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 /* All datapaths of a given type share a single dpif backer instance. */
644 struct timer next_expiration;
645 struct hmap odp_to_ofport_map; /* ODP port to ofport mapping. */
647 struct simap tnl_backers; /* Set of dpif ports backing tunnels. */
649 /* Facet revalidation flags applying to facets which use this backer. */
650 enum revalidate_reason need_revalidate; /* Revalidate every facet. */
651 struct tag_set revalidate_set; /* Revalidate only matching facets. */
653 struct hmap drop_keys; /* Set of dropped odp keys. */
654 bool recv_set_enable; /* Enables or disables receiving packets. */
657 /* All existing ofproto_backer instances, indexed by ofproto->up.type. */
658 static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers);
660 static void drop_key_clear(struct dpif_backer *);
661 static struct ofport_dpif *
662 odp_port_to_ofport(const struct dpif_backer *, uint32_t odp_port);
664 struct avg_subfacet_rates {
665 double add_rate; /* Moving average of new flows created per minute. */
666 double del_rate; /* Moving average of flows deleted per minute. */
668 static void show_dp_rates(struct ds *ds, const char *heading,
669 const struct avg_subfacet_rates *rates);
670 static void exp_mavg(double *avg, int base, double new);
672 struct ofproto_dpif {
673 struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
675 struct dpif_backer *backer;
677 /* Special OpenFlow rules. */
678 struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */
679 struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */
680 struct rule_dpif *drop_frags_rule; /* Used in OFPC_FRAG_DROP mode. */
683 struct netflow *netflow;
684 struct dpif_sflow *sflow;
685 struct dpif_ipfix *ipfix;
686 struct hmap bundles; /* Contains "struct ofbundle"s. */
687 struct mac_learning *ml;
688 struct ofmirror *mirrors[MAX_MIRRORS];
690 bool has_bonded_bundles;
694 struct hmap subfacets;
695 struct governor *governor;
696 long long int consistency_rl;
699 struct table_dpif tables[N_TABLES];
701 /* Support for debugging async flow mods. */
702 struct list completions;
704 bool has_bundle_action; /* True when the first bundle action appears. */
705 struct netdev_stats stats; /* To account packets generated and consumed in
710 long long int stp_last_tick;
712 /* VLAN splinters. */
713 struct hmap realdev_vid_map; /* (realdev,vid) -> vlandev. */
714 struct hmap vlandev_map; /* vlandev -> (realdev,vid). */
717 struct sset ports; /* Set of standard port names. */
718 struct sset ghost_ports; /* Ports with no datapath port. */
719 struct sset port_poll_set; /* Queued names for port_poll() reply. */
720 int port_poll_errno; /* Last errno for port_poll() reply. */
722 /* Per ofproto's dpif stats. */
726 /* Subfacet statistics.
728 * These keep track of the total number of subfacets added and deleted and
729 * flow life span. They are useful for computing the flow rates stats
730 * exposed via "ovs-appctl dpif/show". The goal is to learn about
731 * traffic patterns in ways that we can use later to improve Open vSwitch
732 * performance in new situations. */
733 long long int created; /* Time when it is created. */
734 unsigned int max_n_subfacet; /* Maximum number of flows */
735 unsigned int avg_n_subfacet; /* Average number of flows. */
736 long long int avg_subfacet_life_span;
738 /* The average number of subfacets... */
739 struct avg_subfacet_rates hourly; /* ...over the last hour. */
740 struct avg_subfacet_rates daily; /* ...over the last day. */
741 long long int last_minute; /* Last time 'hourly' was updated. */
743 /* Number of subfacets added or deleted since 'last_minute'. */
744 unsigned int subfacet_add_count;
745 unsigned int subfacet_del_count;
747 /* Number of subfacets added or deleted from 'created' to 'last_minute.' */
748 unsigned long long int total_subfacet_add_count;
749 unsigned long long int total_subfacet_del_count;
751 static void update_moving_averages(struct ofproto_dpif *ofproto);
753 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
754 * for debugging the asynchronous flow_mod implementation.) */
757 /* All existing ofproto_dpif instances, indexed by ->up.name. */
758 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
760 static void ofproto_dpif_unixctl_init(void);
762 static struct ofproto_dpif *
763 ofproto_dpif_cast(const struct ofproto *ofproto)
765 ovs_assert(ofproto->ofproto_class == &ofproto_dpif_class);
766 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
769 static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *,
771 static struct ofport_dpif *get_odp_port(const struct ofproto_dpif *,
773 static void ofproto_trace(struct ofproto_dpif *, const struct flow *,
774 const struct ofpbuf *,
775 const struct initial_vals *, struct ds *);
777 /* Packet processing. */
778 static void update_learning_table(struct ofproto_dpif *,
779 const struct flow *, int vlan,
782 #define FLOW_MISS_MAX_BATCH 50
783 static int handle_upcalls(struct dpif_backer *, unsigned int max_batch);
785 /* Flow expiration. */
786 static int expire(struct dpif_backer *);
789 static void send_netflow_active_timeouts(struct ofproto_dpif *);
792 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
793 static size_t compose_sflow_action(const struct ofproto_dpif *,
794 struct ofpbuf *odp_actions,
795 const struct flow *, uint32_t odp_port);
796 static void compose_ipfix_action(const struct ofproto_dpif *,
797 struct ofpbuf *odp_actions,
798 const struct flow *);
799 static void add_mirror_actions(struct xlate_ctx *ctx,
800 const struct flow *flow);
801 /* Global variables. */
802 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
804 /* Initial mappings of port to bridge mappings. */
805 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
807 /* Factory functions. */
810 init(const struct shash *iface_hints)
812 struct shash_node *node;
814 /* Make a local copy, since we don't own 'iface_hints' elements. */
815 SHASH_FOR_EACH(node, iface_hints) {
816 const struct iface_hint *orig_hint = node->data;
817 struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
819 new_hint->br_name = xstrdup(orig_hint->br_name);
820 new_hint->br_type = xstrdup(orig_hint->br_type);
821 new_hint->ofp_port = orig_hint->ofp_port;
823 shash_add(&init_ofp_ports, node->name, new_hint);
828 enumerate_types(struct sset *types)
830 dp_enumerate_types(types);
834 enumerate_names(const char *type, struct sset *names)
836 struct ofproto_dpif *ofproto;
839 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
840 if (strcmp(type, ofproto->up.type)) {
843 sset_add(names, ofproto->up.name);
850 del(const char *type, const char *name)
855 error = dpif_open(name, type, &dpif);
857 error = dpif_delete(dpif);
864 port_open_type(const char *datapath_type, const char *port_type)
866 return dpif_port_open_type(datapath_type, port_type);
869 /* Type functions. */
871 static struct ofproto_dpif *
872 lookup_ofproto_dpif_by_port_name(const char *name)
874 struct ofproto_dpif *ofproto;
876 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
877 if (sset_contains(&ofproto->ports, name)) {
886 type_run(const char *type)
888 static long long int push_timer = LLONG_MIN;
889 struct dpif_backer *backer;
893 backer = shash_find_data(&all_dpif_backers, type);
895 /* This is not necessarily a problem, since backers are only
896 * created on demand. */
900 dpif_run(backer->dpif);
902 /* The most natural place to push facet statistics is when they're pulled
903 * from the datapath. However, when there are many flows in the datapath,
904 * this expensive operation can occur so frequently, that it reduces our
905 * ability to quickly set up flows. To reduce the cost, we push statistics
907 if (time_msec() > push_timer) {
908 push_timer = time_msec() + 2000;
912 /* If vswitchd started with other_config:flow_restore_wait set as "true",
913 * and the configuration has now changed to "false", enable receiving
914 * packets from the datapath. */
915 if (!backer->recv_set_enable && !ofproto_get_flow_restore_wait()) {
916 backer->recv_set_enable = true;
918 error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
920 VLOG_ERR("Failed to enable receiving packets in dpif.");
923 dpif_flow_flush(backer->dpif);
924 backer->need_revalidate = REV_RECONFIGURE;
927 if (backer->need_revalidate
928 || !tag_set_is_empty(&backer->revalidate_set)) {
929 struct tag_set revalidate_set = backer->revalidate_set;
930 bool need_revalidate = backer->need_revalidate;
931 struct ofproto_dpif *ofproto;
932 struct simap_node *node;
933 struct simap tmp_backers;
935 /* Handle tunnel garbage collection. */
936 simap_init(&tmp_backers);
937 simap_swap(&backer->tnl_backers, &tmp_backers);
939 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
940 struct ofport_dpif *iter;
942 if (backer != ofproto->backer) {
946 HMAP_FOR_EACH (iter, up.hmap_node, &ofproto->up.ports) {
947 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
950 if (!iter->tnl_port) {
954 dp_port = netdev_vport_get_dpif_port(iter->up.netdev,
955 namebuf, sizeof namebuf);
956 node = simap_find(&tmp_backers, dp_port);
958 simap_put(&backer->tnl_backers, dp_port, node->data);
959 simap_delete(&tmp_backers, node);
960 node = simap_find(&backer->tnl_backers, dp_port);
962 node = simap_find(&backer->tnl_backers, dp_port);
964 uint32_t odp_port = UINT32_MAX;
966 if (!dpif_port_add(backer->dpif, iter->up.netdev,
968 simap_put(&backer->tnl_backers, dp_port, odp_port);
969 node = simap_find(&backer->tnl_backers, dp_port);
974 iter->odp_port = node ? node->data : OVSP_NONE;
975 if (tnl_port_reconfigure(&iter->up, iter->odp_port,
977 backer->need_revalidate = REV_RECONFIGURE;
982 SIMAP_FOR_EACH (node, &tmp_backers) {
983 dpif_port_del(backer->dpif, node->data);
985 simap_destroy(&tmp_backers);
987 switch (backer->need_revalidate) {
988 case REV_RECONFIGURE: COVERAGE_INC(rev_reconfigure); break;
989 case REV_STP: COVERAGE_INC(rev_stp); break;
990 case REV_PORT_TOGGLED: COVERAGE_INC(rev_port_toggled); break;
991 case REV_FLOW_TABLE: COVERAGE_INC(rev_flow_table); break;
992 case REV_INCONSISTENCY: COVERAGE_INC(rev_inconsistency); break;
995 if (backer->need_revalidate) {
996 /* Clear the drop_keys in case we should now be accepting some
997 * formerly dropped flows. */
998 drop_key_clear(backer);
1001 /* Clear the revalidation flags. */
1002 tag_set_init(&backer->revalidate_set);
1003 backer->need_revalidate = 0;
1005 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1006 struct facet *facet, *next;
1008 if (ofproto->backer != backer) {
1012 HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &ofproto->facets) {
1014 || tag_set_intersects(&revalidate_set, facet->xout.tags)) {
1015 facet_revalidate(facet);
1022 if (!backer->recv_set_enable) {
1023 /* Wake up before a max of 1000ms. */
1024 timer_set_duration(&backer->next_expiration, 1000);
1025 } else if (timer_expired(&backer->next_expiration)) {
1026 int delay = expire(backer);
1027 timer_set_duration(&backer->next_expiration, delay);
1030 /* Check for port changes in the dpif. */
1031 while ((error = dpif_port_poll(backer->dpif, &devname)) == 0) {
1032 struct ofproto_dpif *ofproto;
1033 struct dpif_port port;
1035 /* Don't report on the datapath's device. */
1036 if (!strcmp(devname, dpif_base_name(backer->dpif))) {
1040 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
1041 &all_ofproto_dpifs) {
1042 if (simap_contains(&ofproto->backer->tnl_backers, devname)) {
1047 ofproto = lookup_ofproto_dpif_by_port_name(devname);
1048 if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
1049 /* The port was removed. If we know the datapath,
1050 * report it through poll_set(). If we don't, it may be
1051 * notifying us of a removal we initiated, so ignore it.
1052 * If there's a pending ENOBUFS, let it stand, since
1053 * everything will be reevaluated. */
1054 if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
1055 sset_add(&ofproto->port_poll_set, devname);
1056 ofproto->port_poll_errno = 0;
1058 } else if (!ofproto) {
1059 /* The port was added, but we don't know with which
1060 * ofproto we should associate it. Delete it. */
1061 dpif_port_del(backer->dpif, port.port_no);
1063 dpif_port_destroy(&port);
1069 if (error != EAGAIN) {
1070 struct ofproto_dpif *ofproto;
1072 /* There was some sort of error, so propagate it to all
1073 * ofprotos that use this backer. */
1074 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
1075 &all_ofproto_dpifs) {
1076 if (ofproto->backer == backer) {
1077 sset_clear(&ofproto->port_poll_set);
1078 ofproto->port_poll_errno = error;
1087 dpif_backer_run_fast(struct dpif_backer *backer, int max_batch)
1091 /* If recv_set_enable is false, we should not handle upcalls. */
1092 if (!backer->recv_set_enable) {
1096 /* Handle one or more batches of upcalls, until there's nothing left to do
1097 * or until we do a fixed total amount of work.
1099 * We do work in batches because it can be much cheaper to set up a number
1100 * of flows and fire off their patches all at once. We do multiple batches
1101 * because in some cases handling a packet can cause another packet to be
1102 * queued almost immediately as part of the return flow. Both
1103 * optimizations can make major improvements on some benchmarks and
1104 * presumably for real traffic as well. */
1106 while (work < max_batch) {
1107 int retval = handle_upcalls(backer, max_batch - work);
1118 type_run_fast(const char *type)
1120 struct dpif_backer *backer;
1122 backer = shash_find_data(&all_dpif_backers, type);
1124 /* This is not necessarily a problem, since backers are only
1125 * created on demand. */
1129 return dpif_backer_run_fast(backer, FLOW_MISS_MAX_BATCH);
1135 static long long int port_rl = LLONG_MIN;
1136 static unsigned int backer_rl = 0;
1138 if (time_msec() >= port_rl) {
1139 struct ofproto_dpif *ofproto;
1140 struct ofport_dpif *ofport;
1142 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1144 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1145 port_run_fast(ofport);
1148 port_rl = time_msec() + 200;
1151 /* XXX: We have to be careful not to do too much work in this function. If
1152 * we call dpif_backer_run_fast() too often, or with too large a batch,
1153 * performance improves signifcantly, but at a cost. It's possible for the
1154 * number of flows in the datapath to increase without bound, and for poll
1155 * loops to take 10s of seconds. The correct solution to this problem,
1156 * long term, is to separate flow miss handling into it's own thread so it
1157 * isn't affected by revalidations, and expirations. Until then, this is
1158 * the best we can do. */
1159 if (++backer_rl >= 10) {
1160 struct shash_node *node;
1163 SHASH_FOR_EACH (node, &all_dpif_backers) {
1164 dpif_backer_run_fast(node->data, 1);
1170 type_wait(const char *type)
1172 struct dpif_backer *backer;
1174 backer = shash_find_data(&all_dpif_backers, type);
1176 /* This is not necessarily a problem, since backers are only
1177 * created on demand. */
1181 timer_wait(&backer->next_expiration);
1184 /* Basic life-cycle. */
1186 static int add_internal_flows(struct ofproto_dpif *);
1188 static struct ofproto *
1191 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
1192 return &ofproto->up;
1196 dealloc(struct ofproto *ofproto_)
1198 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1203 close_dpif_backer(struct dpif_backer *backer)
1205 struct shash_node *node;
1207 ovs_assert(backer->refcount > 0);
1209 if (--backer->refcount) {
1213 drop_key_clear(backer);
1214 hmap_destroy(&backer->drop_keys);
1216 simap_destroy(&backer->tnl_backers);
1217 hmap_destroy(&backer->odp_to_ofport_map);
1218 node = shash_find(&all_dpif_backers, backer->type);
1220 shash_delete(&all_dpif_backers, node);
1221 dpif_close(backer->dpif);
1226 /* Datapath port slated for removal from datapath. */
1227 struct odp_garbage {
1228 struct list list_node;
1233 open_dpif_backer(const char *type, struct dpif_backer **backerp)
1235 struct dpif_backer *backer;
1236 struct dpif_port_dump port_dump;
1237 struct dpif_port port;
1238 struct shash_node *node;
1239 struct list garbage_list;
1240 struct odp_garbage *garbage, *next;
1246 backer = shash_find_data(&all_dpif_backers, type);
1253 backer_name = xasprintf("ovs-%s", type);
1255 /* Remove any existing datapaths, since we assume we're the only
1256 * userspace controlling the datapath. */
1258 dp_enumerate_names(type, &names);
1259 SSET_FOR_EACH(name, &names) {
1260 struct dpif *old_dpif;
1262 /* Don't remove our backer if it exists. */
1263 if (!strcmp(name, backer_name)) {
1267 if (dpif_open(name, type, &old_dpif)) {
1268 VLOG_WARN("couldn't open old datapath %s to remove it", name);
1270 dpif_delete(old_dpif);
1271 dpif_close(old_dpif);
1274 sset_destroy(&names);
1276 backer = xmalloc(sizeof *backer);
1278 error = dpif_create_and_open(backer_name, type, &backer->dpif);
1281 VLOG_ERR("failed to open datapath of type %s: %s", type,
1287 backer->type = xstrdup(type);
1288 backer->refcount = 1;
1289 hmap_init(&backer->odp_to_ofport_map);
1290 hmap_init(&backer->drop_keys);
1291 timer_set_duration(&backer->next_expiration, 1000);
1292 backer->need_revalidate = 0;
1293 simap_init(&backer->tnl_backers);
1294 tag_set_init(&backer->revalidate_set);
1295 backer->recv_set_enable = !ofproto_get_flow_restore_wait();
1298 if (backer->recv_set_enable) {
1299 dpif_flow_flush(backer->dpif);
1302 /* Loop through the ports already on the datapath and remove any
1303 * that we don't need anymore. */
1304 list_init(&garbage_list);
1305 dpif_port_dump_start(&port_dump, backer->dpif);
1306 while (dpif_port_dump_next(&port_dump, &port)) {
1307 node = shash_find(&init_ofp_ports, port.name);
1308 if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
1309 garbage = xmalloc(sizeof *garbage);
1310 garbage->odp_port = port.port_no;
1311 list_push_front(&garbage_list, &garbage->list_node);
1314 dpif_port_dump_done(&port_dump);
1316 LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) {
1317 dpif_port_del(backer->dpif, garbage->odp_port);
1318 list_remove(&garbage->list_node);
1322 shash_add(&all_dpif_backers, type, backer);
1324 error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
1326 VLOG_ERR("failed to listen on datapath of type %s: %s",
1327 type, strerror(error));
1328 close_dpif_backer(backer);
1336 construct(struct ofproto *ofproto_)
1338 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1339 struct shash_node *node, *next;
1344 error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1349 max_ports = dpif_get_max_ports(ofproto->backer->dpif);
1350 ofproto_init_max_ports(ofproto_, MIN(max_ports, OFPP_MAX));
1352 ofproto->netflow = NULL;
1353 ofproto->sflow = NULL;
1354 ofproto->ipfix = NULL;
1355 ofproto->stp = NULL;
1356 hmap_init(&ofproto->bundles);
1357 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1358 for (i = 0; i < MAX_MIRRORS; i++) {
1359 ofproto->mirrors[i] = NULL;
1361 ofproto->has_bonded_bundles = false;
1363 hmap_init(&ofproto->facets);
1364 hmap_init(&ofproto->subfacets);
1365 ofproto->governor = NULL;
1366 ofproto->consistency_rl = LLONG_MIN;
1368 for (i = 0; i < N_TABLES; i++) {
1369 struct table_dpif *table = &ofproto->tables[i];
1371 table->catchall_table = NULL;
1372 table->other_table = NULL;
1373 table->basis = random_uint32();
1376 list_init(&ofproto->completions);
1378 ofproto_dpif_unixctl_init();
1380 ofproto->has_mirrors = false;
1381 ofproto->has_bundle_action = false;
1383 hmap_init(&ofproto->vlandev_map);
1384 hmap_init(&ofproto->realdev_vid_map);
1386 sset_init(&ofproto->ports);
1387 sset_init(&ofproto->ghost_ports);
1388 sset_init(&ofproto->port_poll_set);
1389 ofproto->port_poll_errno = 0;
1391 SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1392 struct iface_hint *iface_hint = node->data;
1394 if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1395 /* Check if the datapath already has this port. */
1396 if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1397 sset_add(&ofproto->ports, node->name);
1400 free(iface_hint->br_name);
1401 free(iface_hint->br_type);
1403 shash_delete(&init_ofp_ports, node);
1407 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1408 hash_string(ofproto->up.name, 0));
1409 memset(&ofproto->stats, 0, sizeof ofproto->stats);
1411 ofproto_init_tables(ofproto_, N_TABLES);
1412 error = add_internal_flows(ofproto);
1413 ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1416 ofproto->n_missed = 0;
1418 ofproto->max_n_subfacet = 0;
1419 ofproto->created = time_msec();
1420 ofproto->last_minute = ofproto->created;
1421 memset(&ofproto->hourly, 0, sizeof ofproto->hourly);
1422 memset(&ofproto->daily, 0, sizeof ofproto->daily);
1423 ofproto->subfacet_add_count = 0;
1424 ofproto->subfacet_del_count = 0;
1425 ofproto->total_subfacet_add_count = 0;
1426 ofproto->total_subfacet_del_count = 0;
1427 ofproto->avg_subfacet_life_span = 0;
1428 ofproto->avg_n_subfacet = 0;
1434 add_internal_flow(struct ofproto_dpif *ofproto, int id,
1435 const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1437 struct ofputil_flow_mod fm;
1440 match_init_catchall(&fm.match);
1442 match_set_reg(&fm.match, 0, id);
1443 fm.new_cookie = htonll(0);
1444 fm.cookie = htonll(0);
1445 fm.cookie_mask = htonll(0);
1446 fm.table_id = TBL_INTERNAL;
1447 fm.command = OFPFC_ADD;
1448 fm.idle_timeout = 0;
1449 fm.hard_timeout = 0;
1453 fm.ofpacts = ofpacts->data;
1454 fm.ofpacts_len = ofpacts->size;
1456 error = ofproto_flow_mod(&ofproto->up, &fm);
1458 VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
1459 id, ofperr_to_string(error));
1463 *rulep = rule_dpif_lookup__(ofproto, &fm.match.flow, TBL_INTERNAL);
1464 ovs_assert(*rulep != NULL);
1470 add_internal_flows(struct ofproto_dpif *ofproto)
1472 struct ofpact_controller *controller;
1473 uint64_t ofpacts_stub[128 / 8];
1474 struct ofpbuf ofpacts;
1478 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1481 controller = ofpact_put_CONTROLLER(&ofpacts);
1482 controller->max_len = UINT16_MAX;
1483 controller->controller_id = 0;
1484 controller->reason = OFPR_NO_MATCH;
1485 ofpact_pad(&ofpacts);
1487 error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule);
1492 ofpbuf_clear(&ofpacts);
1493 error = add_internal_flow(ofproto, id++, &ofpacts,
1494 &ofproto->no_packet_in_rule);
1499 error = add_internal_flow(ofproto, id++, &ofpacts,
1500 &ofproto->drop_frags_rule);
1505 complete_operations(struct ofproto_dpif *ofproto)
1507 struct dpif_completion *c, *next;
1509 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
1510 ofoperation_complete(c->op, 0);
1511 list_remove(&c->list_node);
1517 destruct(struct ofproto *ofproto_)
1519 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1520 struct rule_dpif *rule, *next_rule;
1521 struct oftable *table;
1524 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1525 complete_operations(ofproto);
1527 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1528 struct cls_cursor cursor;
1530 cls_cursor_init(&cursor, &table->cls, NULL);
1531 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1532 ofproto_rule_destroy(&rule->up);
1536 for (i = 0; i < MAX_MIRRORS; i++) {
1537 mirror_destroy(ofproto->mirrors[i]);
1540 netflow_destroy(ofproto->netflow);
1541 dpif_sflow_destroy(ofproto->sflow);
1542 hmap_destroy(&ofproto->bundles);
1543 mac_learning_destroy(ofproto->ml);
1545 hmap_destroy(&ofproto->facets);
1546 hmap_destroy(&ofproto->subfacets);
1547 governor_destroy(ofproto->governor);
1549 hmap_destroy(&ofproto->vlandev_map);
1550 hmap_destroy(&ofproto->realdev_vid_map);
1552 sset_destroy(&ofproto->ports);
1553 sset_destroy(&ofproto->ghost_ports);
1554 sset_destroy(&ofproto->port_poll_set);
1556 close_dpif_backer(ofproto->backer);
1560 run_fast(struct ofproto *ofproto_)
1562 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1563 struct ofport_dpif *ofport;
1565 /* Do not perform any periodic activity required by 'ofproto' while
1566 * waiting for flow restore to complete. */
1567 if (ofproto_get_flow_restore_wait()) {
1571 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1572 port_run_fast(ofport);
1579 run(struct ofproto *ofproto_)
1581 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1582 struct ofport_dpif *ofport;
1583 struct ofbundle *bundle;
1587 complete_operations(ofproto);
1590 /* Do not perform any periodic activity below required by 'ofproto' while
1591 * waiting for flow restore to complete. */
1592 if (ofproto_get_flow_restore_wait()) {
1596 error = run_fast(ofproto_);
1601 if (ofproto->netflow) {
1602 if (netflow_run(ofproto->netflow)) {
1603 send_netflow_active_timeouts(ofproto);
1606 if (ofproto->sflow) {
1607 dpif_sflow_run(ofproto->sflow);
1610 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1613 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1618 mac_learning_run(ofproto->ml, &ofproto->backer->revalidate_set);
1620 /* Check the consistency of a random facet, to aid debugging. */
1621 if (time_msec() >= ofproto->consistency_rl
1622 && !hmap_is_empty(&ofproto->facets)
1623 && !ofproto->backer->need_revalidate) {
1624 struct facet *facet;
1626 ofproto->consistency_rl = time_msec() + 250;
1628 facet = CONTAINER_OF(hmap_random_node(&ofproto->facets),
1629 struct facet, hmap_node);
1630 if (!tag_set_intersects(&ofproto->backer->revalidate_set,
1631 facet->xout.tags)) {
1632 if (!facet_check_consistency(facet)) {
1633 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
1638 if (ofproto->governor) {
1641 governor_run(ofproto->governor);
1643 /* If the governor has shrunk to its minimum size and the number of
1644 * subfacets has dwindled, then drop the governor entirely.
