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. */
374 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
375 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
379 * These should be essentially identical for every subfacet in a facet, but
380 * may differ in trivial ways due to VLAN splinters. */
381 size_t actions_len; /* Number of bytes in actions[]. */
382 struct nlattr *actions; /* Datapath actions. */
384 enum slow_path_reason slow; /* 0 if fast path may be used. */
385 enum subfacet_path path; /* Installed in datapath? */
387 /* Initial values of the packet that may be needed later. */
388 struct initial_vals initial_vals;
390 /* Datapath port the packet arrived on. This is needed to remove
391 * flows for ports that are no longer part of the bridge. Since the
392 * flow definition only has the OpenFlow port number and the port is
393 * no longer part of the bridge, we can't determine the datapath port
394 * number needed to delete the flow from the datapath. */
395 uint32_t odp_in_port;
398 #define SUBFACET_DESTROY_MAX_BATCH 50
400 static struct subfacet *subfacet_create(struct facet *, struct flow_miss *miss,
402 static struct subfacet *subfacet_find(struct ofproto_dpif *,
403 const struct nlattr *key, size_t key_len,
405 static void subfacet_destroy(struct subfacet *);
406 static void subfacet_destroy__(struct subfacet *);
407 static void subfacet_destroy_batch(struct ofproto_dpif *,
408 struct subfacet **, int n);
409 static void subfacet_reset_dp_stats(struct subfacet *,
410 struct dpif_flow_stats *);
411 static void subfacet_update_time(struct subfacet *, long long int used);
412 static void subfacet_update_stats(struct subfacet *,
413 const struct dpif_flow_stats *);
414 static void subfacet_make_actions(struct subfacet *,
415 const struct ofpbuf *packet,
416 struct ofpbuf *odp_actions);
417 static int subfacet_install(struct subfacet *,
418 const struct nlattr *actions, size_t actions_len,
419 struct dpif_flow_stats *, enum slow_path_reason);
420 static void subfacet_uninstall(struct subfacet *);
422 static enum subfacet_path subfacet_want_path(enum slow_path_reason);
424 /* An exact-match instantiation of an OpenFlow flow.
426 * A facet associates a "struct flow", which represents the Open vSwitch
427 * userspace idea of an exact-match flow, with one or more subfacets. Each
428 * subfacet tracks the datapath's idea of the exact-match flow equivalent to
429 * the facet. When the kernel module (or other dpif implementation) and Open
430 * vSwitch userspace agree on the definition of a flow key, there is exactly
431 * one subfacet per facet. If the dpif implementation supports more-specific
432 * flow matching than userspace, however, a facet can have more than one
433 * subfacet, each of which corresponds to some distinction in flow that
434 * userspace simply doesn't understand.
436 * Flow expiration works in terms of subfacets, so a facet must have at least
437 * one subfacet or it will never expire, leaking memory. */
440 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
441 struct list list_node; /* In owning rule's 'facets' list. */
442 struct rule_dpif *rule; /* Owning rule. */
445 struct list subfacets;
446 long long int used; /* Time last used; time created if not used. */
453 * - Do include packets and bytes sent "by hand", e.g. with
456 * - Do include packets and bytes that were obtained from the datapath
457 * when a subfacet's statistics were reset (e.g. dpif_flow_put() with
458 * DPIF_FP_ZERO_STATS).
460 * - Do not include packets or bytes that can be obtained from the
461 * datapath for any existing subfacet.
463 uint64_t packet_count; /* Number of packets received. */
464 uint64_t byte_count; /* Number of bytes received. */
466 /* Resubmit statistics. */
467 uint64_t prev_packet_count; /* Number of packets from last stats push. */
468 uint64_t prev_byte_count; /* Number of bytes from last stats push. */
469 long long int prev_used; /* Used time from last stats push. */
472 uint64_t accounted_bytes; /* Bytes processed by facet_account(). */
473 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
474 uint8_t tcp_flags; /* TCP flags seen for this 'rule'. */
476 /* Properties of datapath actions.
