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"
25 #include "byte-order.h"
30 #include "dynamic-string.h"
31 #include "fail-open.h"
35 #include "mac-learning.h"
36 #include "meta-flow.h"
37 #include "multipath.h"
38 #include "netdev-vport.h"
45 #include "ofp-actions.h"
46 #include "ofp-parse.h"
47 #include "ofp-print.h"
48 #include "ofproto-dpif-governor.h"
49 #include "ofproto-dpif-ipfix.h"
50 #include "ofproto-dpif-sflow.h"
51 #include "poll-loop.h"
56 #include "unaligned.h"
58 #include "vlan-bitmap.h"
61 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
63 COVERAGE_DEFINE(ofproto_dpif_expired);
64 COVERAGE_DEFINE(ofproto_dpif_xlate);
65 COVERAGE_DEFINE(facet_changed_rule);
66 COVERAGE_DEFINE(facet_revalidate);
67 COVERAGE_DEFINE(facet_unexpected);
68 COVERAGE_DEFINE(facet_suppress);
70 /* Maximum depth of flow table recursion (due to resubmit actions) in a
71 * flow translation. */
72 #define MAX_RESUBMIT_RECURSION 64
74 /* Number of implemented OpenFlow tables. */
75 enum { N_TABLES = 255 };
76 enum { TBL_INTERNAL = N_TABLES - 1 }; /* Used for internal hidden rules. */
77 BUILD_ASSERT_DECL(N_TABLES >= 2 && N_TABLES <= 255);
89 * - Do include packets and bytes from facets that have been deleted or
90 * whose own statistics have been folded into the rule.
92 * - Do include packets and bytes sent "by hand" that were accounted to
93 * the rule without any facet being involved (this is a rare corner
94 * case in rule_execute()).
96 * - Do not include packet or bytes that can be obtained from any facet's
97 * packet_count or byte_count member or that can be obtained from the
98 * datapath by, e.g., dpif_flow_get() for any subfacet.
100 uint64_t packet_count; /* Number of packets received. */
101 uint64_t byte_count; /* Number of bytes received. */
103 tag_type tag; /* Caches rule_calculate_tag() result. */
105 struct list facets; /* List of "struct facet"s. */
108 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
110 return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
113 static struct rule_dpif *rule_dpif_lookup(struct ofproto_dpif *,
114 const struct flow *);
115 static struct rule_dpif *rule_dpif_lookup__(struct ofproto_dpif *,
118 static struct rule_dpif *rule_dpif_miss_rule(struct ofproto_dpif *ofproto,
119 const struct flow *flow);
121 static void rule_get_stats(struct rule *, uint64_t *packets, uint64_t *bytes);
122 static void rule_credit_stats(struct rule_dpif *,
123 const struct dpif_flow_stats *);
124 static tag_type rule_calculate_tag(const struct flow *,
125 const struct minimask *, uint32_t basis);
126 static void rule_invalidate(const struct rule_dpif *);
128 #define MAX_MIRRORS 32
129 typedef uint32_t mirror_mask_t;
130 #define MIRROR_MASK_C(X) UINT32_C(X)
131 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
133 struct ofproto_dpif *ofproto; /* Owning ofproto. */
134 size_t idx; /* In ofproto's "mirrors" array. */
135 void *aux; /* Key supplied by ofproto's client. */
136 char *name; /* Identifier for log messages. */
138 /* Selection criteria. */
139 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
140 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
141 unsigned long *vlans; /* Bitmap of chosen VLANs, NULL selects all. */
143 /* Output (exactly one of out == NULL and out_vlan == -1 is true). */
144 struct ofbundle *out; /* Output port or NULL. */
145 int out_vlan; /* Output VLAN or -1. */
146 mirror_mask_t dup_mirrors; /* Bitmap of mirrors with the same output. */
149 int64_t packet_count; /* Number of packets sent. */
150 int64_t byte_count; /* Number of bytes sent. */
153 static void mirror_destroy(struct ofmirror *);
154 static void update_mirror_stats(struct ofproto_dpif *ofproto,
155 mirror_mask_t mirrors,
156 uint64_t packets, uint64_t bytes);
159 struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
160 struct ofproto_dpif *ofproto; /* Owning ofproto. */
161 void *aux; /* Key supplied by ofproto's client. */
162 char *name; /* Identifier for log messages. */
165 struct list ports; /* Contains "struct ofport"s. */
166 enum port_vlan_mode vlan_mode; /* VLAN mode */
167 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
168 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
169 * NULL if all VLANs are trunked. */
170 struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */
171 struct bond *bond; /* Nonnull iff more than one port. */
172 bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */
175 bool floodable; /* True if no port has OFPUTIL_PC_NO_FLOOD set. */
177 /* Port mirroring info. */
178 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
179 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
180 mirror_mask_t mirror_out; /* Mirrors that output to this bundle. */
183 static void bundle_remove(struct ofport *);
184 static void bundle_update(struct ofbundle *);
185 static void bundle_destroy(struct ofbundle *);
186 static void bundle_del_port(struct ofport_dpif *);
187 static void bundle_run(struct ofbundle *);
188 static void bundle_wait(struct ofbundle *);
189 static struct ofbundle *lookup_input_bundle(const struct ofproto_dpif *,
190 uint16_t in_port, bool warn,
191 struct ofport_dpif **in_ofportp);
193 /* A controller may use OFPP_NONE as the ingress port to indicate that
194 * it did not arrive on a "real" port. 'ofpp_none_bundle' exists for
195 * when an input bundle is needed for validation (e.g., mirroring or
196 * OFPP_NORMAL processing). It is not connected to an 'ofproto' or have
197 * any 'port' structs, so care must be taken when dealing with it. */
198 static struct ofbundle ofpp_none_bundle = {
200 .vlan_mode = PORT_VLAN_TRUNK
203 static void stp_run(struct ofproto_dpif *ofproto);
204 static void stp_wait(struct ofproto_dpif *ofproto);
205 static int set_stp_port(struct ofport *,
206 const struct ofproto_port_stp_settings *);
208 static bool ofbundle_includes_vlan(const struct ofbundle *, uint16_t vlan);
212 /* Initial values of fields of the packet that may be changed during
213 * flow processing and needed later. */
214 struct initial_vals {
215 /* This is the value of vlan_tci in the packet as actually received from
216 * dpif. This is the same as the facet's flow.vlan_tci unless the packet
217 * was received via a VLAN splinter. In that case, this value is 0
218 * (because the packet as actually received from the dpif had no 802.1Q
219 * tag) but the facet's flow.vlan_tci is set to the VLAN that the splinter
222 * This member should be removed when the VLAN splinters feature is no
226 /* If received on a tunnel, the IP TOS value of the tunnel. */
227 uint8_t tunnel_ip_tos;
231 tag_type tags; /* Tags associated with actions. */
232 enum slow_path_reason slow; /* 0 if fast path may be used. */
233 bool has_learn; /* Actions include NXAST_LEARN? */
234 bool has_normal; /* Actions output to OFPP_NORMAL? */
235 bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */
236 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
237 mirror_mask_t mirrors; /* Bitmap of associated mirrors. */
239 uint64_t odp_actions_stub[256 / 8];
240 struct ofpbuf odp_actions;
244 struct ofproto_dpif *ofproto;
246 /* Flow to which the OpenFlow actions apply. xlate_actions() will modify
247 * this flow when actions change header fields. */
250 struct initial_vals initial_vals;
252 /* The packet corresponding to 'flow', or a null pointer if we are
253 * revalidating without a packet to refer to. */
254 const struct ofpbuf *packet;
256 /* Should OFPP_NORMAL update the MAC learning table? Should "learn"
257 * actions update the flow table?
259 * We want to update these tables if we are actually processing a packet,
260 * or if we are accounting for packets that the datapath has processed, but
261 * not if we are just revalidating. */
264 /* The rule initiating translation or NULL. */
265 struct rule_dpif *rule;
267 /* The actions to translate. If 'rule' is not NULL, these may be NULL. */
268 const struct ofpact *ofpacts;
271 /* Union of the set of TCP flags seen so far in this flow. (Used only by
272 * NXAST_FIN_TIMEOUT. Set to zero to avoid updating updating rules'
276 /* If nonnull, flow translation calls this function just before executing a
277 * resubmit or OFPP_TABLE action. In addition, disables logging of traces
278 * when the recursion depth is exceeded.
280 * 'rule' is the rule being submitted into. It will be null if the
281 * resubmit or OFPP_TABLE action didn't find a matching rule.
283 * This is normally null so the client has to set it manually after
284 * calling xlate_in_init(). */
285 void (*resubmit_hook)(struct xlate_ctx *, struct rule_dpif *rule);
287 /* If nonnull, flow translation calls this function to report some
288 * significant decision, e.g. to explain why OFPP_NORMAL translation
289 * dropped a packet. */
290 void (*report_hook)(struct xlate_ctx *, const char *s);
292 /* If nonnull, flow translation credits the specified statistics to each
293 * rule reached through a resubmit or OFPP_TABLE action.
295 * This is normally null so the client has to set it manually after
296 * calling xlate_in_init(). */
297 const struct dpif_flow_stats *resubmit_stats;
300 /* Context used by xlate_actions() and its callees. */
302 struct xlate_in *xin;
303 struct xlate_out *xout;
305 struct ofproto_dpif *ofproto;
307 /* Flow at the last commit. */
308 struct flow base_flow;
310 /* Stack for the push and pop actions. Each stack element is of type
311 * "union mf_subvalue". */
312 union mf_subvalue init_stack[1024 / sizeof(union mf_subvalue)];
315 /* The rule that we are currently translating, or NULL. */
316 struct rule_dpif *rule;
318 int recurse; /* Recursion level, via xlate_table_action. */
319 bool max_resubmit_trigger; /* Recursed too deeply during translation. */
320 uint32_t orig_skb_priority; /* Priority when packet arrived. */
321 uint8_t table_id; /* OpenFlow table ID where flow was found. */
322 uint32_t sflow_n_outputs; /* Number of output ports. */
323 uint32_t sflow_odp_port; /* Output port for composing sFlow action. */
324 uint16_t user_cookie_offset;/* Used for user_action_cookie fixup. */
325 bool exit; /* No further actions should be processed. */
328 static void xlate_in_init(struct xlate_in *, struct ofproto_dpif *,
329 const struct flow *, const struct initial_vals *,
330 struct rule_dpif *, uint8_t tcp_flags,
331 const struct ofpbuf *);
333 static void xlate_out_uninit(struct xlate_out *);
335 static void xlate_actions(struct xlate_in *, struct xlate_out *);
337 static void xlate_actions_for_side_effects(struct xlate_in *);
339 static void xlate_table_action(struct xlate_ctx *, uint16_t in_port,
340 uint8_t table_id, bool may_packet_in);
342 static size_t put_userspace_action(const struct ofproto_dpif *,
343 struct ofpbuf *odp_actions,
345 const union user_action_cookie *,
348 static void compose_slow_path(const struct ofproto_dpif *, const struct flow *,
349 enum slow_path_reason,
350 uint64_t *stub, size_t stub_size,
351 const struct nlattr **actionsp,
352 size_t *actions_lenp);
354 static void xlate_report(struct xlate_ctx *ctx, const char *s);
356 /* A subfacet (see "struct subfacet" below) has three possible installation
359 * - SF_NOT_INSTALLED: Not installed in the datapath. This will only be the
360 * case just after the subfacet is created, just before the subfacet is
361 * destroyed, or if the datapath returns an error when we try to install a
364 * - SF_FAST_PATH: The subfacet's actions are installed in the datapath.
366 * - SF_SLOW_PATH: An action that sends every packet for the subfacet through
367 * ofproto_dpif is installed in the datapath.
370 SF_NOT_INSTALLED, /* No datapath flow for this subfacet. */
371 SF_FAST_PATH, /* Full actions are installed. */
372 SF_SLOW_PATH, /* Send-to-userspace action is installed. */
375 /* A dpif flow and actions associated with a facet.
377 * See also the large comment on struct facet. */
380 struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */
381 struct list list_node; /* In struct facet's 'facets' list. */
382 struct facet *facet; /* Owning facet. */
384 enum odp_key_fitness key_fitness;
388 long long int used; /* Time last used; time created if not used. */
389 long long int created; /* Time created. */
391 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
392 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
394 enum subfacet_path path; /* Installed in datapath? */
396 /* Datapath port the packet arrived on. This is needed to remove
397 * flows for ports that are no longer part of the bridge. Since the
398 * flow definition only has the OpenFlow port number and the port is
399 * no longer part of the bridge, we can't determine the datapath port
400 * number needed to delete the flow from the datapath. */
401 uint32_t odp_in_port;
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 tag_type tag; /* Tag associated with this port. */
521 bool may_enable; /* May be enabled in bonds. */
522 long long int carrier_seq; /* Carrier status changes. */
523 struct tnl_port *tnl_port; /* Tunnel handle, or null. */
526 struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */
527 enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */
528 long long int stp_state_entered;
530 struct hmap priorities; /* Map of attached 'priority_to_dscp's. */
532 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
534 * This is deprecated. It is only for compatibility with broken device
535 * drivers in old versions of Linux that do not properly support VLANs when
536 * VLAN devices are not used. When broken device drivers are no longer in
537 * widespread use, we will delete these interfaces. */
538 uint16_t realdev_ofp_port;
542 /* Node in 'ofport_dpif''s 'priorities' map. Used to maintain a map from
543 * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
544 * traffic egressing the 'ofport' with that priority should be marked with. */
545 struct priority_to_dscp {
546 struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'priorities' map. */
547 uint32_t priority; /* Priority of this queue (see struct flow). */
549 uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */
552 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
554 * This is deprecated. It is only for compatibility with broken device drivers
555 * in old versions of Linux that do not properly support VLANs when VLAN
556 * devices are not used. When broken device drivers are no longer in
557 * widespread use, we will delete these interfaces. */
558 struct vlan_splinter {
559 struct hmap_node realdev_vid_node;
560 struct hmap_node vlandev_node;
561 uint16_t realdev_ofp_port;
562 uint16_t vlandev_ofp_port;
566 static uint16_t vsp_realdev_to_vlandev(const struct ofproto_dpif *,
567 uint16_t realdev_ofp_port,
569 static bool vsp_adjust_flow(const struct ofproto_dpif *, struct flow *);
570 static void vsp_remove(struct ofport_dpif *);
571 static void vsp_add(struct ofport_dpif *, uint16_t realdev_ofp_port, int vid);
573 static uint32_t ofp_port_to_odp_port(const struct ofproto_dpif *,
575 static uint16_t odp_port_to_ofp_port(const struct ofproto_dpif *,
578 static struct ofport_dpif *
579 ofport_dpif_cast(const struct ofport *ofport)
581 ovs_assert(ofport->ofproto->ofproto_class == &ofproto_dpif_class);
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_cfm(struct ofport *, const struct cfm_settings *);
589 static void ofport_clear_priorities(struct ofport_dpif *);
590 static void run_fast_rl(void);
592 struct dpif_completion {
593 struct list list_node;
594 struct ofoperation *op;
597 /* Extra information about a classifier table.
598 * Currently used just for optimized flow revalidation. */
600 /* If either of these is nonnull, then this table has a form that allows
601 * flows to be tagged to avoid revalidating most flows for the most common
602 * kinds of flow table changes. */
603 struct cls_table *catchall_table; /* Table that wildcards all fields. */
604 struct cls_table *other_table; /* Table with any other wildcard set. */
605 uint32_t basis; /* Keeps each table's tags separate. */
608 /* Reasons that we might need to revalidate every facet, and corresponding
611 * A value of 0 means that there is no need to revalidate.
613 * It would be nice to have some cleaner way to integrate with coverage
614 * counters, but with only a few reasons I guess this is good enough for
616 enum revalidate_reason {
617 REV_RECONFIGURE = 1, /* Switch configuration changed. */
618 REV_STP, /* Spanning tree protocol port status change. */
619 REV_PORT_TOGGLED, /* Port enabled or disabled by CFM, LACP, ...*/
620 REV_FLOW_TABLE, /* Flow table changed. */
621 REV_INCONSISTENCY /* Facet self-check failed. */
623 COVERAGE_DEFINE(rev_reconfigure);
624 COVERAGE_DEFINE(rev_stp);
625 COVERAGE_DEFINE(rev_port_toggled);
626 COVERAGE_DEFINE(rev_flow_table);
627 COVERAGE_DEFINE(rev_inconsistency);
629 /* Drop keys are odp flow keys which have drop flows installed in the kernel.
630 * These are datapath flows which have no associated ofproto, if they did we
631 * would use facets. */
633 struct hmap_node hmap_node;
638 /* All datapaths of a given type share a single dpif backer instance. */
643 struct timer next_expiration;
644 struct hmap odp_to_ofport_map; /* ODP port to ofport mapping. */
646 struct simap tnl_backers; /* Set of dpif ports backing tunnels. */
648 /* Facet revalidation flags applying to facets which use this backer. */
649 enum revalidate_reason need_revalidate; /* Revalidate every facet. */
650 struct tag_set revalidate_set; /* Revalidate only matching facets. */
652 struct hmap drop_keys; /* Set of dropped odp keys. */
653 bool recv_set_enable; /* Enables or disables receiving packets. */
656 /* All existing ofproto_backer instances, indexed by ofproto->up.type. */
657 static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers);
659 static void drop_key_clear(struct dpif_backer *);
660 static struct ofport_dpif *
661 odp_port_to_ofport(const struct dpif_backer *, uint32_t odp_port);
663 static void dpif_stats_update_hit_count(struct ofproto_dpif *ofproto,
665 struct avg_subfacet_rates {
666 double add_rate; /* Moving average of new flows created per minute. */
667 double del_rate; /* Moving average of flows deleted per minute. */
669 static void show_dp_rates(struct ds *ds, const char *heading,
670 const struct avg_subfacet_rates *rates);
671 static void exp_mavg(double *avg, int base, double new);
673 struct ofproto_dpif {
674 struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
676 struct dpif_backer *backer;
678 /* Special OpenFlow rules. */
679 struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */
680 struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */
681 struct rule_dpif *drop_frags_rule; /* Used in OFPC_FRAG_DROP mode. */
684 struct netflow *netflow;
685 struct dpif_sflow *sflow;
686 struct dpif_ipfix *ipfix;
687 struct hmap bundles; /* Contains "struct ofbundle"s. */
688 struct mac_learning *ml;
689 struct ofmirror *mirrors[MAX_MIRRORS];
691 bool has_bonded_bundles;
695 struct hmap subfacets;
696 struct governor *governor;
697 long long int consistency_rl;
700 struct table_dpif tables[N_TABLES];
702 /* Support for debugging async flow mods. */
703 struct list completions;
705 bool has_bundle_action; /* True when the first bundle action appears. */
706 struct netdev_stats stats; /* To account packets generated and consumed in
711 long long int stp_last_tick;
713 /* VLAN splinters. */
714 struct hmap realdev_vid_map; /* (realdev,vid) -> vlandev. */
715 struct hmap vlandev_map; /* vlandev -> (realdev,vid). */
718 struct sset ports; /* Set of standard port names. */
719 struct sset ghost_ports; /* Ports with no datapath port. */
720 struct sset port_poll_set; /* Queued names for port_poll() reply. */
721 int port_poll_errno; /* Last errno for port_poll() reply. */
723 /* Per ofproto's dpif stats. */
727 /* Subfacet statistics.
729 * These keep track of the total number of subfacets added and deleted and
730 * flow life span. They are useful for computing the flow rates stats
731 * exposed via "ovs-appctl dpif/show". The goal is to learn about
732 * traffic patterns in ways that we can use later to improve Open vSwitch
733 * performance in new situations. */
734 long long int created; /* Time when it is created. */
735 unsigned int max_n_subfacet; /* Maximum number of flows */
737 /* The average number of subfacets... */
738 struct avg_subfacet_rates hourly; /* ...over the last hour. */
739 struct avg_subfacet_rates daily; /* ...over the last day. */
740 long long int last_minute; /* Last time 'hourly' was updated. */
742 /* Number of subfacets added or deleted since 'last_minute'. */
743 unsigned int subfacet_add_count;
744 unsigned int subfacet_del_count;
746 /* Number of subfacets added or deleted from 'created' to 'last_minute.' */
747 unsigned long long int total_subfacet_add_count;
748 unsigned long long int total_subfacet_del_count;
750 /* Sum of the number of milliseconds that each subfacet existed,
751 * over the subfacets that have been added and then later deleted. */
752 unsigned long long int total_subfacet_life_span;
754 /* Incremented by the number of currently existing subfacets, each
755 * time we pull statistics from the kernel. */
756 unsigned long long int total_subfacet_count;
758 /* Number of times we pull statistics from the kernel. */
759 unsigned long long int n_update_stats;
761 static unsigned long long int avg_subfacet_life_span(
762 const struct ofproto_dpif *);
763 static double avg_subfacet_count(const struct ofproto_dpif *ofproto);
764 static void update_moving_averages(struct ofproto_dpif *ofproto);
765 static void dpif_stats_update_hit_count(struct ofproto_dpif *ofproto,
767 static void update_max_subfacet_count(struct ofproto_dpif *ofproto);
769 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
770 * for debugging the asynchronous flow_mod implementation.) */
773 /* All existing ofproto_dpif instances, indexed by ->up.name. */
774 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
776 static void ofproto_dpif_unixctl_init(void);
778 static struct ofproto_dpif *
779 ofproto_dpif_cast(const struct ofproto *ofproto)
781 ovs_assert(ofproto->ofproto_class == &ofproto_dpif_class);
782 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
785 static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *,
787 static struct ofport_dpif *get_odp_port(const struct ofproto_dpif *,
789 static void ofproto_trace(struct ofproto_dpif *, const struct flow *,
790 const struct ofpbuf *,
791 const struct initial_vals *, struct ds *);
793 /* Packet processing. */
794 static void update_learning_table(struct ofproto_dpif *,
795 const struct flow *, int vlan,
798 #define FLOW_MISS_MAX_BATCH 50
799 static int handle_upcalls(struct dpif_backer *, unsigned int max_batch);
801 /* Flow expiration. */
802 static int expire(struct dpif_backer *);
805 static void send_netflow_active_timeouts(struct ofproto_dpif *);
808 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
809 static size_t compose_sflow_action(const struct ofproto_dpif *,
810 struct ofpbuf *odp_actions,
811 const struct flow *, uint32_t odp_port);
812 static void compose_ipfix_action(const struct ofproto_dpif *,
813 struct ofpbuf *odp_actions,
814 const struct flow *);
815 static void add_mirror_actions(struct xlate_ctx *ctx,
816 const struct flow *flow);
817 /* Global variables. */
818 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
820 /* Initial mappings of port to bridge mappings. */
821 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
823 /* Factory functions. */
826 init(const struct shash *iface_hints)
828 struct shash_node *node;
830 /* Make a local copy, since we don't own 'iface_hints' elements. */
831 SHASH_FOR_EACH(node, iface_hints) {
832 const struct iface_hint *orig_hint = node->data;
833 struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
835 new_hint->br_name = xstrdup(orig_hint->br_name);
836 new_hint->br_type = xstrdup(orig_hint->br_type);
837 new_hint->ofp_port = orig_hint->ofp_port;
839 shash_add(&init_ofp_ports, node->name, new_hint);
844 enumerate_types(struct sset *types)
846 dp_enumerate_types(types);
850 enumerate_names(const char *type, struct sset *names)
852 struct ofproto_dpif *ofproto;
855 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
856 if (strcmp(type, ofproto->up.type)) {
859 sset_add(names, ofproto->up.name);
866 del(const char *type, const char *name)
871 error = dpif_open(name, type, &dpif);
873 error = dpif_delete(dpif);
880 port_open_type(const char *datapath_type, const char *port_type)
882 return dpif_port_open_type(datapath_type, port_type);
885 /* Type functions. */
887 static struct ofproto_dpif *
888 lookup_ofproto_dpif_by_port_name(const char *name)
890 struct ofproto_dpif *ofproto;
892 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
893 if (sset_contains(&ofproto->ports, name)) {
902 type_run(const char *type)
904 static long long int push_timer = LLONG_MIN;
905 struct dpif_backer *backer;
909 backer = shash_find_data(&all_dpif_backers, type);
911 /* This is not necessarily a problem, since backers are only
912 * created on demand. */
916 dpif_run(backer->dpif);
918 /* The most natural place to push facet statistics is when they're pulled
919 * from the datapath. However, when there are many flows in the datapath,
920 * this expensive operation can occur so frequently, that it reduces our
921 * ability to quickly set up flows. To reduce the cost, we push statistics
923 if (time_msec() > push_timer) {
924 push_timer = time_msec() + 2000;
928 /* If vswitchd started with other_config:flow_restore_wait set as "true",
929 * and the configuration has now changed to "false", enable receiving
930 * packets from the datapath. */
931 if (!backer->recv_set_enable && !ofproto_get_flow_restore_wait()) {
932 backer->recv_set_enable = true;
934 error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
936 VLOG_ERR("Failed to enable receiving packets in dpif.");
939 dpif_flow_flush(backer->dpif);
940 backer->need_revalidate = REV_RECONFIGURE;
943 if (backer->need_revalidate
944 || !tag_set_is_empty(&backer->revalidate_set)) {
945 struct tag_set revalidate_set = backer->revalidate_set;
946 bool need_revalidate = backer->need_revalidate;
947 struct ofproto_dpif *ofproto;
948 struct simap_node *node;
949 struct simap tmp_backers;
951 /* Handle tunnel garbage collection. */
952 simap_init(&tmp_backers);
953 simap_swap(&backer->tnl_backers, &tmp_backers);
955 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
956 struct ofport_dpif *iter;
958 if (backer != ofproto->backer) {
962 HMAP_FOR_EACH (iter, up.hmap_node, &ofproto->up.ports) {
965 if (!iter->tnl_port) {
969 dp_port = netdev_vport_get_dpif_port(iter->up.netdev);
970 node = simap_find(&tmp_backers, dp_port);
972 simap_put(&backer->tnl_backers, dp_port, node->data);
973 simap_delete(&tmp_backers, node);
974 node = simap_find(&backer->tnl_backers, dp_port);
976 node = simap_find(&backer->tnl_backers, dp_port);
978 uint32_t odp_port = UINT32_MAX;
980 if (!dpif_port_add(backer->dpif, iter->up.netdev,
982 simap_put(&backer->tnl_backers, dp_port, odp_port);
983 node = simap_find(&backer->tnl_backers, dp_port);
988 iter->odp_port = node ? node->data : OVSP_NONE;
989 if (tnl_port_reconfigure(&iter->up, iter->odp_port,
991 backer->need_revalidate = REV_RECONFIGURE;
996 SIMAP_FOR_EACH (node, &tmp_backers) {
997 dpif_port_del(backer->dpif, node->data);
999 simap_destroy(&tmp_backers);
1001 switch (backer->need_revalidate) {
1002 case REV_RECONFIGURE: COVERAGE_INC(rev_reconfigure); break;
1003 case REV_STP: COVERAGE_INC(rev_stp); break;
1004 case REV_PORT_TOGGLED: COVERAGE_INC(rev_port_toggled); break;
1005 case REV_FLOW_TABLE: COVERAGE_INC(rev_flow_table); break;
1006 case REV_INCONSISTENCY: COVERAGE_INC(rev_inconsistency); break;
1009 if (backer->need_revalidate) {
1010 /* Clear the drop_keys in case we should now be accepting some
1011 * formerly dropped flows. */
1012 drop_key_clear(backer);
1015 /* Clear the revalidation flags. */
1016 tag_set_init(&backer->revalidate_set);
1017 backer->need_revalidate = 0;
1019 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1020 struct facet *facet, *next;
1022 if (ofproto->backer != backer) {
1026 HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &ofproto->facets) {
1028 || tag_set_intersects(&revalidate_set, facet->xout.tags)) {
1029 facet_revalidate(facet);
1036 if (!backer->recv_set_enable) {
1037 /* Wake up before a max of 1000ms. */
1038 timer_set_duration(&backer->next_expiration, 1000);
1039 } else if (timer_expired(&backer->next_expiration)) {
1040 int delay = expire(backer);
1041 timer_set_duration(&backer->next_expiration, delay);
1044 /* Check for port changes in the dpif. */
1045 while ((error = dpif_port_poll(backer->dpif, &devname)) == 0) {
1046 struct ofproto_dpif *ofproto;
1047 struct dpif_port port;
1049 /* Don't report on the datapath's device. */
1050 if (!strcmp(devname, dpif_base_name(backer->dpif))) {
1054 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
1055 &all_ofproto_dpifs) {
1056 if (simap_contains(&ofproto->backer->tnl_backers, devname)) {
1061 ofproto = lookup_ofproto_dpif_by_port_name(devname);
1062 if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
1063 /* The port was removed. If we know the datapath,
1064 * report it through poll_set(). If we don't, it may be
1065 * notifying us of a removal we initiated, so ignore it.
1066 * If there's a pending ENOBUFS, let it stand, since
1067 * everything will be reevaluated. */
1068 if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
1069 sset_add(&ofproto->port_poll_set, devname);
1070 ofproto->port_poll_errno = 0;
1072 } else if (!ofproto) {
1073 /* The port was added, but we don't know with which
1074 * ofproto we should associate it. Delete it. */
1075 dpif_port_del(backer->dpif, port.port_no);
1077 dpif_port_destroy(&port);
1083 if (error != EAGAIN) {
1084 struct ofproto_dpif *ofproto;
1086 /* There was some sort of error, so propagate it to all
1087 * ofprotos that use this backer. */
1088 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
1089 &all_ofproto_dpifs) {
1090 if (ofproto->backer == backer) {
1091 sset_clear(&ofproto->port_poll_set);
1092 ofproto->port_poll_errno = error;
1101 dpif_backer_run_fast(struct dpif_backer *backer, int max_batch)
1105 /* If recv_set_enable is false, we should not handle upcalls. */
1106 if (!backer->recv_set_enable) {
1110 /* Handle one or more batches of upcalls, until there's nothing left to do
1111 * or until we do a fixed total amount of work.
