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-sflow.h"
50 #include "poll-loop.h"
55 #include "unaligned.h"
57 #include "vlan-bitmap.h"
60 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
62 COVERAGE_DEFINE(ofproto_dpif_expired);
63 COVERAGE_DEFINE(ofproto_dpif_xlate);
64 COVERAGE_DEFINE(facet_changed_rule);
65 COVERAGE_DEFINE(facet_revalidate);
66 COVERAGE_DEFINE(facet_unexpected);
67 COVERAGE_DEFINE(facet_suppress);
69 /* Maximum depth of flow table recursion (due to resubmit actions) in a
70 * flow translation. */
71 #define MAX_RESUBMIT_RECURSION 64
73 /* Number of implemented OpenFlow tables. */
74 enum { N_TABLES = 255 };
75 enum { TBL_INTERNAL = N_TABLES - 1 }; /* Used for internal hidden rules. */
76 BUILD_ASSERT_DECL(N_TABLES >= 2 && N_TABLES <= 255);
88 * - Do include packets and bytes from facets that have been deleted or
89 * whose own statistics have been folded into the rule.
91 * - Do include packets and bytes sent "by hand" that were accounted to
92 * the rule without any facet being involved (this is a rare corner
93 * case in rule_execute()).
95 * - Do not include packet or bytes that can be obtained from any facet's
96 * packet_count or byte_count member or that can be obtained from the
97 * datapath by, e.g., dpif_flow_get() for any subfacet.
99 uint64_t packet_count; /* Number of packets received. */
100 uint64_t byte_count; /* Number of bytes received. */
102 tag_type tag; /* Caches rule_calculate_tag() result. */
104 struct list facets; /* List of "struct facet"s. */
107 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
109 return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
112 static struct rule_dpif *rule_dpif_lookup(struct ofproto_dpif *,
113 const struct flow *);
114 static struct rule_dpif *rule_dpif_lookup__(struct ofproto_dpif *,
117 static struct rule_dpif *rule_dpif_miss_rule(struct ofproto_dpif *ofproto,
118 const struct flow *flow);
120 static void rule_credit_stats(struct rule_dpif *,
121 const struct dpif_flow_stats *);
122 static void flow_push_stats(struct facet *, const struct dpif_flow_stats *);
123 static tag_type rule_calculate_tag(const struct flow *,
124 const struct minimask *, uint32_t basis);
125 static void rule_invalidate(const struct rule_dpif *);
127 #define MAX_MIRRORS 32
128 typedef uint32_t mirror_mask_t;
129 #define MIRROR_MASK_C(X) UINT32_C(X)
130 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
132 struct ofproto_dpif *ofproto; /* Owning ofproto. */
133 size_t idx; /* In ofproto's "mirrors" array. */
134 void *aux; /* Key supplied by ofproto's client. */
135 char *name; /* Identifier for log messages. */
137 /* Selection criteria. */
138 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
139 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
140 unsigned long *vlans; /* Bitmap of chosen VLANs, NULL selects all. */
142 /* Output (exactly one of out == NULL and out_vlan == -1 is true). */
143 struct ofbundle *out; /* Output port or NULL. */
144 int out_vlan; /* Output VLAN or -1. */
145 mirror_mask_t dup_mirrors; /* Bitmap of mirrors with the same output. */
148 int64_t packet_count; /* Number of packets sent. */
149 int64_t byte_count; /* Number of bytes sent. */
152 static void mirror_destroy(struct ofmirror *);
153 static void update_mirror_stats(struct ofproto_dpif *ofproto,
154 mirror_mask_t mirrors,
155 uint64_t packets, uint64_t bytes);
158 struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
159 struct ofproto_dpif *ofproto; /* Owning ofproto. */
160 void *aux; /* Key supplied by ofproto's client. */
161 char *name; /* Identifier for log messages. */
164 struct list ports; /* Contains "struct ofport"s. */
165 enum port_vlan_mode vlan_mode; /* VLAN mode */
166 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
167 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
168 * NULL if all VLANs are trunked. */
169 struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */
170 struct bond *bond; /* Nonnull iff more than one port. */
171 bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */
174 bool floodable; /* True if no port has OFPUTIL_PC_NO_FLOOD set. */
176 /* Port mirroring info. */
177 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
178 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
179 mirror_mask_t mirror_out; /* Mirrors that output to this bundle. */
182 static void bundle_remove(struct ofport *);
183 static void bundle_update(struct ofbundle *);
184 static void bundle_destroy(struct ofbundle *);
185 static void bundle_del_port(struct ofport_dpif *);
186 static void bundle_run(struct ofbundle *);
187 static void bundle_wait(struct ofbundle *);
188 static struct ofbundle *lookup_input_bundle(const struct ofproto_dpif *,
189 uint16_t in_port, bool warn,
190 struct ofport_dpif **in_ofportp);
192 /* A controller may use OFPP_NONE as the ingress port to indicate that
193 * it did not arrive on a "real" port. 'ofpp_none_bundle' exists for
194 * when an input bundle is needed for validation (e.g., mirroring or
195 * OFPP_NORMAL processing). It is not connected to an 'ofproto' or have
196 * any 'port' structs, so care must be taken when dealing with it. */
197 static struct ofbundle ofpp_none_bundle = {
199 .vlan_mode = PORT_VLAN_TRUNK
202 static void stp_run(struct ofproto_dpif *ofproto);
203 static void stp_wait(struct ofproto_dpif *ofproto);
204 static int set_stp_port(struct ofport *,
205 const struct ofproto_port_stp_settings *);
207 static bool ofbundle_includes_vlan(const struct ofbundle *, uint16_t vlan);
209 struct action_xlate_ctx {
210 /* action_xlate_ctx_init() initializes these members. */
213 struct ofproto_dpif *ofproto;
215 /* Flow to which the OpenFlow actions apply. xlate_actions() will modify
216 * this flow when actions change header fields. */
219 /* stack for the push and pop actions.
220 * Each stack element is of the type "union mf_subvalue". */
222 union mf_subvalue init_stack[1024 / sizeof(union mf_subvalue)];
224 /* The packet corresponding to 'flow', or a null pointer if we are
225 * revalidating without a packet to refer to. */
226 const struct ofpbuf *packet;
228 /* Should OFPP_NORMAL update the MAC learning table? Should "learn"
229 * actions update the flow table?
231 * We want to update these tables if we are actually processing a packet,
232 * or if we are accounting for packets that the datapath has processed, but
233 * not if we are just revalidating. */
236 /* The rule that we are currently translating, or NULL. */
237 struct rule_dpif *rule;
239 /* Union of the set of TCP flags seen so far in this flow. (Used only by
240 * NXAST_FIN_TIMEOUT. Set to zero to avoid updating updating rules'
244 /* If nonnull, flow translation calls this function just before executing a
245 * resubmit or OFPP_TABLE action. In addition, disables logging of traces
246 * when the recursion depth is exceeded.
248 * 'rule' is the rule being submitted into. It will be null if the
249 * resubmit or OFPP_TABLE action didn't find a matching rule.
251 * This is normally null so the client has to set it manually after
252 * calling action_xlate_ctx_init(). */
253 void (*resubmit_hook)(struct action_xlate_ctx *, struct rule_dpif *rule);
255 /* If nonnull, flow translation calls this function to report some
256 * significant decision, e.g. to explain why OFPP_NORMAL translation
257 * dropped a packet. */
258 void (*report_hook)(struct action_xlate_ctx *, const char *s);
260 /* If nonnull, flow translation credits the specified statistics to each
261 * rule reached through a resubmit or OFPP_TABLE action.
263 * This is normally null so the client has to set it manually after
264 * calling action_xlate_ctx_init(). */
265 const struct dpif_flow_stats *resubmit_stats;
267 /* xlate_actions() initializes and uses these members. The client might want
268 * to look at them after it returns. */
270 struct ofpbuf *odp_actions; /* Datapath actions. */
271 tag_type tags; /* Tags associated with actions. */
272 enum slow_path_reason slow; /* 0 if fast path may be used. */
273 bool has_learn; /* Actions include NXAST_LEARN? */
274 bool has_normal; /* Actions output to OFPP_NORMAL? */
275 bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */
276 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
277 mirror_mask_t mirrors; /* Bitmap of associated mirrors. */
279 /* xlate_actions() initializes and uses these members, but the client has no
280 * reason to look at them. */
282 int recurse; /* Recursion level, via xlate_table_action. */
283 bool max_resubmit_trigger; /* Recursed too deeply during translation. */
284 struct flow base_flow; /* Flow at the last commit. */
285 uint32_t orig_skb_priority; /* Priority when packet arrived. */
286 uint8_t table_id; /* OpenFlow table ID where flow was found. */
287 uint32_t sflow_n_outputs; /* Number of output ports. */
288 uint32_t sflow_odp_port; /* Output port for composing sFlow action. */
289 uint16_t user_cookie_offset;/* Used for user_action_cookie fixup. */
290 bool exit; /* No further actions should be processed. */
293 /* Initial values of fields of the packet that may be changed during
294 * flow processing and needed later. */
295 struct initial_vals {
296 /* This is the value of vlan_tci in the packet as actually received from
297 * dpif. This is the same as the facet's flow.vlan_tci unless the packet
298 * was received via a VLAN splinter. In that case, this value is 0
299 * (because the packet as actually received from the dpif had no 802.1Q
300 * tag) but the facet's flow.vlan_tci is set to the VLAN that the splinter
303 * This member should be removed when the VLAN splinters feature is no
307 /* If received on a tunnel, the IP TOS value of the tunnel. */
308 uint8_t tunnel_ip_tos;
311 static void action_xlate_ctx_init(struct action_xlate_ctx *,
312 struct ofproto_dpif *, const struct flow *,
313 const struct initial_vals *initial_vals,
315 uint8_t tcp_flags, const struct ofpbuf *);
316 static void xlate_actions(struct action_xlate_ctx *,
317 const struct ofpact *ofpacts, size_t ofpacts_len,
318 struct ofpbuf *odp_actions);
319 static void xlate_actions_for_side_effects(struct action_xlate_ctx *,
320 const struct ofpact *ofpacts,
322 static void xlate_table_action(struct action_xlate_ctx *, uint16_t in_port,
323 uint8_t table_id, bool may_packet_in);
325 static size_t put_userspace_action(const struct ofproto_dpif *,
326 struct ofpbuf *odp_actions,
328 const union user_action_cookie *);
330 static void compose_slow_path(const struct ofproto_dpif *, const struct flow *,
331 enum slow_path_reason,
332 uint64_t *stub, size_t stub_size,
333 const struct nlattr **actionsp,
334 size_t *actions_lenp);
336 static void xlate_report(struct action_xlate_ctx *ctx, const char *s);
338 /* A subfacet (see "struct subfacet" below) has three possible installation
341 * - SF_NOT_INSTALLED: Not installed in the datapath. This will only be the
342 * case just after the subfacet is created, just before the subfacet is
343 * destroyed, or if the datapath returns an error when we try to install a
346 * - SF_FAST_PATH: The subfacet's actions are installed in the datapath.
348 * - SF_SLOW_PATH: An action that sends every packet for the subfacet through
349 * ofproto_dpif is installed in the datapath.
352 SF_NOT_INSTALLED, /* No datapath flow for this subfacet. */
353 SF_FAST_PATH, /* Full actions are installed. */
354 SF_SLOW_PATH, /* Send-to-userspace action is installed. */
357 static const char *subfacet_path_to_string(enum subfacet_path);
359 /* A dpif flow and actions associated with a facet.
361 * See also the large comment on struct facet. */
364 struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */
365 struct list list_node; /* In struct facet's 'facets' list. */
366 struct facet *facet; /* Owning facet. */
368 enum odp_key_fitness key_fitness;
372 long long int used; /* Time last used; time created if not used. */
373 long long int created; /* Time created. */
375 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
376 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
380 * These should be essentially identical for every subfacet in a facet, but
381 * may differ in trivial ways due to VLAN splinters. */
382 size_t actions_len; /* Number of bytes in actions[]. */
383 struct nlattr *actions; /* Datapath actions. */
385 enum slow_path_reason slow; /* 0 if fast path may be used. */
386 enum subfacet_path path; /* Installed in datapath? */
388 /* Initial values of the packet that may be needed later. */
389 struct initial_vals initial_vals;
391 /* Datapath port the packet arrived on. This is needed to remove
392 * flows for ports that are no longer part of the bridge. Since the
393 * flow definition only has the OpenFlow port number and the port is
394 * no longer part of the bridge, we can't determine the datapath port
395 * number needed to delete the flow from the datapath. */
396 uint32_t odp_in_port;
399 #define SUBFACET_DESTROY_MAX_BATCH 50
401 static struct subfacet *subfacet_create(struct facet *, struct flow_miss *miss,
403 static struct subfacet *subfacet_find(struct ofproto_dpif *,
404 const struct nlattr *key, size_t key_len,
406 static void subfacet_destroy(struct subfacet *);
407 static void subfacet_destroy__(struct subfacet *);
408 static void subfacet_destroy_batch(struct ofproto_dpif *,
409 struct subfacet **, int n);
410 static void subfacet_reset_dp_stats(struct subfacet *,
411 struct dpif_flow_stats *);
412 static void subfacet_update_time(struct subfacet *, long long int used);
413 static void subfacet_update_stats(struct subfacet *,
414 const struct dpif_flow_stats *);
415 static void subfacet_make_actions(struct subfacet *,
416 const struct ofpbuf *packet,
417 struct ofpbuf *odp_actions);
418 static int subfacet_install(struct subfacet *,
419 const struct nlattr *actions, size_t actions_len,
420 struct dpif_flow_stats *, enum slow_path_reason);
421 static void subfacet_uninstall(struct subfacet *);
423 static enum subfacet_path subfacet_want_path(enum slow_path_reason);
425 /* An exact-match instantiation of an OpenFlow flow.
427 * A facet associates a "struct flow", which represents the Open vSwitch
428 * userspace idea of an exact-match flow, with one or more subfacets. Each
429 * subfacet tracks the datapath's idea of the exact-match flow equivalent to
430 * the facet. When the kernel module (or other dpif implementation) and Open
431 * vSwitch userspace agree on the definition of a flow key, there is exactly
432 * one subfacet per facet. If the dpif implementation supports more-specific
433 * flow matching than userspace, however, a facet can have more than one
434 * subfacet, each of which corresponds to some distinction in flow that
435 * userspace simply doesn't understand.
437 * Flow expiration works in terms of subfacets, so a facet must have at least
438 * one subfacet or it will never expire, leaking memory. */
441 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
442 struct list list_node; /* In owning rule's 'facets' list. */
443 struct rule_dpif *rule; /* Owning rule. */
446 struct list subfacets;
447 long long int used; /* Time last used; time created if not used. */
454 * - Do include packets and bytes sent "by hand", e.g. with
457 * - Do include packets and bytes that were obtained from the datapath
458 * when a subfacet's statistics were reset (e.g. dpif_flow_put() with
459 * DPIF_FP_ZERO_STATS).
461 * - Do not include packets or bytes that can be obtained from the
462 * datapath for any existing subfacet.
464 uint64_t packet_count; /* Number of packets received. */
465 uint64_t byte_count; /* Number of bytes received. */
467 /* Resubmit statistics. */
468 uint64_t prev_packet_count; /* Number of packets from last stats push. */
469 uint64_t prev_byte_count; /* Number of bytes from last stats push. */
470 long long int prev_used; /* Used time from last stats push. */
473 uint64_t accounted_bytes; /* Bytes processed by facet_account(). */
474 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
475 uint8_t tcp_flags; /* TCP flags seen for this 'rule'. */
477 /* Properties of datapath actions.
479 * Every subfacet has its own actions because actions can differ slightly
480 * between splintered and non-splintered subfacets due to the VLAN tag
481 * being initially different (present vs. absent). All of them have these
482 * properties in common so we just store one copy of them here. */
483 bool has_learn; /* Actions include NXAST_LEARN? */
484 bool has_normal; /* Actions output to OFPP_NORMAL? */
485 bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */
486 tag_type tags; /* Tags that would require revalidation. */
487 mirror_mask_t mirrors; /* Bitmap of dependent mirrors. */
489 /* Storage for a single subfacet, to reduce malloc() time and space
490 * overhead. (A facet always has at least one subfacet and in the common
491 * case has exactly one subfacet. However, 'one_subfacet' may not
492 * always be valid, since it could have been removed after newer
493 * subfacets were pushed onto the 'subfacets' list.) */
494 struct subfacet one_subfacet;
496 long long int learn_rl; /* Rate limiter for facet_learn(). */
499 static struct facet *facet_create(struct rule_dpif *,
500 const struct flow *, uint32_t hash);
501 static void facet_remove(struct facet *);
502 static void facet_free(struct facet *);
504 static struct facet *facet_find(struct ofproto_dpif *,
505 const struct flow *, uint32_t hash);
506 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
507 const struct flow *, uint32_t hash);
508 static void facet_revalidate(struct facet *);
509 static bool facet_check_consistency(struct facet *);
511 static void facet_flush_stats(struct facet *);
513 static void facet_update_time(struct facet *, long long int used);
514 static void facet_reset_counters(struct facet *);
515 static void facet_push_stats(struct facet *);
516 static void facet_learn(struct facet *);
517 static void facet_account(struct facet *);
518 static void push_all_stats(void);
520 static struct subfacet *facet_get_subfacet(struct facet *);
522 static bool facet_is_controller_flow(struct facet *);
525 struct hmap_node odp_port_node; /* In dpif_backer's "odp_to_ofport_map". */
529 struct ofbundle *bundle; /* Bundle that contains this port, if any. */
530 struct list bundle_node; /* In struct ofbundle's "ports" list. */
531 struct cfm *cfm; /* Connectivity Fault Management, if any. */
532 tag_type tag; /* Tag associated with this port. */
533 bool may_enable; /* May be enabled in bonds. */
534 long long int carrier_seq; /* Carrier status changes. */
535 struct tnl_port *tnl_port; /* Tunnel handle, or null. */
538 struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */
539 enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */
540 long long int stp_state_entered;
542 struct hmap priorities; /* Map of attached 'priority_to_dscp's. */
544 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
546 * This is deprecated. It is only for compatibility with broken device
547 * drivers in old versions of Linux that do not properly support VLANs when
548 * VLAN devices are not used. When broken device drivers are no longer in
549 * widespread use, we will delete these interfaces. */
550 uint16_t realdev_ofp_port;
554 /* Node in 'ofport_dpif''s 'priorities' map. Used to maintain a map from
555 * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
556 * traffic egressing the 'ofport' with that priority should be marked with. */
557 struct priority_to_dscp {
558 struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'priorities' map. */
559 uint32_t priority; /* Priority of this queue (see struct flow). */
561 uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */
564 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
566 * This is deprecated. It is only for compatibility with broken device drivers
567 * in old versions of Linux that do not properly support VLANs when VLAN
568 * devices are not used. When broken device drivers are no longer in
569 * widespread use, we will delete these interfaces. */
570 struct vlan_splinter {
571 struct hmap_node realdev_vid_node;
572 struct hmap_node vlandev_node;
573 uint16_t realdev_ofp_port;
574 uint16_t vlandev_ofp_port;
578 static uint32_t vsp_realdev_to_vlandev(const struct ofproto_dpif *,
579 uint32_t realdev, ovs_be16 vlan_tci);
580 static bool vsp_adjust_flow(const struct ofproto_dpif *, struct flow *);
581 static void vsp_remove(struct ofport_dpif *);
582 static void vsp_add(struct ofport_dpif *, uint16_t realdev_ofp_port, int vid);
584 static uint32_t ofp_port_to_odp_port(const struct ofproto_dpif *,
586 static uint16_t odp_port_to_ofp_port(const struct ofproto_dpif *,
589 static struct ofport_dpif *
590 ofport_dpif_cast(const struct ofport *ofport)
592 ovs_assert(ofport->ofproto->ofproto_class == &ofproto_dpif_class);
593 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
596 static void port_run(struct ofport_dpif *);
597 static void port_run_fast(struct ofport_dpif *);
598 static void port_wait(struct ofport_dpif *);
599 static int set_cfm(struct ofport *, const struct cfm_settings *);
600 static void ofport_clear_priorities(struct ofport_dpif *);
602 struct dpif_completion {
603 struct list list_node;
604 struct ofoperation *op;
607 /* Extra information about a classifier table.
608 * Currently used just for optimized flow revalidation. */
610 /* If either of these is nonnull, then this table has a form that allows
611 * flows to be tagged to avoid revalidating most flows for the most common
612 * kinds of flow table changes. */
613 struct cls_table *catchall_table; /* Table that wildcards all fields. */
614 struct cls_table *other_table; /* Table with any other wildcard set. */
615 uint32_t basis; /* Keeps each table's tags separate. */
618 /* Reasons that we might need to revalidate every facet, and corresponding
621 * A value of 0 means that there is no need to revalidate.
623 * It would be nice to have some cleaner way to integrate with coverage
624 * counters, but with only a few reasons I guess this is good enough for
626 enum revalidate_reason {
627 REV_RECONFIGURE = 1, /* Switch configuration changed. */
628 REV_STP, /* Spanning tree protocol port status change. */
629 REV_PORT_TOGGLED, /* Port enabled or disabled by CFM, LACP, ...*/
630 REV_FLOW_TABLE, /* Flow table changed. */
631 REV_INCONSISTENCY /* Facet self-check failed. */
633 COVERAGE_DEFINE(rev_reconfigure);
634 COVERAGE_DEFINE(rev_stp);
635 COVERAGE_DEFINE(rev_port_toggled);
636 COVERAGE_DEFINE(rev_flow_table);
637 COVERAGE_DEFINE(rev_inconsistency);
639 /* Drop keys are odp flow keys which have drop flows installed in the kernel.
640 * These are datapath flows which have no associated ofproto, if they did we
641 * would use facets. */
643 struct hmap_node hmap_node;
648 /* All datapaths of a given type share a single dpif backer instance. */
653 struct timer next_expiration;
654 struct hmap odp_to_ofport_map; /* ODP port to ofport mapping. */
656 struct simap tnl_backers; /* Set of dpif ports backing tunnels. */
658 /* Facet revalidation flags applying to facets which use this backer. */
659 enum revalidate_reason need_revalidate; /* Revalidate every facet. */
660 struct tag_set revalidate_set; /* Revalidate only matching facets. */
662 struct hmap drop_keys; /* Set of dropped odp keys. */
665 /* All existing ofproto_backer instances, indexed by ofproto->up.type. */
666 static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers);
668 static void drop_key_clear(struct dpif_backer *);
669 static struct ofport_dpif *
670 odp_port_to_ofport(const struct dpif_backer *, uint32_t odp_port);
672 static void dpif_stats_update_hit_count(struct ofproto_dpif *ofproto,
674 struct avg_subfacet_rates {
675 double add_rate; /* Moving average of new flows created per minute. */
676 double del_rate; /* Moving average of flows deleted per minute. */
678 static void show_dp_rates(struct ds *ds, const char *heading,
679 const struct avg_subfacet_rates *rates);
680 static void exp_mavg(double *avg, int base, double new);
682 struct ofproto_dpif {
683 struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
685 struct dpif_backer *backer;
687 /* Special OpenFlow rules. */
688 struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */
689 struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */
695 struct netflow *netflow;
696 struct dpif_sflow *sflow;
697 struct hmap bundles; /* Contains "struct ofbundle"s. */
698 struct mac_learning *ml;
699 struct ofmirror *mirrors[MAX_MIRRORS];
701 bool has_bonded_bundles;
705 struct hmap subfacets;
706 struct governor *governor;
707 long long int consistency_rl;
710 struct table_dpif tables[N_TABLES];
712 /* Support for debugging async flow mods. */
713 struct list completions;
715 bool has_bundle_action; /* True when the first bundle action appears. */
716 struct netdev_stats stats; /* To account packets generated and consumed in
721 long long int stp_last_tick;
723 /* VLAN splinters. */
724 struct hmap realdev_vid_map; /* (realdev,vid) -> vlandev. */
725 struct hmap vlandev_map; /* vlandev -> (realdev,vid). */
728 struct sset ports; /* Set of standard port names. */
729 struct sset ghost_ports; /* Ports with no datapath port. */
730 struct sset port_poll_set; /* Queued names for port_poll() reply. */
731 int port_poll_errno; /* Last errno for port_poll() reply. */
733 /* Per ofproto's dpif stats. */
737 /* Subfacet statistics.
739 * These keep track of the total number of subfacets added and deleted and
740 * flow life span. They are useful for computing the flow rates stats
741 * exposed via "ovs-appctl dpif/show". The goal is to learn about
742 * traffic patterns in ways that we can use later to improve Open vSwitch
743 * performance in new situations. */
744 long long int created; /* Time when it is created. */
745 unsigned int max_n_subfacet; /* Maximum number of flows */
747 /* The average number of subfacets... */
748 struct avg_subfacet_rates hourly; /* ...over the last hour. */
749 struct avg_subfacet_rates daily; /* ...over the last day. */
750 long long int last_minute; /* Last time 'hourly' was updated. */
752 /* Number of subfacets added or deleted since 'last_minute'. */
753 unsigned int subfacet_add_count;
754 unsigned int subfacet_del_count;
756 /* Number of subfacets added or deleted from 'created' to 'last_minute.' */
757 unsigned long long int total_subfacet_add_count;
758 unsigned long long int total_subfacet_del_count;
760 /* Sum of the number of milliseconds that each subfacet existed,
761 * over the subfacets that have been added and then later deleted. */
762 unsigned long long int total_subfacet_life_span;
764 /* Incremented by the number of currently existing subfacets, each
765 * time we pull statistics from the kernel. */
766 unsigned long long int total_subfacet_count;
768 /* Number of times we pull statistics from the kernel. */
769 unsigned long long int n_update_stats;
771 static unsigned long long int avg_subfacet_life_span(
772 const struct ofproto_dpif *);
773 static double avg_subfacet_count(const struct ofproto_dpif *ofproto);
774 static void update_moving_averages(struct ofproto_dpif *ofproto);
775 static void dpif_stats_update_hit_count(struct ofproto_dpif *ofproto,
777 static void update_max_subfacet_count(struct ofproto_dpif *ofproto);
779 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
780 * for debugging the asynchronous flow_mod implementation.) */
783 /* All existing ofproto_dpif instances, indexed by ->up.name. */
784 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
786 static void ofproto_dpif_unixctl_init(void);
788 static struct ofproto_dpif *
789 ofproto_dpif_cast(const struct ofproto *ofproto)
791 ovs_assert(ofproto->ofproto_class == &ofproto_dpif_class);
792 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
795 static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *,
797 static struct ofport_dpif *get_odp_port(const struct ofproto_dpif *,
799 static void ofproto_trace(struct ofproto_dpif *, const struct flow *,
800 const struct ofpbuf *,
801 const struct initial_vals *, struct ds *);
803 /* Packet processing. */
804 static void update_learning_table(struct ofproto_dpif *,
805 const struct flow *, int vlan,
808 #define FLOW_MISS_MAX_BATCH 50
809 static int handle_upcalls(struct dpif_backer *, unsigned int max_batch);
811 /* Flow expiration. */
812 static int expire(struct dpif_backer *);
815 static void send_netflow_active_timeouts(struct ofproto_dpif *);
818 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
819 static size_t compose_sflow_action(const struct ofproto_dpif *,
820 struct ofpbuf *odp_actions,
821 const struct flow *, uint32_t odp_port);
822 static void add_mirror_actions(struct action_xlate_ctx *ctx,
823 const struct flow *flow);
824 /* Global variables. */
825 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
827 /* Initial mappings of port to bridge mappings. */
828 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
830 /* Factory functions. */
833 init(const struct shash *iface_hints)
835 struct shash_node *node;
837 /* Make a local copy, since we don't own 'iface_hints' elements. */
838 SHASH_FOR_EACH(node, iface_hints) {
839 const struct iface_hint *orig_hint = node->data;
840 struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
842 new_hint->br_name = xstrdup(orig_hint->br_name);
843 new_hint->br_type = xstrdup(orig_hint->br_type);
844 new_hint->ofp_port = orig_hint->ofp_port;
846 shash_add(&init_ofp_ports, node->name, new_hint);
851 enumerate_types(struct sset *types)
853 dp_enumerate_types(types);
857 enumerate_names(const char *type, struct sset *names)
859 struct ofproto_dpif *ofproto;
862 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
863 if (strcmp(type, ofproto->up.type)) {
866 sset_add(names, ofproto->up.name);
873 del(const char *type, const char *name)
878 error = dpif_open(name, type, &dpif);
880 error = dpif_delete(dpif);
887 port_open_type(const char *datapath_type, const char *port_type)
889 return dpif_port_open_type(datapath_type, port_type);
892 /* Type functions. */
894 static struct ofproto_dpif *
895 lookup_ofproto_dpif_by_port_name(const char *name)
897 struct ofproto_dpif *ofproto;
899 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
900 if (sset_contains(&ofproto->ports, name)) {
909 type_run(const char *type)
911 struct dpif_backer *backer;
915 backer = shash_find_data(&all_dpif_backers, type);
917 /* This is not necessarily a problem, since backers are only
918 * created on demand. */
922 dpif_run(backer->dpif);
924 if (backer->need_revalidate
925 || !tag_set_is_empty(&backer->revalidate_set)) {
926 struct tag_set revalidate_set = backer->revalidate_set;
927 bool need_revalidate = backer->need_revalidate;
928 struct ofproto_dpif *ofproto;
929 struct simap_node *node;
930 struct simap tmp_backers;
932 /* Handle tunnel garbage collection. */
933 simap_init(&tmp_backers);
934 simap_swap(&backer->tnl_backers, &tmp_backers);
936 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
937 struct ofport_dpif *iter;
939 if (backer != ofproto->backer) {
943 HMAP_FOR_EACH (iter, up.hmap_node, &ofproto->up.ports) {
946 if (!iter->tnl_port) {
950 dp_port = netdev_vport_get_dpif_port(iter->up.netdev);
951 node = simap_find(&tmp_backers, dp_port);
953 simap_put(&backer->tnl_backers, dp_port, node->data);
954 simap_delete(&tmp_backers, node);
955 node = simap_find(&backer->tnl_backers, dp_port);
957 node = simap_find(&backer->tnl_backers, dp_port);
959 uint32_t odp_port = UINT32_MAX;
961 if (!dpif_port_add(backer->dpif, iter->up.netdev,
963 simap_put(&backer->tnl_backers, dp_port, odp_port);
964 node = simap_find(&backer->tnl_backers, dp_port);
969 iter->odp_port = node ? node->data : OVSP_NONE;
970 if (tnl_port_reconfigure(&iter->up, iter->odp_port,
972 backer->need_revalidate = REV_RECONFIGURE;
977 SIMAP_FOR_EACH (node, &tmp_backers) {
978 dpif_port_del(backer->dpif, node->data);
980 simap_destroy(&tmp_backers);
982 switch (backer->need_revalidate) {
983 case REV_RECONFIGURE: COVERAGE_INC(rev_reconfigure); break;
984 case REV_STP: COVERAGE_INC(rev_stp); break;
985 case REV_PORT_TOGGLED: COVERAGE_INC(rev_port_toggled); break;
986 case REV_FLOW_TABLE: COVERAGE_INC(rev_flow_table); break;
987 case REV_INCONSISTENCY: COVERAGE_INC(rev_inconsistency); break;
990 if (backer->need_revalidate) {
991 /* Clear the drop_keys in case we should now be accepting some
992 * formerly dropped flows. */
993 drop_key_clear(backer);
996 /* Clear the revalidation flags. */
997 tag_set_init(&backer->revalidate_set);
998 backer->need_revalidate = 0;
1000 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1001 struct facet *facet, *next;
1003 if (ofproto->backer != backer) {
1007 HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &ofproto->facets) {
1009 || tag_set_intersects(&revalidate_set, facet->tags)) {
1010 facet_revalidate(facet);
1016 if (timer_expired(&backer->next_expiration)) {
1017 int delay = expire(backer);
1018 timer_set_duration(&backer->next_expiration, delay);
1021 /* Check for port changes in the dpif. */
1022 while ((error = dpif_port_poll(backer->dpif, &devname)) == 0) {
1023 struct ofproto_dpif *ofproto;
1024 struct dpif_port port;
1026 /* Don't report on the datapath's device. */
1027 if (!strcmp(devname, dpif_base_name(backer->dpif))) {
1031 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
1032 &all_ofproto_dpifs) {
1033 if (simap_contains(&ofproto->backer->tnl_backers, devname)) {
1038 ofproto = lookup_ofproto_dpif_by_port_name(devname);
1039 if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
1040 /* The port was removed. If we know the datapath,
1041 * report it through poll_set(). If we don't, it may be
1042 * notifying us of a removal we initiated, so ignore it.