478 * Every subfacet has its own actions because actions can differ slightly
479 * between splintered and non-splintered subfacets due to the VLAN tag
480 * being initially different (present vs. absent). All of them have these
481 * properties in common so we just store one copy of them here. */
482 bool has_learn; /* Actions include NXAST_LEARN? */
483 bool has_normal; /* Actions output to OFPP_NORMAL? */
484 bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */
485 tag_type tags; /* Tags that would require revalidation. */
486 mirror_mask_t mirrors; /* Bitmap of dependent mirrors. */
488 /* Storage for a single subfacet, to reduce malloc() time and space
489 * overhead. (A facet always has at least one subfacet and in the common
490 * case has exactly one subfacet. However, 'one_subfacet' may not
491 * always be valid, since it could have been removed after newer
492 * subfacets were pushed onto the 'subfacets' list.) */
493 struct subfacet one_subfacet;
496 static struct facet *facet_create(struct rule_dpif *,
497 const struct flow *, uint32_t hash);
498 static void facet_remove(struct facet *);
499 static void facet_free(struct facet *);
501 static struct facet *facet_find(struct ofproto_dpif *,
502 const struct flow *, uint32_t hash);
503 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
504 const struct flow *, uint32_t hash);
505 static void facet_revalidate(struct facet *);
506 static bool facet_check_consistency(struct facet *);
508 static void facet_flush_stats(struct facet *);
510 static void facet_update_time(struct facet *, long long int used);
511 static void facet_reset_counters(struct facet *);
512 static void facet_push_stats(struct facet *);
513 static void facet_learn(struct facet *);
514 static void facet_account(struct facet *);
516 static struct subfacet *facet_get_subfacet(struct facet *);
518 static bool facet_is_controller_flow(struct facet *);
521 struct hmap_node odp_port_node; /* In dpif_backer's "odp_to_ofport_map". */
525 struct ofbundle *bundle; /* Bundle that contains this port, if any. */
526 struct list bundle_node; /* In struct ofbundle's "ports" list. */
527 struct cfm *cfm; /* Connectivity Fault Management, if any. */
528 tag_type tag; /* Tag associated with this port. */
529 bool may_enable; /* May be enabled in bonds. */
530 long long int carrier_seq; /* Carrier status changes. */
531 struct tnl_port *tnl_port; /* Tunnel handle, or null. */
534 struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */
535 enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */
536 long long int stp_state_entered;
538 struct hmap priorities; /* Map of attached 'priority_to_dscp's. */
540 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
542 * This is deprecated. It is only for compatibility with broken device
543 * drivers in old versions of Linux that do not properly support VLANs when
544 * VLAN devices are not used. When broken device drivers are no longer in
545 * widespread use, we will delete these interfaces. */
546 uint16_t realdev_ofp_port;
550 /* Node in 'ofport_dpif''s 'priorities' map. Used to maintain a map from
551 * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
552 * traffic egressing the 'ofport' with that priority should be marked with. */
553 struct priority_to_dscp {
554 struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'priorities' map. */
555 uint32_t priority; /* Priority of this queue (see struct flow). */
557 uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */
560 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
562 * This is deprecated. It is only for compatibility with broken device drivers
563 * in old versions of Linux that do not properly support VLANs when VLAN
564 * devices are not used. When broken device drivers are no longer in
565 * widespread use, we will delete these interfaces. */
566 struct vlan_splinter {
567 struct hmap_node realdev_vid_node;
568 struct hmap_node vlandev_node;
569 uint16_t realdev_ofp_port;
570 uint16_t vlandev_ofp_port;
574 static uint32_t vsp_realdev_to_vlandev(const struct ofproto_dpif *,
575 uint32_t realdev, ovs_be16 vlan_tci);
576 static bool vsp_adjust_flow(const struct ofproto_dpif *, struct flow *);
577 static void vsp_remove(struct ofport_dpif *);
578 static void vsp_add(struct ofport_dpif *, uint16_t realdev_ofp_port, int vid);
580 static uint32_t ofp_port_to_odp_port(const struct ofproto_dpif *,
582 static uint16_t odp_port_to_ofp_port(const struct ofproto_dpif *,
585 static struct ofport_dpif *
586 ofport_dpif_cast(const struct ofport *ofport)
588 ovs_assert(ofport->ofproto->ofproto_class == &ofproto_dpif_class);
589 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
592 static void port_run(struct ofport_dpif *);
593 static void port_run_fast(struct ofport_dpif *);
594 static void port_wait(struct ofport_dpif *);
595 static int set_cfm(struct ofport *, const struct cfm_settings *);
596 static void ofport_clear_priorities(struct ofport_dpif *);
598 struct dpif_completion {
599 struct list list_node;
600 struct ofoperation *op;
603 /* Extra information about a classifier table.
604 * Currently used just for optimized flow revalidation. */
606 /* If either of these is nonnull, then this table has a form that allows
607 * flows to be tagged to avoid revalidating most flows for the most common
608 * kinds of flow table changes. */
609 struct cls_table *catchall_table; /* Table that wildcards all fields. */
610 struct cls_table *other_table; /* Table with any other wildcard set. */
611 uint32_t basis; /* Keeps each table's tags separate. */
614 /* Reasons that we might need to revalidate every facet, and corresponding
617 * A value of 0 means that there is no need to revalidate.
619 * It would be nice to have some cleaner way to integrate with coverage
620 * counters, but with only a few reasons I guess this is good enough for
622 enum revalidate_reason {
623 REV_RECONFIGURE = 1, /* Switch configuration changed. */
624 REV_STP, /* Spanning tree protocol port status change. */
625 REV_PORT_TOGGLED, /* Port enabled or disabled by CFM, LACP, ...*/
626 REV_FLOW_TABLE, /* Flow table changed. */
627 REV_INCONSISTENCY /* Facet self-check failed. */
629 COVERAGE_DEFINE(rev_reconfigure);
630 COVERAGE_DEFINE(rev_stp);
631 COVERAGE_DEFINE(rev_port_toggled);
632 COVERAGE_DEFINE(rev_flow_table);
633 COVERAGE_DEFINE(rev_inconsistency);
635 /* Drop keys are odp flow keys which have drop flows installed in the kernel.
