mode = PORT_VLAN_TRUNK
205 static void stp_run(struct ofproto_dpif *ofproto);
206 static void stp_wait(struct ofproto_dpif *ofproto);
207 static int set_stp_port(struct ofport *,
208 const struct ofproto_port_stp_settings *);
210 static bool ofbundle_includes_vlan(const struct ofbundle *, uint16_t vlan);
214 /* Initial values of fields of the packet that may be changed during
215 * flow processing and needed later. */
216 struct initial_vals {
217 /* This is the value of vlan_tci in the packet as actually received from
218 * dpif. This is the same as the facet's flow.vlan_tci unless the packet
219 * was received via a VLAN splinter. In that case, this value is 0
220 * (because the packet as actually received from the dpif had no 802.1Q
221 * tag) but the facet's flow.vlan_tci is set to the VLAN that the splinter
224 * This member should be removed when the VLAN splinters feature is no
230 tag_type tags; /* Tags associated with actions. */
231 enum slow_path_reason slow; /* 0 if fast path may be used. */
232 bool has_learn; /* Actions include NXAST_LEARN? */
233 bool has_normal; /* Actions output to OFPP_NORMAL? */
234 bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */
235 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
236 mirror_mask_t mirrors; /* Bitmap of associated mirrors. */
238 uint64_t odp_actions_stub[256 / 8];
239 struct ofpbuf odp_actions;
243 struct ofproto_dpif *ofproto;
245 /* Flow to which the OpenFlow actions apply. xlate_actions() will modify
246 * this flow when actions change header fields. */
249 struct initial_vals initial_vals;
251 /* The packet corresponding to 'flow', or a null pointer if we are
252 * revalidating without a packet to refer to. */
253 const struct ofpbuf *packet;
255 /* Should OFPP_NORMAL update the MAC learning table? Should "learn"
256 * actions update the flow table?
258 * We want to update these tables if we are actually processing a packet,
259 * or if we are accounting for packets that the datapath has processed, but
260 * not if we are just revalidating. */
263 /* The rule initiating translation or NULL. */
264 struct rule_dpif *rule;
266 /* The actions to translate. If 'rule' is not NULL, these may be NULL. */
267 const struct ofpact *ofpacts;
270 /* Union of the set of TCP flags seen so far in this flow. (Used only by
271 * NXAST_FIN_TIMEOUT. Set to zero to avoid updating updating rules'
275 /* If nonnull, flow translation calls this function just before executing a
276 * resubmit or OFPP_TABLE action. In addition, disables logging of traces
277 * when the recursion depth is exceeded.
279 * 'rule' is the rule being submitted into. It will be null if the
280 * resubmit or OFPP_TABLE action didn't find a matching rule.
282 * This is normally null so the client has to set it manually after
283 * calling xlate_in_init(). */
284 void (*resubmit_hook)(struct xlate_ctx *, struct rule_dpif *rule);
286 /* If nonnull, flow translation calls this function to report some
287 * significant decision, e.g. to explain why OFPP_NORMAL translation
288 * dropped a packet. */
289 void (*report_hook)(struct xlate_ctx *, const char *s);
291 /* If nonnull, flow translation credits the specified statistics to each
292 * rule reached through a resubmit or OFPP_TABLE action.
294 * This is normally null so the client has to set it manually after
295 * calling xlate_in_init(). */
296 const struct dpif_flow_stats *resubmit_stats;
299 /* Context used by xlate_actions() and its callees. */
301 struct xlate_in *xin;
302 struct xlate_out *xout;
304 struct ofproto_dpif *ofproto;
306 /* Flow at the last commit. */
307 struct flow base_flow;
309 /* Tunnel IP destination address as received. This is stored separately
310 * as the base_flow.tunnel is cleared on init to reflect the datapath
311 * behavior. Used to make sure not to send tunneled output to ourselves,
312 * which might lead to an infinite loop. This could happen easily
313 * if a tunnel is marked as 'ip_remote=flow', and the flow does not
314 * actually set the tun_dst field. */
315 ovs_be32 orig_tunnel_ip_dst;
317 /* Stack for the push and pop actions. Each stack element is of type
318 * "union mf_subvalue". */
319 union mf_subvalue init_stack[1024 / sizeof(union mf_subvalue)];
322 /* The rule that we are currently translating, or NULL. */
323 struct rule_dpif *rule;
325 int recurse; /* Recursion level, via xlate_table_action. */
326 bool max_resubmit_trigger; /* Recursed too deeply during translation. */
327 uint32_t orig_skb_priority; /* Priority when packet arrived. */
328 uint8_t table_id; /* OpenFlow table ID where flow was found. */
329 uint32_t sflow_n_outputs; /* Number of output ports. */
330 uint32_t sflow_odp_port; /* Output port for composing sFlow action. */
331 uint16_t user_cookie_offset;/* Used for user_action_cookie fixup. */
332 bool exit; /* No further actions should be processed. */
335 static void xlate_in_init(struct xlate_in *, struct ofproto_dpif *,
336 const struct flow *, const struct initial_vals *,
337 struct rule_dpif *, uint8_t tcp_flags,
338 const struct ofpbuf *);
340 static void xlate_out_uninit(struct xlate_out *);
342 static void xlate_actions(struct xlate_in *, struct xlate_out *);
344 static void xlate_actions_for_side_effects(struct xlate_in *);
346 static void xlate_table_action(struct xlate_ctx *, uint16_t in_port,
347 uint8_t table_id, bool may_packet_in);
349 static size_t put_userspace_action(const struct ofproto_dpif *,
350 struct ofpbuf *odp_actions,
352 const union user_action_cookie *,
355 static void compose_slow_path(const struct ofproto_dpif *, const struct flow *,
356 enum slow_path_reason,
357 uint64_t *stub, size_t stub_size,
358 const struct nlattr **actionsp,
359 size_t *actions_lenp);
361 static void xlate_report(struct xlate_ctx *ctx, const char *s);
363 /* A subfacet (see "struct subfacet" below) has three possible installation
366 * - SF_NOT_INSTALLED: Not installed in the datapath. This will only be the
367 * case just after the subfacet is created, just before the subfacet is
368 * destroyed, or if the datapath returns an error when we try to install a
371 * - SF_FAST_PATH: The subfacet's actions are installed in the datapath.
373 * - SF_SLOW_PATH: An action that sends every packet for the subfacet through
374 * ofproto_dpif is installed in the datapath.
377 SF_NOT_INSTALLED, /* No datapath flow for this subfacet. */
378 SF_FAST_PATH, /* Full actions are installed. */
379 SF_SLOW_PATH, /* Send-to-userspace action is installed. */
382 /* A dpif flow and actions associated with a facet.
384 * See also the large comment on struct facet. */
387 struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */
388 struct list list_node; /* In struct facet's 'facets' list. */
389 struct facet *facet; /* Owning facet. */
391 enum odp_key_fitness key_fitness;
395 long long int used; /* Time last used; time created if not used. */
396 long long int created; /* Time created. */
398 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
399 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
401 enum subfacet_path path; /* Installed in datapath? */
404 #define SUBFACET_DESTROY_MAX_BATCH 50
406 static struct subfacet *subfacet_create(struct facet *, struct flow_miss *miss,
408 static struct subfacet *subfacet_find(struct ofproto_dpif *,
409 const struct nlattr *key, size_t key_len,
411 static void subfacet_destroy(struct subfacet *);
412 static void subfacet_destroy__(struct subfacet *);
413 static void subfacet_destroy_batch(struct ofproto_dpif *,
414 struct subfacet **, int n);
415 static void subfacet_reset_dp_stats(struct subfacet *,
416 struct dpif_flow_stats *);
417 static void subfacet_update_stats(struct subfacet *,
418 const struct dpif_flow_stats *);
419 static int subfacet_install(struct subfacet *,
420 const struct ofpbuf *odp_actions,
421 struct dpif_flow_stats *);
422 static void subfacet_uninstall(struct subfacet *);
424 /* An exact-match instantiation of an OpenFlow flow.
426 * A facet associates a "struct flow", which represents the Open vSwitch
427 * userspace idea of an exact-match flow, with one or more subfacets. Each
428 * subfacet tracks the datapath's idea of the exact-match flow equivalent to
429 * the facet. When the kernel module (or other dpif implementation) and Open
430 * vSwitch userspace agree on the definition of a flow key, there is exactly
431 * one subfacet per facet. If the dpif implementation supports more-specific
432 * flow matching than userspace, however, a facet can have more than one
433 * subfacet, each of which corresponds to some distinction in flow that
434 * userspace simply doesn't understand.
436 * Flow expiration works in terms of subfacets, so a facet must have at least
437 * one subfacet or it will never expire, leaking memory. */
440 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
441 struct list list_node; /* In owning rule's 'facets' list. */
442 struct rule_dpif *rule; /* Owning rule. */
445 struct list subfacets;
446 long long int used; /* Time last used; time created if not used. */
453 * - Do include packets and bytes sent "by hand", e.g. with
456 * - Do include packets and bytes that were obtained from the datapath
457 * when a subfacet's statistics were reset (e.g. dpif_flow_put() with
458 * DPIF_FP_ZERO_STATS).
460 * - Do not include packets or bytes that can be obtained from the
461 * datapath for any existing subfacet.
463 uint64_t packet_count; /* Number of packets received. */
464 uint64_t byte_count; /* Number of bytes received. */
466 /* Resubmit statistics. */
467 uint64_t prev_packet_count; /* Number of packets from last stats push. */
468 uint64_t prev_byte_count; /* Number of bytes from last stats push. */
469 long long int prev_used; /* Used time from last stats push. */
472 uint64_t accounted_bytes; /* Bytes processed by facet_account(). */
473 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
474 uint8_t tcp_flags; /* TCP flags seen for this 'rule'. */
476 struct xlate_out xout;
478 /* Initial values of the packet that may be needed later. */
479 struct initial_vals initial_vals;
481 /* Storage for a single subfacet, to reduce malloc() time and space
482 * overhead. (A facet always has at least one subfacet and in the common
483 * case has exactly one subfacet. However, 'one_subfacet' may not
484 * always be valid, since it could have been removed after newer
485 * subfacets were pushed onto the 'subfacets' list.) */
486 struct subfacet one_subfacet;
488 long long int learn_rl; /* Rate limiter for facet_learn(). */
491 static struct facet *facet_create(const struct flow_miss *, uint32_t hash);
492 static void facet_remove(struct facet *);
493 static void facet_free(struct facet *);
495 static struct facet *facet_find(struct ofproto_dpif *,
496 const struct flow *, uint32_t hash);
497 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
498 const struct flow *, uint32_t hash);
499 static bool facet_revalidate(struct facet *);
500 static bool facet_check_consistency(struct facet *);
502 static void facet_flush_stats(struct facet *);
504 static void facet_reset_counters(struct facet *);
505 static void facet_push_stats(struct facet *, bool may_learn);
506 static void facet_learn(struct facet *);
507 static void facet_account(struct facet *);
508 static void push_all_stats(void);
510 static bool facet_is_controller_flow(struct facet *);
513 struct hmap_node odp_port_node; /* In dpif_backer's "odp_to_ofport_map". */
517 struct ofbundle *bundle; /* Bundle that contains this port, if any. */
518 struct list bundle_node; /* In struct ofbundle's "ports" list. */
519 struct cfm *cfm; /* Connectivity Fault Management, if any. */
520 struct bfd *bfd; /* BFD, if any. */
521 tag_type tag; /* Tag associated with this port. */
522 bool may_enable; /* May be enabled in bonds. */
523 long long int carrier_seq; /* Carrier status changes. */
524 struct tnl_port *tnl_port; /* Tunnel handle, or null. */
527 struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */
528 enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */
529 long long int stp_state_entered;
531 struct hmap priorities; /* Map of attached 'priority_to_dscp's. */
533 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
535 * This is deprecated. It is only for compatibility with broken device
536 * drivers in old versions of Linux that do not properly support VLANs when
537 * VLAN devices are not used. When broken device drivers are no longer in
538 * widespread use, we will delete these interfaces. */
539 uint16_t realdev_ofp_port;
543 /* Node in 'ofport_dpif''s 'priorities' map. Used to maintain a map from
544 * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
545 * traffic egressing the 'ofport' with that priority should be marked with. */
546 struct priority_to_dscp {
547 struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'priorities' map. */
548 uint32_t priority; /* Priority of this queue (see struct flow). */
550 uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */
553 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
555 * This is deprecated. It is only for compatibility with broken device drivers
556 * in old versions of Linux that do not properly support VLANs when VLAN
557 * devices are not used. When broken device drivers are no longer in
558 * widespread use, we will delete these interfaces. */
559 struct vlan_splinter {
560 struct hmap_node realdev_vid_node;
561 struct hmap_node vlandev_node;
562 uint16_t realdev_ofp_port;
563 uint16_t vlandev_ofp_port;
567 static uint16_t vsp_realdev_to_vlandev(const struct ofproto_dpif *,
568 uint16_t realdev_ofp_port,
570 static bool vsp_adjust_flow(const struct ofproto_dpif *, struct flow *);
571 static void vsp_remove(struct ofport_dpif *);
572 static void vsp_add(struct ofport_dpif *, uint16_t realdev_ofp_port, int vid);
574 static uint32_t ofp_port_to_odp_port(const struct ofproto_dpif *,
576 static uint16_t odp_port_to_ofp_port(const struct ofproto_dpif *,
579 static struct ofport_dpif *
580 ofport_dpif_cast(const struct ofport *ofport)
582 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
585 static void port_run(struct ofport_dpif *);
586 static void port_run_fast(struct ofport_dpif *);
587 static void port_wait(struct ofport_dpif *);
588 static int set_bfd(struct ofport *, const struct smap *);
589 static int set_cfm(struct ofport *, const struct cfm_settings *);
590 static void ofport_clear_priorities(struct ofport_dpif *);
591 static void run_fast_rl(void);
593 struct dpif_completion {
594 struct list list_node;
595 struct ofoperation *op;
598 /* Extra information about a classifier table.
599 * Currently used just for optimized flow revalidation. */
601 /* If either of these is nonnull, then this table has a form that allows
602 * flows to be tagged to avoid revalidating most flows for the most common
603 * kinds of flow table changes. */
604 struct cls_table *catchall_table; /* Table that wildcards all fields. */
605 struct cls_table *other_table; /* Table with any other wildcard set. */
606 uint32_t basis; /* Keeps each table's tags separate. */
609 /* Reasons that we might need to revalidate every facet, and corresponding
612 * A value of 0 means that there is no need to revalidate.
614 * It would be nice to have some cleaner way to integrate with coverage
615 * counters, but with only a few reasons I guess this is good enough for
617 enum revalidate_reason {
618 REV_RECONFIGURE = 1, /* Switch configuration changed. */
619 REV_STP, /* Spanning tree protocol port status change. */
620 REV_PORT_TOGGLED, /* Port enabled or disabled by CFM, LACP, ...*/
621 REV_FLOW_TABLE, /* Flow table changed. */
622 REV_INCONSISTENCY /* Facet self-check failed. */
624 COVERAGE_DEFINE(rev_reconfigure);
625 COVERAGE_DEFINE(rev_stp);
626 COVERAGE_DEFINE(rev_port_toggled);
627 COVERAGE_DEFINE(rev_flow_table);
628 COVERAGE_DEFINE(rev_inconsistency);
630 /* Drop keys are odp flow keys which have drop flows installed in the kernel.
631 * These are datapath flows which have no associated ofproto, if they did we
632 * would use facets. */
634 struct hmap_node hmap_node;
639 struct avg_subfacet_rates {
640 double add_rate; /* Moving average of new flows created per minute. */
641 double del_rate; /* Moving average of flows deleted per minute. */
644 /* All datapaths of a given type share a single dpif backer instance. */
649 struct timer next_expiration;
650 struct hmap odp_to_ofport_map; /* ODP port to ofport mapping. */
652 struct simap tnl_backers; /* Set of dpif ports backing tunnels. */
654 /* Facet revalidation flags applying to facets which use this backer. */
655 enum revalidate_reason need_revalidate; /* Revalidate every facet. */
656 struct tag_set revalidate_set; /* Revalidate only matching facets. */
658 struct hmap drop_keys; /* Set of dropped odp keys. */
659 bool recv_set_enable; /* Enables or disables receiving packets. */
661 /* Subfacet statistics.
663 * These keep track of the total number of subfacets added and deleted and
664 * flow life span. They are useful for computing the flow rates stats
665 * exposed via "ovs-appctl dpif/show". The goal is to learn about
666 * traffic patterns in ways that we can use later to improve Open vSwitch
667 * performance in new situations. */
668 long long int created; /* Time when it is created. */
669 unsigned max_n_subfacet; /* Maximum number of flows */
670 unsigned avg_n_subfacet; /* Average number of flows. */
671 long long int avg_subfacet_life; /* Average life span of subfacets. */
673 /* The average number of subfacets... */
674 struct avg_subfacet_rates hourly; /* ...over the last hour. */
675 struct avg_subfacet_rates daily; /* ...over the last day. */
676 struct avg_subfacet_rates lifetime; /* ...over the switch lifetime. */
677 long long int last_minute; /* Last time 'hourly' was updated. */
679 /* Number of subfacets added or deleted since 'last_minute'. */
680 unsigned subfacet_add_count;
681 unsigned subfacet_del_count;
683 /* Number of subfacets added or deleted from 'created' to 'last_minute.' */
684 unsigned long long int total_subfacet_add_count;
685 unsigned long long int total_subfacet_del_count;
688 /* All existing ofproto_backer instances, indexed by ofproto->up.type. */
689 static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers);
691 static void drop_key_clear(struct dpif_backer *);
692 static struct ofport_dpif *
693 odp_port_to_ofport(const struct dpif_backer *, uint32_t odp_port);
694 static void update_moving_averages(struct dpif_backer *backer);
696 struct ofproto_dpif {
697 struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
699 struct dpif_backer *backer;
701 /* Special OpenFlow rules. */
702 struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */
703 struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */
704 struct rule_dpif *drop_frags_rule; /* Used in OFPC_FRAG_DROP mode. */
707 struct netflow *netflow;
708 struct dpif_sflow *sflow;
709 struct dpif_ipfix *ipfix;
710 struct hmap bundles; /* Contains "struct ofbundle"s. */
711 struct mac_learning *ml;
712 struct ofmirror *mirrors[MAX_MIRRORS];
714 bool has_bonded_bundles;
718 struct hmap subfacets;
719 struct governor *governor;
720 long long int consistency_rl;
723 struct table_dpif tables[N_TABLES];
725 /* Support for debugging async flow mods. */
726 struct list completions;
728 bool has_bundle_action; /* True when the first bundle action appears. */
729 struct netdev_stats stats; /* To account packets generated and consumed in
734 long long int stp_last_tick;
736 /* VLAN splinters. */
737 struct hmap realdev_vid_map; /* (realdev,vid) -> vlandev. */
738 struct hmap vlandev_map; /* vlandev -> (realdev,vid). */
741 struct sset ports; /* Set of standard port names. */
742 struct sset ghost_ports; /* Ports with no datapath port. */
743 struct sset port_poll_set; /* Queued names for port_poll() reply. */
744 int port_poll_errno; /* Last errno for port_poll() reply. */
746 /* Per ofproto's dpif stats. */
751 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
752 * for debugging the asynchronous flow_mod implementation.) */
755 /* All existing ofproto_dpif instances, indexed by ->up.name. */
756 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
758 static void ofproto_dpif_unixctl_init(void);
760 static struct ofproto_dpif *
761 ofproto_dpif_cast(const struct ofproto *ofproto)
763 ovs_assert(ofproto->ofproto_class == &ofproto_dpif_class);
764 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
767 static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *,
769 static struct ofport_dpif *get_odp_port(const struct ofproto_dpif *,
771 static void ofproto_trace(struct ofproto_dpif *, const struct flow *,
772 const struct ofpbuf *,
773 const struct initial_vals *, struct ds *);
775 /* Packet processing. */
776 static void update_learning_table(struct ofproto_dpif *,
777 const struct flow *, int vlan,
780 #define FLOW_MISS_MAX_BATCH 50
781 static int handle_upcalls(struct dpif_backer *, unsigned int max_batch);
783 /* Flow expiration. */
784 static int expire(struct dpif_backer *);
787 static void send_netflow_active_timeouts(struct ofproto_dpif *);
790 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
791 static size_t compose_sflow_action(const struct ofproto_dpif *,
792 struct ofpbuf *odp_actions,
793 const struct flow *, uint32_t odp_port);
794 static void compose_ipfix_action(const struct ofproto_dpif *,
795 struct ofpbuf *odp_actions,
796 const struct flow *);
797 static void add_mirror_actions(struct xlate_ctx *ctx,
798 const struct flow *flow);
799 /* Global variables. */
800 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
802 /* Initial mappings of port to bridge mappings. */
803 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
805 /* Factory functions. */
808 init(const struct shash *iface_hints)
810 struct shash_node *node;
812 /* Make a local copy, since we don't own 'iface_hints' elements. */
813 SHASH_FOR_EACH(node, iface_hints) {
814 const struct iface_hint *orig_hint = node->data;
815 struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
817 new_hint->br_name = xstrdup(orig_hint->br_name);
818 new_hint->br_type = xstrdup(orig_hint->br_type);
819 new_hint->ofp_port = orig_hint->ofp_port;
821 shash_add(&init_ofp_ports, node->name, new_hint);
826 enumerate_types(struct sset *types)
828 dp_enumerate_types(types);
832 enumerate_names(const char *type, struct sset *names)
834 struct ofproto_dpif *ofproto;
837 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
838 if (strcmp(type, ofproto->up.type)) {
841 sset_add(names, ofproto->up.name);
848 del(const char *type, const char *name)
853 error = dpif_open(name, type, &dpif);
855 error = dpif_delete(dpif);
862 port_open_type(const char *datapath_type, const char *port_type)
864 return dpif_port_open_type(datapath_type, port_type);
867 /* Type functions. */
869 static struct ofproto_dpif *
870 lookup_ofproto_dpif_by_port_name(const char *name)
872 struct ofproto_dpif *ofproto;
874 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
875 if (sset_contains(&ofproto->ports, name)) {
884 type_run(const char *type)
886 static long long int push_timer = LLONG_MIN;
887 struct dpif_backer *backer;
891 backer = shash_find_data(&all_dpif_backers, type);
893 /* This is not necessarily a problem, since backers are only
894 * created on demand. */
898 dpif_run(backer->dpif);
900 /* The most natural place to push facet statistics is when they're pulled
901 * from the datapath. However, when there are many flows in the datapath,
902 * this expensive operation can occur so frequently, that it reduces our
903 * ability to quickly set up flows. To reduce the cost, we push statistics
905 if (time_msec() > push_timer) {
906 push_timer = time_msec() + 2000;
910 /* If vswitchd started with other_config:flow_restore_wait set as "true",
911 * and the configuration has now changed to "false", enable receiving
912 * packets from the datapath. */
913 if (!backer->recv_set_enable && !ofproto_get_flow_restore_wait()) {
914 backer->recv_set_enable = true;
916 error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
918 VLOG_ERR("Failed to enable receiving packets in dpif.");
921 dpif_flow_flush(backer->dpif);
922 backer->need_revalidate = REV_RECONFIGURE;
925 if (backer->need_revalidate
926 || !tag_set_is_empty(&backer->revalidate_set)) {
927 struct tag_set revalidate_set = backer->revalidate_set;
928 bool need_revalidate = backer->need_revalidate;
929 struct ofproto_dpif *ofproto;
930 struct simap_node *node;
931 struct simap tmp_backers;
933 /* Handle tunnel garbage collection. */
934 simap_init(&tmp_backers);
935 simap_swap(&backer->tnl_backers, &tmp_backers);
937 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
938 struct ofport_dpif *iter;
940 if (backer != ofproto->backer) {
944 HMAP_FOR_EACH (iter, up.hmap_node, &ofproto->up.ports) {
945 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
948 if (!iter->tnl_port) {
952 dp_port = netdev_vport_get_dpif_port(iter->up.netdev,
953 namebuf, sizeof namebuf);
954 node = simap_find(&tmp_backers, dp_port);
956 simap_put(&backer->tnl_backers, dp_port, node->data);
957 simap_delete(&tmp_backers, node);
958 node = simap_find(&backer->tnl_backers, dp_port);
960 node = simap_find(&backer->tnl_backers, dp_port);
962 uint32_t odp_port = UINT32_MAX;
964 if (!dpif_port_add(backer->dpif, iter->up.netdev,
966 simap_put(&backer->tnl_backers, dp_port, odp_port);
967 node = simap_find(&backer->tnl_backers, dp_port);
972 iter->odp_port = node ? node->data : OVSP_NONE;
973 if (tnl_port_reconfigure(&iter->up, iter->odp_port,
975 backer->need_revalidate = REV_RECONFIGURE;
980 SIMAP_FOR_EACH (node, &tmp_backers) {
981 dpif_port_del(backer->dpif, node->data);
983 simap_destroy(&tmp_backers);
985 switch (backer->need_revalidate) {
986 case REV_RECONFIGURE: COVERAGE_INC(rev_reconfigure); break;
987 case REV_STP: COVERAGE_INC(rev_stp); break;
988 case REV_PORT_TOGGLED: COVERAGE_INC(rev_port_toggled); break;
989 case REV_FLOW_TABLE: COVERAGE_INC(rev_flow_table); break;
990 case REV_INCONSISTENCY: COVERAGE_INC(rev_inconsistency); break;
993 if (backer->need_revalidate) {
994 /* Clear the drop_keys in case we should now be accepting some
995 * formerly dropped flows. */
996 drop_key_clear(backer);
999 /* Clear the revalidation flags. */
1000 tag_set_init(&backer->revalidate_set);
1001 backer->need_revalidate = 0;
1003 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1004 struct facet *facet, *next;
1006 if (ofproto->backer != backer) {
1010 HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &ofproto->facets) {
1012 || tag_set_intersects(&revalidate_set, facet->xout.tags)) {
1013 facet_revalidate(facet);
1020 if (!backer->recv_set_enable) {
1021 /* Wake up before a max of 1000ms. */
1022 timer_set_duration(&backer->next_expiration, 1000);
1023 } else if (timer_expired(&backer->next_expiration)) {
1024 int delay = expire(backer);
1025 timer_set_duration(&backer->next_expiration, delay);
1028 /* Check for port changes in the dpif. */
1029 while ((error = dpif_port_poll(backer->dpif, &devname)) == 0) {
1030 struct ofproto_dpif *ofproto;
1031 struct dpif_port port;
1033 /* Don't report on the datapath's device. */
1034 if (!strcmp(devname, dpif_base_name(backer->dpif))) {
1038 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
1039 &all_ofproto_dpifs) {
1040 if (simap_contains(&ofproto->backer->tnl_backers, devname)) {
1045 ofproto = lookup_ofproto_dpif_by_port_name(devname);
1046 if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
1047 /* The port was removed. If we know the datapath,
1048 * report it through poll_set(). If we don't, it may be
1049 * notifying us of a removal we initiated, so ignore it.
1050 * If there's a pending ENOBUFS, let it stand, since
1051 * everything will be reevaluated. */
1052 if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
1053 sset_add(&ofproto->port_poll_set, devname);
1054 ofproto->port_poll_errno = 0;
1056 } else if (!ofproto) {
1057 /* The port was added, but we don't know with which
1058 * ofproto we should associate it. Delete it. */
1059 dpif_port_del(backer->dpif, port.port_no);
1061 dpif_port_destroy(&port);
1067 if (error != EAGAIN) {
1068 struct ofproto_dpif *ofproto;
1070 /* There was some sort of error, so propagate it to all
1071 * ofprotos that use this backer. */
1072 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
1073 &all_ofproto_dpifs) {
1074 if (ofproto->backer == backer) {
1075 sset_clear(&ofproto->port_poll_set);
1076 ofproto->port_poll_errno = error;
1085 dpif_backer_run_fast(struct dpif_backer *backer, int max_batch)
1089 /* If recv_set_enable is false, we should not handle upcalls. */
1090 if (!backer->recv_set_enable) {
1094 /* Handle one or more batches of upcalls, until there's nothing left to do
1095 * or until we do a fixed total amount of work.