1646 * For hysteresis, the number of subfacets to drop the governor is
1647 * smaller than the number needed to trigger its creation. */
1648 n_subfacets = hmap_count(&ofproto->subfacets);
1649 if (n_subfacets * 4 < ofproto->up.flow_eviction_threshold
1650 && governor_is_idle(ofproto->governor)) {
1651 governor_destroy(ofproto->governor);
1652 ofproto->governor = NULL;
1660 wait(struct ofproto *ofproto_)
1662 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1663 struct ofport_dpif *ofport;
1664 struct ofbundle *bundle;
1666 if (!clogged && !list_is_empty(&ofproto->completions)) {
1667 poll_immediate_wake();
1670 if (ofproto_get_flow_restore_wait()) {
1674 dpif_wait(ofproto->backer->dpif);
1675 dpif_recv_wait(ofproto->backer->dpif);
1676 if (ofproto->sflow) {
1677 dpif_sflow_wait(ofproto->sflow);
1679 if (!tag_set_is_empty(&ofproto->backer->revalidate_set)) {
1680 poll_immediate_wake();
1682 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1685 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1686 bundle_wait(bundle);
1688 if (ofproto->netflow) {
1689 netflow_wait(ofproto->netflow);
1691 mac_learning_wait(ofproto->ml);
1693 if (ofproto->backer->need_revalidate) {
1694 /* Shouldn't happen, but if it does just go around again. */
1695 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1696 poll_immediate_wake();
1698 if (ofproto->governor) {
1699 governor_wait(ofproto->governor);
1704 get_memory_usage(const struct ofproto *ofproto_, struct simap *usage)
1706 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1708 simap_increase(usage, "facets", hmap_count(&ofproto->facets));
1709 simap_increase(usage, "subfacets", hmap_count(&ofproto->subfacets));
1713 flush(struct ofproto *ofproto_)
1715 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1716 struct subfacet *subfacet, *next_subfacet;
1717 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
1721 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
1722 &ofproto->subfacets) {
1723 if (subfacet->path != SF_NOT_INSTALLED) {
1724 batch[n_batch++] = subfacet;
1725 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
1726 subfacet_destroy_batch(ofproto, batch, n_batch);
1730 subfacet_destroy(subfacet);
1735 subfacet_destroy_batch(ofproto, batch, n_batch);
1740 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1741 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1743 *arp_match_ip = true;
1744 *actions = (OFPUTIL_A_OUTPUT |
1745 OFPUTIL_A_SET_VLAN_VID |
1746 OFPUTIL_A_SET_VLAN_PCP |
1747 OFPUTIL_A_STRIP_VLAN |
1748 OFPUTIL_A_SET_DL_SRC |
1749 OFPUTIL_A_SET_DL_DST |
1750 OFPUTIL_A_SET_NW_SRC |
1751 OFPUTIL_A_SET_NW_DST |
1752 OFPUTIL_A_SET_NW_TOS |
1753 OFPUTIL_A_SET_TP_SRC |
1754 OFPUTIL_A_SET_TP_DST |
1759 get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots)
1761 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1762 struct dpif_dp_stats s;
1763 uint64_t n_miss, n_no_pkt_in, n_bytes, n_dropped_frags;
1766 strcpy(ots->name, "classifier");
1768 dpif_get_dp_stats(ofproto->backer->dpif, &s);
1769 rule_get_stats(&ofproto->miss_rule->up, &n_miss, &n_bytes);
1770 rule_get_stats(&ofproto->no_packet_in_rule->up, &n_no_pkt_in, &n_bytes);
1771 rule_get_stats(&ofproto->drop_frags_rule->up, &n_dropped_frags, &n_bytes);
1773 n_lookup = s.n_hit + s.n_missed - n_dropped_frags;
1774 ots->lookup_count = htonll(n_lookup);
1775 ots->matched_count = htonll(n_lookup - n_miss - n_no_pkt_in);
1778 static struct ofport *
1781 struct ofport_dpif *port = xmalloc(sizeof *port);
1786 port_dealloc(struct ofport *port_)
1788 struct ofport_dpif *port = ofport_dpif_cast(port_);
1793 port_construct(struct ofport *port_)
1795 struct ofport_dpif *port = ofport_dpif_cast(port_);
1796 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1797 const struct netdev *netdev = port->up.netdev;
1798 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1799 struct dpif_port dpif_port;
1802 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1803 port->bundle = NULL;
1806 port->tag = tag_create_random();
1807 port->may_enable = true;
1808 port->stp_port = NULL;
1809 port->stp_state = STP_DISABLED;
1810 port->tnl_port = NULL;
1811 hmap_init(&port->priorities);
1812 port->realdev_ofp_port = 0;
1813 port->vlandev_vid = 0;
1814 port->carrier_seq = netdev_get_carrier_resets(netdev);
1816 if (netdev_vport_is_patch(netdev)) {
1817 /* By bailing out here, we don't submit the port to the sFlow module
1818 * to be considered for counter polling export. This is correct
1819 * because the patch port represents an interface that sFlow considers
1820 * to be "internal" to the switch as a whole, and therefore not an
1821 * candidate for counter polling. */
1822 port->odp_port = OVSP_NONE;
1826 error = dpif_port_query_by_name(ofproto->backer->dpif,
1827 netdev_vport_get_dpif_port(netdev, namebuf,
1834 port->odp_port = dpif_port.port_no;
1836 if (netdev_get_tunnel_config(netdev)) {
1837 port->tnl_port = tnl_port_add(&port->up, port->odp_port);
1839 /* Sanity-check that a mapping doesn't already exist. This
1840 * shouldn't happen for non-tunnel ports. */
1841 if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1842 VLOG_ERR("port %s already has an OpenFlow port number",
1844 dpif_port_destroy(&dpif_port);
1848 hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1849 hash_int(port->odp_port, 0));
1851 dpif_port_destroy(&dpif_port);
1853 if (ofproto->sflow) {
1854 dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1861 port_destruct(struct ofport *port_)
1863 struct ofport_dpif *port = ofport_dpif_cast(port_);
1864 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1865 const char *devname = netdev_get_name(port->up.netdev);
1866 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1867 const char *dp_port_name;
1869 dp_port_name = netdev_vport_get_dpif_port(port->up.netdev, namebuf,
1871 if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1872 /* The underlying device is still there, so delete it. This
1873 * happens when the ofproto is being destroyed, since the caller
1874 * assumes that removal of attached ports will happen as part of
1876 if (!port->tnl_port) {
1877 dpif_port_del(ofproto->backer->dpif, port->odp_port);
1879 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1882 if (port->odp_port != OVSP_NONE && !port->tnl_port) {
1883 hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1886 tnl_port_del(port->tnl_port);
1887 sset_find_and_delete(&ofproto->ports, devname);
1888 sset_find_and_delete(&ofproto->ghost_ports, devname);
1889 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1890 bundle_remove(port_);
1891 set_cfm(port_, NULL);
1892 set_bfd(port_, NULL);
1893 if (ofproto->sflow) {
1894 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1897 ofport_clear_priorities(port);
1898 hmap_destroy(&port->priorities);
1902 port_modified(struct ofport *port_)
1904 struct ofport_dpif *port = ofport_dpif_cast(port_);
1906 if (port->bundle && port->bundle->bond) {
1907 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1911 cfm_set_netdev(port->cfm, port->up.netdev);
1916 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1918 struct ofport_dpif *port = ofport_dpif_cast(port_);
1919 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1920 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1922 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1923 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1924 OFPUTIL_PC_NO_PACKET_IN)) {
1925 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1927 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1928 bundle_update(port->bundle);
1934 set_sflow(struct ofproto *ofproto_,
1935 const struct ofproto_sflow_options *sflow_options)
1937 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1938 struct dpif_sflow *ds = ofproto->sflow;
1940 if (sflow_options) {
1942 struct ofport_dpif *ofport;
1944 ds = ofproto->sflow = dpif_sflow_create();
1945 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1946 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1948 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1950 dpif_sflow_set_options(ds, sflow_options);
1953 dpif_sflow_destroy(ds);
1954 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1955 ofproto->sflow = NULL;
1963 struct ofproto *ofproto_,
1964 const struct ofproto_ipfix_bridge_exporter_options *bridge_exporter_options,
1965 const struct ofproto_ipfix_flow_exporter_options *flow_exporters_options,
1966 size_t n_flow_exporters_options)
1968 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1969 struct dpif_ipfix *di = ofproto->ipfix;
1971 if (bridge_exporter_options || flow_exporters_options) {
1973 di = ofproto->ipfix = dpif_ipfix_create();
1975 dpif_ipfix_set_options(
1976 di, bridge_exporter_options, flow_exporters_options,
1977 n_flow_exporters_options);
1980 dpif_ipfix_destroy(di);
1981 ofproto->ipfix = NULL;
1988 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1990 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1997 struct ofproto_dpif *ofproto;
1999 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2000 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2001 ofport->cfm = cfm_create(ofport->up.netdev);
2004 if (cfm_configure(ofport->cfm, s)) {
2010 cfm_destroy(ofport->cfm);
2016 get_cfm_status(const struct ofport *ofport_,
2017 struct ofproto_cfm_status *status)
2019 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2022 status->faults = cfm_get_fault(ofport->cfm);
2023 status->remote_opstate = cfm_get_opup(ofport->cfm);
2024 status->health = cfm_get_health(ofport->cfm);
2025 cfm_get_remote_mpids(ofport->cfm, &status->rmps, &status->n_rmps);
2033 set_bfd(struct ofport *ofport_, const struct smap *cfg)
2035 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
2036 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2040 ofport->bfd = bfd_configure(old, netdev_get_name(ofport->up.netdev), cfg);
2041 if (ofport->bfd != old) {
2042 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2049 get_bfd_status(struct ofport *ofport_, struct smap *smap)
2051 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2054 bfd_get_status(ofport->bfd, smap);
2061 /* Spanning Tree. */
2064 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
2066 struct ofproto_dpif *ofproto = ofproto_;
2067 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
2068 struct ofport_dpif *ofport;
2070 ofport = stp_port_get_aux(sp);
2072 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
2073 ofproto->up.name, port_num);
2075 struct eth_header *eth = pkt->l2;
2077 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
2078 if (eth_addr_is_zero(eth->eth_src)) {
2079 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
2080 "with unknown MAC", ofproto->up.name, port_num);
2082 send_packet(ofport, pkt);
2088 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
2090 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
2092 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2094 /* Only revalidate flows if the configuration changed. */
2095 if (!s != !ofproto->stp) {
2096 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2100 if (!ofproto->stp) {
2101 ofproto->stp = stp_create(ofproto_->name, s->system_id,
2102 send_bpdu_cb, ofproto);
2103 ofproto->stp_last_tick = time_msec();
2106 stp_set_bridge_id(ofproto->stp, s->system_id);
2107 stp_set_bridge_priority(ofproto->stp, s->priority);
2108 stp_set_hello_time(ofproto->stp, s->hello_time);
2109 stp_set_max_age(ofproto->stp, s->max_age);
2110 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
2112 struct ofport *ofport;
2114 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
2115 set_stp_port(ofport, NULL);
2118 stp_destroy(ofproto->stp);
2119 ofproto->stp = NULL;
2126 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
2128 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2132 s->bridge_id = stp_get_bridge_id(ofproto->stp);
2133 s->designated_root = stp_get_designated_root(ofproto->stp);
2134 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
2143 update_stp_port_state(struct ofport_dpif *ofport)
2145 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2146 enum stp_state state;
2148 /* Figure out new state. */
2149 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
2153 if (ofport->stp_state != state) {
2154 enum ofputil_port_state of_state;
2157 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
2158 netdev_get_name(ofport->up.netdev),
2159 stp_state_name(ofport->stp_state),
2160 stp_state_name(state));
2161 if (stp_learn_in_state(ofport->stp_state)
2162 != stp_learn_in_state(state)) {
2163 /* xxx Learning action flows should also be flushed. */
2164 mac_learning_flush(ofproto->ml,
2165 &ofproto->backer->revalidate_set);
2167 fwd_change = stp_forward_in_state(ofport->stp_state)
2168 != stp_forward_in_state(state);
2170 ofproto->backer->need_revalidate = REV_STP;
2171 ofport->stp_state = state;
2172 ofport->stp_state_entered = time_msec();
2174 if (fwd_change && ofport->bundle) {
2175 bundle_update(ofport->bundle);
2178 /* Update the STP state bits in the OpenFlow port description. */
2179 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
2180 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
2181 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
2182 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
2183 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
2185 ofproto_port_set_state(&ofport->up, of_state);
2189 /* Configures STP on 'ofport_' using the settings defined in 's'. The
2190 * caller is responsible for assigning STP port numbers and ensuring
2191 * there are no duplicates. */
2193 set_stp_port(struct ofport *ofport_,
2194 const struct ofproto_port_stp_settings *s)
2196 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2197 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2198 struct stp_port *sp = ofport->stp_port;
2200 if (!s || !s->enable) {
2202 ofport->stp_port = NULL;
2203 stp_port_disable(sp);
2204 update_stp_port_state(ofport);
2207 } else if (sp && stp_port_no(sp) != s->port_num
2208 && ofport == stp_port_get_aux(sp)) {
2209 /* The port-id changed, so disable the old one if it's not
2210 * already in use by another port. */
2211 stp_port_disable(sp);
2214 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
2215 stp_port_enable(sp);
2217 stp_port_set_aux(sp, ofport);
2218 stp_port_set_priority(sp, s->priority);
2219 stp_port_set_path_cost(sp, s->path_cost);
2221 update_stp_port_state(ofport);
2227 get_stp_port_status(struct ofport *ofport_,
2228 struct ofproto_port_stp_status *s)
2230 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2231 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2232 struct stp_port *sp = ofport->stp_port;
2234 if (!ofproto->stp || !sp) {
2240 s->port_id = stp_port_get_id(sp);
2241 s->state = stp_port_get_state(sp);
2242 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
2243 s->role = stp_port_get_role(sp);
2244 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
2250 stp_run(struct ofproto_dpif *ofproto)
2253 long long int now = time_msec();
2254 long long int elapsed = now - ofproto->stp_last_tick;
2255 struct stp_port *sp;
2258 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
2259 ofproto->stp_last_tick = now;
2261 while (stp_get_changed_port(ofproto->stp, &sp)) {
2262 struct ofport_dpif *ofport = stp_port_get_aux(sp);
2265 update_stp_port_state(ofport);
2269 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
2270 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
2276 stp_wait(struct ofproto_dpif *ofproto)
2279 poll_timer_wait(1000);
2283 /* Returns true if STP should process 'flow'. */
2285 stp_should_process_flow(const struct flow *flow)
2287 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
2291 stp_process_packet(const struct ofport_dpif *ofport,
2292 const struct ofpbuf *packet)
2294 struct ofpbuf payload = *packet;
2295 struct eth_header *eth = payload.data;
2296 struct stp_port *sp = ofport->stp_port;
2298 /* Sink packets on ports that have STP disabled when the bridge has
2300 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
2304 /* Trim off padding on payload. */
2305 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
2306 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
2309 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
2310 stp_received_bpdu(sp, payload.data, payload.size);
2314 static struct priority_to_dscp *
2315 get_priority(const struct ofport_dpif *ofport, uint32_t priority)
2317 struct priority_to_dscp *pdscp;
2320 hash = hash_int(priority, 0);
2321 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) {
2322 if (pdscp->priority == priority) {
2330 ofport_clear_priorities(struct ofport_dpif *ofport)
2332 struct priority_to_dscp *pdscp, *next;
2334 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) {
2335 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
2341 set_queues(struct ofport *ofport_,
2342 const struct ofproto_port_queue *qdscp_list,
2345 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2346 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2347 struct hmap new = HMAP_INITIALIZER(&new);
2350 for (i = 0; i < n_qdscp; i++) {
2351 struct priority_to_dscp *pdscp;
2355 dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
2356 if (dpif_queue_to_priority(ofproto->backer->dpif, qdscp_list[i].queue,
2361 pdscp = get_priority(ofport, priority);
2363 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
2365 pdscp = xmalloc(sizeof *pdscp);
2366 pdscp->priority = priority;
2368 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2371 if (pdscp->dscp != dscp) {
2373 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2376 hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0));
2379 if (!hmap_is_empty(&ofport->priorities)) {
2380 ofport_clear_priorities(ofport);
2381 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2384 hmap_swap(&new, &ofport->priorities);
2392 /* Expires all MAC learning entries associated with 'bundle' and forces its
2393 * ofproto to revalidate every flow.
2395 * Normally MAC learning entries are removed only from the ofproto associated
2396 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2397 * are removed from every ofproto. When patch ports and SLB bonds are in use
2398 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2399 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2400 * with the host from which it migrated. */
2402 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2404 struct ofproto_dpif *ofproto = bundle->ofproto;
2405 struct mac_learning *ml = ofproto->ml;
2406 struct mac_entry *mac, *next_mac;
2408 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2409 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2410 if (mac->port.p == bundle) {
2412 struct ofproto_dpif *o;
2414 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2416 struct mac_entry *e;
2418 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan,
2421 mac_learning_expire(o->ml, e);
2427 mac_learning_expire(ml, mac);
2432 static struct ofbundle *
2433 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2435 struct ofbundle *bundle;
2437 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2438 &ofproto->bundles) {
2439 if (bundle->aux == aux) {
2446 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
2447 * ones that are found to 'bundles'. */
2449 bundle_lookup_multiple(struct ofproto_dpif *ofproto,
2450 void **auxes, size_t n_auxes,
2451 struct hmapx *bundles)
2455 hmapx_init(bundles);
2456 for (i = 0; i < n_auxes; i++) {
2457 struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
2459 hmapx_add(bundles, bundle);
2465 bundle_update(struct ofbundle *bundle)
2467 struct ofport_dpif *port;
2469 bundle->floodable = true;
2470 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2471 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2472 || !stp_forward_in_state(port->stp_state)) {
2473 bundle->floodable = false;
2480 bundle_del_port(struct ofport_dpif *port)
2482 struct ofbundle *bundle = port->bundle;
2484 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2486 list_remove(&port->bundle_node);
2487 port->bundle = NULL;
2490 lacp_slave_unregister(bundle->lacp, port);
2493 bond_slave_unregister(bundle->bond, port);
2496 bundle_update(bundle);
2500 bundle_add_port(struct ofbundle *bundle, uint16_t ofp_port,
2501 struct lacp_slave_settings *lacp)
2503 struct ofport_dpif *port;
2505 port = get_ofp_port(bundle->ofproto, ofp_port);
2510 if (port->bundle != bundle) {
2511 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2513 bundle_del_port(port);
2516 port->bundle = bundle;
2517 list_push_back(&bundle->ports, &port->bundle_node);
2518 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2519 || !stp_forward_in_state(port->stp_state)) {
2520 bundle->floodable = false;
2524 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2525 lacp_slave_register(bundle->lacp, port, lacp);
2532 bundle_destroy(struct ofbundle *bundle)
2534 struct ofproto_dpif *ofproto;
2535 struct ofport_dpif *port, *next_port;
2542 ofproto = bundle->ofproto;
2543 for (i = 0; i < MAX_MIRRORS; i++) {
2544 struct ofmirror *m = ofproto->mirrors[i];
2546 if (m->out == bundle) {
2548 } else if (hmapx_find_and_delete(&m->srcs, bundle)
2549 || hmapx_find_and_delete(&m->dsts, bundle)) {
2550 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2555 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2556 bundle_del_port(port);
2559 bundle_flush_macs(bundle, true);
2560 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2562 free(bundle->trunks);
2563 lacp_destroy(bundle->lacp);
2564 bond_destroy(bundle->bond);
2569 bundle_set(struct ofproto *ofproto_, void *aux,
2570 const struct ofproto_bundle_settings *s)
2572 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2573 bool need_flush = false;
2574 struct ofport_dpif *port;
2575 struct ofbundle *bundle;
2576 unsigned long *trunks;
2582 bundle_destroy(bundle_lookup(ofproto, aux));
2586 ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2587 ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2589 bundle = bundle_lookup(ofproto, aux);
2591 bundle = xmalloc(sizeof *bundle);
2593 bundle->ofproto = ofproto;
2594 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2595 hash_pointer(aux, 0));
2597 bundle->name = NULL;
2599 list_init(&bundle->ports);
2600 bundle->vlan_mode = PORT_VLAN_TRUNK;
2602 bundle->trunks = NULL;
2603 bundle->use_priority_tags = s->use_priority_tags;
2604 bundle->lacp = NULL;
2605 bundle->bond = NULL;
2607 bundle->floodable = true;
2609 bundle->src_mirrors = 0;
2610 bundle->dst_mirrors = 0;
2611 bundle->mirror_out = 0;
2614 if (!bundle->name || strcmp(s->name, bundle->name)) {
2616 bundle->name = xstrdup(s->name);
2621 if (!bundle->lacp) {
2622 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2623 bundle->lacp = lacp_create();
2625 lacp_configure(bundle->lacp, s->lacp);
2627 lacp_destroy(bundle->lacp);
2628 bundle->lacp = NULL;
2631 /* Update set of ports. */
2633 for (i = 0; i < s->n_slaves; i++) {
2634 if (!bundle_add_port(bundle, s->slaves[i],
2635 s->lacp ? &s->lacp_slaves[i] : NULL)) {
2639 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2640 struct ofport_dpif *next_port;
2642 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2643 for (i = 0; i < s->n_slaves; i++) {
2644 if (s->slaves[i] == port->up.ofp_port) {
2649 bundle_del_port(port);
2653 ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2655 if (list_is_empty(&bundle->ports)) {
2656 bundle_destroy(bundle);
2660 /* Set VLAN tagging mode */
2661 if (s->vlan_mode != bundle->vlan_mode
2662 || s->use_priority_tags != bundle->use_priority_tags) {
2663 bundle->vlan_mode = s->vlan_mode;
2664 bundle->use_priority_tags = s->use_priority_tags;
2669 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2670 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2672 if (vlan != bundle->vlan) {
2673 bundle->vlan = vlan;
2677 /* Get trunked VLANs. */
2678 switch (s->vlan_mode) {
2679 case PORT_VLAN_ACCESS:
2683 case PORT_VLAN_TRUNK:
2684 trunks = CONST_CAST(unsigned long *, s->trunks);
2687 case PORT_VLAN_NATIVE_UNTAGGED:
2688 case PORT_VLAN_NATIVE_TAGGED:
2689 if (vlan != 0 && (!s->trunks
2690 || !bitmap_is_set(s->trunks, vlan)
2691 || bitmap_is_set(s->trunks, 0))) {
2692 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2694 trunks = bitmap_clone(s->trunks, 4096);
2696 trunks = bitmap_allocate1(4096);
2698 bitmap_set1(trunks, vlan);
2699 bitmap_set0(trunks, 0);
2701 trunks = CONST_CAST(unsigned long *, s->trunks);
2708 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2709 free(bundle->trunks);
2710 if (trunks == s->trunks) {
2711 bundle->trunks = vlan_bitmap_clone(trunks);
2713 bundle->trunks = trunks;
2718 if (trunks != s->trunks) {
2723 if (!list_is_short(&bundle->ports)) {
2724 bundle->ofproto->has_bonded_bundles = true;
2726 if (bond_reconfigure(bundle->bond, s->bond)) {
2727 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2730 bundle->bond = bond_create(s->bond);
2731 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2734 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2735 bond_slave_register(bundle->bond, port, port->up.netdev);
2738 bond_destroy(bundle->bond);
2739 bundle->bond = NULL;
2742 /* If we changed something that would affect MAC learning, un-learn
2743 * everything on this port and force flow revalidation. */
2745 bundle_flush_macs(bundle, false);
2752 bundle_remove(struct ofport *port_)
2754 struct ofport_dpif *port = ofport_dpif_cast(port_);
2755 struct ofbundle *bundle = port->bundle;
2758 bundle_del_port(port);
2759 if (list_is_empty(&bundle->ports)) {
2760 bundle_destroy(bundle);
2761 } else if (list_is_short(&bundle->ports)) {
2762 bond_destroy(bundle->bond);
2763 bundle->bond = NULL;
2769 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2771 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2772 struct ofport_dpif *port = port_;
2773 uint8_t ea[ETH_ADDR_LEN];
2776 error = netdev_get_etheraddr(port->up.netdev, ea);
2778 struct ofpbuf packet;
2781 ofpbuf_init(&packet, 0);
2782 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2784 memcpy(packet_pdu, pdu, pdu_size);
2786 send_packet(port, &packet);
2787 ofpbuf_uninit(&packet);
2789 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2790 "%s (%s)", port->bundle->name,
2791 netdev_get_name(port->up.netdev), strerror(error));
2796 bundle_send_learning_packets(struct ofbundle *bundle)
2798 struct ofproto_dpif *ofproto = bundle->ofproto;
2799 int error, n_packets, n_errors;
2800 struct mac_entry *e;
2802 error = n_packets = n_errors = 0;
2803 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2804 if (e->port.p != bundle) {
2805 struct ofpbuf *learning_packet;
2806 struct ofport_dpif *port;
2810 /* The assignment to "port" is unnecessary but makes "grep"ing for
2811 * struct ofport_dpif more effective. */
2812 learning_packet = bond_compose_learning_packet(bundle->bond,
2816 ret = send_packet(port, learning_packet);
2817 ofpbuf_delete(learning_packet);
2827 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2828 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2829 "packets, last error was: %s",
2830 bundle->name, n_errors, n_packets, strerror(error));
2832 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2833 bundle->name, n_packets);
2838 bundle_run(struct ofbundle *bundle)
2841 lacp_run(bundle->lacp, send_pdu_cb);
2844 struct ofport_dpif *port;
2846 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2847 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2850 bond_run(bundle->bond, &bundle->ofproto->backer->revalidate_set,
2851 lacp_status(bundle->lacp));
2852 if (bond_should_send_learning_packets(bundle->bond)) {
2853 bundle_send_learning_packets(bundle);
2859 bundle_wait(struct ofbundle *bundle)
2862 lacp_wait(bundle->lacp);
2865 bond_wait(bundle->bond);
2872 mirror_scan(struct ofproto_dpif *ofproto)
2876 for (idx = 0; idx < MAX_MIRRORS; idx++) {
2877 if (!ofproto->mirrors[idx]) {
2884 static struct ofmirror *
2885 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
2889 for (i = 0; i < MAX_MIRRORS; i++) {
2890 struct ofmirror *mirror = ofproto->mirrors[i];
2891 if (mirror && mirror->aux == aux) {
2899 /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */
2901 mirror_update_dups(struct ofproto_dpif *ofproto)
2905 for (i = 0; i < MAX_MIRRORS; i++) {
2906 struct ofmirror *m = ofproto->mirrors[i];
2909 m->dup_mirrors = MIRROR_MASK_C(1) << i;
2913 for (i = 0; i < MAX_MIRRORS; i++) {
2914 struct ofmirror *m1 = ofproto->mirrors[i];
2921 for (j = i + 1; j < MAX_MIRRORS; j++) {