1113 * We do work in batches because it can be much cheaper to set up a number
1114 * of flows and fire off their patches all at once. We do multiple batches
1115 * because in some cases handling a packet can cause another packet to be
1116 * queued almost immediately as part of the return flow. Both
1117 * optimizations can make major improvements on some benchmarks and
1118 * presumably for real traffic as well. */
1120 while (work < max_batch) {
1121 int retval = handle_upcalls(backer, max_batch - work);
1132 type_run_fast(const char *type)
1134 struct dpif_backer *backer;
1136 backer = shash_find_data(&all_dpif_backers, type);
1138 /* This is not necessarily a problem, since backers are only
1139 * created on demand. */
1143 return dpif_backer_run_fast(backer, FLOW_MISS_MAX_BATCH);
1149 static long long int port_rl = LLONG_MIN;
1150 static unsigned int backer_rl = 0;
1152 if (time_msec() >= port_rl) {
1153 struct ofproto_dpif *ofproto;
1154 struct ofport_dpif *ofport;
1156 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1158 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1159 port_run_fast(ofport);
1162 port_rl = time_msec() + 200;
1165 /* XXX: We have to be careful not to do too much work in this function. If
1166 * we call dpif_backer_run_fast() too often, or with too large a batch,
1167 * performance improves signifcantly, but at a cost. It's possible for the
1168 * number of flows in the datapath to increase without bound, and for poll
1169 * loops to take 10s of seconds. The correct solution to this problem,
1170 * long term, is to separate flow miss handling into it's own thread so it
1171 * isn't affected by revalidations, and expirations. Until then, this is
1172 * the best we can do. */
1173 if (++backer_rl >= 10) {
1174 struct shash_node *node;
1177 SHASH_FOR_EACH (node, &all_dpif_backers) {
1178 dpif_backer_run_fast(node->data, 1);
1184 type_wait(const char *type)
1186 struct dpif_backer *backer;
1188 backer = shash_find_data(&all_dpif_backers, type);
1190 /* This is not necessarily a problem, since backers are only
1191 * created on demand. */
1195 timer_wait(&backer->next_expiration);
1198 /* Basic life-cycle. */
1200 static int add_internal_flows(struct ofproto_dpif *);
1202 static struct ofproto *
1205 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
1206 return &ofproto->up;
1210 dealloc(struct ofproto *ofproto_)
1212 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1217 close_dpif_backer(struct dpif_backer *backer)
1219 struct shash_node *node;
1221 ovs_assert(backer->refcount > 0);
1223 if (--backer->refcount) {
1227 drop_key_clear(backer);
1228 hmap_destroy(&backer->drop_keys);
1230 simap_destroy(&backer->tnl_backers);
1231 hmap_destroy(&backer->odp_to_ofport_map);
1232 node = shash_find(&all_dpif_backers, backer->type);
1234 shash_delete(&all_dpif_backers, node);
1235 dpif_close(backer->dpif);
1240 /* Datapath port slated for removal from datapath. */
1241 struct odp_garbage {
1242 struct list list_node;
1247 open_dpif_backer(const char *type, struct dpif_backer **backerp)
1249 struct dpif_backer *backer;
1250 struct dpif_port_dump port_dump;
1251 struct dpif_port port;
1252 struct shash_node *node;
1253 struct list garbage_list;
1254 struct odp_garbage *garbage, *next;
1260 backer = shash_find_data(&all_dpif_backers, type);
1267 backer_name = xasprintf("ovs-%s", type);
1269 /* Remove any existing datapaths, since we assume we're the only
1270 * userspace controlling the datapath. */
1272 dp_enumerate_names(type, &names);
1273 SSET_FOR_EACH(name, &names) {
1274 struct dpif *old_dpif;
1276 /* Don't remove our backer if it exists. */
1277 if (!strcmp(name, backer_name)) {
1281 if (dpif_open(name, type, &old_dpif)) {
1282 VLOG_WARN("couldn't open old datapath %s to remove it", name);
1284 dpif_delete(old_dpif);
1285 dpif_close(old_dpif);
1288 sset_destroy(&names);
1290 backer = xmalloc(sizeof *backer);
1292 error = dpif_create_and_open(backer_name, type, &backer->dpif);
1295 VLOG_ERR("failed to open datapath of type %s: %s", type,
1301 backer->type = xstrdup(type);
1302 backer->refcount = 1;
1303 hmap_init(&backer->odp_to_ofport_map);
1304 hmap_init(&backer->drop_keys);
1305 timer_set_duration(&backer->next_expiration, 1000);
1306 backer->need_revalidate = 0;
1307 simap_init(&backer->tnl_backers);
1308 tag_set_init(&backer->revalidate_set);
1309 backer->recv_set_enable = !ofproto_get_flow_restore_wait();
1312 if (backer->recv_set_enable) {
1313 dpif_flow_flush(backer->dpif);
1316 /* Loop through the ports already on the datapath and remove any
1317 * that we don't need anymore. */
1318 list_init(&garbage_list);
1319 dpif_port_dump_start(&port_dump, backer->dpif);
1320 while (dpif_port_dump_next(&port_dump, &port)) {
1321 node = shash_find(&init_ofp_ports, port.name);
1322 if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
1323 garbage = xmalloc(sizeof *garbage);
1324 garbage->odp_port = port.port_no;
1325 list_push_front(&garbage_list, &garbage->list_node);
1328 dpif_port_dump_done(&port_dump);
1330 LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) {
1331 dpif_port_del(backer->dpif, garbage->odp_port);
1332 list_remove(&garbage->list_node);
1336 shash_add(&all_dpif_backers, type, backer);
1338 error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
1340 VLOG_ERR("failed to listen on datapath of type %s: %s",
1341 type, strerror(error));
1342 close_dpif_backer(backer);
1350 construct(struct ofproto *ofproto_)
1352 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1353 struct shash_node *node, *next;
1358 error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1363 max_ports = dpif_get_max_ports(ofproto->backer->dpif);
1364 ofproto_init_max_ports(ofproto_, MIN(max_ports, OFPP_MAX));
1366 ofproto->netflow = NULL;
1367 ofproto->sflow = NULL;
1368 ofproto->ipfix = NULL;
1369 ofproto->stp = NULL;
1370 hmap_init(&ofproto->bundles);
1371 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1372 for (i = 0; i < MAX_MIRRORS; i++) {
1373 ofproto->mirrors[i] = NULL;
1375 ofproto->has_bonded_bundles = false;
1377 hmap_init(&ofproto->facets);
1378 hmap_init(&ofproto->subfacets);
1379 ofproto->governor = NULL;
1380 ofproto->consistency_rl = LLONG_MIN;
1382 for (i = 0; i < N_TABLES; i++) {
1383 struct table_dpif *table = &ofproto->tables[i];
1385 table->catchall_table = NULL;
1386 table->other_table = NULL;
1387 table->basis = random_uint32();
1390 list_init(&ofproto->completions);
1392 ofproto_dpif_unixctl_init();
1394 ofproto->has_mirrors = false;
1395 ofproto->has_bundle_action = false;
1397 hmap_init(&ofproto->vlandev_map);
1398 hmap_init(&ofproto->realdev_vid_map);
1400 sset_init(&ofproto->ports);
1401 sset_init(&ofproto->ghost_ports);
1402 sset_init(&ofproto->port_poll_set);
1403 ofproto->port_poll_errno = 0;
1405 SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1406 struct iface_hint *iface_hint = node->data;
1408 if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1409 /* Check if the datapath already has this port. */
1410 if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1411 sset_add(&ofproto->ports, node->name);
1414 free(iface_hint->br_name);
1415 free(iface_hint->br_type);
1417 shash_delete(&init_ofp_ports, node);
1421 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1422 hash_string(ofproto->up.name, 0));
1423 memset(&ofproto->stats, 0, sizeof ofproto->stats);
1425 ofproto_init_tables(ofproto_, N_TABLES);
1426 error = add_internal_flows(ofproto);
1427 ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1430 ofproto->n_missed = 0;
1432 ofproto->max_n_subfacet = 0;
1433 ofproto->created = time_msec();
1434 ofproto->last_minute = ofproto->created;
1435 memset(&ofproto->hourly, 0, sizeof ofproto->hourly);
1436 memset(&ofproto->daily, 0, sizeof ofproto->daily);
1437 ofproto->subfacet_add_count = 0;
1438 ofproto->subfacet_del_count = 0;
1439 ofproto->total_subfacet_add_count = 0;
1440 ofproto->total_subfacet_del_count = 0;
1441 ofproto->total_subfacet_life_span = 0;
1442 ofproto->total_subfacet_count = 0;
1443 ofproto->n_update_stats = 0;
1449 add_internal_flow(struct ofproto_dpif *ofproto, int id,
1450 const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1452 struct ofputil_flow_mod fm;
1455 match_init_catchall(&fm.match);
1457 match_set_reg(&fm.match, 0, id);
1458 fm.new_cookie = htonll(0);
1459 fm.cookie = htonll(0);
1460 fm.cookie_mask = htonll(0);
1461 fm.table_id = TBL_INTERNAL;
1462 fm.command = OFPFC_ADD;
1463 fm.idle_timeout = 0;
1464 fm.hard_timeout = 0;
1468 fm.ofpacts = ofpacts->data;
1469 fm.ofpacts_len = ofpacts->size;
1471 error = ofproto_flow_mod(&ofproto->up, &fm);
1473 VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
1474 id, ofperr_to_string(error));
1478 *rulep = rule_dpif_lookup__(ofproto, &fm.match.flow, TBL_INTERNAL);
1479 ovs_assert(*rulep != NULL);
1485 add_internal_flows(struct ofproto_dpif *ofproto)
1487 struct ofpact_controller *controller;
1488 uint64_t ofpacts_stub[128 / 8];
1489 struct ofpbuf ofpacts;
1493 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1496 controller = ofpact_put_CONTROLLER(&ofpacts);
1497 controller->max_len = UINT16_MAX;
1498 controller->controller_id = 0;
1499 controller->reason = OFPR_NO_MATCH;
1500 ofpact_pad(&ofpacts);
1502 error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule);
1507 ofpbuf_clear(&ofpacts);
1508 error = add_internal_flow(ofproto, id++, &ofpacts,
1509 &ofproto->no_packet_in_rule);
1514 error = add_internal_flow(ofproto, id++, &ofpacts,
1515 &ofproto->drop_frags_rule);
1520 complete_operations(struct ofproto_dpif *ofproto)
1522 struct dpif_completion *c, *next;
1524 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
1525 ofoperation_complete(c->op, 0);
1526 list_remove(&c->list_node);
1532 destruct(struct ofproto *ofproto_)
1534 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1535 struct rule_dpif *rule, *next_rule;
1536 struct oftable *table;
1539 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1540 complete_operations(ofproto);
1542 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1543 struct cls_cursor cursor;
1545 cls_cursor_init(&cursor, &table->cls, NULL);
1546 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1547 ofproto_rule_destroy(&rule->up);
1551 for (i = 0; i < MAX_MIRRORS; i++) {
1552 mirror_destroy(ofproto->mirrors[i]);
1555 netflow_destroy(ofproto->netflow);
1556 dpif_sflow_destroy(ofproto->sflow);
1557 hmap_destroy(&ofproto->bundles);
1558 mac_learning_destroy(ofproto->ml);
1560 hmap_destroy(&ofproto->facets);
1561 hmap_destroy(&ofproto->subfacets);
1562 governor_destroy(ofproto->governor);
1564 hmap_destroy(&ofproto->vlandev_map);
1565 hmap_destroy(&ofproto->realdev_vid_map);
1567 sset_destroy(&ofproto->ports);
1568 sset_destroy(&ofproto->ghost_ports);
1569 sset_destroy(&ofproto->port_poll_set);
1571 close_dpif_backer(ofproto->backer);
1575 run_fast(struct ofproto *ofproto_)
1577 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1578 struct ofport_dpif *ofport;
1580 /* Do not perform any periodic activity required by 'ofproto' while
1581 * waiting for flow restore to complete. */
1582 if (ofproto_get_flow_restore_wait()) {
1586 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1587 port_run_fast(ofport);
1594 run(struct ofproto *ofproto_)
1596 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1597 struct ofport_dpif *ofport;
1598 struct ofbundle *bundle;
1602 complete_operations(ofproto);
1605 /* Do not perform any periodic activity below required by 'ofproto' while
1606 * waiting for flow restore to complete. */
1607 if (ofproto_get_flow_restore_wait()) {
1611 error = run_fast(ofproto_);
1616 if (ofproto->netflow) {
1617 if (netflow_run(ofproto->netflow)) {
1618 send_netflow_active_timeouts(ofproto);
1621 if (ofproto->sflow) {
1622 dpif_sflow_run(ofproto->sflow);
1625 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1628 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1633 mac_learning_run(ofproto->ml, &ofproto->backer->revalidate_set);
1635 /* Check the consistency of a random facet, to aid debugging. */
1636 if (time_msec() >= ofproto->consistency_rl
1637 && !hmap_is_empty(&ofproto->facets)
1638 && !ofproto->backer->need_revalidate) {
1639 struct facet *facet;
1641 ofproto->consistency_rl = time_msec() + 250;
1643 facet = CONTAINER_OF(hmap_random_node(&ofproto->facets),
1644 struct facet, hmap_node);
1645 if (!tag_set_intersects(&ofproto->backer->revalidate_set,
1646 facet->xout.tags)) {
1647 if (!facet_check_consistency(facet)) {
1648 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
1653 if (ofproto->governor) {
1656 governor_run(ofproto->governor);
1658 /* If the governor has shrunk to its minimum size and the number of
1659 * subfacets has dwindled, then drop the governor entirely.
1661 * For hysteresis, the number of subfacets to drop the governor is
1662 * smaller than the number needed to trigger its creation. */
1663 n_subfacets = hmap_count(&ofproto->subfacets);
1664 if (n_subfacets * 4 < ofproto->up.flow_eviction_threshold
1665 && governor_is_idle(ofproto->governor)) {
1666 governor_destroy(ofproto->governor);
1667 ofproto->governor = NULL;
1675 wait(struct ofproto *ofproto_)
1677 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1678 struct ofport_dpif *ofport;
1679 struct ofbundle *bundle;
1681 if (!clogged && !list_is_empty(&ofproto->completions)) {
1682 poll_immediate_wake();
1685 if (ofproto_get_flow_restore_wait()) {
1689 dpif_wait(ofproto->backer->dpif);
1690 dpif_recv_wait(ofproto->backer->dpif);
1691 if (ofproto->sflow) {
1692 dpif_sflow_wait(ofproto->sflow);
1694 if (!tag_set_is_empty(&ofproto->backer->revalidate_set)) {
1695 poll_immediate_wake();
1697 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1700 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1701 bundle_wait(bundle);
1703 if (ofproto->netflow) {
1704 netflow_wait(ofproto->netflow);
1706 mac_learning_wait(ofproto->ml);
1708 if (ofproto->backer->need_revalidate) {
1709 /* Shouldn't happen, but if it does just go around again. */
1710 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1711 poll_immediate_wake();
1713 if (ofproto->governor) {
1714 governor_wait(ofproto->governor);
1719 get_memory_usage(const struct ofproto *ofproto_, struct simap *usage)
1721 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1723 simap_increase(usage, "facets", hmap_count(&ofproto->facets));
1724 simap_increase(usage, "subfacets", hmap_count(&ofproto->subfacets));
1728 flush(struct ofproto *ofproto_)
1730 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1731 struct subfacet *subfacet, *next_subfacet;
1732 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
1736 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
1737 &ofproto->subfacets) {
1738 if (subfacet->path != SF_NOT_INSTALLED) {
1739 batch[n_batch++] = subfacet;
1740 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
1741 subfacet_destroy_batch(ofproto, batch, n_batch);
1745 subfacet_destroy(subfacet);
1750 subfacet_destroy_batch(ofproto, batch, n_batch);
1755 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1756 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1758 *arp_match_ip = true;
1759 *actions = (OFPUTIL_A_OUTPUT |
1760 OFPUTIL_A_SET_VLAN_VID |
1761 OFPUTIL_A_SET_VLAN_PCP |
1762 OFPUTIL_A_STRIP_VLAN |
1763 OFPUTIL_A_SET_DL_SRC |
1764 OFPUTIL_A_SET_DL_DST |
1765 OFPUTIL_A_SET_NW_SRC |
1766 OFPUTIL_A_SET_NW_DST |
1767 OFPUTIL_A_SET_NW_TOS |
1768 OFPUTIL_A_SET_TP_SRC |
1769 OFPUTIL_A_SET_TP_DST |
1774 get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots)
1776 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1777 struct dpif_dp_stats s;
1778 uint64_t n_miss, n_no_pkt_in, n_bytes, n_dropped_frags;
1781 strcpy(ots->name, "classifier");
1783 dpif_get_dp_stats(ofproto->backer->dpif, &s);
1784 rule_get_stats(&ofproto->miss_rule->up, &n_miss, &n_bytes);
1785 rule_get_stats(&ofproto->no_packet_in_rule->up, &n_no_pkt_in, &n_bytes);
1786 rule_get_stats(&ofproto->drop_frags_rule->up, &n_dropped_frags, &n_bytes);
1788 n_lookup = s.n_hit + s.n_missed - n_dropped_frags;
1789 ots->lookup_count = htonll(n_lookup);
1790 ots->matched_count = htonll(n_lookup - n_miss - n_no_pkt_in);
1793 static struct ofport *
1796 struct ofport_dpif *port = xmalloc(sizeof *port);
1801 port_dealloc(struct ofport *port_)
1803 struct ofport_dpif *port = ofport_dpif_cast(port_);
1808 port_construct(struct ofport *port_)
1810 struct ofport_dpif *port = ofport_dpif_cast(port_);
1811 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1812 const struct netdev *netdev = port->up.netdev;
1813 struct dpif_port dpif_port;
1816 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1817 port->bundle = NULL;
1819 port->tag = tag_create_random();
1820 port->may_enable = true;
1821 port->stp_port = NULL;
1822 port->stp_state = STP_DISABLED;
1823 port->tnl_port = NULL;
1824 hmap_init(&port->priorities);
1825 port->realdev_ofp_port = 0;
1826 port->vlandev_vid = 0;
1827 port->carrier_seq = netdev_get_carrier_resets(netdev);
1829 if (netdev_vport_is_patch(netdev)) {
1830 /* By bailing out here, we don't submit the port to the sFlow module
1831 * to be considered for counter polling export. This is correct
1832 * because the patch port represents an interface that sFlow considers
1833 * to be "internal" to the switch as a whole, and therefore not an
1834 * candidate for counter polling. */
1835 port->odp_port = OVSP_NONE;
1839 error = dpif_port_query_by_name(ofproto->backer->dpif,
1840 netdev_vport_get_dpif_port(netdev),
1846 port->odp_port = dpif_port.port_no;
1848 if (netdev_get_tunnel_config(netdev)) {
1849 port->tnl_port = tnl_port_add(&port->up, port->odp_port);
1851 /* Sanity-check that a mapping doesn't already exist. This
1852 * shouldn't happen for non-tunnel ports. */
1853 if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1854 VLOG_ERR("port %s already has an OpenFlow port number",
1856 dpif_port_destroy(&dpif_port);
1860 hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1861 hash_int(port->odp_port, 0));
1863 dpif_port_destroy(&dpif_port);
1865 if (ofproto->sflow) {
1866 dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1873 port_destruct(struct ofport *port_)
1875 struct ofport_dpif *port = ofport_dpif_cast(port_);
1876 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1877 const char *dp_port_name = netdev_vport_get_dpif_port(port->up.netdev);
1878 const char *devname = netdev_get_name(port->up.netdev);
1880 if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1881 /* The underlying device is still there, so delete it. This
1882 * happens when the ofproto is being destroyed, since the caller
1883 * assumes that removal of attached ports will happen as part of
1885 if (!port->tnl_port) {
1886 dpif_port_del(ofproto->backer->dpif, port->odp_port);
1888 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1891 if (port->odp_port != OVSP_NONE && !port->tnl_port) {
1892 hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1895 tnl_port_del(port->tnl_port);
1896 sset_find_and_delete(&ofproto->ports, devname);
1897 sset_find_and_delete(&ofproto->ghost_ports, devname);
1898 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1899 bundle_remove(port_);
1900 set_cfm(port_, NULL);
1901 if (ofproto->sflow) {
1902 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1905 ofport_clear_priorities(port);
1906 hmap_destroy(&port->priorities);
1910 port_modified(struct ofport *port_)
1912 struct ofport_dpif *port = ofport_dpif_cast(port_);
1914 if (port->bundle && port->bundle->bond) {
1915 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1919 cfm_set_netdev(port->cfm, port->up.netdev);
1924 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1926 struct ofport_dpif *port = ofport_dpif_cast(port_);
1927 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1928 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1930 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1931 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1932 OFPUTIL_PC_NO_PACKET_IN)) {
1933 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1935 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1936 bundle_update(port->bundle);
1942 set_sflow(struct ofproto *ofproto_,
1943 const struct ofproto_sflow_options *sflow_options)
1945 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1946 struct dpif_sflow *ds = ofproto->sflow;
1948 if (sflow_options) {
1950 struct ofport_dpif *ofport;
1952 ds = ofproto->sflow = dpif_sflow_create();
1953 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1954 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1956 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1958 dpif_sflow_set_options(ds, sflow_options);
1961 dpif_sflow_destroy(ds);
1962 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1963 ofproto->sflow = NULL;
1971 struct ofproto *ofproto_,
1972 const struct ofproto_ipfix_bridge_exporter_options *bridge_exporter_options,
1973 const struct ofproto_ipfix_flow_exporter_options *flow_exporters_options,
1974 size_t n_flow_exporters_options)
1976 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1977 struct dpif_ipfix *di = ofproto->ipfix;
1979 if (bridge_exporter_options || flow_exporters_options) {
1981 di = ofproto->ipfix = dpif_ipfix_create();
1983 dpif_ipfix_set_options(
1984 di, bridge_exporter_options, flow_exporters_options,
1985 n_flow_exporters_options);
1988 dpif_ipfix_destroy(di);
1989 ofproto->ipfix = NULL;
1996 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1998 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2005 struct ofproto_dpif *ofproto;
2007 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2008 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2009 ofport->cfm = cfm_create(ofport->up.netdev);
2012 if (cfm_configure(ofport->cfm, s)) {
2018 cfm_destroy(ofport->cfm);
2024 get_cfm_status(const struct ofport *ofport_,
2025 struct ofproto_cfm_status *status)
2027 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2030 status->faults = cfm_get_fault(ofport->cfm);
2031 status->remote_opstate = cfm_get_opup(ofport->cfm);
2032 status->health = cfm_get_health(ofport->cfm);
2033 cfm_get_remote_mpids(ofport->cfm, &status->rmps, &status->n_rmps);
2040 /* Spanning Tree. */
2043 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
2045 struct ofproto_dpif *ofproto = ofproto_;
2046 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
2047 struct ofport_dpif *ofport;
2049 ofport = stp_port_get_aux(sp);
2051 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
2052 ofproto->up.name, port_num);
2054 struct eth_header *eth = pkt->l2;
2056 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
2057 if (eth_addr_is_zero(eth->eth_src)) {
2058 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
2059 "with unknown MAC", ofproto->up.name, port_num);
2061 send_packet(ofport, pkt);
2067 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
2069 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
2071 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2073 /* Only revalidate flows if the configuration changed. */
2074 if (!s != !ofproto->stp) {
2075 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2079 if (!ofproto->stp) {
2080 ofproto->stp = stp_create(ofproto_->name, s->system_id,
2081 send_bpdu_cb, ofproto);
2082 ofproto->stp_last_tick = time_msec();
2085 stp_set_bridge_id(ofproto->stp, s->system_id);
2086 stp_set_bridge_priority(ofproto->stp, s->priority);
2087 stp_set_hello_time(ofproto->stp, s->hello_time);
2088 stp_set_max_age(ofproto->stp, s->max_age);
2089 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
2091 struct ofport *ofport;
2093 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
2094 set_stp_port(ofport, NULL);
2097 stp_destroy(ofproto->stp);
2098 ofproto->stp = NULL;
2105 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
2107 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2111 s->bridge_id = stp_get_bridge_id(ofproto->stp);
2112 s->designated_root = stp_get_designated_root(ofproto->stp);
2113 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
2122 update_stp_port_state(struct ofport_dpif *ofport)
2124 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2125 enum stp_state state;
2127 /* Figure out new state. */
2128 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
2132 if (ofport->stp_state != state) {
2133 enum ofputil_port_state of_state;
2136 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
2137 netdev_get_name(ofport->up.netdev),
2138 stp_state_name(ofport->stp_state),
2139 stp_state_name(state));
2140 if (stp_learn_in_state(ofport->stp_state)
2141 != stp_learn_in_state(state)) {
2142 /* xxx Learning action flows should also be flushed. */
2143 mac_learning_flush(ofproto->ml,
2144 &ofproto->backer->revalidate_set);
2146 fwd_change = stp_forward_in_state(ofport->stp_state)
2147 != stp_forward_in_state(state);
2149 ofproto->backer->need_revalidate = REV_STP;
2150 ofport->stp_state = state;
2151 ofport->stp_state_entered = time_msec();
2153 if (fwd_change && ofport->bundle) {
2154 bundle_update(ofport->bundle);
2157 /* Update the STP state bits in the OpenFlow port description. */
2158 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
2159 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
2160 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
2161 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
2162 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
2164 ofproto_port_set_state(&ofport->up, of_state);
2168 /* Configures STP on 'ofport_' using the settings defined in 's'. The
2169 * caller is responsible for assigning STP port numbers and ensuring
2170 * there are no duplicates. */
2172 set_stp_port(struct ofport *ofport_,
2173 const struct ofproto_port_stp_settings *s)
2175 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2176 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2177 struct stp_port *sp = ofport->stp_port;
2179 if (!s || !s->enable) {
2181 ofport->stp_port = NULL;
2182 stp_port_disable(sp);
2183 update_stp_port_state(ofport);
2186 } else if (sp && stp_port_no(sp) != s->port_num
2187 && ofport == stp_port_get_aux(sp)) {
2188 /* The port-id changed, so disable the old one if it's not
2189 * already in use by another port. */
2190 stp_port_disable(sp);
2193 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
2194 stp_port_enable(sp);
2196 stp_port_set_aux(sp, ofport);
2197 stp_port_set_priority(sp, s->priority);
2198 stp_port_set_path_cost(sp, s->path_cost);
2200 update_stp_port_state(ofport);
2206 get_stp_port_status(struct ofport *ofport_,
2207 struct ofproto_port_stp_status *s)
2209 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2210 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2211 struct stp_port *sp = ofport->stp_port;
2213 if (!ofproto->stp || !sp) {
2219 s->port_id = stp_port_get_id(sp);
2220 s->state = stp_port_get_state(sp);
2221 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
2222 s->role = stp_port_get_role(sp);
2223 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
2229 stp_run(struct ofproto_dpif *ofproto)
2232 long long int now = time_msec();
2233 long long int elapsed = now - ofproto->stp_last_tick;
2234 struct stp_port *sp;
2237 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
2238 ofproto->stp_last_tick = now;
2240 while (stp_get_changed_port(ofproto->stp, &sp)) {
2241 struct ofport_dpif *ofport = stp_port_get_aux(sp);
2244 update_stp_port_state(ofport);
2248 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
2249 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
2255 stp_wait(struct ofproto_dpif *ofproto)
2258 poll_timer_wait(1000);
2262 /* Returns true if STP should process 'flow'. */
2264 stp_should_process_flow(const struct flow *flow)
2266 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
2270 stp_process_packet(const struct ofport_dpif *ofport,
2271 const struct ofpbuf *packet)
2273 struct ofpbuf payload = *packet;
2274 struct eth_header *eth = payload.data;
2275 struct stp_port *sp = ofport->stp_port;
2277 /* Sink packets on ports that have STP disabled when the bridge has
2279 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
2283 /* Trim off padding on payload. */
2284 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
2285 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
2288 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
2289 stp_received_bpdu(sp, payload.data, payload.size);
2293 static struct priority_to_dscp *
2294 get_priority(const struct ofport_dpif *ofport, uint32_t priority)
2296 struct priority_to_dscp *pdscp;
2299 hash = hash_int(priority, 0);
2300 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) {
2301 if (pdscp->priority == priority) {
2309 ofport_clear_priorities(struct ofport_dpif *ofport)
2311 struct priority_to_dscp *pdscp, *next;
2313 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) {
2314 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
2320 set_queues(struct ofport *ofport_,
2321 const struct ofproto_port_queue *qdscp_list,
2324 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2325 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2326 struct hmap new = HMAP_INITIALIZER(&new);
2329 for (i = 0; i < n_qdscp; i++) {
2330 struct priority_to_dscp *pdscp;
2334 dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
2335 if (dpif_queue_to_priority(ofproto->backer->dpif, qdscp_list[i].queue,
2340 pdscp = get_priority(ofport, priority);
2342 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
2344 pdscp = xmalloc(sizeof *pdscp);
2345 pdscp->priority = priority;
2347 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2350 if (pdscp->dscp != dscp) {
2352 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2355 hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0));
2358 if (!hmap_is_empty(&ofport->priorities)) {
2359 ofport_clear_priorities(ofport);
2360 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2363 hmap_swap(&new, &ofport->priorities);
2371 /* Expires all MAC learning entries associated with 'bundle' and forces its
2372 * ofproto to revalidate every flow.