636 * These are datapath flows which have no associated ofproto, if they did we
637 * would use facets. */
639 struct hmap_node hmap_node;
644 /* All datapaths of a given type share a single dpif backer instance. */
649 struct timer next_expiration;
650 struct hmap odp_to_ofport_map; /* ODP port to ofport mapping. */
652 struct simap tnl_backers; /* Set of dpif ports backing tunnels. */
654 /* Facet revalidation flags applying to facets which use this backer. */
655 enum revalidate_reason need_revalidate; /* Revalidate every facet. */
656 struct tag_set revalidate_set; /* Revalidate only matching facets. */
658 struct hmap drop_keys; /* Set of dropped odp keys. */
661 /* All existing ofproto_backer instances, indexed by ofproto->up.type. */
662 static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers);
664 static void drop_key_clear(struct dpif_backer *);
665 static struct ofport_dpif *
666 odp_port_to_ofport(const struct dpif_backer *, uint32_t odp_port);
668 static void dpif_stats_update_hit_count(struct ofproto_dpif *ofproto,
671 struct ofproto_dpif {
672 struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
674 struct dpif_backer *backer;
676 /* Special OpenFlow rules. */
677 struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */
678 struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */
684 struct netflow *netflow;
685 struct dpif_sflow *sflow;
686 struct hmap bundles; /* Contains "struct ofbundle"s. */
687 struct mac_learning *ml;
688 struct ofmirror *mirrors[MAX_MIRRORS];
690 bool has_bonded_bundles;
694 struct hmap subfacets;
695 struct governor *governor;
696 long long int consistency_rl;
699 struct table_dpif tables[N_TABLES];
701 /* Support for debugging async flow mods. */
702 struct list completions;
704 bool has_bundle_action; /* True when the first bundle action appears. */
705 struct netdev_stats stats; /* To account packets generated and consumed in
710 long long int stp_last_tick;
712 /* VLAN splinters. */
713 struct hmap realdev_vid_map; /* (realdev,vid) -> vlandev. */
714 struct hmap vlandev_map; /* vlandev -> (realdev,vid). */
717 struct sset ports; /* Set of standard port names. */
718 struct sset ghost_ports; /* Ports with no datapath port. */
719 struct sset port_poll_set; /* Queued names for port_poll() reply. */
720 int port_poll_errno; /* Last errno for port_poll() reply. */
722 /* Per ofproto's dpif stats. */
727 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
728 * for debugging the asynchronous flow_mod implementation.) */
731 /* All existing ofproto_dpif instances, indexed by ->up.name. */
732 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
734 static void ofproto_dpif_unixctl_init(void);
736 static struct ofproto_dpif *
737 ofproto_dpif_cast(const struct ofproto *ofproto)
739 ovs_assert(ofproto->ofproto_class == &ofproto_dpif_class);
740 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
743 static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *,
745 static struct ofport_dpif *get_odp_port(const struct ofproto_dpif *,
747 static void ofproto_trace(struct ofproto_dpif *, const struct flow *,
748 const struct ofpbuf *,
749 const struct initial_vals *, struct ds *);
751 /* Packet processing. */
752 static void update_learning_table(struct ofproto_dpif *,
753 const struct flow *, int vlan,
756 #define FLOW_MISS_MAX_BATCH 50
757 static int handle_upcalls(struct dpif_backer *, unsigned int max_batch);
759 /* Flow expiration. */
760 static int expire(struct dpif_backer *);
763 static void send_netflow_active_timeouts(struct ofproto_dpif *);
766 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
767 static size_t compose_sflow_action(const struct ofproto_dpif *,
768 struct ofpbuf *odp_actions,
769 const struct flow *, uint32_t odp_port);
770 static void add_mirror_actions(struct action_xlate_ctx *ctx,
771 const struct flow *flow);
772 /* Global variables. */
773 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
775 /* Initial mappings of port to bridge mappings. */
776 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
778 /* Factory functions. */
781 init(const struct shash *iface_hints)
783 struct shash_node *node;
785 /* Make a local copy, since we don't own 'iface_hints' elements. */
786 SHASH_FOR_EACH(node, iface_hints) {
787 const struct iface_hint *orig_hint = node->data;
788 struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
790 new_hint->br_name = xstrdup(orig_hint->br_name);
791 new_hint->br_type = xstrdup(orig_hint->br_type);
792 new_hint->ofp_port = orig_hint->ofp_port;
794 shash_add(&init_ofp_ports, node->name, new_hint);
799 enumerate_types(struct sset *types)
801 dp_enumerate_types(types);
805 enumerate_names(const char *type, struct sset *names)
807 struct ofproto_dpif *ofproto;
810 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
811 if (strcmp(type, ofproto->up.type)) {
814 sset_add(names, ofproto->up.name);
821 del(const char *type, const char *name)
826 error = dpif_open(name, type, &dpif);
828 error = dpif_delete(dpif);
835 port_open_type(const char *datapath_type, const char *port_type)