1097 * We do work in batches because it can be much cheaper to set up a number
1098 * of flows and fire off their patches all at once. We do multiple batches
1099 * because in some cases handling a packet can cause another packet to be
1100 * queued almost immediately as part of the return flow. Both
1101 * optimizations can make major improvements on some benchmarks and
1102 * presumably for real traffic as well. */
1104 while (work < max_batch) {
1105 int retval = handle_upcalls(backer, max_batch - work);
1116 type_run_fast(const char *type)
1118 struct dpif_backer *backer;
1120 backer = shash_find_data(&all_dpif_backers, type);
1122 /* This is not necessarily a problem, since backers are only
1123 * created on demand. */
1127 return dpif_backer_run_fast(backer, FLOW_MISS_MAX_BATCH);
1133 static long long int port_rl = LLONG_MIN;
1134 static unsigned int backer_rl = 0;
1136 if (time_msec() >= port_rl) {
1137 struct ofproto_dpif *ofproto;
1138 struct ofport_dpif *ofport;
1140 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1142 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1143 port_run_fast(ofport);
1146 port_rl = time_msec() + 200;
1149 /* XXX: We have to be careful not to do too much work in this function. If
1150 * we call dpif_backer_run_fast() too often, or with too large a batch,
1151 * performance improves signifcantly, but at a cost. It's possible for the
1152 * number of flows in the datapath to increase without bound, and for poll
1153 * loops to take 10s of seconds. The correct solution to this problem,
1154 * long term, is to separate flow miss handling into it's own thread so it
1155 * isn't affected by revalidations, and expirations. Until then, this is
1156 * the best we can do. */
1157 if (++backer_rl >= 10) {
1158 struct shash_node *node;
1161 SHASH_FOR_EACH (node, &all_dpif_backers) {
1162 dpif_backer_run_fast(node->data, 1);
1168 type_wait(const char *type)
1170 struct dpif_backer *backer;
1172 backer = shash_find_data(&all_dpif_backers, type);
1174 /* This is not necessarily a problem, since backers are only
1175 * created on demand. */
1179 timer_wait(&backer->next_expiration);
1182 /* Basic life-cycle. */
1184 static int add_internal_flows(struct ofproto_dpif *);
1186 static struct ofproto *
1189 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
1190 return &ofproto->up;
1194 dealloc(struct ofproto *ofproto_)
1196 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1201 close_dpif_backer(struct dpif_backer *backer)
1203 struct shash_node *node;
1205 ovs_assert(backer->refcount > 0);
1207 if (--backer->refcount) {
1211 drop_key_clear(backer);
1212 hmap_destroy(&backer->drop_keys);
1214 simap_destroy(&backer->tnl_backers);
1215 hmap_destroy(&backer->odp_to_ofport_map);
1216 node = shash_find(&all_dpif_backers, backer->type);
1218 shash_delete(&all_dpif_backers, node);
1219 dpif_close(backer->dpif);
1224 /* Datapath port slated for removal from datapath. */
1225 struct odp_garbage {
1226 struct list list_node;
1231 open_dpif_backer(const char *type, struct dpif_backer **backerp)
1233 struct dpif_backer *backer;
1234 struct dpif_port_dump port_dump;
1235 struct dpif_port port;
1236 struct shash_node *node;
1237 struct list garbage_list;
1238 struct odp_garbage *garbage, *next;
1244 backer = shash_find_data(&all_dpif_backers, type);
1251 backer_name = xasprintf("ovs-%s", type);
1253 /* Remove any existing datapaths, since we assume we're the only
1254 * userspace controlling the datapath. */
1256 dp_enumerate_names(type, &names);
1257 SSET_FOR_EACH(name, &names) {
1258 struct dpif *old_dpif;
1260 /* Don't remove our backer if it exists. */
1261 if (!strcmp(name, backer_name)) {
1265 if (dpif_open(name, type, &old_dpif)) {
1266 VLOG_WARN("couldn't open old datapath %s to remove it", name);
1268 dpif_delete(old_dpif);
1269 dpif_close(old_dpif);
1272 sset_destroy(&names);
1274 backer = xmalloc(sizeof *backer);
1276 error = dpif_create_and_open(backer_name, type, &backer->dpif);
1279 VLOG_ERR("failed to open datapath of type %s: %s", type,
1285 backer->type = xstrdup(type);
1286 backer->refcount = 1;
1287 hmap_init(&backer->odp_to_ofport_map);
1288 hmap_init(&backer->drop_keys);
1289 timer_set_duration(&backer->next_expiration, 1000);
1290 backer->need_revalidate = 0;
1291 simap_init(&backer->tnl_backers);
1292 tag_set_init(&backer->revalidate_set);
1293 backer->recv_set_enable = !ofproto_get_flow_restore_wait();
1296 if (backer->recv_set_enable) {
1297 dpif_flow_flush(backer->dpif);
1300 /* Loop through the ports already on the datapath and remove any
1301 * that we don't need anymore. */
1302 list_init(&garbage_list);
1303 dpif_port_dump_start(&port_dump, backer->dpif);
1304 while (dpif_port_dump_next(&port_dump, &port)) {
1305 node = shash_find(&init_ofp_ports, port.name);
1306 if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
1307 garbage = xmalloc(sizeof *garbage);
1308 garbage->odp_port = port.port_no;
1309 list_push_front(&garbage_list, &garbage->list_node);
1312 dpif_port_dump_done(&port_dump);
1314 LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) {
1315 dpif_port_del(backer->dpif, garbage->odp_port);
1316 list_remove(&garbage->list_node);
1320 shash_add(&all_dpif_backers, type, backer);
1322 error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
1324 VLOG_ERR("failed to listen on datapath of type %s: %s",
1325 type, strerror(error));
1326 close_dpif_backer(backer);
1330 backer->max_n_subfacet = 0;
1331 backer->created = time_msec();
1332 backer->last_minute = backer->created;
1333 memset(&backer->hourly, 0, sizeof backer->hourly);
1334 memset(&backer->daily, 0, sizeof backer->daily);
1335 memset(&backer->lifetime, 0, sizeof backer->lifetime);
1336 backer->subfacet_add_count = 0;
1337 backer->subfacet_del_count = 0;
1338 backer->total_subfacet_add_count = 0;
1339 backer->total_subfacet_del_count = 0;
1340 backer->avg_n_subfacet = 0;
1341 backer->avg_subfacet_life = 0;
1347 construct(struct ofproto *ofproto_)
1349 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1350 struct shash_node *node, *next;
1355 error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1360 max_ports = dpif_get_max_ports(ofproto->backer->dpif);
1361 ofproto_init_max_ports(ofproto_, MIN(max_ports, OFPP_MAX));
1363 ofproto->netflow = NULL;
1364 ofproto->sflow = NULL;
1365 ofproto->ipfix = NULL;
1366 ofproto->stp = NULL;
1367 hmap_init(&ofproto->bundles);
1368 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1369 for (i = 0; i < MAX_MIRRORS; i++) {
1370 ofproto->mirrors[i] = NULL;
1372 ofproto->has_bonded_bundles = false;
1374 hmap_init(&ofproto->facets);
1375 hmap_init(&ofproto->subfacets);
1376 ofproto->governor = NULL;
1377 ofproto->consistency_rl = LLONG_MIN;
1379 for (i = 0; i < N_TABLES; i++) {
1380 struct table_dpif *table = &ofproto->tables[i];
1382 table->catchall_table = NULL;
1383 table->other_table = NULL;
1384 table->basis = random_uint32();
1387 list_init(&ofproto->completions);
1389 ofproto_dpif_unixctl_init();
1391 ofproto->has_mirrors = false;
1392 ofproto->has_bundle_action = false;
1394 hmap_init(&ofproto->vlandev_map);
1395 hmap_init(&ofproto->realdev_vid_map);
1397 sset_init(&ofproto->ports);
1398 sset_init(&ofproto->ghost_ports);
1399 sset_init(&ofproto->port_poll_set);
1400 ofproto->port_poll_errno = 0;
1402 SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1403 struct iface_hint *iface_hint = node->data;
1405 if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1406 /* Check if the datapath already has this port. */
1407 if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1408 sset_add(&ofproto->ports, node->name);
1411 free(iface_hint->br_name);
1412 free(iface_hint->br_type);
1414 shash_delete(&init_ofp_ports, node);
1418 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1419 hash_string(ofproto->up.name, 0));
1420 memset(&ofproto->stats, 0, sizeof ofproto->stats);
1422 ofproto_init_tables(ofproto_, N_TABLES);
1423 error = add_internal_flows(ofproto);
1424 ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1427 ofproto->n_missed = 0;
1433 add_internal_flow(struct ofproto_dpif *ofproto, int id,
1434 const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1436 struct ofputil_flow_mod fm;
1439 match_init_catchall(&fm.match);
1441 match_set_reg(&fm.match, 0, id);
1442 fm.new_cookie = htonll(0);
1443 fm.cookie = htonll(0);
1444 fm.cookie_mask = htonll(0);
1445 fm.table_id = TBL_INTERNAL;
1446 fm.command = OFPFC_ADD;
1447 fm.idle_timeout = 0;
1448 fm.hard_timeout = 0;
1452 fm.ofpacts = ofpacts->data;
1453 fm.ofpacts_len = ofpacts->size;
1455 error = ofproto_flow_mod(&ofproto->up, &fm);
1457 VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
1458 id, ofperr_to_string(error));
1462 *rulep = rule_dpif_lookup__(ofproto, &fm.match.flow, TBL_INTERNAL);
1463 ovs_assert(*rulep != NULL);
1469 add_internal_flows(struct ofproto_dpif *ofproto)
1471 struct ofpact_controller *controller;
1472 uint64_t ofpacts_stub[128 / 8];
1473 struct ofpbuf ofpacts;
1477 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1480 controller = ofpact_put_CONTROLLER(&ofpacts);
1481 controller->max_len = UINT16_MAX;
1482 controller->controller_id = 0;
1483 controller->reason = OFPR_NO_MATCH;
1484 ofpact_pad(&ofpacts);
1486 error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule);
1491 ofpbuf_clear(&ofpacts);
1492 error = add_internal_flow(ofproto, id++, &ofpacts,
1493 &ofproto->no_packet_in_rule);
1498 error = add_internal_flow(ofproto, id++, &ofpacts,
1499 &ofproto->drop_frags_rule);
1504 complete_operations(struct ofproto_dpif *ofproto)
1506 struct dpif_completion *c, *next;
1508 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
1509 ofoperation_complete(c->op, 0);
1510 list_remove(&c->list_node);
1516 destruct(struct ofproto *ofproto_)
1518 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1519 struct rule_dpif *rule, *next_rule;
1520 struct oftable *table;
1523 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1524 complete_operations(ofproto);
1526 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1527 struct cls_cursor cursor;
1529 cls_cursor_init(&cursor, &table->cls, NULL);
1530 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1531 ofproto_rule_destroy(&rule->up);
1535 for (i = 0; i < MAX_MIRRORS; i++) {
1536 mirror_destroy(ofproto->mirrors[i]);
1539 netflow_destroy(ofproto->netflow);
1540 dpif_sflow_destroy(ofproto->sflow);
1541 hmap_destroy(&ofproto->bundles);
1542 mac_learning_destroy(ofproto->ml);
1544 hmap_destroy(&ofproto->facets);
1545 hmap_destroy(&ofproto->subfacets);
1546 governor_destroy(ofproto->governor);
1548 hmap_destroy(&ofproto->vlandev_map);
1549 hmap_destroy(&ofproto->realdev_vid_map);
1551 sset_destroy(&ofproto->ports);
1552 sset_destroy(&ofproto->ghost_ports);
1553 sset_destroy(&ofproto->port_poll_set);
1555 close_dpif_backer(ofproto->backer);
1559 run_fast(struct ofproto *ofproto_)
1561 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1562 struct ofport_dpif *ofport;
1564 /* Do not perform any periodic activity required by 'ofproto' while
1565 * waiting for flow restore to complete. */
1566 if (ofproto_get_flow_restore_wait()) {
1570 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1571 port_run_fast(ofport);
1578 run(struct ofproto *ofproto_)
1580 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1581 struct ofport_dpif *ofport;
1582 struct ofbundle *bundle;
1586 complete_operations(ofproto);
1589 /* Do not perform any periodic activity below required by 'ofproto' while
1590 * waiting for flow restore to complete. */
1591 if (ofproto_get_flow_restore_wait()) {
1595 error = run_fast(ofproto_);
1600 if (ofproto->netflow) {
1601 if (netflow_run(ofproto->netflow)) {
1602 send_netflow_active_timeouts(ofproto);
1605 if (ofproto->sflow) {
1606 dpif_sflow_run(ofproto->sflow);
1609 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1612 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1617 mac_learning_run(ofproto->ml, &ofproto->backer->revalidate_set);
1619 /* Check the consistency of a random facet, to aid debugging. */
1620 if (time_msec() >= ofproto->consistency_rl
1621 && !hmap_is_empty(&ofproto->facets)
1622 && !ofproto->backer->need_revalidate) {
1623 struct facet *facet;
1625 ofproto->consistency_rl = time_msec() + 250;
1627 facet = CONTAINER_OF(hmap_random_node(&ofproto->facets),
1628 struct facet, hmap_node);
1629 if (!tag_set_intersects(&ofproto->backer->revalidate_set,
1630 facet->xout.tags)) {
1631 if (!facet_check_consistency(facet)) {
1632 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
1637 if (ofproto->governor) {
1640 governor_run(ofproto->governor);
1642 /* If the governor has shrunk to its minimum size and the number of
1643 * subfacets has dwindled, then drop the governor entirely.
1645 * For hysteresis, the number of subfacets to drop the governor is
1646 * smaller than the number needed to trigger its creation. */
1647 n_subfacets = hmap_count(&ofproto->subfacets);
1648 if (n_subfacets * 4 < ofproto->up.flow_eviction_threshold
1649 && governor_is_idle(ofproto->governor)) {
1650 governor_destroy(ofproto->governor);
1651 ofproto->governor = NULL;
1659 wait(struct ofproto *ofproto_)
1661 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1662 struct ofport_dpif *ofport;
1663 struct ofbundle *bundle;
1665 if (!clogged && !list_is_empty(&ofproto->completions)) {
1666 poll_immediate_wake();
1669 if (ofproto_get_flow_restore_wait()) {
1673 dpif_wait(ofproto->backer->dpif);
1674 dpif_recv_wait(ofproto->backer->dpif);
1675 if (ofproto->sflow) {
1676 dpif_sflow_wait(ofproto->sflow);
1678 if (!tag_set_is_empty(&ofproto->backer->revalidate_set)) {
1679 poll_immediate_wake();
1681 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1684 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1685 bundle_wait(bundle);
1687 if (ofproto->netflow) {
1688 netflow_wait(ofproto->netflow);
1690 mac_learning_wait(ofproto->ml);
1692 if (ofproto->backer->need_revalidate) {
1693 /* Shouldn't happen, but if it does just go around again. */
1694 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1695 poll_immediate_wake();
1697 if (ofproto->governor) {
1698 governor_wait(ofproto->governor);
1703 get_memory_usage(const struct ofproto *ofproto_, struct simap *usage)
1705 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1707 simap_increase(usage, "facets", hmap_count(&ofproto->facets));
1708 simap_increase(usage, "subfacets", hmap_count(&ofproto->subfacets));
1712 flush(struct ofproto *ofproto_)
1714 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1715 struct subfacet *subfacet, *next_subfacet;
1716 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
1720 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
1721 &ofproto->subfacets) {
1722 if (subfacet->path != SF_NOT_INSTALLED) {
1723 batch[n_batch++] = subfacet;
1724 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
1725 subfacet_destroy_batch(ofproto, batch, n_batch);
1729 subfacet_destroy(subfacet);
1734 subfacet_destroy_batch(ofproto, batch, n_batch);
1739 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1740 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1742 *arp_match_ip = true;
1743 *actions = (OFPUTIL_A_OUTPUT |
1744 OFPUTIL_A_SET_VLAN_VID |
1745 OFPUTIL_A_SET_VLAN_PCP |
1746 OFPUTIL_A_STRIP_VLAN |
1747 OFPUTIL_A_SET_DL_SRC |
1748 OFPUTIL_A_SET_DL_DST |
1749 OFPUTIL_A_SET_NW_SRC |
1750 OFPUTIL_A_SET_NW_DST |
1751 OFPUTIL_A_SET_NW_TOS |
1752 OFPUTIL_A_SET_TP_SRC |
1753 OFPUTIL_A_SET_TP_DST |
1758 get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots)
1760 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1761 struct dpif_dp_stats s;
1762 uint64_t n_miss, n_no_pkt_in, n_bytes, n_dropped_frags;
1765 strcpy(ots->name, "classifier");
1767 dpif_get_dp_stats(ofproto->backer->dpif, &s);
1768 rule_get_stats(&ofproto->miss_rule->up, &n_miss, &n_bytes);
1769 rule_get_stats(&ofproto->no_packet_in_rule->up, &n_no_pkt_in, &n_bytes);
1770 rule_get_stats(&ofproto->drop_frags_rule->up, &n_dropped_frags, &n_bytes);
1772 n_lookup = s.n_hit + s.n_missed - n_dropped_frags;
1773 ots->lookup_count = htonll(n_lookup);
1774 ots->matched_count = htonll(n_lookup - n_miss - n_no_pkt_in);
1777 static struct ofport *
1780 struct ofport_dpif *port = xmalloc(sizeof *port);
1785 port_dealloc(struct ofport *port_)
1787 struct ofport_dpif *port = ofport_dpif_cast(port_);
1792 port_construct(struct ofport *port_)
1794 struct ofport_dpif *port = ofport_dpif_cast(port_);
1795 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1796 const struct netdev *netdev = port->up.netdev;
1797 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1798 struct dpif_port dpif_port;
1801 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1802 port->bundle = NULL;
1805 port->tag = tag_create_random();
1806 port->may_enable = true;
1807 port->stp_port = NULL;
1808 port->stp_state = STP_DISABLED;
1809 port->tnl_port = NULL;
1810 hmap_init(&port->priorities);
1811 port->realdev_ofp_port = 0;
1812 port->vlandev_vid = 0;
1813 port->carrier_seq = netdev_get_carrier_resets(netdev);
1815 if (netdev_vport_is_patch(netdev)) {
1816 /* By bailing out here, we don't submit the port to the sFlow module
1817 * to be considered for counter polling export. This is correct
1818 * because the patch port represents an interface that sFlow considers
1819 * to be "internal" to the switch as a whole, and therefore not an
1820 * candidate for counter polling. */
1821 port->odp_port = OVSP_NONE;
1825 error = dpif_port_query_by_name(ofproto->backer->dpif,
1826 netdev_vport_get_dpif_port(netdev, namebuf,
1833 port->odp_port = dpif_port.port_no;
1835 if (netdev_get_tunnel_config(netdev)) {
1836 port->tnl_port = tnl_port_add(&port->up, port->odp_port);
1838 /* Sanity-check that a mapping doesn't already exist. This
1839 * shouldn't happen for non-tunnel ports. */
1840 if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1841 VLOG_ERR("port %s already has an OpenFlow port number",
1843 dpif_port_destroy(&dpif_port);
1847 hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1848 hash_int(port->odp_port, 0));
1850 dpif_port_destroy(&dpif_port);
1852 if (ofproto->sflow) {
1853 dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1860 port_destruct(struct ofport *port_)
1862 struct ofport_dpif *port = ofport_dpif_cast(port_);
1863 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1864 const char *devname = netdev_get_name(port->up.netdev);
1865 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1866 const char *dp_port_name;
1868 dp_port_name = netdev_vport_get_dpif_port(port->up.netdev, namebuf,
1870 if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1871 /* The underlying device is still there, so delete it. This
1872 * happens when the ofproto is being destroyed, since the caller
1873 * assumes that removal of attached ports will happen as part of
1875 if (!port->tnl_port) {
1876 dpif_port_del(ofproto->backer->dpif, port->odp_port);
1878 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1881 if (port->odp_port != OVSP_NONE && !port->tnl_port) {
1882 hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1885 tnl_port_del(port->tnl_port);
1886 sset_find_and_delete(&ofproto->ports, devname);
1887 sset_find_and_delete(&ofproto->ghost_ports, devname);
1888 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1889 bundle_remove(port_);
1890 set_cfm(port_, NULL);
1891 set_bfd(port_, NULL);
1892 if (ofproto->sflow) {
1893 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1896 ofport_clear_priorities(port);
1897 hmap_destroy(&port->priorities);
1901 port_modified(struct ofport *port_)
1903 struct ofport_dpif *port = ofport_dpif_cast(port_);
1905 if (port->bundle && port->bundle->bond) {
1906 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1910 cfm_set_netdev(port->cfm, port->up.netdev);
1915 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1917 struct ofport_dpif *port = ofport_dpif_cast(port_);
1918 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1919 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1921 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1922 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1923 OFPUTIL_PC_NO_PACKET_IN)) {
1924 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1926 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1927 bundle_update(port->bundle);
1933 set_sflow(struct ofproto *ofproto_,
1934 const struct ofproto_sflow_options *sflow_options)
1936 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1937 struct dpif_sflow *ds = ofproto->sflow;
1939 if (sflow_options) {
1941 struct ofport_dpif *ofport;
1943 ds = ofproto->sflow = dpif_sflow_create();
1944 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1945 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1947 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1949 dpif_sflow_set_options(ds, sflow_options);
1952 dpif_sflow_destroy(ds);
1953 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1954 ofproto->sflow = NULL;
1962 struct ofproto *ofproto_,
1963 const struct ofproto_ipfix_bridge_exporter_options *bridge_exporter_options,
1964 const struct ofproto_ipfix_flow_exporter_options *flow_exporters_options,
1965 size_t n_flow_exporters_options)
1967 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1968 struct dpif_ipfix *di = ofproto->ipfix;
1970 if (bridge_exporter_options || flow_exporters_options) {
1972 di = ofproto->ipfix = dpif_ipfix_create();
1974 dpif_ipfix_set_options(
1975 di, bridge_exporter_options, flow_exporters_options,
1976 n_flow_exporters_options);
1979 dpif_ipfix_destroy(di);
1980 ofproto->ipfix = NULL;
1987 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1989 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1996 struct ofproto_dpif *ofproto;
1998 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1999 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2000 ofport->cfm = cfm_create(ofport->up.netdev);
2003 if (cfm_configure(ofport->cfm, s)) {
2009 cfm_destroy(ofport->cfm);
2015 get_cfm_status(const struct ofport *ofport_,
2016 struct ofproto_cfm_status *status)
2018 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2021 status->faults = cfm_get_fault(ofport->cfm);
2022 status->remote_opstate = cfm_get_opup(ofport->cfm);
2023 status->health = cfm_get_health(ofport->cfm);
2024 cfm_get_remote_mpids(ofport->cfm, &status->rmps, &status->n_rmps);
2032 set_bfd(struct ofport *ofport_, const struct smap *cfg)
2034 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
2035 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2039 ofport->bfd = bfd_configure(old, netdev_get_name(ofport->up.netdev), cfg);
2040 if (ofport->bfd != old) {
2041 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2048 get_bfd_status(struct ofport *ofport_, struct smap *smap)
2050 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2053 bfd_get_status(ofport->bfd, smap);
2060 /* Spanning Tree. */
2063 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
2065 struct ofproto_dpif *ofproto = ofproto_;
2066 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
2067 struct ofport_dpif *ofport;
2069 ofport = stp_port_get_aux(sp);
2071 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
2072 ofproto->up.name, port_num);
2074 struct eth_header *eth = pkt->l2;
2076 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
2077 if (eth_addr_is_zero(eth->eth_src)) {
2078 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
2079 "with unknown MAC", ofproto->up.name, port_num);
2081 send_packet(ofport, pkt);
2087 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
2089 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
2091 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2093 /* Only revalidate flows if the configuration changed. */
2094 if (!s != !ofproto->stp) {
2095 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2099 if (!ofproto->stp) {
2100 ofproto->stp = stp_create(ofproto_->name, s->system_id,
2101 send_bpdu_cb, ofproto);
2102 ofproto->stp_last_tick = time_msec();
2105 stp_set_bridge_id(ofproto->stp, s->system_id);
2106 stp_set_bridge_priority(ofproto->stp, s->priority);
2107 stp_set_hello_time(ofproto->stp, s->hello_time);
2108 stp_set_max_age(ofproto->stp, s->max_age);
2109 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
2111 struct ofport *ofport;
2113 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
2114 set_stp_port(ofport, NULL);
2117 stp_destroy(ofproto->stp);
2118 ofproto->stp = NULL;
2125 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
2127 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2131 s->bridge_id = stp_get_bridge_id(ofproto->stp);
2132 s->designated_root = stp_get_designated_root(ofproto->stp);
2133 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
2142 update_stp_port_state(struct ofport_dpif *ofport)
2144 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2145 enum stp_state state;
2147 /* Figure out new state. */
2148 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
2152 if (ofport->stp_state != state) {
2153 enum ofputil_port_state of_state;
2156 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
2157 netdev_get_name(ofport->up.netdev),
2158 stp_state_name(ofport->stp_state),
2159 stp_state_name(state));
2160 if (stp_learn_in_state(ofport->stp_state)
2161 != stp_learn_in_state(state)) {
2162 /* xxx Learning action flows should also be flushed. */
2163 mac_learning_flush(ofproto->ml,
2164 &ofproto->backer->revalidate_set);
2166 fwd_change = stp_forward_in_state(ofport->stp_state)
2167 != stp_forward_in_state(state);
2169 ofproto->backer->need_revalidate = REV_STP;
2170 ofport->stp_state = state;
2171 ofport->stp_state_entered = time_msec();
2173 if (fwd_change && ofport->bundle) {
2174 bundle_update(ofport->bundle);
2177 /* Update the STP state bits in the OpenFlow port description. */
2178 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
2179 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
2180 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
2181 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
2182 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
2184 ofproto_port_set_state(&ofport->up, of_state);
2188 /* Configures STP on 'ofport_' using the settings defined in 's'. The
2189 * caller is responsible for assigning STP port numbers and ensuring
2190 * there are no duplicates. */
2192 set_stp_port(struct ofport *ofport_,
2193 const struct ofproto_port_stp_settings *s)
2195 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2196 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2197 struct stp_port *sp = ofport->stp_port;
2199 if (!s || !s->enable) {
2201 ofport->stp_port = NULL;
2202 stp_port_disable(sp);
2203 update_stp_port_state(ofport);
2206 } else if (sp && stp_port_no(sp) != s->port_num
2207 && ofport == stp_port_get_aux(sp)) {
2208 /* The port-id changed, so disable the old one if it's not
2209 * already in use by another port. */
2210 stp_port_disable(sp);
2213 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
2214 stp_port_enable(sp);
2216 stp_port_set_aux(sp, ofport);
2217 stp_port_set_priority(sp, s->priority);
2218 stp_port_set_path_cost(sp, s->path_cost);
2220 update_stp_port_state(ofport);
2226 get_stp_port_status(struct ofport *ofport_,
2227 struct ofproto_port_stp_status *s)
2229 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2230 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2231 struct stp_port *sp = ofport->stp_port;
2233 if (!ofproto->stp || !sp) {
2239 s->port_id = stp_port_get_id(sp);
2240 s->state = stp_port_get_state(sp);
2241 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
2242 s->role = stp_port_get_role(sp);
2243 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
2249 stp_run(struct ofproto_dpif *ofproto)
2252 long long int now = time_msec();
2253 long long int elapsed = now - ofproto->stp_last_tick;
2254 struct stp_port *sp;
2257 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
2258 ofproto->stp_last_tick = now;
2260 while (stp_get_changed_port(ofproto->stp, &sp)) {
2261 struct ofport_dpif *ofport = stp_port_get_aux(sp);
2264 update_stp_port_state(ofport);
2268 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
2269 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
2275 stp_wait(struct ofproto_dpif *ofproto)
2278 poll_timer_wait(1000);
2282 /* Returns true if STP should process 'flow'. */
2284 stp_should_process_flow(const struct flow *flow)
2286 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
2290 stp_process_packet(const struct ofport_dpif *ofport,
2291 const struct ofpbuf *packet)
2293 struct ofpbuf payload = *packet;
2294 struct eth_header *eth = payload.data;
2295 struct stp_port *sp = ofport->stp_port;
2297 /* Sink packets on ports that have STP disabled when the bridge has
2299 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
2303 /* Trim off padding on payload. */
2304 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
2305 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
2308 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
2309 stp_received_bpdu(sp, payload.data, payload.size);
2313 static struct priority_to_dscp *
2314 get_priority(const struct ofport_dpif *ofport, uint32_t priority)
2316 struct priority_to_dscp *pdscp;
2319 hash = hash_int(priority, 0);
2320 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) {
2321 if (pdscp->priority == priority) {
2329 ofport_clear_priorities(struct ofport_dpif *ofport)
2331 struct priority_to_dscp *pdscp, *next;
2333 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) {
2334 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
2340 set_queues(struct ofport *ofport_,
2341 const struct ofproto_port_queue *qdscp_list,
2344 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2345 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2346 struct hmap new = HMAP_INITIALIZER(&new);
2349 for (i = 0; i < n_qdscp; i++) {
2350 struct priority_to_dscp *pdscp;
2354 dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
2355 if (dpif_queue_to_priority(ofproto->backer->dpif, qdscp_list[i].queue,
2360 pdscp = get_priority(ofport, priority);
2362 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
2364 pdscp = xmalloc(sizeof *pdscp);
2365 pdscp->priority = priority;
2367 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2370 if (pdscp->dscp != dscp) {
2372 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2375 hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0));
2378 if (!hmap_is_empty(&ofport->priorities)) {
2379 ofport_clear_priorities(ofport);
2380 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2383 hmap_swap(&new, &ofport->priorities);
2391 /* Expires all MAC learning entries associated with 'bundle' and forces its
2392 * ofproto to revalidate every flow.