2374 * Normally MAC learning entries are removed only from the ofproto associated
2375 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2376 * are removed from every ofproto. When patch ports and SLB bonds are in use
2377 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2378 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2379 * with the host from which it migrated. */
2381 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2383 struct ofproto_dpif *ofproto = bundle->ofproto;
2384 struct mac_learning *ml = ofproto->ml;
2385 struct mac_entry *mac, *next_mac;
2387 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2388 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2389 if (mac->port.p == bundle) {
2391 struct ofproto_dpif *o;
2393 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2395 struct mac_entry *e;
2397 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan,
2400 mac_learning_expire(o->ml, e);
2406 mac_learning_expire(ml, mac);
2411 static struct ofbundle *
2412 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2414 struct ofbundle *bundle;
2416 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2417 &ofproto->bundles) {
2418 if (bundle->aux == aux) {
2425 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
2426 * ones that are found to 'bundles'. */
2428 bundle_lookup_multiple(struct ofproto_dpif *ofproto,
2429 void **auxes, size_t n_auxes,
2430 struct hmapx *bundles)
2434 hmapx_init(bundles);
2435 for (i = 0; i < n_auxes; i++) {
2436 struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
2438 hmapx_add(bundles, bundle);
2444 bundle_update(struct ofbundle *bundle)
2446 struct ofport_dpif *port;
2448 bundle->floodable = true;
2449 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2450 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2451 || !stp_forward_in_state(port->stp_state)) {
2452 bundle->floodable = false;
2459 bundle_del_port(struct ofport_dpif *port)
2461 struct ofbundle *bundle = port->bundle;
2463 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2465 list_remove(&port->bundle_node);
2466 port->bundle = NULL;
2469 lacp_slave_unregister(bundle->lacp, port);
2472 bond_slave_unregister(bundle->bond, port);
2475 bundle_update(bundle);
2479 bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port,
2480 struct lacp_slave_settings *lacp)
2482 struct ofport_dpif *port;
2484 port = get_ofp_port(bundle->ofproto, ofp_port);
2489 if (port->bundle != bundle) {
2490 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2492 bundle_del_port(port);
2495 port->bundle = bundle;
2496 list_push_back(&bundle->ports, &port->bundle_node);
2497 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2498 || !stp_forward_in_state(port->stp_state)) {
2499 bundle->floodable = false;
2503 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2504 lacp_slave_register(bundle->lacp, port, lacp);
2511 bundle_destroy(struct ofbundle *bundle)
2513 struct ofproto_dpif *ofproto;
2514 struct ofport_dpif *port, *next_port;
2521 ofproto = bundle->ofproto;
2522 for (i = 0; i < MAX_MIRRORS; i++) {
2523 struct ofmirror *m = ofproto->mirrors[i];
2525 if (m->out == bundle) {
2527 } else if (hmapx_find_and_delete(&m->srcs, bundle)
2528 || hmapx_find_and_delete(&m->dsts, bundle)) {
2529 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2534 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2535 bundle_del_port(port);
2538 bundle_flush_macs(bundle, true);
2539 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2541 free(bundle->trunks);
2542 lacp_destroy(bundle->lacp);
2543 bond_destroy(bundle->bond);
2548 bundle_set(struct ofproto *ofproto_, void *aux,
2549 const struct ofproto_bundle_settings *s)
2551 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2552 bool need_flush = false;
2553 struct ofport_dpif *port;
2554 struct ofbundle *bundle;
2555 unsigned long *trunks;
2561 bundle_destroy(bundle_lookup(ofproto, aux));
2565 ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2566 ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2568 bundle = bundle_lookup(ofproto, aux);
2570 bundle = xmalloc(sizeof *bundle);
2572 bundle->ofproto = ofproto;
2573 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2574 hash_pointer(aux, 0));
2576 bundle->name = NULL;
2578 list_init(&bundle->ports);
2579 bundle->vlan_mode = PORT_VLAN_TRUNK;
2581 bundle->trunks = NULL;
2582 bundle->use_priority_tags = s->use_priority_tags;
2583 bundle->lacp = NULL;
2584 bundle->bond = NULL;
2586 bundle->floodable = true;
2588 bundle->src_mirrors = 0;
2589 bundle->dst_mirrors = 0;
2590 bundle->mirror_out = 0;
2593 if (!bundle->name || strcmp(s->name, bundle->name)) {
2595 bundle->name = xstrdup(s->name);
2600 if (!bundle->lacp) {
2601 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2602 bundle->lacp = lacp_create();
2604 lacp_configure(bundle->lacp, s->lacp);
2606 lacp_destroy(bundle->lacp);
2607 bundle->lacp = NULL;
2610 /* Update set of ports. */
2612 for (i = 0; i < s->n_slaves; i++) {
2613 if (!bundle_add_port(bundle, s->slaves[i],
2614 s->lacp ? &s->lacp_slaves[i] : NULL)) {
2618 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2619 struct ofport_dpif *next_port;
2621 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2622 for (i = 0; i < s->n_slaves; i++) {
2623 if (s->slaves[i] == port->up.ofp_port) {
2628 bundle_del_port(port);
2632 ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2634 if (list_is_empty(&bundle->ports)) {
2635 bundle_destroy(bundle);
2639 /* Set VLAN tagging mode */
2640 if (s->vlan_mode != bundle->vlan_mode
2641 || s->use_priority_tags != bundle->use_priority_tags) {
2642 bundle->vlan_mode = s->vlan_mode;
2643 bundle->use_priority_tags = s->use_priority_tags;
2648 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2649 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2651 if (vlan != bundle->vlan) {
2652 bundle->vlan = vlan;
2656 /* Get trunked VLANs. */
2657 switch (s->vlan_mode) {
2658 case PORT_VLAN_ACCESS:
2662 case PORT_VLAN_TRUNK:
2663 trunks = CONST_CAST(unsigned long *, s->trunks);
2666 case PORT_VLAN_NATIVE_UNTAGGED:
2667 case PORT_VLAN_NATIVE_TAGGED:
2668 if (vlan != 0 && (!s->trunks
2669 || !bitmap_is_set(s->trunks, vlan)
2670 || bitmap_is_set(s->trunks, 0))) {
2671 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2673 trunks = bitmap_clone(s->trunks, 4096);
2675 trunks = bitmap_allocate1(4096);
2677 bitmap_set1(trunks, vlan);
2678 bitmap_set0(trunks, 0);
2680 trunks = CONST_CAST(unsigned long *, s->trunks);
2687 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2688 free(bundle->trunks);
2689 if (trunks == s->trunks) {
2690 bundle->trunks = vlan_bitmap_clone(trunks);
2692 bundle->trunks = trunks;
2697 if (trunks != s->trunks) {
2702 if (!list_is_short(&bundle->ports)) {
2703 bundle->ofproto->has_bonded_bundles = true;
2705 if (bond_reconfigure(bundle->bond, s->bond)) {
2706 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2709 bundle->bond = bond_create(s->bond);
2710 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2713 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2714 bond_slave_register(bundle->bond, port, port->up.netdev);
2717 bond_destroy(bundle->bond);
2718 bundle->bond = NULL;
2721 /* If we changed something that would affect MAC learning, un-learn
2722 * everything on this port and force flow revalidation. */
2724 bundle_flush_macs(bundle, false);
2731 bundle_remove(struct ofport *port_)
2733 struct ofport_dpif *port = ofport_dpif_cast(port_);
2734 struct ofbundle *bundle = port->bundle;
2737 bundle_del_port(port);
2738 if (list_is_empty(&bundle->ports)) {
2739 bundle_destroy(bundle);
2740 } else if (list_is_short(&bundle->ports)) {
2741 bond_destroy(bundle->bond);
2742 bundle->bond = NULL;
2748 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2750 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2751 struct ofport_dpif *port = port_;
2752 uint8_t ea[ETH_ADDR_LEN];
2755 error = netdev_get_etheraddr(port->up.netdev, ea);
2757 struct ofpbuf packet;
2760 ofpbuf_init(&packet, 0);
2761 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2763 memcpy(packet_pdu, pdu, pdu_size);
2765 send_packet(port, &packet);
2766 ofpbuf_uninit(&packet);
2768 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2769 "%s (%s)", port->bundle->name,
2770 netdev_get_name(port->up.netdev), strerror(error));
2775 bundle_send_learning_packets(struct ofbundle *bundle)
2777 struct ofproto_dpif *ofproto = bundle->ofproto;
2778 int error, n_packets, n_errors;
2779 struct mac_entry *e;
2781 error = n_packets = n_errors = 0;
2782 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2783 if (e->port.p != bundle) {
2784 struct ofpbuf *learning_packet;
2785 struct ofport_dpif *port;
2789 /* The assignment to "port" is unnecessary but makes "grep"ing for
2790 * struct ofport_dpif more effective. */
2791 learning_packet = bond_compose_learning_packet(bundle->bond,
2795 ret = send_packet(port, learning_packet);
2796 ofpbuf_delete(learning_packet);
2806 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2807 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2808 "packets, last error was: %s",
2809 bundle->name, n_errors, n_packets, strerror(error));
2811 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2812 bundle->name, n_packets);
2817 bundle_run(struct ofbundle *bundle)
2820 lacp_run(bundle->lacp, send_pdu_cb);
2823 struct ofport_dpif *port;
2825 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2826 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2829 bond_run(bundle->bond, &bundle->ofproto->backer->revalidate_set,
2830 lacp_status(bundle->lacp));
2831 if (bond_should_send_learning_packets(bundle->bond)) {
2832 bundle_send_learning_packets(bundle);
2838 bundle_wait(struct ofbundle *bundle)
2841 lacp_wait(bundle->lacp);
2844 bond_wait(bundle->bond);
2851 mirror_scan(struct ofproto_dpif *ofproto)
2855 for (idx = 0; idx < MAX_MIRRORS; idx++) {
2856 if (!ofproto->mirrors[idx]) {
2863 static struct ofmirror *
2864 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
2868 for (i = 0; i < MAX_MIRRORS; i++) {
2869 struct ofmirror *mirror = ofproto->mirrors[i];
2870 if (mirror && mirror->aux == aux) {
2878 /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */
2880 mirror_update_dups(struct ofproto_dpif *ofproto)
2884 for (i = 0; i < MAX_MIRRORS; i++) {
2885 struct ofmirror *m = ofproto->mirrors[i];
2888 m->dup_mirrors = MIRROR_MASK_C(1) << i;
2892 for (i = 0; i < MAX_MIRRORS; i++) {
2893 struct ofmirror *m1 = ofproto->mirrors[i];
2900 for (j = i + 1; j < MAX_MIRRORS; j++) {
2901 struct ofmirror *m2 = ofproto->mirrors[j];
2903 if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) {
2904 m1->dup_mirrors |= MIRROR_MASK_C(1) << j;
2905 m2->dup_mirrors |= m1->dup_mirrors;
2912 mirror_set(struct ofproto *ofproto_, void *aux,
2913 const struct ofproto_mirror_settings *s)
2915 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2916 mirror_mask_t mirror_bit;
2917 struct ofbundle *bundle;
2918 struct ofmirror *mirror;
2919 struct ofbundle *out;
2920 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
2921 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
2924 mirror = mirror_lookup(ofproto, aux);
2926 mirror_destroy(mirror);
2932 idx = mirror_scan(ofproto);
2934 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
2936 ofproto->up.name, MAX_MIRRORS, s->name);
2940 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
2941 mirror->ofproto = ofproto;
2944 mirror->out_vlan = -1;
2945 mirror->name = NULL;
2948 if (!mirror->name || strcmp(s->name, mirror->name)) {
2950 mirror->name = xstrdup(s->name);
2953 /* Get the new configuration. */
2954 if (s->out_bundle) {
2955 out = bundle_lookup(ofproto, s->out_bundle);
2957 mirror_destroy(mirror);
2963 out_vlan = s->out_vlan;
2965 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
2966 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
2968 /* If the configuration has not changed, do nothing. */
2969 if (hmapx_equals(&srcs, &mirror->srcs)
2970 && hmapx_equals(&dsts, &mirror->dsts)
2971 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
2972 && mirror->out == out
2973 && mirror->out_vlan == out_vlan)
2975 hmapx_destroy(&srcs);
2976 hmapx_destroy(&dsts);
2980 hmapx_swap(&srcs, &mirror->srcs);
2981 hmapx_destroy(&srcs);
2983 hmapx_swap(&dsts, &mirror->dsts);
2984 hmapx_destroy(&dsts);
2986 free(mirror->vlans);
2987 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
2990 mirror->out_vlan = out_vlan;
2992 /* Update bundles. */
2993 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2994 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
2995 if (hmapx_contains(&mirror->srcs, bundle)) {
2996 bundle->src_mirrors |= mirror_bit;
2998 bundle->src_mirrors &= ~mirror_bit;
3001 if (hmapx_contains(&mirror->dsts, bundle)) {
3002 bundle->dst_mirrors |= mirror_bit;
3004 bundle->dst_mirrors &= ~mirror_bit;
3007 if (mirror->out == bundle) {
3008 bundle->mirror_out |= mirror_bit;
3010 bundle->mirror_out &= ~mirror_bit;
3014 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3015 ofproto->has_mirrors = true;
3016 mac_learning_flush(ofproto->ml,
3017 &ofproto->backer->revalidate_set);
3018 mirror_update_dups(ofproto);
3024 mirror_destroy(struct ofmirror *mirror)
3026 struct ofproto_dpif *ofproto;
3027 mirror_mask_t mirror_bit;
3028 struct ofbundle *bundle;
3035 ofproto = mirror->ofproto;
3036 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3037 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
3039 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
3040 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
3041 bundle->src_mirrors &= ~mirror_bit;
3042 bundle->dst_mirrors &= ~mirror_bit;
3043 bundle->mirror_out &= ~mirror_bit;
3046 hmapx_destroy(&mirror->srcs);
3047 hmapx_destroy(&mirror->dsts);
3048 free(mirror->vlans);
3050 ofproto->mirrors[mirror->idx] = NULL;
3054 mirror_update_dups(ofproto);
3056 ofproto->has_mirrors = false;
3057 for (i = 0; i < MAX_MIRRORS; i++) {
3058 if (ofproto->mirrors[i]) {
3059 ofproto->has_mirrors = true;
3066 mirror_get_stats(struct ofproto *ofproto_, void *aux,
3067 uint64_t *packets, uint64_t *bytes)
3069 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3070 struct ofmirror *mirror = mirror_lookup(ofproto, aux);
3073 *packets = *bytes = UINT64_MAX;
3079 *packets = mirror->packet_count;
3080 *bytes = mirror->byte_count;
3086 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
3088 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3089 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
3090 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
3096 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
3098 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3099 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
3100 return bundle && bundle->mirror_out != 0;
3104 forward_bpdu_changed(struct ofproto *ofproto_)
3106 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3107 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3111 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
3114 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3115 mac_learning_set_idle_time(ofproto->ml, idle_time);
3116 mac_learning_set_max_entries(ofproto->ml, max_entries);
3121 static struct ofport_dpif *
3122 get_ofp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
3124 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
3125 return ofport ? ofport_dpif_cast(ofport) : NULL;
3128 static struct ofport_dpif *
3129 get_odp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
3131 struct ofport_dpif *port = odp_port_to_ofport(ofproto->backer, odp_port);
3132 return port && &ofproto->up == port->up.ofproto ? port : NULL;
3136 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
3137 struct ofproto_port *ofproto_port,
3138 struct dpif_port *dpif_port)
3140 ofproto_port->name = dpif_port->name;
3141 ofproto_port->type = dpif_port->type;
3142 ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
3145 static struct ofport_dpif *
3146 ofport_get_peer(const struct ofport_dpif *ofport_dpif)
3148 const struct ofproto_dpif *ofproto;
3151 peer = netdev_vport_patch_peer(ofport_dpif->up.netdev);
3156 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
3157 struct ofport *ofport;
3159 ofport = shash_find_data(&ofproto->up.port_by_name, peer);
3160 if (ofport && ofport->ofproto->ofproto_class == &ofproto_dpif_class) {
3161 return ofport_dpif_cast(ofport);
3168 port_run_fast(struct ofport_dpif *ofport)
3170 if (ofport->cfm && cfm_should_send_ccm(ofport->cfm)) {
3171 struct ofpbuf packet;
3173 ofpbuf_init(&packet, 0);
3174 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
3175 send_packet(ofport, &packet);
3176 ofpbuf_uninit(&packet);
3181 port_run(struct ofport_dpif *ofport)
3183 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
3184 bool carrier_changed = carrier_seq != ofport->carrier_seq;
3185 bool enable = netdev_get_carrier(ofport->up.netdev);
3187 ofport->carrier_seq = carrier_seq;
3189 port_run_fast(ofport);
3191 if (ofport->tnl_port
3192 && tnl_port_reconfigure(&ofport->up, ofport->odp_port,
3193 &ofport->tnl_port)) {
3194 ofproto_dpif_cast(ofport->up.ofproto)->backer->need_revalidate = true;
3198 int cfm_opup = cfm_get_opup(ofport->cfm);
3200 cfm_run(ofport->cfm);
3201 enable = enable && !cfm_get_fault(ofport->cfm);
3203 if (cfm_opup >= 0) {
3204 enable = enable && cfm_opup;
3208 if (ofport->bundle) {
3209 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
3210 if (carrier_changed) {
3211 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
3215 if (ofport->may_enable != enable) {
3216 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3218 if (ofproto->has_bundle_action) {
3219 ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
3223 ofport->may_enable = enable;
3227 port_wait(struct ofport_dpif *ofport)
3230 cfm_wait(ofport->cfm);
3235 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
3236 struct ofproto_port *ofproto_port)
3238 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3239 struct dpif_port dpif_port;
3242 if (sset_contains(&ofproto->ghost_ports, devname)) {
3243 const char *type = netdev_get_type_from_name(devname);
3245 /* We may be called before ofproto->up.port_by_name is populated with
3246 * the appropriate ofport. For this reason, we must get the name and
3247 * type from the netdev layer directly. */
3249 const struct ofport *ofport;
3251 ofport = shash_find_data(&ofproto->up.port_by_name, devname);
3252 ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
3253 ofproto_port->name = xstrdup(devname);
3254 ofproto_port->type = xstrdup(type);
3260 if (!sset_contains(&ofproto->ports, devname)) {
3263 error = dpif_port_query_by_name(ofproto->backer->dpif,
3264 devname, &dpif_port);
3266 ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
3272 port_add(struct ofproto *ofproto_, struct netdev *netdev)
3274 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3275 const char *dp_port_name = netdev_vport_get_dpif_port(netdev);
3276 const char *devname = netdev_get_name(netdev);
3278 if (netdev_vport_is_patch(netdev)) {
3279 sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
3283 if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
3284 uint32_t port_no = UINT32_MAX;
3287 error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
3291 if (netdev_get_tunnel_config(netdev)) {
3292 simap_put(&ofproto->backer->tnl_backers, dp_port_name, port_no);
3296 if (netdev_get_tunnel_config(netdev)) {
3297 sset_add(&ofproto->ghost_ports, devname);
3299 sset_add(&ofproto->ports, devname);
3305 port_del(struct ofproto *ofproto_, uint16_t ofp_port)
3307 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3308 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
3315 sset_find_and_delete(&ofproto->ghost_ports,
3316 netdev_get_name(ofport->up.netdev));
3317 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3318 if (!ofport->tnl_port) {
3319 error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
3321 /* The caller is going to close ofport->up.netdev. If this is a
3322 * bonded port, then the bond is using that netdev, so remove it
3323 * from the bond. The client will need to reconfigure everything
3324 * after deleting ports, so then the slave will get re-added. */
3325 bundle_remove(&ofport->up);
3332 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
3334 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3339 error = netdev_get_stats(ofport->up.netdev, stats);
3341 if (!error && ofport_->ofp_port == OFPP_LOCAL) {
3342 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3344 /* ofproto->stats.tx_packets represents packets that we created
3345 * internally and sent to some port (e.g. packets sent with
3346 * send_packet()). Account for them as if they had come from
3347 * OFPP_LOCAL and got forwarded. */
3349 if (stats->rx_packets != UINT64_MAX) {
3350 stats->rx_packets += ofproto->stats.tx_packets;
3353 if (stats->rx_bytes != UINT64_MAX) {
3354 stats->rx_bytes += ofproto->stats.tx_bytes;
3357 /* ofproto->stats.rx_packets represents packets that were received on
3358 * some port and we processed internally and dropped (e.g. STP).
3359 * Account for them as if they had been forwarded to OFPP_LOCAL. */
3361 if (stats->tx_packets != UINT64_MAX) {
3362 stats->tx_packets += ofproto->stats.rx_packets;
3365 if (stats->tx_bytes != UINT64_MAX) {
3366 stats->tx_bytes += ofproto->stats.rx_bytes;
3373 /* Account packets for LOCAL port. */
3375 ofproto_update_local_port_stats(const struct ofproto *ofproto_,
3376 size_t tx_size, size_t rx_size)
3378 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3381 ofproto->stats.rx_packets++;
3382 ofproto->stats.rx_bytes += rx_size;
3385 ofproto->stats.tx_packets++;
3386 ofproto->stats.tx_bytes += tx_size;
3390 struct port_dump_state {
3395 struct ofproto_port port;
3400 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
3402 *statep = xzalloc(sizeof(struct port_dump_state));
3407 port_dump_next(const struct ofproto *ofproto_, void *state_,
3408 struct ofproto_port *port)
3410 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3411 struct port_dump_state *state = state_;
3412 const struct sset *sset;
3413 struct sset_node *node;
3415 if (state->has_port) {
3416 ofproto_port_destroy(&state->port);
3417 state->has_port = false;
3419 sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
3420 while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
3423 error = port_query_by_name(ofproto_, node->name, &state->port);
3425 *port = state->port;
3426 state->has_port = true;
3428 } else if (error != ENODEV) {
3433 if (!state->ghost) {
3434 state->ghost = true;
3437 return port_dump_next(ofproto_, state_, port);
3444 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3446 struct port_dump_state *state = state_;
3448 if (state->has_port) {
3449 ofproto_port_destroy(&state->port);
3456 port_poll(const struct ofproto *ofproto_, char **devnamep)
3458 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3460 if (ofproto->port_poll_errno) {
3461 int error = ofproto->port_poll_errno;
3462 ofproto->port_poll_errno = 0;
3466 if (sset_is_empty(&ofproto->port_poll_set)) {
3470 *devnamep = sset_pop(&ofproto->port_poll_set);
3475 port_poll_wait(const struct ofproto *ofproto_)
3477 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3478 dpif_port_poll_wait(ofproto->backer->dpif);
3482 port_is_lacp_current(const struct ofport *ofport_)
3484 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3485 return (ofport->bundle && ofport->bundle->lacp
3486 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3490 /* Upcall handling. */
3492 /* Flow miss batching.
3494 * Some dpifs implement operations faster when you hand them off in a batch.
3495 * To allow batching, "struct flow_miss" queues the dpif-related work needed
3496 * for a given flow. Each "struct flow_miss" corresponds to sending one or
3497 * more packets, plus possibly installing the flow in the dpif.
3499 * So far we only batch the operations that affect flow setup time the most.
3500 * It's possible to batch more than that, but the benefit might be minimal. */
3502 struct hmap_node hmap_node;
3503 struct ofproto_dpif *ofproto;
3505 enum odp_key_fitness key_fitness;
3506 const struct nlattr *key;
3508 struct initial_vals initial_vals;
3509 struct list packets;
3510 enum dpif_upcall_type upcall_type;
3511 uint32_t odp_in_port;
3514 struct flow_miss_op {
3515 struct dpif_op dpif_op;
3517 uint64_t slow_stub[128 / 8]; /* Buffer for compose_slow_path() */
3518 struct xlate_out xout;
3519 bool xout_garbage; /* 'xout' needs to be uninitialized? */
3522 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
3523 * OpenFlow controller as necessary according to their individual
3524 * configurations. */
3526 send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet,
3527 const struct flow *flow)
3529 struct ofputil_packet_in pin;
3531 pin.packet = packet->data;
3532 pin.packet_len = packet->size;
3533 pin.reason = OFPR_NO_MATCH;
3534 pin.controller_id = 0;
3539 pin.send_len = 0; /* not used for flow table misses */
3541 flow_get_metadata(flow, &pin.fmd);
3543 connmgr_send_packet_in(ofproto->up.connmgr, &pin);
3546 static enum slow_path_reason
3547 process_special(struct ofproto_dpif *ofproto, const struct flow *flow,
3548 const struct ofport_dpif *ofport, const struct ofpbuf *packet)
3552 } else if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) {
3554 cfm_process_heartbeat(ofport->cfm, packet);
3557 } else if (ofport->bundle && ofport->bundle->lacp
3558 && flow->dl_type == htons(ETH_TYPE_LACP)) {
3560 lacp_process_packet(ofport->bundle->lacp, ofport, packet);
3563 } else if (ofproto->stp && stp_should_process_flow(flow)) {
3565 stp_process_packet(ofport, packet);
3573 static struct flow_miss *
3574 flow_miss_find(struct hmap *todo, const struct ofproto_dpif *ofproto,
3575 const struct flow *flow, uint32_t hash)
3577 struct flow_miss *miss;
3579 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
3580 if (miss->ofproto == ofproto && flow_equal(&miss->flow, flow)) {
3588 /* Partially Initializes 'op' as an "execute" operation for 'miss' and
3589 * 'packet'. The caller must initialize op->actions and op->actions_len. If
3590 * 'miss' is associated with a subfacet the caller must also initialize the
3591 * returned op->subfacet, and if anything needs to be freed after processing
3592 * the op, the caller must initialize op->garbage also. */
3594 init_flow_miss_execute_op(struct flow_miss *miss, struct ofpbuf *packet,
3595 struct flow_miss_op *op)
3597 if (miss->flow.vlan_tci != miss->initial_vals.vlan_tci) {
3598 /* This packet was received on a VLAN splinter port. We
3599 * added a VLAN to the packet to make the packet resemble
3600 * the flow, but the actions were composed assuming that
3601 * the packet contained no VLAN. So, we must remove the
3602 * VLAN header from the packet before trying to execute the
3604 eth_pop_vlan(packet);
3607 op->xout_garbage = false;
3608 op->dpif_op.type = DPIF_OP_EXECUTE;
3609 op->dpif_op.u.execute.key = miss->key;
3610 op->dpif_op.u.execute.key_len = miss->key_len;
3611 op->dpif_op.u.execute.packet = packet;
3614 /* Helper for handle_flow_miss_without_facet() and
3615 * handle_flow_miss_with_facet(). */
3617 handle_flow_miss_common(struct rule_dpif *rule,
3618 struct ofpbuf *packet, const struct flow *flow)
3620 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3622 if (rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
3624 * Extra-special case for fail-open mode.
3626 * We are in fail-open mode and the packet matched the fail-open
3627 * rule, but we are connected to a controller too. We should send
3628 * the packet up to the controller in the hope that it will try to
3629 * set up a flow and thereby allow us to exit fail-open.
3631 * See the top-level comment in fail-open.c for more information.
3633 send_packet_in_miss(ofproto, packet, flow);
3637 /* Figures out whether a flow that missed in 'ofproto', whose details are in
3638 * 'miss', is likely to be worth tracking in detail in userspace and (usually)
3639 * installing a datapath flow. The answer is usually "yes" (a return value of
3640 * true). However, for short flows the cost of bookkeeping is much higher than
3641 * the benefits, so when the datapath holds a large number of flows we impose
3642 * some heuristics to decide which flows are likely to be worth tracking. */
3644 flow_miss_should_make_facet(struct ofproto_dpif *ofproto,
3645 struct flow_miss *miss, uint32_t hash)
3647 if (!ofproto->governor) {
3650 n_subfacets = hmap_count(&ofproto->subfacets);
3651 if (n_subfacets * 2 <= ofproto->up.flow_eviction_threshold) {
3655 ofproto->governor = governor_create(ofproto->up.name);
3658 return governor_should_install_flow(ofproto->governor, hash,
3659 list_size(&miss->packets));
3662 /* Handles 'miss', which matches 'rule', without creating a facet or subfacet
3663 * or creating any datapath flow. May add an "execute" operation to 'ops' and
3664 * increment '*n_ops'. */
3666 handle_flow_miss_without_facet(struct flow_miss *miss,
3667 struct flow_miss_op *ops, size_t *n_ops)
3669 struct rule_dpif *rule = rule_dpif_lookup(miss->ofproto, &miss->flow);
3670 long long int now = time_msec();
3671 struct ofpbuf *packet;
3672 struct xlate_in xin;
3674 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3675 struct flow_miss_op *op = &ops[*n_ops];
3676 struct dpif_flow_stats stats;
3678 COVERAGE_INC(facet_suppress);
3680 handle_flow_miss_common(rule, packet, &miss->flow);
3682 dpif_flow_stats_extract(&miss->flow, packet, now, &stats);
3683 rule_credit_stats(rule, &stats);
3685 xlate_in_init(&xin, miss->ofproto, &miss->flow, &miss->initial_vals,
3686 rule, stats.tcp_flags, packet);
3687 xin.resubmit_stats = &stats;
3688 xlate_actions(&xin, &op->xout);
3690 if (op->xout.odp_actions.size) {
3691 struct dpif_execute *execute = &op->dpif_op.u.execute;
3693 init_flow_miss_execute_op(miss, packet, op);
3694 execute->actions = op->xout.odp_actions.data;
3695 execute->actions_len = op->xout.odp_actions.size;
3696 op->xout_garbage = true;
3700 xlate_out_uninit(&op->xout);
3705 /* Handles 'miss', which matches 'facet'. May add any required datapath
3706 * operations to 'ops', incrementing '*n_ops' for each new op.
3708 * All of the packets in 'miss' are considered to have arrived at time 'now'.