837 return dpif_port_open_type(datapath_type, port_type);
840 /* Type functions. */
842 static struct ofproto_dpif *
843 lookup_ofproto_dpif_by_port_name(const char *name)
845 struct ofproto_dpif *ofproto;
847 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
848 if (sset_contains(&ofproto->ports, name)) {
857 type_run(const char *type)
859 struct dpif_backer *backer;
863 backer = shash_find_data(&all_dpif_backers, type);
865 /* This is not necessarily a problem, since backers are only
866 * created on demand. */
870 dpif_run(backer->dpif);
872 if (backer->need_revalidate
873 || !tag_set_is_empty(&backer->revalidate_set)) {
874 struct tag_set revalidate_set = backer->revalidate_set;
875 bool need_revalidate = backer->need_revalidate;
876 struct ofproto_dpif *ofproto;
877 struct simap_node *node;
878 struct simap tmp_backers;
880 /* Handle tunnel garbage collection. */
881 simap_init(&tmp_backers);
882 simap_swap(&backer->tnl_backers, &tmp_backers);
884 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
885 struct ofport_dpif *iter;
887 if (backer != ofproto->backer) {
891 HMAP_FOR_EACH (iter, up.hmap_node, &ofproto->up.ports) {
894 if (!iter->tnl_port) {
898 dp_port = netdev_vport_get_dpif_port(iter->up.netdev);
899 node = simap_find(&tmp_backers, dp_port);
901 simap_put(&backer->tnl_backers, dp_port, node->data);
902 simap_delete(&tmp_backers, node);
903 node = simap_find(&backer->tnl_backers, dp_port);
905 node = simap_find(&backer->tnl_backers, dp_port);
907 uint32_t odp_port = UINT32_MAX;
909 if (!dpif_port_add(backer->dpif, iter->up.netdev,
911 simap_put(&backer->tnl_backers, dp_port, odp_port);
912 node = simap_find(&backer->tnl_backers, dp_port);
917 iter->odp_port = node ? node->data : OVSP_NONE;
918 if (tnl_port_reconfigure(&iter->up, iter->odp_port,
920 backer->need_revalidate = REV_RECONFIGURE;
925 SIMAP_FOR_EACH (node, &tmp_backers) {
926 dpif_port_del(backer->dpif, node->data);
928 simap_destroy(&tmp_backers);
930 switch (backer->need_revalidate) {
931 case REV_RECONFIGURE: COVERAGE_INC(rev_reconfigure); break;
932 case REV_STP: COVERAGE_INC(rev_stp); break;
933 case REV_PORT_TOGGLED: COVERAGE_INC(rev_port_toggled); break;
934 case REV_FLOW_TABLE: COVERAGE_INC(rev_flow_table); break;
935 case REV_INCONSISTENCY: COVERAGE_INC(rev_inconsistency); break;
938 if (backer->need_revalidate) {
939 /* Clear the drop_keys in case we should now be accepting some
940 * formerly dropped flows. */
941 drop_key_clear(backer);
944 /* Clear the revalidation flags. */
945 tag_set_init(&backer->revalidate_set);
946 backer->need_revalidate = 0;
948 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
949 struct facet *facet, *next;
951 if (ofproto->backer != backer) {
955 HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &ofproto->facets) {
957 || tag_set_intersects(&revalidate_set, facet->tags)) {
958 facet_revalidate(facet);
964 if (timer_expired(&backer->next_expiration)) {
965 int delay = expire(backer);
966 timer_set_duration(&backer->next_expiration, delay);
969 /* Check for port changes in the dpif. */
970 while ((error = dpif_port_poll(backer->dpif, &devname)) == 0) {
971 struct ofproto_dpif *ofproto;
972 struct dpif_port port;
974 /* Don't report on the datapath's device. */
975 if (!strcmp(devname, dpif_base_name(backer->dpif))) {
979 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
980 &all_ofproto_dpifs) {
981 if (simap_contains(&ofproto->backer->tnl_backers, devname)) {
986 ofproto = lookup_ofproto_dpif_by_port_name(devname);
987 if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
988 /* The port was removed. If we know the datapath,
989 * report it through poll_set(). If we don't, it may be
990 * notifying us of a removal we initiated, so ignore it.
991 * If there's a pending ENOBUFS, let it stand, since
992 * everything will be reevaluated. */
993 if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
994 sset_add(&ofproto->port_poll_set, devname);
995 ofproto->port_poll_errno = 0;
997 } else if (!ofproto) {
998 /* The port was added, but we don't know with which
999 * ofproto we should associate it. Delete it. */
1000 dpif_port_del(backer->dpif, port.port_no);
1002 dpif_port_destroy(&port);
1008 if (error != EAGAIN) {
1009 struct ofproto_dpif *ofproto;
1011 /* There was some sort of error, so propagate it to all
1012 * ofprotos that use this backer. */
1013 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
1014 &all_ofproto_dpifs) {
1015 if (ofproto->backer == backer) {
1016 sset_clear(&ofproto->port_poll_set);
1017 ofproto->port_poll_errno = error;
1026 type_run_fast(const char *type)
1028 struct dpif_backer *backer;
1031 backer = shash_find_data(&all_dpif_backers, type);
1033 /* This is not necessarily a problem, since backers are only
1034 * created on demand. */
1038 /* Handle one or more batches of upcalls, until there's nothing left to do
1039 * or until we do a fixed total amount of work.
1041 * We do work in batches because it can be much cheaper to set up a number
1042 * of flows and fire off their patches all at once. We do multiple batches