2394 * Normally MAC learning entries are removed only from the ofproto associated
2395 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2396 * are removed from every ofproto. When patch ports and SLB bonds are in use
2397 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2398 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2399 * with the host from which it migrated. */
2401 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2403 struct ofproto_dpif *ofproto = bundle->ofproto;
2404 struct mac_learning *ml = ofproto->ml;
2405 struct mac_entry *mac, *next_mac;
2407 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2408 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2409 if (mac->port.p == bundle) {
2411 struct ofproto_dpif *o;
2413 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2415 struct mac_entry *e;
2417 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan,
2420 mac_learning_expire(o->ml, e);
2426 mac_learning_expire(ml, mac);
2431 static struct ofbundle *
2432 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2434 struct ofbundle *bundle;
2436 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2437 &ofproto->bundles) {
2438 if (bundle->aux == aux) {
2445 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
2446 * ones that are found to 'bundles'. */
2448 bundle_lookup_multiple(struct ofproto_dpif *ofproto,
2449 void **auxes, size_t n_auxes,
2450 struct hmapx *bundles)
2454 hmapx_init(bundles);
2455 for (i = 0; i < n_auxes; i++) {
2456 struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
2458 hmapx_add(bundles, bundle);
2464 bundle_update(struct ofbundle *bundle)
2466 struct ofport_dpif *port;
2468 bundle->floodable = true;
2469 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2470 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2471 || !stp_forward_in_state(port->stp_state)) {
2472 bundle->floodable = false;
2479 bundle_del_port(struct ofport_dpif *port)
2481 struct ofbundle *bundle = port->bundle;
2483 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2485 list_remove(&port->bundle_node);
2486 port->bundle = NULL;
2489 lacp_slave_unregister(bundle->lacp, port);
2492 bond_slave_unregister(bundle->bond, port);
2495 bundle_update(bundle);
2499 bundle_add_port(struct ofbundle *bundle, uint16_t ofp_port,
2500 struct lacp_slave_settings *lacp)
2502 struct ofport_dpif *port;
2504 port = get_ofp_port(bundle->ofproto, ofp_port);
2509 if (port->bundle != bundle) {
2510 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2512 bundle_del_port(port);
2515 port->bundle = bundle;
2516 list_push_back(&bundle->ports, &port->bundle_node);
2517 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2518 || !stp_forward_in_state(port->stp_state)) {
2519 bundle->floodable = false;
2523 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2524 lacp_slave_register(bundle->lacp, port, lacp);
2531 bundle_destroy(struct ofbundle *bundle)
2533 struct ofproto_dpif *ofproto;
2534 struct ofport_dpif *port, *next_port;
2541 ofproto = bundle->ofproto;
2542 for (i = 0; i < MAX_MIRRORS; i++) {
2543 struct ofmirror *m = ofproto->mirrors[i];
2545 if (m->out == bundle) {
2547 } else if (hmapx_find_and_delete(&m->srcs, bundle)
2548 || hmapx_find_and_delete(&m->dsts, bundle)) {
2549 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2554 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2555 bundle_del_port(port);
2558 bundle_flush_macs(bundle, true);
2559 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2561 free(bundle->trunks);
2562 lacp_destroy(bundle->lacp);
2563 bond_destroy(bundle->bond);
2568 bundle_set(struct ofproto *ofproto_, void *aux,
2569 const struct ofproto_bundle_settings *s)
2571 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2572 bool need_flush = false;
2573 struct ofport_dpif *port;
2574 struct ofbundle *bundle;
2575 unsigned long *trunks;
2581 bundle_destroy(bundle_lookup(ofproto, aux));
2585 ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2586 ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2588 bundle = bundle_lookup(ofproto, aux);
2590 bundle = xmalloc(sizeof *bundle);
2592 bundle->ofproto = ofproto;
2593 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2594 hash_pointer(aux, 0));
2596 bundle->name = NULL;
2598 list_init(&bundle->ports);
2599 bundle->vlan_mode = PORT_VLAN_TRUNK;
2601 bundle->trunks = NULL;
2602 bundle->use_priority_tags = s->use_priority_tags;
2603 bundle->lacp = NULL;
2604 bundle->bond = NULL;
2606 bundle->floodable = true;
2608 bundle->src_mirrors = 0;
2609 bundle->dst_mirrors = 0;
2610 bundle->mirror_out = 0;
2613 if (!bundle->name || strcmp(s->name, bundle->name)) {
2615 bundle->name = xstrdup(s->name);
2620 if (!bundle->lacp) {
2621 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2622 bundle->lacp = lacp_create();
2624 lacp_configure(bundle->lacp, s->lacp);
2626 lacp_destroy(bundle->lacp);
2627 bundle->lacp = NULL;
2630 /* Update set of ports. */
2632 for (i = 0; i < s->n_slaves; i++) {
2633 if (!bundle_add_port(bundle, s->slaves[i],
2634 s->lacp ? &s->lacp_slaves[i] : NULL)) {
2638 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2639 struct ofport_dpif *next_port;
2641 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2642 for (i = 0; i < s->n_slaves; i++) {
2643 if (s->slaves[i] == port->up.ofp_port) {
2648 bundle_del_port(port);
2652 ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2654 if (list_is_empty(&bundle->ports)) {
2655 bundle_destroy(bundle);
2659 /* Set VLAN tagging mode */
2660 if (s->vlan_mode != bundle->vlan_mode
2661 || s->use_priority_tags != bundle->use_priority_tags) {
2662 bundle->vlan_mode = s->vlan_mode;
2663 bundle->use_priority_tags = s->use_priority_tags;
2668 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2669 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2671 if (vlan != bundle->vlan) {
2672 bundle->vlan = vlan;
2676 /* Get trunked VLANs. */
2677 switch (s->vlan_mode) {
2678 case PORT_VLAN_ACCESS:
2682 case PORT_VLAN_TRUNK:
2683 trunks = CONST_CAST(unsigned long *, s->trunks);
2686 case PORT_VLAN_NATIVE_UNTAGGED:
2687 case PORT_VLAN_NATIVE_TAGGED:
2688 if (vlan != 0 && (!s->trunks
2689 || !bitmap_is_set(s->trunks, vlan)
2690 || bitmap_is_set(s->trunks, 0))) {
2691 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2693 trunks = bitmap_clone(s->trunks, 4096);
2695 trunks = bitmap_allocate1(4096);
2697 bitmap_set1(trunks, vlan);
2698 bitmap_set0(trunks, 0);
2700 trunks = CONST_CAST(unsigned long *, s->trunks);
2707 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2708 free(bundle->trunks);
2709 if (trunks == s->trunks) {
2710 bundle->trunks = vlan_bitmap_clone(trunks);
2712 bundle->trunks = trunks;
2717 if (trunks != s->trunks) {
2722 if (!list_is_short(&bundle->ports)) {
2723 bundle->ofproto->has_bonded_bundles = true;
2725 if (bond_reconfigure(bundle->bond, s->bond)) {
2726 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2729 bundle->bond = bond_create(s->bond);
2730 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2733 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2734 bond_slave_register(bundle->bond, port, port->up.netdev);
2737 bond_destroy(bundle->bond);
2738 bundle->bond = NULL;
2741 /* If we changed something that would affect MAC learning, un-learn
2742 * everything on this port and force flow revalidation. */
2744 bundle_flush_macs(bundle, false);
2751 bundle_remove(struct ofport *port_)
2753 struct ofport_dpif *port = ofport_dpif_cast(port_);
2754 struct ofbundle *bundle = port->bundle;
2757 bundle_del_port(port);
2758 if (list_is_empty(&bundle->ports)) {
2759 bundle_destroy(bundle);
2760 } else if (list_is_short(&bundle->ports)) {
2761 bond_destroy(bundle->bond);
2762 bundle->bond = NULL;
2768 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2770 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2771 struct ofport_dpif *port = port_;
2772 uint8_t ea[ETH_ADDR_LEN];
2775 error = netdev_get_etheraddr(port->up.netdev, ea);
2777 struct ofpbuf packet;
2780 ofpbuf_init(&packet, 0);
2781 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2783 memcpy(packet_pdu, pdu, pdu_size);
2785 send_packet(port, &packet);
2786 ofpbuf_uninit(&packet);
2788 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2789 "%s (%s)", port->bundle->name,
2790 netdev_get_name(port->up.netdev), strerror(error));
2795 bundle_send_learning_packets(struct ofbundle *bundle)
2797 struct ofproto_dpif *ofproto = bundle->ofproto;
2798 int error, n_packets, n_errors;
2799 struct mac_entry *e;
2801 error = n_packets = n_errors = 0;
2802 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2803 if (e->port.p != bundle) {
2804 struct ofpbuf *learning_packet;
2805 struct ofport_dpif *port;
2809 /* The assignment to "port" is unnecessary but makes "grep"ing for
2810 * struct ofport_dpif more effective. */
2811 learning_packet = bond_compose_learning_packet(bundle->bond,
2815 ret = send_packet(port, learning_packet);
2816 ofpbuf_delete(learning_packet);
2826 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2827 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2828 "packets, last error was: %s",
2829 bundle->name, n_errors, n_packets, strerror(error));
2831 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2832 bundle->name, n_packets);
2837 bundle_run(struct ofbundle *bundle)
2840 lacp_run(bundle->lacp, send_pdu_cb);
2843 struct ofport_dpif *port;
2845 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2846 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2849 bond_run(bundle->bond, &bundle->ofproto->backer->revalidate_set,
2850 lacp_status(bundle->lacp));
2851 if (bond_should_send_learning_packets(bundle->bond)) {
2852 bundle_send_learning_packets(bundle);
2858 bundle_wait(struct ofbundle *bundle)
2861 lacp_wait(bundle->lacp);
2864 bond_wait(bundle->bond);
2871 mirror_scan(struct ofproto_dpif *ofproto)
2875 for (idx = 0; idx < MAX_MIRRORS; idx++) {
2876 if (!ofproto->mirrors[idx]) {
2883 static struct ofmirror *
2884 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
2888 for (i = 0; i < MAX_MIRRORS; i++) {
2889 struct ofmirror *mirror = ofproto->mirrors[i];
2890 if (mirror && mirror->aux == aux) {
2898 /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */
2900 mirror_update_dups(struct ofproto_dpif *ofproto)
2904 for (i = 0; i < MAX_MIRRORS; i++) {
2905 struct ofmirror *m = ofproto->mirrors[i];
2908 m->dup_mirrors = MIRROR_MASK_C(1) << i;
2912 for (i = 0; i < MAX_MIRRORS; i++) {
2913 struct ofmirror *m1 = ofproto->mirrors[i];
2920 for (j = i + 1; j < MAX_MIRRORS; j++) {
2921 struct ofmirror *m2 = ofproto->mirrors[j];
2923 if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) {
2924 m1->dup_mirrors |= MIRROR_MASK_C(1) << j;
2925 m2->dup_mirrors |= m1->dup_mirrors;
2932 mirror_set(struct ofproto *ofproto_, void *aux,
2933 const struct ofproto_mirror_settings *s)
2935 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2936 mirror_mask_t mirror_bit;
2937 struct ofbundle *bundle;
2938 struct ofmirror *mirror;
2939 struct ofbundle *out;
2940 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
2941 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
2944 mirror = mirror_lookup(ofproto, aux);
2946 mirror_destroy(mirror);
2952 idx = mirror_scan(ofproto);
2954 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
2956 ofproto->up.name, MAX_MIRRORS, s->name);
2960 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
2961 mirror->ofproto = ofproto;
2964 mirror->out_vlan = -1;
2965 mirror->name = NULL;
2968 if (!mirror->name || strcmp(s->name, mirror->name)) {
2970 mirror->name = xstrdup(s->name);
2973 /* Get the new configuration. */
2974 if (s->out_bundle) {
2975 out = bundle_lookup(ofproto, s->out_bundle);
2977 mirror_destroy(mirror);
2983 out_vlan = s->out_vlan;
2985 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
2986 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
2988 /* If the configuration has not changed, do nothing. */
2989 if (hmapx_equals(&srcs, &mirror->srcs)
2990 && hmapx_equals(&dsts, &mirror->dsts)
2991 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
2992 && mirror->out == out
2993 && mirror->out_vlan == out_vlan)
2995 hmapx_destroy(&srcs);
2996 hmapx_destroy(&dsts);
3000 hmapx_swap(&srcs, &mirror->srcs);
3001 hmapx_destroy(&srcs);
3003 hmapx_swap(&dsts, &mirror->dsts);
3004 hmapx_destroy(&dsts);
3006 free(mirror->vlans);
3007 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
3010 mirror->out_vlan = out_vlan;
3012 /* Update bundles. */
3013 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
3014 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
3015 if (hmapx_contains(&mirror->srcs, bundle)) {
3016 bundle->src_mirrors |= mirror_bit;
3018 bundle->src_mirrors &= ~mirror_bit;
3021 if (hmapx_contains(&mirror->dsts, bundle)) {
3022 bundle->dst_mirrors |= mirror_bit;
3024 bundle->dst_mirrors &= ~mirror_bit;
3027 if (mirror->out == bundle) {
3028 bundle->mirror_out |= mirror_bit;
3030 bundle->mirror_out &= ~mirror_bit;
3034 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3035 ofproto->has_mirrors = true;
3036 mac_learning_flush(ofproto->ml,
3037 &ofproto->backer->revalidate_set);
3038 mirror_update_dups(ofproto);
3044 mirror_destroy(struct ofmirror *mirror)
3046 struct ofproto_dpif *ofproto;
3047 mirror_mask_t mirror_bit;
3048 struct ofbundle *bundle;
3055 ofproto = mirror->ofproto;
3056 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3057 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
3059 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
3060 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
3061 bundle->src_mirrors &= ~mirror_bit;
3062 bundle->dst_mirrors &= ~mirror_bit;
3063 bundle->mirror_out &= ~mirror_bit;
3066 hmapx_destroy(&mirror->srcs);
3067 hmapx_destroy(&mirror->dsts);
3068 free(mirror->vlans);
3070 ofproto->mirrors[mirror->idx] = NULL;
3074 mirror_update_dups(ofproto);
3076 ofproto->has_mirrors = false;
3077 for (i = 0; i < MAX_MIRRORS; i++) {
3078 if (ofproto->mirrors[i]) {
3079 ofproto->has_mirrors = true;
3086 mirror_get_stats(struct ofproto *ofproto_, void *aux,
3087 uint64_t *packets, uint64_t *bytes)
3089 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3090 struct ofmirror *mirror = mirror_lookup(ofproto, aux);
3093 *packets = *bytes = UINT64_MAX;
3099 *packets = mirror->packet_count;
3100 *bytes = mirror->byte_count;
3106 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
3108 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3109 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
3110 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
3116 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
3118 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3119 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
3120 return bundle && bundle->mirror_out != 0;
3124 forward_bpdu_changed(struct ofproto *ofproto_)
3126 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3127 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3131 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
3134 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3135 mac_learning_set_idle_time(ofproto->ml, idle_time);
3136 mac_learning_set_max_entries(ofproto->ml, max_entries);
3141 static struct ofport_dpif *
3142 get_ofp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
3144 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
3145 return ofport ? ofport_dpif_cast(ofport) : NULL;
3148 static struct ofport_dpif *
3149 get_odp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
3151 struct ofport_dpif *port = odp_port_to_ofport(ofproto->backer, odp_port);
3152 return port && &ofproto->up == port->up.ofproto ? port : NULL;
3156 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
3157 struct ofproto_port *ofproto_port,
3158 struct dpif_port *dpif_port)
3160 ofproto_port->name = dpif_port->name;
3161 ofproto_port->type = dpif_port->type;
3162 ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
3165 static struct ofport_dpif *
3166 ofport_get_peer(const struct ofport_dpif *ofport_dpif)
3168 const struct ofproto_dpif *ofproto;
3171 peer = netdev_vport_patch_peer(ofport_dpif->up.netdev);
3176 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
3177 struct ofport *ofport;
3179 ofport = shash_find_data(&ofproto->up.port_by_name, peer);
3180 if (ofport && ofport->ofproto->ofproto_class == &ofproto_dpif_class) {
3181 return ofport_dpif_cast(ofport);
3188 port_run_fast(struct ofport_dpif *ofport)
3190 if (ofport->cfm && cfm_should_send_ccm(ofport->cfm)) {
3191 struct ofpbuf packet;
3193 ofpbuf_init(&packet, 0);
3194 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
3195 send_packet(ofport, &packet);
3196 ofpbuf_uninit(&packet);
3199 if (ofport->bfd && bfd_should_send_packet(ofport->bfd)) {
3200 struct ofpbuf packet;
3202 ofpbuf_init(&packet, 0);
3203 bfd_put_packet(ofport->bfd, &packet, ofport->up.pp.hw_addr);
3204 send_packet(ofport, &packet);
3205 ofpbuf_uninit(&packet);
3210 port_run(struct ofport_dpif *ofport)
3212 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
3213 bool carrier_changed = carrier_seq != ofport->carrier_seq;
3214 bool enable = netdev_get_carrier(ofport->up.netdev);
3216 ofport->carrier_seq = carrier_seq;
3218 port_run_fast(ofport);
3220 if (ofport->tnl_port
3221 && tnl_port_reconfigure(&ofport->up, ofport->odp_port,
3222 &ofport->tnl_port)) {
3223 ofproto_dpif_cast(ofport->up.ofproto)->backer->need_revalidate = true;
3227 int cfm_opup = cfm_get_opup(ofport->cfm);
3229 cfm_run(ofport->cfm);
3230 enable = enable && !cfm_get_fault(ofport->cfm);
3232 if (cfm_opup >= 0) {
3233 enable = enable && cfm_opup;
3238 bfd_run(ofport->bfd);
3239 enable = enable && bfd_forwarding(ofport->bfd);
3242 if (ofport->bundle) {
3243 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
3244 if (carrier_changed) {
3245 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
3249 if (ofport->may_enable != enable) {
3250 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3252 if (ofproto->has_bundle_action) {
3253 ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
3257 ofport->may_enable = enable;
3261 port_wait(struct ofport_dpif *ofport)
3264 cfm_wait(ofport->cfm);
3268 bfd_wait(ofport->bfd);
3273 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
3274 struct ofproto_port *ofproto_port)
3276 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3277 struct dpif_port dpif_port;
3280 if (sset_contains(&ofproto->ghost_ports, devname)) {
3281 const char *type = netdev_get_type_from_name(devname);
3283 /* We may be called before ofproto->up.port_by_name is populated with
3284 * the appropriate ofport. For this reason, we must get the name and
3285 * type from the netdev layer directly. */
3287 const struct ofport *ofport;
3289 ofport = shash_find_data(&ofproto->up.port_by_name, devname);
3290 ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
3291 ofproto_port->name = xstrdup(devname);
3292 ofproto_port->type = xstrdup(type);
3298 if (!sset_contains(&ofproto->ports, devname)) {
3301 error = dpif_port_query_by_name(ofproto->backer->dpif,
3302 devname, &dpif_port);
3304 ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
3310 port_add(struct ofproto *ofproto_, struct netdev *netdev)
3312 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3313 const char *devname = netdev_get_name(netdev);
3314 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
3315 const char *dp_port_name;
3317 if (netdev_vport_is_patch(netdev)) {
3318 sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
3322 dp_port_name = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
3323 if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
3324 uint32_t port_no = UINT32_MAX;
3327 error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
3331 if (netdev_get_tunnel_config(netdev)) {
3332 simap_put(&ofproto->backer->tnl_backers, dp_port_name, port_no);
3336 if (netdev_get_tunnel_config(netdev)) {
3337 sset_add(&ofproto->ghost_ports, devname);
3339 sset_add(&ofproto->ports, devname);
3345 port_del(struct ofproto *ofproto_, uint16_t ofp_port)
3347 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3348 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
3355 sset_find_and_delete(&ofproto->ghost_ports,
3356 netdev_get_name(ofport->up.netdev));
3357 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3358 if (!ofport->tnl_port) {
3359 error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
3361 /* The caller is going to close ofport->up.netdev. If this is a
3362 * bonded port, then the bond is using that netdev, so remove it
3363 * from the bond. The client will need to reconfigure everything
3364 * after deleting ports, so then the slave will get re-added. */
3365 bundle_remove(&ofport->up);
3372 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
3374 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3379 error = netdev_get_stats(ofport->up.netdev, stats);
3381 if (!error && ofport_->ofp_port == OFPP_LOCAL) {
3382 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3384 /* ofproto->stats.tx_packets represents packets that we created
3385 * internally and sent to some port (e.g. packets sent with
3386 * send_packet()). Account for them as if they had come from
3387 * OFPP_LOCAL and got forwarded. */
3389 if (stats->rx_packets != UINT64_MAX) {
3390 stats->rx_packets += ofproto->stats.tx_packets;
3393 if (stats->rx_bytes != UINT64_MAX) {
3394 stats->rx_bytes += ofproto->stats.tx_bytes;
3397 /* ofproto->stats.rx_packets represents packets that were received on
3398 * some port and we processed internally and dropped (e.g. STP).
3399 * Account for them as if they had been forwarded to OFPP_LOCAL. */
3401 if (stats->tx_packets != UINT64_MAX) {
3402 stats->tx_packets += ofproto->stats.rx_packets;
3405 if (stats->tx_bytes != UINT64_MAX) {
3406 stats->tx_bytes += ofproto->stats.rx_bytes;
3413 struct port_dump_state {
3418 struct ofproto_port port;
3423 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
3425 *statep = xzalloc(sizeof(struct port_dump_state));
3430 port_dump_next(const struct ofproto *ofproto_, void *state_,
3431 struct ofproto_port *port)
3433 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3434 struct port_dump_state *state = state_;
3435 const struct sset *sset;
3436 struct sset_node *node;
3438 if (state->has_port) {
3439 ofproto_port_destroy(&state->port);
3440 state->has_port = false;
3442 sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
3443 while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
3446 error = port_query_by_name(ofproto_, node->name, &state->port);
3448 *port = state->port;
3449 state->has_port = true;
3451 } else if (error != ENODEV) {
3456 if (!state->ghost) {
3457 state->ghost = true;
3460 return port_dump_next(ofproto_, state_, port);
3467 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3469 struct port_dump_state *state = state_;
3471 if (state->has_port) {
3472 ofproto_port_destroy(&state->port);
3479 port_poll(const struct ofproto *ofproto_, char **devnamep)
3481 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3483 if (ofproto->port_poll_errno) {
3484 int error = ofproto->port_poll_errno;
3485 ofproto->port_poll_errno = 0;
3489 if (sset_is_empty(&ofproto->port_poll_set)) {
3493 *devnamep = sset_pop(&ofproto->port_poll_set);
3498 port_poll_wait(const struct ofproto *ofproto_)
3500 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3501 dpif_port_poll_wait(ofproto->backer->dpif);
3505 port_is_lacp_current(const struct ofport *ofport_)
3507 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3508 return (ofport->bundle && ofport->bundle->lacp
3509 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3513 /* Upcall handling. */
3515 /* Flow miss batching.
3517 * Some dpifs implement operations faster when you hand them off in a batch.
3518 * To allow batching, "struct flow_miss" queues the dpif-related work needed
3519 * for a given flow. Each "struct flow_miss" corresponds to sending one or
3520 * more packets, plus possibly installing the flow in the dpif.
3522 * So far we only batch the operations that affect flow setup time the most.
3523 * It's possible to batch more than that, but the benefit might be minimal. */
3525 struct hmap_node hmap_node;
3526 struct ofproto_dpif *ofproto;
3528 enum odp_key_fitness key_fitness;
3529 const struct nlattr *key;
3531 struct initial_vals initial_vals;
3532 struct list packets;
3533 enum dpif_upcall_type upcall_type;
3536 struct flow_miss_op {
3537 struct dpif_op dpif_op;
3539 uint64_t slow_stub[128 / 8]; /* Buffer for compose_slow_path() */
3540 struct xlate_out xout;
3541 bool xout_garbage; /* 'xout' needs to be uninitialized? */
3544 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
3545 * OpenFlow controller as necessary according to their individual
3546 * configurations. */
3548 send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet,
3549 const struct flow *flow)
3551 struct ofputil_packet_in pin;
3553 pin.packet = packet->data;
3554 pin.packet_len = packet->size;
3555 pin.reason = OFPR_NO_MATCH;
3556 pin.controller_id = 0;
3561 pin.send_len = 0; /* not used for flow table misses */
3563 flow_get_metadata(flow, &pin.fmd);
3565 connmgr_send_packet_in(ofproto->up.connmgr, &pin);
3568 static enum slow_path_reason
3569 process_special(struct ofproto_dpif *ofproto, const struct flow *flow,
3570 const struct ofport_dpif *ofport, const struct ofpbuf *packet)
3574 } else if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) {
3576 cfm_process_heartbeat(ofport->cfm, packet);
3579 } else if (ofport->bfd && bfd_should_process_flow(flow)) {
3581 bfd_process_packet(ofport->bfd, flow, packet);
3584 } else if (ofport->bundle && ofport->bundle->lacp
3585 && flow->dl_type == htons(ETH_TYPE_LACP)) {
3587 lacp_process_packet(ofport->bundle->lacp, ofport, packet);
3590 } else if (ofproto->stp && stp_should_process_flow(flow)) {
3592 stp_process_packet(ofport, packet);
3600 static struct flow_miss *
3601 flow_miss_find(struct hmap *todo, const struct ofproto_dpif *ofproto,
3602 const struct flow *flow, uint32_t hash)
3604 struct flow_miss *miss;
3606 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
3607 if (miss->ofproto == ofproto && flow_equal(&miss->flow, flow)) {
3615 /* Partially Initializes 'op' as an "execute" operation for 'miss' and
3616 * 'packet'. The caller must initialize op->actions and op->actions_len. If
3617 * 'miss' is associated with a subfacet the caller must also initialize the
3618 * returned op->subfacet, and if anything needs to be freed after processing
3619 * the op, the caller must initialize op->garbage also. */
3621 init_flow_miss_execute_op(struct flow_miss *miss, struct ofpbuf *packet,
3622 struct flow_miss_op *op)
3624 if (miss->flow.vlan_tci != miss->initial_vals.vlan_tci) {
3625 /* This packet was received on a VLAN splinter port. We
3626 * added a VLAN to the packet to make the packet resemble
3627 * the flow, but the actions were composed assuming that
3628 * the packet contained no VLAN. So, we must remove the
3629 * VLAN header from the packet before trying to execute the
3631 eth_pop_vlan(packet);
3634 op->xout_garbage = false;
3635 op->dpif_op.type = DPIF_OP_EXECUTE;
3636 op->dpif_op.u.execute.key = miss->key;
3637 op->dpif_op.u.execute.key_len = miss->key_len;
3638 op->dpif_op.u.execute.packet = packet;
3641 /* Helper for handle_flow_miss_without_facet() and
3642 * handle_flow_miss_with_facet(). */
3644 handle_flow_miss_common(struct rule_dpif *rule,
3645 struct ofpbuf *packet, const struct flow *flow)
3647 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3649 if (rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
3651 * Extra-special case for fail-open mode.
3653 * We are in fail-open mode and the packet matched the fail-open
3654 * rule, but we are connected to a controller too. We should send
3655 * the packet up to the controller in the hope that it will try to
3656 * set up a flow and thereby allow us to exit fail-open.
3658 * See the top-level comment in fail-open.c for more information.
3660 send_packet_in_miss(ofproto, packet, flow);
3664 /* Figures out whether a flow that missed in 'ofproto', whose details are in
3665 * 'miss', is likely to be worth tracking in detail in userspace and (usually)
3666 * installing a datapath flow. The answer is usually "yes" (a return value of
3667 * true). However, for short flows the cost of bookkeeping is much higher than
3668 * the benefits, so when the datapath holds a large number of flows we impose
3669 * some heuristics to decide which flows are likely to be worth tracking. */
3671 flow_miss_should_make_facet(struct ofproto_dpif *ofproto,
3672 struct flow_miss *miss, uint32_t hash)
3674 if (!ofproto->governor) {
3677 n_subfacets = hmap_count(&ofproto->subfacets);
3678 if (n_subfacets * 2 <= ofproto->up.flow_eviction_threshold) {
3682 ofproto->governor = governor_create(ofproto->up.name);
3685 return governor_should_install_flow(ofproto->governor, hash,
3686 list_size(&miss->packets));
3689 /* Handles 'miss', which matches 'rule', without creating a facet or subfacet
3690 * or creating any datapath flow. May add an "execute" operation to 'ops' and
3691 * increment '*n_ops'. */
3693 handle_flow_miss_without_facet(struct flow_miss *miss,
3694 struct flow_miss_op *ops, size_t *n_ops)
3696 struct rule_dpif *rule = rule_dpif_lookup(miss->ofproto, &miss->flow);
3697 long long int now = time_msec();
3698 struct ofpbuf *packet;
3699 struct xlate_in xin;
3701 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3702 struct flow_miss_op *op = &ops[*n_ops];
3703 struct dpif_flow_stats stats;
3705 COVERAGE_INC(facet_suppress);
3707 handle_flow_miss_common(rule, packet, &miss->flow);
3709 dpif_flow_stats_extract(&miss->flow, packet, now, &stats);
3710 rule_credit_stats(rule, &stats);
3712 xlate_in_init(&xin, miss->ofproto, &miss->flow, &miss->initial_vals,
3713 rule, stats.tcp_flags, packet);
3714 xin.resubmit_stats = &stats;
3715 xlate_actions(&xin, &op->xout);
3717 if (op->xout.odp_actions.size) {
3718 struct dpif_execute *execute = &op->dpif_op.u.execute;
3720 init_flow_miss_execute_op(miss, packet, op);
3721 execute->actions = op->xout.odp_actions.data;
3722 execute->actions_len = op->xout.odp_actions.size;
3723 op->xout_garbage = true;
3727 xlate_out_uninit(&op->xout);
3732 /* Handles 'miss', which matches 'facet'. May add any required datapath
3733 * operations to 'ops', incrementing '*n_ops' for each new op.
3735 * All of the packets in 'miss' are considered to have arrived at time 'now'.