3709 * This is really important only for new facets: if we just called time_msec()
3710 * here, then the new subfacet or its packets could look (occasionally) as
3711 * though it was used some time after the facet was used. That can make a
3712 * one-packet flow look like it has a nonzero duration, which looks odd in
3713 * e.g. NetFlow statistics. */
3715 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
3717 struct flow_miss_op *ops, size_t *n_ops)
3719 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3720 enum subfacet_path want_path;
3721 struct subfacet *subfacet;
3722 struct ofpbuf *packet;
3724 subfacet = subfacet_create(facet, miss, now);
3725 want_path = subfacet->facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
3727 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3728 struct flow_miss_op *op = &ops[*n_ops];
3729 struct dpif_flow_stats stats;
3731 handle_flow_miss_common(facet->rule, packet, &miss->flow);
3733 if (want_path != SF_FAST_PATH) {
3734 struct xlate_in xin;
3736 xlate_in_init(&xin, ofproto, &facet->flow, &facet->initial_vals,
3737 facet->rule, 0, packet);
3738 xlate_actions_for_side_effects(&xin);
3741 dpif_flow_stats_extract(&facet->flow, packet, now, &stats);
3742 subfacet_update_stats(subfacet, &stats);
3744 if (facet->xout.odp_actions.size) {
3745 struct dpif_execute *execute = &op->dpif_op.u.execute;
3747 init_flow_miss_execute_op(miss, packet, op);
3748 execute->actions = facet->xout.odp_actions.data,
3749 execute->actions_len = facet->xout.odp_actions.size;
3754 if (miss->upcall_type == DPIF_UC_MISS || subfacet->path != want_path) {
3755 struct flow_miss_op *op = &ops[(*n_ops)++];
3756 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
3758 subfacet->path = want_path;
3760 op->xout_garbage = false;
3761 op->dpif_op.type = DPIF_OP_FLOW_PUT;
3762 put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
3763 put->key = miss->key;
3764 put->key_len = miss->key_len;
3765 if (want_path == SF_FAST_PATH) {
3766 put->actions = facet->xout.odp_actions.data;
3767 put->actions_len = facet->xout.odp_actions.size;
3769 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
3770 op->slow_stub, sizeof op->slow_stub,
3771 &put->actions, &put->actions_len);
3777 /* Handles flow miss 'miss'. May add any required datapath operations
3778 * to 'ops', incrementing '*n_ops' for each new op. */
3780 handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops,
3783 struct ofproto_dpif *ofproto = miss->ofproto;
3784 struct facet *facet;
3788 /* The caller must ensure that miss->hmap_node.hash contains
3789 * flow_hash(miss->flow, 0). */
3790 hash = miss->hmap_node.hash;
3792 facet = facet_lookup_valid(ofproto, &miss->flow, hash);
3794 /* There does not exist a bijection between 'struct flow' and datapath
3795 * flow keys with fitness ODP_FIT_TO_LITTLE. This breaks a fundamental
3796 * assumption used throughout the facet and subfacet handling code.
3797 * Since we have to handle these misses in userspace anyway, we simply
3798 * skip facet creation, avoiding the problem alltogether. */
3799 if (miss->key_fitness == ODP_FIT_TOO_LITTLE
3800 || !flow_miss_should_make_facet(ofproto, miss, hash)) {
3801 handle_flow_miss_without_facet(miss, ops, n_ops);
3805 facet = facet_create(miss, hash);
3810 handle_flow_miss_with_facet(miss, facet, now, ops, n_ops);
3813 static struct drop_key *
3814 drop_key_lookup(const struct dpif_backer *backer, const struct nlattr *key,
3817 struct drop_key *drop_key;
3819 HMAP_FOR_EACH_WITH_HASH (drop_key, hmap_node, hash_bytes(key, key_len, 0),
3820 &backer->drop_keys) {
3821 if (drop_key->key_len == key_len
3822 && !memcmp(drop_key->key, key, key_len)) {
3830 drop_key_clear(struct dpif_backer *backer)
3832 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3833 struct drop_key *drop_key, *next;
3835 HMAP_FOR_EACH_SAFE (drop_key, next, hmap_node, &backer->drop_keys) {
3838 error = dpif_flow_del(backer->dpif, drop_key->key, drop_key->key_len,
3840 if (error && !VLOG_DROP_WARN(&rl)) {
3841 struct ds ds = DS_EMPTY_INITIALIZER;
3842 odp_flow_key_format(drop_key->key, drop_key->key_len, &ds);
3843 VLOG_WARN("Failed to delete drop key (%s) (%s)", strerror(error),
3848 hmap_remove(&backer->drop_keys, &drop_key->hmap_node);
3849 free(drop_key->key);
3854 /* Given a datpath, packet, and flow metadata ('backer', 'packet', and 'key'
3855 * respectively), populates 'flow' with the result of odp_flow_key_to_flow().
3856 * Optionally, if nonnull, populates 'fitnessp' with the fitness of 'flow' as
3857 * returned by odp_flow_key_to_flow(). Also, optionally populates 'ofproto'
3858 * with the ofproto_dpif, and 'odp_in_port' with the datapath in_port, that
3859 * 'packet' ingressed.
3861 * If 'ofproto' is nonnull, requires 'flow''s in_port to exist. Otherwise sets
3862 * 'flow''s in_port to OFPP_NONE.
3864 * This function does post-processing on data returned from
3865 * odp_flow_key_to_flow() to help make VLAN splinters transparent to the rest
3866 * of the upcall processing logic. In particular, if the extracted in_port is
3867 * a VLAN splinter port, it replaces flow->in_port by the "real" port, sets
3868 * flow->vlan_tci correctly for the VLAN of the VLAN splinter port, and pushes
3869 * a VLAN header onto 'packet' (if it is nonnull).
3871 * Optionally, if 'initial_vals' is nonnull, sets 'initial_vals->vlan_tci'
3872 * to the VLAN TCI with which the packet was really received, that is, the
3873 * actual VLAN TCI extracted by odp_flow_key_to_flow(). (This differs from
3874 * the value returned in flow->vlan_tci only for packets received on
3875 * VLAN splinters.) Also, if received on an IP tunnel, sets
3876 * 'initial_vals->tunnel_ip_tos' to the tunnel's IP TOS.
3878 * Similarly, this function also includes some logic to help with tunnels. It
3879 * may modify 'flow' as necessary to make the tunneling implementation
3880 * transparent to the upcall processing logic.
3882 * Returns 0 if successful, ENODEV if the parsed flow has no associated ofport,
3883 * or some other positive errno if there are other problems. */
3885 ofproto_receive(const struct dpif_backer *backer, struct ofpbuf *packet,
3886 const struct nlattr *key, size_t key_len,
3887 struct flow *flow, enum odp_key_fitness *fitnessp,
3888 struct ofproto_dpif **ofproto, uint32_t *odp_in_port,
3889 struct initial_vals *initial_vals)
3891 const struct ofport_dpif *port;
3892 enum odp_key_fitness fitness;
3895 fitness = odp_flow_key_to_flow(key, key_len, flow);
3896 if (fitness == ODP_FIT_ERROR) {
3902 initial_vals->vlan_tci = flow->vlan_tci;
3903 initial_vals->tunnel_ip_tos = flow->tunnel.ip_tos;
3907 *odp_in_port = flow->in_port;
3910 if (tnl_port_should_receive(flow)) {
3911 const struct ofport *ofport = tnl_port_receive(flow);
3913 flow->in_port = OFPP_NONE;
3916 port = ofport_dpif_cast(ofport);
3918 /* We can't reproduce 'key' from 'flow'. */
3919 fitness = fitness == ODP_FIT_PERFECT ? ODP_FIT_TOO_MUCH : fitness;
3921 /* XXX: Since the tunnel module is not scoped per backer, it's
3922 * theoretically possible that we'll receive an ofport belonging to an
3923 * entirely different datapath. In practice, this can't happen because
3924 * no platforms has two separate datapaths which each support
3926 ovs_assert(ofproto_dpif_cast(port->up.ofproto)->backer == backer);
3928 port = odp_port_to_ofport(backer, flow->in_port);
3930 flow->in_port = OFPP_NONE;
3934 flow->in_port = port->up.ofp_port;
3935 if (vsp_adjust_flow(ofproto_dpif_cast(port->up.ofproto), flow)) {
3937 /* Make the packet resemble the flow, so that it gets sent to
3938 * an OpenFlow controller properly, so that it looks correct
3939 * for sFlow, and so that flow_extract() will get the correct
3940 * vlan_tci if it is called on 'packet'.
3942 * The allocated space inside 'packet' probably also contains
3943 * 'key', that is, both 'packet' and 'key' are probably part of
3944 * a struct dpif_upcall (see the large comment on that
3945 * structure definition), so pushing data on 'packet' is in
3946 * general not a good idea since it could overwrite 'key' or
3947 * free it as a side effect. However, it's OK in this special
3948 * case because we know that 'packet' is inside a Netlink
3949 * attribute: pushing 4 bytes will just overwrite the 4-byte
3950 * "struct nlattr", which is fine since we don't need that
3951 * header anymore. */
3952 eth_push_vlan(packet, flow->vlan_tci);
3954 /* We can't reproduce 'key' from 'flow'. */
3955 fitness = fitness == ODP_FIT_PERFECT ? ODP_FIT_TOO_MUCH : fitness;
3961 *ofproto = ofproto_dpif_cast(port->up.ofproto);
3966 *fitnessp = fitness;
3972 handle_miss_upcalls(struct dpif_backer *backer, struct dpif_upcall *upcalls,
3975 struct dpif_upcall *upcall;
3976 struct flow_miss *miss;
3977 struct flow_miss misses[FLOW_MISS_MAX_BATCH];
3978 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
3979 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
3989 /* Construct the to-do list.
3991 * This just amounts to extracting the flow from each packet and sticking
3992 * the packets that have the same flow in the same "flow_miss" structure so
3993 * that we can process them together. */
3996 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
3997 struct flow_miss *miss = &misses[n_misses];
3998 struct flow_miss *existing_miss;
3999 struct ofproto_dpif *ofproto;
4000 uint32_t odp_in_port;
4005 error = ofproto_receive(backer, upcall->packet, upcall->key,
4006 upcall->key_len, &flow, &miss->key_fitness,
4007 &ofproto, &odp_in_port, &miss->initial_vals);
4008 if (error == ENODEV) {
4009 struct drop_key *drop_key;
4011 /* Received packet on port for which we couldn't associate
4012 * an ofproto. This can happen if a port is removed while
4013 * traffic is being received. Print a rate-limited message
4014 * in case it happens frequently. Install a drop flow so
4015 * that future packets of the flow are inexpensively dropped
4017 VLOG_INFO_RL(&rl, "received packet on unassociated port %"PRIu32,
4020 drop_key = drop_key_lookup(backer, upcall->key, upcall->key_len);
4022 drop_key = xmalloc(sizeof *drop_key);
4023 drop_key->key = xmemdup(upcall->key, upcall->key_len);
4024 drop_key->key_len = upcall->key_len;
4026 hmap_insert(&backer->drop_keys, &drop_key->hmap_node,
4027 hash_bytes(drop_key->key, drop_key->key_len, 0));
4028 dpif_flow_put(backer->dpif, DPIF_FP_CREATE | DPIF_FP_MODIFY,
4029 drop_key->key, drop_key->key_len, NULL, 0, NULL);
4037 ofproto->n_missed++;
4038 flow_extract(upcall->packet, flow.skb_priority, flow.skb_mark,
4039 &flow.tunnel, flow.in_port, &miss->flow);
4041 /* Add other packets to a to-do list. */
4042 hash = flow_hash(&miss->flow, 0);
4043 existing_miss = flow_miss_find(&todo, ofproto, &miss->flow, hash);
4044 if (!existing_miss) {
4045 hmap_insert(&todo, &miss->hmap_node, hash);
4046 miss->ofproto = ofproto;
4047 miss->key = upcall->key;
4048 miss->key_len = upcall->key_len;
4049 miss->upcall_type = upcall->type;
4050 miss->odp_in_port = odp_in_port;
4051 list_init(&miss->packets);
4055 miss = existing_miss;
4057 list_push_back(&miss->packets, &upcall->packet->list_node);
4060 /* Process each element in the to-do list, constructing the set of
4061 * operations to batch. */
4063 HMAP_FOR_EACH (miss, hmap_node, &todo) {
4064 handle_flow_miss(miss, flow_miss_ops, &n_ops);
4066 ovs_assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
4068 /* Execute batch. */
4069 for (i = 0; i < n_ops; i++) {
4070 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
4072 dpif_operate(backer->dpif, dpif_ops, n_ops);
4075 for (i = 0; i < n_ops; i++) {
4076 if (flow_miss_ops[i].xout_garbage) {
4077 xlate_out_uninit(&flow_miss_ops[i].xout);
4080 hmap_destroy(&todo);
4083 static enum { SFLOW_UPCALL, MISS_UPCALL, BAD_UPCALL, FLOW_SAMPLE_UPCALL,
4085 classify_upcall(const struct dpif_upcall *upcall)
4087 size_t userdata_len;
4088 union user_action_cookie cookie;
4090 /* First look at the upcall type. */
4091 switch (upcall->type) {
4092 case DPIF_UC_ACTION:
4098 case DPIF_N_UC_TYPES:
4100 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
4104 /* "action" upcalls need a closer look. */
4105 if (!upcall->userdata) {
4106 VLOG_WARN_RL(&rl, "action upcall missing cookie");
4109 userdata_len = nl_attr_get_size(upcall->userdata);
4110 if (userdata_len < sizeof cookie.type
4111 || userdata_len > sizeof cookie) {
4112 VLOG_WARN_RL(&rl, "action upcall cookie has unexpected size %zu",
4116 memset(&cookie, 0, sizeof cookie);
4117 memcpy(&cookie, nl_attr_get(upcall->userdata), userdata_len);
4118 if (userdata_len == sizeof cookie.sflow
4119 && cookie.type == USER_ACTION_COOKIE_SFLOW) {
4120 return SFLOW_UPCALL;
4121 } else if (userdata_len == sizeof cookie.slow_path
4122 && cookie.type == USER_ACTION_COOKIE_SLOW_PATH) {
4124 } else if (userdata_len == sizeof cookie.flow_sample
4125 && cookie.type == USER_ACTION_COOKIE_FLOW_SAMPLE) {
4126 return FLOW_SAMPLE_UPCALL;
4127 } else if (userdata_len == sizeof cookie.ipfix
4128 && cookie.type == USER_ACTION_COOKIE_IPFIX) {
4129 return IPFIX_UPCALL;
4131 VLOG_WARN_RL(&rl, "invalid user cookie of type %"PRIu16
4132 " and size %zu", cookie.type, userdata_len);
4138 handle_sflow_upcall(struct dpif_backer *backer,
4139 const struct dpif_upcall *upcall)
4141 struct ofproto_dpif *ofproto;
4142 union user_action_cookie cookie;
4144 uint32_t odp_in_port;
4146 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
4147 &flow, NULL, &ofproto, &odp_in_port, NULL)
4148 || !ofproto->sflow) {
4152 memset(&cookie, 0, sizeof cookie);
4153 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.sflow);
4154 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
4155 odp_in_port, &cookie);
4159 handle_flow_sample_upcall(struct dpif_backer *backer,
4160 const struct dpif_upcall *upcall)
4162 struct ofproto_dpif *ofproto;
4163 union user_action_cookie cookie;
4166 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
4167 &flow, NULL, &ofproto, NULL, NULL)
4168 || !ofproto->ipfix) {
4172 memset(&cookie, 0, sizeof cookie);
4173 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.flow_sample);
4175 /* The flow reflects exactly the contents of the packet. Sample
4176 * the packet using it. */
4177 dpif_ipfix_flow_sample(ofproto->ipfix, upcall->packet, &flow,
4178 cookie.flow_sample.collector_set_id,
4179 cookie.flow_sample.probability,
4180 cookie.flow_sample.obs_domain_id,
4181 cookie.flow_sample.obs_point_id);
4185 handle_ipfix_upcall(struct dpif_backer *backer,
4186 const struct dpif_upcall *upcall)
4188 struct ofproto_dpif *ofproto;
4191 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
4192 &flow, NULL, &ofproto, NULL, NULL)
4193 || !ofproto->ipfix) {
4197 /* The flow reflects exactly the contents of the packet. Sample
4198 * the packet using it. */
4199 dpif_ipfix_bridge_sample(ofproto->ipfix, upcall->packet, &flow);
4203 handle_upcalls(struct dpif_backer *backer, unsigned int max_batch)
4205 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
4206 struct ofpbuf miss_bufs[FLOW_MISS_MAX_BATCH];
4207 uint64_t miss_buf_stubs[FLOW_MISS_MAX_BATCH][4096 / 8];
4212 ovs_assert(max_batch <= FLOW_MISS_MAX_BATCH);
4215 for (n_processed = 0; n_processed < max_batch; n_processed++) {
4216 struct dpif_upcall *upcall = &misses[n_misses];
4217 struct ofpbuf *buf = &miss_bufs[n_misses];
4220 ofpbuf_use_stub(buf, miss_buf_stubs[n_misses],
4221 sizeof miss_buf_stubs[n_misses]);
4222 error = dpif_recv(backer->dpif, upcall, buf);
4228 switch (classify_upcall(upcall)) {
4230 /* Handle it later. */
4235 handle_sflow_upcall(backer, upcall);
4239 case FLOW_SAMPLE_UPCALL:
4240 handle_flow_sample_upcall(backer, upcall);
4245 handle_ipfix_upcall(backer, upcall);
4255 /* Handle deferred MISS_UPCALL processing. */
4256 handle_miss_upcalls(backer, misses, n_misses);
4257 for (i = 0; i < n_misses; i++) {
4258 ofpbuf_uninit(&miss_bufs[i]);
4264 /* Flow expiration. */
4266 static int subfacet_max_idle(const struct ofproto_dpif *);
4267 static void update_stats(struct dpif_backer *);
4268 static void rule_expire(struct rule_dpif *);
4269 static void expire_subfacets(struct ofproto_dpif *, int dp_max_idle);
4271 /* This function is called periodically by run(). Its job is to collect
4272 * updates for the flows that have been installed into the datapath, most
4273 * importantly when they last were used, and then use that information to
4274 * expire flows that have not been used recently.
4276 * Returns the number of milliseconds after which it should be called again. */
4278 expire(struct dpif_backer *backer)
4280 struct ofproto_dpif *ofproto;
4281 int max_idle = INT32_MAX;
4283 /* Periodically clear out the drop keys in an effort to keep them
4284 * relatively few. */
4285 drop_key_clear(backer);
4287 /* Update stats for each flow in the backer. */
4288 update_stats(backer);
4290 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4291 struct rule *rule, *next_rule;
4294 if (ofproto->backer != backer) {
4298 /* Keep track of the max number of flows per ofproto_dpif. */
4299 update_max_subfacet_count(ofproto);
4301 /* Expire subfacets that have been idle too long. */
4302 dp_max_idle = subfacet_max_idle(ofproto);
4303 expire_subfacets(ofproto, dp_max_idle);
4305 max_idle = MIN(max_idle, dp_max_idle);
4307 /* Expire OpenFlow flows whose idle_timeout or hard_timeout
4309 LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
4310 &ofproto->up.expirable) {
4311 rule_expire(rule_dpif_cast(rule));
4314 /* All outstanding data in existing flows has been accounted, so it's a
4315 * good time to do bond rebalancing. */
4316 if (ofproto->has_bonded_bundles) {
4317 struct ofbundle *bundle;
4319 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
4321 bond_rebalance(bundle->bond, &backer->revalidate_set);
4327 return MIN(max_idle, 1000);
4330 /* Updates flow table statistics given that the datapath just reported 'stats'
4331 * as 'subfacet''s statistics. */
4333 update_subfacet_stats(struct subfacet *subfacet,
4334 const struct dpif_flow_stats *stats)
4336 struct facet *facet = subfacet->facet;
4337 struct dpif_flow_stats diff;
4339 diff.tcp_flags = stats->tcp_flags;
4340 diff.used = stats->used;
4342 if (stats->n_packets >= subfacet->dp_packet_count) {
4343 diff.n_packets = stats->n_packets - subfacet->dp_packet_count;
4345 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
4349 if (stats->n_bytes >= subfacet->dp_byte_count) {
4350 diff.n_bytes = stats->n_bytes - subfacet->dp_byte_count;
4352 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
4356 subfacet->dp_packet_count = stats->n_packets;
4357 subfacet->dp_byte_count = stats->n_bytes;
4358 subfacet_update_stats(subfacet, &diff);
4360 if (facet->accounted_bytes < facet->byte_count) {
4362 facet_account(facet);
4363 facet->accounted_bytes = facet->byte_count;
4367 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
4368 * about, or a flow that shouldn't be installed but was anyway. Delete it. */
4370 delete_unexpected_flow(struct ofproto_dpif *ofproto,
4371 const struct nlattr *key, size_t key_len)
4373 if (!VLOG_DROP_WARN(&rl)) {
4377 odp_flow_key_format(key, key_len, &s);
4378 VLOG_WARN("unexpected flow on %s: %s", ofproto->up.name, ds_cstr(&s));
4382 COVERAGE_INC(facet_unexpected);
4383 dpif_flow_del(ofproto->backer->dpif, key, key_len, NULL);
4386 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
4388 * This function also pushes statistics updates to rules which each facet
4389 * resubmits into. Generally these statistics will be accurate. However, if a
4390 * facet changes the rule it resubmits into at some time in between
4391 * update_stats() runs, it is possible that statistics accrued to the
4392 * old rule will be incorrectly attributed to the new rule. This could be
4393 * avoided by calling update_stats() whenever rules are created or
4394 * deleted. However, the performance impact of making so many calls to the
4395 * datapath do not justify the benefit of having perfectly accurate statistics.
4397 * In addition, this function maintains per ofproto flow hit counts. The patch
4398 * port is not treated specially. e.g. A packet ingress from br0 patched into
4399 * br1 will increase the hit count of br0 by 1, however, does not affect
4400 * the hit or miss counts of br1.
4403 update_stats(struct dpif_backer *backer)
4405 const struct dpif_flow_stats *stats;
4406 struct dpif_flow_dump dump;
4407 const struct nlattr *key;
4408 struct ofproto_dpif *ofproto;
4411 dpif_flow_dump_start(&dump, backer->dpif);
4412 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
4414 struct subfacet *subfacet;
4417 if (ofproto_receive(backer, NULL, key, key_len, &flow, NULL, &ofproto,
4422 ofproto->total_subfacet_count += hmap_count(&ofproto->subfacets);
4423 ofproto->n_update_stats++;
4425 key_hash = odp_flow_key_hash(key, key_len);
4426 subfacet = subfacet_find(ofproto, key, key_len, key_hash);
4427 switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
4429 /* Update ofproto_dpif's hit count. */
4430 if (stats->n_packets > subfacet->dp_packet_count) {
4431 uint64_t delta = stats->n_packets - subfacet->dp_packet_count;
4432 dpif_stats_update_hit_count(ofproto, delta);
4435 update_subfacet_stats(subfacet, stats);
4439 /* Stats are updated per-packet. */
4442 case SF_NOT_INSTALLED:
4444 delete_unexpected_flow(ofproto, key, key_len);
4449 dpif_flow_dump_done(&dump);
4451 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4452 update_moving_averages(ofproto);
4457 /* Calculates and returns the number of milliseconds of idle time after which
4458 * subfacets should expire from the datapath. When a subfacet expires, we fold
4459 * its statistics into its facet, and when a facet's last subfacet expires, we
4460 * fold its statistic into its rule. */
4462 subfacet_max_idle(const struct ofproto_dpif *ofproto)
4465 * Idle time histogram.
4467 * Most of the time a switch has a relatively small number of subfacets.
4468 * When this is the case we might as well keep statistics for all of them
4469 * in userspace and to cache them in the kernel datapath for performance as
4472 * As the number of subfacets increases, the memory required to maintain
4473 * statistics about them in userspace and in the kernel becomes
4474 * significant. However, with a large number of subfacets it is likely
4475 * that only a few of them are "heavy hitters" that consume a large amount
4476 * of bandwidth. At this point, only heavy hitters are worth caching in
4477 * the kernel and maintaining in userspaces; other subfacets we can
4480 * The technique used to compute the idle time is to build a histogram with
4481 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
4482 * that is installed in the kernel gets dropped in the appropriate bucket.
4483 * After the histogram has been built, we compute the cutoff so that only
4484 * the most-recently-used 1% of subfacets (but at least
4485 * ofproto->up.flow_eviction_threshold flows) are kept cached. At least
4486 * the most-recently-used bucket of subfacets is kept, so actually an
4487 * arbitrary number of subfacets can be kept in any given expiration run
4488 * (though the next run will delete most of those unless they receive
4491 * This requires a second pass through the subfacets, in addition to the
4492 * pass made by update_stats(), because the former function never looks at
4493 * uninstallable subfacets.
4495 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
4496 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
4497 int buckets[N_BUCKETS] = { 0 };
4498 int total, subtotal, bucket;
4499 struct subfacet *subfacet;
4503 total = hmap_count(&ofproto->subfacets);
4504 if (total <= ofproto->up.flow_eviction_threshold) {
4505 return N_BUCKETS * BUCKET_WIDTH;
4508 /* Build histogram. */
4510 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
4511 long long int idle = now - subfacet->used;
4512 int bucket = (idle <= 0 ? 0
4513 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
4514 : (unsigned int) idle / BUCKET_WIDTH);
4518 /* Find the first bucket whose flows should be expired. */
4519 subtotal = bucket = 0;
4521 subtotal += buckets[bucket++];
4522 } while (bucket < N_BUCKETS &&
4523 subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100));
4525 if (VLOG_IS_DBG_ENABLED()) {
4529 ds_put_cstr(&s, "keep");
4530 for (i = 0; i < N_BUCKETS; i++) {
4532 ds_put_cstr(&s, ", drop");
4535 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
4538 VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s));
4542 return bucket * BUCKET_WIDTH;
4546 expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle)
4548 /* Cutoff time for most flows. */
4549 long long int normal_cutoff = time_msec() - dp_max_idle;
4551 /* We really want to keep flows for special protocols around, so use a more
4552 * conservative cutoff. */
4553 long long int special_cutoff = time_msec() - 10000;
4555 struct subfacet *subfacet, *next_subfacet;
4556 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
4560 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
4561 &ofproto->subfacets) {
4562 long long int cutoff;
4564 cutoff = (subfacet->facet->xout.slow & (SLOW_CFM | SLOW_LACP
4568 if (subfacet->used < cutoff) {
4569 if (subfacet->path != SF_NOT_INSTALLED) {
4570 batch[n_batch++] = subfacet;
4571 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
4572 subfacet_destroy_batch(ofproto, batch, n_batch);
4576 subfacet_destroy(subfacet);
4582 subfacet_destroy_batch(ofproto, batch, n_batch);
4586 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
4587 * then delete it entirely. */
4589 rule_expire(struct rule_dpif *rule)
4591 struct facet *facet, *next_facet;
4595 if (rule->up.pending) {
4596 /* We'll have to expire it later. */
4600 /* Has 'rule' expired? */
4602 if (rule->up.hard_timeout
4603 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
4604 reason = OFPRR_HARD_TIMEOUT;
4605 } else if (rule->up.idle_timeout
4606 && now > rule->up.used + rule->up.idle_timeout * 1000) {
4607 reason = OFPRR_IDLE_TIMEOUT;
4612 COVERAGE_INC(ofproto_dpif_expired);
4614 /* Update stats. (This is a no-op if the rule expired due to an idle
4615 * timeout, because that only happens when the rule has no facets left.) */
4616 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4617 facet_remove(facet);
4620 /* Get rid of the rule. */
4621 ofproto_rule_expire(&rule->up, reason);
4626 /* Creates and returns a new facet based on 'miss'.
4628 * The caller must already have determined that no facet with an identical
4629 * 'miss->flow' exists in 'miss->ofproto'.
4631 * 'hash' must be the return value of flow_hash(miss->flow, 0).
4633 * The facet will initially have no subfacets. The caller should create (at
4634 * least) one subfacet with subfacet_create(). */
4635 static struct facet *
4636 facet_create(const struct flow_miss *miss, uint32_t hash)
4638 struct ofproto_dpif *ofproto = miss->ofproto;
4639 struct xlate_in xin;
4640 struct facet *facet;
4642 facet = xzalloc(sizeof *facet);
4643 facet->used = time_msec();
4644 facet->flow = miss->flow;
4645 facet->initial_vals = miss->initial_vals;
4646 facet->rule = rule_dpif_lookup(ofproto, &facet->flow);
4647 facet->learn_rl = time_msec() + 500;
4649 hmap_insert(&ofproto->facets, &facet->hmap_node, hash);
4650 list_push_back(&facet->rule->facets, &facet->list_node);
4651 list_init(&facet->subfacets);
4652 netflow_flow_init(&facet->nf_flow);
4653 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
4655 xlate_in_init(&xin, ofproto, &facet->flow, &facet->initial_vals,
4656 facet->rule, 0, NULL);
4657 xin.may_learn = true;
4658 xlate_actions(&xin, &facet->xout);
4659 facet->nf_flow.output_iface = facet->xout.nf_output_iface;
4665 facet_free(struct facet *facet)
4668 xlate_out_uninit(&facet->xout);
4673 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
4674 * 'packet', which arrived on 'in_port'. */
4676 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
4677 const struct nlattr *odp_actions, size_t actions_len,
4678 struct ofpbuf *packet)
4680 struct odputil_keybuf keybuf;
4684 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4685 odp_flow_key_from_flow(&key, flow,
4686 ofp_port_to_odp_port(ofproto, flow->in_port));
4688 error = dpif_execute(ofproto->backer->dpif, key.data, key.size,
4689 odp_actions, actions_len, packet);
4693 /* Remove 'facet' from 'ofproto' and free up the associated memory:
4695 * - If 'facet' was installed in the datapath, uninstalls it and updates its
4696 * rule's statistics, via subfacet_uninstall().
4698 * - Removes 'facet' from its rule and from ofproto->facets.
4701 facet_remove(struct facet *facet)
4703 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4704 struct subfacet *subfacet, *next_subfacet;
4706 ovs_assert(!list_is_empty(&facet->subfacets));
4708 /* First uninstall all of the subfacets to get final statistics. */
4709 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4710 subfacet_uninstall(subfacet);
4713 /* Flush the final stats to the rule.