1043 * because in some cases handling a packet can cause another packet to be
1044 * queued almost immediately as part of the return flow. Both
1045 * optimizations can make major improvements on some benchmarks and
1046 * presumably for real traffic as well. */
1048 while (work < FLOW_MISS_MAX_BATCH) {
1049 int retval = handle_upcalls(backer, FLOW_MISS_MAX_BATCH - work);
1060 type_wait(const char *type)
1062 struct dpif_backer *backer;
1064 backer = shash_find_data(&all_dpif_backers, type);
1066 /* This is not necessarily a problem, since backers are only
1067 * created on demand. */
1071 timer_wait(&backer->next_expiration);
1074 /* Basic life-cycle. */
1076 static int add_internal_flows(struct ofproto_dpif *);
1078 static struct ofproto *
1081 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
1082 return &ofproto->up;
1086 dealloc(struct ofproto *ofproto_)
1088 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1093 close_dpif_backer(struct dpif_backer *backer)
1095 struct shash_node *node;
1097 ovs_assert(backer->refcount > 0);
1099 if (--backer->refcount) {
1103 drop_key_clear(backer);
1104 hmap_destroy(&backer->drop_keys);
1106 simap_destroy(&backer->tnl_backers);
1107 hmap_destroy(&backer->odp_to_ofport_map);
1108 node = shash_find(&all_dpif_backers, backer->type);
1110 shash_delete(&all_dpif_backers, node);
1111 dpif_close(backer->dpif);
1116 /* Datapath port slated for removal from datapath. */
1117 struct odp_garbage {
1118 struct list list_node;
1123 open_dpif_backer(const char *type, struct dpif_backer **backerp)
1125 struct dpif_backer *backer;
1126 struct dpif_port_dump port_dump;
1127 struct dpif_port port;
1128 struct shash_node *node;
1129 struct list garbage_list;
1130 struct odp_garbage *garbage, *next;
1136 backer = shash_find_data(&all_dpif_backers, type);
1143 backer_name = xasprintf("ovs-%s", type);
1145 /* Remove any existing datapaths, since we assume we're the only
1146 * userspace controlling the datapath. */
1148 dp_enumerate_names(type, &names);
1149 SSET_FOR_EACH(name, &names) {
1150 struct dpif *old_dpif;
1152 /* Don't remove our backer if it exists. */
1153 if (!strcmp(name, backer_name)) {
1157 if (dpif_open(name, type, &old_dpif)) {
1158 VLOG_WARN("couldn't open old datapath %s to remove it", name);
1160 dpif_delete(old_dpif);
1161 dpif_close(old_dpif);
1164 sset_destroy(&names);
1166 backer = xmalloc(sizeof *backer);
1168 error = dpif_create_and_open(backer_name, type, &backer->dpif);
1171 VLOG_ERR("failed to open datapath of type %s: %s", type,
1177 backer->type = xstrdup(type);
1178 backer->refcount = 1;
1179 hmap_init(&backer->odp_to_ofport_map);
1180 hmap_init(&backer->drop_keys);
1181 timer_set_duration(&backer->next_expiration, 1000);
1182 backer->need_revalidate = 0;
1183 simap_init(&backer->tnl_backers);
1184 tag_set_init(&backer->revalidate_set);
1187 dpif_flow_flush(backer->dpif);
1189 /* Loop through the ports already on the datapath and remove any
1190 * that we don't need anymore. */
1191 list_init(&garbage_list);
1192 dpif_port_dump_start(&port_dump, backer->dpif);
1193 while (dpif_port_dump_next(&port_dump, &port)) {
1194 node = shash_find(&init_ofp_ports, port.name);
1195 if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
1196 garbage = xmalloc(sizeof *garbage);
1197 garbage->odp_port = port.port_no;
1198 list_push_front(&garbage_list, &garbage->list_node);
1201 dpif_port_dump_done(&port_dump);
1203 LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) {
1204 dpif_port_del(backer->dpif, garbage->odp_port);
1205 list_remove(&garbage->list_node);
1209 shash_add(&all_dpif_backers, type, backer);
1211 error = dpif_recv_set(backer->dpif, true);
1213 VLOG_ERR("failed to listen on datapath of type %s: %s",
1214 type, strerror(error));
1215 close_dpif_backer(backer);
1223 construct(struct ofproto *ofproto_)
1225 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1226 struct shash_node *node, *next;
1231 error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1236 max_ports = dpif_get_max_ports(ofproto->backer->dpif);
1237 ofproto_init_max_ports(ofproto_, MIN(max_ports, OFPP_MAX));
1239 ofproto->n_matches = 0;
1241 ofproto->netflow = NULL;
1242 ofproto->sflow = NULL;
1243 ofproto->stp = NULL;
1244 hmap_init(&ofproto->bundles);
1245 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1246 for (i = 0; i < MAX_MIRRORS; i++) {
1247 ofproto->mirrors[i] = NULL;
1249 ofproto->has_bonded_bundles = false;
1251 hmap_init(&ofproto->facets);
1252 hmap_init(&ofproto->subfacets);
1253 ofproto->governor = NULL;
1254 ofproto->consistency_rl = LLONG_MIN;
1256 for (i = 0; i < N_TABLES; i++) {
1257 struct table_dpif *table = &ofproto->tables[i];
1259 table->catchall_table = NULL;
1260 table->other_table = NULL;
1261 table->basis = random_uint32();
1264 list_init(&ofproto->completions);
1266 ofproto_dpif_unixctl_init();
1268 ofproto->has_mirrors = false;
1269 ofproto->has_bundle_action = false;
1271 hmap_init(&ofproto->vlandev_map);
1272 hmap_init(&ofproto->realdev_vid_map);
1274 sset_init(&ofproto->ports);
1275 sset_init(&ofproto->ghost_ports);