1043 * If there's a pending ENOBUFS, let it stand, since
1044 * everything will be reevaluated. */
1045 if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
1046 sset_add(&ofproto->port_poll_set, devname);
1047 ofproto->port_poll_errno = 0;
1049 } else if (!ofproto) {
1050 /* The port was added, but we don't know with which
1051 * ofproto we should associate it. Delete it. */
1052 dpif_port_del(backer->dpif, port.port_no);
1054 dpif_port_destroy(&port);
1060 if (error != EAGAIN) {
1061 struct ofproto_dpif *ofproto;
1063 /* There was some sort of error, so propagate it to all
1064 * ofprotos that use this backer. */
1065 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
1066 &all_ofproto_dpifs) {
1067 if (ofproto->backer == backer) {
1068 sset_clear(&ofproto->port_poll_set);
1069 ofproto->port_poll_errno = error;
1078 type_run_fast(const char *type)
1080 struct dpif_backer *backer;
1083 backer = shash_find_data(&all_dpif_backers, type);
1085 /* This is not necessarily a problem, since backers are only
1086 * created on demand. */
1090 /* Handle one or more batches of upcalls, until there's nothing left to do
1091 * or until we do a fixed total amount of work.
1093 * We do work in batches because it can be much cheaper to set up a number
1094 * of flows and fire off their patches all at once. We do multiple batches
1095 * because in some cases handling a packet can cause another packet to be
1096 * queued almost immediately as part of the return flow. Both
1097 * optimizations can make major improvements on some benchmarks and
1098 * presumably for real traffic as well. */
1100 while (work < FLOW_MISS_MAX_BATCH) {
1101 int retval = handle_upcalls(backer, FLOW_MISS_MAX_BATCH - work);
1112 type_wait(const char *type)
1114 struct dpif_backer *backer;
1116 backer = shash_find_data(&all_dpif_backers, type);
1118 /* This is not necessarily a problem, since backers are only
1119 * created on demand. */
1123 timer_wait(&backer->next_expiration);
1126 /* Basic life-cycle. */
1128 static int add_internal_flows(struct ofproto_dpif *);
1130 static struct ofproto *
1133 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
1134 return &ofproto->up;
1138 dealloc(struct ofproto *ofproto_)
1140 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1145 close_dpif_backer(struct dpif_backer *backer)
1147 struct shash_node *node;
1149 ovs_assert(backer->refcount > 0);
1151 if (--backer->refcount) {
1155 drop_key_clear(backer);
1156 hmap_destroy(&backer->drop_keys);
1158 simap_destroy(&backer->tnl_backers);
1159 hmap_destroy(&backer->odp_to_ofport_map);
1160 node = shash_find(&all_dpif_backers, backer->type);
1162 shash_delete(&all_dpif_backers, node);
1163 dpif_close(backer->dpif);
1168 /* Datapath port slated for removal from datapath. */
1169 struct odp_garbage {
1170 struct list list_node;
1175 open_dpif_backer(const char *type, struct dpif_backer **backerp)
1177 struct dpif_backer *backer;
1178 struct dpif_port_dump port_dump;
1179 struct dpif_port port;
1180 struct shash_node *node;
1181 struct list garbage_list;
1182 struct odp_garbage *garbage, *next;
1188 backer = shash_find_data(&all_dpif_backers, type);
1195 backer_name = xasprintf("ovs-%s", type);
1197 /* Remove any existing datapaths, since we assume we're the only
1198 * userspace controlling the datapath. */
1200 dp_enumerate_names(type, &names);
1201 SSET_FOR_EACH(name, &names) {
1202 struct dpif *old_dpif;
1204 /* Don't remove our backer if it exists. */
1205 if (!strcmp(name, backer_name)) {
1209 if (dpif_open(name, type, &old_dpif)) {
1210 VLOG_WARN("couldn't open old datapath %s to remove it", name);
1212 dpif_delete(old_dpif);
1213 dpif_close(old_dpif);
1216 sset_destroy(&names);
1218 backer = xmalloc(sizeof *backer);
1220 error = dpif_create_and_open(backer_name, type, &backer->dpif);
1223 VLOG_ERR("failed to open datapath of type %s: %s", type,
1229 backer->type = xstrdup(type);
1230 backer->refcount = 1;
1231 hmap_init(&backer->odp_to_ofport_map);
1232 hmap_init(&backer->drop_keys);
1233 timer_set_duration(&backer->next_expiration, 1000);
1234 backer->need_revalidate = 0;
1235 simap_init(&backer->tnl_backers);
1236 tag_set_init(&backer->revalidate_set);
1239 dpif_flow_flush(backer->dpif);
1241 /* Loop through the ports already on the datapath and remove any
1242 * that we don't need anymore. */
1243 list_init(&garbage_list);
1244 dpif_port_dump_start(&port_dump, backer->dpif);
1245 while (dpif_port_dump_next(&port_dump, &port)) {
1246 node = shash_find(&init_ofp_ports, port.name);
1247 if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
1248 garbage = xmalloc(sizeof *garbage);
1249 garbage->odp_port = port.port_no;
1250 list_push_front(&garbage_list, &garbage->list_node);
1253 dpif_port_dump_done(&port_dump);
1255 LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) {
1256 dpif_port_del(backer->dpif, garbage->odp_port);
1257 list_remove(&garbage->list_node);
1261 shash_add(&all_dpif_backers, type, backer);
1263 error = dpif_recv_set(backer->dpif, true);
1265 VLOG_ERR("failed to listen on datapath of type %s: %s",
1266 type, strerror(error));
1267 close_dpif_backer(backer);
1275 construct(struct ofproto *ofproto_)
1277 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1278 struct shash_node *node, *next;
1283 error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1288 max_ports = dpif_get_max_ports(ofproto->backer->dpif);
1289 ofproto_init_max_ports(ofproto_, MIN(max_ports, OFPP_MAX));
1291 ofproto->n_matches = 0;
1293 ofproto->netflow = NULL;
1294 ofproto->sflow = NULL;
1295 ofproto->stp = NULL;
1296 hmap_init(&ofproto->bundles);
1297 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1298 for (i = 0; i < MAX_MIRRORS; i++) {
1299 ofproto->mirrors[i] = NULL;
1301 ofproto->has_bonded_bundles = false;
1303 hmap_init(&ofproto->facets);
1304 hmap_init(&ofproto->subfacets);
1305 ofproto->governor = NULL;
1306 ofproto->consistency_rl = LLONG_MIN;
1308 for (i = 0; i < N_TABLES; i++) {
1309 struct table_dpif *table = &ofproto->tables[i];
1311 table->catchall_table = NULL;
1312 table->other_table = NULL;
1313 table->basis = random_uint32();
1316 list_init(&ofproto->completions);
1318 ofproto_dpif_unixctl_init();
1320 ofproto->has_mirrors = false;
1321 ofproto->has_bundle_action = false;
1323 hmap_init(&ofproto->vlandev_map);
1324 hmap_init(&ofproto->realdev_vid_map);
1326 sset_init(&ofproto->ports);
1327 sset_init(&ofproto->ghost_ports);
1328 sset_init(&ofproto->port_poll_set);
1329 ofproto->port_poll_errno = 0;
1331 SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1332 struct iface_hint *iface_hint = node->data;
1334 if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1335 /* Check if the datapath already has this port. */
1336 if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1337 sset_add(&ofproto->ports, node->name);
1340 free(iface_hint->br_name);
1341 free(iface_hint->br_type);
1343 shash_delete(&init_ofp_ports, node);
1347 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1348 hash_string(ofproto->up.name, 0));
1349 memset(&ofproto->stats, 0, sizeof ofproto->stats);
1351 ofproto_init_tables(ofproto_, N_TABLES);
1352 error = add_internal_flows(ofproto);
1353 ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1356 ofproto->n_missed = 0;
1358 ofproto->max_n_subfacet = 0;
1359 ofproto->created = time_msec();
1360 ofproto->last_minute = ofproto->created;
1361 memset(&ofproto->hourly, 0, sizeof ofproto->hourly);
1362 memset(&ofproto->daily, 0, sizeof ofproto->daily);
1363 ofproto->subfacet_add_count = 0;
1364 ofproto->subfacet_del_count = 0;
1365 ofproto->total_subfacet_add_count = 0;
1366 ofproto->total_subfacet_del_count = 0;
1367 ofproto->total_subfacet_life_span = 0;
1368 ofproto->total_subfacet_count = 0;
1369 ofproto->n_update_stats = 0;
1375 add_internal_flow(struct ofproto_dpif *ofproto, int id,
1376 const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1378 struct ofputil_flow_mod fm;
1381 match_init_catchall(&fm.match);
1383 match_set_reg(&fm.match, 0, id);
1384 fm.new_cookie = htonll(0);
1385 fm.cookie = htonll(0);
1386 fm.cookie_mask = htonll(0);
1387 fm.table_id = TBL_INTERNAL;
1388 fm.command = OFPFC_ADD;
1389 fm.idle_timeout = 0;
1390 fm.hard_timeout = 0;
1394 fm.ofpacts = ofpacts->data;
1395 fm.ofpacts_len = ofpacts->size;
1397 error = ofproto_flow_mod(&ofproto->up, &fm);
1399 VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
1400 id, ofperr_to_string(error));
1404 *rulep = rule_dpif_lookup__(ofproto, &fm.match.flow, TBL_INTERNAL);
1405 ovs_assert(*rulep != NULL);
1411 add_internal_flows(struct ofproto_dpif *ofproto)
1413 struct ofpact_controller *controller;
1414 uint64_t ofpacts_stub[128 / 8];
1415 struct ofpbuf ofpacts;
1419 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1422 controller = ofpact_put_CONTROLLER(&ofpacts);
1423 controller->max_len = UINT16_MAX;
1424 controller->controller_id = 0;
1425 controller->reason = OFPR_NO_MATCH;
1426 ofpact_pad(&ofpacts);
1428 error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule);
1433 ofpbuf_clear(&ofpacts);
1434 error = add_internal_flow(ofproto, id++, &ofpacts,
1435 &ofproto->no_packet_in_rule);
1440 complete_operations(struct ofproto_dpif *ofproto)
1442 struct dpif_completion *c, *next;
1444 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
1445 ofoperation_complete(c->op, 0);
1446 list_remove(&c->list_node);
1452 destruct(struct ofproto *ofproto_)
1454 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1455 struct rule_dpif *rule, *next_rule;
1456 struct oftable *table;
1459 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1460 complete_operations(ofproto);
1462 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1463 struct cls_cursor cursor;
1465 cls_cursor_init(&cursor, &table->cls, NULL);
1466 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1467 ofproto_rule_destroy(&rule->up);
1471 for (i = 0; i < MAX_MIRRORS; i++) {
1472 mirror_destroy(ofproto->mirrors[i]);
1475 netflow_destroy(ofproto->netflow);
1476 dpif_sflow_destroy(ofproto->sflow);
1477 hmap_destroy(&ofproto->bundles);
1478 mac_learning_destroy(ofproto->ml);
1480 hmap_destroy(&ofproto->facets);
1481 hmap_destroy(&ofproto->subfacets);
1482 governor_destroy(ofproto->governor);
1484 hmap_destroy(&ofproto->vlandev_map);
1485 hmap_destroy(&ofproto->realdev_vid_map);
1487 sset_destroy(&ofproto->ports);
1488 sset_destroy(&ofproto->ghost_ports);
1489 sset_destroy(&ofproto->port_poll_set);
1491 close_dpif_backer(ofproto->backer);
1495 run_fast(struct ofproto *ofproto_)
1497 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1498 struct ofport_dpif *ofport;
1500 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1501 port_run_fast(ofport);
1508 run(struct ofproto *ofproto_)
1510 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1511 struct ofport_dpif *ofport;
1512 struct ofbundle *bundle;
1516 complete_operations(ofproto);
1519 error = run_fast(ofproto_);
1524 if (ofproto->netflow) {
1525 if (netflow_run(ofproto->netflow)) {
1526 send_netflow_active_timeouts(ofproto);
1529 if (ofproto->sflow) {
1530 dpif_sflow_run(ofproto->sflow);
1533 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1536 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1541 mac_learning_run(ofproto->ml, &ofproto->backer->revalidate_set);
1543 /* Check the consistency of a random facet, to aid debugging. */
1544 if (time_msec() >= ofproto->consistency_rl
1545 && !hmap_is_empty(&ofproto->facets)
1546 && !ofproto->backer->need_revalidate) {
1547 struct facet *facet;
1549 ofproto->consistency_rl = time_msec() + 250;
1551 facet = CONTAINER_OF(hmap_random_node(&ofproto->facets),
1552 struct facet, hmap_node);
1553 if (!tag_set_intersects(&ofproto->backer->revalidate_set,
1555 if (!facet_check_consistency(facet)) {
1556 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
1561 if (ofproto->governor) {
1564 governor_run(ofproto->governor);
1566 /* If the governor has shrunk to its minimum size and the number of
1567 * subfacets has dwindled, then drop the governor entirely.
1569 * For hysteresis, the number of subfacets to drop the governor is
1570 * smaller than the number needed to trigger its creation. */
1571 n_subfacets = hmap_count(&ofproto->subfacets);
1572 if (n_subfacets * 4 < ofproto->up.flow_eviction_threshold
1573 && governor_is_idle(ofproto->governor)) {
1574 governor_destroy(ofproto->governor);
1575 ofproto->governor = NULL;
1583 wait(struct ofproto *ofproto_)
1585 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1586 struct ofport_dpif *ofport;
1587 struct ofbundle *bundle;
1589 if (!clogged && !list_is_empty(&ofproto->completions)) {
1590 poll_immediate_wake();
1593 dpif_wait(ofproto->backer->dpif);
1594 dpif_recv_wait(ofproto->backer->dpif);
1595 if (ofproto->sflow) {
1596 dpif_sflow_wait(ofproto->sflow);
1598 if (!tag_set_is_empty(&ofproto->backer->revalidate_set)) {
1599 poll_immediate_wake();
1601 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1604 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1605 bundle_wait(bundle);
1607 if (ofproto->netflow) {
1608 netflow_wait(ofproto->netflow);
1610 mac_learning_wait(ofproto->ml);
1612 if (ofproto->backer->need_revalidate) {
1613 /* Shouldn't happen, but if it does just go around again. */
1614 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1615 poll_immediate_wake();
1617 if (ofproto->governor) {
1618 governor_wait(ofproto->governor);
1623 get_memory_usage(const struct ofproto *ofproto_, struct simap *usage)
1625 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1627 simap_increase(usage, "facets", hmap_count(&ofproto->facets));
1628 simap_increase(usage, "subfacets", hmap_count(&ofproto->subfacets));
1632 flush(struct ofproto *ofproto_)
1634 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1635 struct subfacet *subfacet, *next_subfacet;
1636 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
1640 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
1641 &ofproto->subfacets) {
1642 if (subfacet->path != SF_NOT_INSTALLED) {
1643 batch[n_batch++] = subfacet;
1644 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
1645 subfacet_destroy_batch(ofproto, batch, n_batch);
1649 subfacet_destroy(subfacet);
1654 subfacet_destroy_batch(ofproto, batch, n_batch);
1659 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1660 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1662 *arp_match_ip = true;
1663 *actions = (OFPUTIL_A_OUTPUT |
1664 OFPUTIL_A_SET_VLAN_VID |
1665 OFPUTIL_A_SET_VLAN_PCP |
1666 OFPUTIL_A_STRIP_VLAN |
1667 OFPUTIL_A_SET_DL_SRC |
1668 OFPUTIL_A_SET_DL_DST |
1669 OFPUTIL_A_SET_NW_SRC |
1670 OFPUTIL_A_SET_NW_DST |
1671 OFPUTIL_A_SET_NW_TOS |
1672 OFPUTIL_A_SET_TP_SRC |
1673 OFPUTIL_A_SET_TP_DST |
1678 get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots)
1680 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1681 struct dpif_dp_stats s;
1683 strcpy(ots->name, "classifier");
1685 dpif_get_dp_stats(ofproto->backer->dpif, &s);
1687 ots->lookup_count = htonll(s.n_hit + s.n_missed);
1688 ots->matched_count = htonll(s.n_hit + ofproto->n_matches);
1691 static struct ofport *
1694 struct ofport_dpif *port = xmalloc(sizeof *port);
1699 port_dealloc(struct ofport *port_)
1701 struct ofport_dpif *port = ofport_dpif_cast(port_);
1706 port_construct(struct ofport *port_)
1708 struct ofport_dpif *port = ofport_dpif_cast(port_);
1709 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1710 const struct netdev *netdev = port->up.netdev;
1711 struct dpif_port dpif_port;
1714 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1715 port->bundle = NULL;
1717 port->tag = tag_create_random();
1718 port->may_enable = true;
1719 port->stp_port = NULL;
1720 port->stp_state = STP_DISABLED;
1721 port->tnl_port = NULL;
1722 hmap_init(&port->priorities);
1723 port->realdev_ofp_port = 0;
1724 port->vlandev_vid = 0;
1725 port->carrier_seq = netdev_get_carrier_resets(netdev);
1727 if (netdev_vport_is_patch(netdev)) {
1728 /* XXX By bailing out here, we don't do required sFlow work. */
1729 port->odp_port = OVSP_NONE;
1733 error = dpif_port_query_by_name(ofproto->backer->dpif,
1734 netdev_vport_get_dpif_port(netdev),
1740 port->odp_port = dpif_port.port_no;
1742 if (netdev_get_tunnel_config(netdev)) {
1743 port->tnl_port = tnl_port_add(&port->up, port->odp_port);
1745 /* Sanity-check that a mapping doesn't already exist. This
1746 * shouldn't happen for non-tunnel ports. */
1747 if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1748 VLOG_ERR("port %s already has an OpenFlow port number",
1750 dpif_port_destroy(&dpif_port);
1754 hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1755 hash_int(port->odp_port, 0));
1757 dpif_port_destroy(&dpif_port);
1759 if (ofproto->sflow) {
1760 dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1767 port_destruct(struct ofport *port_)
1769 struct ofport_dpif *port = ofport_dpif_cast(port_);
1770 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1771 const char *dp_port_name = netdev_vport_get_dpif_port(port->up.netdev);
1772 const char *devname = netdev_get_name(port->up.netdev);
1774 if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1775 /* The underlying device is still there, so delete it. This
1776 * happens when the ofproto is being destroyed, since the caller
1777 * assumes that removal of attached ports will happen as part of
1779 if (!port->tnl_port) {
1780 dpif_port_del(ofproto->backer->dpif, port->odp_port);
1782 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1785 if (port->odp_port != OVSP_NONE && !port->tnl_port) {
1786 hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1789 tnl_port_del(port->tnl_port);
1790 sset_find_and_delete(&ofproto->ports, devname);
1791 sset_find_and_delete(&ofproto->ghost_ports, devname);
1792 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1793 bundle_remove(port_);
1794 set_cfm(port_, NULL);
1795 if (ofproto->sflow) {
1796 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1799 ofport_clear_priorities(port);
1800 hmap_destroy(&port->priorities);
1804 port_modified(struct ofport *port_)
1806 struct ofport_dpif *port = ofport_dpif_cast(port_);
1808 if (port->bundle && port->bundle->bond) {
1809 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1814 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1816 struct ofport_dpif *port = ofport_dpif_cast(port_);
1817 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1818 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1820 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1821 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1822 OFPUTIL_PC_NO_PACKET_IN)) {
1823 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1825 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1826 bundle_update(port->bundle);
1832 set_sflow(struct ofproto *ofproto_,
1833 const struct ofproto_sflow_options *sflow_options)
1835 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1836 struct dpif_sflow *ds = ofproto->sflow;
1838 if (sflow_options) {
1840 struct ofport_dpif *ofport;
1842 ds = ofproto->sflow = dpif_sflow_create();
1843 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1844 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1846 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1848 dpif_sflow_set_options(ds, sflow_options);
1851 dpif_sflow_destroy(ds);
1852 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1853 ofproto->sflow = NULL;
1860 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1862 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1869 struct ofproto_dpif *ofproto;
1871 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1872 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1873 ofport->cfm = cfm_create(netdev_get_name(ofport->up.netdev));
1876 if (cfm_configure(ofport->cfm, s)) {
1882 cfm_destroy(ofport->cfm);
1888 get_cfm_status(const struct ofport *ofport_,
1889 struct ofproto_cfm_status *status)
1891 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1894 status->faults = cfm_get_fault(ofport->cfm);
1895 status->remote_opstate = cfm_get_opup(ofport->cfm);
1896 status->health = cfm_get_health(ofport->cfm);
1897 cfm_get_remote_mpids(ofport->cfm, &status->rmps, &status->n_rmps);
1904 /* Spanning Tree. */
1907 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
1909 struct ofproto_dpif *ofproto = ofproto_;
1910 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
1911 struct ofport_dpif *ofport;
1913 ofport = stp_port_get_aux(sp);
1915 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
1916 ofproto->up.name, port_num);
1918 struct eth_header *eth = pkt->l2;
1920 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
1921 if (eth_addr_is_zero(eth->eth_src)) {
1922 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
1923 "with unknown MAC", ofproto->up.name, port_num);
1925 send_packet(ofport, pkt);
1931 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
1933 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
1935 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1937 /* Only revalidate flows if the configuration changed. */
1938 if (!s != !ofproto->stp) {
1939 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1943 if (!ofproto->stp) {
1944 ofproto->stp = stp_create(ofproto_->name, s->system_id,
1945 send_bpdu_cb, ofproto);
1946 ofproto->stp_last_tick = time_msec();
1949 stp_set_bridge_id(ofproto->stp, s->system_id);
1950 stp_set_bridge_priority(ofproto->stp, s->priority);
1951 stp_set_hello_time(ofproto->stp, s->hello_time);
1952 stp_set_max_age(ofproto->stp, s->max_age);
1953 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
1955 struct ofport *ofport;
1957 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
1958 set_stp_port(ofport, NULL);
1961 stp_destroy(ofproto->stp);
1962 ofproto->stp = NULL;
1969 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
1971 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1975 s->bridge_id = stp_get_bridge_id(ofproto->stp);
1976 s->designated_root = stp_get_designated_root(ofproto->stp);
1977 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
1986 update_stp_port_state(struct ofport_dpif *ofport)
1988 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1989 enum stp_state state;
1991 /* Figure out new state. */
1992 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
1996 if (ofport->stp_state != state) {
1997 enum ofputil_port_state of_state;
2000 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
2001 netdev_get_name(ofport->up.netdev),
2002 stp_state_name(ofport->stp_state),
2003 stp_state_name(state));
2004 if (stp_learn_in_state(ofport->stp_state)
2005 != stp_learn_in_state(state)) {
2006 /* xxx Learning action flows should also be flushed. */
2007 mac_learning_flush(ofproto->ml,
2008 &ofproto->backer->revalidate_set);
2010 fwd_change = stp_forward_in_state(ofport->stp_state)
2011 != stp_forward_in_state(state);
2013 ofproto->backer->need_revalidate = REV_STP;
2014 ofport->stp_state = state;
2015 ofport->stp_state_entered = time_msec();
2017 if (fwd_change && ofport->bundle) {
2018 bundle_update(ofport->bundle);
2021 /* Update the STP state bits in the OpenFlow port description. */
2022 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
2023 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
2024 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
2025 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
2026 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
2028 ofproto_port_set_state(&ofport->up, of_state);
2032 /* Configures STP on 'ofport_' using the settings defined in 's'. The
2033 * caller is responsible for assigning STP port numbers and ensuring
2034 * there are no duplicates. */
2036 set_stp_port(struct ofport *ofport_,
2037 const struct ofproto_port_stp_settings *s)
2039 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2040 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2041 struct stp_port *sp = ofport->stp_port;
2043 if (!s || !s->enable) {
2045 ofport->stp_port = NULL;
2046 stp_port_disable(sp);
2047 update_stp_port_state(ofport);
2050 } else if (sp && stp_port_no(sp) != s->port_num
2051 && ofport == stp_port_get_aux(sp)) {
2052 /* The port-id changed, so disable the old one if it's not
2053 * already in use by another port. */
2054 stp_port_disable(sp);
2057 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
2058 stp_port_enable(sp);
2060 stp_port_set_aux(sp, ofport);
2061 stp_port_set_priority(sp, s->priority);
2062 stp_port_set_path_cost(sp, s->path_cost);
2064 update_stp_port_state(ofport);
2070 get_stp_port_status(struct ofport *ofport_,
2071 struct ofproto_port_stp_status *s)
2073 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2074 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2075 struct stp_port *sp = ofport->stp_port;
2077 if (!ofproto->stp || !sp) {
2083 s->port_id = stp_port_get_id(sp);
2084 s->state = stp_port_get_state(sp);
2085 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
2086 s->role = stp_port_get_role(sp);
2087 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
2093 stp_run(struct ofproto_dpif *ofproto)
2096 long long int now = time_msec();
2097 long long int elapsed = now - ofproto->stp_last_tick;
2098 struct stp_port *sp;
2101 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
2102 ofproto->stp_last_tick = now;
2104 while (stp_get_changed_port(ofproto->stp, &sp)) {
2105 struct ofport_dpif *ofport = stp_port_get_aux(sp);
2108 update_stp_port_state(ofport);
2112 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
2113 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
2119 stp_wait(struct ofproto_dpif *ofproto)
2122 poll_timer_wait(1000);
2126 /* Returns true if STP should process 'flow'. */
2128 stp_should_process_flow(const struct flow *flow)
2130 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
2134 stp_process_packet(const struct ofport_dpif *ofport,
2135 const struct ofpbuf *packet)
2137 struct ofpbuf payload = *packet;
2138 struct eth_header *eth = payload.data;
2139 struct stp_port *sp = ofport->stp_port;
2141 /* Sink packets on ports that have STP disabled when the bridge has
2143 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
2147 /* Trim off padding on payload. */
2148 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
2149 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
2152 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
2153 stp_received_bpdu(sp, payload.data, payload.size);
2157 static struct priority_to_dscp *
2158 get_priority(const struct ofport_dpif *ofport, uint32_t priority)
2160 struct priority_to_dscp *pdscp;
2163 hash = hash_int(priority, 0);
2164 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) {
2165 if (pdscp->priority == priority) {
2173 ofport_clear_priorities(struct ofport_dpif *ofport)
2175 struct priority_to_dscp *pdscp, *next;
2177 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) {
2178 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
2184 set_queues(struct ofport *ofport_,
2185 const struct ofproto_port_queue *qdscp_list,
2188 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2189 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2190 struct hmap new = HMAP_INITIALIZER(&new);
2193 for (i = 0; i < n_qdscp; i++) {
2194 struct priority_to_dscp *pdscp;
2198 dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
2199 if (dpif_queue_to_priority(ofproto->backer->dpif, qdscp_list[i].queue,
2204 pdscp = get_priority(ofport, priority);
2206 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
2208 pdscp = xmalloc(sizeof *pdscp);
2209 pdscp->priority = priority;
2211 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2214 if (pdscp->dscp != dscp) {
2216 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2219 hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0));
2222 if (!hmap_is_empty(&ofport->priorities)) {
2223 ofport_clear_priorities(ofport);
2224 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2227 hmap_swap(&new, &ofport->priorities);
2235 /* Expires all MAC learning entries associated with 'bundle' and forces its
2236 * ofproto to revalidate every flow.