4715 * This might require us to have at least one subfacet around so that we
4716 * can use its actions for accounting in facet_account(), which is why we
4717 * have uninstalled but not yet destroyed the subfacets. */
4718 facet_flush_stats(facet);
4720 /* Now we're really all done so destroy everything. */
4721 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
4722 &facet->subfacets) {
4723 subfacet_destroy__(subfacet);
4725 hmap_remove(&ofproto->facets, &facet->hmap_node);
4726 list_remove(&facet->list_node);
4730 /* Feed information from 'facet' back into the learning table to keep it in
4731 * sync with what is actually flowing through the datapath. */
4733 facet_learn(struct facet *facet)
4735 long long int now = time_msec();
4737 if (!facet->xout.has_fin_timeout && now < facet->learn_rl) {
4741 facet->learn_rl = now + 500;
4743 if (!facet->xout.has_learn
4744 && !facet->xout.has_normal
4745 && (!facet->xout.has_fin_timeout
4746 || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
4750 facet_push_stats(facet, true);
4754 facet_account(struct facet *facet)
4756 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4757 const struct nlattr *a;
4762 if (!facet->xout.has_normal || !ofproto->has_bonded_bundles) {
4765 n_bytes = facet->byte_count - facet->accounted_bytes;
4767 /* This loop feeds byte counters to bond_account() for rebalancing to use
4768 * as a basis. We also need to track the actual VLAN on which the packet
4769 * is going to be sent to ensure that it matches the one passed to
4770 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
4773 * We use the actions from an arbitrary subfacet because they should all
4774 * be equally valid for our purpose. */
4775 vlan_tci = facet->flow.vlan_tci;
4776 NL_ATTR_FOR_EACH_UNSAFE (a, left, facet->xout.odp_actions.data,
4777 facet->xout.odp_actions.size) {
4778 const struct ovs_action_push_vlan *vlan;
4779 struct ofport_dpif *port;
4781 switch (nl_attr_type(a)) {
4782 case OVS_ACTION_ATTR_OUTPUT:
4783 port = get_odp_port(ofproto, nl_attr_get_u32(a));
4784 if (port && port->bundle && port->bundle->bond) {
4785 bond_account(port->bundle->bond, &facet->flow,
4786 vlan_tci_to_vid(vlan_tci), n_bytes);
4790 case OVS_ACTION_ATTR_POP_VLAN:
4791 vlan_tci = htons(0);
4794 case OVS_ACTION_ATTR_PUSH_VLAN:
4795 vlan = nl_attr_get(a);
4796 vlan_tci = vlan->vlan_tci;
4802 /* Returns true if the only action for 'facet' is to send to the controller.
4803 * (We don't report NetFlow expiration messages for such facets because they
4804 * are just part of the control logic for the network, not real traffic). */
4806 facet_is_controller_flow(struct facet *facet)
4809 const struct rule *rule = &facet->rule->up;
4810 const struct ofpact *ofpacts = rule->ofpacts;
4811 size_t ofpacts_len = rule->ofpacts_len;
4813 if (ofpacts_len > 0 &&
4814 ofpacts->type == OFPACT_CONTROLLER &&
4815 ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len)) {
4822 /* Folds all of 'facet''s statistics into its rule. Also updates the
4823 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
4824 * 'facet''s statistics in the datapath should have been zeroed and folded into
4825 * its packet and byte counts before this function is called. */
4827 facet_flush_stats(struct facet *facet)
4829 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4830 struct subfacet *subfacet;
4832 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4833 ovs_assert(!subfacet->dp_byte_count);
4834 ovs_assert(!subfacet->dp_packet_count);
4837 facet_push_stats(facet, false);
4838 if (facet->accounted_bytes < facet->byte_count) {
4839 facet_account(facet);
4840 facet->accounted_bytes = facet->byte_count;
4843 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
4844 struct ofexpired expired;
4845 expired.flow = facet->flow;
4846 expired.packet_count = facet->packet_count;
4847 expired.byte_count = facet->byte_count;
4848 expired.used = facet->used;
4849 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4852 /* Reset counters to prevent double counting if 'facet' ever gets
4854 facet_reset_counters(facet);
4856 netflow_flow_clear(&facet->nf_flow);
4857 facet->tcp_flags = 0;
4860 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
4861 * Returns it if found, otherwise a null pointer.
4863 * 'hash' must be the return value of flow_hash(flow, 0).
4865 * The returned facet might need revalidation; use facet_lookup_valid()
4866 * instead if that is important. */
4867 static struct facet *
4868 facet_find(struct ofproto_dpif *ofproto,
4869 const struct flow *flow, uint32_t hash)
4871 struct facet *facet;
4873 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, hash, &ofproto->facets) {
4874 if (flow_equal(flow, &facet->flow)) {
4882 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
4883 * Returns it if found, otherwise a null pointer.
4885 * 'hash' must be the return value of flow_hash(flow, 0).
4887 * The returned facet is guaranteed to be valid. */
4888 static struct facet *
4889 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow,
4892 struct facet *facet;
4894 facet = facet_find(ofproto, flow, hash);
4896 && (ofproto->backer->need_revalidate
4897 || tag_set_intersects(&ofproto->backer->revalidate_set,
4899 && !facet_revalidate(facet)) {
4907 facet_check_consistency(struct facet *facet)
4909 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
4911 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4913 struct xlate_out xout;
4914 struct xlate_in xin;
4916 struct rule_dpif *rule;
4919 /* Check the rule for consistency. */
4920 rule = rule_dpif_lookup(ofproto, &facet->flow);
4921 if (rule != facet->rule) {
4922 if (!VLOG_DROP_WARN(&rl)) {
4923 struct ds s = DS_EMPTY_INITIALIZER;
4925 flow_format(&s, &facet->flow);
4926 ds_put_format(&s, ": facet associated with wrong rule (was "
4927 "table=%"PRIu8",", facet->rule->up.table_id);
4928 cls_rule_format(&facet->rule->up.cr, &s);
4929 ds_put_format(&s, ") (should have been table=%"PRIu8",",
4931 cls_rule_format(&rule->up.cr, &s);
4932 ds_put_cstr(&s, ")\n");
4939 /* Check the datapath actions for consistency. */
4940 xlate_in_init(&xin, ofproto, &facet->flow, &facet->initial_vals, rule,
4942 xlate_actions(&xin, &xout);
4944 ok = ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)
4945 && facet->xout.slow == xout.slow;
4946 if (!ok && !VLOG_DROP_WARN(&rl)) {
4947 struct ds s = DS_EMPTY_INITIALIZER;
4949 flow_format(&s, &facet->flow);
4950 ds_put_cstr(&s, ": inconsistency in facet");
4952 if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4953 ds_put_cstr(&s, " (actions were: ");
4954 format_odp_actions(&s, facet->xout.odp_actions.data,
4955 facet->xout.odp_actions.size);
4956 ds_put_cstr(&s, ") (correct actions: ");
4957 format_odp_actions(&s, xout.odp_actions.data,
4958 xout.odp_actions.size);
4959 ds_put_cstr(&s, ")");
4962 if (facet->xout.slow != xout.slow) {
4963 ds_put_format(&s, " slow path incorrect. should be %d", xout.slow);
4968 xlate_out_uninit(&xout);
4973 /* Re-searches the classifier for 'facet':
4975 * - If the rule found is different from 'facet''s current rule, moves
4976 * 'facet' to the new rule and recompiles its actions.
4978 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
4979 * where it is and recompiles its actions anyway.
4981 * - If any of 'facet''s subfacets correspond to a new flow according to
4982 * ofproto_receive(), 'facet' is removed.
4984 * Returns true if 'facet' is still valid. False if 'facet' was removed. */
4986 facet_revalidate(struct facet *facet)
4988 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4989 struct rule_dpif *new_rule;
4990 struct subfacet *subfacet;
4991 struct xlate_out xout;
4992 struct xlate_in xin;
4994 COVERAGE_INC(facet_revalidate);
4996 /* Check that child subfacets still correspond to this facet. Tunnel
4997 * configuration changes could cause a subfacet's OpenFlow in_port to
4999 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
5000 struct ofproto_dpif *recv_ofproto;
5001 struct flow recv_flow;
5004 error = ofproto_receive(ofproto->backer, NULL, subfacet->key,
5005 subfacet->key_len, &recv_flow, NULL,
5006 &recv_ofproto, NULL, NULL);
5008 || recv_ofproto != ofproto
5009 || memcmp(&recv_flow, &facet->flow, sizeof recv_flow)) {
5010 facet_remove(facet);
5015 new_rule = rule_dpif_lookup(ofproto, &facet->flow);
5017 /* Calculate new datapath actions.
5019 * We do not modify any 'facet' state yet, because we might need to, e.g.,
5020 * emit a NetFlow expiration and, if so, we need to have the old state
5021 * around to properly compose it. */
5022 xlate_in_init(&xin, ofproto, &facet->flow, &facet->initial_vals, new_rule,
5024 xlate_actions(&xin, &xout);
5026 /* A facet's slow path reason should only change under dramatic
5027 * circumstances. Rather than try to update everything, it's simpler to
5028 * remove the facet and start over. */
5029 if (facet->xout.slow != xout.slow) {
5030 facet_remove(facet);
5031 xlate_out_uninit(&xout);
5035 if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
5036 LIST_FOR_EACH(subfacet, list_node, &facet->subfacets) {
5037 if (subfacet->path == SF_FAST_PATH) {
5038 struct dpif_flow_stats stats;
5040 subfacet_install(subfacet, &xout.odp_actions, &stats);
5041 subfacet_update_stats(subfacet, &stats);
5045 facet_flush_stats(facet);
5047 ofpbuf_clear(&facet->xout.odp_actions);
5048 ofpbuf_put(&facet->xout.odp_actions, xout.odp_actions.data,
5049 xout.odp_actions.size);
5052 /* Update 'facet' now that we've taken care of all the old state. */
5053 facet->xout.tags = xout.tags;
5054 facet->xout.slow = xout.slow;
5055 facet->xout.has_learn = xout.has_learn;
5056 facet->xout.has_normal = xout.has_normal;
5057 facet->xout.has_fin_timeout = xout.has_fin_timeout;
5058 facet->xout.nf_output_iface = xout.nf_output_iface;
5059 facet->xout.mirrors = xout.mirrors;
5060 facet->nf_flow.output_iface = facet->xout.nf_output_iface;
5062 if (facet->rule != new_rule) {
5063 COVERAGE_INC(facet_changed_rule);
5064 list_remove(&facet->list_node);
5065 list_push_back(&new_rule->facets, &facet->list_node);
5066 facet->rule = new_rule;
5067 facet->used = new_rule->up.created;
5068 facet->prev_used = facet->used;
5071 xlate_out_uninit(&xout);
5076 facet_reset_counters(struct facet *facet)
5078 facet->packet_count = 0;
5079 facet->byte_count = 0;
5080 facet->prev_packet_count = 0;
5081 facet->prev_byte_count = 0;
5082 facet->accounted_bytes = 0;
5086 facet_push_stats(struct facet *facet, bool may_learn)
5088 struct dpif_flow_stats stats;
5090 ovs_assert(facet->packet_count >= facet->prev_packet_count);
5091 ovs_assert(facet->byte_count >= facet->prev_byte_count);
5092 ovs_assert(facet->used >= facet->prev_used);
5094 stats.n_packets = facet->packet_count - facet->prev_packet_count;
5095 stats.n_bytes = facet->byte_count - facet->prev_byte_count;
5096 stats.used = facet->used;
5097 stats.tcp_flags = facet->tcp_flags;
5099 if (may_learn || stats.n_packets || facet->used > facet->prev_used) {
5100 struct ofproto_dpif *ofproto =
5101 ofproto_dpif_cast(facet->rule->up.ofproto);
5103 struct ofport_dpif *in_port;
5104 struct xlate_in xin;
5106 facet->prev_packet_count = facet->packet_count;
5107 facet->prev_byte_count = facet->byte_count;
5108 facet->prev_used = facet->used;
5110 in_port = get_ofp_port(ofproto, facet->flow.in_port);
5111 if (in_port && in_port->tnl_port) {
5112 netdev_vport_inc_rx(in_port->up.netdev, &stats);
5115 rule_credit_stats(facet->rule, &stats);
5116 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow,
5118 netflow_flow_update_flags(&facet->nf_flow, facet->tcp_flags);
5119 update_mirror_stats(ofproto, facet->xout.mirrors, stats.n_packets,
5122 xlate_in_init(&xin, ofproto, &facet->flow, &facet->initial_vals,
5123 facet->rule, stats.tcp_flags, NULL);
5124 xin.resubmit_stats = &stats;
5125 xin.may_learn = may_learn;
5126 xlate_actions_for_side_effects(&xin);
5131 push_all_stats__(bool run_fast)
5133 static long long int rl = LLONG_MIN;
5134 struct ofproto_dpif *ofproto;
5136 if (time_msec() < rl) {
5140 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5141 struct facet *facet;
5143 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
5144 facet_push_stats(facet, false);
5151 rl = time_msec() + 100;
5155 push_all_stats(void)
5157 push_all_stats__(true);
5161 rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats)
5163 rule->packet_count += stats->n_packets;
5164 rule->byte_count += stats->n_bytes;
5165 ofproto_rule_update_used(&rule->up, stats->used);
5170 static struct subfacet *
5171 subfacet_find(struct ofproto_dpif *ofproto,
5172 const struct nlattr *key, size_t key_len, uint32_t key_hash)
5174 struct subfacet *subfacet;
5176 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
5177 &ofproto->subfacets) {
5178 if (subfacet->key_len == key_len
5179 && !memcmp(key, subfacet->key, key_len)) {
5187 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
5188 * 'key_fitness', 'key', and 'key_len' members in 'miss'. Returns the
5189 * existing subfacet if there is one, otherwise creates and returns a
5191 static struct subfacet *
5192 subfacet_create(struct facet *facet, struct flow_miss *miss,
5195 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5196 enum odp_key_fitness key_fitness = miss->key_fitness;
5197 const struct nlattr *key = miss->key;
5198 size_t key_len = miss->key_len;
5200 struct subfacet *subfacet;
5202 key_hash = odp_flow_key_hash(key, key_len);
5204 if (list_is_empty(&facet->subfacets)) {
5205 subfacet = &facet->one_subfacet;
5207 subfacet = subfacet_find(ofproto, key, key_len, key_hash);
5209 if (subfacet->facet == facet) {
5213 /* This shouldn't happen. */
5214 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
5215 subfacet_destroy(subfacet);
5218 subfacet = xmalloc(sizeof *subfacet);
5221 hmap_insert(&ofproto->subfacets, &subfacet->hmap_node, key_hash);
5222 list_push_back(&facet->subfacets, &subfacet->list_node);
5223 subfacet->facet = facet;
5224 subfacet->key_fitness = key_fitness;
5225 subfacet->key = xmemdup(key, key_len);
5226 subfacet->key_len = key_len;
5227 subfacet->used = now;
5228 subfacet->created = now;
5229 subfacet->dp_packet_count = 0;
5230 subfacet->dp_byte_count = 0;
5231 subfacet->path = SF_NOT_INSTALLED;
5232 subfacet->odp_in_port = miss->odp_in_port;
5234 ofproto->subfacet_add_count++;
5238 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
5239 * its facet within 'ofproto', and frees it. */
5241 subfacet_destroy__(struct subfacet *subfacet)
5243 struct facet *facet = subfacet->facet;
5244 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5246 /* Update ofproto stats before uninstall the subfacet. */
5247 ofproto->subfacet_del_count++;
5248 ofproto->total_subfacet_life_span += (time_msec() - subfacet->created);
5250 subfacet_uninstall(subfacet);
5251 hmap_remove(&ofproto->subfacets, &subfacet->hmap_node);
5252 list_remove(&subfacet->list_node);
5253 free(subfacet->key);
5254 if (subfacet != &facet->one_subfacet) {
5259 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
5260 * last remaining subfacet in its facet destroys the facet too. */
5262 subfacet_destroy(struct subfacet *subfacet)
5264 struct facet *facet = subfacet->facet;
5266 if (list_is_singleton(&facet->subfacets)) {
5267 /* facet_remove() needs at least one subfacet (it will remove it). */
5268 facet_remove(facet);
5270 subfacet_destroy__(subfacet);
5275 subfacet_destroy_batch(struct ofproto_dpif *ofproto,
5276 struct subfacet **subfacets, int n)
5278 struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH];
5279 struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH];
5280 struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH];
5283 for (i = 0; i < n; i++) {
5284 ops[i].type = DPIF_OP_FLOW_DEL;
5285 ops[i].u.flow_del.key = subfacets[i]->key;
5286 ops[i].u.flow_del.key_len = subfacets[i]->key_len;
5287 ops[i].u.flow_del.stats = &stats[i];
5291 dpif_operate(ofproto->backer->dpif, opsp, n);
5292 for (i = 0; i < n; i++) {
5293 subfacet_reset_dp_stats(subfacets[i], &stats[i]);
5294 subfacets[i]->path = SF_NOT_INSTALLED;
5295 subfacet_destroy(subfacets[i]);
5300 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
5301 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
5302 * in the datapath will be zeroed and 'stats' will be updated with traffic new
5303 * since 'subfacet' was last updated.
5305 * Returns 0 if successful, otherwise a positive errno value. */
5307 subfacet_install(struct subfacet *subfacet, const struct ofpbuf *odp_actions,
5308 struct dpif_flow_stats *stats)
5310 struct facet *facet = subfacet->facet;
5311 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5312 enum subfacet_path path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
5313 const struct nlattr *actions = odp_actions->data;
5314 size_t actions_len = odp_actions->size;
5316 uint64_t slow_path_stub[128 / 8];
5317 enum dpif_flow_put_flags flags;
5320 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
5322 flags |= DPIF_FP_ZERO_STATS;
5325 if (path == SF_SLOW_PATH) {
5326 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
5327 slow_path_stub, sizeof slow_path_stub,
5328 &actions, &actions_len);
5331 ret = dpif_flow_put(ofproto->backer->dpif, flags, subfacet->key,
5332 subfacet->key_len, actions, actions_len, stats);
5335 subfacet_reset_dp_stats(subfacet, stats);
5339 subfacet->path = path;
5344 /* If 'subfacet' is installed in the datapath, uninstalls it. */
5346 subfacet_uninstall(struct subfacet *subfacet)
5348 if (subfacet->path != SF_NOT_INSTALLED) {
5349 struct rule_dpif *rule = subfacet->facet->rule;
5350 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5351 struct dpif_flow_stats stats;
5354 error = dpif_flow_del(ofproto->backer->dpif, subfacet->key,
5355 subfacet->key_len, &stats);
5356 subfacet_reset_dp_stats(subfacet, &stats);
5358 subfacet_update_stats(subfacet, &stats);
5360 subfacet->path = SF_NOT_INSTALLED;
5362 ovs_assert(subfacet->dp_packet_count == 0);
5363 ovs_assert(subfacet->dp_byte_count == 0);
5367 /* Resets 'subfacet''s datapath statistics counters. This should be called
5368 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
5369 * non-null, it should contain the statistics returned by dpif when 'subfacet'
5370 * was reset in the datapath. 'stats' will be modified to include only
5371 * statistics new since 'subfacet' was last updated. */
5373 subfacet_reset_dp_stats(struct subfacet *subfacet,
5374 struct dpif_flow_stats *stats)
5377 && subfacet->dp_packet_count <= stats->n_packets
5378 && subfacet->dp_byte_count <= stats->n_bytes) {
5379 stats->n_packets -= subfacet->dp_packet_count;
5380 stats->n_bytes -= subfacet->dp_byte_count;
5383 subfacet->dp_packet_count = 0;
5384 subfacet->dp_byte_count = 0;
5387 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
5389 * Because of the meaning of a subfacet's counters, it only makes sense to do
5390 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
5391 * represents a packet that was sent by hand or if it represents statistics
5392 * that have been cleared out of the datapath. */
5394 subfacet_update_stats(struct subfacet *subfacet,
5395 const struct dpif_flow_stats *stats)
5397 if (stats->n_packets || stats->used > subfacet->used) {
5398 struct facet *facet = subfacet->facet;
5400 subfacet->used = MAX(subfacet->used, stats->used);
5401 facet->used = MAX(facet->used, stats->used);
5402 facet->packet_count += stats->n_packets;
5403 facet->byte_count += stats->n_bytes;
5404 facet->tcp_flags |= stats->tcp_flags;
5410 static struct rule_dpif *
5411 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow)
5413 struct rule_dpif *rule;
5415 rule = rule_dpif_lookup__(ofproto, flow, 0);
5420 return rule_dpif_miss_rule(ofproto, flow);
5423 static struct rule_dpif *
5424 rule_dpif_lookup__(struct ofproto_dpif *ofproto, const struct flow *flow,
5427 struct cls_rule *cls_rule;
5428 struct classifier *cls;
5431 if (table_id >= N_TABLES) {
5435 cls = &ofproto->up.tables[table_id].cls;
5436 frag = (flow->nw_frag & FLOW_NW_FRAG_ANY) != 0;
5437 if (frag && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
5438 /* We must pretend that transport ports are unavailable. */
5439 struct flow ofpc_normal_flow = *flow;
5440 ofpc_normal_flow.tp_src = htons(0);
5441 ofpc_normal_flow.tp_dst = htons(0);
5442 cls_rule = classifier_lookup(cls, &ofpc_normal_flow);
5443 } else if (frag && ofproto->up.frag_handling == OFPC_FRAG_DROP) {
5444 cls_rule = &ofproto->drop_frags_rule->up.cr;
5446 cls_rule = classifier_lookup(cls, flow);
5448 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
5451 static struct rule_dpif *
5452 rule_dpif_miss_rule(struct ofproto_dpif *ofproto, const struct flow *flow)
5454 struct ofport_dpif *port;
5456 port = get_ofp_port(ofproto, flow->in_port);
5458 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, flow->in_port);
5459 return ofproto->miss_rule;
5462 if (port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN) {
5463 return ofproto->no_packet_in_rule;
5465 return ofproto->miss_rule;
5469 complete_operation(struct rule_dpif *rule)
5471 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5473 rule_invalidate(rule);
5475 struct dpif_completion *c = xmalloc(sizeof *c);
5476 c->op = rule->up.pending;
5477 list_push_back(&ofproto->completions, &c->list_node);
5479 ofoperation_complete(rule->up.pending, 0);
5483 static struct rule *
5486 struct rule_dpif *rule = xmalloc(sizeof *rule);
5491 rule_dealloc(struct rule *rule_)
5493 struct rule_dpif *rule = rule_dpif_cast(rule_);
5498 rule_construct(struct rule *rule_)
5500 struct rule_dpif *rule = rule_dpif_cast(rule_);
5501 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5502 struct rule_dpif *victim;
5505 rule->packet_count = 0;
5506 rule->byte_count = 0;
5508 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
5509 if (victim && !list_is_empty(&victim->facets)) {
5510 struct facet *facet;
5512 rule->facets = victim->facets;
5513 list_moved(&rule->facets);
5514 LIST_FOR_EACH (facet, list_node, &rule->facets) {
5515 /* XXX: We're only clearing our local counters here. It's possible
5516 * that quite a few packets are unaccounted for in the datapath
5517 * statistics. These will be accounted to the new rule instead of
5518 * cleared as required. This could be fixed by clearing out the
5519 * datapath statistics for this facet, but currently it doesn't
5521 facet_reset_counters(facet);
5525 /* Must avoid list_moved() in this case. */
5526 list_init(&rule->facets);
5529 table_id = rule->up.table_id;
5531 rule->tag = victim->tag;
5532 } else if (table_id == 0) {
5537 miniflow_expand(&rule->up.cr.match.flow, &flow);
5538 rule->tag = rule_calculate_tag(&flow, &rule->up.cr.match.mask,
5539 ofproto->tables[table_id].basis);
5542 complete_operation(rule);
5547 rule_destruct(struct rule *rule_)
5549 struct rule_dpif *rule = rule_dpif_cast(rule_);
5550 struct facet *facet, *next_facet;
5552 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
5553 facet_revalidate(facet);
5556 complete_operation(rule);
5560 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
5562 struct rule_dpif *rule = rule_dpif_cast(rule_);
5564 /* push_all_stats() can handle flow misses which, when using the learn
5565 * action, can cause rules to be added and deleted. This can corrupt our
5566 * caller's datastructures which assume that rule_get_stats() doesn't have
5567 * an impact on the flow table. To be safe, we disable miss handling. */
5568 push_all_stats__(false);
5570 /* Start from historical data for 'rule' itself that are no longer tracked
5571 * in facets. This counts, for example, facets that have expired. */
5572 *packets = rule->packet_count;
5573 *bytes = rule->byte_count;
5577 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
5578 struct ofpbuf *packet)
5580 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5581 struct initial_vals initial_vals;
5582 struct dpif_flow_stats stats;
5583 struct xlate_out xout;
5584 struct xlate_in xin;
5586 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
5587 rule_credit_stats(rule, &stats);
5589 initial_vals.vlan_tci = flow->vlan_tci;
5590 initial_vals.tunnel_ip_tos = flow->tunnel.ip_tos;
5591 xlate_in_init(&xin, ofproto, flow, &initial_vals, rule, stats.tcp_flags,
5593 xin.resubmit_stats = &stats;
5594 xlate_actions(&xin, &xout);
5596 execute_odp_actions(ofproto, flow, xout.odp_actions.data,
5597 xout.odp_actions.size, packet);
5599 xlate_out_uninit(&xout);
5603 rule_execute(struct rule *rule, const struct flow *flow,
5604 struct ofpbuf *packet)
5606 rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
5607 ofpbuf_delete(packet);
5612 rule_modify_actions(struct rule *rule_)
5614 struct rule_dpif *rule = rule_dpif_cast(rule_);
5616 complete_operation(rule);
5619 /* Sends 'packet' out 'ofport'.
5620 * May modify 'packet'.
5621 * Returns 0 if successful, otherwise a positive errno value. */
5623 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
5625 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
5626 uint64_t odp_actions_stub[1024 / 8];
5627 struct ofpbuf key, odp_actions;
5628 struct odputil_keybuf keybuf;
5633 flow_extract(packet, 0, 0, NULL, OFPP_LOCAL, &flow);
5634 if (netdev_vport_is_patch(ofport->up.netdev)) {
5635 struct ofproto_dpif *peer_ofproto;
5636 struct dpif_flow_stats stats;
5637 struct ofport_dpif *peer;
5638 struct rule_dpif *rule;
5640 peer = ofport_get_peer(ofport);
5645 dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
5646 netdev_vport_inc_tx(ofport->up.netdev, &stats);
5647 netdev_vport_inc_rx(peer->up.netdev, &stats);
5649 flow.in_port = peer->up.ofp_port;
5650 peer_ofproto = ofproto_dpif_cast(peer->up.ofproto);
5651 rule = rule_dpif_lookup(peer_ofproto, &flow);
5652 rule_dpif_execute(rule, &flow, packet);
5657 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5659 if (ofport->tnl_port) {
5660 struct dpif_flow_stats stats;
5662 odp_port = tnl_port_send(ofport->tnl_port, &flow);
5663 if (odp_port == OVSP_NONE) {
5667 dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
5668 netdev_vport_inc_tx(ofport->up.netdev, &stats);
5669 odp_put_tunnel_action(&flow.tunnel, &odp_actions);
5670 odp_put_skb_mark_action(flow.skb_mark, &odp_actions);
5672 odp_port = vsp_realdev_to_vlandev(ofproto, ofport->odp_port,
5674 if (odp_port != ofport->odp_port) {
5675 eth_pop_vlan(packet);
5676 flow.vlan_tci = htons(0);
5680 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5681 odp_flow_key_from_flow(&key, &flow,
5682 ofp_port_to_odp_port(ofproto, flow.in_port));
5684 compose_sflow_action(ofproto, &odp_actions, &flow, odp_port);
5685 compose_ipfix_action(ofproto, &odp_actions, &flow);
5687 nl_msg_put_u32(&odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port);
5688 error = dpif_execute(ofproto->backer->dpif,
5690 odp_actions.data, odp_actions.size,
5692 ofpbuf_uninit(&odp_actions);
5695 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)",
5696 ofproto->up.name, odp_port, strerror(error));
5698 ofproto_update_local_port_stats(ofport->up.ofproto, packet->size, 0);
5702 /* OpenFlow to datapath action translation. */
5704 static bool may_receive(const struct ofport_dpif *, struct xlate_ctx *);
5705 static void do_xlate_actions(const struct ofpact *, size_t ofpacts_len,
5706 struct xlate_ctx *);
5707 static void xlate_normal(struct xlate_ctx *);
5709 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
5710 * The action will state 'slow' as the reason that the action is in the slow
5711 * path. (This is purely informational: it allows a human viewing "ovs-dpctl
5712 * dump-flows" output to see why a flow is in the slow path.)
5714 * The 'stub_size' bytes in 'stub' will be used to store the action.
5715 * 'stub_size' must be large enough for the action.
5717 * The action and its size will be stored in '*actionsp' and '*actions_lenp',
5720 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
5721 enum slow_path_reason slow,
5722 uint64_t *stub, size_t stub_size,
5723 const struct nlattr **actionsp, size_t *actions_lenp)
5725 union user_action_cookie cookie;
5728 cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
5729 cookie.slow_path.unused = 0;
5730 cookie.slow_path.reason = slow;
5732 ofpbuf_use_stack(&buf, stub, stub_size);
5733 if (slow & (SLOW_CFM | SLOW_LACP | SLOW_STP)) {
5734 uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif, UINT32_MAX);
5735 odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
5737 put_userspace_action(ofproto, &buf, flow, &cookie,
5738 sizeof cookie.slow_path);
5740 *actionsp = buf.data;
5741 *actions_lenp = buf.size;
5745 put_userspace_action(const struct ofproto_dpif *ofproto,
5746 struct ofpbuf *odp_actions,
5747 const struct flow *flow,
5748 const union user_action_cookie *cookie,
5749 const size_t cookie_size)
5753 pid = dpif_port_get_pid(ofproto->backer->dpif,
5754 ofp_port_to_odp_port(ofproto, flow->in_port));
5756 return odp_put_userspace_action(pid, cookie, cookie_size, odp_actions);
5759 /* Compose SAMPLE action for sFlow or IPFIX. The given probability is
5760 * the number of packets out of UINT32_MAX to sample. The given
5761 * cookie is passed back in the callback for each sampled packet.