2922 struct ofmirror *m2 = ofproto->mirrors[j];
2924 if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) {
2925 m1->dup_mirrors |= MIRROR_MASK_C(1) << j;
2926 m2->dup_mirrors |= m1->dup_mirrors;
2933 mirror_set(struct ofproto *ofproto_, void *aux,
2934 const struct ofproto_mirror_settings *s)
2936 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2937 mirror_mask_t mirror_bit;
2938 struct ofbundle *bundle;
2939 struct ofmirror *mirror;
2940 struct ofbundle *out;
2941 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
2942 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
2945 mirror = mirror_lookup(ofproto, aux);
2947 mirror_destroy(mirror);
2953 idx = mirror_scan(ofproto);
2955 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
2957 ofproto->up.name, MAX_MIRRORS, s->name);
2961 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
2962 mirror->ofproto = ofproto;
2965 mirror->out_vlan = -1;
2966 mirror->name = NULL;
2969 if (!mirror->name || strcmp(s->name, mirror->name)) {
2971 mirror->name = xstrdup(s->name);
2974 /* Get the new configuration. */
2975 if (s->out_bundle) {
2976 out = bundle_lookup(ofproto, s->out_bundle);
2978 mirror_destroy(mirror);
2984 out_vlan = s->out_vlan;
2986 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
2987 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
2989 /* If the configuration has not changed, do nothing. */
2990 if (hmapx_equals(&srcs, &mirror->srcs)
2991 && hmapx_equals(&dsts, &mirror->dsts)
2992 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
2993 && mirror->out == out
2994 && mirror->out_vlan == out_vlan)
2996 hmapx_destroy(&srcs);
2997 hmapx_destroy(&dsts);
3001 hmapx_swap(&srcs, &mirror->srcs);
3002 hmapx_destroy(&srcs);
3004 hmapx_swap(&dsts, &mirror->dsts);
3005 hmapx_destroy(&dsts);
3007 free(mirror->vlans);
3008 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
3011 mirror->out_vlan = out_vlan;
3013 /* Update bundles. */
3014 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
3015 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
3016 if (hmapx_contains(&mirror->srcs, bundle)) {
3017 bundle->src_mirrors |= mirror_bit;
3019 bundle->src_mirrors &= ~mirror_bit;
3022 if (hmapx_contains(&mirror->dsts, bundle)) {
3023 bundle->dst_mirrors |= mirror_bit;
3025 bundle->dst_mirrors &= ~mirror_bit;
3028 if (mirror->out == bundle) {
3029 bundle->mirror_out |= mirror_bit;
3031 bundle->mirror_out &= ~mirror_bit;
3035 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3036 ofproto->has_mirrors = true;
3037 mac_learning_flush(ofproto->ml,
3038 &ofproto->backer->revalidate_set);
3039 mirror_update_dups(ofproto);
3045 mirror_destroy(struct ofmirror *mirror)
3047 struct ofproto_dpif *ofproto;
3048 mirror_mask_t mirror_bit;
3049 struct ofbundle *bundle;
3056 ofproto = mirror->ofproto;
3057 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3058 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
3060 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
3061 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
3062 bundle->src_mirrors &= ~mirror_bit;
3063 bundle->dst_mirrors &= ~mirror_bit;
3064 bundle->mirror_out &= ~mirror_bit;
3067 hmapx_destroy(&mirror->srcs);
3068 hmapx_destroy(&mirror->dsts);
3069 free(mirror->vlans);
3071 ofproto->mirrors[mirror->idx] = NULL;
3075 mirror_update_dups(ofproto);
3077 ofproto->has_mirrors = false;
3078 for (i = 0; i < MAX_MIRRORS; i++) {
3079 if (ofproto->mirrors[i]) {
3080 ofproto->has_mirrors = true;
3087 mirror_get_stats(struct ofproto *ofproto_, void *aux,
3088 uint64_t *packets, uint64_t *bytes)
3090 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3091 struct ofmirror *mirror = mirror_lookup(ofproto, aux);
3094 *packets = *bytes = UINT64_MAX;
3100 *packets = mirror->packet_count;
3101 *bytes = mirror->byte_count;
3107 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
3109 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3110 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
3111 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
3117 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
3119 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3120 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
3121 return bundle && bundle->mirror_out != 0;
3125 forward_bpdu_changed(struct ofproto *ofproto_)
3127 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3128 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3132 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
3135 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3136 mac_learning_set_idle_time(ofproto->ml, idle_time);
3137 mac_learning_set_max_entries(ofproto->ml, max_entries);
3142 static struct ofport_dpif *
3143 get_ofp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
3145 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
3146 return ofport ? ofport_dpif_cast(ofport) : NULL;
3149 static struct ofport_dpif *
3150 get_odp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
3152 struct ofport_dpif *port = odp_port_to_ofport(ofproto->backer, odp_port);
3153 return port && &ofproto->up == port->up.ofproto ? port : NULL;
3157 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
3158 struct ofproto_port *ofproto_port,
3159 struct dpif_port *dpif_port)
3161 ofproto_port->name = dpif_port->name;
3162 ofproto_port->type = dpif_port->type;
3163 ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
3166 static struct ofport_dpif *
3167 ofport_get_peer(const struct ofport_dpif *ofport_dpif)
3169 const struct ofproto_dpif *ofproto;
3172 peer = netdev_vport_patch_peer(ofport_dpif->up.netdev);
3177 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
3178 struct ofport *ofport;
3180 ofport = shash_find_data(&ofproto->up.port_by_name, peer);
3181 if (ofport && ofport->ofproto->ofproto_class == &ofproto_dpif_class) {
3182 return ofport_dpif_cast(ofport);
3189 port_run_fast(struct ofport_dpif *ofport)
3191 if (ofport->cfm && cfm_should_send_ccm(ofport->cfm)) {
3192 struct ofpbuf packet;
3194 ofpbuf_init(&packet, 0);
3195 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
3196 send_packet(ofport, &packet);
3197 ofpbuf_uninit(&packet);
3200 if (ofport->bfd && bfd_should_send_packet(ofport->bfd)) {
3201 struct ofpbuf packet;
3203 ofpbuf_init(&packet, 0);
3204 bfd_put_packet(ofport->bfd, &packet, ofport->up.pp.hw_addr);
3205 send_packet(ofport, &packet);
3206 ofpbuf_uninit(&packet);
3211 port_run(struct ofport_dpif *ofport)
3213 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
3214 bool carrier_changed = carrier_seq != ofport->carrier_seq;
3215 bool enable = netdev_get_carrier(ofport->up.netdev);
3217 ofport->carrier_seq = carrier_seq;
3219 port_run_fast(ofport);
3221 if (ofport->tnl_port
3222 && tnl_port_reconfigure(&ofport->up, ofport->odp_port,
3223 &ofport->tnl_port)) {
3224 ofproto_dpif_cast(ofport->up.ofproto)->backer->need_revalidate = true;
3228 int cfm_opup = cfm_get_opup(ofport->cfm);
3230 cfm_run(ofport->cfm);
3231 enable = enable && !cfm_get_fault(ofport->cfm);
3233 if (cfm_opup >= 0) {
3234 enable = enable && cfm_opup;
3239 bfd_run(ofport->bfd);
3240 enable = enable && bfd_forwarding(ofport->bfd);
3243 if (ofport->bundle) {
3244 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
3245 if (carrier_changed) {
3246 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
3250 if (ofport->may_enable != enable) {
3251 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3253 if (ofproto->has_bundle_action) {
3254 ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
3258 ofport->may_enable = enable;
3262 port_wait(struct ofport_dpif *ofport)
3265 cfm_wait(ofport->cfm);
3269 bfd_wait(ofport->bfd);
3274 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
3275 struct ofproto_port *ofproto_port)
3277 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3278 struct dpif_port dpif_port;
3281 if (sset_contains(&ofproto->ghost_ports, devname)) {
3282 const char *type = netdev_get_type_from_name(devname);
3284 /* We may be called before ofproto->up.port_by_name is populated with
3285 * the appropriate ofport. For this reason, we must get the name and
3286 * type from the netdev layer directly. */
3288 const struct ofport *ofport;
3290 ofport = shash_find_data(&ofproto->up.port_by_name, devname);
3291 ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
3292 ofproto_port->name = xstrdup(devname);
3293 ofproto_port->type = xstrdup(type);
3299 if (!sset_contains(&ofproto->ports, devname)) {
3302 error = dpif_port_query_by_name(ofproto->backer->dpif,
3303 devname, &dpif_port);
3305 ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
3311 port_add(struct ofproto *ofproto_, struct netdev *netdev)
3313 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3314 const char *devname = netdev_get_name(netdev);
3315 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
3316 const char *dp_port_name;
3318 if (netdev_vport_is_patch(netdev)) {
3319 sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
3323 dp_port_name = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
3324 if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
3325 uint32_t port_no = UINT32_MAX;
3328 error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
3332 if (netdev_get_tunnel_config(netdev)) {
3333 simap_put(&ofproto->backer->tnl_backers, dp_port_name, port_no);
3337 if (netdev_get_tunnel_config(netdev)) {
3338 sset_add(&ofproto->ghost_ports, devname);
3340 sset_add(&ofproto->ports, devname);
3346 port_del(struct ofproto *ofproto_, uint16_t ofp_port)
3348 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3349 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
3356 sset_find_and_delete(&ofproto->ghost_ports,
3357 netdev_get_name(ofport->up.netdev));
3358 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3359 if (!ofport->tnl_port) {
3360 error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
3362 /* The caller is going to close ofport->up.netdev. If this is a
3363 * bonded port, then the bond is using that netdev, so remove it
3364 * from the bond. The client will need to reconfigure everything
3365 * after deleting ports, so then the slave will get re-added. */
3366 bundle_remove(&ofport->up);
3373 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
3375 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3380 error = netdev_get_stats(ofport->up.netdev, stats);
3382 if (!error && ofport_->ofp_port == OFPP_LOCAL) {
3383 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3385 /* ofproto->stats.tx_packets represents packets that we created
3386 * internally and sent to some port (e.g. packets sent with
3387 * send_packet()). Account for them as if they had come from
3388 * OFPP_LOCAL and got forwarded. */
3390 if (stats->rx_packets != UINT64_MAX) {
3391 stats->rx_packets += ofproto->stats.tx_packets;
3394 if (stats->rx_bytes != UINT64_MAX) {
3395 stats->rx_bytes += ofproto->stats.tx_bytes;
3398 /* ofproto->stats.rx_packets represents packets that were received on
3399 * some port and we processed internally and dropped (e.g. STP).
3400 * Account for them as if they had been forwarded to OFPP_LOCAL. */
3402 if (stats->tx_packets != UINT64_MAX) {
3403 stats->tx_packets += ofproto->stats.rx_packets;
3406 if (stats->tx_bytes != UINT64_MAX) {
3407 stats->tx_bytes += ofproto->stats.rx_bytes;
3414 struct port_dump_state {
3419 struct ofproto_port port;
3424 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
3426 *statep = xzalloc(sizeof(struct port_dump_state));
3431 port_dump_next(const struct ofproto *ofproto_, void *state_,
3432 struct ofproto_port *port)
3434 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3435 struct port_dump_state *state = state_;
3436 const struct sset *sset;
3437 struct sset_node *node;
3439 if (state->has_port) {
3440 ofproto_port_destroy(&state->port);
3441 state->has_port = false;
3443 sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
3444 while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
3447 error = port_query_by_name(ofproto_, node->name, &state->port);
3449 *port = state->port;
3450 state->has_port = true;
3452 } else if (error != ENODEV) {
3457 if (!state->ghost) {
3458 state->ghost = true;
3461 return port_dump_next(ofproto_, state_, port);
3468 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3470 struct port_dump_state *state = state_;
3472 if (state->has_port) {
3473 ofproto_port_destroy(&state->port);
3480 port_poll(const struct ofproto *ofproto_, char **devnamep)
3482 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3484 if (ofproto->port_poll_errno) {
3485 int error = ofproto->port_poll_errno;
3486 ofproto->port_poll_errno = 0;
3490 if (sset_is_empty(&ofproto->port_poll_set)) {
3494 *devnamep = sset_pop(&ofproto->port_poll_set);
3499 port_poll_wait(const struct ofproto *ofproto_)
3501 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3502 dpif_port_poll_wait(ofproto->backer->dpif);
3506 port_is_lacp_current(const struct ofport *ofport_)
3508 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3509 return (ofport->bundle && ofport->bundle->lacp
3510 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3514 /* Upcall handling. */
3516 /* Flow miss batching.
3518 * Some dpifs implement operations faster when you hand them off in a batch.
3519 * To allow batching, "struct flow_miss" queues the dpif-related work needed
3520 * for a given flow. Each "struct flow_miss" corresponds to sending one or
3521 * more packets, plus possibly installing the flow in the dpif.
3523 * So far we only batch the operations that affect flow setup time the most.
3524 * It's possible to batch more than that, but the benefit might be minimal. */
3526 struct hmap_node hmap_node;
3527 struct ofproto_dpif *ofproto;
3529 enum odp_key_fitness key_fitness;
3530 const struct nlattr *key;
3532 struct initial_vals initial_vals;
3533 struct list packets;
3534 enum dpif_upcall_type upcall_type;
3537 struct flow_miss_op {
3538 struct dpif_op dpif_op;
3540 uint64_t slow_stub[128 / 8]; /* Buffer for compose_slow_path() */
3541 struct xlate_out xout;
3542 bool xout_garbage; /* 'xout' needs to be uninitialized? */
3545 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
3546 * OpenFlow controller as necessary according to their individual
3547 * configurations. */
3549 send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet,
3550 const struct flow *flow)
3552 struct ofputil_packet_in pin;
3554 pin.packet = packet->data;
3555 pin.packet_len = packet->size;
3556 pin.reason = OFPR_NO_MATCH;
3557 pin.controller_id = 0;
3562 pin.send_len = 0; /* not used for flow table misses */
3564 flow_get_metadata(flow, &pin.fmd);
3566 connmgr_send_packet_in(ofproto->up.connmgr, &pin);
3569 static enum slow_path_reason
3570 process_special(struct ofproto_dpif *ofproto, const struct flow *flow,
3571 const struct ofport_dpif *ofport, const struct ofpbuf *packet)
3575 } else if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) {
3577 cfm_process_heartbeat(ofport->cfm, packet);
3580 } else if (ofport->bfd && bfd_should_process_flow(flow)) {
3582 bfd_process_packet(ofport->bfd, flow, packet);
3585 } else if (ofport->bundle && ofport->bundle->lacp
3586 && flow->dl_type == htons(ETH_TYPE_LACP)) {
3588 lacp_process_packet(ofport->bundle->lacp, ofport, packet);
3591 } else if (ofproto->stp && stp_should_process_flow(flow)) {
3593 stp_process_packet(ofport, packet);
3601 static struct flow_miss *
3602 flow_miss_find(struct hmap *todo, const struct ofproto_dpif *ofproto,
3603 const struct flow *flow, uint32_t hash)
3605 struct flow_miss *miss;
3607 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
3608 if (miss->ofproto == ofproto && flow_equal(&miss->flow, flow)) {
3616 /* Partially Initializes 'op' as an "execute" operation for 'miss' and
3617 * 'packet'. The caller must initialize op->actions and op->actions_len. If
3618 * 'miss' is associated with a subfacet the caller must also initialize the
3619 * returned op->subfacet, and if anything needs to be freed after processing
3620 * the op, the caller must initialize op->garbage also. */
3622 init_flow_miss_execute_op(struct flow_miss *miss, struct ofpbuf *packet,
3623 struct flow_miss_op *op)
3625 if (miss->flow.vlan_tci != miss->initial_vals.vlan_tci) {
3626 /* This packet was received on a VLAN splinter port. We
3627 * added a VLAN to the packet to make the packet resemble
3628 * the flow, but the actions were composed assuming that
3629 * the packet contained no VLAN. So, we must remove the
3630 * VLAN header from the packet before trying to execute the
3632 eth_pop_vlan(packet);
3635 op->xout_garbage = false;
3636 op->dpif_op.type = DPIF_OP_EXECUTE;
3637 op->dpif_op.u.execute.key = miss->key;
3638 op->dpif_op.u.execute.key_len = miss->key_len;
3639 op->dpif_op.u.execute.packet = packet;
3642 /* Helper for handle_flow_miss_without_facet() and
3643 * handle_flow_miss_with_facet(). */
3645 handle_flow_miss_common(struct rule_dpif *rule,
3646 struct ofpbuf *packet, const struct flow *flow)
3648 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3650 if (rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
3652 * Extra-special case for fail-open mode.
3654 * We are in fail-open mode and the packet matched the fail-open
3655 * rule, but we are connected to a controller too. We should send
3656 * the packet up to the controller in the hope that it will try to
3657 * set up a flow and thereby allow us to exit fail-open.
3659 * See the top-level comment in fail-open.c for more information.
3661 send_packet_in_miss(ofproto, packet, flow);
3665 /* Figures out whether a flow that missed in 'ofproto', whose details are in
3666 * 'miss', is likely to be worth tracking in detail in userspace and (usually)
3667 * installing a datapath flow. The answer is usually "yes" (a return value of
3668 * true). However, for short flows the cost of bookkeeping is much higher than
3669 * the benefits, so when the datapath holds a large number of flows we impose
3670 * some heuristics to decide which flows are likely to be worth tracking. */
3672 flow_miss_should_make_facet(struct ofproto_dpif *ofproto,
3673 struct flow_miss *miss, uint32_t hash)
3675 if (!ofproto->governor) {
3678 n_subfacets = hmap_count(&ofproto->subfacets);
3679 if (n_subfacets * 2 <= ofproto->up.flow_eviction_threshold) {
3683 ofproto->governor = governor_create(ofproto->up.name);
3686 return governor_should_install_flow(ofproto->governor, hash,
3687 list_size(&miss->packets));
3690 /* Handles 'miss', which matches 'rule', without creating a facet or subfacet
3691 * or creating any datapath flow. May add an "execute" operation to 'ops' and
3692 * increment '*n_ops'. */
3694 handle_flow_miss_without_facet(struct flow_miss *miss,
3695 struct flow_miss_op *ops, size_t *n_ops)
3697 struct rule_dpif *rule = rule_dpif_lookup(miss->ofproto, &miss->flow);
3698 long long int now = time_msec();
3699 struct ofpbuf *packet;
3700 struct xlate_in xin;
3702 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3703 struct flow_miss_op *op = &ops[*n_ops];
3704 struct dpif_flow_stats stats;
3706 COVERAGE_INC(facet_suppress);
3708 handle_flow_miss_common(rule, packet, &miss->flow);
3710 dpif_flow_stats_extract(&miss->flow, packet, now, &stats);
3711 rule_credit_stats(rule, &stats);
3713 xlate_in_init(&xin, miss->ofproto, &miss->flow, &miss->initial_vals,
3714 rule, stats.tcp_flags, packet);
3715 xin.resubmit_stats = &stats;
3716 xlate_actions(&xin, &op->xout);
3718 if (op->xout.odp_actions.size) {
3719 struct dpif_execute *execute = &op->dpif_op.u.execute;
3721 init_flow_miss_execute_op(miss, packet, op);
3722 execute->actions = op->xout.odp_actions.data;
3723 execute->actions_len = op->xout.odp_actions.size;
3724 op->xout_garbage = true;
3728 xlate_out_uninit(&op->xout);
3733 /* Handles 'miss', which matches 'facet'. May add any required datapath
3734 * operations to 'ops', incrementing '*n_ops' for each new op.
3736 * All of the packets in 'miss' are considered to have arrived at time 'now'.
3737 * This is really important only for new facets: if we just called time_msec()
3738 * here, then the new subfacet or its packets could look (occasionally) as
3739 * though it was used some time after the facet was used. That can make a
3740 * one-packet flow look like it has a nonzero duration, which looks odd in
3741 * e.g. NetFlow statistics. */
3743 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
3745 struct flow_miss_op *ops, size_t *n_ops)
3747 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3748 enum subfacet_path want_path;
3749 struct subfacet *subfacet;
3750 struct ofpbuf *packet;
3752 subfacet = subfacet_create(facet, miss, now);
3753 want_path = subfacet->facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
3755 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3756 struct flow_miss_op *op = &ops[*n_ops];
3757 struct dpif_flow_stats stats;
3759 handle_flow_miss_common(facet->rule, packet, &miss->flow);
3761 if (want_path != SF_FAST_PATH) {
3762 struct xlate_in xin;
3764 xlate_in_init(&xin, ofproto, &facet->flow, &facet->initial_vals,
3765 facet->rule, 0, packet);
3766 xlate_actions_for_side_effects(&xin);
3769 dpif_flow_stats_extract(&facet->flow, packet, now, &stats);
3770 subfacet_update_stats(subfacet, &stats);
3772 if (facet->xout.odp_actions.size) {
3773 struct dpif_execute *execute = &op->dpif_op.u.execute;
3775 init_flow_miss_execute_op(miss, packet, op);
3776 execute->actions = facet->xout.odp_actions.data,
3777 execute->actions_len = facet->xout.odp_actions.size;
3782 if (miss->upcall_type == DPIF_UC_MISS || subfacet->path != want_path) {
3783 struct flow_miss_op *op = &ops[(*n_ops)++];
3784 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
3786 subfacet->path = want_path;
3788 op->xout_garbage = false;
3789 op->dpif_op.type = DPIF_OP_FLOW_PUT;
3790 put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
3791 put->key = miss->key;
3792 put->key_len = miss->key_len;
3793 if (want_path == SF_FAST_PATH) {
3794 put->actions = facet->xout.odp_actions.data;
3795 put->actions_len = facet->xout.odp_actions.size;
3797 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
3798 op->slow_stub, sizeof op->slow_stub,
3799 &put->actions, &put->actions_len);
3805 /* Handles flow miss 'miss'. May add any required datapath operations
3806 * to 'ops', incrementing '*n_ops' for each new op. */
3808 handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops,
3811 struct ofproto_dpif *ofproto = miss->ofproto;
3812 struct facet *facet;
3816 /* The caller must ensure that miss->hmap_node.hash contains
3817 * flow_hash(miss->flow, 0). */
3818 hash = miss->hmap_node.hash;
3820 facet = facet_lookup_valid(ofproto, &miss->flow, hash);
3822 /* There does not exist a bijection between 'struct flow' and datapath
3823 * flow keys with fitness ODP_FIT_TO_LITTLE. This breaks a fundamental
3824 * assumption used throughout the facet and subfacet handling code.
3825 * Since we have to handle these misses in userspace anyway, we simply
3826 * skip facet creation, avoiding the problem altogether. */
3827 if (miss->key_fitness == ODP_FIT_TOO_LITTLE
3828 || !flow_miss_should_make_facet(ofproto, miss, hash)) {
3829 handle_flow_miss_without_facet(miss, ops, n_ops);
3833 facet = facet_create(miss, hash);
3838 handle_flow_miss_with_facet(miss, facet, now, ops, n_ops);
3841 static struct drop_key *
3842 drop_key_lookup(const struct dpif_backer *backer, const struct nlattr *key,
3845 struct drop_key *drop_key;
3847 HMAP_FOR_EACH_WITH_HASH (drop_key, hmap_node, hash_bytes(key, key_len, 0),
3848 &backer->drop_keys) {
3849 if (drop_key->key_len == key_len
3850 && !memcmp(drop_key->key, key, key_len)) {
3858 drop_key_clear(struct dpif_backer *backer)
3860 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3861 struct drop_key *drop_key, *next;
3863 HMAP_FOR_EACH_SAFE (drop_key, next, hmap_node, &backer->drop_keys) {
3866 error = dpif_flow_del(backer->dpif, drop_key->key, drop_key->key_len,
3868 if (error && !VLOG_DROP_WARN(&rl)) {
3869 struct ds ds = DS_EMPTY_INITIALIZER;
3870 odp_flow_key_format(drop_key->key, drop_key->key_len, &ds);
3871 VLOG_WARN("Failed to delete drop key (%s) (%s)", strerror(error),
3876 hmap_remove(&backer->drop_keys, &drop_key->hmap_node);
3877 free(drop_key->key);
3882 /* Given a datpath, packet, and flow metadata ('backer', 'packet', and 'key'
3883 * respectively), populates 'flow' with the result of odp_flow_key_to_flow().
3884 * Optionally, if nonnull, populates 'fitnessp' with the fitness of 'flow' as
3885 * returned by odp_flow_key_to_flow(). Also, optionally populates 'ofproto'
3886 * with the ofproto_dpif, and 'odp_in_port' with the datapath in_port, that
3887 * 'packet' ingressed.
3889 * If 'ofproto' is nonnull, requires 'flow''s in_port to exist. Otherwise sets
3890 * 'flow''s in_port to OFPP_NONE.
3892 * This function does post-processing on data returned from
3893 * odp_flow_key_to_flow() to help make VLAN splinters transparent to the rest
3894 * of the upcall processing logic. In particular, if the extracted in_port is
3895 * a VLAN splinter port, it replaces flow->in_port by the "real" port, sets
3896 * flow->vlan_tci correctly for the VLAN of the VLAN splinter port, and pushes
3897 * a VLAN header onto 'packet' (if it is nonnull).
3899 * Optionally, if 'initial_vals' is nonnull, sets 'initial_vals->vlan_tci'
3900 * to the VLAN TCI with which the packet was really received, that is, the
3901 * actual VLAN TCI extracted by odp_flow_key_to_flow(). (This differs from
3902 * the value returned in flow->vlan_tci only for packets received on
3905 * Similarly, this function also includes some logic to help with tunnels. It
3906 * may modify 'flow' as necessary to make the tunneling implementation
3907 * transparent to the upcall processing logic.
3909 * Returns 0 if successful, ENODEV if the parsed flow has no associated ofport,
3910 * or some other positive errno if there are other problems. */
3912 ofproto_receive(const struct dpif_backer *backer, struct ofpbuf *packet,
3913 const struct nlattr *key, size_t key_len,
3914 struct flow *flow, enum odp_key_fitness *fitnessp,
3915 struct ofproto_dpif **ofproto, uint32_t *odp_in_port,
3916 struct initial_vals *initial_vals)
3918 const struct ofport_dpif *port;
3919 enum odp_key_fitness fitness;
3922 fitness = odp_flow_key_to_flow(key, key_len, flow);
3923 if (fitness == ODP_FIT_ERROR) {
3929 initial_vals->vlan_tci = flow->vlan_tci;
3933 *odp_in_port = flow->in_port;
3936 port = (tnl_port_should_receive(flow)
3937 ? ofport_dpif_cast(tnl_port_receive(flow))
3938 : odp_port_to_ofport(backer, flow->in_port));
3939 flow->in_port = port ? port->up.ofp_port : OFPP_NONE;
3944 /* XXX: Since the tunnel module is not scoped per backer, for a tunnel port
3945 * it's theoretically possible that we'll receive an ofport belonging to an
3946 * entirely different datapath. In practice, this can't happen because no
3947 * platforms has two separate datapaths which each support tunneling. */
3948 ovs_assert(ofproto_dpif_cast(port->up.ofproto)->backer == backer);
3950 if (vsp_adjust_flow(ofproto_dpif_cast(port->up.ofproto), flow)) {
3952 /* Make the packet resemble the flow, so that it gets sent to
3953 * an OpenFlow controller properly, so that it looks correct
3954 * for sFlow, and so that flow_extract() will get the correct
3955 * vlan_tci if it is called on 'packet'.
3957 * The allocated space inside 'packet' probably also contains
3958 * 'key', that is, both 'packet' and 'key' are probably part of
3959 * a struct dpif_upcall (see the large comment on that
3960 * structure definition), so pushing data on 'packet' is in
3961 * general not a good idea since it could overwrite 'key' or
3962 * free it as a side effect. However, it's OK in this special
3963 * case because we know that 'packet' is inside a Netlink
3964 * attribute: pushing 4 bytes will just overwrite the 4-byte
3965 * "struct nlattr", which is fine since we don't need that
3966 * header anymore. */
3967 eth_push_vlan(packet, flow->vlan_tci);
3969 /* We can't reproduce 'key' from 'flow'. */
3970 fitness = fitness == ODP_FIT_PERFECT ? ODP_FIT_TOO_MUCH : fitness;
3975 *ofproto = ofproto_dpif_cast(port->up.ofproto);
3980 *fitnessp = fitness;
3986 handle_miss_upcalls(struct dpif_backer *backer, struct dpif_upcall *upcalls,
3989 struct dpif_upcall *upcall;
3990 struct flow_miss *miss;
3991 struct flow_miss misses[FLOW_MISS_MAX_BATCH];
3992 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
3993 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
4003 /* Construct the to-do list.