1276 sset_init(&ofproto->port_poll_set);
1277 ofproto->port_poll_errno = 0;
1279 SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1280 struct iface_hint *iface_hint = node->data;
1282 if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1283 /* Check if the datapath already has this port. */
1284 if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1285 sset_add(&ofproto->ports, node->name);
1288 free(iface_hint->br_name);
1289 free(iface_hint->br_type);
1291 shash_delete(&init_ofp_ports, node);
1295 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1296 hash_string(ofproto->up.name, 0));
1297 memset(&ofproto->stats, 0, sizeof ofproto->stats);
1299 ofproto_init_tables(ofproto_, N_TABLES);
1300 error = add_internal_flows(ofproto);
1301 ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1304 ofproto->n_missed = 0;
1310 add_internal_flow(struct ofproto_dpif *ofproto, int id,
1311 const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1313 struct ofputil_flow_mod fm;
1316 match_init_catchall(&fm.match);
1318 match_set_reg(&fm.match, 0, id);
1319 fm.new_cookie = htonll(0);
1320 fm.cookie = htonll(0);
1321 fm.cookie_mask = htonll(0);
1322 fm.table_id = TBL_INTERNAL;
1323 fm.command = OFPFC_ADD;
1324 fm.idle_timeout = 0;
1325 fm.hard_timeout = 0;
1329 fm.ofpacts = ofpacts->data;
1330 fm.ofpacts_len = ofpacts->size;
1332 error = ofproto_flow_mod(&ofproto->up, &fm);
1334 VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
1335 id, ofperr_to_string(error));
1339 *rulep = rule_dpif_lookup__(ofproto, &fm.match.flow, TBL_INTERNAL);
1340 ovs_assert(*rulep != NULL);
1346 add_internal_flows(struct ofproto_dpif *ofproto)
1348 struct ofpact_controller *controller;
1349 uint64_t ofpacts_stub[128 / 8];
1350 struct ofpbuf ofpacts;
1354 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1357 controller = ofpact_put_CONTROLLER(&ofpacts);
1358 controller->max_len = UINT16_MAX;
1359 controller->controller_id = 0;
1360 controller->reason = OFPR_NO_MATCH;
1361 ofpact_pad(&ofpacts);
1363 error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule);
1368 ofpbuf_clear(&ofpacts);
1369 error = add_internal_flow(ofproto, id++, &ofpacts,
1370 &ofproto->no_packet_in_rule);
1375 complete_operations(struct ofproto_dpif *ofproto)
1377 struct dpif_completion *c, *next;
1379 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
1380 ofoperation_complete(c->op, 0);
1381 list_remove(&c->list_node);
1387 destruct(struct ofproto *ofproto_)
1389 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1390 struct rule_dpif *rule, *next_rule;
1391 struct oftable *table;
1394 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1395 complete_operations(ofproto);
1397 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1398 struct cls_cursor cursor;
1400 cls_cursor_init(&cursor, &table->cls, NULL);
1401 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1402 ofproto_rule_destroy(&rule->up);
1406 for (i = 0; i < MAX_MIRRORS; i++) {
1407 mirror_destroy(ofproto->mirrors[i]);
1410 netflow_destroy(ofproto->netflow);
1411 dpif_sflow_destroy(ofproto->sflow);
1412 hmap_destroy(&ofproto->bundles);
1413 mac_learning_destroy(ofproto->ml);
1415 hmap_destroy(&ofproto->facets);
1416 hmap_destroy(&ofproto->subfacets);
1417 governor_destroy(ofproto->governor);
1419 hmap_destroy(&ofproto->vlandev_map);
1420 hmap_destroy(&ofproto->realdev_vid_map);
1422 sset_destroy(&ofproto->ports);
1423 sset_destroy(&ofproto->ghost_ports);
1424 sset_destroy(&ofproto->port_poll_set);
1426 close_dpif_backer(ofproto->backer);
1430 run_fast(struct ofproto *ofproto_)
1432 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1433 struct ofport_dpif *ofport;
1435 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1436 port_run_fast(ofport);
1443 run(struct ofproto *ofproto_)
1445 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1446 struct ofport_dpif *ofport;
1447 struct ofbundle *bundle;
1451 complete_operations(ofproto);
1454 error = run_fast(ofproto_);
1459 if (ofproto->netflow) {
1460 if (netflow_run(ofproto->netflow)) {
1461 send_netflow_active_timeouts(ofproto);
1464 if (ofproto->sflow) {
1465 dpif_sflow_run(ofproto->sflow);
1468 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1471 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1476 mac_learning_run(ofproto->ml, &ofproto->backer->revalidate_set);
1478 /* Check the consistency of a random facet, to aid debugging. */
1479 if (time_msec() >= ofproto->consistency_rl
1480 && !hmap_is_empty(&ofproto->facets)
1481 && !ofproto->backer->need_revalidate) {
1482 struct facet *facet;
1484 ofproto->consistency_rl = time_msec() + 250;
1486 facet = CONTAINER_OF(hmap_random_node(&ofproto->facets),
1487 struct facet, hmap_node);
1488 if (!tag_set_intersects(&ofproto->backer->revalidate_set,
1490 if (!facet_check_consistency(facet)) {
1491 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
1496 if (ofproto->governor) {
1499 governor_run(ofproto->governor);
1501 /* If the governor has shrunk to its minimum size and the number of
1502 * subfacets has dwindled, then drop the governor entirely.