5764 compose_sample_action(const struct ofproto_dpif *ofproto,
5765 struct ofpbuf *odp_actions,
5766 const struct flow *flow,
5767 const uint32_t probability,
5768 const union user_action_cookie *cookie,
5769 const size_t cookie_size)
5771 size_t sample_offset, actions_offset;
5774 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
5776 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
5778 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
5779 cookie_offset = put_userspace_action(ofproto, odp_actions, flow, cookie,
5782 nl_msg_end_nested(odp_actions, actions_offset);
5783 nl_msg_end_nested(odp_actions, sample_offset);
5784 return cookie_offset;
5788 compose_sflow_cookie(const struct ofproto_dpif *ofproto,
5789 ovs_be16 vlan_tci, uint32_t odp_port,
5790 unsigned int n_outputs, union user_action_cookie *cookie)
5794 cookie->type = USER_ACTION_COOKIE_SFLOW;
5795 cookie->sflow.vlan_tci = vlan_tci;
5797 /* See http://www.sflow.org/sflow_version_5.txt (search for "Input/output
5798 * port information") for the interpretation of cookie->output. */
5799 switch (n_outputs) {
5801 /* 0x40000000 | 256 means "packet dropped for unknown reason". */
5802 cookie->sflow.output = 0x40000000 | 256;
5806 ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port);
5808 cookie->sflow.output = ifindex;
5813 /* 0x80000000 means "multiple output ports. */
5814 cookie->sflow.output = 0x80000000 | n_outputs;
5819 /* Compose SAMPLE action for sFlow bridge sampling. */
5821 compose_sflow_action(const struct ofproto_dpif *ofproto,
5822 struct ofpbuf *odp_actions,
5823 const struct flow *flow,
5826 uint32_t probability;
5827 union user_action_cookie cookie;
5829 if (!ofproto->sflow || flow->in_port == OFPP_NONE) {
5833 probability = dpif_sflow_get_probability(ofproto->sflow);
5834 compose_sflow_cookie(ofproto, htons(0), odp_port,
5835 odp_port == OVSP_NONE ? 0 : 1, &cookie);
5837 return compose_sample_action(ofproto, odp_actions, flow, probability,
5838 &cookie, sizeof cookie.sflow);
5842 compose_flow_sample_cookie(uint16_t probability, uint32_t collector_set_id,
5843 uint32_t obs_domain_id, uint32_t obs_point_id,
5844 union user_action_cookie *cookie)
5846 cookie->type = USER_ACTION_COOKIE_FLOW_SAMPLE;
5847 cookie->flow_sample.probability = probability;
5848 cookie->flow_sample.collector_set_id = collector_set_id;
5849 cookie->flow_sample.obs_domain_id = obs_domain_id;
5850 cookie->flow_sample.obs_point_id = obs_point_id;
5854 compose_ipfix_cookie(union user_action_cookie *cookie)
5856 cookie->type = USER_ACTION_COOKIE_IPFIX;
5859 /* Compose SAMPLE action for IPFIX bridge sampling. */
5861 compose_ipfix_action(const struct ofproto_dpif *ofproto,
5862 struct ofpbuf *odp_actions,
5863 const struct flow *flow)
5865 uint32_t probability;
5866 union user_action_cookie cookie;
5868 if (!ofproto->ipfix || flow->in_port == OFPP_NONE) {
5872 probability = dpif_ipfix_get_bridge_exporter_probability(ofproto->ipfix);
5873 compose_ipfix_cookie(&cookie);
5875 compose_sample_action(ofproto, odp_actions, flow, probability,
5876 &cookie, sizeof cookie.ipfix);
5879 /* SAMPLE action for sFlow must be first action in any given list of
5880 * actions. At this point we do not have all information required to
5881 * build it. So try to build sample action as complete as possible. */
5883 add_sflow_action(struct xlate_ctx *ctx)
5885 ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto,
5886 &ctx->xout->odp_actions,
5887 &ctx->xin->flow, OVSP_NONE);
5888 ctx->sflow_odp_port = 0;
5889 ctx->sflow_n_outputs = 0;
5892 /* SAMPLE action for IPFIX must be 1st or 2nd action in any given list
5893 * of actions, eventually after the SAMPLE action for sFlow. */
5895 add_ipfix_action(struct xlate_ctx *ctx)
5897 compose_ipfix_action(ctx->ofproto, &ctx->xout->odp_actions,
5901 /* Fix SAMPLE action according to data collected while composing ODP actions.
5902 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
5903 * USERSPACE action's user-cookie which is required for sflow. */
5905 fix_sflow_action(struct xlate_ctx *ctx)
5907 const struct flow *base = &ctx->base_flow;
5908 union user_action_cookie *cookie;
5910 if (!ctx->user_cookie_offset) {
5914 cookie = ofpbuf_at(&ctx->xout->odp_actions, ctx->user_cookie_offset,
5915 sizeof cookie->sflow);
5916 ovs_assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
5918 compose_sflow_cookie(ctx->ofproto, base->vlan_tci,
5919 ctx->sflow_odp_port, ctx->sflow_n_outputs, cookie);
5923 compose_output_action__(struct xlate_ctx *ctx, uint16_t ofp_port,
5926 const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
5927 ovs_be16 flow_vlan_tci;
5928 uint32_t flow_skb_mark;
5929 uint8_t flow_nw_tos;
5930 struct priority_to_dscp *pdscp;
5931 uint32_t out_port, odp_port;
5933 /* If 'struct flow' gets additional metadata, we'll need to zero it out
5934 * before traversing a patch port. */
5935 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 20);
5938 xlate_report(ctx, "Nonexistent output port");
5940 } else if (ofport->up.pp.config & OFPUTIL_PC_NO_FWD) {
5941 xlate_report(ctx, "OFPPC_NO_FWD set, skipping output");
5943 } else if (check_stp && !stp_forward_in_state(ofport->stp_state)) {
5944 xlate_report(ctx, "STP not in forwarding state, skipping output");
5948 if (netdev_vport_is_patch(ofport->up.netdev)) {
5949 struct ofport_dpif *peer = ofport_get_peer(ofport);
5950 struct flow old_flow = ctx->xin->flow;
5951 const struct ofproto_dpif *peer_ofproto;
5952 enum slow_path_reason special;
5953 struct ofport_dpif *in_port;
5956 xlate_report(ctx, "Nonexistent patch port peer");
5960 peer_ofproto = ofproto_dpif_cast(peer->up.ofproto);
5961 if (peer_ofproto->backer != ctx->ofproto->backer) {
5962 xlate_report(ctx, "Patch port peer on a different datapath");
5966 ctx->ofproto = ofproto_dpif_cast(peer->up.ofproto);
5967 ctx->xin->flow.in_port = peer->up.ofp_port;
5968 ctx->xin->flow.metadata = htonll(0);
5969 memset(&ctx->xin->flow.tunnel, 0, sizeof ctx->xin->flow.tunnel);
5970 memset(ctx->xin->flow.regs, 0, sizeof ctx->xin->flow.regs);
5972 in_port = get_ofp_port(ctx->ofproto, ctx->xin->flow.in_port);
5973 special = process_special(ctx->ofproto, &ctx->xin->flow, in_port,
5976 ctx->xout->slow = special;
5977 } else if (!in_port || may_receive(in_port, ctx)) {
5978 if (!in_port || stp_forward_in_state(in_port->stp_state)) {
5979 xlate_table_action(ctx, ctx->xin->flow.in_port, 0, true);
5981 /* Forwarding is disabled by STP. Let OFPP_NORMAL and the
5982 * learning action look at the packet, then drop it. */
5983 struct flow old_base_flow = ctx->base_flow;
5984 size_t old_size = ctx->xout->odp_actions.size;
5985 xlate_table_action(ctx, ctx->xin->flow.in_port, 0, true);
5986 ctx->base_flow = old_base_flow;
5987 ctx->xout->odp_actions.size = old_size;
5991 ctx->xin->flow = old_flow;
5992 ctx->ofproto = ofproto_dpif_cast(ofport->up.ofproto);
5994 if (ctx->xin->resubmit_stats) {
5995 netdev_vport_inc_tx(ofport->up.netdev, ctx->xin->resubmit_stats);
5996 netdev_vport_inc_rx(peer->up.netdev, ctx->xin->resubmit_stats);
6002 flow_vlan_tci = ctx->xin->flow.vlan_tci;
6003 flow_skb_mark = ctx->xin->flow.skb_mark;
6004 flow_nw_tos = ctx->xin->flow.nw_tos;
6006 pdscp = get_priority(ofport, ctx->xin->flow.skb_priority);
6008 ctx->xin->flow.nw_tos &= ~IP_DSCP_MASK;
6009 ctx->xin->flow.nw_tos |= pdscp->dscp;
6012 if (ofport->tnl_port) {
6013 /* Save tunnel metadata so that changes made due to
6014 * the Logical (tunnel) Port are not visible for any further
6015 * matches, while explicit set actions on tunnel metadata are.
6017 struct flow_tnl flow_tnl = ctx->xin->flow.tunnel;
6018 odp_port = tnl_port_send(ofport->tnl_port, &ctx->xin->flow);
6019 if (odp_port == OVSP_NONE) {
6020 xlate_report(ctx, "Tunneling decided against output");
6021 goto out; /* restore flow_nw_tos */
6024 if (ctx->xin->resubmit_stats) {
6025 netdev_vport_inc_tx(ofport->up.netdev, ctx->xin->resubmit_stats);
6027 out_port = odp_port;
6028 commit_odp_tunnel_action(&ctx->xin->flow, &ctx->base_flow,
6029 &ctx->xout->odp_actions);
6030 ctx->xin->flow.tunnel = flow_tnl; /* Restore tunnel metadata */
6032 uint16_t vlandev_port;
6033 odp_port = ofport->odp_port;
6034 vlandev_port = vsp_realdev_to_vlandev(ctx->ofproto, ofp_port,
6035 ctx->xin->flow.vlan_tci);
6036 if (vlandev_port == ofp_port) {
6037 out_port = odp_port;
6039 out_port = ofp_port_to_odp_port(ctx->ofproto, vlandev_port);
6040 ctx->xin->flow.vlan_tci = htons(0);
6042 ctx->xin->flow.skb_mark &= ~IPSEC_MARK;
6044 commit_odp_actions(&ctx->xin->flow, &ctx->base_flow,
6045 &ctx->xout->odp_actions);
6046 nl_msg_put_u32(&ctx->xout->odp_actions, OVS_ACTION_ATTR_OUTPUT, out_port);
6048 ctx->sflow_odp_port = odp_port;
6049 ctx->sflow_n_outputs++;
6050 ctx->xout->nf_output_iface = ofp_port;
6053 ctx->xin->flow.vlan_tci = flow_vlan_tci;
6054 ctx->xin->flow.skb_mark = flow_skb_mark;
6056 ctx->xin->flow.nw_tos = flow_nw_tos;
6060 compose_output_action(struct xlate_ctx *ctx, uint16_t ofp_port)
6062 compose_output_action__(ctx, ofp_port, true);
6066 tag_the_flow(struct xlate_ctx *ctx, struct rule_dpif *rule)
6068 struct ofproto_dpif *ofproto = ctx->ofproto;
6069 uint8_t table_id = ctx->table_id;
6071 if (table_id > 0 && table_id < N_TABLES) {
6072 struct table_dpif *table = &ofproto->tables[table_id];
6073 if (table->other_table) {
6074 ctx->xout->tags |= (rule && rule->tag
6076 : rule_calculate_tag(&ctx->xin->flow,
6077 &table->other_table->mask,
6083 /* Common rule processing in one place to avoid duplicating code. */
6084 static struct rule_dpif *
6085 ctx_rule_hooks(struct xlate_ctx *ctx, struct rule_dpif *rule,
6088 if (ctx->xin->resubmit_hook) {
6089 ctx->xin->resubmit_hook(ctx, rule);
6091 if (rule == NULL && may_packet_in) {
6093 * check if table configuration flags
6094 * OFPTC_TABLE_MISS_CONTROLLER, default.
6095 * OFPTC_TABLE_MISS_CONTINUE,
6096 * OFPTC_TABLE_MISS_DROP
6097 * When OF1.0, OFPTC_TABLE_MISS_CONTINUE is used. What to do?
6099 rule = rule_dpif_miss_rule(ctx->ofproto, &ctx->xin->flow);
6101 if (rule && ctx->xin->resubmit_stats) {
6102 rule_credit_stats(rule, ctx->xin->resubmit_stats);
6108 xlate_table_action(struct xlate_ctx *ctx,
6109 uint16_t in_port, uint8_t table_id, bool may_packet_in)
6111 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
6112 struct rule_dpif *rule;
6113 uint16_t old_in_port = ctx->xin->flow.in_port;
6114 uint8_t old_table_id = ctx->table_id;
6116 ctx->table_id = table_id;
6118 /* Look up a flow with 'in_port' as the input port. */
6119 ctx->xin->flow.in_port = in_port;
6120 rule = rule_dpif_lookup__(ctx->ofproto, &ctx->xin->flow, table_id);
6122 tag_the_flow(ctx, rule);
6124 /* Restore the original input port. Otherwise OFPP_NORMAL and
6125 * OFPP_IN_PORT will have surprising behavior. */
6126 ctx->xin->flow.in_port = old_in_port;
6128 rule = ctx_rule_hooks(ctx, rule, may_packet_in);
6131 struct rule_dpif *old_rule = ctx->rule;
6135 do_xlate_actions(rule->up.ofpacts, rule->up.ofpacts_len, ctx);
6136 ctx->rule = old_rule;
6140 ctx->table_id = old_table_id;
6142 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
6144 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
6145 MAX_RESUBMIT_RECURSION);
6146 ctx->max_resubmit_trigger = true;
6151 xlate_ofpact_resubmit(struct xlate_ctx *ctx,
6152 const struct ofpact_resubmit *resubmit)
6157 in_port = resubmit->in_port;
6158 if (in_port == OFPP_IN_PORT) {
6159 in_port = ctx->xin->flow.in_port;
6162 table_id = resubmit->table_id;
6163 if (table_id == 255) {
6164 table_id = ctx->table_id;
6167 xlate_table_action(ctx, in_port, table_id, false);
6171 flood_packets(struct xlate_ctx *ctx, bool all)
6173 struct ofport_dpif *ofport;
6175 HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) {
6176 uint16_t ofp_port = ofport->up.ofp_port;
6178 if (ofp_port == ctx->xin->flow.in_port) {
6183 compose_output_action__(ctx, ofp_port, false);
6184 } else if (!(ofport->up.pp.config & OFPUTIL_PC_NO_FLOOD)) {
6185 compose_output_action(ctx, ofp_port);
6189 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
6193 execute_controller_action(struct xlate_ctx *ctx, int len,
6194 enum ofp_packet_in_reason reason,
6195 uint16_t controller_id)
6197 struct ofputil_packet_in pin;
6198 struct ofpbuf *packet;
6200 ovs_assert(!ctx->xout->slow || ctx->xout->slow == SLOW_CONTROLLER);
6201 ctx->xout->slow = SLOW_CONTROLLER;
6202 if (!ctx->xin->packet) {
6206 packet = ofpbuf_clone(ctx->xin->packet);
6208 if (packet->l2 && packet->l3) {
6209 struct eth_header *eh;
6210 uint16_t mpls_depth;
6212 eth_pop_vlan(packet);
6215 memcpy(eh->eth_src, ctx->xin->flow.dl_src, sizeof eh->eth_src);
6216 memcpy(eh->eth_dst, ctx->xin->flow.dl_dst, sizeof eh->eth_dst);
6218 if (ctx->xin->flow.vlan_tci & htons(VLAN_CFI)) {
6219 eth_push_vlan(packet, ctx->xin->flow.vlan_tci);
6222 mpls_depth = eth_mpls_depth(packet);
6224 if (mpls_depth < ctx->xin->flow.mpls_depth) {
6225 push_mpls(packet, ctx->xin->flow.dl_type, ctx->xin->flow.mpls_lse);
6226 } else if (mpls_depth > ctx->xin->flow.mpls_depth) {
6227 pop_mpls(packet, ctx->xin->flow.dl_type);
6228 } else if (mpls_depth) {
6229 set_mpls_lse(packet, ctx->xin->flow.mpls_lse);
6233 if (ctx->xin->flow.dl_type == htons(ETH_TYPE_IP)) {
6234 packet_set_ipv4(packet, ctx->xin->flow.nw_src,
6235 ctx->xin->flow.nw_dst, ctx->xin->flow.nw_tos,
6236 ctx->xin->flow.nw_ttl);
6240 if (ctx->xin->flow.nw_proto == IPPROTO_TCP) {
6241 packet_set_tcp_port(packet, ctx->xin->flow.tp_src,
6242 ctx->xin->flow.tp_dst);
6243 } else if (ctx->xin->flow.nw_proto == IPPROTO_UDP) {
6244 packet_set_udp_port(packet, ctx->xin->flow.tp_src,
6245 ctx->xin->flow.tp_dst);
6251 pin.packet = packet->data;
6252 pin.packet_len = packet->size;
6253 pin.reason = reason;
6254 pin.controller_id = controller_id;
6255 pin.table_id = ctx->table_id;
6256 pin.cookie = ctx->rule ? ctx->rule->up.flow_cookie : 0;
6259 flow_get_metadata(&ctx->xin->flow, &pin.fmd);
6261 connmgr_send_packet_in(ctx->ofproto->up.connmgr, &pin);
6262 ofpbuf_delete(packet);
6266 execute_mpls_push_action(struct xlate_ctx *ctx, ovs_be16 eth_type)
6268 ovs_assert(eth_type_mpls(eth_type));
6270 if (ctx->base_flow.mpls_depth) {
6271 ctx->xin->flow.mpls_lse &= ~htonl(MPLS_BOS_MASK);
6272 ctx->xin->flow.mpls_depth++;
6277 if (ctx->xin->flow.dl_type == htons(ETH_TYPE_IPV6)) {
6278 label = htonl(0x2); /* IPV6 Explicit Null. */
6280 label = htonl(0x0); /* IPV4 Explicit Null. */
6282 tc = (ctx->xin->flow.nw_tos & IP_DSCP_MASK) >> 2;
6283 ttl = ctx->xin->flow.nw_ttl ? ctx->xin->flow.nw_ttl : 0x40;
6284 ctx->xin->flow.mpls_lse = set_mpls_lse_values(ttl, tc, 1, label);
6285 ctx->xin->flow.mpls_depth = 1;
6287 ctx->xin->flow.dl_type = eth_type;
6291 execute_mpls_pop_action(struct xlate_ctx *ctx, ovs_be16 eth_type)
6293 ovs_assert(eth_type_mpls(ctx->xin->flow.dl_type));
6294 ovs_assert(!eth_type_mpls(eth_type));
6296 if (ctx->xin->flow.mpls_depth) {
6297 ctx->xin->flow.mpls_depth--;
6298 ctx->xin->flow.mpls_lse = htonl(0);
6299 if (!ctx->xin->flow.mpls_depth) {
6300 ctx->xin->flow.dl_type = eth_type;
6306 compose_dec_ttl(struct xlate_ctx *ctx, struct ofpact_cnt_ids *ids)
6308 if (ctx->xin->flow.dl_type != htons(ETH_TYPE_IP) &&
6309 ctx->xin->flow.dl_type != htons(ETH_TYPE_IPV6)) {
6313 if (ctx->xin->flow.nw_ttl > 1) {
6314 ctx->xin->flow.nw_ttl--;
6319 for (i = 0; i < ids->n_controllers; i++) {
6320 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL,
6324 /* Stop processing for current table. */
6330 execute_set_mpls_ttl_action(struct xlate_ctx *ctx, uint8_t ttl)
6332 if (!eth_type_mpls(ctx->xin->flow.dl_type)) {
6336 set_mpls_lse_ttl(&ctx->xin->flow.mpls_lse, ttl);
6341 execute_dec_mpls_ttl_action(struct xlate_ctx *ctx)
6343 uint8_t ttl = mpls_lse_to_ttl(ctx->xin->flow.mpls_lse);
6345 if (!eth_type_mpls(ctx->xin->flow.dl_type)) {
6351 set_mpls_lse_ttl(&ctx->xin->flow.mpls_lse, ttl);
6354 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL, 0);
6356 /* Stop processing for current table. */
6362 xlate_output_action(struct xlate_ctx *ctx,
6363 uint16_t port, uint16_t max_len, bool may_packet_in)
6365 uint16_t prev_nf_output_iface = ctx->xout->nf_output_iface;
6367 ctx->xout->nf_output_iface = NF_OUT_DROP;
6371 compose_output_action(ctx, ctx->xin->flow.in_port);
6374 xlate_table_action(ctx, ctx->xin->flow.in_port, 0, may_packet_in);
6380 flood_packets(ctx, false);
6383 flood_packets(ctx, true);
6385 case OFPP_CONTROLLER:
6386 execute_controller_action(ctx, max_len, OFPR_ACTION, 0);
6392 if (port != ctx->xin->flow.in_port) {
6393 compose_output_action(ctx, port);
6395 xlate_report(ctx, "skipping output to input port");
6400 if (prev_nf_output_iface == NF_OUT_FLOOD) {
6401 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
6402 } else if (ctx->xout->nf_output_iface == NF_OUT_DROP) {
6403 ctx->xout->nf_output_iface = prev_nf_output_iface;
6404 } else if (prev_nf_output_iface != NF_OUT_DROP &&
6405 ctx->xout->nf_output_iface != NF_OUT_FLOOD) {
6406 ctx->xout->nf_output_iface = NF_OUT_MULTI;
6411 xlate_output_reg_action(struct xlate_ctx *ctx,
6412 const struct ofpact_output_reg *or)
6414 uint64_t port = mf_get_subfield(&or->src, &ctx->xin->flow);
6415 if (port <= UINT16_MAX) {
6416 xlate_output_action(ctx, port, or->max_len, false);
6421 xlate_enqueue_action(struct xlate_ctx *ctx,
6422 const struct ofpact_enqueue *enqueue)
6424 uint16_t ofp_port = enqueue->port;
6425 uint32_t queue_id = enqueue->queue;
6426 uint32_t flow_priority, priority;
6429 /* Translate queue to priority. */
6430 error = dpif_queue_to_priority(ctx->ofproto->backer->dpif,
6431 queue_id, &priority);
6433 /* Fall back to ordinary output action. */
6434 xlate_output_action(ctx, enqueue->port, 0, false);
6438 /* Check output port. */
6439 if (ofp_port == OFPP_IN_PORT) {
6440 ofp_port = ctx->xin->flow.in_port;
6441 } else if (ofp_port == ctx->xin->flow.in_port) {
6445 /* Add datapath actions. */
6446 flow_priority = ctx->xin->flow.skb_priority;
6447 ctx->xin->flow.skb_priority = priority;
6448 compose_output_action(ctx, ofp_port);
6449 ctx->xin->flow.skb_priority = flow_priority;
6451 /* Update NetFlow output port. */
6452 if (ctx->xout->nf_output_iface == NF_OUT_DROP) {
6453 ctx->xout->nf_output_iface = ofp_port;
6454 } else if (ctx->xout->nf_output_iface != NF_OUT_FLOOD) {
6455 ctx->xout->nf_output_iface = NF_OUT_MULTI;
6460 xlate_set_queue_action(struct xlate_ctx *ctx, uint32_t queue_id)
6462 uint32_t skb_priority;
6464 if (!dpif_queue_to_priority(ctx->ofproto->backer->dpif,
6465 queue_id, &skb_priority)) {
6466 ctx->xin->flow.skb_priority = skb_priority;
6468 /* Couldn't translate queue to a priority. Nothing to do. A warning
6469 * has already been logged. */
6474 slave_enabled_cb(uint16_t ofp_port, void *ofproto_)
6476 struct ofproto_dpif *ofproto = ofproto_;
6477 struct ofport_dpif *port;
6487 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
6490 port = get_ofp_port(ofproto, ofp_port);
6491 return port ? port->may_enable : false;
6496 xlate_bundle_action(struct xlate_ctx *ctx,
6497 const struct ofpact_bundle *bundle)
6501 port = bundle_execute(bundle, &ctx->xin->flow, slave_enabled_cb,
6503 if (bundle->dst.field) {
6504 nxm_reg_load(&bundle->dst, port, &ctx->xin->flow);
6506 xlate_output_action(ctx, port, 0, false);
6511 xlate_learn_action(struct xlate_ctx *ctx,
6512 const struct ofpact_learn *learn)
6514 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
6515 struct ofputil_flow_mod fm;
6516 uint64_t ofpacts_stub[1024 / 8];
6517 struct ofpbuf ofpacts;
6520 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
6521 learn_execute(learn, &ctx->xin->flow, &fm, &ofpacts);
6523 error = ofproto_flow_mod(&ctx->ofproto->up, &fm);
6524 if (error && !VLOG_DROP_WARN(&rl)) {
6525 VLOG_WARN("learning action failed to modify flow table (%s)",
6526 ofperr_get_name(error));
6529 ofpbuf_uninit(&ofpacts);
6532 /* Reduces '*timeout' to no more than 'max'. A value of zero in either case
6533 * means "infinite". */
6535 reduce_timeout(uint16_t max, uint16_t *timeout)
6537 if (max && (!*timeout || *timeout > max)) {
6543 xlate_fin_timeout(struct xlate_ctx *ctx,
6544 const struct ofpact_fin_timeout *oft)
6546 if (ctx->xin->tcp_flags & (TCP_FIN | TCP_RST) && ctx->rule) {
6547 struct rule_dpif *rule = ctx->rule;
6549 reduce_timeout(oft->fin_idle_timeout, &rule->up.idle_timeout);
6550 reduce_timeout(oft->fin_hard_timeout, &rule->up.hard_timeout);
6555 xlate_sample_action(struct xlate_ctx *ctx,
6556 const struct ofpact_sample *os)
6558 union user_action_cookie cookie;
6559 /* Scale the probability from 16-bit to 32-bit while representing
6560 * the same percentage. */
6561 uint32_t probability = (os->probability << 16) | os->probability;
6563 commit_odp_actions(&ctx->xin->flow, &ctx->base_flow,
6564 &ctx->xout->odp_actions);
6566 compose_flow_sample_cookie(os->probability, os->collector_set_id,
6567 os->obs_domain_id, os->obs_point_id, &cookie);
6568 compose_sample_action(ctx->ofproto, &ctx->xout->odp_actions, &ctx->xin->flow,
6569 probability, &cookie, sizeof cookie.flow_sample);
6573 may_receive(const struct ofport_dpif *port, struct xlate_ctx *ctx)
6575 if (port->up.pp.config & (eth_addr_equals(ctx->xin->flow.dl_dst,
6577 ? OFPUTIL_PC_NO_RECV_STP
6578 : OFPUTIL_PC_NO_RECV)) {
6582 /* Only drop packets here if both forwarding and learning are
6583 * disabled. If just learning is enabled, we need to have
6584 * OFPP_NORMAL and the learning action have a look at the packet
6585 * before we can drop it. */
6586 if (!stp_forward_in_state(port->stp_state)
6587 && !stp_learn_in_state(port->stp_state)) {
6595 tunnel_ecn_ok(struct xlate_ctx *ctx)
6597 if (is_ip_any(&ctx->base_flow)
6598 && (ctx->base_flow.tunnel.ip_tos & IP_ECN_MASK) == IP_ECN_CE) {
6599 if ((ctx->base_flow.nw_tos & IP_ECN_MASK) == IP_ECN_NOT_ECT) {
6600 VLOG_WARN_RL(&rl, "dropping tunnel packet marked ECN CE"
6601 " but is not ECN capable");
6604 /* Set the ECN CE value in the tunneled packet. */
6605 ctx->xin->flow.nw_tos |= IP_ECN_CE;
6613 do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len,
6614 struct xlate_ctx *ctx)
6616 bool was_evictable = true;
6617 const struct ofpact *a;
6620 /* Don't let the rule we're working on get evicted underneath us. */
6621 was_evictable = ctx->rule->up.evictable;
6622 ctx->rule->up.evictable = false;
6625 do_xlate_actions_again:
6626 OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) {
6627 struct ofpact_controller *controller;
6628 const struct ofpact_metadata *metadata;
6636 xlate_output_action(ctx, ofpact_get_OUTPUT(a)->port,
6637 ofpact_get_OUTPUT(a)->max_len, true);
6640 case OFPACT_CONTROLLER:
6641 controller = ofpact_get_CONTROLLER(a);
6642 execute_controller_action(ctx, controller->max_len,
6644 controller->controller_id);
6647 case OFPACT_ENQUEUE:
6648 xlate_enqueue_action(ctx, ofpact_get_ENQUEUE(a));
6651 case OFPACT_SET_VLAN_VID:
6652 ctx->xin->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
6653 ctx->xin->flow.vlan_tci |=
6654 (htons(ofpact_get_SET_VLAN_VID(a)->vlan_vid)
6658 case OFPACT_SET_VLAN_PCP:
6659 ctx->xin->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
6660 ctx->xin->flow.vlan_tci |=
6661 htons((ofpact_get_SET_VLAN_PCP(a)->vlan_pcp << VLAN_PCP_SHIFT)
6665 case OFPACT_STRIP_VLAN:
6666 ctx->xin->flow.vlan_tci = htons(0);
6669 case OFPACT_PUSH_VLAN:
6670 /* XXX 802.1AD(QinQ) */
6671 ctx->xin->flow.vlan_tci = htons(VLAN_CFI);
6674 case OFPACT_SET_ETH_SRC:
6675 memcpy(ctx->xin->flow.dl_src, ofpact_get_SET_ETH_SRC(a)->mac,
6679 case OFPACT_SET_ETH_DST:
6680 memcpy(ctx->xin->flow.dl_dst, ofpact_get_SET_ETH_DST(a)->mac,
6684 case OFPACT_SET_IPV4_SRC:
6685 if (ctx->xin->flow.dl_type == htons(ETH_TYPE_IP)) {
6686 ctx->xin->flow.nw_src = ofpact_get_SET_IPV4_SRC(a)->ipv4;
6690 case OFPACT_SET_IPV4_DST:
6691 if (ctx->xin->flow.dl_type == htons(ETH_TYPE_IP)) {
6692 ctx->xin->flow.nw_dst = ofpact_get_SET_IPV4_DST(a)->ipv4;
6696 case OFPACT_SET_IPV4_DSCP:
6697 /* OpenFlow 1.0 only supports IPv4. */
6698 if (ctx->xin->flow.dl_type == htons(ETH_TYPE_IP)) {
6699 ctx->xin->flow.nw_tos &= ~IP_DSCP_MASK;
6700 ctx->xin->flow.nw_tos |= ofpact_get_SET_IPV4_DSCP(a)->dscp;
6704 case OFPACT_SET_L4_SRC_PORT:
6705 if (is_ip_any(&ctx->xin->flow)) {
6706 ctx->xin->flow.tp_src =
6707 htons(ofpact_get_SET_L4_SRC_PORT(a)->port);
6711 case OFPACT_SET_L4_DST_PORT:
6712 if (is_ip_any(&ctx->xin->flow)) {
6713 ctx->xin->flow.tp_dst =
6714 htons(ofpact_get_SET_L4_DST_PORT(a)->port);
6718 case OFPACT_RESUBMIT:
6719 xlate_ofpact_resubmit(ctx, ofpact_get_RESUBMIT(a));
6722 case OFPACT_SET_TUNNEL:
6723 ctx->xin->flow.tunnel.tun_id =
6724 htonll(ofpact_get_SET_TUNNEL(a)->tun_id);
6727 case OFPACT_SET_QUEUE:
6728 xlate_set_queue_action(ctx, ofpact_get_SET_QUEUE(a)->queue_id);
6731 case OFPACT_POP_QUEUE:
6732 ctx->xin->flow.skb_priority = ctx->orig_skb_priority;
6735 case OFPACT_REG_MOVE:
6736 nxm_execute_reg_move(ofpact_get_REG_MOVE(a), &ctx->xin->flow);
6739 case OFPACT_REG_LOAD:
6740 nxm_execute_reg_load(ofpact_get_REG_LOAD(a), &ctx->xin->flow);
6743 case OFPACT_STACK_PUSH:
6744 nxm_execute_stack_push(ofpact_get_STACK_PUSH(a), &ctx->xin->flow,
6748 case OFPACT_STACK_POP:
6749 nxm_execute_stack_pop(ofpact_get_STACK_POP(a), &ctx->xin->flow,
6753 case OFPACT_PUSH_MPLS:
6754 execute_mpls_push_action(ctx, ofpact_get_PUSH_MPLS(a)->ethertype);
6757 case OFPACT_POP_MPLS:
6758 execute_mpls_pop_action(ctx, ofpact_get_POP_MPLS(a)->ethertype);
6761 case OFPACT_SET_MPLS_TTL:
6762 if (execute_set_mpls_ttl_action(ctx,
6763 ofpact_get_SET_MPLS_TTL(a)->ttl)) {
6768 case OFPACT_DEC_MPLS_TTL:
6769 if (execute_dec_mpls_ttl_action(ctx)) {
6774 case OFPACT_DEC_TTL:
6775 if (compose_dec_ttl(ctx, ofpact_get_DEC_TTL(a))) {
6781 /* Nothing to do. */
6784 case OFPACT_MULTIPATH:
6785 multipath_execute(ofpact_get_MULTIPATH(a), &ctx->xin->flow);
6789 ctx->ofproto->has_bundle_action = true;
6790 xlate_bundle_action(ctx, ofpact_get_BUNDLE(a));
6793 case OFPACT_OUTPUT_REG:
6794 xlate_output_reg_action(ctx, ofpact_get_OUTPUT_REG(a));
6798 ctx->xout->has_learn = true;
6799 if (ctx->xin->may_learn) {
6800 xlate_learn_action(ctx, ofpact_get_LEARN(a));
6808 case OFPACT_FIN_TIMEOUT:
6809 ctx->xout->has_fin_timeout = true;
6810 xlate_fin_timeout(ctx, ofpact_get_FIN_TIMEOUT(a));
6813 case OFPACT_CLEAR_ACTIONS:
6815 * Nothing to do because writa-actions is not supported for now.