4005 * This just amounts to extracting the flow from each packet and sticking
4006 * the packets that have the same flow in the same "flow_miss" structure so
4007 * that we can process them together. */
4010 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
4011 struct flow_miss *miss = &misses[n_misses];
4012 struct flow_miss *existing_miss;
4013 struct ofproto_dpif *ofproto;
4014 uint32_t odp_in_port;
4019 error = ofproto_receive(backer, upcall->packet, upcall->key,
4020 upcall->key_len, &flow, &miss->key_fitness,
4021 &ofproto, &odp_in_port, &miss->initial_vals);
4022 if (error == ENODEV) {
4023 struct drop_key *drop_key;
4025 /* Received packet on datapath port for which we couldn't
4026 * associate an ofproto. This can happen if a port is removed
4027 * while traffic is being received. Print a rate-limited message
4028 * in case it happens frequently. Install a drop flow so
4029 * that future packets of the flow are inexpensively dropped
4031 VLOG_INFO_RL(&rl, "received packet on unassociated datapath port "
4032 "%"PRIu32, odp_in_port);
4034 drop_key = drop_key_lookup(backer, upcall->key, upcall->key_len);
4036 drop_key = xmalloc(sizeof *drop_key);
4037 drop_key->key = xmemdup(upcall->key, upcall->key_len);
4038 drop_key->key_len = upcall->key_len;
4040 hmap_insert(&backer->drop_keys, &drop_key->hmap_node,
4041 hash_bytes(drop_key->key, drop_key->key_len, 0));
4042 dpif_flow_put(backer->dpif, DPIF_FP_CREATE | DPIF_FP_MODIFY,
4043 drop_key->key, drop_key->key_len, NULL, 0, NULL);
4051 ofproto->n_missed++;
4052 flow_extract(upcall->packet, flow.skb_priority, flow.skb_mark,
4053 &flow.tunnel, flow.in_port, &miss->flow);
4055 /* Add other packets to a to-do list. */
4056 hash = flow_hash(&miss->flow, 0);
4057 existing_miss = flow_miss_find(&todo, ofproto, &miss->flow, hash);
4058 if (!existing_miss) {
4059 hmap_insert(&todo, &miss->hmap_node, hash);
4060 miss->ofproto = ofproto;
4061 miss->key = upcall->key;
4062 miss->key_len = upcall->key_len;
4063 miss->upcall_type = upcall->type;
4064 list_init(&miss->packets);
4068 miss = existing_miss;
4070 list_push_back(&miss->packets, &upcall->packet->list_node);
4073 /* Process each element in the to-do list, constructing the set of
4074 * operations to batch. */
4076 HMAP_FOR_EACH (miss, hmap_node, &todo) {
4077 handle_flow_miss(miss, flow_miss_ops, &n_ops);
4079 ovs_assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
4081 /* Execute batch. */
4082 for (i = 0; i < n_ops; i++) {
4083 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
4085 dpif_operate(backer->dpif, dpif_ops, n_ops);
4088 for (i = 0; i < n_ops; i++) {
4089 if (flow_miss_ops[i].xout_garbage) {
4090 xlate_out_uninit(&flow_miss_ops[i].xout);
4093 hmap_destroy(&todo);
4096 static enum { SFLOW_UPCALL, MISS_UPCALL, BAD_UPCALL, FLOW_SAMPLE_UPCALL,
4098 classify_upcall(const struct dpif_upcall *upcall)
4100 size_t userdata_len;
4101 union user_action_cookie cookie;
4103 /* First look at the upcall type. */
4104 switch (upcall->type) {
4105 case DPIF_UC_ACTION:
4111 case DPIF_N_UC_TYPES:
4113 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
4117 /* "action" upcalls need a closer look. */
4118 if (!upcall->userdata) {
4119 VLOG_WARN_RL(&rl, "action upcall missing cookie");
4122 userdata_len = nl_attr_get_size(upcall->userdata);
4123 if (userdata_len < sizeof cookie.type
4124 || userdata_len > sizeof cookie) {
4125 VLOG_WARN_RL(&rl, "action upcall cookie has unexpected size %zu",
4129 memset(&cookie, 0, sizeof cookie);
4130 memcpy(&cookie, nl_attr_get(upcall->userdata), userdata_len);
4131 if (userdata_len == sizeof cookie.sflow
4132 && cookie.type == USER_ACTION_COOKIE_SFLOW) {
4133 return SFLOW_UPCALL;
4134 } else if (userdata_len == sizeof cookie.slow_path
4135 && cookie.type == USER_ACTION_COOKIE_SLOW_PATH) {
4137 } else if (userdata_len == sizeof cookie.flow_sample
4138 && cookie.type == USER_ACTION_COOKIE_FLOW_SAMPLE) {
4139 return FLOW_SAMPLE_UPCALL;
4140 } else if (userdata_len == sizeof cookie.ipfix
4141 && cookie.type == USER_ACTION_COOKIE_IPFIX) {
4142 return IPFIX_UPCALL;
4144 VLOG_WARN_RL(&rl, "invalid user cookie of type %"PRIu16
4145 " and size %zu", cookie.type, userdata_len);
4151 handle_sflow_upcall(struct dpif_backer *backer,
4152 const struct dpif_upcall *upcall)
4154 struct ofproto_dpif *ofproto;
4155 union user_action_cookie cookie;
4157 uint32_t odp_in_port;
4159 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
4160 &flow, NULL, &ofproto, &odp_in_port, NULL)
4161 || !ofproto->sflow) {
4165 memset(&cookie, 0, sizeof cookie);
4166 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.sflow);
4167 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
4168 odp_in_port, &cookie);
4172 handle_flow_sample_upcall(struct dpif_backer *backer,
4173 const struct dpif_upcall *upcall)
4175 struct ofproto_dpif *ofproto;
4176 union user_action_cookie cookie;
4179 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
4180 &flow, NULL, &ofproto, NULL, NULL)
4181 || !ofproto->ipfix) {
4185 memset(&cookie, 0, sizeof cookie);
4186 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.flow_sample);
4188 /* The flow reflects exactly the contents of the packet. Sample
4189 * the packet using it. */
4190 dpif_ipfix_flow_sample(ofproto->ipfix, upcall->packet, &flow,
4191 cookie.flow_sample.collector_set_id,
4192 cookie.flow_sample.probability,
4193 cookie.flow_sample.obs_domain_id,
4194 cookie.flow_sample.obs_point_id);
4198 handle_ipfix_upcall(struct dpif_backer *backer,
4199 const struct dpif_upcall *upcall)
4201 struct ofproto_dpif *ofproto;
4204 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
4205 &flow, NULL, &ofproto, NULL, NULL)
4206 || !ofproto->ipfix) {
4210 /* The flow reflects exactly the contents of the packet. Sample
4211 * the packet using it. */
4212 dpif_ipfix_bridge_sample(ofproto->ipfix, upcall->packet, &flow);
4216 handle_upcalls(struct dpif_backer *backer, unsigned int max_batch)
4218 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
4219 struct ofpbuf miss_bufs[FLOW_MISS_MAX_BATCH];
4220 uint64_t miss_buf_stubs[FLOW_MISS_MAX_BATCH][4096 / 8];
4225 ovs_assert(max_batch <= FLOW_MISS_MAX_BATCH);
4228 for (n_processed = 0; n_processed < max_batch; n_processed++) {
4229 struct dpif_upcall *upcall = &misses[n_misses];
4230 struct ofpbuf *buf = &miss_bufs[n_misses];
4233 ofpbuf_use_stub(buf, miss_buf_stubs[n_misses],
4234 sizeof miss_buf_stubs[n_misses]);
4235 error = dpif_recv(backer->dpif, upcall, buf);
4241 switch (classify_upcall(upcall)) {
4243 /* Handle it later. */
4248 handle_sflow_upcall(backer, upcall);
4252 case FLOW_SAMPLE_UPCALL:
4253 handle_flow_sample_upcall(backer, upcall);
4258 handle_ipfix_upcall(backer, upcall);
4268 /* Handle deferred MISS_UPCALL processing. */
4269 handle_miss_upcalls(backer, misses, n_misses);
4270 for (i = 0; i < n_misses; i++) {
4271 ofpbuf_uninit(&miss_bufs[i]);
4277 /* Flow expiration. */
4279 static int subfacet_max_idle(const struct ofproto_dpif *);
4280 static void update_stats(struct dpif_backer *);
4281 static void rule_expire(struct rule_dpif *);
4282 static void expire_subfacets(struct ofproto_dpif *, int dp_max_idle);
4284 /* This function is called periodically by run(). Its job is to collect
4285 * updates for the flows that have been installed into the datapath, most
4286 * importantly when they last were used, and then use that information to
4287 * expire flows that have not been used recently.
4289 * Returns the number of milliseconds after which it should be called again. */
4291 expire(struct dpif_backer *backer)
4293 struct ofproto_dpif *ofproto;
4294 int max_idle = INT32_MAX;
4296 /* Periodically clear out the drop keys in an effort to keep them
4297 * relatively few. */
4298 drop_key_clear(backer);
4300 /* Update stats for each flow in the backer. */
4301 update_stats(backer);
4303 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4304 long long int avg_subfacet_life_span;
4305 struct rule *rule, *next_rule;
4306 struct subfacet *subfacet;
4309 if (ofproto->backer != backer) {
4313 avg_subfacet_life_span = 0;
4314 if (!hmap_is_empty(&ofproto->subfacets)) {
4315 long long int now = time_msec();
4316 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
4317 avg_subfacet_life_span += now - subfacet->created;
4319 avg_subfacet_life_span /= hmap_count(&ofproto->subfacets);
4321 ofproto->avg_subfacet_life_span += avg_subfacet_life_span;
4322 ofproto->avg_subfacet_life_span /= 2;
4324 ofproto->avg_n_subfacet += hmap_count(&ofproto->subfacets);
4325 ofproto->avg_n_subfacet /= 2;
4327 ofproto->max_n_subfacet = MAX(ofproto->max_n_subfacet,
4328 hmap_count(&ofproto->subfacets));
4330 /* Expire subfacets that have been idle too long. */
4331 dp_max_idle = subfacet_max_idle(ofproto);
4332 expire_subfacets(ofproto, dp_max_idle);
4334 max_idle = MIN(max_idle, dp_max_idle);
4336 /* Expire OpenFlow flows whose idle_timeout or hard_timeout
4338 LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
4339 &ofproto->up.expirable) {
4340 rule_expire(rule_dpif_cast(rule));
4343 /* All outstanding data in existing flows has been accounted, so it's a
4344 * good time to do bond rebalancing. */
4345 if (ofproto->has_bonded_bundles) {
4346 struct ofbundle *bundle;
4348 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
4350 bond_rebalance(bundle->bond, &backer->revalidate_set);
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;
4439 struct ofproto_dpif *ofproto;
4442 dpif_flow_dump_start(&dump, backer->dpif);
4443 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
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 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4474 update_moving_averages(ofproto);
4478 /* Calculates and returns the number of milliseconds of idle time after which
4479 * subfacets should expire from the datapath. When a subfacet expires, we fold
4480 * its statistics into its facet, and when a facet's last subfacet expires, we
4481 * fold its statistic into its rule. */
4483 subfacet_max_idle(const struct ofproto_dpif *ofproto)
4486 * Idle time histogram.
4488 * Most of the time a switch has a relatively small number of subfacets.
4489 * When this is the case we might as well keep statistics for all of them
4490 * in userspace and to cache them in the kernel datapath for performance as
4493 * As the number of subfacets increases, the memory required to maintain
4494 * statistics about them in userspace and in the kernel becomes
4495 * significant. However, with a large number of subfacets it is likely
4496 * that only a few of them are "heavy hitters" that consume a large amount
4497 * of bandwidth. At this point, only heavy hitters are worth caching in
4498 * the kernel and maintaining in userspaces; other subfacets we can
4501 * The technique used to compute the idle time is to build a histogram with
4502 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
4503 * that is installed in the kernel gets dropped in the appropriate bucket.
4504 * After the histogram has been built, we compute the cutoff so that only
4505 * the most-recently-used 1% of subfacets (but at least
4506 * ofproto->up.flow_eviction_threshold flows) are kept cached. At least
4507 * the most-recently-used bucket of subfacets is kept, so actually an
4508 * arbitrary number of subfacets can be kept in any given expiration run
4509 * (though the next run will delete most of those unless they receive
4512 * This requires a second pass through the subfacets, in addition to the
4513 * pass made by update_stats(), because the former function never looks at
4514 * uninstallable subfacets.
4516 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
4517 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
4518 int buckets[N_BUCKETS] = { 0 };
4519 int total, subtotal, bucket;
4520 struct subfacet *subfacet;
4524 total = hmap_count(&ofproto->subfacets);
4525 if (total <= ofproto->up.flow_eviction_threshold) {
4526 return N_BUCKETS * BUCKET_WIDTH;
4529 /* Build histogram. */
4531 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
4532 long long int idle = now - subfacet->used;
4533 int bucket = (idle <= 0 ? 0
4534 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
4535 : (unsigned int) idle / BUCKET_WIDTH);
4539 /* Find the first bucket whose flows should be expired. */
4540 subtotal = bucket = 0;
4542 subtotal += buckets[bucket++];
4543 } while (bucket < N_BUCKETS &&
4544 subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100));
4546 if (VLOG_IS_DBG_ENABLED()) {
4550 ds_put_cstr(&s, "keep");
4551 for (i = 0; i < N_BUCKETS; i++) {
4553 ds_put_cstr(&s, ", drop");
4556 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
4559 VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s));
4563 return bucket * BUCKET_WIDTH;
4567 expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle)
4569 /* Cutoff time for most flows. */
4570 long long int normal_cutoff = time_msec() - dp_max_idle;
4572 /* We really want to keep flows for special protocols around, so use a more
4573 * conservative cutoff. */
4574 long long int special_cutoff = time_msec() - 10000;
4576 struct subfacet *subfacet, *next_subfacet;
4577 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
4581 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
4582 &ofproto->subfacets) {
4583 long long int cutoff;
4585 cutoff = (subfacet->facet->xout.slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP
4589 if (subfacet->used < cutoff) {
4590 if (subfacet->path != SF_NOT_INSTALLED) {
4591 batch[n_batch++] = subfacet;
4592 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
4593 subfacet_destroy_batch(ofproto, batch, n_batch);
4597 subfacet_destroy(subfacet);
4603 subfacet_destroy_batch(ofproto, batch, n_batch);
4607 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
4608 * then delete it entirely. */
4610 rule_expire(struct rule_dpif *rule)
4612 struct facet *facet, *next_facet;
4616 if (rule->up.pending) {
4617 /* We'll have to expire it later. */
4621 /* Has 'rule' expired? */
4623 if (rule->up.hard_timeout
4624 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
4625 reason = OFPRR_HARD_TIMEOUT;
4626 } else if (rule->up.idle_timeout
4627 && now > rule->up.used + rule->up.idle_timeout * 1000) {
4628 reason = OFPRR_IDLE_TIMEOUT;
4633 COVERAGE_INC(ofproto_dpif_expired);
4635 /* Update stats. (This is a no-op if the rule expired due to an idle
4636 * timeout, because that only happens when the rule has no facets left.) */
4637 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4638 facet_remove(facet);
4641 /* Get rid of the rule. */
4642 ofproto_rule_expire(&rule->up, reason);
4647 /* Creates and returns a new facet based on 'miss'.
4649 * The caller must already have determined that no facet with an identical
4650 * 'miss->flow' exists in 'miss->ofproto'.
4652 * 'hash' must be the return value of flow_hash(miss->flow, 0).
4654 * The facet will initially have no subfacets. The caller should create (at
4655 * least) one subfacet with subfacet_create(). */
4656 static struct facet *
4657 facet_create(const struct flow_miss *miss, uint32_t hash)
4659 struct ofproto_dpif *ofproto = miss->ofproto;
4660 struct xlate_in xin;
4661 struct facet *facet;
4663 facet = xzalloc(sizeof *facet);
4664 facet->used = time_msec();
4665 facet->flow = miss->flow;
4666 facet->initial_vals = miss->initial_vals;
4667 facet->rule = rule_dpif_lookup(ofproto, &facet->flow);
4668 facet->learn_rl = time_msec() + 500;
4670 hmap_insert(&ofproto->facets, &facet->hmap_node, hash);
4671 list_push_back(&facet->rule->facets, &facet->list_node);
4672 list_init(&facet->subfacets);
4673 netflow_flow_init(&facet->nf_flow);
4674 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
4676 xlate_in_init(&xin, ofproto, &facet->flow, &facet->initial_vals,
4677 facet->rule, 0, NULL);
4678 xin.may_learn = true;
4679 xlate_actions(&xin, &facet->xout);
4680 facet->nf_flow.output_iface = facet->xout.nf_output_iface;
4686 facet_free(struct facet *facet)
4689 xlate_out_uninit(&facet->xout);
4694 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
4695 * 'packet', which arrived on 'in_port'. */
4697 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
4698 const struct nlattr *odp_actions, size_t actions_len,
4699 struct ofpbuf *packet)
4701 struct odputil_keybuf keybuf;
4705 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4706 odp_flow_key_from_flow(&key, flow,
4707 ofp_port_to_odp_port(ofproto, flow->in_port));
4709 error = dpif_execute(ofproto->backer->dpif, key.data, key.size,
4710 odp_actions, actions_len, packet);
4714 /* Remove 'facet' from 'ofproto' and free up the associated memory:
4716 * - If 'facet' was installed in the datapath, uninstalls it and updates its
4717 * rule's statistics, via subfacet_uninstall().
4719 * - Removes 'facet' from its rule and from ofproto->facets.
4722 facet_remove(struct facet *facet)
4724 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4725 struct subfacet *subfacet, *next_subfacet;
4727 ovs_assert(!list_is_empty(&facet->subfacets));
4729 /* First uninstall all of the subfacets to get final statistics. */
4730 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4731 subfacet_uninstall(subfacet);
4734 /* Flush the final stats to the rule.
4736 * This might require us to have at least one subfacet around so that we
4737 * can use its actions for accounting in facet_account(), which is why we
4738 * have uninstalled but not yet destroyed the subfacets. */
4739 facet_flush_stats(facet);
4741 /* Now we're really all done so destroy everything. */
4742 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
4743 &facet->subfacets) {
4744 subfacet_destroy__(subfacet);
4746 hmap_remove(&ofproto->facets, &facet->hmap_node);
4747 list_remove(&facet->list_node);
4751 /* Feed information from 'facet' back into the learning table to keep it in
4752 * sync with what is actually flowing through the datapath. */
4754 facet_learn(struct facet *facet)
4756 long long int now = time_msec();
4758 if (!facet->xout.has_fin_timeout && now < facet->learn_rl) {
4762 facet->learn_rl = now + 500;
4764 if (!facet->xout.has_learn
4765 && !facet->xout.has_normal
4766 && (!facet->xout.has_fin_timeout
4767 || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
4771 facet_push_stats(facet, true);
4775 facet_account(struct facet *facet)
4777 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4778 const struct nlattr *a;
4783 if (!facet->xout.has_normal || !ofproto->has_bonded_bundles) {
4786 n_bytes = facet->byte_count - facet->accounted_bytes;
4788 /* This loop feeds byte counters to bond_account() for rebalancing to use
4789 * as a basis. We also need to track the actual VLAN on which the packet
4790 * is going to be sent to ensure that it matches the one passed to
4791 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
4794 * We use the actions from an arbitrary subfacet because they should all
4795 * be equally valid for our purpose. */
4796 vlan_tci = facet->flow.vlan_tci;
4797 NL_ATTR_FOR_EACH_UNSAFE (a, left, facet->xout.odp_actions.data,
4798 facet->xout.odp_actions.size) {
4799 const struct ovs_action_push_vlan *vlan;
4800 struct ofport_dpif *port;
4802 switch (nl_attr_type(a)) {
4803 case OVS_ACTION_ATTR_OUTPUT:
4804 port = get_odp_port(ofproto, nl_attr_get_u32(a));
4805 if (port && port->bundle && port->bundle->bond) {
4806 bond_account(port->bundle->bond, &facet->flow,
4807 vlan_tci_to_vid(vlan_tci), n_bytes);
4811 case OVS_ACTION_ATTR_POP_VLAN:
4812 vlan_tci = htons(0);
4815 case OVS_ACTION_ATTR_PUSH_VLAN:
4816 vlan = nl_attr_get(a);
4817 vlan_tci = vlan->vlan_tci;
4823 /* Returns true if the only action for 'facet' is to send to the controller.
4824 * (We don't report NetFlow expiration messages for such facets because they
4825 * are just part of the control logic for the network, not real traffic). */
4827 facet_is_controller_flow(struct facet *facet)
4830 const struct rule *rule = &facet->rule->up;
4831 const struct ofpact *ofpacts = rule->ofpacts;
4832 size_t ofpacts_len = rule->ofpacts_len;
4834 if (ofpacts_len > 0 &&
4835 ofpacts->type == OFPACT_CONTROLLER &&
4836 ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len)) {
4843 /* Folds all of 'facet''s statistics into its rule. Also updates the
4844 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
4845 * 'facet''s statistics in the datapath should have been zeroed and folded into
4846 * its packet and byte counts before this function is called. */
4848 facet_flush_stats(struct facet *facet)
4850 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4851 struct subfacet *subfacet;
4853 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4854 ovs_assert(!subfacet->dp_byte_count);
4855 ovs_assert(!subfacet->dp_packet_count);
4858 facet_push_stats(facet, false);
4859 if (facet->accounted_bytes < facet->byte_count) {
4860 facet_account(facet);
4861 facet->accounted_bytes = facet->byte_count;
4864 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
4865 struct ofexpired expired;
4866 expired.flow = facet->flow;
4867 expired.packet_count = facet->packet_count;
4868 expired.byte_count = facet->byte_count;
4869 expired.used = facet->used;
4870 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4873 /* Reset counters to prevent double counting if 'facet' ever gets
4875 facet_reset_counters(facet);
4877 netflow_flow_clear(&facet->nf_flow);
4878 facet->tcp_flags = 0;
4881 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
4882 * Returns it if found, otherwise a null pointer.
4884 * 'hash' must be the return value of flow_hash(flow, 0).
4886 * The returned facet might need revalidation; use facet_lookup_valid()
4887 * instead if that is important. */
4888 static struct facet *
4889 facet_find(struct ofproto_dpif *ofproto,
4890 const struct flow *flow, uint32_t hash)
4892 struct facet *facet;
4894 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, hash, &ofproto->facets) {
4895 if (flow_equal(flow, &facet->flow)) {
4903 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
4904 * Returns it if found, otherwise a null pointer.
4906 * 'hash' must be the return value of flow_hash(flow, 0).
4908 * The returned facet is guaranteed to be valid. */
4909 static struct facet *
4910 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow,
4913 struct facet *facet;
4915 facet = facet_find(ofproto, flow, hash);
4917 && (ofproto->backer->need_revalidate
4918 || tag_set_intersects(&ofproto->backer->revalidate_set,
4920 && !facet_revalidate(facet)) {
4928 facet_check_consistency(struct facet *facet)
4930 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
4932 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4934 struct xlate_out xout;
4935 struct xlate_in xin;
4937 struct rule_dpif *rule;
4940 /* Check the rule for consistency. */
4941 rule = rule_dpif_lookup(ofproto, &facet->flow);
4942 if (rule != facet->rule) {
4943 if (!VLOG_DROP_WARN(&rl)) {
4944 struct ds s = DS_EMPTY_INITIALIZER;
4946 flow_format(&s, &facet->flow);
4947 ds_put_format(&s, ": facet associated with wrong rule (was "
4948 "table=%"PRIu8",", facet->rule->up.table_id);
4949 cls_rule_format(&facet->rule->up.cr, &s);
4950 ds_put_format(&s, ") (should have been table=%"PRIu8",",
4952 cls_rule_format(&rule->up.cr, &s);
4953 ds_put_char(&s, ')');
4955 VLOG_WARN("%s", ds_cstr(&s));
4961 /* Check the datapath actions for consistency. */
4962 xlate_in_init(&xin, ofproto, &facet->flow, &facet->initial_vals, rule,
4964 xlate_actions(&xin, &xout);
4966 ok = ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)
4967 && facet->xout.slow == xout.slow;
4968 if (!ok && !VLOG_DROP_WARN(&rl)) {
4969 struct ds s = DS_EMPTY_INITIALIZER;
4971 flow_format(&s, &facet->flow);
4972 ds_put_cstr(&s, ": inconsistency in facet");
4974 if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4975 ds_put_cstr(&s, " (actions were: ");
4976 format_odp_actions(&s, facet->xout.odp_actions.data,
4977 facet->xout.odp_actions.size);
4978 ds_put_cstr(&s, ") (correct actions: ");
4979 format_odp_actions(&s, xout.odp_actions.data,
4980 xout.odp_actions.size);
4981 ds_put_char(&s, ')');
4984 if (facet->xout.slow != xout.slow) {
4985 ds_put_format(&s, " slow path incorrect. should be %d", xout.slow);
4988 VLOG_WARN("%s", ds_cstr(&s));
4991 xlate_out_uninit(&xout);
4996 /* Re-searches the classifier for 'facet':
4998 * - If the rule found is different from 'facet''s current rule, moves
4999 * 'facet' to the new rule and recompiles its actions.
5001 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
5002 * where it is and recompiles its actions anyway.
5004 * - If any of 'facet''s subfacets correspond to a new flow according to
5005 * ofproto_receive(), 'facet' is removed.
5007 * Returns true if 'facet' is still valid. False if 'facet' was removed. */
5009 facet_revalidate(struct facet *facet)
5011 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5012 struct rule_dpif *new_rule;
5013 struct subfacet *subfacet;
5014 struct xlate_out xout;
5015 struct xlate_in xin;
5017 COVERAGE_INC(facet_revalidate);
5019 /* Check that child subfacets still correspond to this facet. Tunnel
5020 * configuration changes could cause a subfacet's OpenFlow in_port to
5022 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
5023 struct ofproto_dpif *recv_ofproto;
5024 struct flow recv_flow;
5027 error = ofproto_receive(ofproto->backer, NULL, subfacet->key,
5028 subfacet->key_len, &recv_flow, NULL,
5029 &recv_ofproto, NULL, NULL);
5031 || recv_ofproto != ofproto
5032 || memcmp(&recv_flow, &facet->flow, sizeof recv_flow)) {
5033 facet_remove(facet);
5038 new_rule = rule_dpif_lookup(ofproto, &facet->flow);
5040 /* Calculate new datapath actions.