2238 * Normally MAC learning entries are removed only from the ofproto associated
2239 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2240 * are removed from every ofproto. When patch ports and SLB bonds are in use
2241 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2242 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2243 * with the host from which it migrated. */
2245 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2247 struct ofproto_dpif *ofproto = bundle->ofproto;
2248 struct mac_learning *ml = ofproto->ml;
2249 struct mac_entry *mac, *next_mac;
2251 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2252 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2253 if (mac->port.p == bundle) {
2255 struct ofproto_dpif *o;
2257 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2259 struct mac_entry *e;
2261 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan,
2264 mac_learning_expire(o->ml, e);
2270 mac_learning_expire(ml, mac);
2275 static struct ofbundle *
2276 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2278 struct ofbundle *bundle;
2280 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2281 &ofproto->bundles) {
2282 if (bundle->aux == aux) {
2289 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
2290 * ones that are found to 'bundles'. */
2292 bundle_lookup_multiple(struct ofproto_dpif *ofproto,
2293 void **auxes, size_t n_auxes,
2294 struct hmapx *bundles)
2298 hmapx_init(bundles);
2299 for (i = 0; i < n_auxes; i++) {
2300 struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
2302 hmapx_add(bundles, bundle);
2308 bundle_update(struct ofbundle *bundle)
2310 struct ofport_dpif *port;
2312 bundle->floodable = true;
2313 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2314 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2315 || !stp_forward_in_state(port->stp_state)) {
2316 bundle->floodable = false;
2323 bundle_del_port(struct ofport_dpif *port)
2325 struct ofbundle *bundle = port->bundle;
2327 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2329 list_remove(&port->bundle_node);
2330 port->bundle = NULL;
2333 lacp_slave_unregister(bundle->lacp, port);
2336 bond_slave_unregister(bundle->bond, port);
2339 bundle_update(bundle);
2343 bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port,
2344 struct lacp_slave_settings *lacp)
2346 struct ofport_dpif *port;
2348 port = get_ofp_port(bundle->ofproto, ofp_port);
2353 if (port->bundle != bundle) {
2354 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2356 bundle_del_port(port);
2359 port->bundle = bundle;
2360 list_push_back(&bundle->ports, &port->bundle_node);
2361 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2362 || !stp_forward_in_state(port->stp_state)) {
2363 bundle->floodable = false;
2367 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2368 lacp_slave_register(bundle->lacp, port, lacp);
2375 bundle_destroy(struct ofbundle *bundle)
2377 struct ofproto_dpif *ofproto;
2378 struct ofport_dpif *port, *next_port;
2385 ofproto = bundle->ofproto;
2386 for (i = 0; i < MAX_MIRRORS; i++) {
2387 struct ofmirror *m = ofproto->mirrors[i];
2389 if (m->out == bundle) {
2391 } else if (hmapx_find_and_delete(&m->srcs, bundle)
2392 || hmapx_find_and_delete(&m->dsts, bundle)) {
2393 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2398 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2399 bundle_del_port(port);
2402 bundle_flush_macs(bundle, true);
2403 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2405 free(bundle->trunks);
2406 lacp_destroy(bundle->lacp);
2407 bond_destroy(bundle->bond);
2412 bundle_set(struct ofproto *ofproto_, void *aux,
2413 const struct ofproto_bundle_settings *s)
2415 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2416 bool need_flush = false;
2417 struct ofport_dpif *port;
2418 struct ofbundle *bundle;
2419 unsigned long *trunks;
2425 bundle_destroy(bundle_lookup(ofproto, aux));
2429 ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2430 ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2432 bundle = bundle_lookup(ofproto, aux);
2434 bundle = xmalloc(sizeof *bundle);
2436 bundle->ofproto = ofproto;
2437 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2438 hash_pointer(aux, 0));
2440 bundle->name = NULL;
2442 list_init(&bundle->ports);
2443 bundle->vlan_mode = PORT_VLAN_TRUNK;
2445 bundle->trunks = NULL;
2446 bundle->use_priority_tags = s->use_priority_tags;
2447 bundle->lacp = NULL;
2448 bundle->bond = NULL;
2450 bundle->floodable = true;
2452 bundle->src_mirrors = 0;
2453 bundle->dst_mirrors = 0;
2454 bundle->mirror_out = 0;
2457 if (!bundle->name || strcmp(s->name, bundle->name)) {
2459 bundle->name = xstrdup(s->name);
2464 if (!bundle->lacp) {
2465 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2466 bundle->lacp = lacp_create();
2468 lacp_configure(bundle->lacp, s->lacp);
2470 lacp_destroy(bundle->lacp);
2471 bundle->lacp = NULL;
2474 /* Update set of ports. */
2476 for (i = 0; i < s->n_slaves; i++) {
2477 if (!bundle_add_port(bundle, s->slaves[i],
2478 s->lacp ? &s->lacp_slaves[i] : NULL)) {
2482 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2483 struct ofport_dpif *next_port;
2485 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2486 for (i = 0; i < s->n_slaves; i++) {
2487 if (s->slaves[i] == port->up.ofp_port) {
2492 bundle_del_port(port);
2496 ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2498 if (list_is_empty(&bundle->ports)) {
2499 bundle_destroy(bundle);
2503 /* Set VLAN tagging mode */
2504 if (s->vlan_mode != bundle->vlan_mode
2505 || s->use_priority_tags != bundle->use_priority_tags) {
2506 bundle->vlan_mode = s->vlan_mode;
2507 bundle->use_priority_tags = s->use_priority_tags;
2512 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2513 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2515 if (vlan != bundle->vlan) {
2516 bundle->vlan = vlan;
2520 /* Get trunked VLANs. */
2521 switch (s->vlan_mode) {
2522 case PORT_VLAN_ACCESS:
2526 case PORT_VLAN_TRUNK:
2527 trunks = CONST_CAST(unsigned long *, s->trunks);
2530 case PORT_VLAN_NATIVE_UNTAGGED:
2531 case PORT_VLAN_NATIVE_TAGGED:
2532 if (vlan != 0 && (!s->trunks
2533 || !bitmap_is_set(s->trunks, vlan)
2534 || bitmap_is_set(s->trunks, 0))) {
2535 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2537 trunks = bitmap_clone(s->trunks, 4096);
2539 trunks = bitmap_allocate1(4096);
2541 bitmap_set1(trunks, vlan);
2542 bitmap_set0(trunks, 0);
2544 trunks = CONST_CAST(unsigned long *, s->trunks);
2551 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2552 free(bundle->trunks);
2553 if (trunks == s->trunks) {
2554 bundle->trunks = vlan_bitmap_clone(trunks);
2556 bundle->trunks = trunks;
2561 if (trunks != s->trunks) {
2566 if (!list_is_short(&bundle->ports)) {
2567 bundle->ofproto->has_bonded_bundles = true;
2569 if (bond_reconfigure(bundle->bond, s->bond)) {
2570 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2573 bundle->bond = bond_create(s->bond);
2574 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2577 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2578 bond_slave_register(bundle->bond, port, port->up.netdev);
2581 bond_destroy(bundle->bond);
2582 bundle->bond = NULL;
2585 /* If we changed something that would affect MAC learning, un-learn
2586 * everything on this port and force flow revalidation. */
2588 bundle_flush_macs(bundle, false);
2595 bundle_remove(struct ofport *port_)
2597 struct ofport_dpif *port = ofport_dpif_cast(port_);
2598 struct ofbundle *bundle = port->bundle;
2601 bundle_del_port(port);
2602 if (list_is_empty(&bundle->ports)) {
2603 bundle_destroy(bundle);
2604 } else if (list_is_short(&bundle->ports)) {
2605 bond_destroy(bundle->bond);
2606 bundle->bond = NULL;
2612 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2614 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2615 struct ofport_dpif *port = port_;
2616 uint8_t ea[ETH_ADDR_LEN];
2619 error = netdev_get_etheraddr(port->up.netdev, ea);
2621 struct ofpbuf packet;
2624 ofpbuf_init(&packet, 0);
2625 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2627 memcpy(packet_pdu, pdu, pdu_size);
2629 send_packet(port, &packet);
2630 ofpbuf_uninit(&packet);
2632 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2633 "%s (%s)", port->bundle->name,
2634 netdev_get_name(port->up.netdev), strerror(error));
2639 bundle_send_learning_packets(struct ofbundle *bundle)
2641 struct ofproto_dpif *ofproto = bundle->ofproto;
2642 int error, n_packets, n_errors;
2643 struct mac_entry *e;
2645 error = n_packets = n_errors = 0;
2646 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2647 if (e->port.p != bundle) {
2648 struct ofpbuf *learning_packet;
2649 struct ofport_dpif *port;
2653 /* The assignment to "port" is unnecessary but makes "grep"ing for
2654 * struct ofport_dpif more effective. */
2655 learning_packet = bond_compose_learning_packet(bundle->bond,
2659 ret = send_packet(port, learning_packet);
2660 ofpbuf_delete(learning_packet);
2670 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2671 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2672 "packets, last error was: %s",
2673 bundle->name, n_errors, n_packets, strerror(error));
2675 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2676 bundle->name, n_packets);
2681 bundle_run(struct ofbundle *bundle)
2684 lacp_run(bundle->lacp, send_pdu_cb);
2687 struct ofport_dpif *port;
2689 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2690 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2693 bond_run(bundle->bond, &bundle->ofproto->backer->revalidate_set,
2694 lacp_status(bundle->lacp));
2695 if (bond_should_send_learning_packets(bundle->bond)) {
2696 bundle_send_learning_packets(bundle);
2702 bundle_wait(struct ofbundle *bundle)
2705 lacp_wait(bundle->lacp);
2708 bond_wait(bundle->bond);
2715 mirror_scan(struct ofproto_dpif *ofproto)
2719 for (idx = 0; idx < MAX_MIRRORS; idx++) {
2720 if (!ofproto->mirrors[idx]) {
2727 static struct ofmirror *
2728 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
2732 for (i = 0; i < MAX_MIRRORS; i++) {
2733 struct ofmirror *mirror = ofproto->mirrors[i];
2734 if (mirror && mirror->aux == aux) {
2742 /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */
2744 mirror_update_dups(struct ofproto_dpif *ofproto)
2748 for (i = 0; i < MAX_MIRRORS; i++) {
2749 struct ofmirror *m = ofproto->mirrors[i];
2752 m->dup_mirrors = MIRROR_MASK_C(1) << i;
2756 for (i = 0; i < MAX_MIRRORS; i++) {
2757 struct ofmirror *m1 = ofproto->mirrors[i];
2764 for (j = i + 1; j < MAX_MIRRORS; j++) {
2765 struct ofmirror *m2 = ofproto->mirrors[j];
2767 if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) {
2768 m1->dup_mirrors |= MIRROR_MASK_C(1) << j;
2769 m2->dup_mirrors |= m1->dup_mirrors;
2776 mirror_set(struct ofproto *ofproto_, void *aux,
2777 const struct ofproto_mirror_settings *s)
2779 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2780 mirror_mask_t mirror_bit;
2781 struct ofbundle *bundle;
2782 struct ofmirror *mirror;
2783 struct ofbundle *out;
2784 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
2785 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
2788 mirror = mirror_lookup(ofproto, aux);
2790 mirror_destroy(mirror);
2796 idx = mirror_scan(ofproto);
2798 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
2800 ofproto->up.name, MAX_MIRRORS, s->name);
2804 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
2805 mirror->ofproto = ofproto;
2808 mirror->out_vlan = -1;
2809 mirror->name = NULL;
2812 if (!mirror->name || strcmp(s->name, mirror->name)) {
2814 mirror->name = xstrdup(s->name);
2817 /* Get the new configuration. */
2818 if (s->out_bundle) {
2819 out = bundle_lookup(ofproto, s->out_bundle);
2821 mirror_destroy(mirror);
2827 out_vlan = s->out_vlan;
2829 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
2830 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
2832 /* If the configuration has not changed, do nothing. */
2833 if (hmapx_equals(&srcs, &mirror->srcs)
2834 && hmapx_equals(&dsts, &mirror->dsts)
2835 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
2836 && mirror->out == out
2837 && mirror->out_vlan == out_vlan)
2839 hmapx_destroy(&srcs);
2840 hmapx_destroy(&dsts);
2844 hmapx_swap(&srcs, &mirror->srcs);
2845 hmapx_destroy(&srcs);
2847 hmapx_swap(&dsts, &mirror->dsts);
2848 hmapx_destroy(&dsts);
2850 free(mirror->vlans);
2851 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
2854 mirror->out_vlan = out_vlan;
2856 /* Update bundles. */
2857 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2858 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
2859 if (hmapx_contains(&mirror->srcs, bundle)) {
2860 bundle->src_mirrors |= mirror_bit;
2862 bundle->src_mirrors &= ~mirror_bit;
2865 if (hmapx_contains(&mirror->dsts, bundle)) {
2866 bundle->dst_mirrors |= mirror_bit;
2868 bundle->dst_mirrors &= ~mirror_bit;
2871 if (mirror->out == bundle) {
2872 bundle->mirror_out |= mirror_bit;
2874 bundle->mirror_out &= ~mirror_bit;
2878 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2879 ofproto->has_mirrors = true;
2880 mac_learning_flush(ofproto->ml,
2881 &ofproto->backer->revalidate_set);
2882 mirror_update_dups(ofproto);
2888 mirror_destroy(struct ofmirror *mirror)
2890 struct ofproto_dpif *ofproto;
2891 mirror_mask_t mirror_bit;
2892 struct ofbundle *bundle;
2899 ofproto = mirror->ofproto;
2900 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2901 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
2903 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2904 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
2905 bundle->src_mirrors &= ~mirror_bit;
2906 bundle->dst_mirrors &= ~mirror_bit;
2907 bundle->mirror_out &= ~mirror_bit;
2910 hmapx_destroy(&mirror->srcs);
2911 hmapx_destroy(&mirror->dsts);
2912 free(mirror->vlans);
2914 ofproto->mirrors[mirror->idx] = NULL;
2918 mirror_update_dups(ofproto);
2920 ofproto->has_mirrors = false;
2921 for (i = 0; i < MAX_MIRRORS; i++) {
2922 if (ofproto->mirrors[i]) {
2923 ofproto->has_mirrors = true;
2930 mirror_get_stats(struct ofproto *ofproto_, void *aux,
2931 uint64_t *packets, uint64_t *bytes)
2933 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2934 struct ofmirror *mirror = mirror_lookup(ofproto, aux);
2937 *packets = *bytes = UINT64_MAX;
2943 *packets = mirror->packet_count;
2944 *bytes = mirror->byte_count;
2950 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2952 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2953 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2954 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
2960 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2962 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2963 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2964 return bundle && bundle->mirror_out != 0;
2968 forward_bpdu_changed(struct ofproto *ofproto_)
2970 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2971 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2975 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
2978 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2979 mac_learning_set_idle_time(ofproto->ml, idle_time);
2980 mac_learning_set_max_entries(ofproto->ml, max_entries);
2985 static struct ofport_dpif *
2986 get_ofp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
2988 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2989 return ofport ? ofport_dpif_cast(ofport) : NULL;
2992 static struct ofport_dpif *
2993 get_odp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
2995 struct ofport_dpif *port = odp_port_to_ofport(ofproto->backer, odp_port);
2996 return port && &ofproto->up == port->up.ofproto ? port : NULL;
3000 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
3001 struct ofproto_port *ofproto_port,
3002 struct dpif_port *dpif_port)
3004 ofproto_port->name = dpif_port->name;
3005 ofproto_port->type = dpif_port->type;
3006 ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
3009 static struct ofport_dpif *
3010 ofport_get_peer(const struct ofport_dpif *ofport_dpif)
3012 const struct ofproto_dpif *ofproto;
3015 peer = netdev_vport_patch_peer(ofport_dpif->up.netdev);
3020 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
3021 struct ofport *ofport;
3023 ofport = shash_find_data(&ofproto->up.port_by_name, peer);
3024 if (ofport && ofport->ofproto->ofproto_class == &ofproto_dpif_class) {
3025 return ofport_dpif_cast(ofport);
3032 port_run_fast(struct ofport_dpif *ofport)
3034 if (ofport->cfm && cfm_should_send_ccm(ofport->cfm)) {
3035 struct ofpbuf packet;
3037 ofpbuf_init(&packet, 0);
3038 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
3039 send_packet(ofport, &packet);
3040 ofpbuf_uninit(&packet);
3045 port_run(struct ofport_dpif *ofport)
3047 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
3048 bool carrier_changed = carrier_seq != ofport->carrier_seq;
3049 bool enable = netdev_get_carrier(ofport->up.netdev);
3051 ofport->carrier_seq = carrier_seq;
3053 port_run_fast(ofport);
3055 if (ofport->tnl_port
3056 && tnl_port_reconfigure(&ofport->up, ofport->odp_port,
3057 &ofport->tnl_port)) {
3058 ofproto_dpif_cast(ofport->up.ofproto)->backer->need_revalidate = true;
3062 int cfm_opup = cfm_get_opup(ofport->cfm);
3064 cfm_run(ofport->cfm);
3065 enable = enable && !cfm_get_fault(ofport->cfm);
3067 if (cfm_opup >= 0) {
3068 enable = enable && cfm_opup;
3072 if (ofport->bundle) {
3073 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
3074 if (carrier_changed) {
3075 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
3079 if (ofport->may_enable != enable) {
3080 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3082 if (ofproto->has_bundle_action) {
3083 ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
3087 ofport->may_enable = enable;
3091 port_wait(struct ofport_dpif *ofport)
3094 cfm_wait(ofport->cfm);
3099 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
3100 struct ofproto_port *ofproto_port)
3102 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3103 struct dpif_port dpif_port;
3106 if (sset_contains(&ofproto->ghost_ports, devname)) {
3107 const char *type = netdev_get_type_from_name(devname);
3109 /* We may be called before ofproto->up.port_by_name is populated with
3110 * the appropriate ofport. For this reason, we must get the name and
3111 * type from the netdev layer directly. */
3113 const struct ofport *ofport;
3115 ofport = shash_find_data(&ofproto->up.port_by_name, devname);
3116 ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
3117 ofproto_port->name = xstrdup(devname);
3118 ofproto_port->type = xstrdup(type);
3124 if (!sset_contains(&ofproto->ports, devname)) {
3127 error = dpif_port_query_by_name(ofproto->backer->dpif,
3128 devname, &dpif_port);
3130 ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
3136 port_add(struct ofproto *ofproto_, struct netdev *netdev)
3138 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3139 const char *dp_port_name = netdev_vport_get_dpif_port(netdev);
3140 const char *devname = netdev_get_name(netdev);
3142 if (netdev_vport_is_patch(netdev)) {
3143 sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
3147 if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
3148 uint32_t port_no = UINT32_MAX;
3151 error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
3155 if (netdev_get_tunnel_config(netdev)) {
3156 simap_put(&ofproto->backer->tnl_backers, dp_port_name, port_no);
3160 if (netdev_get_tunnel_config(netdev)) {
3161 sset_add(&ofproto->ghost_ports, devname);
3163 sset_add(&ofproto->ports, devname);
3169 port_del(struct ofproto *ofproto_, uint16_t ofp_port)
3171 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3172 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
3179 sset_find_and_delete(&ofproto->ghost_ports,
3180 netdev_get_name(ofport->up.netdev));
3181 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3182 if (!ofport->tnl_port) {
3183 error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
3185 /* The caller is going to close ofport->up.netdev. If this is a
3186 * bonded port, then the bond is using that netdev, so remove it
3187 * from the bond. The client will need to reconfigure everything
3188 * after deleting ports, so then the slave will get re-added. */
3189 bundle_remove(&ofport->up);
3196 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
3198 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3203 error = netdev_get_stats(ofport->up.netdev, stats);
3205 if (!error && ofport_->ofp_port == OFPP_LOCAL) {
3206 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3208 /* ofproto->stats.tx_packets represents packets that we created
3209 * internally and sent to some port (e.g. packets sent with
3210 * send_packet()). Account for them as if they had come from
3211 * OFPP_LOCAL and got forwarded. */
3213 if (stats->rx_packets != UINT64_MAX) {
3214 stats->rx_packets += ofproto->stats.tx_packets;
3217 if (stats->rx_bytes != UINT64_MAX) {
3218 stats->rx_bytes += ofproto->stats.tx_bytes;
3221 /* ofproto->stats.rx_packets represents packets that were received on
3222 * some port and we processed internally and dropped (e.g. STP).
3223 * Account for them as if they had been forwarded to OFPP_LOCAL. */
3225 if (stats->tx_packets != UINT64_MAX) {
3226 stats->tx_packets += ofproto->stats.rx_packets;
3229 if (stats->tx_bytes != UINT64_MAX) {
3230 stats->tx_bytes += ofproto->stats.rx_bytes;
3237 /* Account packets for LOCAL port. */
3239 ofproto_update_local_port_stats(const struct ofproto *ofproto_,
3240 size_t tx_size, size_t rx_size)
3242 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3245 ofproto->stats.rx_packets++;
3246 ofproto->stats.rx_bytes += rx_size;
3249 ofproto->stats.tx_packets++;
3250 ofproto->stats.tx_bytes += tx_size;
3254 struct port_dump_state {
3259 struct ofproto_port port;
3264 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
3266 *statep = xzalloc(sizeof(struct port_dump_state));
3271 port_dump_next(const struct ofproto *ofproto_, void *state_,
3272 struct ofproto_port *port)
3274 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3275 struct port_dump_state *state = state_;
3276 const struct sset *sset;
3277 struct sset_node *node;
3279 if (state->has_port) {
3280 ofproto_port_destroy(&state->port);
3281 state->has_port = false;
3283 sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
3284 while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
3287 error = port_query_by_name(ofproto_, node->name, &state->port);
3289 *port = state->port;
3290 state->has_port = true;
3292 } else if (error != ENODEV) {
3297 if (!state->ghost) {
3298 state->ghost = true;
3301 return port_dump_next(ofproto_, state_, port);
3308 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3310 struct port_dump_state *state = state_;
3312 if (state->has_port) {
3313 ofproto_port_destroy(&state->port);
3320 port_poll(const struct ofproto *ofproto_, char **devnamep)
3322 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3324 if (ofproto->port_poll_errno) {
3325 int error = ofproto->port_poll_errno;
3326 ofproto->port_poll_errno = 0;
3330 if (sset_is_empty(&ofproto->port_poll_set)) {
3334 *devnamep = sset_pop(&ofproto->port_poll_set);
3339 port_poll_wait(const struct ofproto *ofproto_)
3341 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3342 dpif_port_poll_wait(ofproto->backer->dpif);
3346 port_is_lacp_current(const struct ofport *ofport_)
3348 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3349 return (ofport->bundle && ofport->bundle->lacp
3350 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3354 /* Upcall handling. */
3356 /* Flow miss batching.
3358 * Some dpifs implement operations faster when you hand them off in a batch.
3359 * To allow batching, "struct flow_miss" queues the dpif-related work needed
3360 * for a given flow. Each "struct flow_miss" corresponds to sending one or
3361 * more packets, plus possibly installing the flow in the dpif.
3363 * So far we only batch the operations that affect flow setup time the most.
3364 * It's possible to batch more than that, but the benefit might be minimal. */
3366 struct hmap_node hmap_node;
3367 struct ofproto_dpif *ofproto;
3369 enum odp_key_fitness key_fitness;
3370 const struct nlattr *key;
3372 struct initial_vals initial_vals;
3373 struct list packets;
3374 enum dpif_upcall_type upcall_type;
3375 uint32_t odp_in_port;
3378 struct flow_miss_op {
3379 struct dpif_op dpif_op;
3380 void *garbage; /* Pointer to pass to free(), NULL if none. */
3381 uint64_t stub[1024 / 8]; /* Temporary buffer. */
3384 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
3385 * OpenFlow controller as necessary according to their individual
3386 * configurations. */
3388 send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet,
3389 const struct flow *flow)
3391 struct ofputil_packet_in pin;
3393 pin.packet = packet->data;
3394 pin.packet_len = packet->size;
3395 pin.reason = OFPR_NO_MATCH;
3396 pin.controller_id = 0;
3401 pin.send_len = 0; /* not used for flow table misses */
3403 flow_get_metadata(flow, &pin.fmd);
3405 connmgr_send_packet_in(ofproto->up.connmgr, &pin);
3408 static enum slow_path_reason
3409 process_special(struct ofproto_dpif *ofproto, const struct flow *flow,
3410 const struct ofport_dpif *ofport, const struct ofpbuf *packet)
3414 } else if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) {
3416 cfm_process_heartbeat(ofport->cfm, packet);
3419 } else if (ofport->bundle && ofport->bundle->lacp
3420 && flow->dl_type == htons(ETH_TYPE_LACP)) {
3422 lacp_process_packet(ofport->bundle->lacp, ofport, packet);
3425 } else if (ofproto->stp && stp_should_process_flow(flow)) {
3427 stp_process_packet(ofport, packet);
3435 static struct flow_miss *
3436 flow_miss_find(struct hmap *todo, const struct ofproto_dpif *ofproto,
3437 const struct flow *flow, uint32_t hash)
3439 struct flow_miss *miss;
3441 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
3442 if (miss->ofproto == ofproto && flow_equal(&miss->flow, flow)) {
3450 /* Partially Initializes 'op' as an "execute" operation for 'miss' and
3451 * 'packet'. The caller must initialize op->actions and op->actions_len. If
3452 * 'miss' is associated with a subfacet the caller must also initialize the
3453 * returned op->subfacet, and if anything needs to be freed after processing
3454 * the op, the caller must initialize op->garbage also. */
3456 init_flow_miss_execute_op(struct flow_miss *miss, struct ofpbuf *packet,
3457 struct flow_miss_op *op)
3459 if (miss->flow.vlan_tci != miss->initial_vals.vlan_tci) {
3460 /* This packet was received on a VLAN splinter port. We
3461 * added a VLAN to the packet to make the packet resemble
3462 * the flow, but the actions were composed assuming that
3463 * the packet contained no VLAN. So, we must remove the
3464 * VLAN header from the packet before trying to execute the
3466 eth_pop_vlan(packet);
3470 op->dpif_op.type = DPIF_OP_EXECUTE;
3471 op->dpif_op.u.execute.key = miss->key;
3472 op->dpif_op.u.execute.key_len = miss->key_len;
3473 op->dpif_op.u.execute.packet = packet;
3476 /* Helper for handle_flow_miss_without_facet() and
3477 * handle_flow_miss_with_facet(). */
3479 handle_flow_miss_common(struct rule_dpif *rule,
3480 struct ofpbuf *packet, const struct flow *flow)
3482 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3484 ofproto->n_matches++;
3486 if (rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
3488 * Extra-special case for fail-open mode.
3490 * We are in fail-open mode and the packet matched the fail-open
3491 * rule, but we are connected to a controller too. We should send
3492 * the packet up to the controller in the hope that it will try to
3493 * set up a flow and thereby allow us to exit fail-open.
3495 * See the top-level comment in fail-open.c for more information.
3497 send_packet_in_miss(ofproto, packet, flow);
3501 /* Figures out whether a flow that missed in 'ofproto', whose details are in
3502 * 'miss', is likely to be worth tracking in detail in userspace and (usually)
3503 * installing a datapath flow. The answer is usually "yes" (a return value of
3504 * true). However, for short flows the cost of bookkeeping is much higher than
3505 * the benefits, so when the datapath holds a large number of flows we impose
3506 * some heuristics to decide which flows are likely to be worth tracking. */
3508 flow_miss_should_make_facet(struct ofproto_dpif *ofproto,
3509 struct flow_miss *miss, uint32_t hash)
3511 if (!ofproto->governor) {
3514 n_subfacets = hmap_count(&ofproto->subfacets);
3515 if (n_subfacets * 2 <= ofproto->up.flow_eviction_threshold) {
3519 ofproto->governor = governor_create(ofproto->up.name);
3522 return governor_should_install_flow(ofproto->governor, hash,
3523 list_size(&miss->packets));
3526 /* Handles 'miss', which matches 'rule', without creating a facet or subfacet
3527 * or creating any datapath flow. May add an "execute" operation to 'ops' and
3528 * increment '*n_ops'. */
3530 handle_flow_miss_without_facet(struct flow_miss *miss,
3531 struct rule_dpif *rule,
3532 struct flow_miss_op *ops, size_t *n_ops)
3534 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3535 long long int now = time_msec();
3536 struct action_xlate_ctx ctx;
3537 struct ofpbuf *packet;
3539 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3540 struct flow_miss_op *op = &ops[*n_ops];
3541 struct dpif_flow_stats stats;
3542 struct ofpbuf odp_actions;
3544 COVERAGE_INC(facet_suppress);
3546 ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub);
3548 dpif_flow_stats_extract(&miss->flow, packet, now, &stats);
3549 rule_credit_stats(rule, &stats);
3551 action_xlate_ctx_init(&ctx, ofproto, &miss->flow,
3552 &miss->initial_vals, rule, 0, packet);
3553 ctx.resubmit_stats = &stats;
3554 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len,
3557 if (odp_actions.size) {
3558 struct dpif_execute *execute = &op->dpif_op.u.execute;
3560 init_flow_miss_execute_op(miss, packet, op);
3561 execute->actions = odp_actions.data;
3562 execute->actions_len = odp_actions.size;
3563 op->garbage = ofpbuf_get_uninit_pointer(&odp_actions);
3567 ofpbuf_uninit(&odp_actions);
3572 /* Handles 'miss', which matches 'facet'. May add any required datapath
3573 * operations to 'ops', incrementing '*n_ops' for each new op.