6816 * When writa-actions is supported, clear-actions also must
6817 * be supported at the same time.
6821 case OFPACT_WRITE_METADATA:
6822 metadata = ofpact_get_WRITE_METADATA(a);
6823 ctx->xin->flow.metadata &= ~metadata->mask;
6824 ctx->xin->flow.metadata |= metadata->metadata & metadata->mask;
6827 case OFPACT_GOTO_TABLE: {
6828 /* It is assumed that goto-table is the last action. */
6829 struct ofpact_goto_table *ogt = ofpact_get_GOTO_TABLE(a);
6830 struct rule_dpif *rule;
6832 ovs_assert(ctx->table_id < ogt->table_id);
6834 ctx->table_id = ogt->table_id;
6836 /* Look up a flow from the new table. */
6837 rule = rule_dpif_lookup__(ctx->ofproto, &ctx->xin->flow, ctx->table_id);
6839 tag_the_flow(ctx, rule);
6841 rule = ctx_rule_hooks(ctx, rule, true);
6845 ctx->rule->up.evictable = was_evictable;
6848 was_evictable = rule->up.evictable;
6849 rule->up.evictable = false;
6851 /* Tail recursion removal. */
6852 ofpacts = rule->up.ofpacts;
6853 ofpacts_len = rule->up.ofpacts_len;
6854 goto do_xlate_actions_again;
6860 xlate_sample_action(ctx, ofpact_get_SAMPLE(a));
6867 ctx->rule->up.evictable = was_evictable;
6872 xlate_in_init(struct xlate_in *xin, struct ofproto_dpif *ofproto,
6873 const struct flow *flow,
6874 const struct initial_vals *initial_vals,
6875 struct rule_dpif *rule, uint8_t tcp_flags,
6876 const struct ofpbuf *packet)
6878 xin->ofproto = ofproto;
6880 xin->packet = packet;
6881 xin->may_learn = packet != NULL;
6883 xin->ofpacts = NULL;
6884 xin->ofpacts_len = 0;
6885 xin->tcp_flags = tcp_flags;
6886 xin->resubmit_hook = NULL;
6887 xin->report_hook = NULL;
6888 xin->resubmit_stats = NULL;
6891 xin->initial_vals = *initial_vals;
6893 xin->initial_vals.vlan_tci = xin->flow.vlan_tci;
6894 xin->initial_vals.tunnel_ip_tos = xin->flow.tunnel.ip_tos;
6899 xlate_out_uninit(struct xlate_out *xout)
6902 ofpbuf_uninit(&xout->odp_actions);
6906 /* Translates the 'ofpacts_len' bytes of "struct ofpacts" starting at 'ofpacts'
6907 * into datapath actions in 'odp_actions', using 'ctx'. */
6909 xlate_actions(struct xlate_in *xin, struct xlate_out *xout)
6911 /* Normally false. Set to true if we ever hit MAX_RESUBMIT_RECURSION, so
6912 * that in the future we always keep a copy of the original flow for
6913 * tracing purposes. */
6914 static bool hit_resubmit_limit;
6916 enum slow_path_reason special;
6917 const struct ofpact *ofpacts;
6918 struct ofport_dpif *in_port;
6919 struct flow orig_flow;
6920 struct xlate_ctx ctx;
6923 COVERAGE_INC(ofproto_dpif_xlate);
6925 /* Flow initialization rules:
6926 * - 'base_flow' must match the kernel's view of the packet at the
6927 * time that action processing starts. 'flow' represents any
6928 * transformations we wish to make through actions.
6929 * - By default 'base_flow' and 'flow' are the same since the input
6930 * packet matches the output before any actions are applied.
6931 * - When using VLAN splinters, 'base_flow''s VLAN is set to the value
6932 * of the received packet as seen by the kernel. If we later output
6933 * to another device without any modifications this will cause us to
6934 * insert a new tag since the original one was stripped off by the
6936 * - Tunnel 'flow' is largely cleared when transitioning between
6937 * the input and output stages since it does not make sense to output
6938 * a packet with the exact headers that it was received with (i.e.
6939 * the destination IP is us). The one exception is the tun_id, which
6940 * is preserved to allow use in later resubmit lookups and loads into
6942 * - Tunnel 'base_flow' is completely cleared since that is what the
6943 * kernel does. If we wish to maintain the original values an action
6944 * needs to be generated. */
6949 ctx.ofproto = xin->ofproto;
6950 ctx.rule = xin->rule;
6952 ctx.base_flow = ctx.xin->flow;
6953 memset(&ctx.base_flow.tunnel, 0, sizeof ctx.base_flow.tunnel);
6954 ctx.base_flow.vlan_tci = xin->initial_vals.vlan_tci;
6955 ctx.base_flow.tunnel.ip_tos = xin->initial_vals.tunnel_ip_tos;
6959 ctx.xout->has_learn = false;
6960 ctx.xout->has_normal = false;
6961 ctx.xout->has_fin_timeout = false;
6962 ctx.xout->nf_output_iface = NF_OUT_DROP;
6963 ctx.xout->mirrors = 0;
6965 ofpbuf_use_stub(&ctx.xout->odp_actions, ctx.xout->odp_actions_stub,
6966 sizeof ctx.xout->odp_actions_stub);
6967 ofpbuf_reserve(&ctx.xout->odp_actions, NL_A_U32_SIZE);
6970 ctx.max_resubmit_trigger = false;
6971 ctx.orig_skb_priority = ctx.xin->flow.skb_priority;
6976 ofpacts = xin->ofpacts;
6977 ofpacts_len = xin->ofpacts_len;
6978 } else if (xin->rule) {
6979 ofpacts = xin->rule->up.ofpacts;
6980 ofpacts_len = xin->rule->up.ofpacts_len;
6985 ofpbuf_use_stub(&ctx.stack, ctx.init_stack, sizeof ctx.init_stack);
6987 if (ctx.ofproto->has_mirrors || hit_resubmit_limit) {
6988 /* Do this conditionally because the copy is expensive enough that it
6989 * shows up in profiles. */
6990 orig_flow = ctx.xin->flow;
6993 if (ctx.xin->flow.nw_frag & FLOW_NW_FRAG_ANY) {
6994 switch (ctx.ofproto->up.frag_handling) {
6995 case OFPC_FRAG_NORMAL:
6996 /* We must pretend that transport ports are unavailable. */
6997 ctx.xin->flow.tp_src = ctx.base_flow.tp_src = htons(0);
6998 ctx.xin->flow.tp_dst = ctx.base_flow.tp_dst = htons(0);
7001 case OFPC_FRAG_DROP:
7004 case OFPC_FRAG_REASM:
7007 case OFPC_FRAG_NX_MATCH:
7008 /* Nothing to do. */
7011 case OFPC_INVALID_TTL_TO_CONTROLLER:
7016 in_port = get_ofp_port(ctx.ofproto, ctx.xin->flow.in_port);
7017 special = process_special(ctx.ofproto, &ctx.xin->flow, in_port,
7020 ctx.xout->slow = special;
7022 static struct vlog_rate_limit trace_rl = VLOG_RATE_LIMIT_INIT(1, 1);
7023 struct initial_vals initial_vals;
7024 size_t sample_actions_len;
7025 uint32_t local_odp_port;
7027 initial_vals.vlan_tci = ctx.base_flow.vlan_tci;
7028 initial_vals.tunnel_ip_tos = ctx.base_flow.tunnel.ip_tos;
7030 add_sflow_action(&ctx);
7031 add_ipfix_action(&ctx);
7032 sample_actions_len = ctx.xout->odp_actions.size;
7034 if (tunnel_ecn_ok(&ctx) && (!in_port || may_receive(in_port, &ctx))) {
7035 do_xlate_actions(ofpacts, ofpacts_len, &ctx);
7037 /* We've let OFPP_NORMAL and the learning action look at the
7038 * packet, so drop it now if forwarding is disabled. */
7039 if (in_port && !stp_forward_in_state(in_port->stp_state)) {
7040 ctx.xout->odp_actions.size = sample_actions_len;
7044 if (ctx.max_resubmit_trigger && !ctx.xin->resubmit_hook) {
7045 if (!hit_resubmit_limit) {
7046 /* We didn't record the original flow. Make sure we do from
7048 hit_resubmit_limit = true;
7049 } else if (!VLOG_DROP_ERR(&trace_rl)) {
7050 struct ds ds = DS_EMPTY_INITIALIZER;
7052 ofproto_trace(ctx.ofproto, &orig_flow, ctx.xin->packet,
7053 &initial_vals, &ds);
7054 VLOG_ERR("Trace triggered by excessive resubmit "
7055 "recursion:\n%s", ds_cstr(&ds));
7060 local_odp_port = ofp_port_to_odp_port(ctx.ofproto, OFPP_LOCAL);
7061 if (!connmgr_must_output_local(ctx.ofproto->up.connmgr, &ctx.xin->flow,
7063 ctx.xout->odp_actions.data,
7064 ctx.xout->odp_actions.size)) {
7065 compose_output_action(&ctx, OFPP_LOCAL);
7067 if (ctx.ofproto->has_mirrors) {
7068 add_mirror_actions(&ctx, &orig_flow);
7070 fix_sflow_action(&ctx);
7073 ofpbuf_uninit(&ctx.stack);
7076 /* Translates the 'ofpacts_len' bytes of "struct ofpact"s starting at 'ofpacts'
7077 * into datapath actions, using 'ctx', and discards the datapath actions. */
7079 xlate_actions_for_side_effects(struct xlate_in *xin)
7081 struct xlate_out xout;
7083 xlate_actions(xin, &xout);
7084 xlate_out_uninit(&xout);
7088 xlate_report(struct xlate_ctx *ctx, const char *s)
7090 if (ctx->xin->report_hook) {
7091 ctx->xin->report_hook(ctx, s);
7095 /* OFPP_NORMAL implementation. */
7097 static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *);
7099 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
7100 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle',
7101 * the bundle on which the packet was received, returns the VLAN to which the
7104 * Both 'vid' and the return value are in the range 0...4095. */
7106 input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid)
7108 switch (in_bundle->vlan_mode) {
7109 case PORT_VLAN_ACCESS:
7110 return in_bundle->vlan;
7113 case PORT_VLAN_TRUNK:
7116 case PORT_VLAN_NATIVE_UNTAGGED:
7117 case PORT_VLAN_NATIVE_TAGGED:
7118 return vid ? vid : in_bundle->vlan;
7125 /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'.
7126 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
7129 * 'vid' should be the VID obtained from the 802.1Q header that was received as
7130 * part of a packet (specify 0 if there was no 802.1Q header), in the range
7133 input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn)
7135 /* Allow any VID on the OFPP_NONE port. */
7136 if (in_bundle == &ofpp_none_bundle) {
7140 switch (in_bundle->vlan_mode) {
7141 case PORT_VLAN_ACCESS:
7144 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7145 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
7146 "packet received on port %s configured as VLAN "
7147 "%"PRIu16" access port",
7148 in_bundle->ofproto->up.name, vid,
7149 in_bundle->name, in_bundle->vlan);
7155 case PORT_VLAN_NATIVE_UNTAGGED:
7156 case PORT_VLAN_NATIVE_TAGGED:
7158 /* Port must always carry its native VLAN. */
7162 case PORT_VLAN_TRUNK:
7163 if (!ofbundle_includes_vlan(in_bundle, vid)) {
7165 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7166 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet "
7167 "received on port %s not configured for trunking "
7169 in_bundle->ofproto->up.name, vid,
7170 in_bundle->name, vid);
7182 /* Given 'vlan', the VLAN that a packet belongs to, and
7183 * 'out_bundle', a bundle on which the packet is to be output, returns the VID
7184 * that should be included in the 802.1Q header. (If the return value is 0,
7185 * then the 802.1Q header should only be included in the packet if there is a
7188 * Both 'vlan' and the return value are in the range 0...4095. */
7190 output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan)
7192 switch (out_bundle->vlan_mode) {
7193 case PORT_VLAN_ACCESS:
7196 case PORT_VLAN_TRUNK:
7197 case PORT_VLAN_NATIVE_TAGGED:
7200 case PORT_VLAN_NATIVE_UNTAGGED:
7201 return vlan == out_bundle->vlan ? 0 : vlan;
7209 output_normal(struct xlate_ctx *ctx, const struct ofbundle *out_bundle,
7212 struct ofport_dpif *port;
7214 ovs_be16 tci, old_tci;
7216 vid = output_vlan_to_vid(out_bundle, vlan);
7217 if (!out_bundle->bond) {
7218 port = ofbundle_get_a_port(out_bundle);
7220 port = bond_choose_output_slave(out_bundle->bond, &ctx->xin->flow,
7221 vid, &ctx->xout->tags);
7223 /* No slaves enabled, so drop packet. */
7228 old_tci = ctx->xin->flow.vlan_tci;
7230 if (tci || out_bundle->use_priority_tags) {
7231 tci |= ctx->xin->flow.vlan_tci & htons(VLAN_PCP_MASK);
7233 tci |= htons(VLAN_CFI);
7236 ctx->xin->flow.vlan_tci = tci;
7238 compose_output_action(ctx, port->up.ofp_port);
7239 ctx->xin->flow.vlan_tci = old_tci;
7243 mirror_mask_ffs(mirror_mask_t mask)
7245 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
7250 ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
7252 return (bundle->vlan_mode != PORT_VLAN_ACCESS
7253 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
7257 ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan)
7259 return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan);
7262 /* Returns an arbitrary interface within 'bundle'. */
7263 static struct ofport_dpif *
7264 ofbundle_get_a_port(const struct ofbundle *bundle)
7266 return CONTAINER_OF(list_front(&bundle->ports),
7267 struct ofport_dpif, bundle_node);
7271 vlan_is_mirrored(const struct ofmirror *m, int vlan)
7273 return !m->vlans || bitmap_is_set(m->vlans, vlan);
7277 add_mirror_actions(struct xlate_ctx *ctx, const struct flow *orig_flow)
7279 struct ofproto_dpif *ofproto = ctx->ofproto;
7280 mirror_mask_t mirrors;
7281 struct ofbundle *in_bundle;
7284 const struct nlattr *a;
7287 in_bundle = lookup_input_bundle(ctx->ofproto, orig_flow->in_port,
7288 ctx->xin->packet != NULL, NULL);
7292 mirrors = in_bundle->src_mirrors;
7294 /* Drop frames on bundles reserved for mirroring. */
7295 if (in_bundle->mirror_out) {
7296 if (ctx->xin->packet != NULL) {
7297 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7298 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
7299 "%s, which is reserved exclusively for mirroring",
7300 ctx->ofproto->up.name, in_bundle->name);
7306 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
7307 if (!input_vid_is_valid(vid, in_bundle, ctx->xin->packet != NULL)) {
7310 vlan = input_vid_to_vlan(in_bundle, vid);
7312 /* Look at the output ports to check for destination selections. */
7314 NL_ATTR_FOR_EACH (a, left, ctx->xout->odp_actions.data,
7315 ctx->xout->odp_actions.size) {
7316 enum ovs_action_attr type = nl_attr_type(a);
7317 struct ofport_dpif *ofport;
7319 if (type != OVS_ACTION_ATTR_OUTPUT) {
7323 ofport = get_odp_port(ofproto, nl_attr_get_u32(a));
7324 if (ofport && ofport->bundle) {
7325 mirrors |= ofport->bundle->dst_mirrors;
7333 /* Restore the original packet before adding the mirror actions. */
7334 ctx->xin->flow = *orig_flow;
7339 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
7341 if (!vlan_is_mirrored(m, vlan)) {
7342 mirrors = zero_rightmost_1bit(mirrors);
7346 mirrors &= ~m->dup_mirrors;
7347 ctx->xout->mirrors |= m->dup_mirrors;
7349 output_normal(ctx, m->out, vlan);
7350 } else if (vlan != m->out_vlan
7351 && !eth_addr_is_reserved(orig_flow->dl_dst)) {
7352 struct ofbundle *bundle;
7354 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
7355 if (ofbundle_includes_vlan(bundle, m->out_vlan)
7356 && !bundle->mirror_out) {
7357 output_normal(ctx, bundle, m->out_vlan);
7365 update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors,
7366 uint64_t packets, uint64_t bytes)
7372 for (; mirrors; mirrors = zero_rightmost_1bit(mirrors)) {
7375 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
7378 /* In normal circumstances 'm' will not be NULL. However,
7379 * if mirrors are reconfigured, we can temporarily get out
7380 * of sync in facet_revalidate(). We could "correct" the
7381 * mirror list before reaching here, but doing that would
7382 * not properly account the traffic stats we've currently
7383 * accumulated for previous mirror configuration. */
7387 m->packet_count += packets;
7388 m->byte_count += bytes;
7392 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
7393 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
7394 * indicate this; newer upstream kernels use gratuitous ARP requests. */
7396 is_gratuitous_arp(const struct flow *flow)
7398 return (flow->dl_type == htons(ETH_TYPE_ARP)
7399 && eth_addr_is_broadcast(flow->dl_dst)
7400 && (flow->nw_proto == ARP_OP_REPLY
7401 || (flow->nw_proto == ARP_OP_REQUEST
7402 && flow->nw_src == flow->nw_dst)));
7406 update_learning_table(struct ofproto_dpif *ofproto,
7407 const struct flow *flow, int vlan,
7408 struct ofbundle *in_bundle)
7410 struct mac_entry *mac;
7412 /* Don't learn the OFPP_NONE port. */
7413 if (in_bundle == &ofpp_none_bundle) {
7417 if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
7421 mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan);
7422 if (is_gratuitous_arp(flow)) {
7423 /* We don't want to learn from gratuitous ARP packets that are
7424 * reflected back over bond slaves so we lock the learning table. */
7425 if (!in_bundle->bond) {
7426 mac_entry_set_grat_arp_lock(mac);
7427 } else if (mac_entry_is_grat_arp_locked(mac)) {
7432 if (mac_entry_is_new(mac) || mac->port.p != in_bundle) {
7433 /* The log messages here could actually be useful in debugging,
7434 * so keep the rate limit relatively high. */
7435 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
7436 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
7437 "on port %s in VLAN %d",
7438 ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src),
7439 in_bundle->name, vlan);
7441 mac->port.p = in_bundle;
7442 tag_set_add(&ofproto->backer->revalidate_set,
7443 mac_learning_changed(ofproto->ml, mac));
7447 static struct ofbundle *
7448 lookup_input_bundle(const struct ofproto_dpif *ofproto, uint16_t in_port,
7449 bool warn, struct ofport_dpif **in_ofportp)
7451 struct ofport_dpif *ofport;
7453 /* Find the port and bundle for the received packet. */
7454 ofport = get_ofp_port(ofproto, in_port);
7456 *in_ofportp = ofport;
7458 if (ofport && ofport->bundle) {
7459 return ofport->bundle;
7462 /* Special-case OFPP_NONE, which a controller may use as the ingress
7463 * port for traffic that it is sourcing. */
7464 if (in_port == OFPP_NONE) {
7465 return &ofpp_none_bundle;
7468 /* Odd. A few possible reasons here:
7470 * - We deleted a port but there are still a few packets queued up
7473 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
7474 * we don't know about.
7476 * - The ofproto client didn't configure the port as part of a bundle.
7477 * This is particularly likely to happen if a packet was received on the
7478 * port after it was created, but before the client had a chance to
7479 * configure its bundle.
7482 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7484 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
7485 "port %"PRIu16, ofproto->up.name, in_port);
7490 /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or
7491 * dropped. Returns true if they may be forwarded, false if they should be
7494 * 'in_port' must be the ofport_dpif that corresponds to flow->in_port.
7495 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
7497 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
7498 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
7499 * checked by input_vid_is_valid().
7501 * May also add tags to '*tags', although the current implementation only does
7502 * so in one special case.
7505 is_admissible(struct xlate_ctx *ctx, struct ofport_dpif *in_port,
7508 struct ofproto_dpif *ofproto = ctx->ofproto;
7509 struct flow *flow = &ctx->xin->flow;
7510 struct ofbundle *in_bundle = in_port->bundle;
7512 /* Drop frames for reserved multicast addresses
7513 * only if forward_bpdu option is absent. */
7514 if (!ofproto->up.forward_bpdu && eth_addr_is_reserved(flow->dl_dst)) {
7515 xlate_report(ctx, "packet has reserved destination MAC, dropping");
7519 if (in_bundle->bond) {
7520 struct mac_entry *mac;
7522 switch (bond_check_admissibility(in_bundle->bond, in_port,
7523 flow->dl_dst, &ctx->xout->tags)) {
7528 xlate_report(ctx, "bonding refused admissibility, dropping");
7531 case BV_DROP_IF_MOVED:
7532 mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL);
7533 if (mac && mac->port.p != in_bundle &&
7534 (!is_gratuitous_arp(flow)
7535 || mac_entry_is_grat_arp_locked(mac))) {
7536 xlate_report(ctx, "SLB bond thinks this packet looped back, "
7548 xlate_normal(struct xlate_ctx *ctx)
7550 struct ofport_dpif *in_port;
7551 struct ofbundle *in_bundle;
7552 struct mac_entry *mac;
7556 ctx->xout->has_normal = true;
7558 in_bundle = lookup_input_bundle(ctx->ofproto, ctx->xin->flow.in_port,
7559 ctx->xin->packet != NULL, &in_port);
7561 xlate_report(ctx, "no input bundle, dropping");
7565 /* Drop malformed frames. */
7566 if (ctx->xin->flow.dl_type == htons(ETH_TYPE_VLAN) &&
7567 !(ctx->xin->flow.vlan_tci & htons(VLAN_CFI))) {
7568 if (ctx->xin->packet != NULL) {
7569 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7570 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
7571 "VLAN tag received on port %s",
7572 ctx->ofproto->up.name, in_bundle->name);
7574 xlate_report(ctx, "partial VLAN tag, dropping");
7578 /* Drop frames on bundles reserved for mirroring. */
7579 if (in_bundle->mirror_out) {
7580 if (ctx->xin->packet != NULL) {
7581 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7582 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
7583 "%s, which is reserved exclusively for mirroring",
7584 ctx->ofproto->up.name, in_bundle->name);
7586 xlate_report(ctx, "input port is mirror output port, dropping");
7591 vid = vlan_tci_to_vid(ctx->xin->flow.vlan_tci);
7592 if (!input_vid_is_valid(vid, in_bundle, ctx->xin->packet != NULL)) {
7593 xlate_report(ctx, "disallowed VLAN VID for this input port, dropping");
7596 vlan = input_vid_to_vlan(in_bundle, vid);
7598 /* Check other admissibility requirements. */
7599 if (in_port && !is_admissible(ctx, in_port, vlan)) {
7603 /* Learn source MAC. */
7604 if (ctx->xin->may_learn) {
7605 update_learning_table(ctx->ofproto, &ctx->xin->flow, vlan, in_bundle);
7608 /* Determine output bundle. */
7609 mac = mac_learning_lookup(ctx->ofproto->ml, ctx->xin->flow.dl_dst, vlan,
7612 if (mac->port.p != in_bundle) {
7613 xlate_report(ctx, "forwarding to learned port");
7614 output_normal(ctx, mac->port.p, vlan);
7616 xlate_report(ctx, "learned port is input port, dropping");
7619 struct ofbundle *bundle;
7621 xlate_report(ctx, "no learned MAC for destination, flooding");
7622 HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
7623 if (bundle != in_bundle
7624 && ofbundle_includes_vlan(bundle, vlan)
7625 && bundle->floodable
7626 && !bundle->mirror_out) {
7627 output_normal(ctx, bundle, vlan);
7630 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
7634 /* Optimized flow revalidation.
7636 * It's a difficult problem, in general, to tell which facets need to have
7637 * their actions recalculated whenever the OpenFlow flow table changes. We
7638 * don't try to solve that general problem: for most kinds of OpenFlow flow
7639 * table changes, we recalculate the actions for every facet. This is
7640 * relatively expensive, but it's good enough if the OpenFlow flow table
7641 * doesn't change very often.
7643 * However, we can expect one particular kind of OpenFlow flow table change to
7644 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
7645 * of CPU on revalidating every facet whenever MAC learning modifies the flow
7646 * table, we add a special case that applies to flow tables in which every rule
7647 * has the same form (that is, the same wildcards), except that the table is
7648 * also allowed to have a single "catch-all" flow that matches all packets. We
7649 * optimize this case by tagging all of the facets that resubmit into the table
7650 * and invalidating the same tag whenever a flow changes in that table. The
7651 * end result is that we revalidate just the facets that need it (and sometimes
7652 * a few more, but not all of the facets or even all of the facets that
7653 * resubmit to the table modified by MAC learning). */
7655 /* Calculates the tag to use for 'flow' and mask 'mask' when it is inserted
7656 * into an OpenFlow table with the given 'basis'. */
7658 rule_calculate_tag(const struct flow *flow, const struct minimask *mask,
7661 if (minimask_is_catchall(mask)) {
7664 uint32_t hash = flow_hash_in_minimask(flow, mask, secret);
7665 return tag_create_deterministic(hash);
7669 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
7670 * taggability of that table.
7672 * This function must be called after *each* change to a flow table. If you
7673 * skip calling it on some changes then the pointer comparisons at the end can
7674 * be invalid if you get unlucky. For example, if a flow removal causes a
7675 * cls_table to be destroyed and then a flow insertion causes a cls_table with
7676 * different wildcards to be created with the same address, then this function
7677 * will incorrectly skip revalidation. */
7679 table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
7681 struct table_dpif *table = &ofproto->tables[table_id];
7682 const struct oftable *oftable = &ofproto->up.tables[table_id];
7683 struct cls_table *catchall, *other;
7684 struct cls_table *t;
7686 catchall = other = NULL;
7688 switch (hmap_count(&oftable->cls.tables)) {
7690 /* We could tag this OpenFlow table but it would make the logic a
7691 * little harder and it's a corner case that doesn't seem worth it
7697 HMAP_FOR_EACH (t, hmap_node, &oftable->cls.tables) {
7698 if (cls_table_is_catchall(t)) {
7700 } else if (!other) {
7703 /* Indicate that we can't tag this by setting both tables to
7704 * NULL. (We know that 'catchall' is already NULL.) */
7711 /* Can't tag this table. */
7715 if (table->catchall_table != catchall || table->other_table != other) {
7716 table->catchall_table = catchall;
7717 table->other_table = other;
7718 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
7722 /* Given 'rule' that has changed in some way (either it is a rule being
7723 * inserted, a rule being deleted, or a rule whose actions are being
7724 * modified), marks facets for revalidation to ensure that packets will be
7725 * forwarded correctly according to the new state of the flow table.