3736 * This is really important only for new facets: if we just called time_msec()
3737 * here, then the new subfacet or its packets could look (occasionally) as
3738 * though it was used some time after the facet was used. That can make a
3739 * one-packet flow look like it has a nonzero duration, which looks odd in
3740 * e.g. NetFlow statistics. */
3742 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
3744 struct flow_miss_op *ops, size_t *n_ops)
3746 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3747 enum subfacet_path want_path;
3748 struct subfacet *subfacet;
3749 struct ofpbuf *packet;
3751 subfacet = subfacet_create(facet, miss, now);
3752 want_path = subfacet->facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
3754 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3755 struct flow_miss_op *op = &ops[*n_ops];
3756 struct dpif_flow_stats stats;
3758 handle_flow_miss_common(facet->rule, packet, &miss->flow);
3760 if (want_path != SF_FAST_PATH) {
3761 struct xlate_in xin;
3763 xlate_in_init(&xin, ofproto, &facet->flow, &facet->initial_vals,
3764 facet->rule, 0, packet);
3765 xlate_actions_for_side_effects(&xin);
3768 dpif_flow_stats_extract(&facet->flow, packet, now, &stats);
3769 subfacet_update_stats(subfacet, &stats);
3771 if (facet->xout.odp_actions.size) {
3772 struct dpif_execute *execute = &op->dpif_op.u.execute;
3774 init_flow_miss_execute_op(miss, packet, op);
3775 execute->actions = facet->xout.odp_actions.data,
3776 execute->actions_len = facet->xout.odp_actions.size;
3781 if (miss->upcall_type == DPIF_UC_MISS || subfacet->path != want_path) {
3782 struct flow_miss_op *op = &ops[(*n_ops)++];
3783 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
3785 subfacet->path = want_path;
3787 op->xout_garbage = false;
3788 op->dpif_op.type = DPIF_OP_FLOW_PUT;
3789 put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
3790 put->key = miss->key;
3791 put->key_len = miss->key_len;
3792 if (want_path == SF_FAST_PATH) {
3793 put->actions = facet->xout.odp_actions.data;
3794 put->actions_len = facet->xout.odp_actions.size;
3796 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
3797 op->slow_stub, sizeof op->slow_stub,
3798 &put->actions, &put->actions_len);
3804 /* Handles flow miss 'miss'. May add any required datapath operations
3805 * to 'ops', incrementing '*n_ops' for each new op. */
3807 handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops,
3810 struct ofproto_dpif *ofproto = miss->ofproto;
3811 struct facet *facet;
3815 /* The caller must ensure that miss->hmap_node.hash contains
3816 * flow_hash(miss->flow, 0). */
3817 hash = miss->hmap_node.hash;
3819 facet = facet_lookup_valid(ofproto, &miss->flow, hash);
3821 /* There does not exist a bijection between 'struct flow' and datapath
3822 * flow keys with fitness ODP_FIT_TO_LITTLE. This breaks a fundamental
3823 * assumption used throughout the facet and subfacet handling code.
3824 * Since we have to handle these misses in userspace anyway, we simply
3825 * skip facet creation, avoiding the problem altogether. */
3826 if (miss->key_fitness == ODP_FIT_TOO_LITTLE
3827 || !flow_miss_should_make_facet(ofproto, miss, hash)) {
3828 handle_flow_miss_without_facet(miss, ops, n_ops);
3832 facet = facet_create(miss, hash);
3837 handle_flow_miss_with_facet(miss, facet, now, ops, n_ops);
3840 static struct drop_key *
3841 drop_key_lookup(const struct dpif_backer *backer, const struct nlattr *key,
3844 struct drop_key *drop_key;
3846 HMAP_FOR_EACH_WITH_HASH (drop_key, hmap_node, hash_bytes(key, key_len, 0),
3847 &backer->drop_keys) {
3848 if (drop_key->key_len == key_len
3849 && !memcmp(drop_key->key, key, key_len)) {
3857 drop_key_clear(struct dpif_backer *backer)
3859 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3860 struct drop_key *drop_key, *next;
3862 HMAP_FOR_EACH_SAFE (drop_key, next, hmap_node, &backer->drop_keys) {
3865 error = dpif_flow_del(backer->dpif, drop_key->key, drop_key->key_len,
3867 if (error && !VLOG_DROP_WARN(&rl)) {
3868 struct ds ds = DS_EMPTY_INITIALIZER;
3869 odp_flow_key_format(drop_key->key, drop_key->key_len, &ds);
3870 VLOG_WARN("Failed to delete drop key (%s) (%s)", strerror(error),
3875 hmap_remove(&backer->drop_keys, &drop_key->hmap_node);
3876 free(drop_key->key);
3881 /* Given a datpath, packet, and flow metadata ('backer', 'packet', and 'key'
3882 * respectively), populates 'flow' with the result of odp_flow_key_to_flow().
3883 * Optionally, if nonnull, populates 'fitnessp' with the fitness of 'flow' as
3884 * returned by odp_flow_key_to_flow(). Also, optionally populates 'ofproto'
3885 * with the ofproto_dpif, and 'odp_in_port' with the datapath in_port, that
3886 * 'packet' ingressed.
3888 * If 'ofproto' is nonnull, requires 'flow''s in_port to exist. Otherwise sets
3889 * 'flow''s in_port to OFPP_NONE.
3891 * This function does post-processing on data returned from
3892 * odp_flow_key_to_flow() to help make VLAN splinters transparent to the rest
3893 * of the upcall processing logic. In particular, if the extracted in_port is
3894 * a VLAN splinter port, it replaces flow->in_port by the "real" port, sets
3895 * flow->vlan_tci correctly for the VLAN of the VLAN splinter port, and pushes
3896 * a VLAN header onto 'packet' (if it is nonnull).
3898 * Optionally, if 'initial_vals' is nonnull, sets 'initial_vals->vlan_tci'
3899 * to the VLAN TCI with which the packet was really received, that is, the
3900 * actual VLAN TCI extracted by odp_flow_key_to_flow(). (This differs from
3901 * the value returned in flow->vlan_tci only for packets received on
3904 * Similarly, this function also includes some logic to help with tunnels. It
3905 * may modify 'flow' as necessary to make the tunneling implementation
3906 * transparent to the upcall processing logic.
3908 * Returns 0 if successful, ENODEV if the parsed flow has no associated ofport,
3909 * or some other positive errno if there are other problems. */
3911 ofproto_receive(const struct dpif_backer *backer, struct ofpbuf *packet,
3912 const struct nlattr *key, size_t key_len,
3913 struct flow *flow, enum odp_key_fitness *fitnessp,
3914 struct ofproto_dpif **ofproto, uint32_t *odp_in_port,
3915 struct initial_vals *initial_vals)
3917 const struct ofport_dpif *port;
3918 enum odp_key_fitness fitness;
3921 fitness = odp_flow_key_to_flow(key, key_len, flow);
3922 if (fitness == ODP_FIT_ERROR) {
3928 initial_vals->vlan_tci = flow->vlan_tci;
3932 *odp_in_port = flow->in_port;
3935 port = (tnl_port_should_receive(flow)
3936 ? ofport_dpif_cast(tnl_port_receive(flow))
3937 : odp_port_to_ofport(backer, flow->in_port));
3938 flow->in_port = port ? port->up.ofp_port : OFPP_NONE;
3943 /* XXX: Since the tunnel module is not scoped per backer, for a tunnel port
3944 * it's theoretically possible that we'll receive an ofport belonging to an
3945 * entirely different datapath. In practice, this can't happen because no
3946 * platforms has two separate datapaths which each support tunneling. */
3947 ovs_assert(ofproto_dpif_cast(port->up.ofproto)->backer == backer);
3949 if (vsp_adjust_flow(ofproto_dpif_cast(port->up.ofproto), flow)) {
3951 /* Make the packet resemble the flow, so that it gets sent to
3952 * an OpenFlow controller properly, so that it looks correct
3953 * for sFlow, and so that flow_extract() will get the correct
3954 * vlan_tci if it is called on 'packet'.
3956 * The allocated space inside 'packet' probably also contains
3957 * 'key', that is, both 'packet' and 'key' are probably part of
3958 * a struct dpif_upcall (see the large comment on that
3959 * structure definition), so pushing data on 'packet' is in
3960 * general not a good idea since it could overwrite 'key' or
3961 * free it as a side effect. However, it's OK in this special
3962 * case because we know that 'packet' is inside a Netlink
3963 * attribute: pushing 4 bytes will just overwrite the 4-byte
3964 * "struct nlattr", which is fine since we don't need that
3965 * header anymore. */
3966 eth_push_vlan(packet, flow->vlan_tci);
3968 /* We can't reproduce 'key' from 'flow'. */
3969 fitness = fitness == ODP_FIT_PERFECT ? ODP_FIT_TOO_MUCH : fitness;
3974 *ofproto = ofproto_dpif_cast(port->up.ofproto);
3979 *fitnessp = fitness;
3985 handle_miss_upcalls(struct dpif_backer *backer, struct dpif_upcall *upcalls,
3988 struct dpif_upcall *upcall;
3989 struct flow_miss *miss;
3990 struct flow_miss misses[FLOW_MISS_MAX_BATCH];
3991 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
3992 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
4002 /* Construct the to-do list.
4004 * This just amounts to extracting the flow from each packet and sticking
4005 * the packets that have the same flow in the same "flow_miss" structure so
4006 * that we can process them together. */
4009 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
4010 struct flow_miss *miss = &misses[n_misses];
4011 struct flow_miss *existing_miss;
4012 struct ofproto_dpif *ofproto;
4013 uint32_t odp_in_port;
4018 error = ofproto_receive(backer, upcall->packet, upcall->key,
4019 upcall->key_len, &flow, &miss->key_fitness,
4020 &ofproto, &odp_in_port, &miss->initial_vals);
4021 if (error == ENODEV) {
4022 struct drop_key *drop_key;
4024 /* Received packet on datapath port for which we couldn't
4025 * associate an ofproto. This can happen if a port is removed
4026 * while traffic is being received. Print a rate-limited message
4027 * in case it happens frequently. Install a drop flow so
4028 * that future packets of the flow are inexpensively dropped
4030 VLOG_INFO_RL(&rl, "received packet on unassociated datapath port "
4031 "%"PRIu32, odp_in_port);
4033 drop_key = drop_key_lookup(backer, upcall->key, upcall->key_len);
4035 drop_key = xmalloc(sizeof *drop_key);
4036 drop_key->key = xmemdup(upcall->key, upcall->key_len);
4037 drop_key->key_len = upcall->key_len;
4039 hmap_insert(&backer->drop_keys, &drop_key->hmap_node,
4040 hash_bytes(drop_key->key, drop_key->key_len, 0));
4041 dpif_flow_put(backer->dpif, DPIF_FP_CREATE | DPIF_FP_MODIFY,
4042 drop_key->key, drop_key->key_len, NULL, 0, NULL);
4050 ofproto->n_missed++;
4051 flow_extract(upcall->packet, flow.skb_priority, flow.skb_mark,
4052 &flow.tunnel, flow.in_port, &miss->flow);
4054 /* Add other packets to a to-do list. */
4055 hash = flow_hash(&miss->flow, 0);
4056 existing_miss = flow_miss_find(&todo, ofproto, &miss->flow, hash);
4057 if (!existing_miss) {
4058 hmap_insert(&todo, &miss->hmap_node, hash);
4059 miss->ofproto = ofproto;
4060 miss->key = upcall->key;
4061 miss->key_len = upcall->key_len;
4062 miss->upcall_type = upcall->type;
4063 list_init(&miss->packets);
4067 miss = existing_miss;
4069 list_push_back(&miss->packets, &upcall->packet->list_node);
4072 /* Process each element in the to-do list, constructing the set of
4073 * operations to batch. */
4075 HMAP_FOR_EACH (miss, hmap_node, &todo) {
4076 handle_flow_miss(miss, flow_miss_ops, &n_ops);
4078 ovs_assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
4080 /* Execute batch. */
4081 for (i = 0; i < n_ops; i++) {
4082 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
4084 dpif_operate(backer->dpif, dpif_ops, n_ops);
4087 for (i = 0; i < n_ops; i++) {
4088 if (flow_miss_ops[i].xout_garbage) {
4089 xlate_out_uninit(&flow_miss_ops[i].xout);
4092 hmap_destroy(&todo);
4095 static enum { SFLOW_UPCALL, MISS_UPCALL, BAD_UPCALL, FLOW_SAMPLE_UPCALL,
4097 classify_upcall(const struct dpif_upcall *upcall)
4099 size_t userdata_len;
4100 union user_action_cookie cookie;
4102 /* First look at the upcall type. */
4103 switch (upcall->type) {
4104 case DPIF_UC_ACTION:
4110 case DPIF_N_UC_TYPES:
4112 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
4116 /* "action" upcalls need a closer look. */
4117 if (!upcall->userdata) {
4118 VLOG_WARN_RL(&rl, "action upcall missing cookie");
4121 userdata_len = nl_attr_get_size(upcall->userdata);
4122 if (userdata_len < sizeof cookie.type
4123 || userdata_len > sizeof cookie) {
4124 VLOG_WARN_RL(&rl, "action upcall cookie has unexpected size %zu",
4128 memset(&cookie, 0, sizeof cookie);
4129 memcpy(&cookie, nl_attr_get(upcall->userdata), userdata_len);
4130 if (userdata_len == sizeof cookie.sflow
4131 && cookie.type == USER_ACTION_COOKIE_SFLOW) {
4132 return SFLOW_UPCALL;
4133 } else if (userdata_len == sizeof cookie.slow_path
4134 && cookie.type == USER_ACTION_COOKIE_SLOW_PATH) {
4136 } else if (userdata_len == sizeof cookie.flow_sample
4137 && cookie.type == USER_ACTION_COOKIE_FLOW_SAMPLE) {
4138 return FLOW_SAMPLE_UPCALL;
4139 } else if (userdata_len == sizeof cookie.ipfix
4140 && cookie.type == USER_ACTION_COOKIE_IPFIX) {
4141 return IPFIX_UPCALL;
4143 VLOG_WARN_RL(&rl, "invalid user cookie of type %"PRIu16
4144 " and size %zu", cookie.type, userdata_len);
4150 handle_sflow_upcall(struct dpif_backer *backer,
4151 const struct dpif_upcall *upcall)
4153 struct ofproto_dpif *ofproto;
4154 union user_action_cookie cookie;
4156 uint32_t odp_in_port;
4158 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
4159 &flow, NULL, &ofproto, &odp_in_port, NULL)
4160 || !ofproto->sflow) {
4164 memset(&cookie, 0, sizeof cookie);
4165 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.sflow);
4166 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
4167 odp_in_port, &cookie);
4171 handle_flow_sample_upcall(struct dpif_backer *backer,
4172 const struct dpif_upcall *upcall)
4174 struct ofproto_dpif *ofproto;
4175 union user_action_cookie cookie;
4178 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
4179 &flow, NULL, &ofproto, NULL, NULL)
4180 || !ofproto->ipfix) {
4184 memset(&cookie, 0, sizeof cookie);
4185 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.flow_sample);
4187 /* The flow reflects exactly the contents of the packet. Sample
4188 * the packet using it. */
4189 dpif_ipfix_flow_sample(ofproto->ipfix, upcall->packet, &flow,
4190 cookie.flow_sample.collector_set_id,
4191 cookie.flow_sample.probability,
4192 cookie.flow_sample.obs_domain_id,
4193 cookie.flow_sample.obs_point_id);
4197 handle_ipfix_upcall(struct dpif_backer *backer,
4198 const struct dpif_upcall *upcall)
4200 struct ofproto_dpif *ofproto;
4203 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
4204 &flow, NULL, &ofproto, NULL, NULL)
4205 || !ofproto->ipfix) {
4209 /* The flow reflects exactly the contents of the packet. Sample
4210 * the packet using it. */
4211 dpif_ipfix_bridge_sample(ofproto->ipfix, upcall->packet, &flow);
4215 handle_upcalls(struct dpif_backer *backer, unsigned int max_batch)
4217 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
4218 struct ofpbuf miss_bufs[FLOW_MISS_MAX_BATCH];
4219 uint64_t miss_buf_stubs[FLOW_MISS_MAX_BATCH][4096 / 8];
4224 ovs_assert(max_batch <= FLOW_MISS_MAX_BATCH);
4227 for (n_processed = 0; n_processed < max_batch; n_processed++) {
4228 struct dpif_upcall *upcall = &misses[n_misses];
4229 struct ofpbuf *buf = &miss_bufs[n_misses];
4232 ofpbuf_use_stub(buf, miss_buf_stubs[n_misses],
4233 sizeof miss_buf_stubs[n_misses]);
4234 error = dpif_recv(backer->dpif, upcall, buf);
4240 switch (classify_upcall(upcall)) {
4242 /* Handle it later. */
4247 handle_sflow_upcall(backer, upcall);
4251 case FLOW_SAMPLE_UPCALL:
4252 handle_flow_sample_upcall(backer, upcall);
4257 handle_ipfix_upcall(backer, upcall);
4267 /* Handle deferred MISS_UPCALL processing. */
4268 handle_miss_upcalls(backer, misses, n_misses);
4269 for (i = 0; i < n_misses; i++) {
4270 ofpbuf_uninit(&miss_bufs[i]);
4276 /* Flow expiration. */
4278 static int subfacet_max_idle(const struct ofproto_dpif *);
4279 static void update_stats(struct dpif_backer *);
4280 static void rule_expire(struct rule_dpif *);
4281 static void expire_subfacets(struct ofproto_dpif *, int dp_max_idle);
4283 /* This function is called periodically by run(). Its job is to collect
4284 * updates for the flows that have been installed into the datapath, most
4285 * importantly when they last were used, and then use that information to
4286 * expire flows that have not been used recently.
4288 * Returns the number of milliseconds after which it should be called again. */
4290 expire(struct dpif_backer *backer)
4292 long long int total_subfacet_life, now;
4293 struct ofproto_dpif *ofproto;
4294 int max_idle = INT32_MAX;
4297 /* Periodically clear out the drop keys in an effort to keep them
4298 * relatively few. */
4299 drop_key_clear(backer);
4301 /* Update stats for each flow in the backer. */
4302 update_stats(backer);
4305 total_subfacet_life = n_subfacets = 0;
4306 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4307 struct rule *rule, *next_rule;
4308 struct subfacet *subfacet;
4311 if (ofproto->backer != backer) {
4315 n_subfacets += hmap_count(&ofproto->subfacets);
4316 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
4317 total_subfacet_life += now - subfacet->created;
4320 /* Expire subfacets that have been idle too long. */
4321 dp_max_idle = subfacet_max_idle(ofproto);
4322 expire_subfacets(ofproto, dp_max_idle);
4324 max_idle = MIN(max_idle, dp_max_idle);
4326 /* Expire OpenFlow flows whose idle_timeout or hard_timeout
4328 LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
4329 &ofproto->up.expirable) {
4330 rule_expire(rule_dpif_cast(rule));
4333 /* All outstanding data in existing flows has been accounted, so it's a
4334 * good time to do bond rebalancing. */
4335 if (ofproto->has_bonded_bundles) {
4336 struct ofbundle *bundle;
4338 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
4340 bond_rebalance(bundle->bond, &backer->revalidate_set);
4347 backer->avg_subfacet_life += total_subfacet_life / n_subfacets;
4349 backer->avg_subfacet_life /= 2;
4351 backer->avg_n_subfacet += n_subfacets;
4352 backer->avg_n_subfacet /= 2;
4354 backer->max_n_subfacet = MAX(backer->max_n_subfacet, n_subfacets);
4356 return MIN(max_idle, 1000);
4359 /* Updates flow table statistics given that the datapath just reported 'stats'
4360 * as 'subfacet''s statistics. */
4362 update_subfacet_stats(struct subfacet *subfacet,
4363 const struct dpif_flow_stats *stats)
4365 struct facet *facet = subfacet->facet;
4366 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4367 struct dpif_flow_stats diff;
4369 diff.tcp_flags = stats->tcp_flags;
4370 diff.used = stats->used;
4372 if (stats->n_packets >= subfacet->dp_packet_count) {
4373 diff.n_packets = stats->n_packets - subfacet->dp_packet_count;
4375 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
4379 if (stats->n_bytes >= subfacet->dp_byte_count) {
4380 diff.n_bytes = stats->n_bytes - subfacet->dp_byte_count;
4382 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
4386 ofproto->n_hit += diff.n_packets;
4387 subfacet->dp_packet_count = stats->n_packets;
4388 subfacet->dp_byte_count = stats->n_bytes;
4389 subfacet_update_stats(subfacet, &diff);
4391 if (facet->accounted_bytes < facet->byte_count) {
4393 facet_account(facet);
4394 facet->accounted_bytes = facet->byte_count;
4398 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
4399 * about, or a flow that shouldn't be installed but was anyway. Delete it. */
4401 delete_unexpected_flow(struct ofproto_dpif *ofproto,
4402 const struct nlattr *key, size_t key_len)
4404 if (!VLOG_DROP_WARN(&rl)) {
4408 odp_flow_key_format(key, key_len, &s);
4409 VLOG_WARN("unexpected flow on %s: %s", ofproto->up.name, ds_cstr(&s));
4413 COVERAGE_INC(facet_unexpected);
4414 dpif_flow_del(ofproto->backer->dpif, key, key_len, NULL);
4417 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
4419 * This function also pushes statistics updates to rules which each facet
4420 * resubmits into. Generally these statistics will be accurate. However, if a
4421 * facet changes the rule it resubmits into at some time in between
4422 * update_stats() runs, it is possible that statistics accrued to the
4423 * old rule will be incorrectly attributed to the new rule. This could be
4424 * avoided by calling update_stats() whenever rules are created or
4425 * deleted. However, the performance impact of making so many calls to the
4426 * datapath do not justify the benefit of having perfectly accurate statistics.
4428 * In addition, this function maintains per ofproto flow hit counts. The patch
4429 * port is not treated specially. e.g. A packet ingress from br0 patched into
4430 * br1 will increase the hit count of br0 by 1, however, does not affect
4431 * the hit or miss counts of br1.
4434 update_stats(struct dpif_backer *backer)
4436 const struct dpif_flow_stats *stats;
4437 struct dpif_flow_dump dump;
4438 const struct nlattr *key;
4441 dpif_flow_dump_start(&dump, backer->dpif);
4442 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
4443 struct ofproto_dpif *ofproto;
4445 struct subfacet *subfacet;
4448 if (ofproto_receive(backer, NULL, key, key_len, &flow, NULL, &ofproto,
4453 key_hash = odp_flow_key_hash(key, key_len);
4454 subfacet = subfacet_find(ofproto, key, key_len, key_hash);
4455 switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
4457 update_subfacet_stats(subfacet, stats);
4461 /* Stats are updated per-packet. */
4464 case SF_NOT_INSTALLED:
4466 delete_unexpected_flow(ofproto, key, key_len);
4471 dpif_flow_dump_done(&dump);
4473 update_moving_averages(backer);
4476 /* Calculates and returns the number of milliseconds of idle time after which
4477 * subfacets should expire from the datapath. When a subfacet expires, we fold
4478 * its statistics into its facet, and when a facet's last subfacet expires, we
4479 * fold its statistic into its rule. */
4481 subfacet_max_idle(const struct ofproto_dpif *ofproto)
4484 * Idle time histogram.
4486 * Most of the time a switch has a relatively small number of subfacets.
4487 * When this is the case we might as well keep statistics for all of them
4488 * in userspace and to cache them in the kernel datapath for performance as
4491 * As the number of subfacets increases, the memory required to maintain
4492 * statistics about them in userspace and in the kernel becomes
4493 * significant. However, with a large number of subfacets it is likely
4494 * that only a few of them are "heavy hitters" that consume a large amount
4495 * of bandwidth. At this point, only heavy hitters are worth caching in
4496 * the kernel and maintaining in userspaces; other subfacets we can
4499 * The technique used to compute the idle time is to build a histogram with
4500 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
4501 * that is installed in the kernel gets dropped in the appropriate bucket.
4502 * After the histogram has been built, we compute the cutoff so that only
4503 * the most-recently-used 1% of subfacets (but at least
4504 * ofproto->up.flow_eviction_threshold flows) are kept cached. At least
4505 * the most-recently-used bucket of subfacets is kept, so actually an
4506 * arbitrary number of subfacets can be kept in any given expiration run
4507 * (though the next run will delete most of those unless they receive
4510 * This requires a second pass through the subfacets, in addition to the
4511 * pass made by update_stats(), because the former function never looks at
4512 * uninstallable subfacets.
4514 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
4515 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
4516 int buckets[N_BUCKETS] = { 0 };
4517 int total, subtotal, bucket;
4518 struct subfacet *subfacet;
4522 total = hmap_count(&ofproto->subfacets);
4523 if (total <= ofproto->up.flow_eviction_threshold) {
4524 return N_BUCKETS * BUCKET_WIDTH;
4527 /* Build histogram. */
4529 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
4530 long long int idle = now - subfacet->used;
4531 int bucket = (idle <= 0 ? 0
4532 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
4533 : (unsigned int) idle / BUCKET_WIDTH);
4537 /* Find the first bucket whose flows should be expired. */
4538 subtotal = bucket = 0;
4540 subtotal += buckets[bucket++];
4541 } while (bucket < N_BUCKETS &&
4542 subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100));
4544 if (VLOG_IS_DBG_ENABLED()) {
4548 ds_put_cstr(&s, "keep");
4549 for (i = 0; i < N_BUCKETS; i++) {
4551 ds_put_cstr(&s, ", drop");
4554 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
4557 VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s));
4561 return bucket * BUCKET_WIDTH;
4565 expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle)
4567 /* Cutoff time for most flows. */
4568 long long int normal_cutoff = time_msec() - dp_max_idle;
4570 /* We really want to keep flows for special protocols around, so use a more
4571 * conservative cutoff. */
4572 long long int special_cutoff = time_msec() - 10000;
4574 struct subfacet *subfacet, *next_subfacet;
4575 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
4579 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
4580 &ofproto->subfacets) {
4581 long long int cutoff;
4583 cutoff = (subfacet->facet->xout.slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP
4587 if (subfacet->used < cutoff) {
4588 if (subfacet->path != SF_NOT_INSTALLED) {
4589 batch[n_batch++] = subfacet;
4590 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
4591 subfacet_destroy_batch(ofproto, batch, n_batch);
4595 subfacet_destroy(subfacet);
4601 subfacet_destroy_batch(ofproto, batch, n_batch);
4605 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
4606 * then delete it entirely. */
4608 rule_expire(struct rule_dpif *rule)
4610 struct facet *facet, *next_facet;
4614 if (rule->up.pending) {
4615 /* We'll have to expire it later. */
4619 /* Has 'rule' expired? */
4621 if (rule->up.hard_timeout
4622 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
4623 reason = OFPRR_HARD_TIMEOUT;
4624 } else if (rule->up.idle_timeout
4625 && now > rule->up.used + rule->up.idle_timeout * 1000) {
4626 reason = OFPRR_IDLE_TIMEOUT;
4631 COVERAGE_INC(ofproto_dpif_expired);
4633 /* Update stats. (This is a no-op if the rule expired due to an idle
4634 * timeout, because that only happens when the rule has no facets left.) */
4635 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4636 facet_remove(facet);
4639 /* Get rid of the rule. */
4640 ofproto_rule_expire(&rule->up, reason);
4645 /* Creates and returns a new facet based on 'miss'.
4647 * The caller must already have determined that no facet with an identical
4648 * 'miss->flow' exists in 'miss->ofproto'.
4650 * 'hash' must be the return value of flow_hash(miss->flow, 0).
4652 * The facet will initially have no subfacets. The caller should create (at
4653 * least) one subfacet with subfacet_create(). */
4654 static struct facet *
4655 facet_create(const struct flow_miss *miss, uint32_t hash)
4657 struct ofproto_dpif *ofproto = miss->ofproto;
4658 struct xlate_in xin;
4659 struct facet *facet;
4661 facet = xzalloc(sizeof *facet);
4662 facet->used = time_msec();
4663 facet->flow = miss->flow;
4664 facet->initial_vals = miss->initial_vals;
4665 facet->rule = rule_dpif_lookup(ofproto, &facet->flow);
4666 facet->learn_rl = time_msec() + 500;
4668 hmap_insert(&ofproto->facets, &facet->hmap_node, hash);
4669 list_push_back(&facet->rule->facets, &facet->list_node);
4670 list_init(&facet->subfacets);
4671 netflow_flow_init(&facet->nf_flow);
4672 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
4674 xlate_in_init(&xin, ofproto, &facet->flow, &facet->initial_vals,
4675 facet->rule, 0, NULL);
4676 xin.may_learn = true;
4677 xlate_actions(&xin, &facet->xout);
4678 facet->nf_flow.output_iface = facet->xout.nf_output_iface;
4684 facet_free(struct facet *facet)
4687 xlate_out_uninit(&facet->xout);
4692 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
4693 * 'packet', which arrived on 'in_port'. */
4695 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
4696 const struct nlattr *odp_actions, size_t actions_len,
4697 struct ofpbuf *packet)
4699 struct odputil_keybuf keybuf;
4703 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4704 odp_flow_key_from_flow(&key, flow,
4705 ofp_port_to_odp_port(ofproto, flow->in_port));
4707 error = dpif_execute(ofproto->backer->dpif, key.data, key.size,
4708 odp_actions, actions_len, packet);
4712 /* Remove 'facet' from 'ofproto' and free up the associated memory:
4714 * - If 'facet' was installed in the datapath, uninstalls it and updates its
4715 * rule's statistics, via subfacet_uninstall().
4717 * - Removes 'facet' from its rule and from ofproto->facets.