5042 * We do not modify any 'facet' state yet, because we might need to, e.g.,
5043 * emit a NetFlow expiration and, if so, we need to have the old state
5044 * around to properly compose it. */
5045 xlate_in_init(&xin, ofproto, &facet->flow, &facet->initial_vals, new_rule,
5047 xlate_actions(&xin, &xout);
5049 /* A facet's slow path reason should only change under dramatic
5050 * circumstances. Rather than try to update everything, it's simpler to
5051 * remove the facet and start over. */
5052 if (facet->xout.slow != xout.slow) {
5053 facet_remove(facet);
5054 xlate_out_uninit(&xout);
5058 if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
5059 LIST_FOR_EACH(subfacet, list_node, &facet->subfacets) {
5060 if (subfacet->path == SF_FAST_PATH) {
5061 struct dpif_flow_stats stats;
5063 subfacet_install(subfacet, &xout.odp_actions, &stats);
5064 subfacet_update_stats(subfacet, &stats);
5068 facet_flush_stats(facet);
5070 ofpbuf_clear(&facet->xout.odp_actions);
5071 ofpbuf_put(&facet->xout.odp_actions, xout.odp_actions.data,
5072 xout.odp_actions.size);
5075 /* Update 'facet' now that we've taken care of all the old state. */
5076 facet->xout.tags = xout.tags;
5077 facet->xout.slow = xout.slow;
5078 facet->xout.has_learn = xout.has_learn;
5079 facet->xout.has_normal = xout.has_normal;
5080 facet->xout.has_fin_timeout = xout.has_fin_timeout;
5081 facet->xout.nf_output_iface = xout.nf_output_iface;
5082 facet->xout.mirrors = xout.mirrors;
5083 facet->nf_flow.output_iface = facet->xout.nf_output_iface;
5085 if (facet->rule != new_rule) {
5086 COVERAGE_INC(facet_changed_rule);
5087 list_remove(&facet->list_node);
5088 list_push_back(&new_rule->facets, &facet->list_node);
5089 facet->rule = new_rule;
5090 facet->used = new_rule->up.created;
5091 facet->prev_used = facet->used;
5094 xlate_out_uninit(&xout);
5099 facet_reset_counters(struct facet *facet)
5101 facet->packet_count = 0;
5102 facet->byte_count = 0;
5103 facet->prev_packet_count = 0;
5104 facet->prev_byte_count = 0;
5105 facet->accounted_bytes = 0;
5109 facet_push_stats(struct facet *facet, bool may_learn)
5111 struct dpif_flow_stats stats;
5113 ovs_assert(facet->packet_count >= facet->prev_packet_count);
5114 ovs_assert(facet->byte_count >= facet->prev_byte_count);
5115 ovs_assert(facet->used >= facet->prev_used);
5117 stats.n_packets = facet->packet_count - facet->prev_packet_count;
5118 stats.n_bytes = facet->byte_count - facet->prev_byte_count;
5119 stats.used = facet->used;
5120 stats.tcp_flags = facet->tcp_flags;
5122 if (may_learn || stats.n_packets || facet->used > facet->prev_used) {
5123 struct ofproto_dpif *ofproto =
5124 ofproto_dpif_cast(facet->rule->up.ofproto);
5126 struct ofport_dpif *in_port;
5127 struct xlate_in xin;
5129 facet->prev_packet_count = facet->packet_count;
5130 facet->prev_byte_count = facet->byte_count;
5131 facet->prev_used = facet->used;
5133 in_port = get_ofp_port(ofproto, facet->flow.in_port);
5134 if (in_port && in_port->tnl_port) {
5135 netdev_vport_inc_rx(in_port->up.netdev, &stats);
5138 rule_credit_stats(facet->rule, &stats);
5139 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow,
5141 netflow_flow_update_flags(&facet->nf_flow, facet->tcp_flags);
5142 update_mirror_stats(ofproto, facet->xout.mirrors, stats.n_packets,
5145 xlate_in_init(&xin, ofproto, &facet->flow, &facet->initial_vals,
5146 facet->rule, stats.tcp_flags, NULL);
5147 xin.resubmit_stats = &stats;
5148 xin.may_learn = may_learn;
5149 xlate_actions_for_side_effects(&xin);
5154 push_all_stats__(bool run_fast)
5156 static long long int rl = LLONG_MIN;
5157 struct ofproto_dpif *ofproto;
5159 if (time_msec() < rl) {
5163 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5164 struct facet *facet;
5166 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
5167 facet_push_stats(facet, false);
5174 rl = time_msec() + 100;
5178 push_all_stats(void)
5180 push_all_stats__(true);
5184 rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats)
5186 rule->packet_count += stats->n_packets;
5187 rule->byte_count += stats->n_bytes;
5188 ofproto_rule_update_used(&rule->up, stats->used);
5193 static struct subfacet *
5194 subfacet_find(struct ofproto_dpif *ofproto,
5195 const struct nlattr *key, size_t key_len, uint32_t key_hash)
5197 struct subfacet *subfacet;
5199 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
5200 &ofproto->subfacets) {
5201 if (subfacet->key_len == key_len
5202 && !memcmp(key, subfacet->key, key_len)) {
5210 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
5211 * 'key_fitness', 'key', and 'key_len' members in 'miss'. Returns the
5212 * existing subfacet if there is one, otherwise creates and returns a
5214 static struct subfacet *
5215 subfacet_create(struct facet *facet, struct flow_miss *miss,
5218 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5219 enum odp_key_fitness key_fitness = miss->key_fitness;
5220 const struct nlattr *key = miss->key;
5221 size_t key_len = miss->key_len;
5223 struct subfacet *subfacet;
5225 key_hash = odp_flow_key_hash(key, key_len);
5227 if (list_is_empty(&facet->subfacets)) {
5228 subfacet = &facet->one_subfacet;
5230 subfacet = subfacet_find(ofproto, key, key_len, key_hash);
5232 if (subfacet->facet == facet) {
5236 /* This shouldn't happen. */
5237 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
5238 subfacet_destroy(subfacet);
5241 subfacet = xmalloc(sizeof *subfacet);
5244 hmap_insert(&ofproto->subfacets, &subfacet->hmap_node, key_hash);
5245 list_push_back(&facet->subfacets, &subfacet->list_node);
5246 subfacet->facet = facet;
5247 subfacet->key_fitness = key_fitness;
5248 subfacet->key = xmemdup(key, key_len);
5249 subfacet->key_len = key_len;
5250 subfacet->used = now;
5251 subfacet->created = now;
5252 subfacet->dp_packet_count = 0;
5253 subfacet->dp_byte_count = 0;
5254 subfacet->path = SF_NOT_INSTALLED;
5256 ofproto->subfacet_add_count++;
5260 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
5261 * its facet within 'ofproto', and frees it. */
5263 subfacet_destroy__(struct subfacet *subfacet)
5265 struct facet *facet = subfacet->facet;
5266 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5268 /* Update ofproto stats before uninstall the subfacet. */
5269 ofproto->subfacet_del_count++;
5271 subfacet_uninstall(subfacet);
5272 hmap_remove(&ofproto->subfacets, &subfacet->hmap_node);
5273 list_remove(&subfacet->list_node);
5274 free(subfacet->key);
5275 if (subfacet != &facet->one_subfacet) {
5280 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
5281 * last remaining subfacet in its facet destroys the facet too. */
5283 subfacet_destroy(struct subfacet *subfacet)
5285 struct facet *facet = subfacet->facet;
5287 if (list_is_singleton(&facet->subfacets)) {
5288 /* facet_remove() needs at least one subfacet (it will remove it). */
5289 facet_remove(facet);
5291 subfacet_destroy__(subfacet);
5296 subfacet_destroy_batch(struct ofproto_dpif *ofproto,
5297 struct subfacet **subfacets, int n)
5299 struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH];
5300 struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH];
5301 struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH];
5304 for (i = 0; i < n; i++) {
5305 ops[i].type = DPIF_OP_FLOW_DEL;
5306 ops[i].u.flow_del.key = subfacets[i]->key;
5307 ops[i].u.flow_del.key_len = subfacets[i]->key_len;
5308 ops[i].u.flow_del.stats = &stats[i];
5312 dpif_operate(ofproto->backer->dpif, opsp, n);
5313 for (i = 0; i < n; i++) {
5314 subfacet_reset_dp_stats(subfacets[i], &stats[i]);
5315 subfacets[i]->path = SF_NOT_INSTALLED;
5316 subfacet_destroy(subfacets[i]);
5321 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
5322 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
5323 * in the datapath will be zeroed and 'stats' will be updated with traffic new
5324 * since 'subfacet' was last updated.
5326 * Returns 0 if successful, otherwise a positive errno value. */
5328 subfacet_install(struct subfacet *subfacet, const struct ofpbuf *odp_actions,
5329 struct dpif_flow_stats *stats)
5331 struct facet *facet = subfacet->facet;
5332 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5333 enum subfacet_path path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
5334 const struct nlattr *actions = odp_actions->data;
5335 size_t actions_len = odp_actions->size;
5337 uint64_t slow_path_stub[128 / 8];
5338 enum dpif_flow_put_flags flags;
5341 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
5343 flags |= DPIF_FP_ZERO_STATS;
5346 if (path == SF_SLOW_PATH) {
5347 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
5348 slow_path_stub, sizeof slow_path_stub,
5349 &actions, &actions_len);
5352 ret = dpif_flow_put(ofproto->backer->dpif, flags, subfacet->key,
5353 subfacet->key_len, actions, actions_len, stats);
5356 subfacet_reset_dp_stats(subfacet, stats);
5360 subfacet->path = path;
5365 /* If 'subfacet' is installed in the datapath, uninstalls it. */
5367 subfacet_uninstall(struct subfacet *subfacet)
5369 if (subfacet->path != SF_NOT_INSTALLED) {
5370 struct rule_dpif *rule = subfacet->facet->rule;
5371 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5372 struct dpif_flow_stats stats;
5375 error = dpif_flow_del(ofproto->backer->dpif, subfacet->key,
5376 subfacet->key_len, &stats);
5377 subfacet_reset_dp_stats(subfacet, &stats);
5379 subfacet_update_stats(subfacet, &stats);
5381 subfacet->path = SF_NOT_INSTALLED;
5383 ovs_assert(subfacet->dp_packet_count == 0);
5384 ovs_assert(subfacet->dp_byte_count == 0);
5388 /* Resets 'subfacet''s datapath statistics counters. This should be called
5389 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
5390 * non-null, it should contain the statistics returned by dpif when 'subfacet'
5391 * was reset in the datapath. 'stats' will be modified to include only
5392 * statistics new since 'subfacet' was last updated. */
5394 subfacet_reset_dp_stats(struct subfacet *subfacet,
5395 struct dpif_flow_stats *stats)
5398 && subfacet->dp_packet_count <= stats->n_packets
5399 && subfacet->dp_byte_count <= stats->n_bytes) {
5400 stats->n_packets -= subfacet->dp_packet_count;
5401 stats->n_bytes -= subfacet->dp_byte_count;
5404 subfacet->dp_packet_count = 0;
5405 subfacet->dp_byte_count = 0;
5408 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
5410 * Because of the meaning of a subfacet's counters, it only makes sense to do
5411 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
5412 * represents a packet that was sent by hand or if it represents statistics
5413 * that have been cleared out of the datapath. */
5415 subfacet_update_stats(struct subfacet *subfacet,
5416 const struct dpif_flow_stats *stats)
5418 if (stats->n_packets || stats->used > subfacet->used) {
5419 struct facet *facet = subfacet->facet;
5421 subfacet->used = MAX(subfacet->used, stats->used);
5422 facet->used = MAX(facet->used, stats->used);
5423 facet->packet_count += stats->n_packets;
5424 facet->byte_count += stats->n_bytes;
5425 facet->tcp_flags |= stats->tcp_flags;
5431 static struct rule_dpif *
5432 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow)
5434 struct rule_dpif *rule;
5436 rule = rule_dpif_lookup__(ofproto, flow, 0);
5441 return rule_dpif_miss_rule(ofproto, flow);
5444 static struct rule_dpif *
5445 rule_dpif_lookup__(struct ofproto_dpif *ofproto, const struct flow *flow,
5448 struct cls_rule *cls_rule;
5449 struct classifier *cls;
5452 if (table_id >= N_TABLES) {
5456 cls = &ofproto->up.tables[table_id].cls;
5457 frag = (flow->nw_frag & FLOW_NW_FRAG_ANY) != 0;
5458 if (frag && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
5459 /* We must pretend that transport ports are unavailable. */
5460 struct flow ofpc_normal_flow = *flow;
5461 ofpc_normal_flow.tp_src = htons(0);
5462 ofpc_normal_flow.tp_dst = htons(0);
5463 cls_rule = classifier_lookup(cls, &ofpc_normal_flow);
5464 } else if (frag && ofproto->up.frag_handling == OFPC_FRAG_DROP) {
5465 cls_rule = &ofproto->drop_frags_rule->up.cr;
5467 cls_rule = classifier_lookup(cls, flow);
5469 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
5472 static struct rule_dpif *
5473 rule_dpif_miss_rule(struct ofproto_dpif *ofproto, const struct flow *flow)
5475 struct ofport_dpif *port;
5477 port = get_ofp_port(ofproto, flow->in_port);
5479 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, flow->in_port);
5480 return ofproto->miss_rule;
5483 if (port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN) {
5484 return ofproto->no_packet_in_rule;
5486 return ofproto->miss_rule;
5490 complete_operation(struct rule_dpif *rule)
5492 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5494 rule_invalidate(rule);
5496 struct dpif_completion *c = xmalloc(sizeof *c);
5497 c->op = rule->up.pending;
5498 list_push_back(&ofproto->completions, &c->list_node);
5500 ofoperation_complete(rule->up.pending, 0);
5504 static struct rule *
5507 struct rule_dpif *rule = xmalloc(sizeof *rule);
5512 rule_dealloc(struct rule *rule_)
5514 struct rule_dpif *rule = rule_dpif_cast(rule_);
5519 rule_construct(struct rule *rule_)
5521 struct rule_dpif *rule = rule_dpif_cast(rule_);
5522 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5523 struct rule_dpif *victim;
5526 rule->packet_count = 0;
5527 rule->byte_count = 0;
5529 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
5530 if (victim && !list_is_empty(&victim->facets)) {
5531 struct facet *facet;
5533 rule->facets = victim->facets;
5534 list_moved(&rule->facets);
5535 LIST_FOR_EACH (facet, list_node, &rule->facets) {
5536 /* XXX: We're only clearing our local counters here. It's possible
5537 * that quite a few packets are unaccounted for in the datapath
5538 * statistics. These will be accounted to the new rule instead of
5539 * cleared as required. This could be fixed by clearing out the
5540 * datapath statistics for this facet, but currently it doesn't
5542 facet_reset_counters(facet);
5546 /* Must avoid list_moved() in this case. */
5547 list_init(&rule->facets);
5550 table_id = rule->up.table_id;
5552 rule->tag = victim->tag;
5553 } else if (table_id == 0) {
5558 miniflow_expand(&rule->up.cr.match.flow, &flow);
5559 rule->tag = rule_calculate_tag(&flow, &rule->up.cr.match.mask,
5560 ofproto->tables[table_id].basis);
5563 complete_operation(rule);
5568 rule_destruct(struct rule *rule_)
5570 struct rule_dpif *rule = rule_dpif_cast(rule_);
5571 struct facet *facet, *next_facet;
5573 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
5574 facet_revalidate(facet);
5577 complete_operation(rule);
5581 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
5583 struct rule_dpif *rule = rule_dpif_cast(rule_);
5585 /* push_all_stats() can handle flow misses which, when using the learn
5586 * action, can cause rules to be added and deleted. This can corrupt our
5587 * caller's datastructures which assume that rule_get_stats() doesn't have
5588 * an impact on the flow table. To be safe, we disable miss handling. */
5589 push_all_stats__(false);
5591 /* Start from historical data for 'rule' itself that are no longer tracked
5592 * in facets. This counts, for example, facets that have expired. */
5593 *packets = rule->packet_count;
5594 *bytes = rule->byte_count;
5598 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
5599 struct ofpbuf *packet)
5601 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5602 struct initial_vals initial_vals;
5603 struct dpif_flow_stats stats;
5604 struct xlate_out xout;
5605 struct xlate_in xin;
5607 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
5608 rule_credit_stats(rule, &stats);
5610 initial_vals.vlan_tci = flow->vlan_tci;
5611 xlate_in_init(&xin, ofproto, flow, &initial_vals, rule, stats.tcp_flags,
5613 xin.resubmit_stats = &stats;
5614 xlate_actions(&xin, &xout);
5616 execute_odp_actions(ofproto, flow, xout.odp_actions.data,
5617 xout.odp_actions.size, packet);
5619 xlate_out_uninit(&xout);
5623 rule_execute(struct rule *rule, const struct flow *flow,
5624 struct ofpbuf *packet)
5626 rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
5627 ofpbuf_delete(packet);
5632 rule_modify_actions(struct rule *rule_)
5634 struct rule_dpif *rule = rule_dpif_cast(rule_);
5636 complete_operation(rule);
5639 /* Sends 'packet' out 'ofport'.
5640 * May modify 'packet'.
5641 * Returns 0 if successful, otherwise a positive errno value. */
5643 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
5645 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
5646 uint64_t odp_actions_stub[1024 / 8];
5647 struct ofpbuf key, odp_actions;
5648 struct dpif_flow_stats stats;
5649 struct odputil_keybuf keybuf;
5650 struct ofpact_output output;
5651 struct xlate_out xout;
5652 struct xlate_in xin;
5656 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5657 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5659 /* Use OFPP_NONE as the in_port to avoid special packet processing. */
5660 flow_extract(packet, 0, 0, NULL, OFPP_NONE, &flow);
5661 odp_flow_key_from_flow(&key, &flow, ofp_port_to_odp_port(ofproto,
5663 dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
5665 ofpact_init(&output.ofpact, OFPACT_OUTPUT, sizeof output);
5666 output.port = ofport->up.ofp_port;
5669 xlate_in_init(&xin, ofproto, &flow, NULL, NULL, 0, packet);
5670 xin.ofpacts_len = sizeof output;
5671 xin.ofpacts = &output.ofpact;
5672 xin.resubmit_stats = &stats;
5673 xlate_actions(&xin, &xout);
5675 error = dpif_execute(ofproto->backer->dpif,
5677 xout.odp_actions.data, xout.odp_actions.size,
5679 xlate_out_uninit(&xout);
5682 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %s (%s)",
5683 ofproto->up.name, netdev_get_name(ofport->up.netdev),
5687 ofproto->stats.tx_packets++;
5688 ofproto->stats.tx_bytes += packet->size;
5692 /* OpenFlow to datapath action translation. */
5694 static bool may_receive(const struct ofport_dpif *, struct xlate_ctx *);
5695 static void do_xlate_actions(const struct ofpact *, size_t ofpacts_len,
5696 struct xlate_ctx *);
5697 static void xlate_normal(struct xlate_ctx *);
5699 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
5700 * The action will state 'slow' as the reason that the action is in the slow
5701 * path. (This is purely informational: it allows a human viewing "ovs-dpctl
5702 * dump-flows" output to see why a flow is in the slow path.)
5704 * The 'stub_size' bytes in 'stub' will be used to store the action.
5705 * 'stub_size' must be large enough for the action.
5707 * The action and its size will be stored in '*actionsp' and '*actions_lenp',
5710 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
5711 enum slow_path_reason slow,
5712 uint64_t *stub, size_t stub_size,
5713 const struct nlattr **actionsp, size_t *actions_lenp)
5715 union user_action_cookie cookie;
5718 cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
5719 cookie.slow_path.unused = 0;
5720 cookie.slow_path.reason = slow;
5722 ofpbuf_use_stack(&buf, stub, stub_size);
5723 if (slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP | SLOW_STP)) {
5724 uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif, UINT32_MAX);
5725 odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
5727 put_userspace_action(ofproto, &buf, flow, &cookie,
5728 sizeof cookie.slow_path);
5730 *actionsp = buf.data;
5731 *actions_lenp = buf.size;
5735 put_userspace_action(const struct ofproto_dpif *ofproto,
5736 struct ofpbuf *odp_actions,
5737 const struct flow *flow,
5738 const union user_action_cookie *cookie,
5739 const size_t cookie_size)
5743 pid = dpif_port_get_pid(ofproto->backer->dpif,
5744 ofp_port_to_odp_port(ofproto, flow->in_port));
5746 return odp_put_userspace_action(pid, cookie, cookie_size, odp_actions);
5749 /* Compose SAMPLE action for sFlow or IPFIX. The given probability is
5750 * the number of packets out of UINT32_MAX to sample. The given
5751 * cookie is passed back in the callback for each sampled packet.
5754 compose_sample_action(const struct ofproto_dpif *ofproto,
5755 struct ofpbuf *odp_actions,
5756 const struct flow *flow,
5757 const uint32_t probability,
5758 const union user_action_cookie *cookie,
5759 const size_t cookie_size)
5761 size_t sample_offset, actions_offset;
5764 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
5766 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
5768 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
5769 cookie_offset = put_userspace_action(ofproto, odp_actions, flow, cookie,
5772 nl_msg_end_nested(odp_actions, actions_offset);
5773 nl_msg_end_nested(odp_actions, sample_offset);
5774 return cookie_offset;
5778 compose_sflow_cookie(const struct ofproto_dpif *ofproto,
5779 ovs_be16 vlan_tci, uint32_t odp_port,
5780 unsigned int n_outputs, union user_action_cookie *cookie)
5784 cookie->type = USER_ACTION_COOKIE_SFLOW;
5785 cookie->sflow.vlan_tci = vlan_tci;
5787 /* See http://www.sflow.org/sflow_version_5.txt (search for "Input/output
5788 * port information") for the interpretation of cookie->output. */
5789 switch (n_outputs) {
5791 /* 0x40000000 | 256 means "packet dropped for unknown reason". */
5792 cookie->sflow.output = 0x40000000 | 256;
5796 ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port);
5798 cookie->sflow.output = ifindex;
5803 /* 0x80000000 means "multiple output ports. */
5804 cookie->sflow.output = 0x80000000 | n_outputs;
5809 /* Compose SAMPLE action for sFlow bridge sampling. */
5811 compose_sflow_action(const struct ofproto_dpif *ofproto,
5812 struct ofpbuf *odp_actions,
5813 const struct flow *flow,
5816 uint32_t probability;
5817 union user_action_cookie cookie;
5819 if (!ofproto->sflow || flow->in_port == OFPP_NONE) {
5823 probability = dpif_sflow_get_probability(ofproto->sflow);
5824 compose_sflow_cookie(ofproto, htons(0), odp_port,
5825 odp_port == OVSP_NONE ? 0 : 1, &cookie);
5827 return compose_sample_action(ofproto, odp_actions, flow, probability,
5828 &cookie, sizeof cookie.sflow);
5832 compose_flow_sample_cookie(uint16_t probability, uint32_t collector_set_id,
5833 uint32_t obs_domain_id, uint32_t obs_point_id,
5834 union user_action_cookie *cookie)
5836 cookie->type = USER_ACTION_COOKIE_FLOW_SAMPLE;
5837 cookie->flow_sample.probability = probability;
5838 cookie->flow_sample.collector_set_id = collector_set_id;
5839 cookie->flow_sample.obs_domain_id = obs_domain_id;
5840 cookie->flow_sample.obs_point_id = obs_point_id;
5844 compose_ipfix_cookie(union user_action_cookie *cookie)
5846 cookie->type = USER_ACTION_COOKIE_IPFIX;
5849 /* Compose SAMPLE action for IPFIX bridge sampling. */
5851 compose_ipfix_action(const struct ofproto_dpif *ofproto,
5852 struct ofpbuf *odp_actions,
5853 const struct flow *flow)
5855 uint32_t probability;
5856 union user_action_cookie cookie;
5858 if (!ofproto->ipfix || flow->in_port == OFPP_NONE) {
5862 probability = dpif_ipfix_get_bridge_exporter_probability(ofproto->ipfix);
5863 compose_ipfix_cookie(&cookie);
5865 compose_sample_action(ofproto, odp_actions, flow, probability,
5866 &cookie, sizeof cookie.ipfix);
5869 /* SAMPLE action for sFlow must be first action in any given list of
5870 * actions. At this point we do not have all information required to
5871 * build it. So try to build sample action as complete as possible. */
5873 add_sflow_action(struct xlate_ctx *ctx)
5875 ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto,
5876 &ctx->xout->odp_actions,
5877 &ctx->xin->flow, OVSP_NONE);
5878 ctx->sflow_odp_port = 0;
5879 ctx->sflow_n_outputs = 0;
5882 /* SAMPLE action for IPFIX must be 1st or 2nd action in any given list
5883 * of actions, eventually after the SAMPLE action for sFlow. */
5885 add_ipfix_action(struct xlate_ctx *ctx)
5887 compose_ipfix_action(ctx->ofproto, &ctx->xout->odp_actions,
5891 /* Fix SAMPLE action according to data collected while composing ODP actions.
5892 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
5893 * USERSPACE action's user-cookie which is required for sflow. */
5895 fix_sflow_action(struct xlate_ctx *ctx)
5897 const struct flow *base = &ctx->base_flow;
5898 union user_action_cookie *cookie;
5900 if (!ctx->user_cookie_offset) {
5904 cookie = ofpbuf_at(&ctx->xout->odp_actions, ctx->user_cookie_offset,
5905 sizeof cookie->sflow);
5906 ovs_assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
5908 compose_sflow_cookie(ctx->ofproto, base->vlan_tci,
5909 ctx->sflow_odp_port, ctx->sflow_n_outputs, cookie);
5913 compose_output_action__(struct xlate_ctx *ctx, uint16_t ofp_port,
5916 const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
5917 ovs_be16 flow_vlan_tci;
5918 uint32_t flow_skb_mark;
5919 uint8_t flow_nw_tos;
5920 struct priority_to_dscp *pdscp;
5921 uint32_t out_port, odp_port;
5923 /* If 'struct flow' gets additional metadata, we'll need to zero it out
5924 * before traversing a patch port. */
5925 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 20);
5928 xlate_report(ctx, "Nonexistent output port");
5930 } else if (ofport->up.pp.config & OFPUTIL_PC_NO_FWD) {
5931 xlate_report(ctx, "OFPPC_NO_FWD set, skipping output");
5933 } else if (check_stp && !stp_forward_in_state(ofport->stp_state)) {
5934 xlate_report(ctx, "STP not in forwarding state, skipping output");
5938 if (netdev_vport_is_patch(ofport->up.netdev)) {
5939 struct ofport_dpif *peer = ofport_get_peer(ofport);
5940 struct flow old_flow = ctx->xin->flow;
5941 const struct ofproto_dpif *peer_ofproto;
5942 enum slow_path_reason special;
5943 struct ofport_dpif *in_port;
5946 xlate_report(ctx, "Nonexistent patch port peer");
5950 peer_ofproto = ofproto_dpif_cast(peer->up.ofproto);
5951 if (peer_ofproto->backer != ctx->ofproto->backer) {
5952 xlate_report(ctx, "Patch port peer on a different datapath");
5956 ctx->ofproto = ofproto_dpif_cast(peer->up.ofproto);
5957 ctx->xin->flow.in_port = peer->up.ofp_port;
5958 ctx->xin->flow.metadata = htonll(0);
5959 memset(&ctx->xin->flow.tunnel, 0, sizeof ctx->xin->flow.tunnel);
5960 memset(ctx->xin->flow.regs, 0, sizeof ctx->xin->flow.regs);
5962 in_port = get_ofp_port(ctx->ofproto, ctx->xin->flow.in_port);
5963 special = process_special(ctx->ofproto, &ctx->xin->flow, in_port,
5966 ctx->xout->slow = special;
5967 } else if (!in_port || may_receive(in_port, ctx)) {
5968 if (!in_port || stp_forward_in_state(in_port->stp_state)) {
5969 xlate_table_action(ctx, ctx->xin->flow.in_port, 0, true);
5971 /* Forwarding is disabled by STP. Let OFPP_NORMAL and the
5972 * learning action look at the packet, then drop it. */
5973 struct flow old_base_flow = ctx->base_flow;
5974 size_t old_size = ctx->xout->odp_actions.size;
5975 xlate_table_action(ctx, ctx->xin->flow.in_port, 0, true);
5976 ctx->base_flow = old_base_flow;
5977 ctx->xout->odp_actions.size = old_size;
5981 ctx->xin->flow = old_flow;
5982 ctx->ofproto = ofproto_dpif_cast(ofport->up.ofproto);
5984 if (ctx->xin->resubmit_stats) {
5985 netdev_vport_inc_tx(ofport->up.netdev, ctx->xin->resubmit_stats);
5986 netdev_vport_inc_rx(peer->up.netdev, ctx->xin->resubmit_stats);
5992 flow_vlan_tci = ctx->xin->flow.vlan_tci;
5993 flow_skb_mark = ctx->xin->flow.skb_mark;
5994 flow_nw_tos = ctx->xin->flow.nw_tos;
5996 pdscp = get_priority(ofport, ctx->xin->flow.skb_priority);
5998 ctx->xin->flow.nw_tos &= ~IP_DSCP_MASK;
5999 ctx->xin->flow.nw_tos |= pdscp->dscp;
6002 if (ofport->tnl_port) {
6003 /* Save tunnel metadata so that changes made due to
6004 * the Logical (tunnel) Port are not visible for any further
6005 * matches, while explicit set actions on tunnel metadata are.