1504 * For hysteresis, the number of subfacets to drop the governor is
1505 * smaller than the number needed to trigger its creation. */
1506 n_subfacets = hmap_count(&ofproto->subfacets);
1507 if (n_subfacets * 4 < ofproto->up.flow_eviction_threshold
1508 && governor_is_idle(ofproto->governor)) {
1509 governor_destroy(ofproto->governor);
1510 ofproto->governor = NULL;
1518 wait(struct ofproto *ofproto_)
1520 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1521 struct ofport_dpif *ofport;
1522 struct ofbundle *bundle;
1524 if (!clogged && !list_is_empty(&ofproto->completions)) {
1525 poll_immediate_wake();
1528 dpif_wait(ofproto->backer->dpif);
1529 dpif_recv_wait(ofproto->backer->dpif);
1530 if (ofproto->sflow) {
1531 dpif_sflow_wait(ofproto->sflow);
1533 if (!tag_set_is_empty(&ofproto->backer->revalidate_set)) {
1534 poll_immediate_wake();
1536 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1539 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1540 bundle_wait(bundle);
1542 if (ofproto->netflow) {
1543 netflow_wait(ofproto->netflow);
1545 mac_learning_wait(ofproto->ml);
1547 if (ofproto->backer->need_revalidate) {
1548 /* Shouldn't happen, but if it does just go around again. */
1549 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1550 poll_immediate_wake();
1552 if (ofproto->governor) {
1553 governor_wait(ofproto->governor);
1558 get_memory_usage(const struct ofproto *ofproto_, struct simap *usage)
1560 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1562 simap_increase(usage, "facets", hmap_count(&ofproto->facets));
1563 simap_increase(usage, "subfacets", hmap_count(&ofproto->subfacets));
1567 flush(struct ofproto *ofproto_)
1569 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1570 struct subfacet *subfacet, *next_subfacet;
1571 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
1575 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
1576 &ofproto->subfacets) {
1577 if (subfacet->path != SF_NOT_INSTALLED) {
1578 batch[n_batch++] = subfacet;
1579 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
1580 subfacet_destroy_batch(ofproto, batch, n_batch);
1584 subfacet_destroy(subfacet);
1589 subfacet_destroy_batch(ofproto, batch, n_batch);
1594 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1595 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1597 *arp_match_ip = true;
1598 *actions = (OFPUTIL_A_OUTPUT |
1599 OFPUTIL_A_SET_VLAN_VID |
1600 OFPUTIL_A_SET_VLAN_PCP |
1601 OFPUTIL_A_STRIP_VLAN |
1602 OFPUTIL_A_SET_DL_SRC |
1603 OFPUTIL_A_SET_DL_DST |
1604 OFPUTIL_A_SET_NW_SRC |
1605 OFPUTIL_A_SET_NW_DST |
1606 OFPUTIL_A_SET_NW_TOS |
1607 OFPUTIL_A_SET_TP_SRC |
1608 OFPUTIL_A_SET_TP_DST |
1613 get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots)
1615 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1616 struct dpif_dp_stats s;
1618 strcpy(ots->name, "classifier");
1620 dpif_get_dp_stats(ofproto->backer->dpif, &s);
1622 ots->lookup_count = htonll(s.n_hit + s.n_missed);
1623 ots->matched_count = htonll(s.n_hit + ofproto->n_matches);
1626 static struct ofport *
1629 struct ofport_dpif *port = xmalloc(sizeof *port);
1634 port_dealloc(struct ofport *port_)
1636 struct ofport_dpif *port = ofport_dpif_cast(port_);
1641 port_construct(struct ofport *port_)
1643 struct ofport_dpif *port = ofport_dpif_cast(port_);
1644 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1645 const struct netdev *netdev = port->up.netdev;
1646 struct dpif_port dpif_port;
1649 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1650 port->bundle = NULL;
1652 port->tag = tag_create_random();
1653 port->may_enable = true;
1654 port->stp_port = NULL;
1655 port->stp_state = STP_DISABLED;
1656 port->tnl_port = NULL;
1657 hmap_init(&port->priorities);
1658 port->realdev_ofp_port = 0;
1659 port->vlandev_vid = 0;
1660 port->carrier_seq = netdev_get_carrier_resets(netdev);
1662 if (netdev_vport_is_patch(netdev)) {
1663 /* XXX By bailing out here, we don't do required sFlow work. */
1664 port->odp_port = OVSP_NONE;
1668 error = dpif_port_query_by_name(ofproto->backer->dpif,
1669 netdev_vport_get_dpif_port(netdev),
1675 port->odp_port = dpif_port.port_no;
1677 if (netdev_get_tunnel_config(netdev)) {
1678 port->tnl_port = tnl_port_add(&port->up, port->odp_port);
1680 /* Sanity-check that a mapping doesn't already exist. This
1681 * shouldn't happen for non-tunnel ports. */
1682 if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1683 VLOG_ERR("port %s already has an OpenFlow port number",
1685 dpif_port_destroy(&dpif_port);
1689 hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1690 hash_int(port->odp_port, 0));
1692 dpif_port_destroy(&dpif_port);
1694 if (ofproto->sflow) {
1695 dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1702 port_destruct(struct ofport *port_)
1704 struct ofport_dpif *port = ofport_dpif_cast(port_);
1705 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1706 const char *dp_port_name = netdev_vport_get_dpif_port(port->up.netdev);
1707 const char *devname = netdev_get_name(port->up.netdev);
1709 if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1710 /* The underlying device is still there, so delete it. This
1711 * happens when the ofproto is being destroyed, since the caller
1712 * assumes that removal of attached ports will happen as part of
1714 if (!port->tnl_port) {
1715 dpif_port_del(ofproto->backer->dpif, port->odp_port);
1717 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1720 if (port->odp_port != OVSP_NONE && !port->tnl_port) {
1721 hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1724 tnl_port_del(port->tnl_port);
1725 sset_find_and_delete(&ofproto->ports, devname);
1726 sset_find_and_delete(&ofproto->ghost_ports, devname);
1727 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1728 bundle_remove(port_);
1729 set_cfm(port_, NULL);
1730 if (ofproto->sflow) {
1731 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1734 ofport_clear_priorities(port);
1735 hmap_destroy(&port->priorities);
1739 port_modified(struct ofport *port_)
1741 struct ofport_dpif *port = ofport_dpif_cast(port_);
1743 if (port->bundle && port->bundle->bond) {
1744 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1749 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1751 struct ofport_dpif *port = ofport_dpif_cast(port_);
1752 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1753 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1755 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1756 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1757 OFPUTIL_PC_NO_PACKET_IN)) {
1758 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1760 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1761 bundle_update(port->bundle);
1767 set_sflow(struct ofproto *ofproto_,