3575 * All of the packets in 'miss' are considered to have arrived at time 'now'.
3576 * This is really important only for new facets: if we just called time_msec()
3577 * here, then the new subfacet or its packets could look (occasionally) as
3578 * though it was used some time after the facet was used. That can make a
3579 * one-packet flow look like it has a nonzero duration, which looks odd in
3580 * e.g. NetFlow statistics. */
3582 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
3584 struct flow_miss_op *ops, size_t *n_ops)
3586 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3587 enum subfacet_path want_path;
3588 struct subfacet *subfacet;
3589 struct ofpbuf *packet;
3591 subfacet = subfacet_create(facet, miss, now);
3593 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3594 struct flow_miss_op *op = &ops[*n_ops];
3595 struct dpif_flow_stats stats;
3596 struct ofpbuf odp_actions;
3598 handle_flow_miss_common(facet->rule, packet, &miss->flow);
3600 ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub);
3601 if (!subfacet->actions || subfacet->slow) {
3602 subfacet_make_actions(subfacet, packet, &odp_actions);
3605 dpif_flow_stats_extract(&facet->flow, packet, now, &stats);
3606 subfacet_update_stats(subfacet, &stats);
3608 if (subfacet->actions_len) {
3609 struct dpif_execute *execute = &op->dpif_op.u.execute;
3611 init_flow_miss_execute_op(miss, packet, op);
3612 if (!subfacet->slow) {
3613 execute->actions = subfacet->actions;
3614 execute->actions_len = subfacet->actions_len;
3615 ofpbuf_uninit(&odp_actions);
3617 execute->actions = odp_actions.data;
3618 execute->actions_len = odp_actions.size;
3619 op->garbage = ofpbuf_get_uninit_pointer(&odp_actions);
3624 ofpbuf_uninit(&odp_actions);
3628 want_path = subfacet_want_path(subfacet->slow);
3629 if (miss->upcall_type == DPIF_UC_MISS || subfacet->path != want_path) {
3630 struct flow_miss_op *op = &ops[(*n_ops)++];
3631 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
3633 subfacet->path = want_path;
3636 op->dpif_op.type = DPIF_OP_FLOW_PUT;
3637 put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
3638 put->key = miss->key;
3639 put->key_len = miss->key_len;
3640 if (want_path == SF_FAST_PATH) {
3641 put->actions = subfacet->actions;
3642 put->actions_len = subfacet->actions_len;
3644 compose_slow_path(ofproto, &facet->flow, subfacet->slow,
3645 op->stub, sizeof op->stub,
3646 &put->actions, &put->actions_len);
3652 /* Handles flow miss 'miss'. May add any required datapath operations
3653 * to 'ops', incrementing '*n_ops' for each new op. */
3655 handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops,
3658 struct ofproto_dpif *ofproto = miss->ofproto;
3659 struct facet *facet;
3663 /* The caller must ensure that miss->hmap_node.hash contains
3664 * flow_hash(miss->flow, 0). */
3665 hash = miss->hmap_node.hash;
3667 facet = facet_lookup_valid(ofproto, &miss->flow, hash);
3669 struct rule_dpif *rule = rule_dpif_lookup(ofproto, &miss->flow);
3671 if (!flow_miss_should_make_facet(ofproto, miss, hash)) {
3672 handle_flow_miss_without_facet(miss, rule, ops, n_ops);
3676 facet = facet_create(rule, &miss->flow, hash);
3681 handle_flow_miss_with_facet(miss, facet, now, ops, n_ops);
3684 static struct drop_key *
3685 drop_key_lookup(const struct dpif_backer *backer, const struct nlattr *key,
3688 struct drop_key *drop_key;
3690 HMAP_FOR_EACH_WITH_HASH (drop_key, hmap_node, hash_bytes(key, key_len, 0),
3691 &backer->drop_keys) {
3692 if (drop_key->key_len == key_len
3693 && !memcmp(drop_key->key, key, key_len)) {
3701 drop_key_clear(struct dpif_backer *backer)
3703 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3704 struct drop_key *drop_key, *next;
3706 HMAP_FOR_EACH_SAFE (drop_key, next, hmap_node, &backer->drop_keys) {
3709 error = dpif_flow_del(backer->dpif, drop_key->key, drop_key->key_len,
3711 if (error && !VLOG_DROP_WARN(&rl)) {
3712 struct ds ds = DS_EMPTY_INITIALIZER;
3713 odp_flow_key_format(drop_key->key, drop_key->key_len, &ds);
3714 VLOG_WARN("Failed to delete drop key (%s) (%s)", strerror(error),
3719 hmap_remove(&backer->drop_keys, &drop_key->hmap_node);
3720 free(drop_key->key);
3725 /* Given a datpath, packet, and flow metadata ('backer', 'packet', and 'key'
3726 * respectively), populates 'flow' with the result of odp_flow_key_to_flow().
3727 * Optionally, if nonnull, populates 'fitnessp' with the fitness of 'flow' as
3728 * returned by odp_flow_key_to_flow(). Also, optionally populates 'ofproto'
3729 * with the ofproto_dpif, and 'odp_in_port' with the datapath in_port, that
3730 * 'packet' ingressed.
3732 * If 'ofproto' is nonnull, requires 'flow''s in_port to exist. Otherwise sets
3733 * 'flow''s in_port to OFPP_NONE.
3735 * This function does post-processing on data returned from
3736 * odp_flow_key_to_flow() to help make VLAN splinters transparent to the rest
3737 * of the upcall processing logic. In particular, if the extracted in_port is
3738 * a VLAN splinter port, it replaces flow->in_port by the "real" port, sets
3739 * flow->vlan_tci correctly for the VLAN of the VLAN splinter port, and pushes
3740 * a VLAN header onto 'packet' (if it is nonnull).
3742 * Optionally, if 'initial_vals' is nonnull, sets 'initial_vals->vlan_tci'
3743 * to the VLAN TCI with which the packet was really received, that is, the
3744 * actual VLAN TCI extracted by odp_flow_key_to_flow(). (This differs from
3745 * the value returned in flow->vlan_tci only for packets received on
3746 * VLAN splinters.) Also, if received on an IP tunnel, sets
3747 * 'initial_vals->tunnel_ip_tos' to the tunnel's IP TOS.
3749 * Similarly, this function also includes some logic to help with tunnels. It
3750 * may modify 'flow' as necessary to make the tunneling implementation
3751 * transparent to the upcall processing logic.
3753 * Returns 0 if successful, ENODEV if the parsed flow has no associated ofport,
3754 * or some other positive errno if there are other problems. */
3756 ofproto_receive(const struct dpif_backer *backer, struct ofpbuf *packet,
3757 const struct nlattr *key, size_t key_len,
3758 struct flow *flow, enum odp_key_fitness *fitnessp,
3759 struct ofproto_dpif **ofproto, uint32_t *odp_in_port,
3760 struct initial_vals *initial_vals)
3762 const struct ofport_dpif *port;
3763 enum odp_key_fitness fitness;
3766 fitness = odp_flow_key_to_flow(key, key_len, flow);
3767 if (fitness == ODP_FIT_ERROR) {
3773 initial_vals->vlan_tci = flow->vlan_tci;
3774 initial_vals->tunnel_ip_tos = flow->tunnel.ip_tos;
3778 *odp_in_port = flow->in_port;
3781 if (tnl_port_should_receive(flow)) {
3782 const struct ofport *ofport = tnl_port_receive(flow);
3784 flow->in_port = OFPP_NONE;
3787 port = ofport_dpif_cast(ofport);
3789 /* We can't reproduce 'key' from 'flow'. */
3790 fitness = fitness == ODP_FIT_PERFECT ? ODP_FIT_TOO_MUCH : fitness;
3792 /* XXX: Since the tunnel module is not scoped per backer, it's
3793 * theoretically possible that we'll receive an ofport belonging to an
3794 * entirely different datapath. In practice, this can't happen because
3795 * no platforms has two separate datapaths which each support
3797 ovs_assert(ofproto_dpif_cast(port->up.ofproto)->backer == backer);
3799 port = odp_port_to_ofport(backer, flow->in_port);
3801 flow->in_port = OFPP_NONE;
3805 flow->in_port = port->up.ofp_port;
3806 if (vsp_adjust_flow(ofproto_dpif_cast(port->up.ofproto), flow)) {
3808 /* Make the packet resemble the flow, so that it gets sent to
3809 * an OpenFlow controller properly, so that it looks correct
3810 * for sFlow, and so that flow_extract() will get the correct
3811 * vlan_tci if it is called on 'packet'.
3813 * The allocated space inside 'packet' probably also contains
3814 * 'key', that is, both 'packet' and 'key' are probably part of
3815 * a struct dpif_upcall (see the large comment on that
3816 * structure definition), so pushing data on 'packet' is in
3817 * general not a good idea since it could overwrite 'key' or
3818 * free it as a side effect. However, it's OK in this special
3819 * case because we know that 'packet' is inside a Netlink
3820 * attribute: pushing 4 bytes will just overwrite the 4-byte
3821 * "struct nlattr", which is fine since we don't need that
3822 * header anymore. */
3823 eth_push_vlan(packet, flow->vlan_tci);
3825 /* We can't reproduce 'key' from 'flow'. */
3826 fitness = fitness == ODP_FIT_PERFECT ? ODP_FIT_TOO_MUCH : fitness;
3832 *ofproto = ofproto_dpif_cast(port->up.ofproto);
3837 *fitnessp = fitness;
3843 handle_miss_upcalls(struct dpif_backer *backer, struct dpif_upcall *upcalls,
3846 struct dpif_upcall *upcall;
3847 struct flow_miss *miss;
3848 struct flow_miss misses[FLOW_MISS_MAX_BATCH];
3849 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
3850 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
3860 /* Construct the to-do list.
3862 * This just amounts to extracting the flow from each packet and sticking
3863 * the packets that have the same flow in the same "flow_miss" structure so
3864 * that we can process them together. */
3867 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
3868 struct flow_miss *miss = &misses[n_misses];
3869 struct flow_miss *existing_miss;
3870 struct ofproto_dpif *ofproto;
3871 uint32_t odp_in_port;
3876 error = ofproto_receive(backer, upcall->packet, upcall->key,
3877 upcall->key_len, &flow, &miss->key_fitness,
3878 &ofproto, &odp_in_port, &miss->initial_vals);
3879 if (error == ENODEV) {
3880 struct drop_key *drop_key;
3882 /* Received packet on port for which we couldn't associate
3883 * an ofproto. This can happen if a port is removed while
3884 * traffic is being received. Print a rate-limited message
3885 * in case it happens frequently. Install a drop flow so
3886 * that future packets of the flow are inexpensively dropped
3888 VLOG_INFO_RL(&rl, "received packet on unassociated port %"PRIu32,
3891 drop_key = drop_key_lookup(backer, upcall->key, upcall->key_len);
3893 drop_key = xmalloc(sizeof *drop_key);
3894 drop_key->key = xmemdup(upcall->key, upcall->key_len);
3895 drop_key->key_len = upcall->key_len;
3897 hmap_insert(&backer->drop_keys, &drop_key->hmap_node,
3898 hash_bytes(drop_key->key, drop_key->key_len, 0));
3899 dpif_flow_put(backer->dpif, DPIF_FP_CREATE | DPIF_FP_MODIFY,
3900 drop_key->key, drop_key->key_len, NULL, 0, NULL);
3908 ofproto->n_missed++;
3909 flow_extract(upcall->packet, flow.skb_priority, flow.skb_mark,
3910 &flow.tunnel, flow.in_port, &miss->flow);
3912 /* Add other packets to a to-do list. */
3913 hash = flow_hash(&miss->flow, 0);
3914 existing_miss = flow_miss_find(&todo, ofproto, &miss->flow, hash);
3915 if (!existing_miss) {
3916 hmap_insert(&todo, &miss->hmap_node, hash);
3917 miss->ofproto = ofproto;
3918 miss->key = upcall->key;
3919 miss->key_len = upcall->key_len;
3920 miss->upcall_type = upcall->type;
3921 miss->odp_in_port = odp_in_port;
3922 list_init(&miss->packets);
3926 miss = existing_miss;
3928 list_push_back(&miss->packets, &upcall->packet->list_node);
3931 /* Process each element in the to-do list, constructing the set of
3932 * operations to batch. */
3934 HMAP_FOR_EACH (miss, hmap_node, &todo) {
3935 handle_flow_miss(miss, flow_miss_ops, &n_ops);
3937 ovs_assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
3939 /* Execute batch. */
3940 for (i = 0; i < n_ops; i++) {
3941 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
3943 dpif_operate(backer->dpif, dpif_ops, n_ops);
3946 for (i = 0; i < n_ops; i++) {
3947 free(flow_miss_ops[i].garbage);
3949 hmap_destroy(&todo);
3952 static enum { SFLOW_UPCALL, MISS_UPCALL, BAD_UPCALL }
3953 classify_upcall(const struct dpif_upcall *upcall)
3955 union user_action_cookie cookie;
3957 /* First look at the upcall type. */
3958 switch (upcall->type) {
3959 case DPIF_UC_ACTION:
3965 case DPIF_N_UC_TYPES:
3967 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
3971 /* "action" upcalls need a closer look. */
3972 if (!upcall->userdata) {
3973 VLOG_WARN_RL(&rl, "action upcall missing cookie");
3976 if (nl_attr_get_size(upcall->userdata) != sizeof(cookie)) {
3977 VLOG_WARN_RL(&rl, "action upcall cookie has unexpected size %zu",
3978 nl_attr_get_size(upcall->userdata));
3981 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof(cookie));
3982 switch (cookie.type) {
3983 case USER_ACTION_COOKIE_SFLOW:
3984 return SFLOW_UPCALL;
3986 case USER_ACTION_COOKIE_SLOW_PATH:
3989 case USER_ACTION_COOKIE_UNSPEC:
3991 VLOG_WARN_RL(&rl, "invalid user cookie : 0x%"PRIx64,
3992 nl_attr_get_u64(upcall->userdata));
3998 handle_sflow_upcall(struct dpif_backer *backer,
3999 const struct dpif_upcall *upcall)
4001 struct ofproto_dpif *ofproto;
4002 union user_action_cookie cookie;
4004 uint32_t odp_in_port;
4006 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
4007 &flow, NULL, &ofproto, &odp_in_port, NULL)
4008 || !ofproto->sflow) {
4012 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof(cookie));
4013 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
4014 odp_in_port, &cookie);
4018 handle_upcalls(struct dpif_backer *backer, unsigned int max_batch)
4020 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
4021 struct ofpbuf miss_bufs[FLOW_MISS_MAX_BATCH];
4022 uint64_t miss_buf_stubs[FLOW_MISS_MAX_BATCH][4096 / 8];
4027 ovs_assert(max_batch <= FLOW_MISS_MAX_BATCH);
4030 for (n_processed = 0; n_processed < max_batch; n_processed++) {
4031 struct dpif_upcall *upcall = &misses[n_misses];
4032 struct ofpbuf *buf = &miss_bufs[n_misses];
4035 ofpbuf_use_stub(buf, miss_buf_stubs[n_misses],
4036 sizeof miss_buf_stubs[n_misses]);
4037 error = dpif_recv(backer->dpif, upcall, buf);
4043 switch (classify_upcall(upcall)) {
4045 /* Handle it later. */
4050 handle_sflow_upcall(backer, upcall);
4060 /* Handle deferred MISS_UPCALL processing. */
4061 handle_miss_upcalls(backer, misses, n_misses);
4062 for (i = 0; i < n_misses; i++) {
4063 ofpbuf_uninit(&miss_bufs[i]);
4069 /* Flow expiration. */
4071 static int subfacet_max_idle(const struct ofproto_dpif *);
4072 static void update_stats(struct dpif_backer *);
4073 static void rule_expire(struct rule_dpif *);
4074 static void expire_subfacets(struct ofproto_dpif *, int dp_max_idle);
4076 /* This function is called periodically by run(). Its job is to collect
4077 * updates for the flows that have been installed into the datapath, most
4078 * importantly when they last were used, and then use that information to
4079 * expire flows that have not been used recently.
4081 * Returns the number of milliseconds after which it should be called again. */
4083 expire(struct dpif_backer *backer)
4085 struct ofproto_dpif *ofproto;
4086 int max_idle = INT32_MAX;
4088 /* Periodically clear out the drop keys in an effort to keep them
4089 * relatively few. */
4090 drop_key_clear(backer);
4092 /* Update stats for each flow in the backer. */
4093 update_stats(backer);
4095 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4096 struct rule *rule, *next_rule;
4099 if (ofproto->backer != backer) {
4103 /* Keep track of the max number of flows per ofproto_dpif. */
4104 update_max_subfacet_count(ofproto);
4106 /* Expire subfacets that have been idle too long. */
4107 dp_max_idle = subfacet_max_idle(ofproto);
4108 expire_subfacets(ofproto, dp_max_idle);
4110 max_idle = MIN(max_idle, dp_max_idle);
4112 /* Expire OpenFlow flows whose idle_timeout or hard_timeout
4114 LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
4115 &ofproto->up.expirable) {
4116 rule_expire(rule_dpif_cast(rule));
4119 /* All outstanding data in existing flows has been accounted, so it's a
4120 * good time to do bond rebalancing. */
4121 if (ofproto->has_bonded_bundles) {
4122 struct ofbundle *bundle;
4124 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
4126 bond_rebalance(bundle->bond, &backer->revalidate_set);
4132 return MIN(max_idle, 1000);
4135 /* Updates flow table statistics given that the datapath just reported 'stats'
4136 * as 'subfacet''s statistics. */
4138 update_subfacet_stats(struct subfacet *subfacet,
4139 const struct dpif_flow_stats *stats)
4141 struct facet *facet = subfacet->facet;
4143 if (stats->n_packets >= subfacet->dp_packet_count) {
4144 uint64_t extra = stats->n_packets - subfacet->dp_packet_count;
4145 facet->packet_count += extra;
4147 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
4150 if (stats->n_bytes >= subfacet->dp_byte_count) {
4151 facet->byte_count += stats->n_bytes - subfacet->dp_byte_count;
4153 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
4156 subfacet->dp_packet_count = stats->n_packets;
4157 subfacet->dp_byte_count = stats->n_bytes;
4159 facet->tcp_flags |= stats->tcp_flags;
4161 subfacet_update_time(subfacet, stats->used);
4162 if (facet->accounted_bytes < facet->byte_count) {
4164 facet_account(facet);
4165 facet->accounted_bytes = facet->byte_count;
4167 facet_push_stats(facet);
4170 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
4171 * about, or a flow that shouldn't be installed but was anyway. Delete it. */
4173 delete_unexpected_flow(struct ofproto_dpif *ofproto,
4174 const struct nlattr *key, size_t key_len)
4176 if (!VLOG_DROP_WARN(&rl)) {
4180 odp_flow_key_format(key, key_len, &s);
4181 VLOG_WARN("unexpected flow on %s: %s", ofproto->up.name, ds_cstr(&s));
4185 COVERAGE_INC(facet_unexpected);
4186 dpif_flow_del(ofproto->backer->dpif, key, key_len, NULL);
4189 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
4191 * This function also pushes statistics updates to rules which each facet
4192 * resubmits into. Generally these statistics will be accurate. However, if a
4193 * facet changes the rule it resubmits into at some time in between
4194 * update_stats() runs, it is possible that statistics accrued to the
4195 * old rule will be incorrectly attributed to the new rule. This could be
4196 * avoided by calling update_stats() whenever rules are created or
4197 * deleted. However, the performance impact of making so many calls to the
4198 * datapath do not justify the benefit of having perfectly accurate statistics.
4200 * In addition, this function maintains per ofproto flow hit counts. The patch
4201 * port is not treated specially. e.g. A packet ingress from br0 patched into
4202 * br1 will increase the hit count of br0 by 1, however, does not affect
4203 * the hit or miss counts of br1.
4206 update_stats(struct dpif_backer *backer)
4208 const struct dpif_flow_stats *stats;
4209 struct dpif_flow_dump dump;
4210 const struct nlattr *key;
4213 dpif_flow_dump_start(&dump, backer->dpif);
4214 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
4216 struct subfacet *subfacet;
4217 struct ofproto_dpif *ofproto;
4218 struct ofport_dpif *ofport;
4221 if (ofproto_receive(backer, NULL, key, key_len, &flow, NULL, &ofproto,
4226 ofproto->total_subfacet_count += hmap_count(&ofproto->subfacets);
4227 ofproto->n_update_stats++;
4228 update_moving_averages(ofproto);
4230 ofport = get_ofp_port(ofproto, flow.in_port);
4231 if (ofport && ofport->tnl_port) {
4232 netdev_vport_inc_rx(ofport->up.netdev, stats);
4235 key_hash = odp_flow_key_hash(key, key_len);
4236 subfacet = subfacet_find(ofproto, key, key_len, key_hash);
4237 switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
4239 /* Update ofproto_dpif's hit count. */
4240 if (stats->n_packets > subfacet->dp_packet_count) {
4241 uint64_t delta = stats->n_packets - subfacet->dp_packet_count;
4242 dpif_stats_update_hit_count(ofproto, delta);
4245 update_subfacet_stats(subfacet, stats);
4249 /* Stats are updated per-packet. */
4252 case SF_NOT_INSTALLED:
4254 delete_unexpected_flow(ofproto, key, key_len);
4258 dpif_flow_dump_done(&dump);
4261 /* Calculates and returns the number of milliseconds of idle time after which
4262 * subfacets should expire from the datapath. When a subfacet expires, we fold
4263 * its statistics into its facet, and when a facet's last subfacet expires, we
4264 * fold its statistic into its rule. */
4266 subfacet_max_idle(const struct ofproto_dpif *ofproto)
4269 * Idle time histogram.
4271 * Most of the time a switch has a relatively small number of subfacets.
4272 * When this is the case we might as well keep statistics for all of them
4273 * in userspace and to cache them in the kernel datapath for performance as
4276 * As the number of subfacets increases, the memory required to maintain
4277 * statistics about them in userspace and in the kernel becomes
4278 * significant. However, with a large number of subfacets it is likely
4279 * that only a few of them are "heavy hitters" that consume a large amount
4280 * of bandwidth. At this point, only heavy hitters are worth caching in
4281 * the kernel and maintaining in userspaces; other subfacets we can
4284 * The technique used to compute the idle time is to build a histogram with
4285 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
4286 * that is installed in the kernel gets dropped in the appropriate bucket.
4287 * After the histogram has been built, we compute the cutoff so that only
4288 * the most-recently-used 1% of subfacets (but at least
4289 * ofproto->up.flow_eviction_threshold flows) are kept cached. At least
4290 * the most-recently-used bucket of subfacets is kept, so actually an
4291 * arbitrary number of subfacets can be kept in any given expiration run
4292 * (though the next run will delete most of those unless they receive
4295 * This requires a second pass through the subfacets, in addition to the
4296 * pass made by update_stats(), because the former function never looks at
4297 * uninstallable subfacets.
4299 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
4300 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
4301 int buckets[N_BUCKETS] = { 0 };
4302 int total, subtotal, bucket;
4303 struct subfacet *subfacet;
4307 total = hmap_count(&ofproto->subfacets);
4308 if (total <= ofproto->up.flow_eviction_threshold) {
4309 return N_BUCKETS * BUCKET_WIDTH;
4312 /* Build histogram. */
4314 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
4315 long long int idle = now - subfacet->used;
4316 int bucket = (idle <= 0 ? 0
4317 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
4318 : (unsigned int) idle / BUCKET_WIDTH);
4322 /* Find the first bucket whose flows should be expired. */
4323 subtotal = bucket = 0;
4325 subtotal += buckets[bucket++];
4326 } while (bucket < N_BUCKETS &&
4327 subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100));
4329 if (VLOG_IS_DBG_ENABLED()) {
4333 ds_put_cstr(&s, "keep");
4334 for (i = 0; i < N_BUCKETS; i++) {
4336 ds_put_cstr(&s, ", drop");
4339 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
4342 VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s));
4346 return bucket * BUCKET_WIDTH;
4350 expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle)
4352 /* Cutoff time for most flows. */
4353 long long int normal_cutoff = time_msec() - dp_max_idle;
4355 /* We really want to keep flows for special protocols around, so use a more
4356 * conservative cutoff. */
4357 long long int special_cutoff = time_msec() - 10000;
4359 struct subfacet *subfacet, *next_subfacet;
4360 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
4364 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
4365 &ofproto->subfacets) {
4366 long long int cutoff;
4368 cutoff = (subfacet->slow & (SLOW_CFM | SLOW_LACP | SLOW_STP)
4371 if (subfacet->used < cutoff) {
4372 if (subfacet->path != SF_NOT_INSTALLED) {
4373 batch[n_batch++] = subfacet;
4374 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
4375 subfacet_destroy_batch(ofproto, batch, n_batch);
4379 subfacet_destroy(subfacet);
4385 subfacet_destroy_batch(ofproto, batch, n_batch);
4389 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
4390 * then delete it entirely. */
4392 rule_expire(struct rule_dpif *rule)
4394 struct facet *facet, *next_facet;
4398 if (rule->up.pending) {
4399 /* We'll have to expire it later. */
4403 /* Has 'rule' expired? */
4405 if (rule->up.hard_timeout
4406 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
4407 reason = OFPRR_HARD_TIMEOUT;
4408 } else if (rule->up.idle_timeout
4409 && now > rule->up.used + rule->up.idle_timeout * 1000) {
4410 reason = OFPRR_IDLE_TIMEOUT;
4415 COVERAGE_INC(ofproto_dpif_expired);
4417 /* Update stats. (This is a no-op if the rule expired due to an idle
4418 * timeout, because that only happens when the rule has no facets left.) */
4419 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4420 facet_remove(facet);
4423 /* Get rid of the rule. */
4424 ofproto_rule_expire(&rule->up, reason);
4429 /* Creates and returns a new facet owned by 'rule', given a 'flow'.
4431 * The caller must already have determined that no facet with an identical
4432 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
4433 * the ofproto's classifier table.
4435 * 'hash' must be the return value of flow_hash(flow, 0).
4437 * The facet will initially have no subfacets. The caller should create (at
4438 * least) one subfacet with subfacet_create(). */
4439 static struct facet *
4440 facet_create(struct rule_dpif *rule, const struct flow *flow, uint32_t hash)
4442 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4443 struct facet *facet;
4445 facet = xzalloc(sizeof *facet);
4446 facet->used = time_msec();
4447 hmap_insert(&ofproto->facets, &facet->hmap_node, hash);
4448 list_push_back(&rule->facets, &facet->list_node);
4450 facet->flow = *flow;
4451 list_init(&facet->subfacets);
4452 netflow_flow_init(&facet->nf_flow);
4453 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
4455 facet->learn_rl = time_msec() + 500;
4461 facet_free(struct facet *facet)
4466 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
4467 * 'packet', which arrived on 'in_port'. */
4469 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
4470 const struct nlattr *odp_actions, size_t actions_len,
4471 struct ofpbuf *packet)
4473 struct odputil_keybuf keybuf;
4477 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4478 odp_flow_key_from_flow(&key, flow,
4479 ofp_port_to_odp_port(ofproto, flow->in_port));
4481 error = dpif_execute(ofproto->backer->dpif, key.data, key.size,
4482 odp_actions, actions_len, packet);
4486 /* Remove 'facet' from 'ofproto' and free up the associated memory:
4488 * - If 'facet' was installed in the datapath, uninstalls it and updates its
4489 * rule's statistics, via subfacet_uninstall().
4491 * - Removes 'facet' from its rule and from ofproto->facets.
4494 facet_remove(struct facet *facet)
4496 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4497 struct subfacet *subfacet, *next_subfacet;
4499 ovs_assert(!list_is_empty(&facet->subfacets));
4501 /* First uninstall all of the subfacets to get final statistics. */
4502 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4503 subfacet_uninstall(subfacet);
4506 /* Flush the final stats to the rule.
4508 * This might require us to have at least one subfacet around so that we
4509 * can use its actions for accounting in facet_account(), which is why we
4510 * have uninstalled but not yet destroyed the subfacets. */
4511 facet_flush_stats(facet);
4513 /* Now we're really all done so destroy everything. */
4514 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
4515 &facet->subfacets) {
4516 subfacet_destroy__(subfacet);
4518 hmap_remove(&ofproto->facets, &facet->hmap_node);
4519 list_remove(&facet->list_node);
4523 /* Feed information from 'facet' back into the learning table to keep it in
4524 * sync with what is actually flowing through the datapath. */
4526 facet_learn(struct facet *facet)
4528 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4529 struct subfacet *subfacet= CONTAINER_OF(list_front(&facet->subfacets),
4530 struct subfacet, list_node);
4531 struct action_xlate_ctx ctx;
4533 if (time_msec() < facet->learn_rl) {
4537 facet->learn_rl = time_msec() + 500;
4539 if (!facet->has_learn
4540 && !facet->has_normal
4541 && (!facet->has_fin_timeout
4542 || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
4546 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
4547 &subfacet->initial_vals,
4548 facet->rule, facet->tcp_flags, NULL);
4549 ctx.may_learn = true;
4550 xlate_actions_for_side_effects(&ctx, facet->rule->up.ofpacts,
4551 facet->rule->up.ofpacts_len);
4555 facet_account(struct facet *facet)
4557 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4558 struct subfacet *subfacet = facet_get_subfacet(facet);
4559 const struct nlattr *a;
4564 if (!facet->has_normal || !ofproto->has_bonded_bundles) {
4567 n_bytes = facet->byte_count - facet->accounted_bytes;
4569 /* This loop feeds byte counters to bond_account() for rebalancing to use
4570 * as a basis. We also need to track the actual VLAN on which the packet
4571 * is going to be sent to ensure that it matches the one passed to
4572 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
4575 * We use the actions from an arbitrary subfacet because they should all
4576 * be equally valid for our purpose. */
4577 vlan_tci = facet->flow.vlan_tci;
4578 NL_ATTR_FOR_EACH_UNSAFE (a, left,
4579 subfacet->actions, subfacet->actions_len) {
4580 const struct ovs_action_push_vlan *vlan;
4581 struct ofport_dpif *port;
4583 switch (nl_attr_type(a)) {
4584 case OVS_ACTION_ATTR_OUTPUT:
4585 port = get_odp_port(ofproto, nl_attr_get_u32(a));
4586 if (port && port->bundle && port->bundle->bond) {
4587 bond_account(port->bundle->bond, &facet->flow,
4588 vlan_tci_to_vid(vlan_tci), n_bytes);
4592 case OVS_ACTION_ATTR_POP_VLAN:
4593 vlan_tci = htons(0);
4596 case OVS_ACTION_ATTR_PUSH_VLAN:
4597 vlan = nl_attr_get(a);
4598 vlan_tci = vlan->vlan_tci;
4604 /* Returns true if the only action for 'facet' is to send to the controller.