4720 facet_remove(struct facet *facet)
4722 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4723 struct subfacet *subfacet, *next_subfacet;
4725 ovs_assert(!list_is_empty(&facet->subfacets));
4727 /* First uninstall all of the subfacets to get final statistics. */
4728 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4729 subfacet_uninstall(subfacet);
4732 /* Flush the final stats to the rule.
4734 * This might require us to have at least one subfacet around so that we
4735 * can use its actions for accounting in facet_account(), which is why we
4736 * have uninstalled but not yet destroyed the subfacets. */
4737 facet_flush_stats(facet);
4739 /* Now we're really all done so destroy everything. */
4740 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
4741 &facet->subfacets) {
4742 subfacet_destroy__(subfacet);
4744 hmap_remove(&ofproto->facets, &facet->hmap_node);
4745 list_remove(&facet->list_node);
4749 /* Feed information from 'facet' back into the learning table to keep it in
4750 * sync with what is actually flowing through the datapath. */
4752 facet_learn(struct facet *facet)
4754 long long int now = time_msec();
4756 if (!facet->xout.has_fin_timeout && now < facet->learn_rl) {
4760 facet->learn_rl = now + 500;
4762 if (!facet->xout.has_learn
4763 && !facet->xout.has_normal
4764 && (!facet->xout.has_fin_timeout
4765 || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
4769 facet_push_stats(facet, true);
4773 facet_account(struct facet *facet)
4775 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4776 const struct nlattr *a;
4781 if (!facet->xout.has_normal || !ofproto->has_bonded_bundles) {
4784 n_bytes = facet->byte_count - facet->accounted_bytes;
4786 /* This loop feeds byte counters to bond_account() for rebalancing to use
4787 * as a basis. We also need to track the actual VLAN on which the packet
4788 * is going to be sent to ensure that it matches the one passed to
4789 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
4792 * We use the actions from an arbitrary subfacet because they should all
4793 * be equally valid for our purpose. */
4794 vlan_tci = facet->flow.vlan_tci;
4795 NL_ATTR_FOR_EACH_UNSAFE (a, left, facet->xout.odp_actions.data,
4796 facet->xout.odp_actions.size) {
4797 const struct ovs_action_push_vlan *vlan;
4798 struct ofport_dpif *port;
4800 switch (nl_attr_type(a)) {
4801 case OVS_ACTION_ATTR_OUTPUT:
4802 port = get_odp_port(ofproto, nl_attr_get_u32(a));
4803 if (port && port->bundle && port->bundle->bond) {
4804 bond_account(port->bundle->bond, &facet->flow,
4805 vlan_tci_to_vid(vlan_tci), n_bytes);
4809 case OVS_ACTION_ATTR_POP_VLAN:
4810 vlan_tci = htons(0);
4813 case OVS_ACTION_ATTR_PUSH_VLAN:
4814 vlan = nl_attr_get(a);
4815 vlan_tci = vlan->vlan_tci;
4821 /* Returns true if the only action for 'facet' is to send to the controller.
4822 * (We don't report NetFlow expiration messages for such facets because they
4823 * are just part of the control logic for the network, not real traffic). */
4825 facet_is_controller_flow(struct facet *facet)
4828 const struct rule *rule = &facet->rule->up;
4829 const struct ofpact *ofpacts = rule->ofpacts;
4830 size_t ofpacts_len = rule->ofpacts_len;
4832 if (ofpacts_len > 0 &&
4833 ofpacts->type == OFPACT_CONTROLLER &&
4834 ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len)) {
4841 /* Folds all of 'facet''s statistics into its rule. Also updates the
4842 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
4843 * 'facet''s statistics in the datapath should have been zeroed and folded into
4844 * its packet and byte counts before this function is called. */
4846 facet_flush_stats(struct facet *facet)
4848 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4849 struct subfacet *subfacet;
4851 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4852 ovs_assert(!subfacet->dp_byte_count);
4853 ovs_assert(!subfacet->dp_packet_count);
4856 facet_push_stats(facet, false);
4857 if (facet->accounted_bytes < facet->byte_count) {
4858 facet_account(facet);
4859 facet->accounted_bytes = facet->byte_count;
4862 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
4863 struct ofexpired expired;
4864 expired.flow = facet->flow;
4865 expired.packet_count = facet->packet_count;
4866 expired.byte_count = facet->byte_count;
4867 expired.used = facet->used;
4868 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4871 /* Reset counters to prevent double counting if 'facet' ever gets
4873 facet_reset_counters(facet);
4875 netflow_flow_clear(&facet->nf_flow);
4876 facet->tcp_flags = 0;
4879 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
4880 * Returns it if found, otherwise a null pointer.
4882 * 'hash' must be the return value of flow_hash(flow, 0).
4884 * The returned facet might need revalidation; use facet_lookup_valid()
4885 * instead if that is important. */
4886 static struct facet *
4887 facet_find(struct ofproto_dpif *ofproto,
4888 const struct flow *flow, uint32_t hash)
4890 struct facet *facet;
4892 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, hash, &ofproto->facets) {
4893 if (flow_equal(flow, &facet->flow)) {
4901 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
4902 * Returns it if found, otherwise a null pointer.
4904 * 'hash' must be the return value of flow_hash(flow, 0).
4906 * The returned facet is guaranteed to be valid. */
4907 static struct facet *
4908 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow,
4911 struct facet *facet;
4913 facet = facet_find(ofproto, flow, hash);
4915 && (ofproto->backer->need_revalidate
4916 || tag_set_intersects(&ofproto->backer->revalidate_set,
4918 && !facet_revalidate(facet)) {
4926 facet_check_consistency(struct facet *facet)
4928 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
4930 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4932 struct xlate_out xout;
4933 struct xlate_in xin;
4935 struct rule_dpif *rule;
4938 /* Check the rule for consistency. */
4939 rule = rule_dpif_lookup(ofproto, &facet->flow);
4940 if (rule != facet->rule) {
4941 if (!VLOG_DROP_WARN(&rl)) {
4942 struct ds s = DS_EMPTY_INITIALIZER;
4944 flow_format(&s, &facet->flow);
4945 ds_put_format(&s, ": facet associated with wrong rule (was "
4946 "table=%"PRIu8",", facet->rule->up.table_id);
4947 cls_rule_format(&facet->rule->up.cr, &s);
4948 ds_put_format(&s, ") (should have been table=%"PRIu8",",
4950 cls_rule_format(&rule->up.cr, &s);
4951 ds_put_char(&s, ')');
4953 VLOG_WARN("%s", ds_cstr(&s));
4959 /* Check the datapath actions for consistency. */
4960 xlate_in_init(&xin, ofproto, &facet->flow, &facet->initial_vals, rule,
4962 xlate_actions(&xin, &xout);
4964 ok = ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)
4965 && facet->xout.slow == xout.slow;
4966 if (!ok && !VLOG_DROP_WARN(&rl)) {
4967 struct ds s = DS_EMPTY_INITIALIZER;
4969 flow_format(&s, &facet->flow);
4970 ds_put_cstr(&s, ": inconsistency in facet");
4972 if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4973 ds_put_cstr(&s, " (actions were: ");
4974 format_odp_actions(&s, facet->xout.odp_actions.data,
4975 facet->xout.odp_actions.size);
4976 ds_put_cstr(&s, ") (correct actions: ");
4977 format_odp_actions(&s, xout.odp_actions.data,
4978 xout.odp_actions.size);
4979 ds_put_char(&s, ')');
4982 if (facet->xout.slow != xout.slow) {
4983 ds_put_format(&s, " slow path incorrect. should be %d", xout.slow);
4986 VLOG_WARN("%s", ds_cstr(&s));
4989 xlate_out_uninit(&xout);
4994 /* Re-searches the classifier for 'facet':
4996 * - If the rule found is different from 'facet''s current rule, moves
4997 * 'facet' to the new rule and recompiles its actions.
4999 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
5000 * where it is and recompiles its actions anyway.
5002 * - If any of 'facet''s subfacets correspond to a new flow according to
5003 * ofproto_receive(), 'facet' is removed.
5005 * Returns true if 'facet' is still valid. False if 'facet' was removed. */
5007 facet_revalidate(struct facet *facet)
5009 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5010 struct rule_dpif *new_rule;
5011 struct subfacet *subfacet;
5012 struct xlate_out xout;
5013 struct xlate_in xin;
5015 COVERAGE_INC(facet_revalidate);
5017 /* Check that child subfacets still correspond to this facet. Tunnel
5018 * configuration changes could cause a subfacet's OpenFlow in_port to
5020 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
5021 struct ofproto_dpif *recv_ofproto;
5022 struct flow recv_flow;
5025 error = ofproto_receive(ofproto->backer, NULL, subfacet->key,
5026 subfacet->key_len, &recv_flow, NULL,
5027 &recv_ofproto, NULL, NULL);
5029 || recv_ofproto != ofproto
5030 || memcmp(&recv_flow, &facet->flow, sizeof recv_flow)) {
5031 facet_remove(facet);
5036 new_rule = rule_dpif_lookup(ofproto, &facet->flow);
5038 /* Calculate new datapath actions.
5040 * We do not modify any 'facet' state yet, because we might need to, e.g.,
5041 * emit a NetFlow expiration and, if so, we need to have the old state
5042 * around to properly compose it. */
5043 xlate_in_init(&xin, ofproto, &facet->flow, &facet->initial_vals, new_rule,
5045 xlate_actions(&xin, &xout);
5047 /* A facet's slow path reason should only change under dramatic
5048 * circumstances. Rather than try to update everything, it's simpler to
5049 * remove the facet and start over. */
5050 if (facet->xout.slow != xout.slow) {
5051 facet_remove(facet);
5052 xlate_out_uninit(&xout);
5056 if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
5057 LIST_FOR_EACH(subfacet, list_node, &facet->subfacets) {
5058 if (subfacet->path == SF_FAST_PATH) {
5059 struct dpif_flow_stats stats;
5061 subfacet_install(subfacet, &xout.odp_actions, &stats);
5062 subfacet_update_stats(subfacet, &stats);
5066 facet_flush_stats(facet);
5068 ofpbuf_clear(&facet->xout.odp_actions);
5069 ofpbuf_put(&facet->xout.odp_actions, xout.odp_actions.data,
5070 xout.odp_actions.size);
5073 /* Update 'facet' now that we've taken care of all the old state. */
5074 facet->xout.tags = xout.tags;
5075 facet->xout.slow = xout.slow;
5076 facet->xout.has_learn = xout.has_learn;
5077 facet->xout.has_normal = xout.has_normal;
5078 facet->xout.has_fin_timeout = xout.has_fin_timeout;
5079 facet->xout.nf_output_iface = xout.nf_output_iface;
5080 facet->xout.mirrors = xout.mirrors;
5081 facet->nf_flow.output_iface = facet->xout.nf_output_iface;
5083 if (facet->rule != new_rule) {
5084 COVERAGE_INC(facet_changed_rule);
5085 list_remove(&facet->list_node);
5086 list_push_back(&new_rule->facets, &facet->list_node);
5087 facet->rule = new_rule;
5088 facet->used = new_rule->up.created;
5089 facet->prev_used = facet->used;
5092 xlate_out_uninit(&xout);
5097 facet_reset_counters(struct facet *facet)
5099 facet->packet_count = 0;
5100 facet->byte_count = 0;
5101 facet->prev_packet_count = 0;
5102 facet->prev_byte_count = 0;
5103 facet->accounted_bytes = 0;
5107 facet_push_stats(struct facet *facet, bool may_learn)
5109 struct dpif_flow_stats stats;
5111 ovs_assert(facet->packet_count >= facet->prev_packet_count);
5112 ovs_assert(facet->byte_count >= facet->prev_byte_count);
5113 ovs_assert(facet->used >= facet->prev_used);
5115 stats.n_packets = facet->packet_count - facet->prev_packet_count;
5116 stats.n_bytes = facet->byte_count - facet->prev_byte_count;
5117 stats.used = facet->used;
5118 stats.tcp_flags = facet->tcp_flags;
5120 if (may_learn || stats.n_packets || facet->used > facet->prev_used) {
5121 struct ofproto_dpif *ofproto =
5122 ofproto_dpif_cast(facet->rule->up.ofproto);
5124 struct ofport_dpif *in_port;
5125 struct xlate_in xin;
5127 facet->prev_packet_count = facet->packet_count;
5128 facet->prev_byte_count = facet->byte_count;
5129 facet->prev_used = facet->used;
5131 in_port = get_ofp_port(ofproto, facet->flow.in_port);
5132 if (in_port && in_port->tnl_port) {
5133 netdev_vport_inc_rx(in_port->up.netdev, &stats);
5136 rule_credit_stats(facet->rule, &stats);
5137 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow,
5139 netflow_flow_update_flags(&facet->nf_flow, facet->tcp_flags);
5140 update_mirror_stats(ofproto, facet->xout.mirrors, stats.n_packets,
5143 xlate_in_init(&xin, ofproto, &facet->flow, &facet->initial_vals,
5144 facet->rule, stats.tcp_flags, NULL);
5145 xin.resubmit_stats = &stats;
5146 xin.may_learn = may_learn;
5147 xlate_actions_for_side_effects(&xin);
5152 push_all_stats__(bool run_fast)
5154 static long long int rl = LLONG_MIN;
5155 struct ofproto_dpif *ofproto;
5157 if (time_msec() < rl) {
5161 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5162 struct facet *facet;
5164 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
5165 facet_push_stats(facet, false);
5172 rl = time_msec() + 100;
5176 push_all_stats(void)
5178 push_all_stats__(true);
5182 rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats)
5184 rule->packet_count += stats->n_packets;
5185 rule->byte_count += stats->n_bytes;
5186 ofproto_rule_update_used(&rule->up, stats->used);
5191 static struct subfacet *
5192 subfacet_find(struct ofproto_dpif *ofproto,
5193 const struct nlattr *key, size_t key_len, uint32_t key_hash)
5195 struct subfacet *subfacet;
5197 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
5198 &ofproto->subfacets) {
5199 if (subfacet->key_len == key_len
5200 && !memcmp(key, subfacet->key, key_len)) {
5208 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
5209 * 'key_fitness', 'key', and 'key_len' members in 'miss'. Returns the
5210 * existing subfacet if there is one, otherwise creates and returns a
5212 static struct subfacet *
5213 subfacet_create(struct facet *facet, struct flow_miss *miss,
5216 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5217 enum odp_key_fitness key_fitness = miss->key_fitness;
5218 const struct nlattr *key = miss->key;
5219 size_t key_len = miss->key_len;
5221 struct subfacet *subfacet;
5223 key_hash = odp_flow_key_hash(key, key_len);
5225 if (list_is_empty(&facet->subfacets)) {
5226 subfacet = &facet->one_subfacet;
5228 subfacet = subfacet_find(ofproto, key, key_len, key_hash);
5230 if (subfacet->facet == facet) {
5234 /* This shouldn't happen. */
5235 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
5236 subfacet_destroy(subfacet);
5239 subfacet = xmalloc(sizeof *subfacet);
5242 hmap_insert(&ofproto->subfacets, &subfacet->hmap_node, key_hash);
5243 list_push_back(&facet->subfacets, &subfacet->list_node);
5244 subfacet->facet = facet;
5245 subfacet->key_fitness = key_fitness;
5246 subfacet->key = xmemdup(key, key_len);
5247 subfacet->key_len = key_len;
5248 subfacet->used = now;
5249 subfacet->created = now;
5250 subfacet->dp_packet_count = 0;
5251 subfacet->dp_byte_count = 0;
5252 subfacet->path = SF_NOT_INSTALLED;
5254 ofproto->backer->subfacet_add_count++;
5258 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
5259 * its facet within 'ofproto', and frees it. */
5261 subfacet_destroy__(struct subfacet *subfacet)
5263 struct facet *facet = subfacet->facet;
5264 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5266 /* Update ofproto stats before uninstall the subfacet. */
5267 ofproto->backer->subfacet_del_count++;
5269 subfacet_uninstall(subfacet);
5270 hmap_remove(&ofproto->subfacets, &subfacet->hmap_node);
5271 list_remove(&subfacet->list_node);
5272 free(subfacet->key);
5273 if (subfacet != &facet->one_subfacet) {
5278 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
5279 * last remaining subfacet in its facet destroys the facet too. */
5281 subfacet_destroy(struct subfacet *subfacet)
5283 struct facet *facet = subfacet->facet;
5285 if (list_is_singleton(&facet->subfacets)) {
5286 /* facet_remove() needs at least one subfacet (it will remove it). */
5287 facet_remove(facet);
5289 subfacet_destroy__(subfacet);
5294 subfacet_destroy_batch(struct ofproto_dpif *ofproto,
5295 struct subfacet **subfacets, int n)
5297 struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH];
5298 struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH];
5299 struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH];
5302 for (i = 0; i < n; i++) {
5303 ops[i].type = DPIF_OP_FLOW_DEL;
5304 ops[i].u.flow_del.key = subfacets[i]->key;
5305 ops[i].u.flow_del.key_len = subfacets[i]->key_len;
5306 ops[i].u.flow_del.stats = &stats[i];
5310 dpif_operate(ofproto->backer->dpif, opsp, n);
5311 for (i = 0; i < n; i++) {
5312 subfacet_reset_dp_stats(subfacets[i], &stats[i]);
5313 subfacets[i]->path = SF_NOT_INSTALLED;
5314 subfacet_destroy(subfacets[i]);
5319 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
5320 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
5321 * in the datapath will be zeroed and 'stats' will be updated with traffic new
5322 * since 'subfacet' was last updated.
5324 * Returns 0 if successful, otherwise a positive errno value. */
5326 subfacet_install(struct subfacet *subfacet, const struct ofpbuf *odp_actions,
5327 struct dpif_flow_stats *stats)
5329 struct facet *facet = subfacet->facet;
5330 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5331 enum subfacet_path path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
5332 const struct nlattr *actions = odp_actions->data;
5333 size_t actions_len = odp_actions->size;
5335 uint64_t slow_path_stub[128 / 8];
5336 enum dpif_flow_put_flags flags;
5339 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
5341 flags |= DPIF_FP_ZERO_STATS;
5344 if (path == SF_SLOW_PATH) {
5345 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
5346 slow_path_stub, sizeof slow_path_stub,
5347 &actions, &actions_len);
5350 ret = dpif_flow_put(ofproto->backer->dpif, flags, subfacet->key,
5351 subfacet->key_len, actions, actions_len, stats);
5354 subfacet_reset_dp_stats(subfacet, stats);
5358 subfacet->path = path;
5363 /* If 'subfacet' is installed in the datapath, uninstalls it. */
5365 subfacet_uninstall(struct subfacet *subfacet)
5367 if (subfacet->path != SF_NOT_INSTALLED) {
5368 struct rule_dpif *rule = subfacet->facet->rule;
5369 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5370 struct dpif_flow_stats stats;
5373 error = dpif_flow_del(ofproto->backer->dpif, subfacet->key,
5374 subfacet->key_len, &stats);
5375 subfacet_reset_dp_stats(subfacet, &stats);
5377 subfacet_update_stats(subfacet, &stats);
5379 subfacet->path = SF_NOT_INSTALLED;
5381 ovs_assert(subfacet->dp_packet_count == 0);
5382 ovs_assert(subfacet->dp_byte_count == 0);
5386 /* Resets 'subfacet''s datapath statistics counters. This should be called
5387 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
5388 * non-null, it should contain the statistics returned by dpif when 'subfacet'
5389 * was reset in the datapath. 'stats' will be modified to include only
5390 * statistics new since 'subfacet' was last updated. */
5392 subfacet_reset_dp_stats(struct subfacet *subfacet,
5393 struct dpif_flow_stats *stats)
5396 && subfacet->dp_packet_count <= stats->n_packets
5397 && subfacet->dp_byte_count <= stats->n_bytes) {
5398 stats->n_packets -= subfacet->dp_packet_count;
5399 stats->n_bytes -= subfacet->dp_byte_count;
5402 subfacet->dp_packet_count = 0;
5403 subfacet->dp_byte_count = 0;
5406 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
5408 * Because of the meaning of a subfacet's counters, it only makes sense to do
5409 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
5410 * represents a packet that was sent by hand or if it represents statistics
5411 * that have been cleared out of the datapath. */
5413 subfacet_update_stats(struct subfacet *subfacet,
5414 const struct dpif_flow_stats *stats)
5416 if (stats->n_packets || stats->used > subfacet->used) {
5417 struct facet *facet = subfacet->facet;
5419 subfacet->used = MAX(subfacet->used, stats->used);
5420 facet->used = MAX(facet->used, stats->used);
5421 facet->packet_count += stats->n_packets;
5422 facet->byte_count += stats->n_bytes;
5423 facet->tcp_flags |= stats->tcp_flags;
5429 static struct rule_dpif *
5430 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow)
5432 struct rule_dpif *rule;
5434 rule = rule_dpif_lookup__(ofproto, flow, 0);
5439 return rule_dpif_miss_rule(ofproto, flow);
5442 static struct rule_dpif *
5443 rule_dpif_lookup__(struct ofproto_dpif *ofproto, const struct flow *flow,
5446 struct cls_rule *cls_rule;
5447 struct classifier *cls;
5450 if (table_id >= N_TABLES) {
5454 cls = &ofproto->up.tables[table_id].cls;
5455 frag = (flow->nw_frag & FLOW_NW_FRAG_ANY) != 0;
5456 if (frag && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
5457 /* We must pretend that transport ports are unavailable. */
5458 struct flow ofpc_normal_flow = *flow;
5459 ofpc_normal_flow.tp_src = htons(0);
5460 ofpc_normal_flow.tp_dst = htons(0);
5461 cls_rule = classifier_lookup(cls, &ofpc_normal_flow);
5462 } else if (frag && ofproto->up.frag_handling == OFPC_FRAG_DROP) {
5463 cls_rule = &ofproto->drop_frags_rule->up.cr;
5465 cls_rule = classifier_lookup(cls, flow);
5467 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
5470 static struct rule_dpif *
5471 rule_dpif_miss_rule(struct ofproto_dpif *ofproto, const struct flow *flow)
5473 struct ofport_dpif *port;
5475 port = get_ofp_port(ofproto, flow->in_port);
5477 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, flow->in_port);
5478 return ofproto->miss_rule;
5481 if (port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN) {
5482 return ofproto->no_packet_in_rule;
5484 return ofproto->miss_rule;
5488 complete_operation(struct rule_dpif *rule)
5490 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5492 rule_invalidate(rule);
5494 struct dpif_completion *c = xmalloc(sizeof *c);
5495 c->op = rule->up.pending;
5496 list_push_back(&ofproto->completions, &c->list_node);
5498 ofoperation_complete(rule->up.pending, 0);
5502 static struct rule *
5505 struct rule_dpif *rule = xmalloc(sizeof *rule);
5510 rule_dealloc(struct rule *rule_)
5512 struct rule_dpif *rule = rule_dpif_cast(rule_);
5517 rule_construct(struct rule *rule_)
5519 struct rule_dpif *rule = rule_dpif_cast(rule_);
5520 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5521 struct rule_dpif *victim;
5524 rule->packet_count = 0;
5525 rule->byte_count = 0;
5527 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
5528 if (victim && !list_is_empty(&victim->facets)) {
5529 struct facet *facet;
5531 rule->facets = victim->facets;
5532 list_moved(&rule->facets);
5533 LIST_FOR_EACH (facet, list_node, &rule->facets) {
5534 /* XXX: We're only clearing our local counters here. It's possible
5535 * that quite a few packets are unaccounted for in the datapath
5536 * statistics. These will be accounted to the new rule instead of
5537 * cleared as required. This could be fixed by clearing out the
5538 * datapath statistics for this facet, but currently it doesn't
5540 facet_reset_counters(facet);
5544 /* Must avoid list_moved() in this case. */
5545 list_init(&rule->facets);
5548 table_id = rule->up.table_id;
5550 rule->tag = victim->tag;
5551 } else if (table_id == 0) {
5556 miniflow_expand(&rule->up.cr.match.flow, &flow);
5557 rule->tag = rule_calculate_tag(&flow, &rule->up.cr.match.mask,
5558 ofproto->tables[table_id].basis);
5561 complete_operation(rule);
5566 rule_destruct(struct rule *rule_)
5568 struct rule_dpif *rule = rule_dpif_cast(rule_);
5569 struct facet *facet, *next_facet;
5571 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
5572 facet_revalidate(facet);
5575 complete_operation(rule);
5579 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
5581 struct rule_dpif *rule = rule_dpif_cast(rule_);
5583 /* push_all_stats() can handle flow misses which, when using the learn
5584 * action, can cause rules to be added and deleted. This can corrupt our
5585 * caller's datastructures which assume that rule_get_stats() doesn't have
5586 * an impact on the flow table. To be safe, we disable miss handling. */
5587 push_all_stats__(false);
5589 /* Start from historical data for 'rule' itself that are no longer tracked
5590 * in facets. This counts, for example, facets that have expired. */
5591 *packets = rule->packet_count;
5592 *bytes = rule->byte_count;
5596 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
5597 struct ofpbuf *packet)
5599 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5600 struct initial_vals initial_vals;
5601 struct dpif_flow_stats stats;
5602 struct xlate_out xout;
5603 struct xlate_in xin;
5605 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
5606 rule_credit_stats(rule, &stats);
5608 initial_vals.vlan_tci = flow->vlan_tci;
5609 xlate_in_init(&xin, ofproto, flow, &initial_vals, rule, stats.tcp_flags,
5611 xin.resubmit_stats = &stats;
5612 xlate_actions(&xin, &xout);
5614 execute_odp_actions(ofproto, flow, xout.odp_actions.data,
5615 xout.odp_actions.size, packet);
5617 xlate_out_uninit(&xout);
5621 rule_execute(struct rule *rule, const struct flow *flow,
5622 struct ofpbuf *packet)
5624 rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
5625 ofpbuf_delete(packet);
5630 rule_modify_actions(struct rule *rule_)
5632 struct rule_dpif *rule = rule_dpif_cast(rule_);
5634 complete_operation(rule);
5637 /* Sends 'packet' out 'ofport'.
5638 * May modify 'packet'.
5639 * Returns 0 if successful, otherwise a positive errno value. */
5641 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
5643 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
5644 uint64_t odp_actions_stub[1024 / 8];
5645 struct ofpbuf key, odp_actions;
5646 struct dpif_flow_stats stats;
5647 struct odputil_keybuf keybuf;
5648 struct ofpact_output output;
5649 struct xlate_out xout;
5650 struct xlate_in xin;
5654 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5655 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5657 /* Use OFPP_NONE as the in_port to avoid special packet processing. */
5658 flow_extract(packet, 0, 0, NULL, OFPP_NONE, &flow);
5659 odp_flow_key_from_flow(&key, &flow, ofp_port_to_odp_port(ofproto,
5661 dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
5663 ofpact_init(&output.ofpact, OFPACT_OUTPUT, sizeof output);
5664 output.port = ofport->up.ofp_port;
5667 xlate_in_init(&xin, ofproto, &flow, NULL, NULL, 0, packet);
5668 xin.ofpacts_len = sizeof output;
5669 xin.ofpacts = &output.ofpact;
5670 xin.resubmit_stats = &stats;
5671 xlate_actions(&xin, &xout);
5673 error = dpif_execute(ofproto->backer->dpif,
5675 xout.odp_actions.data, xout.odp_actions.size,
5677 xlate_out_uninit(&xout);
5680 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %s (%s)",
5681 ofproto->up.name, netdev_get_name(ofport->up.netdev),
5685 ofproto->stats.tx_packets++;
5686 ofproto->stats.tx_bytes += packet->size;
5690 /* OpenFlow to datapath action translation. */
5692 static bool may_receive(const struct ofport_dpif *, struct xlate_ctx *);
5693 static void do_xlate_actions(const struct ofpact *, size_t ofpacts_len,
5694 struct xlate_ctx *);
5695 static void xlate_normal(struct xlate_ctx *);
5697 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
5698 * The action will state 'slow' as the reason that the action is in the slow
5699 * path. (This is purely informational: it allows a human viewing "ovs-dpctl
5700 * dump-flows" output to see why a flow is in the slow path.)
5702 * The 'stub_size' bytes in 'stub' will be used to store the action.
5703 * 'stub_size' must be large enough for the action.
5705 * The action and its size will be stored in '*actionsp' and '*actions_lenp',
5708 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
5709 enum slow_path_reason slow,
5710 uint64_t *stub, size_t stub_size,
5711 const struct nlattr **actionsp, size_t *actions_lenp)
5713 union user_action_cookie cookie;
5716 cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
5717 cookie.slow_path.unused = 0;
5718 cookie.slow_path.reason = slow;
5720 ofpbuf_use_stack(&buf, stub, stub_size);
5721 if (slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP | SLOW_STP)) {
5722 uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif, UINT32_MAX);
5723 odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
5725 put_userspace_action(ofproto, &buf, flow, &cookie,
5726 sizeof cookie.slow_path);
5728 *actionsp = buf.data;
5729 *actions_lenp = buf.size;
5733 put_userspace_action(const struct ofproto_dpif *ofproto,
5734 struct ofpbuf *odp_actions,
5735 const struct flow *flow,
5736 const union user_action_cookie *cookie,
5737 const size_t cookie_size)
5741 pid = dpif_port_get_pid(ofproto->backer->dpif,
5742 ofp_port_to_odp_port(ofproto, flow->in_port));
5744 return odp_put_userspace_action(pid, cookie, cookie_size, odp_actions);
5747 /* Compose SAMPLE action for sFlow or IPFIX. The given probability is
5748 * the number of packets out of UINT32_MAX to sample. The given
5749 * cookie is passed back in the callback for each sampled packet.