6007 struct flow_tnl flow_tnl = ctx->xin->flow.tunnel;
6008 odp_port = tnl_port_send(ofport->tnl_port, &ctx->xin->flow);
6009 if (odp_port == OVSP_NONE) {
6010 xlate_report(ctx, "Tunneling decided against output");
6011 goto out; /* restore flow_nw_tos */
6013 if (ctx->xin->flow.tunnel.ip_dst == ctx->orig_tunnel_ip_dst) {
6014 xlate_report(ctx, "Not tunneling to our own address");
6015 goto out; /* restore flow_nw_tos */
6017 if (ctx->xin->resubmit_stats) {
6018 netdev_vport_inc_tx(ofport->up.netdev, ctx->xin->resubmit_stats);
6020 out_port = odp_port;
6021 commit_odp_tunnel_action(&ctx->xin->flow, &ctx->base_flow,
6022 &ctx->xout->odp_actions);
6023 ctx->xin->flow.tunnel = flow_tnl; /* Restore tunnel metadata */
6025 uint16_t vlandev_port;
6026 odp_port = ofport->odp_port;
6027 vlandev_port = vsp_realdev_to_vlandev(ctx->ofproto, ofp_port,
6028 ctx->xin->flow.vlan_tci);
6029 if (vlandev_port == ofp_port) {
6030 out_port = odp_port;
6032 out_port = ofp_port_to_odp_port(ctx->ofproto, vlandev_port);
6033 ctx->xin->flow.vlan_tci = htons(0);
6035 ctx->xin->flow.skb_mark &= ~IPSEC_MARK;
6037 commit_odp_actions(&ctx->xin->flow, &ctx->base_flow,
6038 &ctx->xout->odp_actions);
6039 nl_msg_put_u32(&ctx->xout->odp_actions, OVS_ACTION_ATTR_OUTPUT, out_port);
6041 ctx->sflow_odp_port = odp_port;
6042 ctx->sflow_n_outputs++;
6043 ctx->xout->nf_output_iface = ofp_port;
6046 ctx->xin->flow.vlan_tci = flow_vlan_tci;
6047 ctx->xin->flow.skb_mark = flow_skb_mark;
6049 ctx->xin->flow.nw_tos = flow_nw_tos;
6053 compose_output_action(struct xlate_ctx *ctx, uint16_t ofp_port)
6055 compose_output_action__(ctx, ofp_port, true);
6059 tag_the_flow(struct xlate_ctx *ctx, struct rule_dpif *rule)
6061 struct ofproto_dpif *ofproto = ctx->ofproto;
6062 uint8_t table_id = ctx->table_id;
6064 if (table_id > 0 && table_id < N_TABLES) {
6065 struct table_dpif *table = &ofproto->tables[table_id];
6066 if (table->other_table) {
6067 ctx->xout->tags |= (rule && rule->tag
6069 : rule_calculate_tag(&ctx->xin->flow,
6070 &table->other_table->mask,
6076 /* Common rule processing in one place to avoid duplicating code. */
6077 static struct rule_dpif *
6078 ctx_rule_hooks(struct xlate_ctx *ctx, struct rule_dpif *rule,
6081 if (ctx->xin->resubmit_hook) {
6082 ctx->xin->resubmit_hook(ctx, rule);
6084 if (rule == NULL && may_packet_in) {
6086 * check if table configuration flags
6087 * OFPTC_TABLE_MISS_CONTROLLER, default.
6088 * OFPTC_TABLE_MISS_CONTINUE,
6089 * OFPTC_TABLE_MISS_DROP
6090 * When OF1.0, OFPTC_TABLE_MISS_CONTINUE is used. What to do?
6092 rule = rule_dpif_miss_rule(ctx->ofproto, &ctx->xin->flow);
6094 if (rule && ctx->xin->resubmit_stats) {
6095 rule_credit_stats(rule, ctx->xin->resubmit_stats);
6101 xlate_table_action(struct xlate_ctx *ctx,
6102 uint16_t in_port, uint8_t table_id, bool may_packet_in)
6104 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
6105 struct rule_dpif *rule;
6106 uint16_t old_in_port = ctx->xin->flow.in_port;
6107 uint8_t old_table_id = ctx->table_id;
6109 ctx->table_id = table_id;
6111 /* Look up a flow with 'in_port' as the input port. */
6112 ctx->xin->flow.in_port = in_port;
6113 rule = rule_dpif_lookup__(ctx->ofproto, &ctx->xin->flow, table_id);
6115 tag_the_flow(ctx, rule);
6117 /* Restore the original input port. Otherwise OFPP_NORMAL and
6118 * OFPP_IN_PORT will have surprising behavior. */
6119 ctx->xin->flow.in_port = old_in_port;
6121 rule = ctx_rule_hooks(ctx, rule, may_packet_in);
6124 struct rule_dpif *old_rule = ctx->rule;
6128 do_xlate_actions(rule->up.ofpacts, rule->up.ofpacts_len, ctx);
6129 ctx->rule = old_rule;
6133 ctx->table_id = old_table_id;
6135 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
6137 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
6138 MAX_RESUBMIT_RECURSION);
6139 ctx->max_resubmit_trigger = true;
6144 xlate_ofpact_resubmit(struct xlate_ctx *ctx,
6145 const struct ofpact_resubmit *resubmit)
6150 in_port = resubmit->in_port;
6151 if (in_port == OFPP_IN_PORT) {
6152 in_port = ctx->xin->flow.in_port;
6155 table_id = resubmit->table_id;
6156 if (table_id == 255) {
6157 table_id = ctx->table_id;
6160 xlate_table_action(ctx, in_port, table_id, false);
6164 flood_packets(struct xlate_ctx *ctx, bool all)
6166 struct ofport_dpif *ofport;
6168 HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) {
6169 uint16_t ofp_port = ofport->up.ofp_port;
6171 if (ofp_port == ctx->xin->flow.in_port) {
6176 compose_output_action__(ctx, ofp_port, false);
6177 } else if (!(ofport->up.pp.config & OFPUTIL_PC_NO_FLOOD)) {
6178 compose_output_action(ctx, ofp_port);
6182 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
6186 execute_controller_action(struct xlate_ctx *ctx, int len,
6187 enum ofp_packet_in_reason reason,
6188 uint16_t controller_id)
6190 struct ofputil_packet_in pin;
6191 struct ofpbuf *packet;
6194 ovs_assert(!ctx->xout->slow || ctx->xout->slow == SLOW_CONTROLLER);
6195 ctx->xout->slow = SLOW_CONTROLLER;
6196 if (!ctx->xin->packet) {
6200 packet = ofpbuf_clone(ctx->xin->packet);
6202 key.skb_priority = 0;
6204 memset(&key.tunnel, 0, sizeof key.tunnel);
6206 commit_odp_actions(&ctx->xin->flow, &ctx->base_flow,
6207 &ctx->xout->odp_actions);
6209 odp_execute_actions(NULL, packet, &key, ctx->xout->odp_actions.data,
6210 ctx->xout->odp_actions.size, NULL, NULL);
6212 pin.packet = packet->data;
6213 pin.packet_len = packet->size;
6214 pin.reason = reason;
6215 pin.controller_id = controller_id;
6216 pin.table_id = ctx->table_id;
6217 pin.cookie = ctx->rule ? ctx->rule->up.flow_cookie : 0;
6220 flow_get_metadata(&ctx->xin->flow, &pin.fmd);
6222 connmgr_send_packet_in(ctx->ofproto->up.connmgr, &pin);
6223 ofpbuf_delete(packet);
6227 execute_mpls_push_action(struct xlate_ctx *ctx, ovs_be16 eth_type)
6229 ovs_assert(eth_type_mpls(eth_type));
6231 if (ctx->base_flow.mpls_depth) {
6232 ctx->xin->flow.mpls_lse &= ~htonl(MPLS_BOS_MASK);
6233 ctx->xin->flow.mpls_depth++;
6238 if (ctx->xin->flow.dl_type == htons(ETH_TYPE_IPV6)) {
6239 label = htonl(0x2); /* IPV6 Explicit Null. */
6241 label = htonl(0x0); /* IPV4 Explicit Null. */
6243 tc = (ctx->xin->flow.nw_tos & IP_DSCP_MASK) >> 2;
6244 ttl = ctx->xin->flow.nw_ttl ? ctx->xin->flow.nw_ttl : 0x40;
6245 ctx->xin->flow.mpls_lse = set_mpls_lse_values(ttl, tc, 1, label);
6246 ctx->xin->flow.mpls_depth = 1;
6248 ctx->xin->flow.dl_type = eth_type;
6252 execute_mpls_pop_action(struct xlate_ctx *ctx, ovs_be16 eth_type)
6254 ovs_assert(eth_type_mpls(ctx->xin->flow.dl_type));
6255 ovs_assert(!eth_type_mpls(eth_type));
6257 if (ctx->xin->flow.mpls_depth) {
6258 ctx->xin->flow.mpls_depth--;
6259 ctx->xin->flow.mpls_lse = htonl(0);
6260 if (!ctx->xin->flow.mpls_depth) {
6261 ctx->xin->flow.dl_type = eth_type;
6267 compose_dec_ttl(struct xlate_ctx *ctx, struct ofpact_cnt_ids *ids)
6269 if (ctx->xin->flow.dl_type != htons(ETH_TYPE_IP) &&
6270 ctx->xin->flow.dl_type != htons(ETH_TYPE_IPV6)) {
6274 if (ctx->xin->flow.nw_ttl > 1) {
6275 ctx->xin->flow.nw_ttl--;
6280 for (i = 0; i < ids->n_controllers; i++) {
6281 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL,
6285 /* Stop processing for current table. */
6291 execute_set_mpls_ttl_action(struct xlate_ctx *ctx, uint8_t ttl)
6293 if (!eth_type_mpls(ctx->xin->flow.dl_type)) {
6297 set_mpls_lse_ttl(&ctx->xin->flow.mpls_lse, ttl);
6302 execute_dec_mpls_ttl_action(struct xlate_ctx *ctx)
6304 uint8_t ttl = mpls_lse_to_ttl(ctx->xin->flow.mpls_lse);
6306 if (!eth_type_mpls(ctx->xin->flow.dl_type)) {
6312 set_mpls_lse_ttl(&ctx->xin->flow.mpls_lse, ttl);
6315 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL, 0);
6317 /* Stop processing for current table. */
6323 xlate_output_action(struct xlate_ctx *ctx,
6324 uint16_t port, uint16_t max_len, bool may_packet_in)
6326 uint16_t prev_nf_output_iface = ctx->xout->nf_output_iface;
6328 ctx->xout->nf_output_iface = NF_OUT_DROP;
6332 compose_output_action(ctx, ctx->xin->flow.in_port);
6335 xlate_table_action(ctx, ctx->xin->flow.in_port, 0, may_packet_in);
6341 flood_packets(ctx, false);
6344 flood_packets(ctx, true);
6346 case OFPP_CONTROLLER:
6347 execute_controller_action(ctx, max_len, OFPR_ACTION, 0);
6353 if (port != ctx->xin->flow.in_port) {
6354 compose_output_action(ctx, port);
6356 xlate_report(ctx, "skipping output to input port");
6361 if (prev_nf_output_iface == NF_OUT_FLOOD) {
6362 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
6363 } else if (ctx->xout->nf_output_iface == NF_OUT_DROP) {
6364 ctx->xout->nf_output_iface = prev_nf_output_iface;
6365 } else if (prev_nf_output_iface != NF_OUT_DROP &&
6366 ctx->xout->nf_output_iface != NF_OUT_FLOOD) {
6367 ctx->xout->nf_output_iface = NF_OUT_MULTI;
6372 xlate_output_reg_action(struct xlate_ctx *ctx,
6373 const struct ofpact_output_reg *or)
6375 uint64_t port = mf_get_subfield(&or->src, &ctx->xin->flow);
6376 if (port <= UINT16_MAX) {
6377 xlate_output_action(ctx, port, or->max_len, false);
6382 xlate_enqueue_action(struct xlate_ctx *ctx,
6383 const struct ofpact_enqueue *enqueue)
6385 uint16_t ofp_port = enqueue->port;
6386 uint32_t queue_id = enqueue->queue;
6387 uint32_t flow_priority, priority;
6390 /* Translate queue to priority. */
6391 error = dpif_queue_to_priority(ctx->ofproto->backer->dpif,
6392 queue_id, &priority);
6394 /* Fall back to ordinary output action. */
6395 xlate_output_action(ctx, enqueue->port, 0, false);
6399 /* Check output port. */
6400 if (ofp_port == OFPP_IN_PORT) {
6401 ofp_port = ctx->xin->flow.in_port;
6402 } else if (ofp_port == ctx->xin->flow.in_port) {
6406 /* Add datapath actions. */
6407 flow_priority = ctx->xin->flow.skb_priority;
6408 ctx->xin->flow.skb_priority = priority;
6409 compose_output_action(ctx, ofp_port);
6410 ctx->xin->flow.skb_priority = flow_priority;
6412 /* Update NetFlow output port. */
6413 if (ctx->xout->nf_output_iface == NF_OUT_DROP) {
6414 ctx->xout->nf_output_iface = ofp_port;
6415 } else if (ctx->xout->nf_output_iface != NF_OUT_FLOOD) {
6416 ctx->xout->nf_output_iface = NF_OUT_MULTI;
6421 xlate_set_queue_action(struct xlate_ctx *ctx, uint32_t queue_id)
6423 uint32_t skb_priority;
6425 if (!dpif_queue_to_priority(ctx->ofproto->backer->dpif,
6426 queue_id, &skb_priority)) {
6427 ctx->xin->flow.skb_priority = skb_priority;
6429 /* Couldn't translate queue to a priority. Nothing to do. A warning
6430 * has already been logged. */
6435 slave_enabled_cb(uint16_t ofp_port, void *ofproto_)
6437 struct ofproto_dpif *ofproto = ofproto_;
6438 struct ofport_dpif *port;
6448 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
6451 port = get_ofp_port(ofproto, ofp_port);
6452 return port ? port->may_enable : false;
6457 xlate_bundle_action(struct xlate_ctx *ctx,
6458 const struct ofpact_bundle *bundle)
6462 port = bundle_execute(bundle, &ctx->xin->flow, slave_enabled_cb,
6464 if (bundle->dst.field) {
6465 nxm_reg_load(&bundle->dst, port, &ctx->xin->flow);
6467 xlate_output_action(ctx, port, 0, false);
6472 xlate_learn_action(struct xlate_ctx *ctx,
6473 const struct ofpact_learn *learn)
6475 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
6476 struct ofputil_flow_mod fm;
6477 uint64_t ofpacts_stub[1024 / 8];
6478 struct ofpbuf ofpacts;
6481 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
6482 learn_execute(learn, &ctx->xin->flow, &fm, &ofpacts);
6484 error = ofproto_flow_mod(&ctx->ofproto->up, &fm);
6485 if (error && !VLOG_DROP_WARN(&rl)) {
6486 VLOG_WARN("learning action failed to modify flow table (%s)",
6487 ofperr_get_name(error));
6490 ofpbuf_uninit(&ofpacts);
6493 /* Reduces '*timeout' to no more than 'max'. A value of zero in either case
6494 * means "infinite". */
6496 reduce_timeout(uint16_t max, uint16_t *timeout)
6498 if (max && (!*timeout || *timeout > max)) {
6504 xlate_fin_timeout(struct xlate_ctx *ctx,
6505 const struct ofpact_fin_timeout *oft)
6507 if (ctx->xin->tcp_flags & (TCP_FIN | TCP_RST) && ctx->rule) {
6508 struct rule_dpif *rule = ctx->rule;
6510 reduce_timeout(oft->fin_idle_timeout, &rule->up.idle_timeout);
6511 reduce_timeout(oft->fin_hard_timeout, &rule->up.hard_timeout);
6516 xlate_sample_action(struct xlate_ctx *ctx,
6517 const struct ofpact_sample *os)
6519 union user_action_cookie cookie;
6520 /* Scale the probability from 16-bit to 32-bit while representing
6521 * the same percentage. */
6522 uint32_t probability = (os->probability << 16) | os->probability;
6524 commit_odp_actions(&ctx->xin->flow, &ctx->base_flow,
6525 &ctx->xout->odp_actions);
6527 compose_flow_sample_cookie(os->probability, os->collector_set_id,
6528 os->obs_domain_id, os->obs_point_id, &cookie);
6529 compose_sample_action(ctx->ofproto, &ctx->xout->odp_actions, &ctx->xin->flow,
6530 probability, &cookie, sizeof cookie.flow_sample);
6534 may_receive(const struct ofport_dpif *port, struct xlate_ctx *ctx)
6536 if (port->up.pp.config & (eth_addr_equals(ctx->xin->flow.dl_dst,
6538 ? OFPUTIL_PC_NO_RECV_STP
6539 : OFPUTIL_PC_NO_RECV)) {
6543 /* Only drop packets here if both forwarding and learning are
6544 * disabled. If just learning is enabled, we need to have
6545 * OFPP_NORMAL and the learning action have a look at the packet
6546 * before we can drop it. */
6547 if (!stp_forward_in_state(port->stp_state)
6548 && !stp_learn_in_state(port->stp_state)) {
6556 tunnel_ecn_ok(struct xlate_ctx *ctx)
6558 if (is_ip_any(&ctx->base_flow)
6559 && (ctx->xin->flow.tunnel.ip_tos & IP_ECN_MASK) == IP_ECN_CE) {
6560 if ((ctx->base_flow.nw_tos & IP_ECN_MASK) == IP_ECN_NOT_ECT) {
6561 VLOG_WARN_RL(&rl, "dropping tunnel packet marked ECN CE"
6562 " but is not ECN capable");
6565 /* Set the ECN CE value in the tunneled packet. */
6566 ctx->xin->flow.nw_tos |= IP_ECN_CE;
6574 do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len,
6575 struct xlate_ctx *ctx)
6577 bool was_evictable = true;
6578 const struct ofpact *a;
6581 /* Don't let the rule we're working on get evicted underneath us. */
6582 was_evictable = ctx->rule->up.evictable;
6583 ctx->rule->up.evictable = false;
6586 do_xlate_actions_again:
6587 OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) {
6588 struct ofpact_controller *controller;
6589 const struct ofpact_metadata *metadata;
6597 xlate_output_action(ctx, ofpact_get_OUTPUT(a)->port,
6598 ofpact_get_OUTPUT(a)->max_len, true);
6601 case OFPACT_CONTROLLER:
6602 controller = ofpact_get_CONTROLLER(a);
6603 execute_controller_action(ctx, controller->max_len,
6605 controller->controller_id);
6608 case OFPACT_ENQUEUE:
6609 xlate_enqueue_action(ctx, ofpact_get_ENQUEUE(a));
6612 case OFPACT_SET_VLAN_VID:
6613 ctx->xin->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
6614 ctx->xin->flow.vlan_tci |=
6615 (htons(ofpact_get_SET_VLAN_VID(a)->vlan_vid)
6619 case OFPACT_SET_VLAN_PCP:
6620 ctx->xin->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
6621 ctx->xin->flow.vlan_tci |=
6622 htons((ofpact_get_SET_VLAN_PCP(a)->vlan_pcp << VLAN_PCP_SHIFT)
6626 case OFPACT_STRIP_VLAN:
6627 ctx->xin->flow.vlan_tci = htons(0);
6630 case OFPACT_PUSH_VLAN:
6631 /* XXX 802.1AD(QinQ) */
6632 ctx->xin->flow.vlan_tci = htons(VLAN_CFI);
6635 case OFPACT_SET_ETH_SRC:
6636 memcpy(ctx->xin->flow.dl_src, ofpact_get_SET_ETH_SRC(a)->mac,
6640 case OFPACT_SET_ETH_DST:
6641 memcpy(ctx->xin->flow.dl_dst, ofpact_get_SET_ETH_DST(a)->mac,
6645 case OFPACT_SET_IPV4_SRC:
6646 if (ctx->xin->flow.dl_type == htons(ETH_TYPE_IP)) {
6647 ctx->xin->flow.nw_src = ofpact_get_SET_IPV4_SRC(a)->ipv4;
6651 case OFPACT_SET_IPV4_DST:
6652 if (ctx->xin->flow.dl_type == htons(ETH_TYPE_IP)) {
6653 ctx->xin->flow.nw_dst = ofpact_get_SET_IPV4_DST(a)->ipv4;
6657 case OFPACT_SET_IPV4_DSCP:
6658 /* OpenFlow 1.0 only supports IPv4. */
6659 if (ctx->xin->flow.dl_type == htons(ETH_TYPE_IP)) {
6660 ctx->xin->flow.nw_tos &= ~IP_DSCP_MASK;
6661 ctx->xin->flow.nw_tos |= ofpact_get_SET_IPV4_DSCP(a)->dscp;
6665 case OFPACT_SET_L4_SRC_PORT:
6666 if (is_ip_any(&ctx->xin->flow)) {
6667 ctx->xin->flow.tp_src =
6668 htons(ofpact_get_SET_L4_SRC_PORT(a)->port);
6672 case OFPACT_SET_L4_DST_PORT:
6673 if (is_ip_any(&ctx->xin->flow)) {
6674 ctx->xin->flow.tp_dst =
6675 htons(ofpact_get_SET_L4_DST_PORT(a)->port);
6679 case OFPACT_RESUBMIT:
6680 xlate_ofpact_resubmit(ctx, ofpact_get_RESUBMIT(a));
6683 case OFPACT_SET_TUNNEL:
6684 ctx->xin->flow.tunnel.tun_id =
6685 htonll(ofpact_get_SET_TUNNEL(a)->tun_id);
6688 case OFPACT_SET_QUEUE:
6689 xlate_set_queue_action(ctx, ofpact_get_SET_QUEUE(a)->queue_id);
6692 case OFPACT_POP_QUEUE:
6693 ctx->xin->flow.skb_priority = ctx->orig_skb_priority;
6696 case OFPACT_REG_MOVE:
6697 nxm_execute_reg_move(ofpact_get_REG_MOVE(a), &ctx->xin->flow);
6700 case OFPACT_REG_LOAD:
6701 nxm_execute_reg_load(ofpact_get_REG_LOAD(a), &ctx->xin->flow);
6704 case OFPACT_STACK_PUSH:
6705 nxm_execute_stack_push(ofpact_get_STACK_PUSH(a), &ctx->xin->flow,
6709 case OFPACT_STACK_POP:
6710 nxm_execute_stack_pop(ofpact_get_STACK_POP(a), &ctx->xin->flow,
6714 case OFPACT_PUSH_MPLS:
6715 execute_mpls_push_action(ctx, ofpact_get_PUSH_MPLS(a)->ethertype);
6718 case OFPACT_POP_MPLS:
6719 execute_mpls_pop_action(ctx, ofpact_get_POP_MPLS(a)->ethertype);
6722 case OFPACT_SET_MPLS_TTL:
6723 if (execute_set_mpls_ttl_action(ctx,
6724 ofpact_get_SET_MPLS_TTL(a)->ttl)) {
6729 case OFPACT_DEC_MPLS_TTL:
6730 if (execute_dec_mpls_ttl_action(ctx)) {
6735 case OFPACT_DEC_TTL:
6736 if (compose_dec_ttl(ctx, ofpact_get_DEC_TTL(a))) {
6742 /* Nothing to do. */
6745 case OFPACT_MULTIPATH:
6746 multipath_execute(ofpact_get_MULTIPATH(a), &ctx->xin->flow);
6750 ctx->ofproto->has_bundle_action = true;
6751 xlate_bundle_action(ctx, ofpact_get_BUNDLE(a));
6754 case OFPACT_OUTPUT_REG:
6755 xlate_output_reg_action(ctx, ofpact_get_OUTPUT_REG(a));
6759 ctx->xout->has_learn = true;
6760 if (ctx->xin->may_learn) {
6761 xlate_learn_action(ctx, ofpact_get_LEARN(a));
6769 case OFPACT_FIN_TIMEOUT:
6770 ctx->xout->has_fin_timeout = true;
6771 xlate_fin_timeout(ctx, ofpact_get_FIN_TIMEOUT(a));
6774 case OFPACT_CLEAR_ACTIONS:
6776 * Nothing to do because writa-actions is not supported for now.
6777 * When writa-actions is supported, clear-actions also must
6778 * be supported at the same time.