4605 * (We don't report NetFlow expiration messages for such facets because they
4606 * are just part of the control logic for the network, not real traffic). */
4608 facet_is_controller_flow(struct facet *facet)
4611 const struct rule *rule = &facet->rule->up;
4612 const struct ofpact *ofpacts = rule->ofpacts;
4613 size_t ofpacts_len = rule->ofpacts_len;
4615 if (ofpacts_len > 0 &&
4616 ofpacts->type == OFPACT_CONTROLLER &&
4617 ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len)) {
4624 /* Folds all of 'facet''s statistics into its rule. Also updates the
4625 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
4626 * 'facet''s statistics in the datapath should have been zeroed and folded into
4627 * its packet and byte counts before this function is called. */
4629 facet_flush_stats(struct facet *facet)
4631 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4632 struct subfacet *subfacet;
4634 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4635 ovs_assert(!subfacet->dp_byte_count);
4636 ovs_assert(!subfacet->dp_packet_count);
4639 facet_push_stats(facet);
4640 if (facet->accounted_bytes < facet->byte_count) {
4641 facet_account(facet);
4642 facet->accounted_bytes = facet->byte_count;
4645 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
4646 struct ofexpired expired;
4647 expired.flow = facet->flow;
4648 expired.packet_count = facet->packet_count;
4649 expired.byte_count = facet->byte_count;
4650 expired.used = facet->used;
4651 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4654 facet->rule->packet_count += facet->packet_count;
4655 facet->rule->byte_count += facet->byte_count;
4657 /* Reset counters to prevent double counting if 'facet' ever gets
4659 facet_reset_counters(facet);
4661 netflow_flow_clear(&facet->nf_flow);
4662 facet->tcp_flags = 0;
4665 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
4666 * Returns it if found, otherwise a null pointer.
4668 * 'hash' must be the return value of flow_hash(flow, 0).
4670 * The returned facet might need revalidation; use facet_lookup_valid()
4671 * instead if that is important. */
4672 static struct facet *
4673 facet_find(struct ofproto_dpif *ofproto,
4674 const struct flow *flow, uint32_t hash)
4676 struct facet *facet;
4678 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, hash, &ofproto->facets) {
4679 if (flow_equal(flow, &facet->flow)) {
4687 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
4688 * Returns it if found, otherwise a null pointer.
4690 * 'hash' must be the return value of flow_hash(flow, 0).
4692 * The returned facet is guaranteed to be valid. */
4693 static struct facet *
4694 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow,
4697 struct facet *facet;
4699 facet = facet_find(ofproto, flow, hash);
4701 && (ofproto->backer->need_revalidate
4702 || tag_set_intersects(&ofproto->backer->revalidate_set,
4704 facet_revalidate(facet);
4706 /* facet_revalidate() may have destroyed 'facet'. */
4707 facet = facet_find(ofproto, flow, hash);
4713 /* Return a subfacet from 'facet'. A facet consists of one or more
4714 * subfacets, and this function returns one of them. */
4715 static struct subfacet *facet_get_subfacet(struct facet *facet)
4717 return CONTAINER_OF(list_front(&facet->subfacets), struct subfacet,
4722 subfacet_path_to_string(enum subfacet_path path)
4725 case SF_NOT_INSTALLED:
4726 return "not installed";
4728 return "in fast path";
4730 return "in slow path";
4736 /* Returns the path in which a subfacet should be installed if its 'slow'
4737 * member has the specified value. */
4738 static enum subfacet_path
4739 subfacet_want_path(enum slow_path_reason slow)
4741 return slow ? SF_SLOW_PATH : SF_FAST_PATH;
4744 /* Returns true if 'subfacet' needs to have its datapath flow updated,
4745 * supposing that its actions have been recalculated as 'want_actions' and that
4746 * 'slow' is nonzero iff 'subfacet' should be in the slow path. */
4748 subfacet_should_install(struct subfacet *subfacet, enum slow_path_reason slow,
4749 const struct ofpbuf *want_actions)
4751 enum subfacet_path want_path = subfacet_want_path(slow);
4752 return (want_path != subfacet->path
4753 || (want_path == SF_FAST_PATH
4754 && (subfacet->actions_len != want_actions->size
4755 || memcmp(subfacet->actions, want_actions->data,
4756 subfacet->actions_len))));
4760 facet_check_consistency(struct facet *facet)
4762 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
4764 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4766 uint64_t odp_actions_stub[1024 / 8];
4767 struct ofpbuf odp_actions;
4769 struct rule_dpif *rule;
4770 struct subfacet *subfacet;
4771 bool may_log = false;
4774 /* Check the rule for consistency. */
4775 rule = rule_dpif_lookup(ofproto, &facet->flow);
4776 ok = rule == facet->rule;
4778 may_log = !VLOG_DROP_WARN(&rl);
4783 flow_format(&s, &facet->flow);
4784 ds_put_format(&s, ": facet associated with wrong rule (was "
4785 "table=%"PRIu8",", facet->rule->up.table_id);
4786 cls_rule_format(&facet->rule->up.cr, &s);
4787 ds_put_format(&s, ") (should have been table=%"PRIu8",",
4789 cls_rule_format(&rule->up.cr, &s);
4790 ds_put_char(&s, ')');
4792 VLOG_WARN("%s", ds_cstr(&s));
4797 /* Check the datapath actions for consistency. */
4798 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
4799 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4800 enum subfacet_path want_path;
4801 struct action_xlate_ctx ctx;
4804 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
4805 &subfacet->initial_vals, rule, 0, NULL);
4806 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len,
4809 if (subfacet->path == SF_NOT_INSTALLED) {
4810 /* This only happens if the datapath reported an error when we
4811 * tried to install the flow. Don't flag another error here. */
4815 want_path = subfacet_want_path(subfacet->slow);
4816 if (want_path == SF_SLOW_PATH && subfacet->path == SF_SLOW_PATH) {
4817 /* The actions for slow-path flows may legitimately vary from one
4818 * packet to the next. We're done. */
4822 if (!subfacet_should_install(subfacet, subfacet->slow, &odp_actions)) {
4826 /* Inconsistency! */
4828 may_log = !VLOG_DROP_WARN(&rl);
4832 /* Rate-limited, skip reporting. */
4837 odp_flow_key_format(subfacet->key, subfacet->key_len, &s);
4839 ds_put_cstr(&s, ": inconsistency in subfacet");
4840 if (want_path != subfacet->path) {
4841 enum odp_key_fitness fitness = subfacet->key_fitness;
4843 ds_put_format(&s, " (%s, fitness=%s)",
4844 subfacet_path_to_string(subfacet->path),
4845 odp_key_fitness_to_string(fitness));
4846 ds_put_format(&s, " (should have been %s)",
4847 subfacet_path_to_string(want_path));
4848 } else if (want_path == SF_FAST_PATH) {
4849 ds_put_cstr(&s, " (actions were: ");
4850 format_odp_actions(&s, subfacet->actions,
4851 subfacet->actions_len);
4852 ds_put_cstr(&s, ") (correct actions: ");
4853 format_odp_actions(&s, odp_actions.data, odp_actions.size);
4854 ds_put_char(&s, ')');
4856 ds_put_cstr(&s, " (actions: ");
4857 format_odp_actions(&s, subfacet->actions,
4858 subfacet->actions_len);
4859 ds_put_char(&s, ')');
4861 VLOG_WARN("%s", ds_cstr(&s));
4864 ofpbuf_uninit(&odp_actions);
4869 /* Re-searches the classifier for 'facet':
4871 * - If the rule found is different from 'facet''s current rule, moves
4872 * 'facet' to the new rule and recompiles its actions.
4874 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
4875 * where it is and recompiles its actions anyway.
4877 * - If any of 'facet''s subfacets correspond to a new flow according to
4878 * ofproto_receive(), 'facet' is removed. */
4880 facet_revalidate(struct facet *facet)
4882 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4884 struct nlattr *odp_actions;
4887 struct actions *new_actions;
4889 struct action_xlate_ctx ctx;
4890 uint64_t odp_actions_stub[1024 / 8];
4891 struct ofpbuf odp_actions;
4893 struct rule_dpif *new_rule;
4894 struct subfacet *subfacet;
4897 COVERAGE_INC(facet_revalidate);
4899 /* Check that child subfacets still correspond to this facet. Tunnel
4900 * configuration changes could cause a subfacet's OpenFlow in_port to
4902 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4903 struct ofproto_dpif *recv_ofproto;
4904 struct flow recv_flow;
4907 error = ofproto_receive(ofproto->backer, NULL, subfacet->key,
4908 subfacet->key_len, &recv_flow, NULL,
4909 &recv_ofproto, NULL, NULL);
4911 || recv_ofproto != ofproto
4912 || memcmp(&recv_flow, &facet->flow, sizeof recv_flow)) {
4913 facet_remove(facet);
4918 new_rule = rule_dpif_lookup(ofproto, &facet->flow);
4920 /* Calculate new datapath actions.
4922 * We do not modify any 'facet' state yet, because we might need to, e.g.,
4923 * emit a NetFlow expiration and, if so, we need to have the old state
4924 * around to properly compose it. */
4926 /* If the datapath actions changed or the installability changed,
4927 * then we need to talk to the datapath. */
4930 memset(&ctx, 0, sizeof ctx);
4931 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
4932 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4933 enum slow_path_reason slow;
4935 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
4936 &subfacet->initial_vals, new_rule, 0, NULL);
4937 xlate_actions(&ctx, new_rule->up.ofpacts, new_rule->up.ofpacts_len,
4940 slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
4941 if (subfacet_should_install(subfacet, slow, &odp_actions)) {
4942 struct dpif_flow_stats stats;
4944 subfacet_install(subfacet,
4945 odp_actions.data, odp_actions.size, &stats, slow);
4946 subfacet_update_stats(subfacet, &stats);
4949 new_actions = xcalloc(list_size(&facet->subfacets),
4950 sizeof *new_actions);
4952 new_actions[i].odp_actions = xmemdup(odp_actions.data,
4954 new_actions[i].actions_len = odp_actions.size;
4959 ofpbuf_uninit(&odp_actions);
4962 facet_flush_stats(facet);
4965 /* Update 'facet' now that we've taken care of all the old state. */
4966 facet->tags = ctx.tags;
4967 facet->nf_flow.output_iface = ctx.nf_output_iface;
4968 facet->has_learn = ctx.has_learn;
4969 facet->has_normal = ctx.has_normal;
4970 facet->has_fin_timeout = ctx.has_fin_timeout;
4971 facet->mirrors = ctx.mirrors;
4974 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4975 subfacet->slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
4977 if (new_actions && new_actions[i].odp_actions) {
4978 free(subfacet->actions);
4979 subfacet->actions = new_actions[i].odp_actions;
4980 subfacet->actions_len = new_actions[i].actions_len;
4986 if (facet->rule != new_rule) {
4987 COVERAGE_INC(facet_changed_rule);
4988 list_remove(&facet->list_node);
4989 list_push_back(&new_rule->facets, &facet->list_node);
4990 facet->rule = new_rule;
4991 facet->used = new_rule->up.created;
4992 facet->prev_used = facet->used;
4996 /* Updates 'facet''s used time. Caller is responsible for calling
4997 * facet_push_stats() to update the flows which 'facet' resubmits into. */
4999 facet_update_time(struct facet *facet, long long int used)
5001 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5002 if (used > facet->used) {
5004 ofproto_rule_update_used(&facet->rule->up, used);
5005 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
5010 facet_reset_counters(struct facet *facet)
5012 facet->packet_count = 0;
5013 facet->byte_count = 0;
5014 facet->prev_packet_count = 0;
5015 facet->prev_byte_count = 0;
5016 facet->accounted_bytes = 0;
5020 facet_push_stats(struct facet *facet)
5022 struct dpif_flow_stats stats;
5024 ovs_assert(facet->packet_count >= facet->prev_packet_count);
5025 ovs_assert(facet->byte_count >= facet->prev_byte_count);
5026 ovs_assert(facet->used >= facet->prev_used);
5028 stats.n_packets = facet->packet_count - facet->prev_packet_count;
5029 stats.n_bytes = facet->byte_count - facet->prev_byte_count;
5030 stats.used = facet->used;
5031 stats.tcp_flags = 0;
5033 if (stats.n_packets || stats.n_bytes || facet->used > facet->prev_used) {
5034 facet->prev_packet_count = facet->packet_count;
5035 facet->prev_byte_count = facet->byte_count;
5036 facet->prev_used = facet->used;
5038 flow_push_stats(facet, &stats);
5040 update_mirror_stats(ofproto_dpif_cast(facet->rule->up.ofproto),
5041 facet->mirrors, stats.n_packets, stats.n_bytes);
5046 push_all_stats(void)
5048 static long long int rl = LLONG_MIN;
5049 struct ofproto_dpif *ofproto;
5051 if (time_msec() < rl) {
5055 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5056 struct facet *facet;
5058 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
5059 facet_push_stats(facet);
5063 rl = time_msec() + 100;
5067 rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats)
5069 rule->packet_count += stats->n_packets;
5070 rule->byte_count += stats->n_bytes;
5071 ofproto_rule_update_used(&rule->up, stats->used);
5074 /* Pushes flow statistics to the rules which 'facet->flow' resubmits
5075 * into given 'facet->rule''s actions and mirrors. */
5077 flow_push_stats(struct facet *facet, const struct dpif_flow_stats *stats)
5079 struct rule_dpif *rule = facet->rule;
5080 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5081 struct subfacet *subfacet = facet_get_subfacet(facet);
5082 struct action_xlate_ctx ctx;
5084 ofproto_rule_update_used(&rule->up, stats->used);
5086 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
5087 &subfacet->initial_vals, rule, 0, NULL);
5088 ctx.resubmit_stats = stats;
5089 xlate_actions_for_side_effects(&ctx, rule->up.ofpacts,
5090 rule->up.ofpacts_len);
5095 static struct subfacet *
5096 subfacet_find(struct ofproto_dpif *ofproto,
5097 const struct nlattr *key, size_t key_len, uint32_t key_hash)
5099 struct subfacet *subfacet;
5101 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
5102 &ofproto->subfacets) {
5103 if (subfacet->key_len == key_len
5104 && !memcmp(key, subfacet->key, key_len)) {
5112 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
5113 * 'key_fitness', 'key', and 'key_len' members in 'miss'. Returns the
5114 * existing subfacet if there is one, otherwise creates and returns a
5117 * If the returned subfacet is new, then subfacet->actions will be NULL, in
5118 * which case the caller must populate the actions with
5119 * subfacet_make_actions(). */
5120 static struct subfacet *
5121 subfacet_create(struct facet *facet, struct flow_miss *miss,
5124 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5125 enum odp_key_fitness key_fitness = miss->key_fitness;
5126 const struct nlattr *key = miss->key;
5127 size_t key_len = miss->key_len;
5129 struct subfacet *subfacet;
5131 key_hash = odp_flow_key_hash(key, key_len);
5133 if (list_is_empty(&facet->subfacets)) {
5134 subfacet = &facet->one_subfacet;
5136 subfacet = subfacet_find(ofproto, key, key_len, key_hash);
5138 if (subfacet->facet == facet) {
5142 /* This shouldn't happen. */
5143 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
5144 subfacet_destroy(subfacet);
5147 subfacet = xmalloc(sizeof *subfacet);
5150 hmap_insert(&ofproto->subfacets, &subfacet->hmap_node, key_hash);
5151 list_push_back(&facet->subfacets, &subfacet->list_node);
5152 subfacet->facet = facet;
5153 subfacet->key_fitness = key_fitness;
5154 subfacet->key = xmemdup(key, key_len);
5155 subfacet->key_len = key_len;
5156 subfacet->used = now;
5157 subfacet->created = now;
5158 subfacet->dp_packet_count = 0;
5159 subfacet->dp_byte_count = 0;
5160 subfacet->actions_len = 0;
5161 subfacet->actions = NULL;
5162 subfacet->slow = (subfacet->key_fitness == ODP_FIT_TOO_LITTLE
5165 subfacet->path = SF_NOT_INSTALLED;
5166 subfacet->initial_vals = miss->initial_vals;
5167 subfacet->odp_in_port = miss->odp_in_port;
5169 ofproto->subfacet_add_count++;
5173 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
5174 * its facet within 'ofproto', and frees it. */
5176 subfacet_destroy__(struct subfacet *subfacet)
5178 struct facet *facet = subfacet->facet;
5179 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5181 /* Update ofproto stats before uninstall the subfacet. */
5182 ofproto->subfacet_del_count++;
5183 ofproto->total_subfacet_life_span += (time_msec() - subfacet->created);
5185 subfacet_uninstall(subfacet);
5186 hmap_remove(&ofproto->subfacets, &subfacet->hmap_node);
5187 list_remove(&subfacet->list_node);
5188 free(subfacet->key);
5189 free(subfacet->actions);
5190 if (subfacet != &facet->one_subfacet) {
5195 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
5196 * last remaining subfacet in its facet destroys the facet too. */
5198 subfacet_destroy(struct subfacet *subfacet)
5200 struct facet *facet = subfacet->facet;
5202 if (list_is_singleton(&facet->subfacets)) {
5203 /* facet_remove() needs at least one subfacet (it will remove it). */
5204 facet_remove(facet);
5206 subfacet_destroy__(subfacet);
5211 subfacet_destroy_batch(struct ofproto_dpif *ofproto,
5212 struct subfacet **subfacets, int n)
5214 struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH];
5215 struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH];
5216 struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH];
5219 for (i = 0; i < n; i++) {
5220 ops[i].type = DPIF_OP_FLOW_DEL;
5221 ops[i].u.flow_del.key = subfacets[i]->key;
5222 ops[i].u.flow_del.key_len = subfacets[i]->key_len;
5223 ops[i].u.flow_del.stats = &stats[i];
5227 dpif_operate(ofproto->backer->dpif, opsp, n);
5228 for (i = 0; i < n; i++) {
5229 subfacet_reset_dp_stats(subfacets[i], &stats[i]);
5230 subfacets[i]->path = SF_NOT_INSTALLED;
5231 subfacet_destroy(subfacets[i]);
5235 /* Composes the datapath actions for 'subfacet' based on its rule's actions.
5236 * Translates the actions into 'odp_actions', which the caller must have
5237 * initialized and is responsible for uninitializing. */
5239 subfacet_make_actions(struct subfacet *subfacet, const struct ofpbuf *packet,
5240 struct ofpbuf *odp_actions)
5242 struct facet *facet = subfacet->facet;
5243 struct rule_dpif *rule = facet->rule;
5244 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5246 struct action_xlate_ctx ctx;
5248 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
5249 &subfacet->initial_vals, rule, 0, packet);
5250 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, odp_actions);
5251 facet->tags = ctx.tags;
5252 facet->has_learn = ctx.has_learn;
5253 facet->has_normal = ctx.has_normal;
5254 facet->has_fin_timeout = ctx.has_fin_timeout;
5255 facet->nf_flow.output_iface = ctx.nf_output_iface;
5256 facet->mirrors = ctx.mirrors;
5258 subfacet->slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
5259 if (subfacet->actions_len != odp_actions->size
5260 || memcmp(subfacet->actions, odp_actions->data, odp_actions->size)) {
5261 free(subfacet->actions);
5262 subfacet->actions_len = odp_actions->size;
5263 subfacet->actions = xmemdup(odp_actions->data, odp_actions->size);
5267 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
5268 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
5269 * in the datapath will be zeroed and 'stats' will be updated with traffic new
5270 * since 'subfacet' was last updated.
5272 * Returns 0 if successful, otherwise a positive errno value. */
5274 subfacet_install(struct subfacet *subfacet,
5275 const struct nlattr *actions, size_t actions_len,
5276 struct dpif_flow_stats *stats,
5277 enum slow_path_reason slow)
5279 struct facet *facet = subfacet->facet;
5280 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5281 enum subfacet_path path = subfacet_want_path(slow);
5282 uint64_t slow_path_stub[128 / 8];
5283 enum dpif_flow_put_flags flags;
5286 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
5288 flags |= DPIF_FP_ZERO_STATS;
5291 if (path == SF_SLOW_PATH) {
5292 compose_slow_path(ofproto, &facet->flow, slow,
5293 slow_path_stub, sizeof slow_path_stub,
5294 &actions, &actions_len);
5297 ret = dpif_flow_put(ofproto->backer->dpif, flags, subfacet->key,
5298 subfacet->key_len, actions, actions_len, stats);
5301 subfacet_reset_dp_stats(subfacet, stats);
5305 subfacet->path = path;
5311 subfacet_reinstall(struct subfacet *subfacet, struct dpif_flow_stats *stats)
5313 return subfacet_install(subfacet, subfacet->actions, subfacet->actions_len,
5314 stats, subfacet->slow);
5317 /* If 'subfacet' is installed in the datapath, uninstalls it. */
5319 subfacet_uninstall(struct subfacet *subfacet)
5321 if (subfacet->path != SF_NOT_INSTALLED) {
5322 struct rule_dpif *rule = subfacet->facet->rule;
5323 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5324 struct dpif_flow_stats stats;
5327 error = dpif_flow_del(ofproto->backer->dpif, subfacet->key,
5328 subfacet->key_len, &stats);
5329 subfacet_reset_dp_stats(subfacet, &stats);
5331 subfacet_update_stats(subfacet, &stats);
5333 subfacet->path = SF_NOT_INSTALLED;
5335 ovs_assert(subfacet->dp_packet_count == 0);
5336 ovs_assert(subfacet->dp_byte_count == 0);
5340 /* Resets 'subfacet''s datapath statistics counters. This should be called
5341 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
5342 * non-null, it should contain the statistics returned by dpif when 'subfacet'
5343 * was reset in the datapath. 'stats' will be modified to include only
5344 * statistics new since 'subfacet' was last updated. */
5346 subfacet_reset_dp_stats(struct subfacet *subfacet,
5347 struct dpif_flow_stats *stats)
5350 && subfacet->dp_packet_count <= stats->n_packets
5351 && subfacet->dp_byte_count <= stats->n_bytes) {
5352 stats->n_packets -= subfacet->dp_packet_count;
5353 stats->n_bytes -= subfacet->dp_byte_count;
5356 subfacet->dp_packet_count = 0;
5357 subfacet->dp_byte_count = 0;
5360 /* Updates 'subfacet''s used time. The caller is responsible for calling
5361 * facet_push_stats() to update the flows which 'subfacet' resubmits into. */
5363 subfacet_update_time(struct subfacet *subfacet, long long int used)
5365 if (used > subfacet->used) {
5366 subfacet->used = used;
5367 facet_update_time(subfacet->facet, used);
5371 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
5373 * Because of the meaning of a subfacet's counters, it only makes sense to do
5374 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
5375 * represents a packet that was sent by hand or if it represents statistics
5376 * that have been cleared out of the datapath. */
5378 subfacet_update_stats(struct subfacet *subfacet,
5379 const struct dpif_flow_stats *stats)
5381 if (stats->n_packets || stats->used > subfacet->used) {
5382 struct facet *facet = subfacet->facet;
5384 subfacet_update_time(subfacet, stats->used);
5385 facet->packet_count += stats->n_packets;
5386 facet->byte_count += stats->n_bytes;
5387 facet->tcp_flags |= stats->tcp_flags;
5388 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
5394 static struct rule_dpif *
5395 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow)
5397 struct rule_dpif *rule;
5399 rule = rule_dpif_lookup__(ofproto, flow, 0);
5404 return rule_dpif_miss_rule(ofproto, flow);
5407 static struct rule_dpif *
5408 rule_dpif_lookup__(struct ofproto_dpif *ofproto, const struct flow *flow,
5411 struct cls_rule *cls_rule;
5412 struct classifier *cls;
5414 if (table_id >= N_TABLES) {
5418 cls = &ofproto->up.tables[table_id].cls;
5419 if (flow->nw_frag & FLOW_NW_FRAG_ANY
5420 && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
5421 /* For OFPC_NORMAL frag_handling, we must pretend that transport ports
5422 * are unavailable. */
5423 struct flow ofpc_normal_flow = *flow;
5424 ofpc_normal_flow.tp_src = htons(0);
5425 ofpc_normal_flow.tp_dst = htons(0);
5426 cls_rule = classifier_lookup(cls, &ofpc_normal_flow);
5428 cls_rule = classifier_lookup(cls, flow);
5430 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
5433 static struct rule_dpif *
5434 rule_dpif_miss_rule(struct ofproto_dpif *ofproto, const struct flow *flow)
5436 struct ofport_dpif *port;
5438 port = get_ofp_port(ofproto, flow->in_port);
5440 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, flow->in_port);
5441 return ofproto->miss_rule;
5444 if (port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN) {
5445 return ofproto->no_packet_in_rule;
5447 return ofproto->miss_rule;
5451 complete_operation(struct rule_dpif *rule)
5453 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5455 rule_invalidate(rule);
5457 struct dpif_completion *c = xmalloc(sizeof *c);
5458 c->op = rule->up.pending;
5459 list_push_back(&ofproto->completions, &c->list_node);
5461 ofoperation_complete(rule->up.pending, 0);
5465 static struct rule *
5468 struct rule_dpif *rule = xmalloc(sizeof *rule);
5473 rule_dealloc(struct rule *rule_)
5475 struct rule_dpif *rule = rule_dpif_cast(rule_);
5480 rule_construct(struct rule *rule_)
5482 struct rule_dpif *rule = rule_dpif_cast(rule_);
5483 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5484 struct rule_dpif *victim;
5487 rule->packet_count = 0;
5488 rule->byte_count = 0;
5490 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
5491 if (victim && !list_is_empty(&victim->facets)) {
5492 struct facet *facet;
5494 rule->facets = victim->facets;
5495 list_moved(&rule->facets);
5496 LIST_FOR_EACH (facet, list_node, &rule->facets) {
5497 /* XXX: We're only clearing our local counters here. It's possible
5498 * that quite a few packets are unaccounted for in the datapath
5499 * statistics. These will be accounted to the new rule instead of
5500 * cleared as required. This could be fixed by clearing out the
5501 * datapath statistics for this facet, but currently it doesn't
5503 facet_reset_counters(facet);
5507 /* Must avoid list_moved() in this case. */
5508 list_init(&rule->facets);
5511 table_id = rule->up.table_id;
5513 rule->tag = victim->tag;
5514 } else if (table_id == 0) {
5519 miniflow_expand(&rule->up.cr.match.flow, &flow);
5520 rule->tag = rule_calculate_tag(&flow, &rule->up.cr.match.mask,
5521 ofproto->tables[table_id].basis);
5524 complete_operation(rule);
5529 rule_destruct(struct rule *rule_)
5531 struct rule_dpif *rule = rule_dpif_cast(rule_);
5532 struct facet *facet, *next_facet;
5534 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
5535 facet_revalidate(facet);
5538 complete_operation(rule);
5542 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
5544 struct rule_dpif *rule = rule_dpif_cast(rule_);
5545 struct facet *facet;
5549 /* Start from historical data for 'rule' itself that are no longer tracked
5550 * in facets. This counts, for example, facets that have expired. */
5551 *packets = rule->packet_count;
5552 *bytes = rule->byte_count;
5554 /* Add any statistics that are tracked by facets. This includes
5555 * statistical data recently updated by ofproto_update_stats() as well as
5556 * stats for packets that were executed "by hand" via dpif_execute(). */
5557 LIST_FOR_EACH (facet, list_node, &rule->facets) {
5558 *packets += facet->packet_count;
5559 *bytes += facet->byte_count;
5564 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
5565 struct ofpbuf *packet)
5567 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5568 struct initial_vals initial_vals;
5569 struct dpif_flow_stats stats;
5570 struct action_xlate_ctx ctx;
5571 uint64_t odp_actions_stub[1024 / 8];
5572 struct ofpbuf odp_actions;
5574 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
5575 rule_credit_stats(rule, &stats);
5577 initial_vals.vlan_tci = flow->vlan_tci;
5578 initial_vals.tunnel_ip_tos = flow->tunnel.ip_tos;
5579 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5580 action_xlate_ctx_init(&ctx, ofproto, flow, &initial_vals,
5581 rule, stats.tcp_flags, packet);
5582 ctx.resubmit_stats = &stats;
5583 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, &odp_actions);
5585 execute_odp_actions(ofproto, flow, odp_actions.data,
5586 odp_actions.size, packet);
5588 ofpbuf_uninit(&odp_actions);
5592 rule_execute(struct rule *rule, const struct flow *flow,
5593 struct ofpbuf *packet)
5595 rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
5596 ofpbuf_delete(packet);
5601 rule_modify_actions(struct rule *rule_)
5603 struct rule_dpif *rule = rule_dpif_cast(rule_);
5605 complete_operation(rule);
5608 /* Sends 'packet' out 'ofport'.