5752 compose_sample_action(const struct ofproto_dpif *ofproto,
5753 struct ofpbuf *odp_actions,
5754 const struct flow *flow,
5755 const uint32_t probability,
5756 const union user_action_cookie *cookie,
5757 const size_t cookie_size)
5759 size_t sample_offset, actions_offset;
5762 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
5764 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
5766 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
5767 cookie_offset = put_userspace_action(ofproto, odp_actions, flow, cookie,
5770 nl_msg_end_nested(odp_actions, actions_offset);
5771 nl_msg_end_nested(odp_actions, sample_offset);
5772 return cookie_offset;
5776 compose_sflow_cookie(const struct ofproto_dpif *ofproto,
5777 ovs_be16 vlan_tci, uint32_t odp_port,
5778 unsigned int n_outputs, union user_action_cookie *cookie)
5782 cookie->type = USER_ACTION_COOKIE_SFLOW;
5783 cookie->sflow.vlan_tci = vlan_tci;
5785 /* See http://www.sflow.org/sflow_version_5.txt (search for "Input/output
5786 * port information") for the interpretation of cookie->output. */
5787 switch (n_outputs) {
5789 /* 0x40000000 | 256 means "packet dropped for unknown reason". */
5790 cookie->sflow.output = 0x40000000 | 256;
5794 ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port);
5796 cookie->sflow.output = ifindex;
5801 /* 0x80000000 means "multiple output ports. */
5802 cookie->sflow.output = 0x80000000 | n_outputs;
5807 /* Compose SAMPLE action for sFlow bridge sampling. */
5809 compose_sflow_action(const struct ofproto_dpif *ofproto,
5810 struct ofpbuf *odp_actions,
5811 const struct flow *flow,
5814 uint32_t probability;
5815 union user_action_cookie cookie;
5817 if (!ofproto->sflow || flow->in_port == OFPP_NONE) {
5821 probability = dpif_sflow_get_probability(ofproto->sflow);
5822 compose_sflow_cookie(ofproto, htons(0), odp_port,
5823 odp_port == OVSP_NONE ? 0 : 1, &cookie);
5825 return compose_sample_action(ofproto, odp_actions, flow, probability,
5826 &cookie, sizeof cookie.sflow);
5830 compose_flow_sample_cookie(uint16_t probability, uint32_t collector_set_id,
5831 uint32_t obs_domain_id, uint32_t obs_point_id,
5832 union user_action_cookie *cookie)
5834 cookie->type = USER_ACTION_COOKIE_FLOW_SAMPLE;
5835 cookie->flow_sample.probability = probability;
5836 cookie->flow_sample.collector_set_id = collector_set_id;
5837 cookie->flow_sample.obs_domain_id = obs_domain_id;
5838 cookie->flow_sample.obs_point_id = obs_point_id;
5842 compose_ipfix_cookie(union user_action_cookie *cookie)
5844 cookie->type = USER_ACTION_COOKIE_IPFIX;
5847 /* Compose SAMPLE action for IPFIX bridge sampling. */
5849 compose_ipfix_action(const struct ofproto_dpif *ofproto,
5850 struct ofpbuf *odp_actions,
5851 const struct flow *flow)
5853 uint32_t probability;
5854 union user_action_cookie cookie;
5856 if (!ofproto->ipfix || flow->in_port == OFPP_NONE) {
5860 probability = dpif_ipfix_get_bridge_exporter_probability(ofproto->ipfix);
5861 compose_ipfix_cookie(&cookie);
5863 compose_sample_action(ofproto, odp_actions, flow, probability,
5864 &cookie, sizeof cookie.ipfix);
5867 /* SAMPLE action for sFlow must be first action in any given list of
5868 * actions. At this point we do not have all information required to
5869 * build it. So try to build sample action as complete as possible. */
5871 add_sflow_action(struct xlate_ctx *ctx)
5873 ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto,
5874 &ctx->xout->odp_actions,
5875 &ctx->xin->flow, OVSP_NONE);
5876 ctx->sflow_odp_port = 0;
5877 ctx->sflow_n_outputs = 0;
5880 /* SAMPLE action for IPFIX must be 1st or 2nd action in any given list
5881 * of actions, eventually after the SAMPLE action for sFlow. */
5883 add_ipfix_action(struct xlate_ctx *ctx)
5885 compose_ipfix_action(ctx->ofproto, &ctx->xout->odp_actions,
5889 /* Fix SAMPLE action according to data collected while composing ODP actions.
5890 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
5891 * USERSPACE action's user-cookie which is required for sflow. */
5893 fix_sflow_action(struct xlate_ctx *ctx)
5895 const struct flow *base = &ctx->base_flow;
5896 union user_action_cookie *cookie;
5898 if (!ctx->user_cookie_offset) {
5902 cookie = ofpbuf_at(&ctx->xout->odp_actions, ctx->user_cookie_offset,
5903 sizeof cookie->sflow);
5904 ovs_assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
5906 compose_sflow_cookie(ctx->ofproto, base->vlan_tci,
5907 ctx->sflow_odp_port, ctx->sflow_n_outputs, cookie);
5911 compose_output_action__(struct xlate_ctx *ctx, uint16_t ofp_port,
5914 const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
5915 ovs_be16 flow_vlan_tci;
5916 uint32_t flow_skb_mark;
5917 uint8_t flow_nw_tos;
5918 struct priority_to_dscp *pdscp;
5919 uint32_t out_port, odp_port;
5921 /* If 'struct flow' gets additional metadata, we'll need to zero it out
5922 * before traversing a patch port. */
5923 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 20);
5926 xlate_report(ctx, "Nonexistent output port");
5928 } else if (ofport->up.pp.config & OFPUTIL_PC_NO_FWD) {
5929 xlate_report(ctx, "OFPPC_NO_FWD set, skipping output");
5931 } else if (check_stp && !stp_forward_in_state(ofport->stp_state)) {
5932 xlate_report(ctx, "STP not in forwarding state, skipping output");
5936 if (netdev_vport_is_patch(ofport->up.netdev)) {
5937 struct ofport_dpif *peer = ofport_get_peer(ofport);
5938 struct flow old_flow = ctx->xin->flow;
5939 const struct ofproto_dpif *peer_ofproto;
5940 enum slow_path_reason special;
5941 struct ofport_dpif *in_port;
5944 xlate_report(ctx, "Nonexistent patch port peer");
5948 peer_ofproto = ofproto_dpif_cast(peer->up.ofproto);
5949 if (peer_ofproto->backer != ctx->ofproto->backer) {
5950 xlate_report(ctx, "Patch port peer on a different datapath");
5954 ctx->ofproto = ofproto_dpif_cast(peer->up.ofproto);
5955 ctx->xin->flow.in_port = peer->up.ofp_port;
5956 ctx->xin->flow.metadata = htonll(0);
5957 memset(&ctx->xin->flow.tunnel, 0, sizeof ctx->xin->flow.tunnel);
5958 memset(ctx->xin->flow.regs, 0, sizeof ctx->xin->flow.regs);
5960 in_port = get_ofp_port(ctx->ofproto, ctx->xin->flow.in_port);
5961 special = process_special(ctx->ofproto, &ctx->xin->flow, in_port,
5964 ctx->xout->slow = special;
5965 } else if (!in_port || may_receive(in_port, ctx)) {
5966 if (!in_port || stp_forward_in_state(in_port->stp_state)) {
5967 xlate_table_action(ctx, ctx->xin->flow.in_port, 0, true);
5969 /* Forwarding is disabled by STP. Let OFPP_NORMAL and the
5970 * learning action look at the packet, then drop it. */
5971 struct flow old_base_flow = ctx->base_flow;
5972 size_t old_size = ctx->xout->odp_actions.size;
5973 xlate_table_action(ctx, ctx->xin->flow.in_port, 0, true);
5974 ctx->base_flow = old_base_flow;
5975 ctx->xout->odp_actions.size = old_size;
5979 ctx->xin->flow = old_flow;
5980 ctx->ofproto = ofproto_dpif_cast(ofport->up.ofproto);
5982 if (ctx->xin->resubmit_stats) {
5983 netdev_vport_inc_tx(ofport->up.netdev, ctx->xin->resubmit_stats);
5984 netdev_vport_inc_rx(peer->up.netdev, ctx->xin->resubmit_stats);
5990 flow_vlan_tci = ctx->xin->flow.vlan_tci;
5991 flow_skb_mark = ctx->xin->flow.skb_mark;
5992 flow_nw_tos = ctx->xin->flow.nw_tos;
5994 pdscp = get_priority(ofport, ctx->xin->flow.skb_priority);
5996 ctx->xin->flow.nw_tos &= ~IP_DSCP_MASK;
5997 ctx->xin->flow.nw_tos |= pdscp->dscp;
6000 if (ofport->tnl_port) {
6001 /* Save tunnel metadata so that changes made due to
6002 * the Logical (tunnel) Port are not visible for any further
6003 * matches, while explicit set actions on tunnel metadata are.
6005 struct flow_tnl flow_tnl = ctx->xin->flow.tunnel;
6006 odp_port = tnl_port_send(ofport->tnl_port, &ctx->xin->flow);
6007 if (odp_port == OVSP_NONE) {
6008 xlate_report(ctx, "Tunneling decided against output");
6009 goto out; /* restore flow_nw_tos */
6011 if (ctx->xin->flow.tunnel.ip_dst == ctx->orig_tunnel_ip_dst) {
6012 xlate_report(ctx, "Not tunneling to our own address");
6013 goto out; /* restore flow_nw_tos */
6015 if (ctx->xin->resubmit_stats) {
6016 netdev_vport_inc_tx(ofport->up.netdev, ctx->xin->resubmit_stats);
6018 out_port = odp_port;
6019 commit_odp_tunnel_action(&ctx->xin->flow, &ctx->base_flow,
6020 &ctx->xout->odp_actions);
6021 ctx->xin->flow.tunnel = flow_tnl; /* Restore tunnel metadata */
6023 uint16_t vlandev_port;
6024 odp_port = ofport->odp_port;
6025 vlandev_port = vsp_realdev_to_vlandev(ctx->ofproto, ofp_port,
6026 ctx->xin->flow.vlan_tci);
6027 if (vlandev_port == ofp_port) {
6028 out_port = odp_port;
6030 out_port = ofp_port_to_odp_port(ctx->ofproto, vlandev_port);
6031 ctx->xin->flow.vlan_tci = htons(0);
6033 ctx->xin->flow.skb_mark &= ~IPSEC_MARK;
6035 commit_odp_actions(&ctx->xin->flow, &ctx->base_flow,
6036 &ctx->xout->odp_actions);
6037 nl_msg_put_u32(&ctx->xout->odp_actions, OVS_ACTION_ATTR_OUTPUT, out_port);
6039 ctx->sflow_odp_port = odp_port;
6040 ctx->sflow_n_outputs++;
6041 ctx->xout->nf_output_iface = ofp_port;
6044 ctx->xin->flow.vlan_tci = flow_vlan_tci;
6045 ctx->xin->flow.skb_mark = flow_skb_mark;
6047 ctx->xin->flow.nw_tos = flow_nw_tos;
6051 compose_output_action(struct xlate_ctx *ctx, uint16_t ofp_port)
6053 compose_output_action__(ctx, ofp_port, true);
6057 tag_the_flow(struct xlate_ctx *ctx, struct rule_dpif *rule)
6059 struct ofproto_dpif *ofproto = ctx->ofproto;
6060 uint8_t table_id = ctx->table_id;
6062 if (table_id > 0 && table_id < N_TABLES) {
6063 struct table_dpif *table = &ofproto->tables[table_id];
6064 if (table->other_table) {
6065 ctx->xout->tags |= (rule && rule->tag
6067 : rule_calculate_tag(&ctx->xin->flow,
6068 &table->other_table->mask,
6074 /* Common rule processing in one place to avoid duplicating code. */
6075 static struct rule_dpif *
6076 ctx_rule_hooks(struct xlate_ctx *ctx, struct rule_dpif *rule,
6079 if (ctx->xin->resubmit_hook) {
6080 ctx->xin->resubmit_hook(ctx, rule);
6082 if (rule == NULL && may_packet_in) {
6084 * check if table configuration flags
6085 * OFPTC_TABLE_MISS_CONTROLLER, default.
6086 * OFPTC_TABLE_MISS_CONTINUE,
6087 * OFPTC_TABLE_MISS_DROP
6088 * When OF1.0, OFPTC_TABLE_MISS_CONTINUE is used. What to do?
6090 rule = rule_dpif_miss_rule(ctx->ofproto, &ctx->xin->flow);
6092 if (rule && ctx->xin->resubmit_stats) {
6093 rule_credit_stats(rule, ctx->xin->resubmit_stats);
6099 xlate_table_action(struct xlate_ctx *ctx,
6100 uint16_t in_port, uint8_t table_id, bool may_packet_in)
6102 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
6103 struct rule_dpif *rule;
6104 uint16_t old_in_port = ctx->xin->flow.in_port;
6105 uint8_t old_table_id = ctx->table_id;
6107 ctx->table_id = table_id;
6109 /* Look up a flow with 'in_port' as the input port. */
6110 ctx->xin->flow.in_port = in_port;
6111 rule = rule_dpif_lookup__(ctx->ofproto, &ctx->xin->flow, table_id);
6113 tag_the_flow(ctx, rule);
6115 /* Restore the original input port. Otherwise OFPP_NORMAL and
6116 * OFPP_IN_PORT will have surprising behavior. */
6117 ctx->xin->flow.in_port = old_in_port;
6119 rule = ctx_rule_hooks(ctx, rule, may_packet_in);
6122 struct rule_dpif *old_rule = ctx->rule;
6126 do_xlate_actions(rule->up.ofpacts, rule->up.ofpacts_len, ctx);
6127 ctx->rule = old_rule;
6131 ctx->table_id = old_table_id;
6133 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
6135 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
6136 MAX_RESUBMIT_RECURSION);
6137 ctx->max_resubmit_trigger = true;
6142 xlate_ofpact_resubmit(struct xlate_ctx *ctx,
6143 const struct ofpact_resubmit *resubmit)
6148 in_port = resubmit->in_port;
6149 if (in_port == OFPP_IN_PORT) {
6150 in_port = ctx->xin->flow.in_port;
6153 table_id = resubmit->table_id;
6154 if (table_id == 255) {
6155 table_id = ctx->table_id;
6158 xlate_table_action(ctx, in_port, table_id, false);
6162 flood_packets(struct xlate_ctx *ctx, bool all)
6164 struct ofport_dpif *ofport;
6166 HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) {
6167 uint16_t ofp_port = ofport->up.ofp_port;
6169 if (ofp_port == ctx->xin->flow.in_port) {
6174 compose_output_action__(ctx, ofp_port, false);
6175 } else if (!(ofport->up.pp.config & OFPUTIL_PC_NO_FLOOD)) {
6176 compose_output_action(ctx, ofp_port);
6180 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
6184 execute_controller_action(struct xlate_ctx *ctx, int len,
6185 enum ofp_packet_in_reason reason,
6186 uint16_t controller_id)
6188 struct ofputil_packet_in pin;
6189 struct ofpbuf *packet;
6192 ovs_assert(!ctx->xout->slow || ctx->xout->slow == SLOW_CONTROLLER);
6193 ctx->xout->slow = SLOW_CONTROLLER;
6194 if (!ctx->xin->packet) {
6198 packet = ofpbuf_clone(ctx->xin->packet);
6200 key.skb_priority = 0;
6202 memset(&key.tunnel, 0, sizeof key.tunnel);
6204 commit_odp_actions(&ctx->xin->flow, &ctx->base_flow,
6205 &ctx->xout->odp_actions);
6207 odp_execute_actions(NULL, packet, &key, ctx->xout->odp_actions.data,
6208 ctx->xout->odp_actions.size, NULL, NULL);
6210 pin.packet = packet->data;
6211 pin.packet_len = packet->size;
6212 pin.reason = reason;
6213 pin.controller_id = controller_id;
6214 pin.table_id = ctx->table_id;
6215 pin.cookie = ctx->rule ? ctx->rule->up.flow_cookie : 0;
6218 flow_get_metadata(&ctx->xin->flow, &pin.fmd);
6220 connmgr_send_packet_in(ctx->ofproto->up.connmgr, &pin);
6221 ofpbuf_delete(packet);
6225 execute_mpls_push_action(struct xlate_ctx *ctx, ovs_be16 eth_type)
6227 ovs_assert(eth_type_mpls(eth_type));
6229 if (ctx->base_flow.mpls_depth) {
6230 ctx->xin->flow.mpls_lse &= ~htonl(MPLS_BOS_MASK);
6231 ctx->xin->flow.mpls_depth++;
6236 if (ctx->xin->flow.dl_type == htons(ETH_TYPE_IPV6)) {
6237 label = htonl(0x2); /* IPV6 Explicit Null. */
6239 label = htonl(0x0); /* IPV4 Explicit Null. */
6241 tc = (ctx->xin->flow.nw_tos & IP_DSCP_MASK) >> 2;
6242 ttl = ctx->xin->flow.nw_ttl ? ctx->xin->flow.nw_ttl : 0x40;
6243 ctx->xin->flow.mpls_lse = set_mpls_lse_values(ttl, tc, 1, label);
6244 ctx->xin->flow.mpls_depth = 1;
6246 ctx->xin->flow.dl_type = eth_type;
6250 execute_mpls_pop_action(struct xlate_ctx *ctx, ovs_be16 eth_type)
6252 ovs_assert(eth_type_mpls(ctx->xin->flow.dl_type));
6253 ovs_assert(!eth_type_mpls(eth_type));
6255 if (ctx->xin->flow.mpls_depth) {
6256 ctx->xin->flow.mpls_depth--;
6257 ctx->xin->flow.mpls_lse = htonl(0);
6258 if (!ctx->xin->flow.mpls_depth) {
6259 ctx->xin->flow.dl_type = eth_type;
6265 compose_dec_ttl(struct xlate_ctx *ctx, struct ofpact_cnt_ids *ids)
6267 if (ctx->xin->flow.dl_type != htons(ETH_TYPE_IP) &&
6268 ctx->xin->flow.dl_type != htons(ETH_TYPE_IPV6)) {
6272 if (ctx->xin->flow.nw_ttl > 1) {
6273 ctx->xin->flow.nw_ttl--;
6278 for (i = 0; i < ids->n_controllers; i++) {
6279 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL,
6283 /* Stop processing for current table. */
6289 execute_set_mpls_ttl_action(struct xlate_ctx *ctx, uint8_t ttl)
6291 if (!eth_type_mpls(ctx->xin->flow.dl_type)) {
6295 set_mpls_lse_ttl(&ctx->xin->flow.mpls_lse, ttl);
6300 execute_dec_mpls_ttl_action(struct xlate_ctx *ctx)
6302 uint8_t ttl = mpls_lse_to_ttl(ctx->xin->flow.mpls_lse);
6304 if (!eth_type_mpls(ctx->xin->flow.dl_type)) {
6310 set_mpls_lse_ttl(&ctx->xin->flow.mpls_lse, ttl);
6313 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL, 0);
6315 /* Stop processing for current table. */
6321 xlate_output_action(struct xlate_ctx *ctx,
6322 uint16_t port, uint16_t max_len, bool may_packet_in)
6324 uint16_t prev_nf_output_iface = ctx->xout->nf_output_iface;
6326 ctx->xout->nf_output_iface = NF_OUT_DROP;
6330 compose_output_action(ctx, ctx->xin->flow.in_port);
6333 xlate_table_action(ctx, ctx->xin->flow.in_port, 0, may_packet_in);
6339 flood_packets(ctx, false);
6342 flood_packets(ctx, true);
6344 case OFPP_CONTROLLER:
6345 execute_controller_action(ctx, max_len, OFPR_ACTION, 0);
6351 if (port != ctx->xin->flow.in_port) {
6352 compose_output_action(ctx, port);
6354 xlate_report(ctx, "skipping output to input port");
6359 if (prev_nf_output_iface == NF_OUT_FLOOD) {
6360 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
6361 } else if (ctx->xout->nf_output_iface == NF_OUT_DROP) {
6362 ctx->xout->nf_output_iface = prev_nf_output_iface;
6363 } else if (prev_nf_output_iface != NF_OUT_DROP &&
6364 ctx->xout->nf_output_iface != NF_OUT_FLOOD) {
6365 ctx->xout->nf_output_iface = NF_OUT_MULTI;
6370 xlate_output_reg_action(struct xlate_ctx *ctx,
6371 const struct ofpact_output_reg *or)
6373 uint64_t port = mf_get_subfield(&or->src, &ctx->xin->flow);
6374 if (port <= UINT16_MAX) {
6375 xlate_output_action(ctx, port, or->max_len, false);
6380 xlate_enqueue_action(struct xlate_ctx *ctx,
6381 const struct ofpact_enqueue *enqueue)
6383 uint16_t ofp_port = enqueue->port;
6384 uint32_t queue_id = enqueue->queue;
6385 uint32_t flow_priority, priority;
6388 /* Translate queue to priority. */
6389 error = dpif_queue_to_priority(ctx->ofproto->backer->dpif,
6390 queue_id, &priority);
6392 /* Fall back to ordinary output action. */
6393 xlate_output_action(ctx, enqueue->port, 0, false);
6397 /* Check output port. */
6398 if (ofp_port == OFPP_IN_PORT) {
6399 ofp_port = ctx->xin->flow.in_port;
6400 } else if (ofp_port == ctx->xin->flow.in_port) {
6404 /* Add datapath actions. */
6405 flow_priority = ctx->xin->flow.skb_priority;
6406 ctx->xin->flow.skb_priority = priority;
6407 compose_output_action(ctx, ofp_port);
6408 ctx->xin->flow.skb_priority = flow_priority;
6410 /* Update NetFlow output port. */
6411 if (ctx->xout->nf_output_iface == NF_OUT_DROP) {
6412 ctx->xout->nf_output_iface = ofp_port;
6413 } else if (ctx->xout->nf_output_iface != NF_OUT_FLOOD) {
6414 ctx->xout->nf_output_iface = NF_OUT_MULTI;
6419 xlate_set_queue_action(struct xlate_ctx *ctx, uint32_t queue_id)
6421 uint32_t skb_priority;
6423 if (!dpif_queue_to_priority(ctx->ofproto->backer->dpif,
6424 queue_id, &skb_priority)) {
6425 ctx->xin->flow.skb_priority = skb_priority;
6427 /* Couldn't translate queue to a priority. Nothing to do. A warning
6428 * has already been logged. */
6433 slave_enabled_cb(uint16_t ofp_port, void *ofproto_)
6435 struct ofproto_dpif *ofproto = ofproto_;
6436 struct ofport_dpif *port;
6446 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
6449 port = get_ofp_port(ofproto, ofp_port);
6450 return port ? port->may_enable : false;
6455 xlate_bundle_action(struct xlate_ctx *ctx,
6456 const struct ofpact_bundle *bundle)
6460 port = bundle_execute(bundle, &ctx->xin->flow, slave_enabled_cb,
6462 if (bundle->dst.field) {
6463 nxm_reg_load(&bundle->dst, port, &ctx->xin->flow);
6465 xlate_output_action(ctx, port, 0, false);
6470 xlate_learn_action(struct xlate_ctx *ctx,
6471 const struct ofpact_learn *learn)
6473 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
6474 struct ofputil_flow_mod fm;
6475 uint64_t ofpacts_stub[1024 / 8];
6476 struct ofpbuf ofpacts;
6479 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
6480 learn_execute(learn, &ctx->xin->flow, &fm, &ofpacts);
6482 error = ofproto_flow_mod(&ctx->ofproto->up, &fm);
6483 if (error && !VLOG_DROP_WARN(&rl)) {
6484 VLOG_WARN("learning action failed to modify flow table (%s)",
6485 ofperr_get_name(error));
6488 ofpbuf_uninit(&ofpacts);
6491 /* Reduces '*timeout' to no more than 'max'. A value of zero in either case
6492 * means "infinite". */
6494 reduce_timeout(uint16_t max, uint16_t *timeout)
6496 if (max && (!*timeout || *timeout > max)) {
6502 xlate_fin_timeout(struct xlate_ctx *ctx,
6503 const struct ofpact_fin_timeout *oft)
6505 if (ctx->xin->tcp_flags & (TCP_FIN | TCP_RST) && ctx->rule) {
6506 struct rule_dpif *rule = ctx->rule;
6508 reduce_timeout(oft->fin_idle_timeout, &rule->up.idle_timeout);
6509 reduce_timeout(oft->fin_hard_timeout, &rule->up.hard_timeout);
6514 xlate_sample_action(struct xlate_ctx *ctx,
6515 const struct ofpact_sample *os)
6517 union user_action_cookie cookie;
6518 /* Scale the probability from 16-bit to 32-bit while representing
6519 * the same percentage. */
6520 uint32_t probability = (os->probability << 16) | os->probability;
6522 commit_odp_actions(&ctx->xin->flow, &ctx->base_flow,
6523 &ctx->xout->odp_actions);
6525 compose_flow_sample_cookie(os->probability, os->collector_set_id,
6526 os->obs_domain_id, os->obs_point_id, &cookie);
6527 compose_sample_action(ctx->ofproto, &ctx->xout->odp_actions, &ctx->xin->flow,
6528 probability, &cookie, sizeof cookie.flow_sample);
6532 may_receive(const struct ofport_dpif *port, struct xlate_ctx *ctx)
6534 if (port->up.pp.config & (eth_addr_equals(ctx->xin->flow.dl_dst,
6536 ? OFPUTIL_PC_NO_RECV_STP
6537 : OFPUTIL_PC_NO_RECV)) {
6541 /* Only drop packets here if both forwarding and learning are
6542 * disabled. If just learning is enabled, we need to have
6543 * OFPP_NORMAL and the learning action have a look at the packet
6544 * before we can drop it. */
6545 if (!stp_forward_in_state(port->stp_state)
6546 && !stp_learn_in_state(port->stp_state)) {
6554 tunnel_ecn_ok(struct xlate_ctx *ctx)
6556 if (is_ip_any(&ctx->base_flow)
6557 && (ctx->xin->flow.tunnel.ip_tos & IP_ECN_MASK) == IP_ECN_CE) {
6558 if ((ctx->base_flow.nw_tos & IP_ECN_MASK) == IP_ECN_NOT_ECT) {
6559 VLOG_WARN_RL(&rl, "dropping tunnel packet marked ECN CE"
6560 " but is not ECN capable");
6563 /* Set the ECN CE value in the tunneled packet. */
6564 ctx->xin->flow.nw_tos |= IP_ECN_CE;
6572 do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len,
6573 struct xlate_ctx *ctx)
6575 bool was_evictable = true;
6576 const struct ofpact *a;
6579 /* Don't let the rule we're working on get evicted underneath us. */
6580 was_evictable = ctx->rule->up.evictable;
6581 ctx->rule->up.evictable = false;
6584 do_xlate_actions_again:
6585 OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) {
6586 struct ofpact_controller *controller;
6587 const struct ofpact_metadata *metadata;
6595 xlate_output_action(ctx, ofpact_get_OUTPUT(a)->port,
6596 ofpact_get_OUTPUT(a)->max_len, true);
6599 case OFPACT_CONTROLLER:
6600 controller = ofpact_get_CONTROLLER(a);
6601 execute_controller_action(ctx, controller->max_len,
6603 controller->controller_id);
6606 case OFPACT_ENQUEUE:
6607 xlate_enqueue_action(ctx, ofpact_get_ENQUEUE(a));
6610 case OFPACT_SET_VLAN_VID:
6611 ctx->xin->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
6612 ctx->xin->flow.vlan_tci |=
6613 (htons(ofpact_get_SET_VLAN_VID(a)->vlan_vid)
6617 case OFPACT_SET_VLAN_PCP:
6618 ctx->xin->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
6619 ctx->xin->flow.vlan_tci |=
6620 htons((ofpact_get_SET_VLAN_PCP(a)->vlan_pcp << VLAN_PCP_SHIFT)
6624 case OFPACT_STRIP_VLAN:
6625 ctx->xin->flow.vlan_tci = htons(0);
6628 case OFPACT_PUSH_VLAN:
6629 /* XXX 802.1AD(QinQ) */
6630 ctx->xin->flow.vlan_tci = htons(VLAN_CFI);
6633 case OFPACT_SET_ETH_SRC:
6634 memcpy(ctx->xin->flow.dl_src, ofpact_get_SET_ETH_SRC(a)->mac,
6638 case OFPACT_SET_ETH_DST:
6639 memcpy(ctx->xin->flow.dl_dst, ofpact_get_SET_ETH_DST(a)->mac,
6643 case OFPACT_SET_IPV4_SRC:
6644 if (ctx->xin->flow.dl_type == htons(ETH_TYPE_IP)) {
6645 ctx->xin->flow.nw_src = ofpact_get_SET_IPV4_SRC(a)->ipv4;
6649 case OFPACT_SET_IPV4_DST:
6650 if (ctx->xin->flow.dl_type == htons(ETH_TYPE_IP)) {
6651 ctx->xin->flow.nw_dst = ofpact_get_SET_IPV4_DST(a)->ipv4;
6655 case OFPACT_SET_IPV4_DSCP:
6656 /* OpenFlow 1.0 only supports IPv4. */
6657 if (ctx->xin->flow.dl_type == htons(ETH_TYPE_IP)) {
6658 ctx->xin->flow.nw_tos &= ~IP_DSCP_MASK;
6659 ctx->xin->flow.nw_tos |= ofpact_get_SET_IPV4_DSCP(a)->dscp;
6663 case OFPACT_SET_L4_SRC_PORT:
6664 if (is_ip_any(&ctx->xin->flow)) {
6665 ctx->xin->flow.tp_src =
6666 htons(ofpact_get_SET_L4_SRC_PORT(a)->port);
6670 case OFPACT_SET_L4_DST_PORT:
6671 if (is_ip_any(&ctx->xin->flow)) {
6672 ctx->xin->flow.tp_dst =
6673 htons(ofpact_get_SET_L4_DST_PORT(a)->port);
6677 case OFPACT_RESUBMIT:
6678 xlate_ofpact_resubmit(ctx, ofpact_get_RESUBMIT(a));
6681 case OFPACT_SET_TUNNEL:
6682 ctx->xin->flow.tunnel.tun_id =
6683 htonll(ofpact_get_SET_TUNNEL(a)->tun_id);
6686 case OFPACT_SET_QUEUE:
6687 xlate_set_queue_action(ctx, ofpact_get_SET_QUEUE(a)->queue_id);
6690 case OFPACT_POP_QUEUE:
6691 ctx->xin->flow.skb_priority = ctx->orig_skb_priority;
6694 case OFPACT_REG_MOVE:
6695 nxm_execute_reg_move(ofpact_get_REG_MOVE(a), &ctx->xin->flow);
6698 case OFPACT_REG_LOAD:
6699 nxm_execute_reg_load(ofpact_get_REG_LOAD(a), &ctx->xin->flow);
6702 case OFPACT_STACK_PUSH:
6703 nxm_execute_stack_push(ofpact_get_STACK_PUSH(a), &ctx->xin->flow,
6707 case OFPACT_STACK_POP:
6708 nxm_execute_stack_pop(ofpact_get_STACK_POP(a), &ctx->xin->flow,
6712 case OFPACT_PUSH_MPLS:
6713 execute_mpls_push_action(ctx, ofpact_get_PUSH_MPLS(a)->ethertype);
6716 case OFPACT_POP_MPLS:
6717 execute_mpls_pop_action(ctx, ofpact_get_POP_MPLS(a)->ethertype);
6720 case OFPACT_SET_MPLS_TTL:
6721 if (execute_set_mpls_ttl_action(ctx,
6722 ofpact_get_SET_MPLS_TTL(a)->ttl)) {
6727 case OFPACT_DEC_MPLS_TTL:
6728 if (execute_dec_mpls_ttl_action(ctx)) {
6733 case OFPACT_DEC_TTL:
6734 if (compose_dec_ttl(ctx, ofpact_get_DEC_TTL(a))) {
6740 /* Nothing to do. */
6743 case OFPACT_MULTIPATH:
6744 multipath_execute(ofpact_get_MULTIPATH(a), &ctx->xin->flow);
6748 ctx->ofproto->has_bundle_action = true;
6749 xlate_bundle_action(ctx, ofpact_get_BUNDLE(a));
6752 case OFPACT_OUTPUT_REG:
6753 xlate_output_reg_action(ctx, ofpact_get_OUTPUT_REG(a));
6757 ctx->xout->has_learn = true;
6758 if (ctx->xin->may_learn) {
6759 xlate_learn_action(ctx, ofpact_get_LEARN(a));
6767 case OFPACT_FIN_TIMEOUT:
6768 ctx->xout->has_fin_timeout = true;
6769 xlate_fin_timeout(ctx, ofpact_get_FIN_TIMEOUT(a));
6772 case OFPACT_CLEAR_ACTIONS:
6774 * Nothing to do because writa-actions is not supported for now.