7727 * This function must be called after *each* change to a flow table. See
7728 * the comment on table_update_taggable() for more information. */
7730 rule_invalidate(const struct rule_dpif *rule)
7732 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
7734 table_update_taggable(ofproto, rule->up.table_id);
7736 if (!ofproto->backer->need_revalidate) {
7737 struct table_dpif *table = &ofproto->tables[rule->up.table_id];
7739 if (table->other_table && rule->tag) {
7740 tag_set_add(&ofproto->backer->revalidate_set, rule->tag);
7742 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
7748 set_frag_handling(struct ofproto *ofproto_,
7749 enum ofp_config_flags frag_handling)
7751 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7752 if (frag_handling != OFPC_FRAG_REASM) {
7753 ofproto->backer->need_revalidate = REV_RECONFIGURE;
7761 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
7762 const struct flow *flow,
7763 const struct ofpact *ofpacts, size_t ofpacts_len)
7765 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7766 struct initial_vals initial_vals;
7767 struct odputil_keybuf keybuf;
7768 struct dpif_flow_stats stats;
7769 struct xlate_out xout;
7770 struct xlate_in xin;
7774 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
7775 odp_flow_key_from_flow(&key, flow,
7776 ofp_port_to_odp_port(ofproto, flow->in_port));
7778 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
7780 initial_vals.vlan_tci = flow->vlan_tci;
7781 initial_vals.tunnel_ip_tos = 0;
7782 xlate_in_init(&xin, ofproto, flow, &initial_vals, NULL, stats.tcp_flags,
7784 xin.resubmit_stats = &stats;
7785 xin.ofpacts_len = ofpacts_len;
7786 xin.ofpacts = ofpacts;
7788 xlate_actions(&xin, &xout);
7789 dpif_execute(ofproto->backer->dpif, key.data, key.size,
7790 xout.odp_actions.data, xout.odp_actions.size, packet);
7791 xlate_out_uninit(&xout);
7799 set_netflow(struct ofproto *ofproto_,
7800 const struct netflow_options *netflow_options)
7802 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7804 if (netflow_options) {
7805 if (!ofproto->netflow) {
7806 ofproto->netflow = netflow_create();
7808 return netflow_set_options(ofproto->netflow, netflow_options);
7810 netflow_destroy(ofproto->netflow);
7811 ofproto->netflow = NULL;
7817 get_netflow_ids(const struct ofproto *ofproto_,
7818 uint8_t *engine_type, uint8_t *engine_id)
7820 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7822 dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
7826 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
7828 if (!facet_is_controller_flow(facet) &&
7829 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
7830 struct subfacet *subfacet;
7831 struct ofexpired expired;
7833 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
7834 if (subfacet->path == SF_FAST_PATH) {
7835 struct dpif_flow_stats stats;
7837 subfacet_install(subfacet, &facet->xout.odp_actions, &stats);
7838 subfacet_update_stats(subfacet, &stats);
7842 expired.flow = facet->flow;
7843 expired.packet_count = facet->packet_count;
7844 expired.byte_count = facet->byte_count;
7845 expired.used = facet->used;
7846 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
7851 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
7853 struct facet *facet;
7855 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
7856 send_active_timeout(ofproto, facet);
7860 static struct ofproto_dpif *
7861 ofproto_dpif_lookup(const char *name)
7863 struct ofproto_dpif *ofproto;
7865 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
7866 hash_string(name, 0), &all_ofproto_dpifs) {
7867 if (!strcmp(ofproto->up.name, name)) {
7875 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
7876 const char *argv[], void *aux OVS_UNUSED)
7878 struct ofproto_dpif *ofproto;
7881 ofproto = ofproto_dpif_lookup(argv[1]);
7883 unixctl_command_reply_error(conn, "no such bridge");
7886 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
7888 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
7889 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
7893 unixctl_command_reply(conn, "table successfully flushed");
7897 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
7898 const char *argv[], void *aux OVS_UNUSED)
7900 struct ds ds = DS_EMPTY_INITIALIZER;
7901 const struct ofproto_dpif *ofproto;
7902 const struct mac_entry *e;
7904 ofproto = ofproto_dpif_lookup(argv[1]);
7906 unixctl_command_reply_error(conn, "no such bridge");
7910 ds_put_cstr(&ds, " port VLAN MAC Age\n");
7911 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
7912 struct ofbundle *bundle = e->port.p;
7913 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
7914 ofbundle_get_a_port(bundle)->odp_port,
7915 e->vlan, ETH_ADDR_ARGS(e->mac),
7916 mac_entry_age(ofproto->ml, e));
7918 unixctl_command_reply(conn, ds_cstr(&ds));
7923 struct xlate_out xout;
7924 struct xlate_in xin;
7930 trace_format_rule(struct ds *result, uint8_t table_id, int level,
7931 const struct rule_dpif *rule)
7933 ds_put_char_multiple(result, '\t', level);
7935 ds_put_cstr(result, "No match\n");
7939 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
7940 table_id, ntohll(rule->up.flow_cookie));
7941 cls_rule_format(&rule->up.cr, result);
7942 ds_put_char(result, '\n');
7944 ds_put_char_multiple(result, '\t', level);
7945 ds_put_cstr(result, "OpenFlow ");
7946 ofpacts_format(rule->up.ofpacts, rule->up.ofpacts_len, result);
7947 ds_put_char(result, '\n');
7951 trace_format_flow(struct ds *result, int level, const char *title,
7952 struct trace_ctx *trace)
7954 ds_put_char_multiple(result, '\t', level);
7955 ds_put_format(result, "%s: ", title);
7956 if (flow_equal(&trace->xin.flow, &trace->flow)) {
7957 ds_put_cstr(result, "unchanged");
7959 flow_format(result, &trace->xin.flow);
7960 trace->flow = trace->xin.flow;
7962 ds_put_char(result, '\n');
7966 trace_format_regs(struct ds *result, int level, const char *title,
7967 struct trace_ctx *trace)
7971 ds_put_char_multiple(result, '\t', level);
7972 ds_put_format(result, "%s:", title);
7973 for (i = 0; i < FLOW_N_REGS; i++) {
7974 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
7976 ds_put_char(result, '\n');
7980 trace_format_odp(struct ds *result, int level, const char *title,
7981 struct trace_ctx *trace)
7983 struct ofpbuf *odp_actions = &trace->xout.odp_actions;
7985 ds_put_char_multiple(result, '\t', level);
7986 ds_put_format(result, "%s: ", title);
7987 format_odp_actions(result, odp_actions->data, odp_actions->size);
7988 ds_put_char(result, '\n');
7992 trace_resubmit(struct xlate_ctx *ctx, struct rule_dpif *rule)
7994 struct trace_ctx *trace = CONTAINER_OF(ctx->xin, struct trace_ctx, xin);
7995 struct ds *result = trace->result;
7997 ds_put_char(result, '\n');
7998 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
7999 trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace);
8000 trace_format_odp(result, ctx->recurse + 1, "Resubmitted odp", trace);
8001 trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule);
8005 trace_report(struct xlate_ctx *ctx, const char *s)
8007 struct trace_ctx *trace = CONTAINER_OF(ctx->xin, struct trace_ctx, xin);
8008 struct ds *result = trace->result;
8010 ds_put_char_multiple(result, '\t', ctx->recurse);
8011 ds_put_cstr(result, s);
8012 ds_put_char(result, '\n');
8016 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
8017 void *aux OVS_UNUSED)
8019 const char *dpname = argv[1];
8020 struct ofproto_dpif *ofproto;
8021 struct ofpbuf odp_key;
8022 struct ofpbuf *packet;
8023 struct initial_vals initial_vals;
8029 ofpbuf_init(&odp_key, 0);
8032 ofproto = ofproto_dpif_lookup(dpname);
8034 unixctl_command_reply_error(conn, "Unknown ofproto (use ofproto/list "
8038 if (argc == 3 || (argc == 4 && !strcmp(argv[3], "-generate"))) {
8039 /* ofproto/trace dpname flow [-generate] */
8040 const char *flow_s = argv[2];
8041 const char *generate_s = argv[3];
8043 /* Allow 'flow_s' to be either a datapath flow or an OpenFlow-like
8044 * flow. We guess which type it is based on whether 'flow_s' contains
8045 * an '(', since a datapath flow always contains '(') but an
8046 * OpenFlow-like flow should not (in fact it's allowed but I believe
8047 * that's not documented anywhere).
8049 * An alternative would be to try to parse 'flow_s' both ways, but then
8050 * it would be tricky giving a sensible error message. After all, do
8051 * you just say "syntax error" or do you present both error messages?
8052 * Both choices seem lousy. */
8053 if (strchr(flow_s, '(')) {
8056 /* Convert string to datapath key. */
8057 ofpbuf_init(&odp_key, 0);
8058 error = odp_flow_key_from_string(flow_s, NULL, &odp_key);
8060 unixctl_command_reply_error(conn, "Bad flow syntax");
8064 /* The user might have specified the wrong ofproto but within the
8065 * same backer. That's OK, ofproto_receive() can find the right
8067 if (ofproto_receive(ofproto->backer, NULL, odp_key.data,
8068 odp_key.size, &flow, NULL, &ofproto, NULL,
8070 unixctl_command_reply_error(conn, "Invalid flow");
8073 ds_put_format(&result, "Bridge: %s\n", ofproto->up.name);
8077 error_s = parse_ofp_exact_flow(&flow, argv[2]);
8079 unixctl_command_reply_error(conn, error_s);
8084 initial_vals.vlan_tci = flow.vlan_tci;
8085 initial_vals.tunnel_ip_tos = flow.tunnel.ip_tos;
8088 /* Generate a packet, if requested. */
8090 packet = ofpbuf_new(0);
8091 flow_compose(packet, &flow);
8093 } else if (argc == 7) {
8094 /* ofproto/trace dpname priority tun_id in_port mark packet */
8095 const char *priority_s = argv[2];
8096 const char *tun_id_s = argv[3];
8097 const char *in_port_s = argv[4];
8098 const char *mark_s = argv[5];
8099 const char *packet_s = argv[6];
8100 uint32_t in_port = atoi(in_port_s);
8101 ovs_be64 tun_id = htonll(strtoull(tun_id_s, NULL, 0));
8102 uint32_t priority = atoi(priority_s);
8103 uint32_t mark = atoi(mark_s);
8106 msg = eth_from_hex(packet_s, &packet);
8108 unixctl_command_reply_error(conn, msg);
8112 ds_put_cstr(&result, "Packet: ");
8113 s = ofp_packet_to_string(packet->data, packet->size);
8114 ds_put_cstr(&result, s);
8117 flow_extract(packet, priority, mark, NULL, in_port, &flow);
8118 flow.tunnel.tun_id = tun_id;
8119 initial_vals.vlan_tci = flow.vlan_tci;
8120 initial_vals.tunnel_ip_tos = flow.tunnel.ip_tos;
8122 unixctl_command_reply_error(conn, "Bad command syntax");
8126 ofproto_trace(ofproto, &flow, packet, &initial_vals, &result);
8127 unixctl_command_reply(conn, ds_cstr(&result));
8130 ds_destroy(&result);
8131 ofpbuf_delete(packet);
8132 ofpbuf_uninit(&odp_key);
8136 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
8137 const struct ofpbuf *packet,
8138 const struct initial_vals *initial_vals, struct ds *ds)
8140 struct rule_dpif *rule;
8142 ds_put_cstr(ds, "Flow: ");
8143 flow_format(ds, flow);
8144 ds_put_char(ds, '\n');
8146 rule = rule_dpif_lookup(ofproto, flow);
8148 trace_format_rule(ds, 0, 0, rule);
8149 if (rule == ofproto->miss_rule) {
8150 ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
8151 } else if (rule == ofproto->no_packet_in_rule) {
8152 ds_put_cstr(ds, "\nNo match, packets dropped because "
8153 "OFPPC_NO_PACKET_IN is set on in_port.\n");
8154 } else if (rule == ofproto->drop_frags_rule) {
8155 ds_put_cstr(ds, "\nPackets dropped because they are IP fragments "
8156 "and the fragment handling mode is \"drop\".\n");
8160 uint64_t odp_actions_stub[1024 / 8];
8161 struct ofpbuf odp_actions;
8163 struct trace_ctx trace;
8166 tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
8169 ofpbuf_use_stub(&odp_actions,
8170 odp_actions_stub, sizeof odp_actions_stub);
8171 xlate_in_init(&trace.xin, ofproto, flow, initial_vals, rule, tcp_flags,
8173 trace.xin.resubmit_hook = trace_resubmit;
8174 trace.xin.report_hook = trace_report;
8175 xlate_actions(&trace.xin, &trace.xout);
8177 ds_put_char(ds, '\n');
8178 trace_format_flow(ds, 0, "Final flow", &trace);
8179 ds_put_cstr(ds, "Datapath actions: ");
8180 format_odp_actions(ds, trace.xout.odp_actions.data,
8181 trace.xout.odp_actions.size);
8183 if (trace.xout.slow) {
8184 ds_put_cstr(ds, "\nThis flow is handled by the userspace "
8185 "slow path because it:");
8186 switch (trace.xout.slow) {
8188 ds_put_cstr(ds, "\n\t- Consists of CFM packets.");
8191 ds_put_cstr(ds, "\n\t- Consists of LACP packets.");
8194 ds_put_cstr(ds, "\n\t- Consists of STP packets.");
8196 case SLOW_CONTROLLER:
8197 ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages "
8198 "to the OpenFlow controller.");
8205 xlate_out_uninit(&trace.xout);
8210 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
8211 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
8214 unixctl_command_reply(conn, NULL);
8218 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
8219 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
8222 unixctl_command_reply(conn, NULL);
8225 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
8226 * 'reply' describing the results. */
8228 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
8230 struct facet *facet;
8234 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
8235 if (!facet_check_consistency(facet)) {
8240 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
8244 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
8245 ofproto->up.name, errors);
8247 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
8252 ofproto_dpif_self_check(struct unixctl_conn *conn,
8253 int argc, const char *argv[], void *aux OVS_UNUSED)
8255 struct ds reply = DS_EMPTY_INITIALIZER;
8256 struct ofproto_dpif *ofproto;
8259 ofproto = ofproto_dpif_lookup(argv[1]);
8261 unixctl_command_reply_error(conn, "Unknown ofproto (use "
8262 "ofproto/list for help)");
8265 ofproto_dpif_self_check__(ofproto, &reply);
8267 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
8268 ofproto_dpif_self_check__(ofproto, &reply);
8272 unixctl_command_reply(conn, ds_cstr(&reply));
8276 /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list
8277 * of the 'ofproto_shash' nodes. It is the responsibility of the caller
8278 * to destroy 'ofproto_shash' and free the returned value. */
8279 static const struct shash_node **
8280 get_ofprotos(struct shash *ofproto_shash)
8282 const struct ofproto_dpif *ofproto;
8284 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
8285 char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
8286 shash_add_nocopy(ofproto_shash, name, ofproto);
8289 return shash_sort(ofproto_shash);
8293 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
8294 const char *argv[] OVS_UNUSED,
8295 void *aux OVS_UNUSED)
8297 struct ds ds = DS_EMPTY_INITIALIZER;
8298 struct shash ofproto_shash;
8299 const struct shash_node **sorted_ofprotos;
8302 shash_init(&ofproto_shash);
8303 sorted_ofprotos = get_ofprotos(&ofproto_shash);
8304 for (i = 0; i < shash_count(&ofproto_shash); i++) {
8305 const struct shash_node *node = sorted_ofprotos[i];
8306 ds_put_format(&ds, "%s\n", node->name);
8309 shash_destroy(&ofproto_shash);
8310 free(sorted_ofprotos);
8312 unixctl_command_reply(conn, ds_cstr(&ds));
8317 show_dp_format(const struct ofproto_dpif *ofproto, struct ds *ds)
8319 const struct shash_node **ports;
8321 struct avg_subfacet_rates lifetime;
8322 unsigned long long int minutes;
8323 const int min_ms = 60 * 1000; /* milliseconds in one minute. */
8325 minutes = (time_msec() - ofproto->created) / min_ms;
8328 lifetime.add_rate = (double)ofproto->total_subfacet_add_count
8330 lifetime.del_rate = (double)ofproto->total_subfacet_del_count
8333 lifetime.add_rate = 0.0;
8334 lifetime.del_rate = 0.0;
8337 ds_put_format(ds, "%s (%s):\n", ofproto->up.name,
8338 dpif_name(ofproto->backer->dpif));
8340 "\tlookups: hit:%"PRIu64" missed:%"PRIu64"\n",
8341 ofproto->n_hit, ofproto->n_missed);
8342 ds_put_format(ds, "\tflows: cur: %zu, avg: %5.3f, max: %d,"
8343 " life span: %llu(ms)\n",
8344 hmap_count(&ofproto->subfacets),
8345 avg_subfacet_count(ofproto),
8346 ofproto->max_n_subfacet,
8347 avg_subfacet_life_span(ofproto));
8348 if (minutes >= 60) {
8349 show_dp_rates(ds, "\t\thourly avg:", &ofproto->hourly);
8351 if (minutes >= 60 * 24) {
8352 show_dp_rates(ds, "\t\tdaily avg:", &ofproto->daily);
8354 show_dp_rates(ds, "\t\toverall avg:", &lifetime);
8356 ports = shash_sort(&ofproto->up.port_by_name);
8357 for (i = 0; i < shash_count(&ofproto->up.port_by_name); i++) {
8358 const struct shash_node *node = ports[i];
8359 struct ofport *ofport = node->data;
8360 const char *name = netdev_get_name(ofport->netdev);
8361 const char *type = netdev_get_type(ofport->netdev);
8364 ds_put_format(ds, "\t%s %u/", name, ofport->ofp_port);
8366 odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
8367 if (odp_port != OVSP_NONE) {
8368 ds_put_format(ds, "%"PRIu32":", odp_port);
8370 ds_put_cstr(ds, "none:");
8373 if (strcmp(type, "system")) {
8374 struct netdev *netdev;
8377 ds_put_format(ds, " (%s", type);
8379 error = netdev_open(name, type, &netdev);
8384 error = netdev_get_config(netdev, &config);
8386 const struct smap_node **nodes;
8389 nodes = smap_sort(&config);
8390 for (i = 0; i < smap_count(&config); i++) {
8391 const struct smap_node *node = nodes[i];
8392 ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
8393 node->key, node->value);
8397 smap_destroy(&config);
8399 netdev_close(netdev);
8401 ds_put_char(ds, ')');
8403 ds_put_char(ds, '\n');
8409 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc,
8410 const char *argv[], void *aux OVS_UNUSED)
8412 struct ds ds = DS_EMPTY_INITIALIZER;
8413 const struct ofproto_dpif *ofproto;
8417 for (i = 1; i < argc; i++) {
8418 ofproto = ofproto_dpif_lookup(argv[i]);
8420 ds_put_format(&ds, "Unknown bridge %s (use dpif/dump-dps "
8421 "for help)", argv[i]);
8422 unixctl_command_reply_error(conn, ds_cstr(&ds));
8425 show_dp_format(ofproto, &ds);
8428 struct shash ofproto_shash;
8429 const struct shash_node **sorted_ofprotos;
8432 shash_init(&ofproto_shash);
8433 sorted_ofprotos = get_ofprotos(&ofproto_shash);
8434 for (i = 0; i < shash_count(&ofproto_shash); i++) {
8435 const struct shash_node *node = sorted_ofprotos[i];
8436 show_dp_format(node->data, &ds);
8439 shash_destroy(&ofproto_shash);
8440 free(sorted_ofprotos);
8443 unixctl_command_reply(conn, ds_cstr(&ds));
8448 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
8449 int argc OVS_UNUSED, const char *argv[],
8450 void *aux OVS_UNUSED)
8452 struct ds ds = DS_EMPTY_INITIALIZER;
8453 const struct ofproto_dpif *ofproto;
8454 struct subfacet *subfacet;
8456 ofproto = ofproto_dpif_lookup(argv[1]);
8458 unixctl_command_reply_error(conn, "no such bridge");
8462 update_stats(ofproto->backer);
8464 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
8465 struct facet *facet = subfacet->facet;
8467 odp_flow_key_format(subfacet->key, subfacet->key_len, &ds);
8469 ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:",
8470 subfacet->dp_packet_count, subfacet->dp_byte_count);
8471 if (subfacet->used) {
8472 ds_put_format(&ds, "%.3fs",
8473 (time_msec() - subfacet->used) / 1000.0);
8475 ds_put_format(&ds, "never");
8477 if (subfacet->facet->tcp_flags) {
8478 ds_put_cstr(&ds, ", flags:");
8479 packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags);
8482 ds_put_cstr(&ds, ", actions:");
8483 if (facet->xout.slow) {
8484 uint64_t slow_path_stub[128 / 8];
8485 const struct nlattr *actions;
8488 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
8489 slow_path_stub, sizeof slow_path_stub,
8490 &actions, &actions_len);
8491 format_odp_actions(&ds, actions, actions_len);
8493 format_odp_actions(&ds, facet->xout.odp_actions.data,
8494 facet->xout.odp_actions.size);
8496 ds_put_char(&ds, '\n');
8499 unixctl_command_reply(conn, ds_cstr(&ds));
8504 ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn,
8505 int argc OVS_UNUSED, const char *argv[],
8506 void *aux OVS_UNUSED)
8508 struct ds ds = DS_EMPTY_INITIALIZER;
8509 struct ofproto_dpif *ofproto;
8511 ofproto = ofproto_dpif_lookup(argv[1]);
8513 unixctl_command_reply_error(conn, "no such bridge");
8517 flush(&ofproto->up);
8519 unixctl_command_reply(conn, ds_cstr(&ds));
8524 ofproto_dpif_unixctl_init(void)
8526 static bool registered;
8532 unixctl_command_register(
8534 "bridge {priority tun_id in_port mark packet | odp_flow [-generate]}",
8535 2, 6, ofproto_unixctl_trace, NULL);
8536 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
8537 ofproto_unixctl_fdb_flush, NULL);
8538 unixctl_command_register("fdb/show", "bridge", 1, 1,
8539 ofproto_unixctl_fdb_show, NULL);
8540 unixctl_command_register("ofproto/clog", "", 0, 0,
8541 ofproto_dpif_clog, NULL);
8542 unixctl_command_register("ofproto/unclog", "", 0, 0,
8543 ofproto_dpif_unclog, NULL);
8544 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
8545 ofproto_dpif_self_check, NULL);
8546 unixctl_command_register("dpif/dump-dps", "", 0, 0,
8547 ofproto_unixctl_dpif_dump_dps, NULL);
8548 unixctl_command_register("dpif/show", "[bridge]", 0, INT_MAX,
8549 ofproto_unixctl_dpif_show, NULL);
8550 unixctl_command_register("dpif/dump-flows", "bridge", 1, 1,
8551 ofproto_unixctl_dpif_dump_flows, NULL);
8552 unixctl_command_register("dpif/del-flows", "bridge", 1, 1,
8553 ofproto_unixctl_dpif_del_flows, NULL);
8556 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
8558 * This is deprecated. It is only for compatibility with broken device drivers
8559 * in old versions of Linux that do not properly support VLANs when VLAN
8560 * devices are not used. When broken device drivers are no longer in
8561 * widespread use, we will delete these interfaces. */
8564 set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid)
8566 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
8567 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
8569 if (realdev_ofp_port == ofport->realdev_ofp_port
8570 && vid == ofport->vlandev_vid) {
8574 ofproto->backer->need_revalidate = REV_RECONFIGURE;
8576 if (ofport->realdev_ofp_port) {
8579 if (realdev_ofp_port && ofport->bundle) {
8580 /* vlandevs are enslaved to their realdevs, so they are not allowed to
8581 * themselves be part of a bundle. */
8582 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
8585 ofport->realdev_ofp_port = realdev_ofp_port;
8586 ofport->vlandev_vid = vid;
8588 if (realdev_ofp_port) {
8589 vsp_add(ofport, realdev_ofp_port, vid);
8596 hash_realdev_vid(uint16_t realdev_ofp_port, int vid)
8598 return hash_2words(realdev_ofp_port, vid);
8601 /* Returns the OFP port number of the Linux VLAN device that corresponds to
8602 * 'vlan_tci' on the network device with port number 'realdev_ofp_port' in
8603 * 'struct ofport_dpif'. For example, given 'realdev_ofp_port' of eth0 and
8604 * 'vlan_tci' 9, it would return the port number of eth0.9.
8606 * Unless VLAN splinters are enabled for port 'realdev_ofp_port', this
8607 * function just returns its 'realdev_ofp_port' argument. */
8609 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
8610 uint16_t realdev_ofp_port, ovs_be16 vlan_tci)
8612 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
8613 int vid = vlan_tci_to_vid(vlan_tci);
8614 const struct vlan_splinter *vsp;
8616 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
8617 hash_realdev_vid(realdev_ofp_port, vid),
8618 &ofproto->realdev_vid_map) {
8619 if (vsp->realdev_ofp_port == realdev_ofp_port
8620 && vsp->vid == vid) {
8621 return vsp->vlandev_ofp_port;
8625 return realdev_ofp_port;
8628 static struct vlan_splinter *
8629 vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port)
8631 struct vlan_splinter *vsp;
8633 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, hash_int(vlandev_ofp_port, 0),
8634 &ofproto->vlandev_map) {
8635 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
8643 /* Returns the OpenFlow port number of the "real" device underlying the Linux
8644 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
8645 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
8646 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
8647 * eth0 and store 9 in '*vid'.
8649 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
8650 * VLAN device. Unless VLAN splinters are enabled, this is what this function
8653 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
8654 uint16_t vlandev_ofp_port, int *vid)
8656 if (!hmap_is_empty(&ofproto->vlandev_map)) {
8657 const struct vlan_splinter *vsp;
8659 vsp = vlandev_find(ofproto, vlandev_ofp_port);
8664 return vsp->realdev_ofp_port;
8670 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
8671 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
8672 * 'flow->in_port' to the "real" device backing the VLAN device, sets
8673 * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is
8674 * always the case unless VLAN splinters are enabled), returns false without
8675 * making any changes. */
8677 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
8682 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port, &vid);
8687 /* Cause the flow to be processed as if it came in on the real device with
8688 * the VLAN device's VLAN ID. */
8689 flow->in_port = realdev;
8690 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
8695 vsp_remove(struct ofport_dpif *port)
8697 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
8698 struct vlan_splinter *vsp;
8700 vsp = vlandev_find(ofproto, port->up.ofp_port);
8702 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
8703 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
8706 port->realdev_ofp_port = 0;
8708 VLOG_ERR("missing vlan device record");
8713 vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid)
8715 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
8717 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
8718 && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid))
8719 == realdev_ofp_port)) {
8720 struct vlan_splinter *vsp;
8722 vsp = xmalloc(sizeof *vsp);
8723 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
8724 hash_int(port->up.ofp_port, 0));
8725 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
8726 hash_realdev_vid(realdev_ofp_port, vid));
8727 vsp->realdev_ofp_port = realdev_ofp_port;
8728 vsp->vlandev_ofp_port = port->up.ofp_port;
8731 port->realdev_ofp_port = realdev_ofp_port;
8733 VLOG_ERR("duplicate vlan device record");
8738 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
8740 const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
8741 return ofport ? ofport->odp_port : OVSP_NONE;
8744 static struct ofport_dpif *
8745 odp_port_to_ofport(const struct dpif_backer *backer, uint32_t odp_port)
8747 struct ofport_dpif *port;
8749 HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node,
8750 hash_int(odp_port, 0),
8751 &backer->odp_to_ofport_map) {
8752 if (port->odp_port == odp_port) {
8761 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
8763 struct ofport_dpif *port;
8765 port = odp_port_to_ofport(ofproto->backer, odp_port);
8766 if (port && &ofproto->up == port->up.ofproto) {
8767 return port->up.ofp_port;
8772 static unsigned long long int
8773 avg_subfacet_life_span(const struct ofproto_dpif *ofproto)
8775 unsigned long long int dc;
8776 unsigned long long int avg;
8778 dc = ofproto->total_subfacet_del_count + ofproto->subfacet_del_count;
8779 avg = dc ? ofproto->total_subfacet_life_span / dc : 0;
8785 avg_subfacet_count(const struct ofproto_dpif *ofproto)
8789 if (ofproto->n_update_stats) {
8790 avg_c = (double)ofproto->total_subfacet_count
8791 / ofproto->n_update_stats;
8798 show_dp_rates(struct ds *ds, const char *heading,
8799 const struct avg_subfacet_rates *rates)
8801 ds_put_format(ds, "%s add rate: %5.3f/min, del rate: %5.3f/min\n",
8802 heading, rates->add_rate, rates->del_rate);
8806 update_max_subfacet_count(struct ofproto_dpif *ofproto)
8808 ofproto->max_n_subfacet = MAX(ofproto->max_n_subfacet,
8809 hmap_count(&ofproto->subfacets));
8812 /* Compute exponentially weighted moving average, adding 'new' as the newest,
8813 * most heavily weighted element. 'base' designates the rate of decay: after
8814 * 'base' further updates, 'new''s weight in the EWMA decays to about 1/e
8817 exp_mavg(double *avg, int base, double new)
8819 *avg = (*avg * (base - 1) + new) / base;
8823 update_moving_averages(struct ofproto_dpif *ofproto)
8825 const int min_ms = 60 * 1000; /* milliseconds in one minute. */
8827 /* Update hourly averages on the minute boundaries. */
8828 if (time_msec() - ofproto->last_minute >= min_ms) {
8829 exp_mavg(&ofproto->hourly.add_rate, 60, ofproto->subfacet_add_count);
8830 exp_mavg(&ofproto->hourly.del_rate, 60, ofproto->subfacet_del_count);
8832 /* Update daily averages on the hour boundaries. */
8833 if ((ofproto->last_minute - ofproto->created) / min_ms % 60 == 59) {
8834 exp_mavg(&ofproto->daily.add_rate, 24, ofproto->hourly.add_rate);
8835 exp_mavg(&ofproto->daily.del_rate, 24, ofproto->hourly.del_rate);
8838 ofproto->total_subfacet_add_count += ofproto->subfacet_add_count;
8839 ofproto->total_subfacet_del_count += ofproto->subfacet_del_count;
8840 ofproto->subfacet_add_count = 0;
8841 ofproto->subfacet_del_count = 0;
8842 ofproto->last_minute += min_ms;
8847 dpif_stats_update_hit_count(struct ofproto_dpif *ofproto, uint64_t delta)
8849 ofproto->n_hit += delta;
8852 const struct ofproto_class ofproto_dpif_class = {
8887 port_is_lacp_current,
8888 NULL, /* rule_choose_table */
8895 rule_modify_actions,
8907 get_stp_port_status,
8914 is_mirror_output_bundle,
8915 forward_bpdu_changed,
8916 set_mac_table_config,