6782 case OFPACT_WRITE_METADATA:
6783 metadata = ofpact_get_WRITE_METADATA(a);
6784 ctx->xin->flow.metadata &= ~metadata->mask;
6785 ctx->xin->flow.metadata |= metadata->metadata & metadata->mask;
6788 case OFPACT_GOTO_TABLE: {
6789 /* It is assumed that goto-table is the last action. */
6790 struct ofpact_goto_table *ogt = ofpact_get_GOTO_TABLE(a);
6791 struct rule_dpif *rule;
6793 ovs_assert(ctx->table_id < ogt->table_id);
6795 ctx->table_id = ogt->table_id;
6797 /* Look up a flow from the new table. */
6798 rule = rule_dpif_lookup__(ctx->ofproto, &ctx->xin->flow, ctx->table_id);
6800 tag_the_flow(ctx, rule);
6802 rule = ctx_rule_hooks(ctx, rule, true);
6806 ctx->rule->up.evictable = was_evictable;
6809 was_evictable = rule->up.evictable;
6810 rule->up.evictable = false;
6812 /* Tail recursion removal. */
6813 ofpacts = rule->up.ofpacts;
6814 ofpacts_len = rule->up.ofpacts_len;
6815 goto do_xlate_actions_again;
6821 xlate_sample_action(ctx, ofpact_get_SAMPLE(a));
6828 ctx->rule->up.evictable = was_evictable;
6833 xlate_in_init(struct xlate_in *xin, struct ofproto_dpif *ofproto,
6834 const struct flow *flow,
6835 const struct initial_vals *initial_vals,
6836 struct rule_dpif *rule, uint8_t tcp_flags,
6837 const struct ofpbuf *packet)
6839 xin->ofproto = ofproto;
6841 xin->packet = packet;
6842 xin->may_learn = packet != NULL;
6844 xin->ofpacts = NULL;
6845 xin->ofpacts_len = 0;
6846 xin->tcp_flags = tcp_flags;
6847 xin->resubmit_hook = NULL;
6848 xin->report_hook = NULL;
6849 xin->resubmit_stats = NULL;
6852 xin->initial_vals = *initial_vals;
6854 xin->initial_vals.vlan_tci = xin->flow.vlan_tci;
6859 xlate_out_uninit(struct xlate_out *xout)
6862 ofpbuf_uninit(&xout->odp_actions);
6866 /* Translates the 'ofpacts_len' bytes of "struct ofpacts" starting at 'ofpacts'
6867 * into datapath actions in 'odp_actions', using 'ctx'. */
6869 xlate_actions(struct xlate_in *xin, struct xlate_out *xout)
6871 /* Normally false. Set to true if we ever hit MAX_RESUBMIT_RECURSION, so
6872 * that in the future we always keep a copy of the original flow for
6873 * tracing purposes. */
6874 static bool hit_resubmit_limit;
6876 enum slow_path_reason special;
6877 const struct ofpact *ofpacts;
6878 struct ofport_dpif *in_port;
6879 struct flow orig_flow;
6880 struct xlate_ctx ctx;
6883 COVERAGE_INC(ofproto_dpif_xlate);
6885 /* Flow initialization rules:
6886 * - 'base_flow' must match the kernel's view of the packet at the
6887 * time that action processing starts. 'flow' represents any
6888 * transformations we wish to make through actions.
6889 * - By default 'base_flow' and 'flow' are the same since the input
6890 * packet matches the output before any actions are applied.
6891 * - When using VLAN splinters, 'base_flow''s VLAN is set to the value
6892 * of the received packet as seen by the kernel. If we later output
6893 * to another device without any modifications this will cause us to
6894 * insert a new tag since the original one was stripped off by the
6896 * - Tunnel metadata as received is retained in 'flow'. This allows
6897 * tunnel metadata matching also in later tables.
6898 * Since a kernel action for setting the tunnel metadata will only be
6899 * generated with actual tunnel output, changing the tunnel metadata
6900 * values in 'flow' (such as tun_id) will only have effect with a later
6901 * tunnel output action.
6902 * - Tunnel 'base_flow' is completely cleared since that is what the
6903 * kernel does. If we wish to maintain the original values an action
6904 * needs to be generated. */
6909 ctx.ofproto = xin->ofproto;
6910 ctx.rule = xin->rule;
6912 ctx.base_flow = ctx.xin->flow;
6913 ctx.base_flow.vlan_tci = xin->initial_vals.vlan_tci;
6914 memset(&ctx.base_flow.tunnel, 0, sizeof ctx.base_flow.tunnel);
6915 ctx.orig_tunnel_ip_dst = ctx.xin->flow.tunnel.ip_dst;
6919 ctx.xout->has_learn = false;
6920 ctx.xout->has_normal = false;
6921 ctx.xout->has_fin_timeout = false;
6922 ctx.xout->nf_output_iface = NF_OUT_DROP;
6923 ctx.xout->mirrors = 0;
6925 ofpbuf_use_stub(&ctx.xout->odp_actions, ctx.xout->odp_actions_stub,
6926 sizeof ctx.xout->odp_actions_stub);
6927 ofpbuf_reserve(&ctx.xout->odp_actions, NL_A_U32_SIZE);
6930 ctx.max_resubmit_trigger = false;
6931 ctx.orig_skb_priority = ctx.xin->flow.skb_priority;
6936 ofpacts = xin->ofpacts;
6937 ofpacts_len = xin->ofpacts_len;
6938 } else if (xin->rule) {
6939 ofpacts = xin->rule->up.ofpacts;
6940 ofpacts_len = xin->rule->up.ofpacts_len;
6945 ofpbuf_use_stub(&ctx.stack, ctx.init_stack, sizeof ctx.init_stack);
6947 if (ctx.ofproto->has_mirrors || hit_resubmit_limit) {
6948 /* Do this conditionally because the copy is expensive enough that it
6949 * shows up in profiles. */
6950 orig_flow = ctx.xin->flow;
6953 if (ctx.xin->flow.nw_frag & FLOW_NW_FRAG_ANY) {
6954 switch (ctx.ofproto->up.frag_handling) {
6955 case OFPC_FRAG_NORMAL:
6956 /* We must pretend that transport ports are unavailable. */
6957 ctx.xin->flow.tp_src = ctx.base_flow.tp_src = htons(0);
6958 ctx.xin->flow.tp_dst = ctx.base_flow.tp_dst = htons(0);
6961 case OFPC_FRAG_DROP:
6964 case OFPC_FRAG_REASM:
6967 case OFPC_FRAG_NX_MATCH:
6968 /* Nothing to do. */
6971 case OFPC_INVALID_TTL_TO_CONTROLLER:
6976 in_port = get_ofp_port(ctx.ofproto, ctx.xin->flow.in_port);
6977 special = process_special(ctx.ofproto, &ctx.xin->flow, in_port,
6980 ctx.xout->slow = special;
6982 static struct vlog_rate_limit trace_rl = VLOG_RATE_LIMIT_INIT(1, 1);
6983 struct initial_vals initial_vals;
6984 size_t sample_actions_len;
6985 uint32_t local_odp_port;
6987 initial_vals.vlan_tci = ctx.base_flow.vlan_tci;
6989 add_sflow_action(&ctx);
6990 add_ipfix_action(&ctx);
6991 sample_actions_len = ctx.xout->odp_actions.size;
6993 if (tunnel_ecn_ok(&ctx) && (!in_port || may_receive(in_port, &ctx))) {
6994 do_xlate_actions(ofpacts, ofpacts_len, &ctx);
6996 /* We've let OFPP_NORMAL and the learning action look at the
6997 * packet, so drop it now if forwarding is disabled. */
6998 if (in_port && !stp_forward_in_state(in_port->stp_state)) {
6999 ctx.xout->odp_actions.size = sample_actions_len;
7003 if (ctx.max_resubmit_trigger && !ctx.xin->resubmit_hook) {
7004 if (!hit_resubmit_limit) {
7005 /* We didn't record the original flow. Make sure we do from
7007 hit_resubmit_limit = true;
7008 } else if (!VLOG_DROP_ERR(&trace_rl)) {
7009 struct ds ds = DS_EMPTY_INITIALIZER;
7011 ofproto_trace(ctx.ofproto, &orig_flow, ctx.xin->packet,
7012 &initial_vals, &ds);
7013 VLOG_ERR("Trace triggered by excessive resubmit "
7014 "recursion:\n%s", ds_cstr(&ds));
7019 local_odp_port = ofp_port_to_odp_port(ctx.ofproto, OFPP_LOCAL);
7020 if (!connmgr_must_output_local(ctx.ofproto->up.connmgr, &ctx.xin->flow,
7022 ctx.xout->odp_actions.data,
7023 ctx.xout->odp_actions.size)) {
7024 compose_output_action(&ctx, OFPP_LOCAL);
7026 if (ctx.ofproto->has_mirrors) {
7027 add_mirror_actions(&ctx, &orig_flow);
7029 fix_sflow_action(&ctx);
7032 ofpbuf_uninit(&ctx.stack);
7035 /* Translates the 'ofpacts_len' bytes of "struct ofpact"s starting at 'ofpacts'
7036 * into datapath actions, using 'ctx', and discards the datapath actions. */
7038 xlate_actions_for_side_effects(struct xlate_in *xin)
7040 struct xlate_out xout;
7042 xlate_actions(xin, &xout);
7043 xlate_out_uninit(&xout);
7047 xlate_report(struct xlate_ctx *ctx, const char *s)
7049 if (ctx->xin->report_hook) {
7050 ctx->xin->report_hook(ctx, s);
7054 /* OFPP_NORMAL implementation. */
7056 static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *);
7058 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
7059 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle',
7060 * the bundle on which the packet was received, returns the VLAN to which the
7063 * Both 'vid' and the return value are in the range 0...4095. */
7065 input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid)
7067 switch (in_bundle->vlan_mode) {
7068 case PORT_VLAN_ACCESS:
7069 return in_bundle->vlan;
7072 case PORT_VLAN_TRUNK:
7075 case PORT_VLAN_NATIVE_UNTAGGED:
7076 case PORT_VLAN_NATIVE_TAGGED:
7077 return vid ? vid : in_bundle->vlan;
7084 /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'.
7085 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
7088 * 'vid' should be the VID obtained from the 802.1Q header that was received as
7089 * part of a packet (specify 0 if there was no 802.1Q header), in the range
7092 input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn)
7094 /* Allow any VID on the OFPP_NONE port. */
7095 if (in_bundle == &ofpp_none_bundle) {
7099 switch (in_bundle->vlan_mode) {
7100 case PORT_VLAN_ACCESS:
7103 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7104 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
7105 "packet received on port %s configured as VLAN "
7106 "%"PRIu16" access port",
7107 in_bundle->ofproto->up.name, vid,
7108 in_bundle->name, in_bundle->vlan);
7114 case PORT_VLAN_NATIVE_UNTAGGED:
7115 case PORT_VLAN_NATIVE_TAGGED:
7117 /* Port must always carry its native VLAN. */
7121 case PORT_VLAN_TRUNK:
7122 if (!ofbundle_includes_vlan(in_bundle, vid)) {
7124 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7125 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet "
7126 "received on port %s not configured for trunking "
7128 in_bundle->ofproto->up.name, vid,
7129 in_bundle->name, vid);
7141 /* Given 'vlan', the VLAN that a packet belongs to, and
7142 * 'out_bundle', a bundle on which the packet is to be output, returns the VID
7143 * that should be included in the 802.1Q header. (If the return value is 0,
7144 * then the 802.1Q header should only be included in the packet if there is a
7147 * Both 'vlan' and the return value are in the range 0...4095. */
7149 output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan)
7151 switch (out_bundle->vlan_mode) {
7152 case PORT_VLAN_ACCESS:
7155 case PORT_VLAN_TRUNK:
7156 case PORT_VLAN_NATIVE_TAGGED:
7159 case PORT_VLAN_NATIVE_UNTAGGED:
7160 return vlan == out_bundle->vlan ? 0 : vlan;
7168 output_normal(struct xlate_ctx *ctx, const struct ofbundle *out_bundle,
7171 struct ofport_dpif *port;
7173 ovs_be16 tci, old_tci;
7175 vid = output_vlan_to_vid(out_bundle, vlan);
7176 if (!out_bundle->bond) {
7177 port = ofbundle_get_a_port(out_bundle);
7179 port = bond_choose_output_slave(out_bundle->bond, &ctx->xin->flow,
7180 vid, &ctx->xout->tags);
7182 /* No slaves enabled, so drop packet. */
7187 old_tci = ctx->xin->flow.vlan_tci;
7189 if (tci || out_bundle->use_priority_tags) {
7190 tci |= ctx->xin->flow.vlan_tci & htons(VLAN_PCP_MASK);
7192 tci |= htons(VLAN_CFI);
7195 ctx->xin->flow.vlan_tci = tci;
7197 compose_output_action(ctx, port->up.ofp_port);
7198 ctx->xin->flow.vlan_tci = old_tci;
7202 mirror_mask_ffs(mirror_mask_t mask)
7204 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
7209 ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
7211 return (bundle->vlan_mode != PORT_VLAN_ACCESS
7212 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
7216 ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan)
7218 return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan);
7221 /* Returns an arbitrary interface within 'bundle'. */
7222 static struct ofport_dpif *
7223 ofbundle_get_a_port(const struct ofbundle *bundle)
7225 return CONTAINER_OF(list_front(&bundle->ports),
7226 struct ofport_dpif, bundle_node);
7230 vlan_is_mirrored(const struct ofmirror *m, int vlan)
7232 return !m->vlans || bitmap_is_set(m->vlans, vlan);
7236 add_mirror_actions(struct xlate_ctx *ctx, const struct flow *orig_flow)
7238 struct ofproto_dpif *ofproto = ctx->ofproto;
7239 mirror_mask_t mirrors;
7240 struct ofbundle *in_bundle;
7243 const struct nlattr *a;
7246 in_bundle = lookup_input_bundle(ctx->ofproto, orig_flow->in_port,
7247 ctx->xin->packet != NULL, NULL);
7251 mirrors = in_bundle->src_mirrors;
7253 /* Drop frames on bundles reserved for mirroring. */
7254 if (in_bundle->mirror_out) {
7255 if (ctx->xin->packet != NULL) {
7256 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7257 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
7258 "%s, which is reserved exclusively for mirroring",
7259 ctx->ofproto->up.name, in_bundle->name);
7265 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
7266 if (!input_vid_is_valid(vid, in_bundle, ctx->xin->packet != NULL)) {
7269 vlan = input_vid_to_vlan(in_bundle, vid);
7271 /* Look at the output ports to check for destination selections. */
7273 NL_ATTR_FOR_EACH (a, left, ctx->xout->odp_actions.data,
7274 ctx->xout->odp_actions.size) {
7275 enum ovs_action_attr type = nl_attr_type(a);
7276 struct ofport_dpif *ofport;
7278 if (type != OVS_ACTION_ATTR_OUTPUT) {
7282 ofport = get_odp_port(ofproto, nl_attr_get_u32(a));
7283 if (ofport && ofport->bundle) {
7284 mirrors |= ofport->bundle->dst_mirrors;
7292 /* Restore the original packet before adding the mirror actions. */
7293 ctx->xin->flow = *orig_flow;
7298 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
7300 if (!vlan_is_mirrored(m, vlan)) {
7301 mirrors = zero_rightmost_1bit(mirrors);
7305 mirrors &= ~m->dup_mirrors;
7306 ctx->xout->mirrors |= m->dup_mirrors;
7308 output_normal(ctx, m->out, vlan);
7309 } else if (vlan != m->out_vlan
7310 && !eth_addr_is_reserved(orig_flow->dl_dst)) {
7311 struct ofbundle *bundle;
7313 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
7314 if (ofbundle_includes_vlan(bundle, m->out_vlan)
7315 && !bundle->mirror_out) {
7316 output_normal(ctx, bundle, m->out_vlan);
7324 update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors,
7325 uint64_t packets, uint64_t bytes)
7331 for (; mirrors; mirrors = zero_rightmost_1bit(mirrors)) {
7334 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
7337 /* In normal circumstances 'm' will not be NULL. However,
7338 * if mirrors are reconfigured, we can temporarily get out
7339 * of sync in facet_revalidate(). We could "correct" the
7340 * mirror list before reaching here, but doing that would
7341 * not properly account the traffic stats we've currently
7342 * accumulated for previous mirror configuration. */
7346 m->packet_count += packets;
7347 m->byte_count += bytes;
7351 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
7352 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
7353 * indicate this; newer upstream kernels use gratuitous ARP requests. */
7355 is_gratuitous_arp(const struct flow *flow)
7357 return (flow->dl_type == htons(ETH_TYPE_ARP)
7358 && eth_addr_is_broadcast(flow->dl_dst)
7359 && (flow->nw_proto == ARP_OP_REPLY
7360 || (flow->nw_proto == ARP_OP_REQUEST
7361 && flow->nw_src == flow->nw_dst)));
7365 update_learning_table(struct ofproto_dpif *ofproto,
7366 const struct flow *flow, int vlan,
7367 struct ofbundle *in_bundle)
7369 struct mac_entry *mac;
7371 /* Don't learn the OFPP_NONE port. */
7372 if (in_bundle == &ofpp_none_bundle) {
7376 if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
7380 mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan);
7381 if (is_gratuitous_arp(flow)) {
7382 /* We don't want to learn from gratuitous ARP packets that are
7383 * reflected back over bond slaves so we lock the learning table. */
7384 if (!in_bundle->bond) {
7385 mac_entry_set_grat_arp_lock(mac);
7386 } else if (mac_entry_is_grat_arp_locked(mac)) {
7391 if (mac_entry_is_new(mac) || mac->port.p != in_bundle) {
7392 /* The log messages here could actually be useful in debugging,
7393 * so keep the rate limit relatively high. */
7394 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
7395 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
7396 "on port %s in VLAN %d",
7397 ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src),
7398 in_bundle->name, vlan);
7400 mac->port.p = in_bundle;
7401 tag_set_add(&ofproto->backer->revalidate_set,
7402 mac_learning_changed(ofproto->ml, mac));
7406 static struct ofbundle *
7407 lookup_input_bundle(const struct ofproto_dpif *ofproto, uint16_t in_port,
7408 bool warn, struct ofport_dpif **in_ofportp)
7410 struct ofport_dpif *ofport;
7412 /* Find the port and bundle for the received packet. */
7413 ofport = get_ofp_port(ofproto, in_port);
7415 *in_ofportp = ofport;
7417 if (ofport && ofport->bundle) {
7418 return ofport->bundle;
7421 /* Special-case OFPP_NONE, which a controller may use as the ingress
7422 * port for traffic that it is sourcing. */
7423 if (in_port == OFPP_NONE) {
7424 return &ofpp_none_bundle;
7427 /* Odd. A few possible reasons here:
7429 * - We deleted a port but there are still a few packets queued up
7432 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
7433 * we don't know about.
7435 * - The ofproto client didn't configure the port as part of a bundle.
7436 * This is particularly likely to happen if a packet was received on the
7437 * port after it was created, but before the client had a chance to
7438 * configure its bundle.
7441 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7443 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
7444 "port %"PRIu16, ofproto->up.name, in_port);
7449 /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or
7450 * dropped. Returns true if they may be forwarded, false if they should be
7453 * 'in_port' must be the ofport_dpif that corresponds to flow->in_port.
7454 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
7456 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
7457 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
7458 * checked by input_vid_is_valid().
7460 * May also add tags to '*tags', although the current implementation only does
7461 * so in one special case.
7464 is_admissible(struct xlate_ctx *ctx, struct ofport_dpif *in_port,
7467 struct ofproto_dpif *ofproto = ctx->ofproto;
7468 struct flow *flow = &ctx->xin->flow;
7469 struct ofbundle *in_bundle = in_port->bundle;
7471 /* Drop frames for reserved multicast addresses
7472 * only if forward_bpdu option is absent. */
7473 if (!ofproto->up.forward_bpdu && eth_addr_is_reserved(flow->dl_dst)) {
7474 xlate_report(ctx, "packet has reserved destination MAC, dropping");
7478 if (in_bundle->bond) {
7479 struct mac_entry *mac;
7481 switch (bond_check_admissibility(in_bundle->bond, in_port,
7482 flow->dl_dst, &ctx->xout->tags)) {
7487 xlate_report(ctx, "bonding refused admissibility, dropping");
7490 case BV_DROP_IF_MOVED:
7491 mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL);
7492 if (mac && mac->port.p != in_bundle &&
7493 (!is_gratuitous_arp(flow)
7494 || mac_entry_is_grat_arp_locked(mac))) {
7495 xlate_report(ctx, "SLB bond thinks this packet looped back, "
7507 xlate_normal(struct xlate_ctx *ctx)
7509 struct ofport_dpif *in_port;
7510 struct ofbundle *in_bundle;
7511 struct mac_entry *mac;
7515 ctx->xout->has_normal = true;
7517 in_bundle = lookup_input_bundle(ctx->ofproto, ctx->xin->flow.in_port,
7518 ctx->xin->packet != NULL, &in_port);
7520 xlate_report(ctx, "no input bundle, dropping");
7524 /* Drop malformed frames. */
7525 if (ctx->xin->flow.dl_type == htons(ETH_TYPE_VLAN) &&
7526 !(ctx->xin->flow.vlan_tci & htons(VLAN_CFI))) {
7527 if (ctx->xin->packet != NULL) {
7528 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7529 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
7530 "VLAN tag received on port %s",
7531 ctx->ofproto->up.name, in_bundle->name);
7533 xlate_report(ctx, "partial VLAN tag, dropping");
7537 /* Drop frames on bundles reserved for mirroring. */
7538 if (in_bundle->mirror_out) {
7539 if (ctx->xin->packet != NULL) {
7540 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7541 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
7542 "%s, which is reserved exclusively for mirroring",
7543 ctx->ofproto->up.name, in_bundle->name);
7545 xlate_report(ctx, "input port is mirror output port, dropping");
7550 vid = vlan_tci_to_vid(ctx->xin->flow.vlan_tci);
7551 if (!input_vid_is_valid(vid, in_bundle, ctx->xin->packet != NULL)) {
7552 xlate_report(ctx, "disallowed VLAN VID for this input port, dropping");
7555 vlan = input_vid_to_vlan(in_bundle, vid);
7557 /* Check other admissibility requirements. */
7558 if (in_port && !is_admissible(ctx, in_port, vlan)) {
7562 /* Learn source MAC. */
7563 if (ctx->xin->may_learn) {
7564 update_learning_table(ctx->ofproto, &ctx->xin->flow, vlan, in_bundle);
7567 /* Determine output bundle. */
7568 mac = mac_learning_lookup(ctx->ofproto->ml, ctx->xin->flow.dl_dst, vlan,
7571 if (mac->port.p != in_bundle) {
7572 xlate_report(ctx, "forwarding to learned port");
7573 output_normal(ctx, mac->port.p, vlan);
7575 xlate_report(ctx, "learned port is input port, dropping");
7578 struct ofbundle *bundle;
7580 xlate_report(ctx, "no learned MAC for destination, flooding");
7581 HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
7582 if (bundle != in_bundle
7583 && ofbundle_includes_vlan(bundle, vlan)
7584 && bundle->floodable
7585 && !bundle->mirror_out) {
7586 output_normal(ctx, bundle, vlan);
7589 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
7593 /* Optimized flow revalidation.
7595 * It's a difficult problem, in general, to tell which facets need to have
7596 * their actions recalculated whenever the OpenFlow flow table changes. We
7597 * don't try to solve that general problem: for most kinds of OpenFlow flow
7598 * table changes, we recalculate the actions for every facet. This is
7599 * relatively expensive, but it's good enough if the OpenFlow flow table
7600 * doesn't change very often.
7602 * However, we can expect one particular kind of OpenFlow flow table change to
7603 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
7604 * of CPU on revalidating every facet whenever MAC learning modifies the flow
7605 * table, we add a special case that applies to flow tables in which every rule
7606 * has the same form (that is, the same wildcards), except that the table is
7607 * also allowed to have a single "catch-all" flow that matches all packets. We
7608 * optimize this case by tagging all of the facets that resubmit into the table
7609 * and invalidating the same tag whenever a flow changes in that table. The
7610 * end result is that we revalidate just the facets that need it (and sometimes
7611 * a few more, but not all of the facets or even all of the facets that
7612 * resubmit to the table modified by MAC learning). */
7614 /* Calculates the tag to use for 'flow' and mask 'mask' when it is inserted
7615 * into an OpenFlow table with the given 'basis'. */
7617 rule_calculate_tag(const struct flow *flow, const struct minimask *mask,
7620 if (minimask_is_catchall(mask)) {
7623 uint32_t hash = flow_hash_in_minimask(flow, mask, secret);
7624 return tag_create_deterministic(hash);
7628 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
7629 * taggability of that table.
7631 * This function must be called after *each* change to a flow table. If you
7632 * skip calling it on some changes then the pointer comparisons at the end can
7633 * be invalid if you get unlucky. For example, if a flow removal causes a
7634 * cls_table to be destroyed and then a flow insertion causes a cls_table with
7635 * different wildcards to be created with the same address, then this function
7636 * will incorrectly skip revalidation. */
7638 table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
7640 struct table_dpif *table = &ofproto->tables[table_id];
7641 const struct oftable *oftable = &ofproto->up.tables[table_id];
7642 struct cls_table *catchall, *other;
7643 struct cls_table *t;
7645 catchall = other = NULL;
7647 switch (hmap_count(&oftable->cls.tables)) {
7649 /* We could tag this OpenFlow table but it would make the logic a
7650 * little harder and it's a corner case that doesn't seem worth it
7656 HMAP_FOR_EACH (t, hmap_node, &oftable->cls.tables) {
7657 if (cls_table_is_catchall(t)) {
7659 } else if (!other) {
7662 /* Indicate that we can't tag this by setting both tables to
7663 * NULL. (We know that 'catchall' is already NULL.) */
7670 /* Can't tag this table. */
7674 if (table->catchall_table != catchall || table->other_table != other) {
7675 table->catchall_table = catchall;
7676 table->other_table = other;
7677 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
7681 /* Given 'rule' that has changed in some way (either it is a rule being
7682 * inserted, a rule being deleted, or a rule whose actions are being
7683 * modified), marks facets for revalidation to ensure that packets will be
7684 * forwarded correctly according to the new state of the flow table.