6775 * When writa-actions is supported, clear-actions also must
6776 * be supported at the same time.
6780 case OFPACT_WRITE_METADATA:
6781 metadata = ofpact_get_WRITE_METADATA(a);
6782 ctx->xin->flow.metadata &= ~metadata->mask;
6783 ctx->xin->flow.metadata |= metadata->metadata & metadata->mask;
6786 case OFPACT_GOTO_TABLE: {
6787 /* It is assumed that goto-table is the last action. */
6788 struct ofpact_goto_table *ogt = ofpact_get_GOTO_TABLE(a);
6789 struct rule_dpif *rule;
6791 ovs_assert(ctx->table_id < ogt->table_id);
6793 ctx->table_id = ogt->table_id;
6795 /* Look up a flow from the new table. */
6796 rule = rule_dpif_lookup__(ctx->ofproto, &ctx->xin->flow, ctx->table_id);
6798 tag_the_flow(ctx, rule);
6800 rule = ctx_rule_hooks(ctx, rule, true);
6804 ctx->rule->up.evictable = was_evictable;
6807 was_evictable = rule->up.evictable;
6808 rule->up.evictable = false;
6810 /* Tail recursion removal. */
6811 ofpacts = rule->up.ofpacts;
6812 ofpacts_len = rule->up.ofpacts_len;
6813 goto do_xlate_actions_again;
6819 xlate_sample_action(ctx, ofpact_get_SAMPLE(a));
6826 ctx->rule->up.evictable = was_evictable;
6831 xlate_in_init(struct xlate_in *xin, struct ofproto_dpif *ofproto,
6832 const struct flow *flow,
6833 const struct initial_vals *initial_vals,
6834 struct rule_dpif *rule, uint8_t tcp_flags,
6835 const struct ofpbuf *packet)
6837 xin->ofproto = ofproto;
6839 xin->packet = packet;
6840 xin->may_learn = packet != NULL;
6842 xin->ofpacts = NULL;
6843 xin->ofpacts_len = 0;
6844 xin->tcp_flags = tcp_flags;
6845 xin->resubmit_hook = NULL;
6846 xin->report_hook = NULL;
6847 xin->resubmit_stats = NULL;
6850 xin->initial_vals = *initial_vals;
6852 xin->initial_vals.vlan_tci = xin->flow.vlan_tci;
6857 xlate_out_uninit(struct xlate_out *xout)
6860 ofpbuf_uninit(&xout->odp_actions);
6864 /* Translates the 'ofpacts_len' bytes of "struct ofpacts" starting at 'ofpacts'
6865 * into datapath actions in 'odp_actions', using 'ctx'. */
6867 xlate_actions(struct xlate_in *xin, struct xlate_out *xout)
6869 /* Normally false. Set to true if we ever hit MAX_RESUBMIT_RECURSION, so
6870 * that in the future we always keep a copy of the original flow for
6871 * tracing purposes. */
6872 static bool hit_resubmit_limit;
6874 enum slow_path_reason special;
6875 const struct ofpact *ofpacts;
6876 struct ofport_dpif *in_port;
6877 struct flow orig_flow;
6878 struct xlate_ctx ctx;
6881 COVERAGE_INC(ofproto_dpif_xlate);
6883 /* Flow initialization rules:
6884 * - 'base_flow' must match the kernel's view of the packet at the
6885 * time that action processing starts. 'flow' represents any
6886 * transformations we wish to make through actions.
6887 * - By default 'base_flow' and 'flow' are the same since the input
6888 * packet matches the output before any actions are applied.
6889 * - When using VLAN splinters, 'base_flow''s VLAN is set to the value
6890 * of the received packet as seen by the kernel. If we later output
6891 * to another device without any modifications this will cause us to
6892 * insert a new tag since the original one was stripped off by the
6894 * - Tunnel metadata as received is retained in 'flow'. This allows
6895 * tunnel metadata matching also in later tables.
6896 * Since a kernel action for setting the tunnel metadata will only be
6897 * generated with actual tunnel output, changing the tunnel metadata
6898 * values in 'flow' (such as tun_id) will only have effect with a later
6899 * tunnel output action.
6900 * - Tunnel 'base_flow' is completely cleared since that is what the
6901 * kernel does. If we wish to maintain the original values an action
6902 * needs to be generated. */
6907 ctx.ofproto = xin->ofproto;
6908 ctx.rule = xin->rule;
6910 ctx.base_flow = ctx.xin->flow;
6911 ctx.base_flow.vlan_tci = xin->initial_vals.vlan_tci;
6912 memset(&ctx.base_flow.tunnel, 0, sizeof ctx.base_flow.tunnel);
6913 ctx.orig_tunnel_ip_dst = ctx.xin->flow.tunnel.ip_dst;
6917 ctx.xout->has_learn = false;
6918 ctx.xout->has_normal = false;
6919 ctx.xout->has_fin_timeout = false;
6920 ctx.xout->nf_output_iface = NF_OUT_DROP;
6921 ctx.xout->mirrors = 0;
6923 ofpbuf_use_stub(&ctx.xout->odp_actions, ctx.xout->odp_actions_stub,
6924 sizeof ctx.xout->odp_actions_stub);
6925 ofpbuf_reserve(&ctx.xout->odp_actions, NL_A_U32_SIZE);
6928 ctx.max_resubmit_trigger = false;
6929 ctx.orig_skb_priority = ctx.xin->flow.skb_priority;
6934 ofpacts = xin->ofpacts;
6935 ofpacts_len = xin->ofpacts_len;
6936 } else if (xin->rule) {
6937 ofpacts = xin->rule->up.ofpacts;
6938 ofpacts_len = xin->rule->up.ofpacts_len;
6943 ofpbuf_use_stub(&ctx.stack, ctx.init_stack, sizeof ctx.init_stack);
6945 if (ctx.ofproto->has_mirrors || hit_resubmit_limit) {
6946 /* Do this conditionally because the copy is expensive enough that it
6947 * shows up in profiles. */
6948 orig_flow = ctx.xin->flow;
6951 if (ctx.xin->flow.nw_frag & FLOW_NW_FRAG_ANY) {
6952 switch (ctx.ofproto->up.frag_handling) {
6953 case OFPC_FRAG_NORMAL:
6954 /* We must pretend that transport ports are unavailable. */
6955 ctx.xin->flow.tp_src = ctx.base_flow.tp_src = htons(0);
6956 ctx.xin->flow.tp_dst = ctx.base_flow.tp_dst = htons(0);
6959 case OFPC_FRAG_DROP:
6962 case OFPC_FRAG_REASM:
6965 case OFPC_FRAG_NX_MATCH:
6966 /* Nothing to do. */
6969 case OFPC_INVALID_TTL_TO_CONTROLLER:
6974 in_port = get_ofp_port(ctx.ofproto, ctx.xin->flow.in_port);
6975 special = process_special(ctx.ofproto, &ctx.xin->flow, in_port,
6978 ctx.xout->slow = special;
6980 static struct vlog_rate_limit trace_rl = VLOG_RATE_LIMIT_INIT(1, 1);
6981 struct initial_vals initial_vals;
6982 size_t sample_actions_len;
6983 uint32_t local_odp_port;
6985 initial_vals.vlan_tci = ctx.base_flow.vlan_tci;
6987 add_sflow_action(&ctx);
6988 add_ipfix_action(&ctx);
6989 sample_actions_len = ctx.xout->odp_actions.size;
6991 if (tunnel_ecn_ok(&ctx) && (!in_port || may_receive(in_port, &ctx))) {
6992 do_xlate_actions(ofpacts, ofpacts_len, &ctx);
6994 /* We've let OFPP_NORMAL and the learning action look at the
6995 * packet, so drop it now if forwarding is disabled. */
6996 if (in_port && !stp_forward_in_state(in_port->stp_state)) {
6997 ctx.xout->odp_actions.size = sample_actions_len;
7001 if (ctx.max_resubmit_trigger && !ctx.xin->resubmit_hook) {
7002 if (!hit_resubmit_limit) {
7003 /* We didn't record the original flow. Make sure we do from
7005 hit_resubmit_limit = true;
7006 } else if (!VLOG_DROP_ERR(&trace_rl)) {
7007 struct ds ds = DS_EMPTY_INITIALIZER;
7009 ofproto_trace(ctx.ofproto, &orig_flow, ctx.xin->packet,
7010 &initial_vals, &ds);
7011 VLOG_ERR("Trace triggered by excessive resubmit "
7012 "recursion:\n%s", ds_cstr(&ds));
7017 local_odp_port = ofp_port_to_odp_port(ctx.ofproto, OFPP_LOCAL);
7018 if (!connmgr_must_output_local(ctx.ofproto->up.connmgr, &ctx.xin->flow,
7020 ctx.xout->odp_actions.data,
7021 ctx.xout->odp_actions.size)) {
7022 compose_output_action(&ctx, OFPP_LOCAL);
7024 if (ctx.ofproto->has_mirrors) {
7025 add_mirror_actions(&ctx, &orig_flow);
7027 fix_sflow_action(&ctx);
7030 ofpbuf_uninit(&ctx.stack);
7033 /* Translates the 'ofpacts_len' bytes of "struct ofpact"s starting at 'ofpacts'
7034 * into datapath actions, using 'ctx', and discards the datapath actions. */
7036 xlate_actions_for_side_effects(struct xlate_in *xin)
7038 struct xlate_out xout;
7040 xlate_actions(xin, &xout);
7041 xlate_out_uninit(&xout);
7045 xlate_report(struct xlate_ctx *ctx, const char *s)
7047 if (ctx->xin->report_hook) {
7048 ctx->xin->report_hook(ctx, s);
7052 /* OFPP_NORMAL implementation. */
7054 static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *);
7056 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
7057 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle',
7058 * the bundle on which the packet was received, returns the VLAN to which the
7061 * Both 'vid' and the return value are in the range 0...4095. */
7063 input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid)
7065 switch (in_bundle->vlan_mode) {
7066 case PORT_VLAN_ACCESS:
7067 return in_bundle->vlan;
7070 case PORT_VLAN_TRUNK:
7073 case PORT_VLAN_NATIVE_UNTAGGED:
7074 case PORT_VLAN_NATIVE_TAGGED:
7075 return vid ? vid : in_bundle->vlan;
7082 /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'.
7083 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
7086 * 'vid' should be the VID obtained from the 802.1Q header that was received as
7087 * part of a packet (specify 0 if there was no 802.1Q header), in the range
7090 input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn)
7092 /* Allow any VID on the OFPP_NONE port. */
7093 if (in_bundle == &ofpp_none_bundle) {
7097 switch (in_bundle->vlan_mode) {
7098 case PORT_VLAN_ACCESS:
7101 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7102 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
7103 "packet received on port %s configured as VLAN "
7104 "%"PRIu16" access port",
7105 in_bundle->ofproto->up.name, vid,
7106 in_bundle->name, in_bundle->vlan);
7112 case PORT_VLAN_NATIVE_UNTAGGED:
7113 case PORT_VLAN_NATIVE_TAGGED:
7115 /* Port must always carry its native VLAN. */
7119 case PORT_VLAN_TRUNK:
7120 if (!ofbundle_includes_vlan(in_bundle, vid)) {
7122 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7123 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet "
7124 "received on port %s not configured for trunking "
7126 in_bundle->ofproto->up.name, vid,
7127 in_bundle->name, vid);
7139 /* Given 'vlan', the VLAN that a packet belongs to, and
7140 * 'out_bundle', a bundle on which the packet is to be output, returns the VID
7141 * that should be included in the 802.1Q header. (If the return value is 0,
7142 * then the 802.1Q header should only be included in the packet if there is a
7145 * Both 'vlan' and the return value are in the range 0...4095. */
7147 output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan)
7149 switch (out_bundle->vlan_mode) {
7150 case PORT_VLAN_ACCESS:
7153 case PORT_VLAN_TRUNK:
7154 case PORT_VLAN_NATIVE_TAGGED:
7157 case PORT_VLAN_NATIVE_UNTAGGED:
7158 return vlan == out_bundle->vlan ? 0 : vlan;
7166 output_normal(struct xlate_ctx *ctx, const struct ofbundle *out_bundle,
7169 struct ofport_dpif *port;
7171 ovs_be16 tci, old_tci;
7173 vid = output_vlan_to_vid(out_bundle, vlan);
7174 if (!out_bundle->bond) {
7175 port = ofbundle_get_a_port(out_bundle);
7177 port = bond_choose_output_slave(out_bundle->bond, &ctx->xin->flow,
7178 vid, &ctx->xout->tags);
7180 /* No slaves enabled, so drop packet. */
7185 old_tci = ctx->xin->flow.vlan_tci;
7187 if (tci || out_bundle->use_priority_tags) {
7188 tci |= ctx->xin->flow.vlan_tci & htons(VLAN_PCP_MASK);
7190 tci |= htons(VLAN_CFI);
7193 ctx->xin->flow.vlan_tci = tci;
7195 compose_output_action(ctx, port->up.ofp_port);
7196 ctx->xin->flow.vlan_tci = old_tci;
7200 mirror_mask_ffs(mirror_mask_t mask)
7202 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
7207 ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
7209 return (bundle->vlan_mode != PORT_VLAN_ACCESS
7210 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
7214 ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan)
7216 return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan);
7219 /* Returns an arbitrary interface within 'bundle'. */
7220 static struct ofport_dpif *
7221 ofbundle_get_a_port(const struct ofbundle *bundle)
7223 return CONTAINER_OF(list_front(&bundle->ports),
7224 struct ofport_dpif, bundle_node);
7228 vlan_is_mirrored(const struct ofmirror *m, int vlan)
7230 return !m->vlans || bitmap_is_set(m->vlans, vlan);
7234 add_mirror_actions(struct xlate_ctx *ctx, const struct flow *orig_flow)
7236 struct ofproto_dpif *ofproto = ctx->ofproto;
7237 mirror_mask_t mirrors;
7238 struct ofbundle *in_bundle;
7241 const struct nlattr *a;
7244 in_bundle = lookup_input_bundle(ctx->ofproto, orig_flow->in_port,
7245 ctx->xin->packet != NULL, NULL);
7249 mirrors = in_bundle->src_mirrors;
7251 /* Drop frames on bundles reserved for mirroring. */
7252 if (in_bundle->mirror_out) {
7253 if (ctx->xin->packet != NULL) {
7254 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7255 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
7256 "%s, which is reserved exclusively for mirroring",
7257 ctx->ofproto->up.name, in_bundle->name);
7263 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
7264 if (!input_vid_is_valid(vid, in_bundle, ctx->xin->packet != NULL)) {
7267 vlan = input_vid_to_vlan(in_bundle, vid);
7269 /* Look at the output ports to check for destination selections. */
7271 NL_ATTR_FOR_EACH (a, left, ctx->xout->odp_actions.data,
7272 ctx->xout->odp_actions.size) {
7273 enum ovs_action_attr type = nl_attr_type(a);
7274 struct ofport_dpif *ofport;
7276 if (type != OVS_ACTION_ATTR_OUTPUT) {
7280 ofport = get_odp_port(ofproto, nl_attr_get_u32(a));
7281 if (ofport && ofport->bundle) {
7282 mirrors |= ofport->bundle->dst_mirrors;
7290 /* Restore the original packet before adding the mirror actions. */
7291 ctx->xin->flow = *orig_flow;
7296 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
7298 if (!vlan_is_mirrored(m, vlan)) {
7299 mirrors = zero_rightmost_1bit(mirrors);
7303 mirrors &= ~m->dup_mirrors;
7304 ctx->xout->mirrors |= m->dup_mirrors;
7306 output_normal(ctx, m->out, vlan);
7307 } else if (vlan != m->out_vlan
7308 && !eth_addr_is_reserved(orig_flow->dl_dst)) {
7309 struct ofbundle *bundle;
7311 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
7312 if (ofbundle_includes_vlan(bundle, m->out_vlan)
7313 && !bundle->mirror_out) {
7314 output_normal(ctx, bundle, m->out_vlan);
7322 update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors,
7323 uint64_t packets, uint64_t bytes)
7329 for (; mirrors; mirrors = zero_rightmost_1bit(mirrors)) {
7332 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
7335 /* In normal circumstances 'm' will not be NULL. However,
7336 * if mirrors are reconfigured, we can temporarily get out
7337 * of sync in facet_revalidate(). We could "correct" the
7338 * mirror list before reaching here, but doing that would
7339 * not properly account the traffic stats we've currently
7340 * accumulated for previous mirror configuration. */
7344 m->packet_count += packets;
7345 m->byte_count += bytes;
7349 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
7350 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
7351 * indicate this; newer upstream kernels use gratuitous ARP requests. */
7353 is_gratuitous_arp(const struct flow *flow)
7355 return (flow->dl_type == htons(ETH_TYPE_ARP)
7356 && eth_addr_is_broadcast(flow->dl_dst)
7357 && (flow->nw_proto == ARP_OP_REPLY
7358 || (flow->nw_proto == ARP_OP_REQUEST
7359 && flow->nw_src == flow->nw_dst)));
7363 update_learning_table(struct ofproto_dpif *ofproto,
7364 const struct flow *flow, int vlan,
7365 struct ofbundle *in_bundle)
7367 struct mac_entry *mac;
7369 /* Don't learn the OFPP_NONE port. */
7370 if (in_bundle == &ofpp_none_bundle) {
7374 if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
7378 mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan);
7379 if (is_gratuitous_arp(flow)) {
7380 /* We don't want to learn from gratuitous ARP packets that are
7381 * reflected back over bond slaves so we lock the learning table. */
7382 if (!in_bundle->bond) {
7383 mac_entry_set_grat_arp_lock(mac);
7384 } else if (mac_entry_is_grat_arp_locked(mac)) {
7389 if (mac_entry_is_new(mac) || mac->port.p != in_bundle) {
7390 /* The log messages here could actually be useful in debugging,
7391 * so keep the rate limit relatively high. */
7392 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
7393 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
7394 "on port %s in VLAN %d",
7395 ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src),
7396 in_bundle->name, vlan);
7398 mac->port.p = in_bundle;
7399 tag_set_add(&ofproto->backer->revalidate_set,
7400 mac_learning_changed(ofproto->ml, mac));
7404 static struct ofbundle *
7405 lookup_input_bundle(const struct ofproto_dpif *ofproto, uint16_t in_port,
7406 bool warn, struct ofport_dpif **in_ofportp)
7408 struct ofport_dpif *ofport;
7410 /* Find the port and bundle for the received packet. */
7411 ofport = get_ofp_port(ofproto, in_port);
7413 *in_ofportp = ofport;
7415 if (ofport && ofport->bundle) {
7416 return ofport->bundle;
7419 /* Special-case OFPP_NONE, which a controller may use as the ingress
7420 * port for traffic that it is sourcing. */
7421 if (in_port == OFPP_NONE) {
7422 return &ofpp_none_bundle;
7425 /* Odd. A few possible reasons here:
7427 * - We deleted a port but there are still a few packets queued up
7430 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
7431 * we don't know about.
7433 * - The ofproto client didn't configure the port as part of a bundle.
7434 * This is particularly likely to happen if a packet was received on the
7435 * port after it was created, but before the client had a chance to
7436 * configure its bundle.
7439 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7441 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
7442 "port %"PRIu16, ofproto->up.name, in_port);
7447 /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or
7448 * dropped. Returns true if they may be forwarded, false if they should be
7451 * 'in_port' must be the ofport_dpif that corresponds to flow->in_port.
7452 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
7454 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
7455 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
7456 * checked by input_vid_is_valid().
7458 * May also add tags to '*tags', although the current implementation only does
7459 * so in one special case.
7462 is_admissible(struct xlate_ctx *ctx, struct ofport_dpif *in_port,
7465 struct ofproto_dpif *ofproto = ctx->ofproto;
7466 struct flow *flow = &ctx->xin->flow;
7467 struct ofbundle *in_bundle = in_port->bundle;
7469 /* Drop frames for reserved multicast addresses
7470 * only if forward_bpdu option is absent. */
7471 if (!ofproto->up.forward_bpdu && eth_addr_is_reserved(flow->dl_dst)) {
7472 xlate_report(ctx, "packet has reserved destination MAC, dropping");
7476 if (in_bundle->bond) {
7477 struct mac_entry *mac;
7479 switch (bond_check_admissibility(in_bundle->bond, in_port,
7480 flow->dl_dst, &ctx->xout->tags)) {
7485 xlate_report(ctx, "bonding refused admissibility, dropping");
7488 case BV_DROP_IF_MOVED:
7489 mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL);
7490 if (mac && mac->port.p != in_bundle &&
7491 (!is_gratuitous_arp(flow)
7492 || mac_entry_is_grat_arp_locked(mac))) {
7493 xlate_report(ctx, "SLB bond thinks this packet looped back, "
7505 xlate_normal(struct xlate_ctx *ctx)
7507 struct ofport_dpif *in_port;
7508 struct ofbundle *in_bundle;
7509 struct mac_entry *mac;
7513 ctx->xout->has_normal = true;
7515 in_bundle = lookup_input_bundle(ctx->ofproto, ctx->xin->flow.in_port,
7516 ctx->xin->packet != NULL, &in_port);
7518 xlate_report(ctx, "no input bundle, dropping");
7522 /* Drop malformed frames. */
7523 if (ctx->xin->flow.dl_type == htons(ETH_TYPE_VLAN) &&
7524 !(ctx->xin->flow.vlan_tci & htons(VLAN_CFI))) {
7525 if (ctx->xin->packet != NULL) {
7526 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7527 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
7528 "VLAN tag received on port %s",
7529 ctx->ofproto->up.name, in_bundle->name);
7531 xlate_report(ctx, "partial VLAN tag, dropping");
7535 /* Drop frames on bundles reserved for mirroring. */
7536 if (in_bundle->mirror_out) {
7537 if (ctx->xin->packet != NULL) {
7538 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7539 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
7540 "%s, which is reserved exclusively for mirroring",
7541 ctx->ofproto->up.name, in_bundle->name);
7543 xlate_report(ctx, "input port is mirror output port, dropping");
7548 vid = vlan_tci_to_vid(ctx->xin->flow.vlan_tci);
7549 if (!input_vid_is_valid(vid, in_bundle, ctx->xin->packet != NULL)) {
7550 xlate_report(ctx, "disallowed VLAN VID for this input port, dropping");
7553 vlan = input_vid_to_vlan(in_bundle, vid);
7555 /* Check other admissibility requirements. */
7556 if (in_port && !is_admissible(ctx, in_port, vlan)) {
7560 /* Learn source MAC. */
7561 if (ctx->xin->may_learn) {
7562 update_learning_table(ctx->ofproto, &ctx->xin->flow, vlan, in_bundle);
7565 /* Determine output bundle. */
7566 mac = mac_learning_lookup(ctx->ofproto->ml, ctx->xin->flow.dl_dst, vlan,
7569 if (mac->port.p != in_bundle) {
7570 xlate_report(ctx, "forwarding to learned port");
7571 output_normal(ctx, mac->port.p, vlan);
7573 xlate_report(ctx, "learned port is input port, dropping");
7576 struct ofbundle *bundle;
7578 xlate_report(ctx, "no learned MAC for destination, flooding");
7579 HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
7580 if (bundle != in_bundle
7581 && ofbundle_includes_vlan(bundle, vlan)
7582 && bundle->floodable
7583 && !bundle->mirror_out) {
7584 output_normal(ctx, bundle, vlan);
7587 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
7591 /* Optimized flow revalidation.
7593 * It's a difficult problem, in general, to tell which facets need to have
7594 * their actions recalculated whenever the OpenFlow flow table changes. We
7595 * don't try to solve that general problem: for most kinds of OpenFlow flow
7596 * table changes, we recalculate the actions for every facet. This is
7597 * relatively expensive, but it's good enough if the OpenFlow flow table
7598 * doesn't change very often.
7600 * However, we can expect one particular kind of OpenFlow flow table change to
7601 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
7602 * of CPU on revalidating every facet whenever MAC learning modifies the flow
7603 * table, we add a special case that applies to flow tables in which every rule
7604 * has the same form (that is, the same wildcards), except that the table is
7605 * also allowed to have a single "catch-all" flow that matches all packets. We
7606 * optimize this case by tagging all of the facets that resubmit into the table
7607 * and invalidating the same tag whenever a flow changes in that table. The
7608 * end result is that we revalidate just the facets that need it (and sometimes
7609 * a few more, but not all of the facets or even all of the facets that
7610 * resubmit to the table modified by MAC learning). */
7612 /* Calculates the tag to use for 'flow' and mask 'mask' when it is inserted
7613 * into an OpenFlow table with the given 'basis'. */
7615 rule_calculate_tag(const struct flow *flow, const struct minimask *mask,
7618 if (minimask_is_catchall(mask)) {
7621 uint32_t hash = flow_hash_in_minimask(flow, mask, secret);
7622 return tag_create_deterministic(hash);
7626 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
7627 * taggability of that table.
7629 * This function must be called after *each* change to a flow table. If you
7630 * skip calling it on some changes then the pointer comparisons at the end can
7631 * be invalid if you get unlucky. For example, if a flow removal causes a
7632 * cls_table to be destroyed and then a flow insertion causes a cls_table with
7633 * different wildcards to be created with the same address, then this function
7634 * will incorrectly skip revalidation. */
7636 table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
7638 struct table_dpif *table = &ofproto->tables[table_id];
7639 const struct oftable *oftable = &ofproto->up.tables[table_id];
7640 struct cls_table *catchall, *other;
7641 struct cls_table *t;
7643 catchall = other = NULL;
7645 switch (hmap_count(&oftable->cls.tables)) {
7647 /* We could tag this OpenFlow table but it would make the logic a
7648 * little harder and it's a corner case that doesn't seem worth it
7654 HMAP_FOR_EACH (t, hmap_node, &oftable->cls.tables) {
7655 if (cls_table_is_catchall(t)) {
7657 } else if (!other) {
7660 /* Indicate that we can't tag this by setting both tables to
7661 * NULL. (We know that 'catchall' is already NULL.) */
7668 /* Can't tag this table. */
7672 if (table->catchall_table != catchall || table->other_table != other) {
7673 table->catchall_table = catchall;
7674 table->other_table = other;
7675 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
7679 /* Given 'rule' that has changed in some way (either it is a rule being
7680 * inserted, a rule being deleted, or a rule whose actions are being
7681 * modified), marks facets for revalidation to ensure that packets will be
7682 * forwarded correctly according to the new state of the flow table.