5609 * May modify 'packet'.
5610 * Returns 0 if successful, otherwise a positive errno value. */
5612 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
5614 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
5615 uint64_t odp_actions_stub[1024 / 8];
5616 struct ofpbuf key, odp_actions;
5617 struct odputil_keybuf keybuf;
5622 flow_extract(packet, 0, 0, NULL, OFPP_LOCAL, &flow);
5623 if (netdev_vport_is_patch(ofport->up.netdev)) {
5624 struct ofproto_dpif *peer_ofproto;
5625 struct dpif_flow_stats stats;
5626 struct ofport_dpif *peer;
5627 struct rule_dpif *rule;
5629 peer = ofport_get_peer(ofport);
5634 dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
5635 netdev_vport_inc_tx(ofport->up.netdev, &stats);
5636 netdev_vport_inc_rx(peer->up.netdev, &stats);
5638 flow.in_port = peer->up.ofp_port;
5639 peer_ofproto = ofproto_dpif_cast(peer->up.ofproto);
5640 rule = rule_dpif_lookup(peer_ofproto, &flow);
5641 rule_dpif_execute(rule, &flow, packet);
5646 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5648 if (ofport->tnl_port) {
5649 struct dpif_flow_stats stats;
5651 odp_port = tnl_port_send(ofport->tnl_port, &flow);
5652 if (odp_port == OVSP_NONE) {
5656 dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
5657 netdev_vport_inc_tx(ofport->up.netdev, &stats);
5658 odp_put_tunnel_action(&flow.tunnel, &odp_actions);
5659 odp_put_skb_mark_action(flow.skb_mark, &odp_actions);
5661 odp_port = vsp_realdev_to_vlandev(ofproto, ofport->odp_port,
5663 if (odp_port != ofport->odp_port) {
5664 eth_pop_vlan(packet);
5665 flow.vlan_tci = htons(0);
5669 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5670 odp_flow_key_from_flow(&key, &flow,
5671 ofp_port_to_odp_port(ofproto, flow.in_port));
5673 compose_sflow_action(ofproto, &odp_actions, &flow, odp_port);
5675 nl_msg_put_u32(&odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port);
5676 error = dpif_execute(ofproto->backer->dpif,
5678 odp_actions.data, odp_actions.size,
5680 ofpbuf_uninit(&odp_actions);
5683 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)",
5684 ofproto->up.name, odp_port, strerror(error));
5686 ofproto_update_local_port_stats(ofport->up.ofproto, packet->size, 0);
5690 /* OpenFlow to datapath action translation. */
5692 static bool may_receive(const struct ofport_dpif *, struct action_xlate_ctx *);
5693 static void do_xlate_actions(const struct ofpact *, size_t ofpacts_len,
5694 struct action_xlate_ctx *);
5695 static void xlate_normal(struct action_xlate_ctx *);
5697 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
5698 * The action will state 'slow' as the reason that the action is in the slow
5699 * path. (This is purely informational: it allows a human viewing "ovs-dpctl
5700 * dump-flows" output to see why a flow is in the slow path.)
5702 * The 'stub_size' bytes in 'stub' will be used to store the action.
5703 * 'stub_size' must be large enough for the action.
5705 * The action and its size will be stored in '*actionsp' and '*actions_lenp',
5708 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
5709 enum slow_path_reason slow,
5710 uint64_t *stub, size_t stub_size,
5711 const struct nlattr **actionsp, size_t *actions_lenp)
5713 union user_action_cookie cookie;
5716 cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
5717 cookie.slow_path.unused = 0;
5718 cookie.slow_path.reason = slow;
5720 ofpbuf_use_stack(&buf, stub, stub_size);
5721 if (slow & (SLOW_CFM | SLOW_LACP | SLOW_STP)) {
5722 uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif, UINT32_MAX);
5723 odp_put_userspace_action(pid, &cookie, sizeof cookie, &buf);
5725 put_userspace_action(ofproto, &buf, flow, &cookie);
5727 *actionsp = buf.data;
5728 *actions_lenp = buf.size;
5732 put_userspace_action(const struct ofproto_dpif *ofproto,
5733 struct ofpbuf *odp_actions,
5734 const struct flow *flow,
5735 const union user_action_cookie *cookie)
5739 pid = dpif_port_get_pid(ofproto->backer->dpif,
5740 ofp_port_to_odp_port(ofproto, flow->in_port));
5742 return odp_put_userspace_action(pid, cookie, sizeof *cookie, odp_actions);
5746 compose_sflow_cookie(const struct ofproto_dpif *ofproto,
5747 ovs_be16 vlan_tci, uint32_t odp_port,
5748 unsigned int n_outputs, union user_action_cookie *cookie)
5752 cookie->type = USER_ACTION_COOKIE_SFLOW;
5753 cookie->sflow.vlan_tci = vlan_tci;
5755 /* See http://www.sflow.org/sflow_version_5.txt (search for "Input/output
5756 * port information") for the interpretation of cookie->output. */
5757 switch (n_outputs) {
5759 /* 0x40000000 | 256 means "packet dropped for unknown reason". */
5760 cookie->sflow.output = 0x40000000 | 256;
5764 ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port);
5766 cookie->sflow.output = ifindex;
5771 /* 0x80000000 means "multiple output ports. */
5772 cookie->sflow.output = 0x80000000 | n_outputs;
5777 /* Compose SAMPLE action for sFlow. */
5779 compose_sflow_action(const struct ofproto_dpif *ofproto,
5780 struct ofpbuf *odp_actions,
5781 const struct flow *flow,
5784 uint32_t probability;
5785 union user_action_cookie cookie;
5786 size_t sample_offset, actions_offset;
5789 if (!ofproto->sflow || flow->in_port == OFPP_NONE) {
5793 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
5795 /* Number of packets out of UINT_MAX to sample. */
5796 probability = dpif_sflow_get_probability(ofproto->sflow);
5797 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
5799 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
5800 compose_sflow_cookie(ofproto, htons(0), odp_port,
5801 odp_port == OVSP_NONE ? 0 : 1, &cookie);
5802 cookie_offset = put_userspace_action(ofproto, odp_actions, flow, &cookie);
5804 nl_msg_end_nested(odp_actions, actions_offset);
5805 nl_msg_end_nested(odp_actions, sample_offset);
5806 return cookie_offset;
5809 /* SAMPLE action must be first action in any given list of actions.
5810 * At this point we do not have all information required to build it. So try to
5811 * build sample action as complete as possible. */
5813 add_sflow_action(struct action_xlate_ctx *ctx)
5815 ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto,
5817 &ctx->flow, OVSP_NONE);
5818 ctx->sflow_odp_port = 0;
5819 ctx->sflow_n_outputs = 0;
5822 /* Fix SAMPLE action according to data collected while composing ODP actions.
5823 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
5824 * USERSPACE action's user-cookie which is required for sflow. */
5826 fix_sflow_action(struct action_xlate_ctx *ctx)
5828 const struct flow *base = &ctx->base_flow;
5829 union user_action_cookie *cookie;
5831 if (!ctx->user_cookie_offset) {
5835 cookie = ofpbuf_at(ctx->odp_actions, ctx->user_cookie_offset,
5837 ovs_assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
5839 compose_sflow_cookie(ctx->ofproto, base->vlan_tci,
5840 ctx->sflow_odp_port, ctx->sflow_n_outputs, cookie);
5844 compose_output_action__(struct action_xlate_ctx *ctx, uint16_t ofp_port,
5847 const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
5848 ovs_be16 flow_vlan_tci = ctx->flow.vlan_tci;
5849 ovs_be64 flow_tun_id = ctx->flow.tunnel.tun_id;
5850 uint8_t flow_nw_tos = ctx->flow.nw_tos;
5851 struct priority_to_dscp *pdscp;
5852 uint32_t out_port, odp_port;
5854 /* If 'struct flow' gets additional metadata, we'll need to zero it out
5855 * before traversing a patch port. */
5856 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 20);
5859 xlate_report(ctx, "Nonexistent output port");
5861 } else if (ofport->up.pp.config & OFPUTIL_PC_NO_FWD) {
5862 xlate_report(ctx, "OFPPC_NO_FWD set, skipping output");
5864 } else if (check_stp && !stp_forward_in_state(ofport->stp_state)) {
5865 xlate_report(ctx, "STP not in forwarding state, skipping output");
5869 if (netdev_vport_is_patch(ofport->up.netdev)) {
5870 struct ofport_dpif *peer = ofport_get_peer(ofport);
5871 struct flow old_flow = ctx->flow;
5872 const struct ofproto_dpif *peer_ofproto;
5873 enum slow_path_reason special;
5874 struct ofport_dpif *in_port;
5877 xlate_report(ctx, "Nonexistent patch port peer");
5881 peer_ofproto = ofproto_dpif_cast(peer->up.ofproto);
5882 if (peer_ofproto->backer != ctx->ofproto->backer) {
5883 xlate_report(ctx, "Patch port peer on a different datapath");
5887 ctx->ofproto = ofproto_dpif_cast(peer->up.ofproto);
5888 ctx->flow.in_port = peer->up.ofp_port;
5889 ctx->flow.metadata = htonll(0);
5890 memset(&ctx->flow.tunnel, 0, sizeof ctx->flow.tunnel);
5891 memset(ctx->flow.regs, 0, sizeof ctx->flow.regs);
5893 in_port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
5894 special = process_special(ctx->ofproto, &ctx->flow, in_port,
5897 ctx->slow |= special;
5898 } else if (!in_port || may_receive(in_port, ctx)) {
5899 if (!in_port || stp_forward_in_state(in_port->stp_state)) {
5900 xlate_table_action(ctx, ctx->flow.in_port, 0, true);
5902 /* Forwarding is disabled by STP. Let OFPP_NORMAL and the
5903 * learning action look at the packet, then drop it. */
5904 struct flow old_base_flow = ctx->base_flow;
5905 size_t old_size = ctx->odp_actions->size;
5906 xlate_table_action(ctx, ctx->flow.in_port, 0, true);
5907 ctx->base_flow = old_base_flow;
5908 ctx->odp_actions->size = old_size;
5912 ctx->flow = old_flow;
5913 ctx->ofproto = ofproto_dpif_cast(ofport->up.ofproto);
5915 if (ctx->resubmit_stats) {
5916 netdev_vport_inc_tx(ofport->up.netdev, ctx->resubmit_stats);
5917 netdev_vport_inc_rx(peer->up.netdev, ctx->resubmit_stats);
5923 pdscp = get_priority(ofport, ctx->flow.skb_priority);
5925 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
5926 ctx->flow.nw_tos |= pdscp->dscp;
5929 if (ofport->tnl_port) {
5930 odp_port = tnl_port_send(ofport->tnl_port, &ctx->flow);
5931 if (odp_port == OVSP_NONE) {
5932 xlate_report(ctx, "Tunneling decided against output");
5936 if (ctx->resubmit_stats) {
5937 netdev_vport_inc_tx(ofport->up.netdev, ctx->resubmit_stats);
5939 out_port = odp_port;
5940 commit_odp_tunnel_action(&ctx->flow, &ctx->base_flow,
5943 odp_port = ofport->odp_port;
5944 out_port = vsp_realdev_to_vlandev(ctx->ofproto, odp_port,
5945 ctx->flow.vlan_tci);
5946 if (out_port != odp_port) {
5947 ctx->flow.vlan_tci = htons(0);
5949 ctx->flow.skb_mark &= ~IPSEC_MARK;
5951 commit_odp_actions(&ctx->flow, &ctx->base_flow, ctx->odp_actions);
5952 nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_OUTPUT, out_port);
5954 ctx->sflow_odp_port = odp_port;
5955 ctx->sflow_n_outputs++;
5956 ctx->nf_output_iface = ofp_port;
5957 ctx->flow.tunnel.tun_id = flow_tun_id;
5958 ctx->flow.vlan_tci = flow_vlan_tci;
5959 ctx->flow.nw_tos = flow_nw_tos;
5963 compose_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port)
5965 compose_output_action__(ctx, ofp_port, true);
5969 tag_the_flow(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
5971 struct ofproto_dpif *ofproto = ctx->ofproto;
5972 uint8_t table_id = ctx->table_id;
5974 if (table_id > 0 && table_id < N_TABLES) {
5975 struct table_dpif *table = &ofproto->tables[table_id];
5976 if (table->other_table) {
5977 ctx->tags |= (rule && rule->tag
5979 : rule_calculate_tag(&ctx->flow,
5980 &table->other_table->mask,
5986 /* Common rule processing in one place to avoid duplicating code. */
5987 static struct rule_dpif *
5988 ctx_rule_hooks(struct action_xlate_ctx *ctx, struct rule_dpif *rule,
5991 if (ctx->resubmit_hook) {
5992 ctx->resubmit_hook(ctx, rule);
5994 if (rule == NULL && may_packet_in) {
5996 * check if table configuration flags
5997 * OFPTC_TABLE_MISS_CONTROLLER, default.
5998 * OFPTC_TABLE_MISS_CONTINUE,
5999 * OFPTC_TABLE_MISS_DROP
6000 * When OF1.0, OFPTC_TABLE_MISS_CONTINUE is used. What to do?
6002 rule = rule_dpif_miss_rule(ctx->ofproto, &ctx->flow);
6004 if (rule && ctx->resubmit_stats) {
6005 rule_credit_stats(rule, ctx->resubmit_stats);
6011 xlate_table_action(struct action_xlate_ctx *ctx,
6012 uint16_t in_port, uint8_t table_id, bool may_packet_in)
6014 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
6015 struct rule_dpif *rule;
6016 uint16_t old_in_port = ctx->flow.in_port;
6017 uint8_t old_table_id = ctx->table_id;
6019 ctx->table_id = table_id;
6021 /* Look up a flow with 'in_port' as the input port. */
6022 ctx->flow.in_port = in_port;
6023 rule = rule_dpif_lookup__(ctx->ofproto, &ctx->flow, table_id);
6025 tag_the_flow(ctx, rule);
6027 /* Restore the original input port. Otherwise OFPP_NORMAL and
6028 * OFPP_IN_PORT will have surprising behavior. */
6029 ctx->flow.in_port = old_in_port;
6031 rule = ctx_rule_hooks(ctx, rule, may_packet_in);
6034 struct rule_dpif *old_rule = ctx->rule;
6038 do_xlate_actions(rule->up.ofpacts, rule->up.ofpacts_len, ctx);
6039 ctx->rule = old_rule;
6043 ctx->table_id = old_table_id;
6045 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
6047 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
6048 MAX_RESUBMIT_RECURSION);
6049 ctx->max_resubmit_trigger = true;
6054 xlate_ofpact_resubmit(struct action_xlate_ctx *ctx,
6055 const struct ofpact_resubmit *resubmit)
6060 in_port = resubmit->in_port;
6061 if (in_port == OFPP_IN_PORT) {
6062 in_port = ctx->flow.in_port;
6065 table_id = resubmit->table_id;
6066 if (table_id == 255) {
6067 table_id = ctx->table_id;
6070 xlate_table_action(ctx, in_port, table_id, false);
6074 flood_packets(struct action_xlate_ctx *ctx, bool all)
6076 struct ofport_dpif *ofport;
6078 HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) {
6079 uint16_t ofp_port = ofport->up.ofp_port;
6081 if (ofp_port == ctx->flow.in_port) {
6086 compose_output_action__(ctx, ofp_port, false);
6087 } else if (!(ofport->up.pp.config & OFPUTIL_PC_NO_FLOOD)) {
6088 compose_output_action(ctx, ofp_port);
6092 ctx->nf_output_iface = NF_OUT_FLOOD;
6096 execute_controller_action(struct action_xlate_ctx *ctx, int len,
6097 enum ofp_packet_in_reason reason,
6098 uint16_t controller_id)
6100 struct ofputil_packet_in pin;
6101 struct ofpbuf *packet;
6103 ctx->slow |= SLOW_CONTROLLER;
6108 packet = ofpbuf_clone(ctx->packet);
6110 if (packet->l2 && packet->l3) {
6111 struct eth_header *eh;
6112 uint16_t mpls_depth;
6114 eth_pop_vlan(packet);
6117 memcpy(eh->eth_src, ctx->flow.dl_src, sizeof eh->eth_src);
6118 memcpy(eh->eth_dst, ctx->flow.dl_dst, sizeof eh->eth_dst);
6120 if (ctx->flow.vlan_tci & htons(VLAN_CFI)) {
6121 eth_push_vlan(packet, ctx->flow.vlan_tci);
6124 mpls_depth = eth_mpls_depth(packet);
6126 if (mpls_depth < ctx->flow.mpls_depth) {
6127 push_mpls(packet, ctx->flow.dl_type, ctx->flow.mpls_lse);
6128 } else if (mpls_depth > ctx->flow.mpls_depth) {
6129 pop_mpls(packet, ctx->flow.dl_type);
6130 } else if (mpls_depth) {
6131 set_mpls_lse(packet, ctx->flow.mpls_lse);
6135 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
6136 packet_set_ipv4(packet, ctx->flow.nw_src, ctx->flow.nw_dst,
6137 ctx->flow.nw_tos, ctx->flow.nw_ttl);
6141 if (ctx->flow.nw_proto == IPPROTO_TCP) {
6142 packet_set_tcp_port(packet, ctx->flow.tp_src,
6144 } else if (ctx->flow.nw_proto == IPPROTO_UDP) {
6145 packet_set_udp_port(packet, ctx->flow.tp_src,
6152 pin.packet = packet->data;
6153 pin.packet_len = packet->size;
6154 pin.reason = reason;
6155 pin.controller_id = controller_id;
6156 pin.table_id = ctx->table_id;
6157 pin.cookie = ctx->rule ? ctx->rule->up.flow_cookie : 0;
6160 flow_get_metadata(&ctx->flow, &pin.fmd);
6162 connmgr_send_packet_in(ctx->ofproto->up.connmgr, &pin);
6163 ofpbuf_delete(packet);
6167 execute_mpls_push_action(struct action_xlate_ctx *ctx, ovs_be16 eth_type)
6169 ovs_assert(eth_type_mpls(eth_type));
6171 if (ctx->base_flow.mpls_depth) {
6172 ctx->flow.mpls_lse &= ~htonl(MPLS_BOS_MASK);
6173 ctx->flow.mpls_depth++;
6178 if (ctx->flow.dl_type == htons(ETH_TYPE_IPV6)) {
6179 label = htonl(0x2); /* IPV6 Explicit Null. */
6181 label = htonl(0x0); /* IPV4 Explicit Null. */
6183 tc = (ctx->flow.nw_tos & IP_DSCP_MASK) >> 2;
6184 ttl = ctx->flow.nw_ttl ? ctx->flow.nw_ttl : 0x40;
6185 ctx->flow.mpls_lse = set_mpls_lse_values(ttl, tc, 1, label);
6186 ctx->flow.mpls_depth = 1;
6188 ctx->flow.dl_type = eth_type;
6192 execute_mpls_pop_action(struct action_xlate_ctx *ctx, ovs_be16 eth_type)
6194 ovs_assert(eth_type_mpls(ctx->flow.dl_type));
6195 ovs_assert(!eth_type_mpls(eth_type));
6197 if (ctx->flow.mpls_depth) {
6198 ctx->flow.mpls_depth--;
6199 ctx->flow.mpls_lse = htonl(0);
6200 if (!ctx->flow.mpls_depth) {
6201 ctx->flow.dl_type = eth_type;
6207 compose_dec_ttl(struct action_xlate_ctx *ctx, struct ofpact_cnt_ids *ids)
6209 if (ctx->flow.dl_type != htons(ETH_TYPE_IP) &&
6210 ctx->flow.dl_type != htons(ETH_TYPE_IPV6)) {
6214 if (ctx->flow.nw_ttl > 1) {
6220 for (i = 0; i < ids->n_controllers; i++) {
6221 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL,
6225 /* Stop processing for current table. */
6231 execute_set_mpls_ttl_action(struct action_xlate_ctx *ctx, uint8_t ttl)
6233 if (!eth_type_mpls(ctx->flow.dl_type)) {
6237 set_mpls_lse_ttl(&ctx->flow.mpls_lse, ttl);
6242 execute_dec_mpls_ttl_action(struct action_xlate_ctx *ctx)
6244 uint8_t ttl = mpls_lse_to_ttl(ctx->flow.mpls_lse);
6246 if (!eth_type_mpls(ctx->flow.dl_type)) {
6252 set_mpls_lse_ttl(&ctx->flow.mpls_lse, ttl);
6255 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL, 0);
6257 /* Stop processing for current table. */
6263 xlate_output_action(struct action_xlate_ctx *ctx,
6264 uint16_t port, uint16_t max_len, bool may_packet_in)
6266 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
6268 ctx->nf_output_iface = NF_OUT_DROP;
6272 compose_output_action(ctx, ctx->flow.in_port);
6275 xlate_table_action(ctx, ctx->flow.in_port, 0, may_packet_in);
6281 flood_packets(ctx, false);
6284 flood_packets(ctx, true);
6286 case OFPP_CONTROLLER:
6287 execute_controller_action(ctx, max_len, OFPR_ACTION, 0);
6293 if (port != ctx->flow.in_port) {
6294 compose_output_action(ctx, port);
6296 xlate_report(ctx, "skipping output to input port");
6301 if (prev_nf_output_iface == NF_OUT_FLOOD) {
6302 ctx->nf_output_iface = NF_OUT_FLOOD;
6303 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
6304 ctx->nf_output_iface = prev_nf_output_iface;
6305 } else if (prev_nf_output_iface != NF_OUT_DROP &&
6306 ctx->nf_output_iface != NF_OUT_FLOOD) {
6307 ctx->nf_output_iface = NF_OUT_MULTI;
6312 xlate_output_reg_action(struct action_xlate_ctx *ctx,
6313 const struct ofpact_output_reg *or)
6315 uint64_t port = mf_get_subfield(&or->src, &ctx->flow);
6316 if (port <= UINT16_MAX) {
6317 xlate_output_action(ctx, port, or->max_len, false);
6322 xlate_enqueue_action(struct action_xlate_ctx *ctx,
6323 const struct ofpact_enqueue *enqueue)
6325 uint16_t ofp_port = enqueue->port;
6326 uint32_t queue_id = enqueue->queue;
6327 uint32_t flow_priority, priority;
6330 /* Translate queue to priority. */
6331 error = dpif_queue_to_priority(ctx->ofproto->backer->dpif,
6332 queue_id, &priority);
6334 /* Fall back to ordinary output action. */
6335 xlate_output_action(ctx, enqueue->port, 0, false);
6339 /* Check output port. */
6340 if (ofp_port == OFPP_IN_PORT) {
6341 ofp_port = ctx->flow.in_port;
6342 } else if (ofp_port == ctx->flow.in_port) {
6346 /* Add datapath actions. */
6347 flow_priority = ctx->flow.skb_priority;
6348 ctx->flow.skb_priority = priority;
6349 compose_output_action(ctx, ofp_port);
6350 ctx->flow.skb_priority = flow_priority;
6352 /* Update NetFlow output port. */
6353 if (ctx->nf_output_iface == NF_OUT_DROP) {
6354 ctx->nf_output_iface = ofp_port;
6355 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
6356 ctx->nf_output_iface = NF_OUT_MULTI;
6361 xlate_set_queue_action(struct action_xlate_ctx *ctx, uint32_t queue_id)
6363 uint32_t skb_priority;
6365 if (!dpif_queue_to_priority(ctx->ofproto->backer->dpif,
6366 queue_id, &skb_priority)) {
6367 ctx->flow.skb_priority = skb_priority;
6369 /* Couldn't translate queue to a priority. Nothing to do. A warning
6370 * has already been logged. */
6374 struct xlate_reg_state {
6380 slave_enabled_cb(uint16_t ofp_port, void *ofproto_)
6382 struct ofproto_dpif *ofproto = ofproto_;
6383 struct ofport_dpif *port;
6393 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
6396 port = get_ofp_port(ofproto, ofp_port);
6397 return port ? port->may_enable : false;
6402 xlate_bundle_action(struct action_xlate_ctx *ctx,
6403 const struct ofpact_bundle *bundle)
6407 port = bundle_execute(bundle, &ctx->flow, slave_enabled_cb, ctx->ofproto);
6408 if (bundle->dst.field) {
6409 nxm_reg_load(&bundle->dst, port, &ctx->flow);
6411 xlate_output_action(ctx, port, 0, false);
6416 xlate_learn_action(struct action_xlate_ctx *ctx,
6417 const struct ofpact_learn *learn)
6419 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
6420 struct ofputil_flow_mod fm;
6421 uint64_t ofpacts_stub[1024 / 8];
6422 struct ofpbuf ofpacts;
6425 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
6426 learn_execute(learn, &ctx->flow, &fm, &ofpacts);
6428 error = ofproto_flow_mod(&ctx->ofproto->up, &fm);
6429 if (error && !VLOG_DROP_WARN(&rl)) {
6430 VLOG_WARN("learning action failed to modify flow table (%s)",
6431 ofperr_get_name(error));
6434 ofpbuf_uninit(&ofpacts);
6437 /* Reduces '*timeout' to no more than 'max'. A value of zero in either case
6438 * means "infinite". */
6440 reduce_timeout(uint16_t max, uint16_t *timeout)
6442 if (max && (!*timeout || *timeout > max)) {
6448 xlate_fin_timeout(struct action_xlate_ctx *ctx,
6449 const struct ofpact_fin_timeout *oft)
6451 if (ctx->tcp_flags & (TCP_FIN | TCP_RST) && ctx->rule) {
6452 struct rule_dpif *rule = ctx->rule;
6454 reduce_timeout(oft->fin_idle_timeout, &rule->up.idle_timeout);
6455 reduce_timeout(oft->fin_hard_timeout, &rule->up.hard_timeout);
6460 may_receive(const struct ofport_dpif *port, struct action_xlate_ctx *ctx)
6462 if (port->up.pp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
6463 ? OFPUTIL_PC_NO_RECV_STP
6464 : OFPUTIL_PC_NO_RECV)) {
6468 /* Only drop packets here if both forwarding and learning are
6469 * disabled. If just learning is enabled, we need to have
6470 * OFPP_NORMAL and the learning action have a look at the packet
6471 * before we can drop it. */
6472 if (!stp_forward_in_state(port->stp_state)
6473 && !stp_learn_in_state(port->stp_state)) {
6481 tunnel_ecn_ok(struct action_xlate_ctx *ctx)
6483 if (is_ip_any(&ctx->base_flow)
6484 && (ctx->base_flow.tunnel.ip_tos & IP_ECN_MASK) == IP_ECN_CE) {
6485 if ((ctx->base_flow.nw_tos & IP_ECN_MASK) == IP_ECN_NOT_ECT) {
6486 VLOG_WARN_RL(&rl, "dropping tunnel packet marked ECN CE"
6487 " but is not ECN capable");
6490 /* Set the ECN CE value in the tunneled packet. */
6491 ctx->flow.nw_tos |= IP_ECN_CE;
6499 do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len,
6500 struct action_xlate_ctx *ctx)
6502 bool was_evictable = true;
6503 const struct ofpact *a;
6506 /* Don't let the rule we're working on get evicted underneath us. */
6507 was_evictable = ctx->rule->up.evictable;
6508 ctx->rule->up.evictable = false;
6511 do_xlate_actions_again:
6512 OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) {
6513 struct ofpact_controller *controller;
6514 const struct ofpact_metadata *metadata;
6522 xlate_output_action(ctx, ofpact_get_OUTPUT(a)->port,
6523 ofpact_get_OUTPUT(a)->max_len, true);
6526 case OFPACT_CONTROLLER:
6527 controller = ofpact_get_CONTROLLER(a);
6528 execute_controller_action(ctx, controller->max_len,
6530 controller->controller_id);
6533 case OFPACT_ENQUEUE:
6534 xlate_enqueue_action(ctx, ofpact_get_ENQUEUE(a));
6537 case OFPACT_SET_VLAN_VID:
6538 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
6539 ctx->flow.vlan_tci |= (htons(ofpact_get_SET_VLAN_VID(a)->vlan_vid)
6543 case OFPACT_SET_VLAN_PCP:
6544 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
6545 ctx->flow.vlan_tci |= htons((ofpact_get_SET_VLAN_PCP(a)->vlan_pcp
6550 case OFPACT_STRIP_VLAN:
6551 ctx->flow.vlan_tci = htons(0);
6554 case OFPACT_PUSH_VLAN:
6555 /* XXX 802.1AD(QinQ) */
6556 ctx->flow.vlan_tci = htons(VLAN_CFI);
6559 case OFPACT_SET_ETH_SRC:
6560 memcpy(ctx->flow.dl_src, ofpact_get_SET_ETH_SRC(a)->mac,
6564 case OFPACT_SET_ETH_DST:
6565 memcpy(ctx->flow.dl_dst, ofpact_get_SET_ETH_DST(a)->mac,
6569 case OFPACT_SET_IPV4_SRC:
6570 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
6571 ctx->flow.nw_src = ofpact_get_SET_IPV4_SRC(a)->ipv4;
6575 case OFPACT_SET_IPV4_DST:
6576 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
6577 ctx->flow.nw_dst = ofpact_get_SET_IPV4_DST(a)->ipv4;
6581 case OFPACT_SET_IPV4_DSCP:
6582 /* OpenFlow 1.0 only supports IPv4. */
6583 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
6584 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
6585 ctx->flow.nw_tos |= ofpact_get_SET_IPV4_DSCP(a)->dscp;
6589 case OFPACT_SET_L4_SRC_PORT:
6590 if (is_ip_any(&ctx->flow)) {
6591 ctx->flow.tp_src = htons(ofpact_get_SET_L4_SRC_PORT(a)->port);
6595 case OFPACT_SET_L4_DST_PORT:
6596 if (is_ip_any(&ctx->flow)) {
6597 ctx->flow.tp_dst = htons(ofpact_get_SET_L4_DST_PORT(a)->port);
6601 case OFPACT_RESUBMIT:
6602 xlate_ofpact_resubmit(ctx, ofpact_get_RESUBMIT(a));
6605 case OFPACT_SET_TUNNEL:
6606 ctx->flow.tunnel.tun_id = htonll(ofpact_get_SET_TUNNEL(a)->tun_id);
6609 case OFPACT_SET_QUEUE:
6610 xlate_set_queue_action(ctx, ofpact_get_SET_QUEUE(a)->queue_id);
6613 case OFPACT_POP_QUEUE:
6614 ctx->flow.skb_priority = ctx->orig_skb_priority;
6617 case OFPACT_REG_MOVE:
6618 nxm_execute_reg_move(ofpact_get_REG_MOVE(a), &ctx->flow);
6621 case OFPACT_REG_LOAD:
6622 nxm_execute_reg_load(ofpact_get_REG_LOAD(a), &ctx->flow);
6625 case OFPACT_STACK_PUSH:
6626 nxm_execute_stack_push(ofpact_get_STACK_PUSH(a), &ctx->flow,
6630 case OFPACT_STACK_POP:
6631 nxm_execute_stack_pop(ofpact_get_STACK_POP(a), &ctx->flow,
6635 case OFPACT_PUSH_MPLS:
6636 execute_mpls_push_action(ctx, ofpact_get_PUSH_MPLS(a)->ethertype);
6639 case OFPACT_POP_MPLS:
6640 execute_mpls_pop_action(ctx, ofpact_get_POP_MPLS(a)->ethertype);
6643 case OFPACT_SET_MPLS_TTL:
6644 if (execute_set_mpls_ttl_action(ctx, ofpact_get_SET_MPLS_TTL(a)->ttl)) {
6649 case OFPACT_DEC_MPLS_TTL:
6650 if (execute_dec_mpls_ttl_action(ctx)) {
6655 case OFPACT_DEC_TTL:
6656 if (compose_dec_ttl(ctx, ofpact_get_DEC_TTL(a))) {
6662 /* Nothing to do. */
6665 case OFPACT_MULTIPATH:
6666 multipath_execute(ofpact_get_MULTIPATH(a), &ctx->flow);
6670 ctx->ofproto->has_bundle_action = true;
6671 xlate_bundle_action(ctx, ofpact_get_BUNDLE(a));
6674 case OFPACT_OUTPUT_REG:
6675 xlate_output_reg_action(ctx, ofpact_get_OUTPUT_REG(a));
6679 ctx->has_learn = true;
6680 if (ctx->may_learn) {
6681 xlate_learn_action(ctx, ofpact_get_LEARN(a));
6689 case OFPACT_FIN_TIMEOUT:
6690 ctx->has_fin_timeout = true;
6691 xlate_fin_timeout(ctx, ofpact_get_FIN_TIMEOUT(a));
6694 case OFPACT_CLEAR_ACTIONS:
6696 * Nothing to do because writa-actions is not supported for now.