7686 * This function must be called after *each* change to a flow table. See
7687 * the comment on table_update_taggable() for more information. */
7689 rule_invalidate(const struct rule_dpif *rule)
7691 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
7693 table_update_taggable(ofproto, rule->up.table_id);
7695 if (!ofproto->backer->need_revalidate) {
7696 struct table_dpif *table = &ofproto->tables[rule->up.table_id];
7698 if (table->other_table && rule->tag) {
7699 tag_set_add(&ofproto->backer->revalidate_set, rule->tag);
7701 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
7707 set_frag_handling(struct ofproto *ofproto_,
7708 enum ofp_config_flags frag_handling)
7710 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7711 if (frag_handling != OFPC_FRAG_REASM) {
7712 ofproto->backer->need_revalidate = REV_RECONFIGURE;
7720 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
7721 const struct flow *flow,
7722 const struct ofpact *ofpacts, size_t ofpacts_len)
7724 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7725 struct initial_vals initial_vals;
7726 struct odputil_keybuf keybuf;
7727 struct dpif_flow_stats stats;
7728 struct xlate_out xout;
7729 struct xlate_in xin;
7733 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
7734 odp_flow_key_from_flow(&key, flow,
7735 ofp_port_to_odp_port(ofproto, flow->in_port));
7737 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
7739 initial_vals.vlan_tci = flow->vlan_tci;
7740 xlate_in_init(&xin, ofproto, flow, &initial_vals, NULL, stats.tcp_flags,
7742 xin.resubmit_stats = &stats;
7743 xin.ofpacts_len = ofpacts_len;
7744 xin.ofpacts = ofpacts;
7746 xlate_actions(&xin, &xout);
7747 dpif_execute(ofproto->backer->dpif, key.data, key.size,
7748 xout.odp_actions.data, xout.odp_actions.size, packet);
7749 xlate_out_uninit(&xout);
7757 set_netflow(struct ofproto *ofproto_,
7758 const struct netflow_options *netflow_options)
7760 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7762 if (netflow_options) {
7763 if (!ofproto->netflow) {
7764 ofproto->netflow = netflow_create();
7766 return netflow_set_options(ofproto->netflow, netflow_options);
7768 netflow_destroy(ofproto->netflow);
7769 ofproto->netflow = NULL;
7775 get_netflow_ids(const struct ofproto *ofproto_,
7776 uint8_t *engine_type, uint8_t *engine_id)
7778 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7780 dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
7784 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
7786 if (!facet_is_controller_flow(facet) &&
7787 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
7788 struct subfacet *subfacet;
7789 struct ofexpired expired;
7791 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
7792 if (subfacet->path == SF_FAST_PATH) {
7793 struct dpif_flow_stats stats;
7795 subfacet_install(subfacet, &facet->xout.odp_actions, &stats);
7796 subfacet_update_stats(subfacet, &stats);
7800 expired.flow = facet->flow;
7801 expired.packet_count = facet->packet_count;
7802 expired.byte_count = facet->byte_count;
7803 expired.used = facet->used;
7804 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
7809 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
7811 struct facet *facet;
7813 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
7814 send_active_timeout(ofproto, facet);
7818 static struct ofproto_dpif *
7819 ofproto_dpif_lookup(const char *name)
7821 struct ofproto_dpif *ofproto;
7823 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
7824 hash_string(name, 0), &all_ofproto_dpifs) {
7825 if (!strcmp(ofproto->up.name, name)) {
7833 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
7834 const char *argv[], void *aux OVS_UNUSED)
7836 struct ofproto_dpif *ofproto;
7839 ofproto = ofproto_dpif_lookup(argv[1]);
7841 unixctl_command_reply_error(conn, "no such bridge");
7844 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
7846 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
7847 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
7851 unixctl_command_reply(conn, "table successfully flushed");
7855 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
7856 const char *argv[], void *aux OVS_UNUSED)
7858 struct ds ds = DS_EMPTY_INITIALIZER;
7859 const struct ofproto_dpif *ofproto;
7860 const struct mac_entry *e;
7862 ofproto = ofproto_dpif_lookup(argv[1]);
7864 unixctl_command_reply_error(conn, "no such bridge");
7868 ds_put_cstr(&ds, " port VLAN MAC Age\n");
7869 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
7870 struct ofbundle *bundle = e->port.p;
7871 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
7872 ofbundle_get_a_port(bundle)->odp_port,
7873 e->vlan, ETH_ADDR_ARGS(e->mac),
7874 mac_entry_age(ofproto->ml, e));
7876 unixctl_command_reply(conn, ds_cstr(&ds));
7881 struct xlate_out xout;
7882 struct xlate_in xin;
7888 trace_format_rule(struct ds *result, uint8_t table_id, int level,
7889 const struct rule_dpif *rule)
7891 ds_put_char_multiple(result, '\t', level);
7893 ds_put_cstr(result, "No match\n");
7897 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
7898 table_id, ntohll(rule->up.flow_cookie));
7899 cls_rule_format(&rule->up.cr, result);
7900 ds_put_char(result, '\n');
7902 ds_put_char_multiple(result, '\t', level);
7903 ds_put_cstr(result, "OpenFlow ");
7904 ofpacts_format(rule->up.ofpacts, rule->up.ofpacts_len, result);
7905 ds_put_char(result, '\n');
7909 trace_format_flow(struct ds *result, int level, const char *title,
7910 struct trace_ctx *trace)
7912 ds_put_char_multiple(result, '\t', level);
7913 ds_put_format(result, "%s: ", title);
7914 if (flow_equal(&trace->xin.flow, &trace->flow)) {
7915 ds_put_cstr(result, "unchanged");
7917 flow_format(result, &trace->xin.flow);
7918 trace->flow = trace->xin.flow;
7920 ds_put_char(result, '\n');
7924 trace_format_regs(struct ds *result, int level, const char *title,
7925 struct trace_ctx *trace)
7929 ds_put_char_multiple(result, '\t', level);
7930 ds_put_format(result, "%s:", title);
7931 for (i = 0; i < FLOW_N_REGS; i++) {
7932 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
7934 ds_put_char(result, '\n');
7938 trace_format_odp(struct ds *result, int level, const char *title,
7939 struct trace_ctx *trace)
7941 struct ofpbuf *odp_actions = &trace->xout.odp_actions;
7943 ds_put_char_multiple(result, '\t', level);
7944 ds_put_format(result, "%s: ", title);
7945 format_odp_actions(result, odp_actions->data, odp_actions->size);
7946 ds_put_char(result, '\n');
7950 trace_resubmit(struct xlate_ctx *ctx, struct rule_dpif *rule)
7952 struct trace_ctx *trace = CONTAINER_OF(ctx->xin, struct trace_ctx, xin);
7953 struct ds *result = trace->result;
7955 ds_put_char(result, '\n');
7956 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
7957 trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace);
7958 trace_format_odp(result, ctx->recurse + 1, "Resubmitted odp", trace);
7959 trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule);
7963 trace_report(struct xlate_ctx *ctx, const char *s)
7965 struct trace_ctx *trace = CONTAINER_OF(ctx->xin, struct trace_ctx, xin);
7966 struct ds *result = trace->result;
7968 ds_put_char_multiple(result, '\t', ctx->recurse);
7969 ds_put_cstr(result, s);
7970 ds_put_char(result, '\n');
7974 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
7975 void *aux OVS_UNUSED)
7977 const struct dpif_backer *backer;
7978 struct ofproto_dpif *ofproto;
7979 struct ofpbuf odp_key;
7980 struct ofpbuf *packet;
7981 struct initial_vals initial_vals;
7989 ofpbuf_init(&odp_key, 0);
7991 /* Handle "-generate" or a hex string as the last argument. */
7992 if (!strcmp(argv[argc - 1], "-generate")) {
7993 packet = ofpbuf_new(0);
7996 const char *error = eth_from_hex(argv[argc - 1], &packet);
7999 } else if (argc == 4) {
8000 /* The 3-argument form must end in "-generate' or a hex string. */
8001 unixctl_command_reply_error(conn, error);
8006 /* Parse the flow and determine whether a datapath or
8007 * bridge is specified. If function odp_flow_key_from_string()
8008 * returns 0, the flow is a odp_flow. If function
8009 * parse_ofp_exact_flow() returns 0, the flow is a br_flow. */
8010 if (!odp_flow_key_from_string(argv[argc - 1], NULL, &odp_key)) {
8011 /* If the odp_flow is the second argument,
8012 * the datapath name is the first argument. */
8014 const char *dp_type;
8015 if (!strncmp(argv[1], "ovs-", 4)) {
8016 dp_type = argv[1] + 4;
8020 backer = shash_find_data(&all_dpif_backers, dp_type);
8022 unixctl_command_reply_error(conn, "Cannot find datapath "
8027 /* No datapath name specified, so there should be only one
8029 struct shash_node *node;
8030 if (shash_count(&all_dpif_backers) != 1) {
8031 unixctl_command_reply_error(conn, "Must specify datapath "
8032 "name, there is more than one type of datapath");
8035 node = shash_first(&all_dpif_backers);
8036 backer = node->data;
8039 /* Extract the ofproto_dpif object from the ofproto_receive()
8041 if (ofproto_receive(backer, NULL, odp_key.data,
8042 odp_key.size, &flow, NULL, &ofproto, NULL,
8044 unixctl_command_reply_error(conn, "Invalid datapath flow");
8047 ds_put_format(&result, "Bridge: %s\n", ofproto->up.name);
8048 } else if (!parse_ofp_exact_flow(&flow, argv[argc - 1])) {
8050 unixctl_command_reply_error(conn, "Must specify bridge name");
8054 ofproto = ofproto_dpif_lookup(argv[1]);
8056 unixctl_command_reply_error(conn, "Unknown bridge name");
8059 initial_vals.vlan_tci = flow.vlan_tci;
8061 unixctl_command_reply_error(conn, "Bad flow syntax");
8065 /* Generate a packet, if requested. */
8067 if (!packet->size) {
8068 flow_compose(packet, &flow);
8070 ds_put_cstr(&result, "Packet: ");
8071 s = ofp_packet_to_string(packet->data, packet->size);
8072 ds_put_cstr(&result, s);
8075 /* Use the metadata from the flow and the packet argument
8076 * to reconstruct the flow. */
8077 flow_extract(packet, flow.skb_priority, flow.skb_mark, NULL,
8078 flow.in_port, &flow);
8079 initial_vals.vlan_tci = flow.vlan_tci;
8083 ofproto_trace(ofproto, &flow, packet, &initial_vals, &result);
8084 unixctl_command_reply(conn, ds_cstr(&result));
8087 ds_destroy(&result);
8088 ofpbuf_delete(packet);
8089 ofpbuf_uninit(&odp_key);
8093 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
8094 const struct ofpbuf *packet,
8095 const struct initial_vals *initial_vals, struct ds *ds)
8097 struct rule_dpif *rule;
8099 ds_put_cstr(ds, "Flow: ");
8100 flow_format(ds, flow);
8101 ds_put_char(ds, '\n');
8103 rule = rule_dpif_lookup(ofproto, flow);
8105 trace_format_rule(ds, 0, 0, rule);
8106 if (rule == ofproto->miss_rule) {
8107 ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
8108 } else if (rule == ofproto->no_packet_in_rule) {
8109 ds_put_cstr(ds, "\nNo match, packets dropped because "
8110 "OFPPC_NO_PACKET_IN is set on in_port.\n");
8111 } else if (rule == ofproto->drop_frags_rule) {
8112 ds_put_cstr(ds, "\nPackets dropped because they are IP fragments "
8113 "and the fragment handling mode is \"drop\".\n");
8117 uint64_t odp_actions_stub[1024 / 8];
8118 struct ofpbuf odp_actions;
8120 struct trace_ctx trace;
8123 tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
8126 ofpbuf_use_stub(&odp_actions,
8127 odp_actions_stub, sizeof odp_actions_stub);
8128 xlate_in_init(&trace.xin, ofproto, flow, initial_vals, rule, tcp_flags,
8130 trace.xin.resubmit_hook = trace_resubmit;
8131 trace.xin.report_hook = trace_report;
8132 xlate_actions(&trace.xin, &trace.xout);
8134 ds_put_char(ds, '\n');
8135 trace_format_flow(ds, 0, "Final flow", &trace);
8136 ds_put_cstr(ds, "Datapath actions: ");
8137 format_odp_actions(ds, trace.xout.odp_actions.data,
8138 trace.xout.odp_actions.size);
8140 if (trace.xout.slow) {
8141 ds_put_cstr(ds, "\nThis flow is handled by the userspace "
8142 "slow path because it:");
8143 switch (trace.xout.slow) {
8145 ds_put_cstr(ds, "\n\t- Consists of CFM packets.");
8148 ds_put_cstr(ds, "\n\t- Consists of LACP packets.");
8151 ds_put_cstr(ds, "\n\t- Consists of STP packets.");
8154 ds_put_cstr(ds, "\n\t- Consists of BFD packets.");
8156 case SLOW_CONTROLLER:
8157 ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages "
8158 "to the OpenFlow controller.");
8165 xlate_out_uninit(&trace.xout);
8170 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
8171 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
8174 unixctl_command_reply(conn, NULL);
8178 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
8179 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
8182 unixctl_command_reply(conn, NULL);
8185 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
8186 * 'reply' describing the results. */
8188 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
8190 struct facet *facet;
8194 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
8195 if (!facet_check_consistency(facet)) {
8200 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
8204 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
8205 ofproto->up.name, errors);
8207 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
8212 ofproto_dpif_self_check(struct unixctl_conn *conn,
8213 int argc, const char *argv[], void *aux OVS_UNUSED)
8215 struct ds reply = DS_EMPTY_INITIALIZER;
8216 struct ofproto_dpif *ofproto;
8219 ofproto = ofproto_dpif_lookup(argv[1]);
8221 unixctl_command_reply_error(conn, "Unknown ofproto (use "
8222 "ofproto/list for help)");
8225 ofproto_dpif_self_check__(ofproto, &reply);
8227 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
8228 ofproto_dpif_self_check__(ofproto, &reply);
8232 unixctl_command_reply(conn, ds_cstr(&reply));
8236 /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list
8237 * of the 'ofproto_shash' nodes. It is the responsibility of the caller
8238 * to destroy 'ofproto_shash' and free the returned value. */
8239 static const struct shash_node **
8240 get_ofprotos(struct shash *ofproto_shash)
8242 const struct ofproto_dpif *ofproto;
8244 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
8245 char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
8246 shash_add_nocopy(ofproto_shash, name, ofproto);
8249 return shash_sort(ofproto_shash);
8253 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
8254 const char *argv[] OVS_UNUSED,
8255 void *aux OVS_UNUSED)
8257 struct ds ds = DS_EMPTY_INITIALIZER;
8258 struct shash ofproto_shash;
8259 const struct shash_node **sorted_ofprotos;
8262 shash_init(&ofproto_shash);
8263 sorted_ofprotos = get_ofprotos(&ofproto_shash);
8264 for (i = 0; i < shash_count(&ofproto_shash); i++) {
8265 const struct shash_node *node = sorted_ofprotos[i];
8266 ds_put_format(&ds, "%s\n", node->name);
8269 shash_destroy(&ofproto_shash);
8270 free(sorted_ofprotos);
8272 unixctl_command_reply(conn, ds_cstr(&ds));
8277 show_dp_format(const struct ofproto_dpif *ofproto, struct ds *ds)
8279 const struct shash_node **ports;
8281 struct avg_subfacet_rates lifetime;
8282 unsigned long long int minutes;
8283 const int min_ms = 60 * 1000; /* milliseconds in one minute. */
8285 minutes = (time_msec() - ofproto->created) / min_ms;
8288 lifetime.add_rate = (double)ofproto->total_subfacet_add_count
8290 lifetime.del_rate = (double)ofproto->total_subfacet_del_count
8293 lifetime.add_rate = 0.0;
8294 lifetime.del_rate = 0.0;
8297 ds_put_format(ds, "%s (%s):\n", ofproto->up.name,
8298 dpif_name(ofproto->backer->dpif));
8300 "\tlookups: hit:%"PRIu64" missed:%"PRIu64"\n",
8301 ofproto->n_hit, ofproto->n_missed);
8302 ds_put_format(ds, "\tflows: cur: %zu, avg: %u, max: %d,"
8303 " life span: %lld(ms)\n",
8304 hmap_count(&ofproto->subfacets),
8305 ofproto->avg_n_subfacet,
8306 ofproto->max_n_subfacet,
8307 ofproto->avg_subfacet_life_span);
8308 if (minutes >= 60) {
8309 show_dp_rates(ds, "\t\thourly avg:", &ofproto->hourly);
8311 if (minutes >= 60 * 24) {
8312 show_dp_rates(ds, "\t\tdaily avg:", &ofproto->daily);
8314 show_dp_rates(ds, "\t\toverall avg:", &lifetime);
8316 ports = shash_sort(&ofproto->up.port_by_name);
8317 for (i = 0; i < shash_count(&ofproto->up.port_by_name); i++) {
8318 const struct shash_node *node = ports[i];
8319 struct ofport *ofport = node->data;
8320 const char *name = netdev_get_name(ofport->netdev);
8321 const char *type = netdev_get_type(ofport->netdev);
8324 ds_put_format(ds, "\t%s %u/", name, ofport->ofp_port);
8326 odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
8327 if (odp_port != OVSP_NONE) {
8328 ds_put_format(ds, "%"PRIu32":", odp_port);
8330 ds_put_cstr(ds, "none:");
8333 if (strcmp(type, "system")) {
8334 struct netdev *netdev;
8337 ds_put_format(ds, " (%s", type);
8339 error = netdev_open(name, type, &netdev);
8344 error = netdev_get_config(netdev, &config);
8346 const struct smap_node **nodes;
8349 nodes = smap_sort(&config);
8350 for (i = 0; i < smap_count(&config); i++) {
8351 const struct smap_node *node = nodes[i];
8352 ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
8353 node->key, node->value);
8357 smap_destroy(&config);
8359 netdev_close(netdev);
8361 ds_put_char(ds, ')');
8363 ds_put_char(ds, '\n');
8369 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc,
8370 const char *argv[], void *aux OVS_UNUSED)
8372 struct ds ds = DS_EMPTY_INITIALIZER;
8373 const struct ofproto_dpif *ofproto;
8377 for (i = 1; i < argc; i++) {
8378 ofproto = ofproto_dpif_lookup(argv[i]);
8380 ds_put_format(&ds, "Unknown bridge %s (use dpif/dump-dps "
8381 "for help)", argv[i]);
8382 unixctl_command_reply_error(conn, ds_cstr(&ds));
8385 show_dp_format(ofproto, &ds);
8388 struct shash ofproto_shash;
8389 const struct shash_node **sorted_ofprotos;
8392 shash_init(&ofproto_shash);
8393 sorted_ofprotos = get_ofprotos(&ofproto_shash);
8394 for (i = 0; i < shash_count(&ofproto_shash); i++) {
8395 const struct shash_node *node = sorted_ofprotos[i];
8396 show_dp_format(node->data, &ds);
8399 shash_destroy(&ofproto_shash);
8400 free(sorted_ofprotos);
8403 unixctl_command_reply(conn, ds_cstr(&ds));
8408 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
8409 int argc OVS_UNUSED, const char *argv[],
8410 void *aux OVS_UNUSED)
8412 struct ds ds = DS_EMPTY_INITIALIZER;
8413 const struct ofproto_dpif *ofproto;
8414 struct subfacet *subfacet;
8416 ofproto = ofproto_dpif_lookup(argv[1]);
8418 unixctl_command_reply_error(conn, "no such bridge");
8422 update_stats(ofproto->backer);
8424 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
8425 struct facet *facet = subfacet->facet;
8427 odp_flow_key_format(subfacet->key, subfacet->key_len, &ds);
8429 ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:",
8430 subfacet->dp_packet_count, subfacet->dp_byte_count);
8431 if (subfacet->used) {
8432 ds_put_format(&ds, "%.3fs",
8433 (time_msec() - subfacet->used) / 1000.0);
8435 ds_put_format(&ds, "never");
8437 if (subfacet->facet->tcp_flags) {
8438 ds_put_cstr(&ds, ", flags:");
8439 packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags);
8442 ds_put_cstr(&ds, ", actions:");
8443 if (facet->xout.slow) {
8444 uint64_t slow_path_stub[128 / 8];
8445 const struct nlattr *actions;
8448 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
8449 slow_path_stub, sizeof slow_path_stub,
8450 &actions, &actions_len);
8451 format_odp_actions(&ds, actions, actions_len);
8453 format_odp_actions(&ds, facet->xout.odp_actions.data,
8454 facet->xout.odp_actions.size);
8456 ds_put_char(&ds, '\n');
8459 unixctl_command_reply(conn, ds_cstr(&ds));
8464 ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn,
8465 int argc OVS_UNUSED, const char *argv[],
8466 void *aux OVS_UNUSED)
8468 struct ds ds = DS_EMPTY_INITIALIZER;
8469 struct ofproto_dpif *ofproto;
8471 ofproto = ofproto_dpif_lookup(argv[1]);
8473 unixctl_command_reply_error(conn, "no such bridge");
8477 flush(&ofproto->up);
8479 unixctl_command_reply(conn, ds_cstr(&ds));
8484 ofproto_dpif_unixctl_init(void)
8486 static bool registered;
8492 unixctl_command_register(
8494 "[dp_name]|bridge odp_flow|br_flow [-generate|packet]",
8495 1, 3, ofproto_unixctl_trace, NULL);
8496 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
8497 ofproto_unixctl_fdb_flush, NULL);
8498 unixctl_command_register("fdb/show", "bridge", 1, 1,
8499 ofproto_unixctl_fdb_show, NULL);
8500 unixctl_command_register("ofproto/clog", "", 0, 0,
8501 ofproto_dpif_clog, NULL);
8502 unixctl_command_register("ofproto/unclog", "", 0, 0,
8503 ofproto_dpif_unclog, NULL);
8504 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
8505 ofproto_dpif_self_check, NULL);
8506 unixctl_command_register("dpif/dump-dps", "", 0, 0,
8507 ofproto_unixctl_dpif_dump_dps, NULL);
8508 unixctl_command_register("dpif/show", "[bridge]", 0, INT_MAX,
8509 ofproto_unixctl_dpif_show, NULL);
8510 unixctl_command_register("dpif/dump-flows", "bridge", 1, 1,
8511 ofproto_unixctl_dpif_dump_flows, NULL);
8512 unixctl_command_register("dpif/del-flows", "bridge", 1, 1,
8513 ofproto_unixctl_dpif_del_flows, NULL);
8516 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
8518 * This is deprecated. It is only for compatibility with broken device drivers
8519 * in old versions of Linux that do not properly support VLANs when VLAN
8520 * devices are not used. When broken device drivers are no longer in
8521 * widespread use, we will delete these interfaces. */
8524 set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid)
8526 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
8527 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
8529 if (realdev_ofp_port == ofport->realdev_ofp_port
8530 && vid == ofport->vlandev_vid) {
8534 ofproto->backer->need_revalidate = REV_RECONFIGURE;
8536 if (ofport->realdev_ofp_port) {
8539 if (realdev_ofp_port && ofport->bundle) {
8540 /* vlandevs are enslaved to their realdevs, so they are not allowed to
8541 * themselves be part of a bundle. */
8542 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
8545 ofport->realdev_ofp_port = realdev_ofp_port;
8546 ofport->vlandev_vid = vid;
8548 if (realdev_ofp_port) {
8549 vsp_add(ofport, realdev_ofp_port, vid);
8556 hash_realdev_vid(uint16_t realdev_ofp_port, int vid)
8558 return hash_2words(realdev_ofp_port, vid);
8561 /* Returns the OFP port number of the Linux VLAN device that corresponds to
8562 * 'vlan_tci' on the network device with port number 'realdev_ofp_port' in
8563 * 'struct ofport_dpif'. For example, given 'realdev_ofp_port' of eth0 and
8564 * 'vlan_tci' 9, it would return the port number of eth0.9.
8566 * Unless VLAN splinters are enabled for port 'realdev_ofp_port', this
8567 * function just returns its 'realdev_ofp_port' argument. */
8569 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
8570 uint16_t realdev_ofp_port, ovs_be16 vlan_tci)
8572 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
8573 int vid = vlan_tci_to_vid(vlan_tci);
8574 const struct vlan_splinter *vsp;
8576 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
8577 hash_realdev_vid(realdev_ofp_port, vid),
8578 &ofproto->realdev_vid_map) {
8579 if (vsp->realdev_ofp_port == realdev_ofp_port
8580 && vsp->vid == vid) {
8581 return vsp->vlandev_ofp_port;
8585 return realdev_ofp_port;
8588 static struct vlan_splinter *
8589 vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port)
8591 struct vlan_splinter *vsp;
8593 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, hash_int(vlandev_ofp_port, 0),
8594 &ofproto->vlandev_map) {
8595 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
8603 /* Returns the OpenFlow port number of the "real" device underlying the Linux
8604 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
8605 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
8606 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
8607 * eth0 and store 9 in '*vid'.
8609 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
8610 * VLAN device. Unless VLAN splinters are enabled, this is what this function
8613 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
8614 uint16_t vlandev_ofp_port, int *vid)
8616 if (!hmap_is_empty(&ofproto->vlandev_map)) {
8617 const struct vlan_splinter *vsp;
8619 vsp = vlandev_find(ofproto, vlandev_ofp_port);
8624 return vsp->realdev_ofp_port;
8630 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
8631 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
8632 * 'flow->in_port' to the "real" device backing the VLAN device, sets
8633 * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is
8634 * always the case unless VLAN splinters are enabled), returns false without
8635 * making any changes. */
8637 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
8642 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port, &vid);
8647 /* Cause the flow to be processed as if it came in on the real device with
8648 * the VLAN device's VLAN ID. */
8649 flow->in_port = realdev;
8650 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
8655 vsp_remove(struct ofport_dpif *port)
8657 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
8658 struct vlan_splinter *vsp;
8660 vsp = vlandev_find(ofproto, port->up.ofp_port);
8662 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
8663 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
8666 port->realdev_ofp_port = 0;
8668 VLOG_ERR("missing vlan device record");
8673 vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid)
8675 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
8677 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
8678 && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid))
8679 == realdev_ofp_port)) {
8680 struct vlan_splinter *vsp;
8682 vsp = xmalloc(sizeof *vsp);
8683 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
8684 hash_int(port->up.ofp_port, 0));
8685 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
8686 hash_realdev_vid(realdev_ofp_port, vid));
8687 vsp->realdev_ofp_port = realdev_ofp_port;
8688 vsp->vlandev_ofp_port = port->up.ofp_port;
8691 port->realdev_ofp_port = realdev_ofp_port;
8693 VLOG_ERR("duplicate vlan device record");
8698 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
8700 const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
8701 return ofport ? ofport->odp_port : OVSP_NONE;
8704 static struct ofport_dpif *
8705 odp_port_to_ofport(const struct dpif_backer *backer, uint32_t odp_port)
8707 struct ofport_dpif *port;
8709 HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node,
8710 hash_int(odp_port, 0),
8711 &backer->odp_to_ofport_map) {
8712 if (port->odp_port == odp_port) {
8721 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
8723 struct ofport_dpif *port;
8725 port = odp_port_to_ofport(ofproto->backer, odp_port);
8726 if (port && &ofproto->up == port->up.ofproto) {
8727 return port->up.ofp_port;
8734 show_dp_rates(struct ds *ds, const char *heading,
8735 const struct avg_subfacet_rates *rates)
8737 ds_put_format(ds, "%s add rate: %5.3f/min, del rate: %5.3f/min\n",
8738 heading, rates->add_rate, rates->del_rate);
8741 /* Compute exponentially weighted moving average, adding 'new' as the newest,
8742 * most heavily weighted element. 'base' designates the rate of decay: after
8743 * 'base' further updates, 'new''s weight in the EWMA decays to about 1/e
8746 exp_mavg(double *avg, int base, double new)
8748 *avg = (*avg * (base - 1) + new) / base;
8752 update_moving_averages(struct ofproto_dpif *ofproto)
8754 const int min_ms = 60 * 1000; /* milliseconds in one minute. */
8756 /* Update hourly averages on the minute boundaries. */
8757 if (time_msec() - ofproto->last_minute >= min_ms) {
8758 exp_mavg(&ofproto->hourly.add_rate, 60, ofproto->subfacet_add_count);
8759 exp_mavg(&ofproto->hourly.del_rate, 60, ofproto->subfacet_del_count);
8761 /* Update daily averages on the hour boundaries. */
8762 if ((ofproto->last_minute - ofproto->created) / min_ms % 60 == 59) {
8763 exp_mavg(&ofproto->daily.add_rate, 24, ofproto->hourly.add_rate);
8764 exp_mavg(&ofproto->daily.del_rate, 24, ofproto->hourly.del_rate);
8767 ofproto->total_subfacet_add_count += ofproto->subfacet_add_count;
8768 ofproto->total_subfacet_del_count += ofproto->subfacet_del_count;
8769 ofproto->subfacet_add_count = 0;
8770 ofproto->subfacet_del_count = 0;
8771 ofproto->last_minute += min_ms;
8775 const struct ofproto_class ofproto_dpif_class = {
8810 port_is_lacp_current,
8811 NULL, /* rule_choose_table */
8818 rule_modify_actions,
8832 get_stp_port_status,
8839 is_mirror_output_bundle,
8840 forward_bpdu_changed,
8841 set_mac_table_config,