7684 * This function must be called after *each* change to a flow table. See
7685 * the comment on table_update_taggable() for more information. */
7687 rule_invalidate(const struct rule_dpif *rule)
7689 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
7691 table_update_taggable(ofproto, rule->up.table_id);
7693 if (!ofproto->backer->need_revalidate) {
7694 struct table_dpif *table = &ofproto->tables[rule->up.table_id];
7696 if (table->other_table && rule->tag) {
7697 tag_set_add(&ofproto->backer->revalidate_set, rule->tag);
7699 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
7705 set_frag_handling(struct ofproto *ofproto_,
7706 enum ofp_config_flags frag_handling)
7708 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7709 if (frag_handling != OFPC_FRAG_REASM) {
7710 ofproto->backer->need_revalidate = REV_RECONFIGURE;
7718 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
7719 const struct flow *flow,
7720 const struct ofpact *ofpacts, size_t ofpacts_len)
7722 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7723 struct initial_vals initial_vals;
7724 struct odputil_keybuf keybuf;
7725 struct dpif_flow_stats stats;
7726 struct xlate_out xout;
7727 struct xlate_in xin;
7731 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
7732 odp_flow_key_from_flow(&key, flow,
7733 ofp_port_to_odp_port(ofproto, flow->in_port));
7735 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
7737 initial_vals.vlan_tci = flow->vlan_tci;
7738 xlate_in_init(&xin, ofproto, flow, &initial_vals, NULL, stats.tcp_flags,
7740 xin.resubmit_stats = &stats;
7741 xin.ofpacts_len = ofpacts_len;
7742 xin.ofpacts = ofpacts;
7744 xlate_actions(&xin, &xout);
7745 dpif_execute(ofproto->backer->dpif, key.data, key.size,
7746 xout.odp_actions.data, xout.odp_actions.size, packet);
7747 xlate_out_uninit(&xout);
7755 set_netflow(struct ofproto *ofproto_,
7756 const struct netflow_options *netflow_options)
7758 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7760 if (netflow_options) {
7761 if (!ofproto->netflow) {
7762 ofproto->netflow = netflow_create();
7764 return netflow_set_options(ofproto->netflow, netflow_options);
7766 netflow_destroy(ofproto->netflow);
7767 ofproto->netflow = NULL;
7773 get_netflow_ids(const struct ofproto *ofproto_,
7774 uint8_t *engine_type, uint8_t *engine_id)
7776 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7778 dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
7782 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
7784 if (!facet_is_controller_flow(facet) &&
7785 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
7786 struct subfacet *subfacet;
7787 struct ofexpired expired;
7789 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
7790 if (subfacet->path == SF_FAST_PATH) {
7791 struct dpif_flow_stats stats;
7793 subfacet_install(subfacet, &facet->xout.odp_actions, &stats);
7794 subfacet_update_stats(subfacet, &stats);
7798 expired.flow = facet->flow;
7799 expired.packet_count = facet->packet_count;
7800 expired.byte_count = facet->byte_count;
7801 expired.used = facet->used;
7802 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
7807 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
7809 struct facet *facet;
7811 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
7812 send_active_timeout(ofproto, facet);
7816 static struct ofproto_dpif *
7817 ofproto_dpif_lookup(const char *name)
7819 struct ofproto_dpif *ofproto;
7821 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
7822 hash_string(name, 0), &all_ofproto_dpifs) {
7823 if (!strcmp(ofproto->up.name, name)) {
7831 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
7832 const char *argv[], void *aux OVS_UNUSED)
7834 struct ofproto_dpif *ofproto;
7837 ofproto = ofproto_dpif_lookup(argv[1]);
7839 unixctl_command_reply_error(conn, "no such bridge");
7842 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
7844 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
7845 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
7849 unixctl_command_reply(conn, "table successfully flushed");
7853 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
7854 const char *argv[], void *aux OVS_UNUSED)
7856 struct ds ds = DS_EMPTY_INITIALIZER;
7857 const struct ofproto_dpif *ofproto;
7858 const struct mac_entry *e;
7860 ofproto = ofproto_dpif_lookup(argv[1]);
7862 unixctl_command_reply_error(conn, "no such bridge");
7866 ds_put_cstr(&ds, " port VLAN MAC Age\n");
7867 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
7868 struct ofbundle *bundle = e->port.p;
7869 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
7870 ofbundle_get_a_port(bundle)->odp_port,
7871 e->vlan, ETH_ADDR_ARGS(e->mac),
7872 mac_entry_age(ofproto->ml, e));
7874 unixctl_command_reply(conn, ds_cstr(&ds));
7879 struct xlate_out xout;
7880 struct xlate_in xin;
7886 trace_format_rule(struct ds *result, uint8_t table_id, int level,
7887 const struct rule_dpif *rule)
7889 ds_put_char_multiple(result, '\t', level);
7891 ds_put_cstr(result, "No match\n");
7895 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
7896 table_id, ntohll(rule->up.flow_cookie));
7897 cls_rule_format(&rule->up.cr, result);
7898 ds_put_char(result, '\n');
7900 ds_put_char_multiple(result, '\t', level);
7901 ds_put_cstr(result, "OpenFlow ");
7902 ofpacts_format(rule->up.ofpacts, rule->up.ofpacts_len, result);
7903 ds_put_char(result, '\n');
7907 trace_format_flow(struct ds *result, int level, const char *title,
7908 struct trace_ctx *trace)
7910 ds_put_char_multiple(result, '\t', level);
7911 ds_put_format(result, "%s: ", title);
7912 if (flow_equal(&trace->xin.flow, &trace->flow)) {
7913 ds_put_cstr(result, "unchanged");
7915 flow_format(result, &trace->xin.flow);
7916 trace->flow = trace->xin.flow;
7918 ds_put_char(result, '\n');
7922 trace_format_regs(struct ds *result, int level, const char *title,
7923 struct trace_ctx *trace)
7927 ds_put_char_multiple(result, '\t', level);
7928 ds_put_format(result, "%s:", title);
7929 for (i = 0; i < FLOW_N_REGS; i++) {
7930 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
7932 ds_put_char(result, '\n');
7936 trace_format_odp(struct ds *result, int level, const char *title,
7937 struct trace_ctx *trace)
7939 struct ofpbuf *odp_actions = &trace->xout.odp_actions;
7941 ds_put_char_multiple(result, '\t', level);
7942 ds_put_format(result, "%s: ", title);
7943 format_odp_actions(result, odp_actions->data, odp_actions->size);
7944 ds_put_char(result, '\n');
7948 trace_resubmit(struct xlate_ctx *ctx, struct rule_dpif *rule)
7950 struct trace_ctx *trace = CONTAINER_OF(ctx->xin, struct trace_ctx, xin);
7951 struct ds *result = trace->result;
7953 ds_put_char(result, '\n');
7954 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
7955 trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace);
7956 trace_format_odp(result, ctx->recurse + 1, "Resubmitted odp", trace);
7957 trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule);
7961 trace_report(struct xlate_ctx *ctx, const char *s)
7963 struct trace_ctx *trace = CONTAINER_OF(ctx->xin, struct trace_ctx, xin);
7964 struct ds *result = trace->result;
7966 ds_put_char_multiple(result, '\t', ctx->recurse);
7967 ds_put_cstr(result, s);
7968 ds_put_char(result, '\n');
7972 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
7973 void *aux OVS_UNUSED)
7975 const struct dpif_backer *backer;
7976 struct ofproto_dpif *ofproto;
7977 struct ofpbuf odp_key;
7978 struct ofpbuf *packet;
7979 struct initial_vals initial_vals;
7987 ofpbuf_init(&odp_key, 0);
7989 /* Handle "-generate" or a hex string as the last argument. */
7990 if (!strcmp(argv[argc - 1], "-generate")) {
7991 packet = ofpbuf_new(0);
7994 const char *error = eth_from_hex(argv[argc - 1], &packet);
7997 } else if (argc == 4) {
7998 /* The 3-argument form must end in "-generate' or a hex string. */
7999 unixctl_command_reply_error(conn, error);
8004 /* Parse the flow and determine whether a datapath or
8005 * bridge is specified. If function odp_flow_key_from_string()
8006 * returns 0, the flow is a odp_flow. If function
8007 * parse_ofp_exact_flow() returns 0, the flow is a br_flow. */
8008 if (!odp_flow_key_from_string(argv[argc - 1], NULL, &odp_key)) {
8009 /* If the odp_flow is the second argument,
8010 * the datapath name is the first argument. */
8012 const char *dp_type;
8013 if (!strncmp(argv[1], "ovs-", 4)) {
8014 dp_type = argv[1] + 4;
8018 backer = shash_find_data(&all_dpif_backers, dp_type);
8020 unixctl_command_reply_error(conn, "Cannot find datapath "
8025 /* No datapath name specified, so there should be only one
8027 struct shash_node *node;
8028 if (shash_count(&all_dpif_backers) != 1) {
8029 unixctl_command_reply_error(conn, "Must specify datapath "
8030 "name, there is more than one type of datapath");
8033 node = shash_first(&all_dpif_backers);
8034 backer = node->data;
8037 /* Extract the ofproto_dpif object from the ofproto_receive()
8039 if (ofproto_receive(backer, NULL, odp_key.data,
8040 odp_key.size, &flow, NULL, &ofproto, NULL,
8042 unixctl_command_reply_error(conn, "Invalid datapath flow");
8045 ds_put_format(&result, "Bridge: %s\n", ofproto->up.name);
8046 } else if (!parse_ofp_exact_flow(&flow, argv[argc - 1])) {
8048 unixctl_command_reply_error(conn, "Must specify bridge name");
8052 ofproto = ofproto_dpif_lookup(argv[1]);
8054 unixctl_command_reply_error(conn, "Unknown bridge name");
8057 initial_vals.vlan_tci = flow.vlan_tci;
8059 unixctl_command_reply_error(conn, "Bad flow syntax");
8063 /* Generate a packet, if requested. */
8065 if (!packet->size) {
8066 flow_compose(packet, &flow);
8068 ds_put_cstr(&result, "Packet: ");
8069 s = ofp_packet_to_string(packet->data, packet->size);
8070 ds_put_cstr(&result, s);
8073 /* Use the metadata from the flow and the packet argument
8074 * to reconstruct the flow. */
8075 flow_extract(packet, flow.skb_priority, flow.skb_mark, NULL,
8076 flow.in_port, &flow);
8077 initial_vals.vlan_tci = flow.vlan_tci;
8081 ofproto_trace(ofproto, &flow, packet, &initial_vals, &result);
8082 unixctl_command_reply(conn, ds_cstr(&result));
8085 ds_destroy(&result);
8086 ofpbuf_delete(packet);
8087 ofpbuf_uninit(&odp_key);
8091 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
8092 const struct ofpbuf *packet,
8093 const struct initial_vals *initial_vals, struct ds *ds)
8095 struct rule_dpif *rule;
8097 ds_put_cstr(ds, "Flow: ");
8098 flow_format(ds, flow);
8099 ds_put_char(ds, '\n');
8101 rule = rule_dpif_lookup(ofproto, flow);
8103 trace_format_rule(ds, 0, 0, rule);
8104 if (rule == ofproto->miss_rule) {
8105 ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
8106 } else if (rule == ofproto->no_packet_in_rule) {
8107 ds_put_cstr(ds, "\nNo match, packets dropped because "
8108 "OFPPC_NO_PACKET_IN is set on in_port.\n");
8109 } else if (rule == ofproto->drop_frags_rule) {
8110 ds_put_cstr(ds, "\nPackets dropped because they are IP fragments "
8111 "and the fragment handling mode is \"drop\".\n");
8115 uint64_t odp_actions_stub[1024 / 8];
8116 struct ofpbuf odp_actions;
8118 struct trace_ctx trace;
8121 tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
8124 ofpbuf_use_stub(&odp_actions,
8125 odp_actions_stub, sizeof odp_actions_stub);
8126 xlate_in_init(&trace.xin, ofproto, flow, initial_vals, rule, tcp_flags,
8128 trace.xin.resubmit_hook = trace_resubmit;
8129 trace.xin.report_hook = trace_report;
8130 xlate_actions(&trace.xin, &trace.xout);
8132 ds_put_char(ds, '\n');
8133 trace_format_flow(ds, 0, "Final flow", &trace);
8134 ds_put_cstr(ds, "Datapath actions: ");
8135 format_odp_actions(ds, trace.xout.odp_actions.data,
8136 trace.xout.odp_actions.size);
8138 if (trace.xout.slow) {
8139 ds_put_cstr(ds, "\nThis flow is handled by the userspace "
8140 "slow path because it:");
8141 switch (trace.xout.slow) {
8143 ds_put_cstr(ds, "\n\t- Consists of CFM packets.");
8146 ds_put_cstr(ds, "\n\t- Consists of LACP packets.");
8149 ds_put_cstr(ds, "\n\t- Consists of STP packets.");
8152 ds_put_cstr(ds, "\n\t- Consists of BFD packets.");
8154 case SLOW_CONTROLLER:
8155 ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages "
8156 "to the OpenFlow controller.");
8163 xlate_out_uninit(&trace.xout);
8168 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
8169 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
8172 unixctl_command_reply(conn, NULL);
8176 ofproto_dpif_unclog(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);
8183 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
8184 * 'reply' describing the results. */
8186 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
8188 struct facet *facet;
8192 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
8193 if (!facet_check_consistency(facet)) {
8198 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
8202 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
8203 ofproto->up.name, errors);
8205 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
8210 ofproto_dpif_self_check(struct unixctl_conn *conn,
8211 int argc, const char *argv[], void *aux OVS_UNUSED)
8213 struct ds reply = DS_EMPTY_INITIALIZER;
8214 struct ofproto_dpif *ofproto;
8217 ofproto = ofproto_dpif_lookup(argv[1]);
8219 unixctl_command_reply_error(conn, "Unknown ofproto (use "
8220 "ofproto/list for help)");
8223 ofproto_dpif_self_check__(ofproto, &reply);
8225 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
8226 ofproto_dpif_self_check__(ofproto, &reply);
8230 unixctl_command_reply(conn, ds_cstr(&reply));
8234 /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list
8235 * of the 'ofproto_shash' nodes. It is the responsibility of the caller
8236 * to destroy 'ofproto_shash' and free the returned value. */
8237 static const struct shash_node **
8238 get_ofprotos(struct shash *ofproto_shash)
8240 const struct ofproto_dpif *ofproto;
8242 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
8243 char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
8244 shash_add_nocopy(ofproto_shash, name, ofproto);
8247 return shash_sort(ofproto_shash);
8251 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
8252 const char *argv[] OVS_UNUSED,
8253 void *aux OVS_UNUSED)
8255 struct ds ds = DS_EMPTY_INITIALIZER;
8256 struct shash ofproto_shash;
8257 const struct shash_node **sorted_ofprotos;
8260 shash_init(&ofproto_shash);
8261 sorted_ofprotos = get_ofprotos(&ofproto_shash);
8262 for (i = 0; i < shash_count(&ofproto_shash); i++) {
8263 const struct shash_node *node = sorted_ofprotos[i];
8264 ds_put_format(&ds, "%s\n", node->name);
8267 shash_destroy(&ofproto_shash);
8268 free(sorted_ofprotos);
8270 unixctl_command_reply(conn, ds_cstr(&ds));
8275 show_dp_rates(struct ds *ds, const char *heading,
8276 const struct avg_subfacet_rates *rates)
8278 ds_put_format(ds, "%s add rate: %5.3f/min, del rate: %5.3f/min\n",
8279 heading, rates->add_rate, rates->del_rate);
8283 dpif_show_backer(const struct dpif_backer *backer, struct ds *ds)
8285 size_t n_hit, n_missed, n_subfacets, i;
8286 const struct shash_node **ofprotos;
8287 struct ofproto_dpif *ofproto;
8288 struct shash ofproto_shash;
8289 long long int minutes;
8291 n_hit = n_missed = n_subfacets = 0;
8292 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
8293 if (ofproto->backer == backer) {
8294 n_subfacets += hmap_count(&ofproto->subfacets);
8295 n_missed += ofproto->n_missed;
8296 n_hit += ofproto->n_hit;
8300 ds_put_format(ds, "%s: hit:%"PRIu64" missed:%"PRIu64"\n",
8301 dpif_name(backer->dpif), n_hit, n_missed);
8302 ds_put_format(ds, "\tflows: cur: %zu, avg: %u, max: %u,"
8303 " life span: %lldms\n", n_subfacets,
8304 backer->avg_n_subfacet, backer->max_n_subfacet,
8305 backer->avg_subfacet_life);
8307 minutes = (time_msec() - backer->created) / (1000 * 60);
8308 if (minutes >= 60) {
8309 show_dp_rates(ds, "\thourly avg:", &backer->hourly);
8311 if (minutes >= 60 * 24) {
8312 show_dp_rates(ds, "\tdaily avg:", &backer->daily);
8314 show_dp_rates(ds, "\toverall avg:", &backer->lifetime);
8316 shash_init(&ofproto_shash);
8317 ofprotos = get_ofprotos(&ofproto_shash);
8318 for (i = 0; i < shash_count(&ofproto_shash); i++) {
8319 struct ofproto_dpif *ofproto = ofprotos[i]->data;
8320 const struct shash_node **ports;
8323 if (ofproto->backer != backer) {
8327 ds_put_format(ds, "\t%s: hit:%"PRIu64" missed:%"PRIu64"\n",
8328 ofproto->up.name, ofproto->n_hit, ofproto->n_missed);
8330 ports = shash_sort(&ofproto->up.port_by_name);
8331 for (j = 0; j < shash_count(&ofproto->up.port_by_name); j++) {
8332 const struct shash_node *node = ports[j];
8333 struct ofport *ofport = node->data;
8337 ds_put_format(ds, "\t\t%s %u/", netdev_get_name(ofport->netdev),
8340 odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
8341 if (odp_port != OVSP_NONE) {
8342 ds_put_format(ds, "%"PRIu32":", odp_port);
8344 ds_put_cstr(ds, "none:");
8347 ds_put_format(ds, " (%s", netdev_get_type(ofport->netdev));
8350 if (!netdev_get_config(ofport->netdev, &config)) {
8351 const struct smap_node **nodes;
8354 nodes = smap_sort(&config);
8355 for (i = 0; i < smap_count(&config); i++) {
8356 const struct smap_node *node = nodes[i];
8357 ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
8358 node->key, node->value);
8362 smap_destroy(&config);
8364 ds_put_char(ds, ')');
8365 ds_put_char(ds, '\n');
8369 shash_destroy(&ofproto_shash);
8374 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
8375 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
8377 struct ds ds = DS_EMPTY_INITIALIZER;
8378 const struct shash_node **backers;
8381 backers = shash_sort(&all_dpif_backers);
8382 for (i = 0; i < shash_count(&all_dpif_backers); i++) {
8383 dpif_show_backer(backers[i]->data, &ds);
8387 unixctl_command_reply(conn, ds_cstr(&ds));
8392 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
8393 int argc OVS_UNUSED, const char *argv[],
8394 void *aux OVS_UNUSED)
8396 struct ds ds = DS_EMPTY_INITIALIZER;
8397 const struct ofproto_dpif *ofproto;
8398 struct subfacet *subfacet;
8400 ofproto = ofproto_dpif_lookup(argv[1]);
8402 unixctl_command_reply_error(conn, "no such bridge");
8406 update_stats(ofproto->backer);
8408 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
8409 struct facet *facet = subfacet->facet;
8411 odp_flow_key_format(subfacet->key, subfacet->key_len, &ds);
8413 ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:",
8414 subfacet->dp_packet_count, subfacet->dp_byte_count);
8415 if (subfacet->used) {
8416 ds_put_format(&ds, "%.3fs",
8417 (time_msec() - subfacet->used) / 1000.0);
8419 ds_put_format(&ds, "never");
8421 if (subfacet->facet->tcp_flags) {
8422 ds_put_cstr(&ds, ", flags:");
8423 packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags);
8426 ds_put_cstr(&ds, ", actions:");
8427 if (facet->xout.slow) {
8428 uint64_t slow_path_stub[128 / 8];
8429 const struct nlattr *actions;
8432 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
8433 slow_path_stub, sizeof slow_path_stub,
8434 &actions, &actions_len);
8435 format_odp_actions(&ds, actions, actions_len);
8437 format_odp_actions(&ds, facet->xout.odp_actions.data,
8438 facet->xout.odp_actions.size);
8440 ds_put_char(&ds, '\n');
8443 unixctl_command_reply(conn, ds_cstr(&ds));
8448 ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn,
8449 int argc OVS_UNUSED, const char *argv[],
8450 void *aux OVS_UNUSED)
8452 struct ds ds = DS_EMPTY_INITIALIZER;
8453 struct ofproto_dpif *ofproto;
8455 ofproto = ofproto_dpif_lookup(argv[1]);
8457 unixctl_command_reply_error(conn, "no such bridge");
8461 flush(&ofproto->up);
8463 unixctl_command_reply(conn, ds_cstr(&ds));
8468 ofproto_dpif_unixctl_init(void)
8470 static bool registered;
8476 unixctl_command_register(
8478 "[dp_name]|bridge odp_flow|br_flow [-generate|packet]",
8479 1, 3, ofproto_unixctl_trace, NULL);
8480 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
8481 ofproto_unixctl_fdb_flush, NULL);
8482 unixctl_command_register("fdb/show", "bridge", 1, 1,
8483 ofproto_unixctl_fdb_show, NULL);
8484 unixctl_command_register("ofproto/clog", "", 0, 0,
8485 ofproto_dpif_clog, NULL);
8486 unixctl_command_register("ofproto/unclog", "", 0, 0,
8487 ofproto_dpif_unclog, NULL);
8488 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
8489 ofproto_dpif_self_check, NULL);
8490 unixctl_command_register("dpif/dump-dps", "", 0, 0,
8491 ofproto_unixctl_dpif_dump_dps, NULL);
8492 unixctl_command_register("dpif/show", "", 0, 0, ofproto_unixctl_dpif_show,
8494 unixctl_command_register("dpif/dump-flows", "bridge", 1, 1,
8495 ofproto_unixctl_dpif_dump_flows, NULL);
8496 unixctl_command_register("dpif/del-flows", "bridge", 1, 1,
8497 ofproto_unixctl_dpif_del_flows, NULL);
8500 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
8502 * This is deprecated. It is only for compatibility with broken device drivers
8503 * in old versions of Linux that do not properly support VLANs when VLAN
8504 * devices are not used. When broken device drivers are no longer in
8505 * widespread use, we will delete these interfaces. */
8508 set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid)
8510 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
8511 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
8513 if (realdev_ofp_port == ofport->realdev_ofp_port
8514 && vid == ofport->vlandev_vid) {
8518 ofproto->backer->need_revalidate = REV_RECONFIGURE;
8520 if (ofport->realdev_ofp_port) {
8523 if (realdev_ofp_port && ofport->bundle) {
8524 /* vlandevs are enslaved to their realdevs, so they are not allowed to
8525 * themselves be part of a bundle. */
8526 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
8529 ofport->realdev_ofp_port = realdev_ofp_port;
8530 ofport->vlandev_vid = vid;
8532 if (realdev_ofp_port) {
8533 vsp_add(ofport, realdev_ofp_port, vid);
8540 hash_realdev_vid(uint16_t realdev_ofp_port, int vid)
8542 return hash_2words(realdev_ofp_port, vid);
8545 /* Returns the OFP port number of the Linux VLAN device that corresponds to
8546 * 'vlan_tci' on the network device with port number 'realdev_ofp_port' in
8547 * 'struct ofport_dpif'. For example, given 'realdev_ofp_port' of eth0 and
8548 * 'vlan_tci' 9, it would return the port number of eth0.9.
8550 * Unless VLAN splinters are enabled for port 'realdev_ofp_port', this
8551 * function just returns its 'realdev_ofp_port' argument. */
8553 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
8554 uint16_t realdev_ofp_port, ovs_be16 vlan_tci)
8556 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
8557 int vid = vlan_tci_to_vid(vlan_tci);
8558 const struct vlan_splinter *vsp;
8560 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
8561 hash_realdev_vid(realdev_ofp_port, vid),
8562 &ofproto->realdev_vid_map) {
8563 if (vsp->realdev_ofp_port == realdev_ofp_port
8564 && vsp->vid == vid) {
8565 return vsp->vlandev_ofp_port;
8569 return realdev_ofp_port;
8572 static struct vlan_splinter *
8573 vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port)
8575 struct vlan_splinter *vsp;
8577 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, hash_int(vlandev_ofp_port, 0),
8578 &ofproto->vlandev_map) {
8579 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
8587 /* Returns the OpenFlow port number of the "real" device underlying the Linux
8588 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
8589 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
8590 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
8591 * eth0 and store 9 in '*vid'.
8593 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
8594 * VLAN device. Unless VLAN splinters are enabled, this is what this function
8597 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
8598 uint16_t vlandev_ofp_port, int *vid)
8600 if (!hmap_is_empty(&ofproto->vlandev_map)) {
8601 const struct vlan_splinter *vsp;
8603 vsp = vlandev_find(ofproto, vlandev_ofp_port);
8608 return vsp->realdev_ofp_port;
8614 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
8615 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
8616 * 'flow->in_port' to the "real" device backing the VLAN device, sets
8617 * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is
8618 * always the case unless VLAN splinters are enabled), returns false without
8619 * making any changes. */
8621 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
8626 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port, &vid);
8631 /* Cause the flow to be processed as if it came in on the real device with
8632 * the VLAN device's VLAN ID. */
8633 flow->in_port = realdev;
8634 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
8639 vsp_remove(struct ofport_dpif *port)
8641 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
8642 struct vlan_splinter *vsp;
8644 vsp = vlandev_find(ofproto, port->up.ofp_port);
8646 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
8647 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
8650 port->realdev_ofp_port = 0;
8652 VLOG_ERR("missing vlan device record");
8657 vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid)
8659 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
8661 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
8662 && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid))
8663 == realdev_ofp_port)) {
8664 struct vlan_splinter *vsp;
8666 vsp = xmalloc(sizeof *vsp);
8667 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
8668 hash_int(port->up.ofp_port, 0));
8669 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
8670 hash_realdev_vid(realdev_ofp_port, vid));
8671 vsp->realdev_ofp_port = realdev_ofp_port;
8672 vsp->vlandev_ofp_port = port->up.ofp_port;
8675 port->realdev_ofp_port = realdev_ofp_port;
8677 VLOG_ERR("duplicate vlan device record");
8682 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
8684 const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
8685 return ofport ? ofport->odp_port : OVSP_NONE;
8688 static struct ofport_dpif *
8689 odp_port_to_ofport(const struct dpif_backer *backer, uint32_t odp_port)
8691 struct ofport_dpif *port;
8693 HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node,
8694 hash_int(odp_port, 0),
8695 &backer->odp_to_ofport_map) {
8696 if (port->odp_port == odp_port) {
8705 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
8707 struct ofport_dpif *port;
8709 port = odp_port_to_ofport(ofproto->backer, odp_port);
8710 if (port && &ofproto->up == port->up.ofproto) {
8711 return port->up.ofp_port;
8717 /* Compute exponentially weighted moving average, adding 'new' as the newest,
8718 * most heavily weighted element. 'base' designates the rate of decay: after
8719 * 'base' further updates, 'new''s weight in the EWMA decays to about 1/e
8722 exp_mavg(double *avg, int base, double new)
8724 *avg = (*avg * (base - 1) + new) / base;
8728 update_moving_averages(struct dpif_backer *backer)
8730 const int min_ms = 60 * 1000; /* milliseconds in one minute. */
8731 long long int minutes = (time_msec() - backer->created) / min_ms;
8734 backer->lifetime.add_rate = (double) backer->total_subfacet_add_count
8736 backer->lifetime.del_rate = (double) backer->total_subfacet_del_count
8739 backer->lifetime.add_rate = 0.0;
8740 backer->lifetime.del_rate = 0.0;
8743 /* Update hourly averages on the minute boundaries. */
8744 if (time_msec() - backer->last_minute >= min_ms) {
8745 exp_mavg(&backer->hourly.add_rate, 60, backer->subfacet_add_count);
8746 exp_mavg(&backer->hourly.del_rate, 60, backer->subfacet_del_count);
8748 /* Update daily averages on the hour boundaries. */
8749 if ((backer->last_minute - backer->created) / min_ms % 60 == 59) {
8750 exp_mavg(&backer->daily.add_rate, 24, backer->hourly.add_rate);
8751 exp_mavg(&backer->daily.del_rate, 24, backer->hourly.del_rate);
8754 backer->total_subfacet_add_count += backer->subfacet_add_count;
8755 backer->total_subfacet_del_count += backer->subfacet_del_count;
8756 backer->subfacet_add_count = 0;
8757 backer->subfacet_del_count = 0;
8758 backer->last_minute += min_ms;
8762 const struct ofproto_class ofproto_dpif_class = {
8797 port_is_lacp_current,
8798 NULL, /* rule_choose_table */
8805 rule_modify_actions,
8819 get_stp_port_status,
8826 is_mirror_output_bundle,
8827 forward_bpdu_changed,
8828 set_mac_table_config,