6697 * When writa-actions is supported, clear-actions also must
6698 * be supported at the same time.
6702 case OFPACT_WRITE_METADATA:
6703 metadata = ofpact_get_WRITE_METADATA(a);
6704 ctx->flow.metadata &= ~metadata->mask;
6705 ctx->flow.metadata |= metadata->metadata & metadata->mask;
6708 case OFPACT_GOTO_TABLE: {
6709 /* It is assumed that goto-table is the last action. */
6710 struct ofpact_goto_table *ogt = ofpact_get_GOTO_TABLE(a);
6711 struct rule_dpif *rule;
6713 ovs_assert(ctx->table_id < ogt->table_id);
6715 ctx->table_id = ogt->table_id;
6717 /* Look up a flow from the new table. */
6718 rule = rule_dpif_lookup__(ctx->ofproto, &ctx->flow, ctx->table_id);
6720 tag_the_flow(ctx, rule);
6722 rule = ctx_rule_hooks(ctx, rule, true);
6726 ctx->rule->up.evictable = was_evictable;
6729 was_evictable = rule->up.evictable;
6730 rule->up.evictable = false;
6732 /* Tail recursion removal. */
6733 ofpacts = rule->up.ofpacts;
6734 ofpacts_len = rule->up.ofpacts_len;
6735 goto do_xlate_actions_again;
6744 ctx->rule->up.evictable = was_evictable;
6749 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
6750 struct ofproto_dpif *ofproto, const struct flow *flow,
6751 const struct initial_vals *initial_vals,
6752 struct rule_dpif *rule,
6753 uint8_t tcp_flags, const struct ofpbuf *packet)
6755 ovs_be64 initial_tun_id = flow->tunnel.tun_id;
6757 /* Flow initialization rules:
6758 * - 'base_flow' must match the kernel's view of the packet at the
6759 * time that action processing starts. 'flow' represents any
6760 * transformations we wish to make through actions.
6761 * - By default 'base_flow' and 'flow' are the same since the input
6762 * packet matches the output before any actions are applied.
6763 * - When using VLAN splinters, 'base_flow''s VLAN is set to the value
6764 * of the received packet as seen by the kernel. If we later output
6765 * to another device without any modifications this will cause us to
6766 * insert a new tag since the original one was stripped off by the
6768 * - Tunnel 'flow' is largely cleared when transitioning between
6769 * the input and output stages since it does not make sense to output
6770 * a packet with the exact headers that it was received with (i.e.
6771 * the destination IP is us). The one exception is the tun_id, which
6772 * is preserved to allow use in later resubmit lookups and loads into
6774 * - Tunnel 'base_flow' is completely cleared since that is what the
6775 * kernel does. If we wish to maintain the original values an action
6776 * needs to be generated. */
6778 ctx->ofproto = ofproto;
6780 memset(&ctx->flow.tunnel, 0, sizeof ctx->flow.tunnel);
6781 ctx->base_flow = ctx->flow;
6782 ctx->base_flow.vlan_tci = initial_vals->vlan_tci;
6783 ctx->base_flow.tunnel.ip_tos = initial_vals->tunnel_ip_tos;
6784 ctx->flow.tunnel.tun_id = initial_tun_id;
6786 ctx->packet = packet;
6787 ctx->may_learn = packet != NULL;
6788 ctx->tcp_flags = tcp_flags;
6789 ctx->resubmit_hook = NULL;
6790 ctx->report_hook = NULL;
6791 ctx->resubmit_stats = NULL;
6794 /* Translates the 'ofpacts_len' bytes of "struct ofpacts" starting at 'ofpacts'
6795 * into datapath actions in 'odp_actions', using 'ctx'. */
6797 xlate_actions(struct action_xlate_ctx *ctx,
6798 const struct ofpact *ofpacts, size_t ofpacts_len,
6799 struct ofpbuf *odp_actions)
6801 /* Normally false. Set to true if we ever hit MAX_RESUBMIT_RECURSION, so
6802 * that in the future we always keep a copy of the original flow for
6803 * tracing purposes. */
6804 static bool hit_resubmit_limit;
6806 enum slow_path_reason special;
6807 struct ofport_dpif *in_port;
6808 struct flow orig_flow;
6810 COVERAGE_INC(ofproto_dpif_xlate);
6812 ofpbuf_clear(odp_actions);
6813 ofpbuf_reserve(odp_actions, NL_A_U32_SIZE);
6815 ctx->odp_actions = odp_actions;
6818 ctx->has_learn = false;
6819 ctx->has_normal = false;
6820 ctx->has_fin_timeout = false;
6821 ctx->nf_output_iface = NF_OUT_DROP;
6824 ctx->max_resubmit_trigger = false;
6825 ctx->orig_skb_priority = ctx->flow.skb_priority;
6829 ofpbuf_use_stub(&ctx->stack, ctx->init_stack, sizeof ctx->init_stack);
6831 if (ctx->ofproto->has_mirrors || hit_resubmit_limit) {
6832 /* Do this conditionally because the copy is expensive enough that it
6833 * shows up in profiles. */
6834 orig_flow = ctx->flow;
6837 if (ctx->flow.nw_frag & FLOW_NW_FRAG_ANY) {
6838 switch (ctx->ofproto->up.frag_handling) {
6839 case OFPC_FRAG_NORMAL:
6840 /* We must pretend that transport ports are unavailable. */
6841 ctx->flow.tp_src = ctx->base_flow.tp_src = htons(0);
6842 ctx->flow.tp_dst = ctx->base_flow.tp_dst = htons(0);
6845 case OFPC_FRAG_DROP:
6848 case OFPC_FRAG_REASM:
6851 case OFPC_FRAG_NX_MATCH:
6852 /* Nothing to do. */
6855 case OFPC_INVALID_TTL_TO_CONTROLLER:
6860 in_port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
6861 special = process_special(ctx->ofproto, &ctx->flow, in_port, ctx->packet);
6863 ctx->slow |= special;
6865 static struct vlog_rate_limit trace_rl = VLOG_RATE_LIMIT_INIT(1, 1);
6866 struct initial_vals initial_vals;
6867 uint32_t local_odp_port;
6869 initial_vals.vlan_tci = ctx->base_flow.vlan_tci;
6870 initial_vals.tunnel_ip_tos = ctx->base_flow.tunnel.ip_tos;
6872 add_sflow_action(ctx);
6874 if (tunnel_ecn_ok(ctx) && (!in_port || may_receive(in_port, ctx))) {
6875 do_xlate_actions(ofpacts, ofpacts_len, ctx);
6877 /* We've let OFPP_NORMAL and the learning action look at the
6878 * packet, so drop it now if forwarding is disabled. */
6879 if (in_port && !stp_forward_in_state(in_port->stp_state)) {
6880 ofpbuf_clear(ctx->odp_actions);
6881 add_sflow_action(ctx);
6885 if (ctx->max_resubmit_trigger && !ctx->resubmit_hook) {
6886 if (!hit_resubmit_limit) {
6887 /* We didn't record the original flow. Make sure we do from
6889 hit_resubmit_limit = true;
6890 } else if (!VLOG_DROP_ERR(&trace_rl)) {
6891 struct ds ds = DS_EMPTY_INITIALIZER;
6893 ofproto_trace(ctx->ofproto, &orig_flow, ctx->packet,
6894 &initial_vals, &ds);
6895 VLOG_ERR("Trace triggered by excessive resubmit "
6896 "recursion:\n%s", ds_cstr(&ds));
6901 local_odp_port = ofp_port_to_odp_port(ctx->ofproto, OFPP_LOCAL);
6902 if (!connmgr_may_set_up_flow(ctx->ofproto->up.connmgr, &ctx->flow,
6904 ctx->odp_actions->data,
6905 ctx->odp_actions->size)) {
6906 ctx->slow |= SLOW_IN_BAND;
6908 && connmgr_msg_in_hook(ctx->ofproto->up.connmgr, &ctx->flow,
6910 compose_output_action(ctx, OFPP_LOCAL);
6913 if (ctx->ofproto->has_mirrors) {
6914 add_mirror_actions(ctx, &orig_flow);
6916 fix_sflow_action(ctx);
6919 ofpbuf_uninit(&ctx->stack);
6922 /* Translates the 'ofpacts_len' bytes of "struct ofpact"s starting at 'ofpacts'
6923 * into datapath actions, using 'ctx', and discards the datapath actions. */
6925 xlate_actions_for_side_effects(struct action_xlate_ctx *ctx,
6926 const struct ofpact *ofpacts,
6929 uint64_t odp_actions_stub[1024 / 8];
6930 struct ofpbuf odp_actions;
6932 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
6933 xlate_actions(ctx, ofpacts, ofpacts_len, &odp_actions);
6934 ofpbuf_uninit(&odp_actions);
6938 xlate_report(struct action_xlate_ctx *ctx, const char *s)
6940 if (ctx->report_hook) {
6941 ctx->report_hook(ctx, s);
6945 /* OFPP_NORMAL implementation. */
6947 static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *);
6949 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
6950 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle',
6951 * the bundle on which the packet was received, returns the VLAN to which the
6954 * Both 'vid' and the return value are in the range 0...4095. */
6956 input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid)
6958 switch (in_bundle->vlan_mode) {
6959 case PORT_VLAN_ACCESS:
6960 return in_bundle->vlan;
6963 case PORT_VLAN_TRUNK:
6966 case PORT_VLAN_NATIVE_UNTAGGED:
6967 case PORT_VLAN_NATIVE_TAGGED:
6968 return vid ? vid : in_bundle->vlan;
6975 /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'.
6976 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
6979 * 'vid' should be the VID obtained from the 802.1Q header that was received as
6980 * part of a packet (specify 0 if there was no 802.1Q header), in the range
6983 input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn)
6985 /* Allow any VID on the OFPP_NONE port. */
6986 if (in_bundle == &ofpp_none_bundle) {
6990 switch (in_bundle->vlan_mode) {
6991 case PORT_VLAN_ACCESS:
6994 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6995 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
6996 "packet received on port %s configured as VLAN "
6997 "%"PRIu16" access port",
6998 in_bundle->ofproto->up.name, vid,
6999 in_bundle->name, in_bundle->vlan);
7005 case PORT_VLAN_NATIVE_UNTAGGED:
7006 case PORT_VLAN_NATIVE_TAGGED:
7008 /* Port must always carry its native VLAN. */
7012 case PORT_VLAN_TRUNK:
7013 if (!ofbundle_includes_vlan(in_bundle, vid)) {
7015 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7016 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet "
7017 "received on port %s not configured for trunking "
7019 in_bundle->ofproto->up.name, vid,
7020 in_bundle->name, vid);
7032 /* Given 'vlan', the VLAN that a packet belongs to, and
7033 * 'out_bundle', a bundle on which the packet is to be output, returns the VID
7034 * that should be included in the 802.1Q header. (If the return value is 0,
7035 * then the 802.1Q header should only be included in the packet if there is a
7038 * Both 'vlan' and the return value are in the range 0...4095. */
7040 output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan)
7042 switch (out_bundle->vlan_mode) {
7043 case PORT_VLAN_ACCESS:
7046 case PORT_VLAN_TRUNK:
7047 case PORT_VLAN_NATIVE_TAGGED:
7050 case PORT_VLAN_NATIVE_UNTAGGED:
7051 return vlan == out_bundle->vlan ? 0 : vlan;
7059 output_normal(struct action_xlate_ctx *ctx, const struct ofbundle *out_bundle,
7062 struct ofport_dpif *port;
7064 ovs_be16 tci, old_tci;
7066 vid = output_vlan_to_vid(out_bundle, vlan);
7067 if (!out_bundle->bond) {
7068 port = ofbundle_get_a_port(out_bundle);
7070 port = bond_choose_output_slave(out_bundle->bond, &ctx->flow,
7073 /* No slaves enabled, so drop packet. */
7078 old_tci = ctx->flow.vlan_tci;
7080 if (tci || out_bundle->use_priority_tags) {
7081 tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK);
7083 tci |= htons(VLAN_CFI);
7086 ctx->flow.vlan_tci = tci;
7088 compose_output_action(ctx, port->up.ofp_port);
7089 ctx->flow.vlan_tci = old_tci;
7093 mirror_mask_ffs(mirror_mask_t mask)
7095 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
7100 ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
7102 return (bundle->vlan_mode != PORT_VLAN_ACCESS
7103 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
7107 ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan)
7109 return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan);
7112 /* Returns an arbitrary interface within 'bundle'. */
7113 static struct ofport_dpif *
7114 ofbundle_get_a_port(const struct ofbundle *bundle)
7116 return CONTAINER_OF(list_front(&bundle->ports),
7117 struct ofport_dpif, bundle_node);
7121 vlan_is_mirrored(const struct ofmirror *m, int vlan)
7123 return !m->vlans || bitmap_is_set(m->vlans, vlan);
7127 add_mirror_actions(struct action_xlate_ctx *ctx, const struct flow *orig_flow)
7129 struct ofproto_dpif *ofproto = ctx->ofproto;
7130 mirror_mask_t mirrors;
7131 struct ofbundle *in_bundle;
7134 const struct nlattr *a;
7137 in_bundle = lookup_input_bundle(ctx->ofproto, orig_flow->in_port,
7138 ctx->packet != NULL, NULL);
7142 mirrors = in_bundle->src_mirrors;
7144 /* Drop frames on bundles reserved for mirroring. */
7145 if (in_bundle->mirror_out) {
7146 if (ctx->packet != NULL) {
7147 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7148 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
7149 "%s, which is reserved exclusively for mirroring",
7150 ctx->ofproto->up.name, in_bundle->name);
7156 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
7157 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
7160 vlan = input_vid_to_vlan(in_bundle, vid);
7162 /* Look at the output ports to check for destination selections. */
7164 NL_ATTR_FOR_EACH (a, left, ctx->odp_actions->data,
7165 ctx->odp_actions->size) {
7166 enum ovs_action_attr type = nl_attr_type(a);
7167 struct ofport_dpif *ofport;
7169 if (type != OVS_ACTION_ATTR_OUTPUT) {
7173 ofport = get_odp_port(ofproto, nl_attr_get_u32(a));
7174 if (ofport && ofport->bundle) {
7175 mirrors |= ofport->bundle->dst_mirrors;
7183 /* Restore the original packet before adding the mirror actions. */
7184 ctx->flow = *orig_flow;
7189 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
7191 if (!vlan_is_mirrored(m, vlan)) {
7192 mirrors = zero_rightmost_1bit(mirrors);
7196 mirrors &= ~m->dup_mirrors;
7197 ctx->mirrors |= m->dup_mirrors;
7199 output_normal(ctx, m->out, vlan);
7200 } else if (vlan != m->out_vlan
7201 && !eth_addr_is_reserved(orig_flow->dl_dst)) {
7202 struct ofbundle *bundle;
7204 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
7205 if (ofbundle_includes_vlan(bundle, m->out_vlan)
7206 && !bundle->mirror_out) {
7207 output_normal(ctx, bundle, m->out_vlan);
7215 update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors,
7216 uint64_t packets, uint64_t bytes)
7222 for (; mirrors; mirrors = zero_rightmost_1bit(mirrors)) {
7225 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
7228 /* In normal circumstances 'm' will not be NULL. However,
7229 * if mirrors are reconfigured, we can temporarily get out
7230 * of sync in facet_revalidate(). We could "correct" the
7231 * mirror list before reaching here, but doing that would
7232 * not properly account the traffic stats we've currently
7233 * accumulated for previous mirror configuration. */
7237 m->packet_count += packets;
7238 m->byte_count += bytes;
7242 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
7243 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
7244 * indicate this; newer upstream kernels use gratuitous ARP requests. */
7246 is_gratuitous_arp(const struct flow *flow)
7248 return (flow->dl_type == htons(ETH_TYPE_ARP)
7249 && eth_addr_is_broadcast(flow->dl_dst)
7250 && (flow->nw_proto == ARP_OP_REPLY
7251 || (flow->nw_proto == ARP_OP_REQUEST
7252 && flow->nw_src == flow->nw_dst)));
7256 update_learning_table(struct ofproto_dpif *ofproto,
7257 const struct flow *flow, int vlan,
7258 struct ofbundle *in_bundle)
7260 struct mac_entry *mac;
7262 /* Don't learn the OFPP_NONE port. */
7263 if (in_bundle == &ofpp_none_bundle) {
7267 if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
7271 mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan);
7272 if (is_gratuitous_arp(flow)) {
7273 /* We don't want to learn from gratuitous ARP packets that are
7274 * reflected back over bond slaves so we lock the learning table. */
7275 if (!in_bundle->bond) {
7276 mac_entry_set_grat_arp_lock(mac);
7277 } else if (mac_entry_is_grat_arp_locked(mac)) {
7282 if (mac_entry_is_new(mac) || mac->port.p != in_bundle) {
7283 /* The log messages here could actually be useful in debugging,
7284 * so keep the rate limit relatively high. */
7285 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
7286 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
7287 "on port %s in VLAN %d",
7288 ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src),
7289 in_bundle->name, vlan);
7291 mac->port.p = in_bundle;
7292 tag_set_add(&ofproto->backer->revalidate_set,
7293 mac_learning_changed(ofproto->ml, mac));
7297 static struct ofbundle *
7298 lookup_input_bundle(const struct ofproto_dpif *ofproto, uint16_t in_port,
7299 bool warn, struct ofport_dpif **in_ofportp)
7301 struct ofport_dpif *ofport;
7303 /* Find the port and bundle for the received packet. */
7304 ofport = get_ofp_port(ofproto, in_port);
7306 *in_ofportp = ofport;
7308 if (ofport && ofport->bundle) {
7309 return ofport->bundle;
7312 /* Special-case OFPP_NONE, which a controller may use as the ingress
7313 * port for traffic that it is sourcing. */
7314 if (in_port == OFPP_NONE) {
7315 return &ofpp_none_bundle;
7318 /* Odd. A few possible reasons here:
7320 * - We deleted a port but there are still a few packets queued up
7323 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
7324 * we don't know about.
7326 * - The ofproto client didn't configure the port as part of a bundle.
7327 * This is particularly likely to happen if a packet was received on the
7328 * port after it was created, but before the client had a chance to
7329 * configure its bundle.
7332 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7334 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
7335 "port %"PRIu16, ofproto->up.name, in_port);
7340 /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or
7341 * dropped. Returns true if they may be forwarded, false if they should be
7344 * 'in_port' must be the ofport_dpif that corresponds to flow->in_port.
7345 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
7347 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
7348 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
7349 * checked by input_vid_is_valid().
7351 * May also add tags to '*tags', although the current implementation only does
7352 * so in one special case.
7355 is_admissible(struct action_xlate_ctx *ctx, struct ofport_dpif *in_port,
7358 struct ofproto_dpif *ofproto = ctx->ofproto;
7359 struct flow *flow = &ctx->flow;
7360 struct ofbundle *in_bundle = in_port->bundle;
7362 /* Drop frames for reserved multicast addresses
7363 * only if forward_bpdu option is absent. */
7364 if (!ofproto->up.forward_bpdu && eth_addr_is_reserved(flow->dl_dst)) {
7365 xlate_report(ctx, "packet has reserved destination MAC, dropping");
7369 if (in_bundle->bond) {
7370 struct mac_entry *mac;
7372 switch (bond_check_admissibility(in_bundle->bond, in_port,
7373 flow->dl_dst, &ctx->tags)) {
7378 xlate_report(ctx, "bonding refused admissibility, dropping");
7381 case BV_DROP_IF_MOVED:
7382 mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL);
7383 if (mac && mac->port.p != in_bundle &&
7384 (!is_gratuitous_arp(flow)
7385 || mac_entry_is_grat_arp_locked(mac))) {
7386 xlate_report(ctx, "SLB bond thinks this packet looped back, "
7398 xlate_normal(struct action_xlate_ctx *ctx)
7400 struct ofport_dpif *in_port;
7401 struct ofbundle *in_bundle;
7402 struct mac_entry *mac;
7406 ctx->has_normal = true;
7408 in_bundle = lookup_input_bundle(ctx->ofproto, ctx->flow.in_port,
7409 ctx->packet != NULL, &in_port);
7411 xlate_report(ctx, "no input bundle, dropping");
7415 /* Drop malformed frames. */
7416 if (ctx->flow.dl_type == htons(ETH_TYPE_VLAN) &&
7417 !(ctx->flow.vlan_tci & htons(VLAN_CFI))) {
7418 if (ctx->packet != NULL) {
7419 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7420 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
7421 "VLAN tag received on port %s",
7422 ctx->ofproto->up.name, in_bundle->name);
7424 xlate_report(ctx, "partial VLAN tag, dropping");
7428 /* Drop frames on bundles reserved for mirroring. */
7429 if (in_bundle->mirror_out) {
7430 if (ctx->packet != NULL) {
7431 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7432 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
7433 "%s, which is reserved exclusively for mirroring",
7434 ctx->ofproto->up.name, in_bundle->name);
7436 xlate_report(ctx, "input port is mirror output port, dropping");
7441 vid = vlan_tci_to_vid(ctx->flow.vlan_tci);
7442 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
7443 xlate_report(ctx, "disallowed VLAN VID for this input port, dropping");
7446 vlan = input_vid_to_vlan(in_bundle, vid);
7448 /* Check other admissibility requirements. */
7449 if (in_port && !is_admissible(ctx, in_port, vlan)) {
7453 /* Learn source MAC. */
7454 if (ctx->may_learn) {
7455 update_learning_table(ctx->ofproto, &ctx->flow, vlan, in_bundle);
7458 /* Determine output bundle. */
7459 mac = mac_learning_lookup(ctx->ofproto->ml, ctx->flow.dl_dst, vlan,
7462 if (mac->port.p != in_bundle) {
7463 xlate_report(ctx, "forwarding to learned port");
7464 output_normal(ctx, mac->port.p, vlan);
7466 xlate_report(ctx, "learned port is input port, dropping");
7469 struct ofbundle *bundle;
7471 xlate_report(ctx, "no learned MAC for destination, flooding");
7472 HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
7473 if (bundle != in_bundle
7474 && ofbundle_includes_vlan(bundle, vlan)
7475 && bundle->floodable
7476 && !bundle->mirror_out) {
7477 output_normal(ctx, bundle, vlan);
7480 ctx->nf_output_iface = NF_OUT_FLOOD;
7484 /* Optimized flow revalidation.
7486 * It's a difficult problem, in general, to tell which facets need to have
7487 * their actions recalculated whenever the OpenFlow flow table changes. We
7488 * don't try to solve that general problem: for most kinds of OpenFlow flow
7489 * table changes, we recalculate the actions for every facet. This is
7490 * relatively expensive, but it's good enough if the OpenFlow flow table
7491 * doesn't change very often.
7493 * However, we can expect one particular kind of OpenFlow flow table change to
7494 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
7495 * of CPU on revalidating every facet whenever MAC learning modifies the flow
7496 * table, we add a special case that applies to flow tables in which every rule
7497 * has the same form (that is, the same wildcards), except that the table is
7498 * also allowed to have a single "catch-all" flow that matches all packets. We
7499 * optimize this case by tagging all of the facets that resubmit into the table
7500 * and invalidating the same tag whenever a flow changes in that table. The
7501 * end result is that we revalidate just the facets that need it (and sometimes
7502 * a few more, but not all of the facets or even all of the facets that
7503 * resubmit to the table modified by MAC learning). */
7505 /* Calculates the tag to use for 'flow' and mask 'mask' when it is inserted
7506 * into an OpenFlow table with the given 'basis'. */
7508 rule_calculate_tag(const struct flow *flow, const struct minimask *mask,
7511 if (minimask_is_catchall(mask)) {
7514 uint32_t hash = flow_hash_in_minimask(flow, mask, secret);
7515 return tag_create_deterministic(hash);
7519 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
7520 * taggability of that table.
7522 * This function must be called after *each* change to a flow table. If you
7523 * skip calling it on some changes then the pointer comparisons at the end can
7524 * be invalid if you get unlucky. For example, if a flow removal causes a
7525 * cls_table to be destroyed and then a flow insertion causes a cls_table with
7526 * different wildcards to be created with the same address, then this function
7527 * will incorrectly skip revalidation. */
7529 table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
7531 struct table_dpif *table = &ofproto->tables[table_id];
7532 const struct oftable *oftable = &ofproto->up.tables[table_id];
7533 struct cls_table *catchall, *other;
7534 struct cls_table *t;
7536 catchall = other = NULL;
7538 switch (hmap_count(&oftable->cls.tables)) {
7540 /* We could tag this OpenFlow table but it would make the logic a
7541 * little harder and it's a corner case that doesn't seem worth it
7547 HMAP_FOR_EACH (t, hmap_node, &oftable->cls.tables) {
7548 if (cls_table_is_catchall(t)) {
7550 } else if (!other) {
7553 /* Indicate that we can't tag this by setting both tables to
7554 * NULL. (We know that 'catchall' is already NULL.) */
7561 /* Can't tag this table. */
7565 if (table->catchall_table != catchall || table->other_table != other) {
7566 table->catchall_table = catchall;
7567 table->other_table = other;
7568 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
7572 /* Given 'rule' that has changed in some way (either it is a rule being
7573 * inserted, a rule being deleted, or a rule whose actions are being
7574 * modified), marks facets for revalidation to ensure that packets will be
7575 * forwarded correctly according to the new state of the flow table.
7577 * This function must be called after *each* change to a flow table. See
7578 * the comment on table_update_taggable() for more information. */
7580 rule_invalidate(const struct rule_dpif *rule)
7582 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
7584 table_update_taggable(ofproto, rule->up.table_id);
7586 if (!ofproto->backer->need_revalidate) {
7587 struct table_dpif *table = &ofproto->tables[rule->up.table_id];
7589 if (table->other_table && rule->tag) {
7590 tag_set_add(&ofproto->backer->revalidate_set, rule->tag);
7592 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
7598 set_frag_handling(struct ofproto *ofproto_,
7599 enum ofp_config_flags frag_handling)
7601 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7602 if (frag_handling != OFPC_FRAG_REASM) {
7603 ofproto->backer->need_revalidate = REV_RECONFIGURE;
7611 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
7612 const struct flow *flow,
7613 const struct ofpact *ofpacts, size_t ofpacts_len)
7615 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7616 struct initial_vals initial_vals;
7617 struct odputil_keybuf keybuf;
7618 struct dpif_flow_stats stats;
7622 struct action_xlate_ctx ctx;
7623 uint64_t odp_actions_stub[1024 / 8];
7624 struct ofpbuf odp_actions;
7626 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
7627 odp_flow_key_from_flow(&key, flow,
7628 ofp_port_to_odp_port(ofproto, flow->in_port));
7630 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
7632 initial_vals.vlan_tci = flow->vlan_tci;
7633 initial_vals.tunnel_ip_tos = 0;
7634 action_xlate_ctx_init(&ctx, ofproto, flow, &initial_vals, NULL,
7635 packet_get_tcp_flags(packet, flow), packet);
7636 ctx.resubmit_stats = &stats;
7638 ofpbuf_use_stub(&odp_actions,
7639 odp_actions_stub, sizeof odp_actions_stub);
7640 xlate_actions(&ctx, ofpacts, ofpacts_len, &odp_actions);
7641 dpif_execute(ofproto->backer->dpif, key.data, key.size,
7642 odp_actions.data, odp_actions.size, packet);
7643 ofpbuf_uninit(&odp_actions);
7651 set_netflow(struct ofproto *ofproto_,
7652 const struct netflow_options *netflow_options)
7654 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7656 if (netflow_options) {
7657 if (!ofproto->netflow) {
7658 ofproto->netflow = netflow_create();
7660 return netflow_set_options(ofproto->netflow, netflow_options);
7662 netflow_destroy(ofproto->netflow);
7663 ofproto->netflow = NULL;
7669 get_netflow_ids(const struct ofproto *ofproto_,
7670 uint8_t *engine_type, uint8_t *engine_id)
7672 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7674 dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
7678 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
7680 if (!facet_is_controller_flow(facet) &&
7681 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
7682 struct subfacet *subfacet;
7683 struct ofexpired expired;
7685 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
7686 if (subfacet->path == SF_FAST_PATH) {
7687 struct dpif_flow_stats stats;
7689 subfacet_reinstall(subfacet, &stats);
7690 subfacet_update_stats(subfacet, &stats);
7694 expired.flow = facet->flow;
7695 expired.packet_count = facet->packet_count;
7696 expired.byte_count = facet->byte_count;
7697 expired.used = facet->used;
7698 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
7703 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
7705 struct facet *facet;
7707 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
7708 send_active_timeout(ofproto, facet);
7712 static struct ofproto_dpif *
7713 ofproto_dpif_lookup(const char *name)
7715 struct ofproto_dpif *ofproto;
7717 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
7718 hash_string(name, 0), &all_ofproto_dpifs) {
7719 if (!strcmp(ofproto->up.name, name)) {
7727 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
7728 const char *argv[], void *aux OVS_UNUSED)
7730 struct ofproto_dpif *ofproto;
7733 ofproto = ofproto_dpif_lookup(argv[1]);
7735 unixctl_command_reply_error(conn, "no such bridge");
7738 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
7740 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
7741 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
7745 unixctl_command_reply(conn, "table successfully flushed");
7749 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
7750 const char *argv[], void *aux OVS_UNUSED)
7752 struct ds ds = DS_EMPTY_INITIALIZER;
7753 const struct ofproto_dpif *ofproto;
7754 const struct mac_entry *e;
7756 ofproto = ofproto_dpif_lookup(argv[1]);
7758 unixctl_command_reply_error(conn, "no such bridge");
7762 ds_put_cstr(&ds, " port VLAN MAC Age\n");
7763 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
7764 struct ofbundle *bundle = e->port.p;
7765 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
7766 ofbundle_get_a_port(bundle)->odp_port,
7767 e->vlan, ETH_ADDR_ARGS(e->mac),
7768 mac_entry_age(ofproto->ml, e));
7770 unixctl_command_reply(conn, ds_cstr(&ds));
7775 struct action_xlate_ctx ctx;
7781 trace_format_rule(struct ds *result, uint8_t table_id, int level,
7782 const struct rule_dpif *rule)
7784 ds_put_char_multiple(result, '\t', level);
7786 ds_put_cstr(result, "No match\n");
7790 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
7791 table_id, ntohll(rule->up.flow_cookie));
7792 cls_rule_format(&rule->up.cr, result);
7793 ds_put_char(result, '\n');
7795 ds_put_char_multiple(result, '\t', level);
7796 ds_put_cstr(result, "OpenFlow ");
7797 ofpacts_format(rule->up.ofpacts, rule->up.ofpacts_len, result);
7798 ds_put_char(result, '\n');
7802 trace_format_flow(struct ds *result, int level, const char *title,
7803 struct trace_ctx *trace)
7805 ds_put_char_multiple(result, '\t', level);
7806 ds_put_format(result, "%s: ", title);
7807 if (flow_equal(&trace->ctx.flow, &trace->flow)) {
7808 ds_put_cstr(result, "unchanged");
7810 flow_format(result, &trace->ctx.flow);
7811 trace->flow = trace->ctx.flow;
7813 ds_put_char(result, '\n');
7817 trace_format_regs(struct ds *result, int level, const char *title,
7818 struct trace_ctx *trace)
7822 ds_put_char_multiple(result, '\t', level);
7823 ds_put_format(result, "%s:", title);
7824 for (i = 0; i < FLOW_N_REGS; i++) {
7825 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
7827 ds_put_char(result, '\n');
7831 trace_format_odp(struct ds *result, int level, const char *title,
7832 struct trace_ctx *trace)
7834 struct ofpbuf *odp_actions = trace->ctx.odp_actions;
7836 ds_put_char_multiple(result, '\t', level);
7837 ds_put_format(result, "%s: ", title);
7838 format_odp_actions(result, odp_actions->data, odp_actions->size);
7839 ds_put_char(result, '\n');
7843 trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
7845 struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx);
7846 struct ds *result = trace->result;
7848 ds_put_char(result, '\n');
7849 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
7850 trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace);
7851 trace_format_odp(result, ctx->recurse + 1, "Resubmitted odp", trace);
7852 trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule);
7856 trace_report(struct action_xlate_ctx *ctx, const char *s)
7858 struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx);
7859 struct ds *result = trace->result;
7861 ds_put_char_multiple(result, '\t', ctx->recurse);
7862 ds_put_cstr(result, s);
7863 ds_put_char(result, '\n');
7867 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
7868 void *aux OVS_UNUSED)
7870 const char *dpname = argv[1];
7871 struct ofproto_dpif *ofproto;
7872 struct ofpbuf odp_key;
7873 struct ofpbuf *packet;
7874 struct initial_vals initial_vals;
7880 ofpbuf_init(&odp_key, 0);
7883 ofproto = ofproto_dpif_lookup(dpname);
7885 unixctl_command_reply_error(conn, "Unknown ofproto (use ofproto/list "
7889 if (argc == 3 || (argc == 4 && !strcmp(argv[3], "-generate"))) {
7890 /* ofproto/trace dpname flow [-generate] */
7891 const char *flow_s = argv[2];
7892 const char *generate_s = argv[3];
7894 /* Allow 'flow_s' to be either a datapath flow or an OpenFlow-like
7895 * flow. We guess which type it is based on whether 'flow_s' contains
7896 * an '(', since a datapath flow always contains '(') but an
7897 * OpenFlow-like flow should not (in fact it's allowed but I believe
7898 * that's not documented anywhere).
7900 * An alternative would be to try to parse 'flow_s' both ways, but then
7901 * it would be tricky giving a sensible error message. After all, do
7902 * you just say "syntax error" or do you present both error messages?
7903 * Both choices seem lousy. */
7904 if (strchr(flow_s, '(')) {
7907 /* Convert string to datapath key. */
7908 ofpbuf_init(&odp_key, 0);
7909 error = odp_flow_key_from_string(flow_s, NULL, &odp_key);
7911 unixctl_command_reply_error(conn, "Bad flow syntax");
7915 /* The user might have specified the wrong ofproto but within the
7916 * same backer. That's OK, ofproto_receive() can find the right
7918 if (ofproto_receive(ofproto->backer, NULL, odp_key.data,
7919 odp_key.size, &flow, NULL, &ofproto, NULL,
7921 unixctl_command_reply_error(conn, "Invalid flow");
7924 ds_put_format(&result, "Bridge: %s\n", ofproto->up.name);
7928 error_s = parse_ofp_exact_flow(&flow, argv[2]);
7930 unixctl_command_reply_error(conn, error_s);
7935 initial_vals.vlan_tci = flow.vlan_tci;
7936 initial_vals.tunnel_ip_tos = flow.tunnel.ip_tos;
7939 /* Generate a packet, if requested. */
7941 packet = ofpbuf_new(0);
7942 flow_compose(packet, &flow);
7944 } else if (argc == 7) {
7945 /* ofproto/trace dpname priority tun_id in_port mark packet */
7946 const char *priority_s = argv[2];
7947 const char *tun_id_s = argv[3];
7948 const char *in_port_s = argv[4];
7949 const char *mark_s = argv[5];
7950 const char *packet_s = argv[6];
7951 uint32_t in_port = atoi(in_port_s);
7952 ovs_be64 tun_id = htonll(strtoull(tun_id_s, NULL, 0));
7953 uint32_t priority = atoi(priority_s);
7954 uint32_t mark = atoi(mark_s);
7957 msg = eth_from_hex(packet_s, &packet);
7959 unixctl_command_reply_error(conn, msg);
7963 ds_put_cstr(&result, "Packet: ");
7964 s = ofp_packet_to_string(packet->data, packet->size);
7965 ds_put_cstr(&result, s);
7968 flow_extract(packet, priority, mark, NULL, in_port, &flow);
7969 flow.tunnel.tun_id = tun_id;
7970 initial_vals.vlan_tci = flow.vlan_tci;
7971 initial_vals.tunnel_ip_tos = flow.tunnel.ip_tos;
7973 unixctl_command_reply_error(conn, "Bad command syntax");
7977 ofproto_trace(ofproto, &flow, packet, &initial_vals, &result);
7978 unixctl_command_reply(conn, ds_cstr(&result));
7981 ds_destroy(&result);
7982 ofpbuf_delete(packet);
7983 ofpbuf_uninit(&odp_key);
7987 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
7988 const struct ofpbuf *packet,
7989 const struct initial_vals *initial_vals, struct ds *ds)
7991 struct rule_dpif *rule;
7993 ds_put_cstr(ds, "Flow: ");
7994 flow_format(ds, flow);
7995 ds_put_char(ds, '\n');
7997 rule = rule_dpif_lookup(ofproto, flow);
7999 trace_format_rule(ds, 0, 0, rule);
8000 if (rule == ofproto->miss_rule) {
8001 ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
8002 } else if (rule == ofproto->no_packet_in_rule) {
8003 ds_put_cstr(ds, "\nNo match, packets dropped because "
8004 "OFPPC_NO_PACKET_IN is set on in_port.\n");
8008 uint64_t odp_actions_stub[1024 / 8];
8009 struct ofpbuf odp_actions;
8011 struct trace_ctx trace;
8014 tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
8017 ofpbuf_use_stub(&odp_actions,
8018 odp_actions_stub, sizeof odp_actions_stub);
8019 action_xlate_ctx_init(&trace.ctx, ofproto, flow, initial_vals,
8020 rule, tcp_flags, packet);
8021 trace.ctx.resubmit_hook = trace_resubmit;
8022 trace.ctx.report_hook = trace_report;
8023 xlate_actions(&trace.ctx, rule->up.ofpacts, rule->up.ofpacts_len,
8026 ds_put_char(ds, '\n');
8027 trace_format_flow(ds, 0, "Final flow", &trace);
8028 ds_put_cstr(ds, "Datapath actions: ");
8029 format_odp_actions(ds, odp_actions.data, odp_actions.size);
8030 ofpbuf_uninit(&odp_actions);
8032 if (trace.ctx.slow) {
8033 enum slow_path_reason slow;
8035 ds_put_cstr(ds, "\nThis flow is handled by the userspace "
8036 "slow path because it:");
8037 for (slow = trace.ctx.slow; slow; ) {
8038 enum slow_path_reason bit = rightmost_1bit(slow);
8042 ds_put_cstr(ds, "\n\t- Consists of CFM packets.");
8045 ds_put_cstr(ds, "\n\t- Consists of LACP packets.");
8048 ds_put_cstr(ds, "\n\t- Consists of STP packets.");
8051 ds_put_cstr(ds, "\n\t- Needs in-band special case "
8054 ds_put_cstr(ds, "\n\t (The datapath actions are "
8055 "incomplete--for complete actions, "
8056 "please supply a packet.)");
8059 case SLOW_CONTROLLER:
8060 ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages "
8061 "to the OpenFlow controller.");
8064 ds_put_cstr(ds, "\n\t- Needs more specific matching "
8065 "than the datapath supports.");
8072 if (slow & ~SLOW_MATCH) {
8073 ds_put_cstr(ds, "\nThe datapath actions above do not reflect "
8074 "the special slow-path processing.");
8081 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
8082 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
8085 unixctl_command_reply(conn, NULL);
8089 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
8090 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
8093 unixctl_command_reply(conn, NULL);
8096 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
8097 * 'reply' describing the results. */
8099 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
8101 struct facet *facet;
8105 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
8106 if (!facet_check_consistency(facet)) {
8111 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
8115 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
8116 ofproto->up.name, errors);
8118 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
8123 ofproto_dpif_self_check(struct unixctl_conn *conn,
8124 int argc, const char *argv[], void *aux OVS_UNUSED)
8126 struct ds reply = DS_EMPTY_INITIALIZER;
8127 struct ofproto_dpif *ofproto;
8130 ofproto = ofproto_dpif_lookup(argv[1]);
8132 unixctl_command_reply_error(conn, "Unknown ofproto (use "
8133 "ofproto/list for help)");
8136 ofproto_dpif_self_check__(ofproto, &reply);
8138 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
8139 ofproto_dpif_self_check__(ofproto, &reply);
8143 unixctl_command_reply(conn, ds_cstr(&reply));
8147 /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list
8148 * of the 'ofproto_shash' nodes. It is the responsibility of the caller
8149 * to destroy 'ofproto_shash' and free the returned value. */
8150 static const struct shash_node **
8151 get_ofprotos(struct shash *ofproto_shash)
8153 const struct ofproto_dpif *ofproto;
8155 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
8156 char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
8157 shash_add_nocopy(ofproto_shash, name, ofproto);
8160 return shash_sort(ofproto_shash);
8164 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
8165 const char *argv[] OVS_UNUSED,
8166 void *aux OVS_UNUSED)
8168 struct ds ds = DS_EMPTY_INITIALIZER;
8169 struct shash ofproto_shash;
8170 const struct shash_node **sorted_ofprotos;
8173 shash_init(&ofproto_shash);
8174 sorted_ofprotos = get_ofprotos(&ofproto_shash);
8175 for (i = 0; i < shash_count(&ofproto_shash); i++) {
8176 const struct shash_node *node = sorted_ofprotos[i];
8177 ds_put_format(&ds, "%s\n", node->name);
8180 shash_destroy(&ofproto_shash);
8181 free(sorted_ofprotos);
8183 unixctl_command_reply(conn, ds_cstr(&ds));
8188 show_dp_format(const struct ofproto_dpif *ofproto, struct ds *ds)
8190 const struct shash_node **ports;
8192 struct avg_subfacet_rates lifetime;
8193 unsigned long long int minutes;
8194 const int min_ms = 60 * 1000; /* milliseconds in one minute. */
8196 minutes = (time_msec() - ofproto->created) / min_ms;
8199 lifetime.add_rate = (double)ofproto->total_subfacet_add_count
8201 lifetime.del_rate = (double)ofproto->total_subfacet_del_count
8204 lifetime.add_rate = 0.0;
8205 lifetime.del_rate = 0.0;
8208 ds_put_format(ds, "%s (%s):\n", ofproto->up.name,
8209 dpif_name(ofproto->backer->dpif));
8211 "\tlookups: hit:%"PRIu64" missed:%"PRIu64"\n",
8212 ofproto->n_hit, ofproto->n_missed);
8213 ds_put_format(ds, "\tflows: cur: %zu, avg: %5.3f, max: %d,"
8214 " life span: %llu(ms)\n",
8215 hmap_count(&ofproto->subfacets),
8216 avg_subfacet_count(ofproto),
8217 ofproto->max_n_subfacet,
8218 avg_subfacet_life_span(ofproto));
8219 if (minutes >= 60) {
8220 show_dp_rates(ds, "\t\thourly avg:", &ofproto->hourly);
8222 if (minutes >= 60 * 24) {
8223 show_dp_rates(ds, "\t\tdaily avg:", &ofproto->daily);
8225 show_dp_rates(ds, "\t\toverall avg:", &lifetime);
8227 ports = shash_sort(&ofproto->up.port_by_name);
8228 for (i = 0; i < shash_count(&ofproto->up.port_by_name); i++) {
8229 const struct shash_node *node = ports[i];
8230 struct ofport *ofport = node->data;
8231 const char *name = netdev_get_name(ofport->netdev);
8232 const char *type = netdev_get_type(ofport->netdev);
8235 ds_put_format(ds, "\t%s %u/", name, ofport->ofp_port);
8237 odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
8238 if (odp_port != OVSP_NONE) {
8239 ds_put_format(ds, "%"PRIu32":", odp_port);
8241 ds_put_cstr(ds, "none:");
8244 if (strcmp(type, "system")) {
8245 struct netdev *netdev;
8248 ds_put_format(ds, " (%s", type);
8250 error = netdev_open(name, type, &netdev);
8255 error = netdev_get_config(netdev, &config);
8257 const struct smap_node **nodes;
8260 nodes = smap_sort(&config);
8261 for (i = 0; i < smap_count(&config); i++) {
8262 const struct smap_node *node = nodes[i];
8263 ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
8264 node->key, node->value);
8268 smap_destroy(&config);
8270 netdev_close(netdev);
8272 ds_put_char(ds, ')');
8274 ds_put_char(ds, '\n');
8280 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc,
8281 const char *argv[], void *aux OVS_UNUSED)
8283 struct ds ds = DS_EMPTY_INITIALIZER;
8284 const struct ofproto_dpif *ofproto;
8288 for (i = 1; i < argc; i++) {
8289 ofproto = ofproto_dpif_lookup(argv[i]);
8291 ds_put_format(&ds, "Unknown bridge %s (use dpif/dump-dps "
8292 "for help)", argv[i]);
8293 unixctl_command_reply_error(conn, ds_cstr(&ds));
8296 show_dp_format(ofproto, &ds);
8299 struct shash ofproto_shash;
8300 const struct shash_node **sorted_ofprotos;
8303 shash_init(&ofproto_shash);
8304 sorted_ofprotos = get_ofprotos(&ofproto_shash);
8305 for (i = 0; i < shash_count(&ofproto_shash); i++) {
8306 const struct shash_node *node = sorted_ofprotos[i];
8307 show_dp_format(node->data, &ds);
8310 shash_destroy(&ofproto_shash);
8311 free(sorted_ofprotos);
8314 unixctl_command_reply(conn, ds_cstr(&ds));
8319 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
8320 int argc OVS_UNUSED, const char *argv[],
8321 void *aux OVS_UNUSED)
8323 struct ds ds = DS_EMPTY_INITIALIZER;
8324 const struct ofproto_dpif *ofproto;
8325 struct subfacet *subfacet;
8327 ofproto = ofproto_dpif_lookup(argv[1]);
8329 unixctl_command_reply_error(conn, "no such bridge");
8333 update_stats(ofproto->backer);
8335 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
8336 odp_flow_key_format(subfacet->key, subfacet->key_len, &ds);
8338 ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:",
8339 subfacet->dp_packet_count, subfacet->dp_byte_count);
8340 if (subfacet->used) {
8341 ds_put_format(&ds, "%.3fs",
8342 (time_msec() - subfacet->used) / 1000.0);
8344 ds_put_format(&ds, "never");
8346 if (subfacet->facet->tcp_flags) {
8347 ds_put_cstr(&ds, ", flags:");
8348 packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags);
8351 ds_put_cstr(&ds, ", actions:");
8352 if (subfacet->slow) {
8353 uint64_t slow_path_stub[128 / 8];
8354 const struct nlattr *actions;
8357 compose_slow_path(ofproto, &subfacet->facet->flow, subfacet->slow,
8358 slow_path_stub, sizeof slow_path_stub,
8359 &actions, &actions_len);
8360 format_odp_actions(&ds, actions, actions_len);
8362 format_odp_actions(&ds, subfacet->actions, subfacet->actions_len);
8364 ds_put_char(&ds, '\n');
8367 unixctl_command_reply(conn, ds_cstr(&ds));
8372 ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn,
8373 int argc OVS_UNUSED, const char *argv[],
8374 void *aux OVS_UNUSED)
8376 struct ds ds = DS_EMPTY_INITIALIZER;
8377 struct ofproto_dpif *ofproto;
8379 ofproto = ofproto_dpif_lookup(argv[1]);
8381 unixctl_command_reply_error(conn, "no such bridge");
8385 flush(&ofproto->up);
8387 unixctl_command_reply(conn, ds_cstr(&ds));
8392 ofproto_dpif_unixctl_init(void)
8394 static bool registered;
8400 unixctl_command_register(
8402 "bridge {priority tun_id in_port mark packet | odp_flow [-generate]}",
8403 2, 6, ofproto_unixctl_trace, NULL);
8404 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
8405 ofproto_unixctl_fdb_flush, NULL);
8406 unixctl_command_register("fdb/show", "bridge", 1, 1,
8407 ofproto_unixctl_fdb_show, NULL);
8408 unixctl_command_register("ofproto/clog", "", 0, 0,
8409 ofproto_dpif_clog, NULL);
8410 unixctl_command_register("ofproto/unclog", "", 0, 0,
8411 ofproto_dpif_unclog, NULL);
8412 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
8413 ofproto_dpif_self_check, NULL);
8414 unixctl_command_register("dpif/dump-dps", "", 0, 0,
8415 ofproto_unixctl_dpif_dump_dps, NULL);
8416 unixctl_command_register("dpif/show", "[bridge]", 0, INT_MAX,
8417 ofproto_unixctl_dpif_show, NULL);
8418 unixctl_command_register("dpif/dump-flows", "bridge", 1, 1,
8419 ofproto_unixctl_dpif_dump_flows, NULL);
8420 unixctl_command_register("dpif/del-flows", "bridge", 1, 1,
8421 ofproto_unixctl_dpif_del_flows, NULL);
8424 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
8426 * This is deprecated. It is only for compatibility with broken device drivers
8427 * in old versions of Linux that do not properly support VLANs when VLAN
8428 * devices are not used. When broken device drivers are no longer in
8429 * widespread use, we will delete these interfaces. */
8432 set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid)
8434 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
8435 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
8437 if (realdev_ofp_port == ofport->realdev_ofp_port
8438 && vid == ofport->vlandev_vid) {
8442 ofproto->backer->need_revalidate = REV_RECONFIGURE;
8444 if (ofport->realdev_ofp_port) {
8447 if (realdev_ofp_port && ofport->bundle) {
8448 /* vlandevs are enslaved to their realdevs, so they are not allowed to
8449 * themselves be part of a bundle. */
8450 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
8453 ofport->realdev_ofp_port = realdev_ofp_port;
8454 ofport->vlandev_vid = vid;
8456 if (realdev_ofp_port) {
8457 vsp_add(ofport, realdev_ofp_port, vid);
8464 hash_realdev_vid(uint16_t realdev_ofp_port, int vid)
8466 return hash_2words(realdev_ofp_port, vid);
8469 /* Returns the ODP port number of the Linux VLAN device that corresponds to
8470 * 'vlan_tci' on the network device with port number 'realdev_odp_port' in
8471 * 'ofproto'. For example, given 'realdev_odp_port' of eth0 and 'vlan_tci' 9,
8472 * it would return the port number of eth0.9.
8474 * Unless VLAN splinters are enabled for port 'realdev_odp_port', this
8475 * function just returns its 'realdev_odp_port' argument. */
8477 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
8478 uint32_t realdev_odp_port, ovs_be16 vlan_tci)
8480 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
8481 uint16_t realdev_ofp_port;
8482 int vid = vlan_tci_to_vid(vlan_tci);
8483 const struct vlan_splinter *vsp;
8485 realdev_ofp_port = odp_port_to_ofp_port(ofproto, realdev_odp_port);
8486 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
8487 hash_realdev_vid(realdev_ofp_port, vid),
8488 &ofproto->realdev_vid_map) {
8489 if (vsp->realdev_ofp_port == realdev_ofp_port
8490 && vsp->vid == vid) {
8491 return ofp_port_to_odp_port(ofproto, vsp->vlandev_ofp_port);
8495 return realdev_odp_port;
8498 static struct vlan_splinter *
8499 vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port)
8501 struct vlan_splinter *vsp;
8503 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, hash_int(vlandev_ofp_port, 0),
8504 &ofproto->vlandev_map) {
8505 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
8513 /* Returns the OpenFlow port number of the "real" device underlying the Linux
8514 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
8515 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
8516 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
8517 * eth0 and store 9 in '*vid'.
8519 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
8520 * VLAN device. Unless VLAN splinters are enabled, this is what this function
8523 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
8524 uint16_t vlandev_ofp_port, int *vid)
8526 if (!hmap_is_empty(&ofproto->vlandev_map)) {
8527 const struct vlan_splinter *vsp;
8529 vsp = vlandev_find(ofproto, vlandev_ofp_port);
8534 return vsp->realdev_ofp_port;
8540 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
8541 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
8542 * 'flow->in_port' to the "real" device backing the VLAN device, sets
8543 * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is
8544 * always the case unless VLAN splinters are enabled), returns false without
8545 * making any changes. */
8547 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
8552 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port, &vid);
8557 /* Cause the flow to be processed as if it came in on the real device with
8558 * the VLAN device's VLAN ID. */
8559 flow->in_port = realdev;
8560 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
8565 vsp_remove(struct ofport_dpif *port)
8567 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
8568 struct vlan_splinter *vsp;
8570 vsp = vlandev_find(ofproto, port->up.ofp_port);
8572 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
8573 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
8576 port->realdev_ofp_port = 0;
8578 VLOG_ERR("missing vlan device record");
8583 vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid)
8585 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
8587 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
8588 && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid))
8589 == realdev_ofp_port)) {
8590 struct vlan_splinter *vsp;
8592 vsp = xmalloc(sizeof *vsp);
8593 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
8594 hash_int(port->up.ofp_port, 0));
8595 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
8596 hash_realdev_vid(realdev_ofp_port, vid));
8597 vsp->realdev_ofp_port = realdev_ofp_port;
8598 vsp->vlandev_ofp_port = port->up.ofp_port;
8601 port->realdev_ofp_port = realdev_ofp_port;
8603 VLOG_ERR("duplicate vlan device record");
8608 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
8610 const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
8611 return ofport ? ofport->odp_port : OVSP_NONE;
8614 static struct ofport_dpif *
8615 odp_port_to_ofport(const struct dpif_backer *backer, uint32_t odp_port)
8617 struct ofport_dpif *port;
8619 HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node,
8620 hash_int(odp_port, 0),
8621 &backer->odp_to_ofport_map) {
8622 if (port->odp_port == odp_port) {
8631 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
8633 struct ofport_dpif *port;
8635 port = odp_port_to_ofport(ofproto->backer, odp_port);
8636 if (port && &ofproto->up == port->up.ofproto) {
8637 return port->up.ofp_port;
8642 static unsigned long long int
8643 avg_subfacet_life_span(const struct ofproto_dpif *ofproto)
8645 unsigned long long int dc;
8646 unsigned long long int avg;
8648 dc = ofproto->total_subfacet_del_count + ofproto->subfacet_del_count;
8649 avg = dc ? ofproto->total_subfacet_life_span / dc : 0;
8655 avg_subfacet_count(const struct ofproto_dpif *ofproto)
8659 if (ofproto->n_update_stats) {
8660 avg_c = (double)ofproto->total_subfacet_count
8661 / ofproto->n_update_stats;
8668 show_dp_rates(struct ds *ds, const char *heading,
8669 const struct avg_subfacet_rates *rates)
8671 ds_put_format(ds, "%s add rate: %5.3f/min, del rate: %5.3f/min\n",
8672 heading, rates->add_rate, rates->del_rate);
8676 update_max_subfacet_count(struct ofproto_dpif *ofproto)
8678 ofproto->max_n_subfacet = MAX(ofproto->max_n_subfacet,
8679 hmap_count(&ofproto->subfacets));
8682 /* Compute exponentially weighted moving average, adding 'new' as the newest,
8683 * most heavily weighted element. 'base' designates the rate of decay: after
8684 * 'base' further updates, 'new''s weight in the EWMA decays to about 1/e
8687 exp_mavg(double *avg, int base, double new)
8689 *avg = (*avg * (base - 1) + new) / base;
8693 update_moving_averages(struct ofproto_dpif *ofproto)
8695 const int min_ms = 60 * 1000; /* milliseconds in one minute. */
8697 /* Update hourly averages on the minute boundaries. */
8698 if (time_msec() - ofproto->last_minute >= min_ms) {
8699 exp_mavg(&ofproto->hourly.add_rate, 60, ofproto->subfacet_add_count);
8700 exp_mavg(&ofproto->hourly.del_rate, 60, ofproto->subfacet_del_count);
8702 /* Update daily averages on the hour boundaries. */
8703 if ((ofproto->last_minute - ofproto->created) / min_ms % 60 == 59) {
8704 exp_mavg(&ofproto->daily.add_rate, 24, ofproto->hourly.add_rate);
8705 exp_mavg(&ofproto->daily.del_rate, 24, ofproto->hourly.del_rate);
8708 ofproto->total_subfacet_add_count += ofproto->subfacet_add_count;
8709 ofproto->total_subfacet_del_count += ofproto->subfacet_del_count;
8710 ofproto->subfacet_add_count = 0;
8711 ofproto->subfacet_del_count = 0;
8712 ofproto->last_minute += min_ms;
8717 dpif_stats_update_hit_count(struct ofproto_dpif *ofproto, uint64_t delta)
8719 ofproto->n_hit += delta;
8722 const struct ofproto_class ofproto_dpif_class = {
8757 port_is_lacp_current,
8758 NULL, /* rule_choose_table */
8765 rule_modify_actions,
8776 get_stp_port_status,
8783 is_mirror_output_bundle,
8784 forward_bpdu_changed,
8785 set_mac_table_config,