1768 const struct ofproto_sflow_options *sflow_options)
1770 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1771 struct dpif_sflow *ds = ofproto->sflow;
1773 if (sflow_options) {
1775 struct ofport_dpif *ofport;
1777 ds = ofproto->sflow = dpif_sflow_create();
1778 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1779 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1781 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1783 dpif_sflow_set_options(ds, sflow_options);
1786 dpif_sflow_destroy(ds);
1787 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1788 ofproto->sflow = NULL;
1795 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1797 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1804 struct ofproto_dpif *ofproto;
1806 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1807 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1808 ofport->cfm = cfm_create(netdev_get_name(ofport->up.netdev));
1811 if (cfm_configure(ofport->cfm, s)) {
1817 cfm_destroy(ofport->cfm);
1823 get_cfm_status(const struct ofport *ofport_,
1824 struct ofproto_cfm_status *status)
1826 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1829 status->faults = cfm_get_fault(ofport->cfm);
1830 status->remote_opstate = cfm_get_opup(ofport->cfm);
1831 status->health = cfm_get_health(ofport->cfm);
1832 cfm_get_remote_mpids(ofport->cfm, &status->rmps, &status->n_rmps);
1839 /* Spanning Tree. */
1842 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
1844 struct ofproto_dpif *ofproto = ofproto_;
1845 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
1846 struct ofport_dpif *ofport;
1848 ofport = stp_port_get_aux(sp);
1850 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
1851 ofproto->up.name, port_num);
1853 struct eth_header *eth = pkt->l2;
1855 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
1856 if (eth_addr_is_zero(eth->eth_src)) {
1857 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
1858 "with unknown MAC", ofproto->up.name, port_num);
1860 send_packet(ofport, pkt);
1866 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
1868 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
1870 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1872 /* Only revalidate flows if the configuration changed. */
1873 if (!s != !ofproto->stp) {
1874 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1878 if (!ofproto->stp) {
1879 ofproto->stp = stp_create(ofproto_->name, s->system_id,
1880 send_bpdu_cb, ofproto);
1881 ofproto->stp_last_tick = time_msec();
1884 stp_set_bridge_id(ofproto->stp, s->system_id);
1885 stp_set_bridge_priority(ofproto->stp, s->priority);
1886 stp_set_hello_time(ofproto->stp, s->hello_time);
1887 stp_set_max_age(ofproto->stp, s->max_age);
1888 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
1890 struct ofport *ofport;
1892 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
1893 set_stp_port(ofport, NULL);
1896 stp_destroy(ofproto->stp);
1897 ofproto->stp = NULL;
1904 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
1906 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1910 s->bridge_id = stp_get_bridge_id(ofproto->stp);
1911 s->designated_root = stp_get_designated_root(ofproto->stp);
1912 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
1921 update_stp_port_state(struct ofport_dpif *ofport)
1923 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1924 enum stp_state state;
1926 /* Figure out new state. */
1927 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
1931 if (ofport->stp_state != state) {
1932 enum ofputil_port_state of_state;
1935 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
1936 netdev_get_name(ofport->up.netdev),
1937 stp_state_name(ofport->stp_state),
1938 stp_state_name(state));
1939 if (stp_learn_in_state(ofport->stp_state)
1940 != stp_learn_in_state(state)) {
1941 /* xxx Learning action flows should also be flushed. */
1942 mac_learning_flush(ofproto->ml,
1943 &ofproto->backer->revalidate_set);
1945 fwd_change = stp_forward_in_state(ofport->stp_state)
1946 != stp_forward_in_state(state);
1948 ofproto->backer->need_revalidate = REV_STP;
1949 ofport->stp_state = state;
1950 ofport->stp_state_entered = time_msec();
1952 if (fwd_change && ofport->bundle) {
1953 bundle_update(ofport->bundle);
1956 /* Update the STP state bits in the OpenFlow port description. */
1957 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
1958 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
1959 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
1960 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
1961 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
1963 ofproto_port_set_state(&ofport->up, of_state);
1967 /* Configures STP on 'ofport_' using the settings defined in 's'. The
1968 * caller is responsible for assigning STP port numbers and ensuring
1969 * there are no duplicates. */
1971 set_stp_port(struct ofport *ofport_,
1972 const struct ofproto_port_stp_settings *s)
1974 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1975 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1976 struct stp_port *sp = ofport->stp_port;
1978 if (!s || !s->enable) {
1980 ofport->stp_port = NULL;
1981 stp_port_disable(sp);
1982 update_stp_port_state(ofport);
1985 } else if (sp && stp_port_no(sp) != s->port_num
1986 && ofport == stp_port_get_aux(sp)) {
1987 /* The port-id changed, so disable the old one if it's not
1988 * already in use by another port. */
1989 stp_port_disable(sp);
1992 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
1993 stp_port_enable(sp);
1995 stp_port_set_aux(sp, ofport);
1996 stp_port_set_priority(sp, s->priority);
1997 stp_port_set_path_cost(sp, s->path_cost);
1999 update_stp_port_state(ofport);
2005 get_stp_port_status(struct ofport *ofport_,
2006 struct ofproto_port_stp_status *s)
2008 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2009 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2010 struct stp_port *sp = ofport->stp_port;
2012 if (!ofproto->stp || !sp) {
2018 s->port_id = stp_port_get_id(sp);
2019 s->state = stp_port_get_state(sp);
2020 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
2021 s->role = stp_port_get_role(sp);
2022 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
2028 stp_run(struct ofproto_dpif *ofproto)
2031 long long int now = time_msec();
2032 long long int elapsed = now - ofproto->stp_last_tick;
2033 struct stp_port *sp;
2036 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
2037 ofproto->stp_last_tick = now;
2039 while (stp_get_changed_port(ofproto->stp, &sp)) {
2040 struct ofport_dpif *ofport = stp_port_get_aux(sp);
2043 update_stp_port_state(ofport);
2047 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
2048 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
2054 stp_wait(struct ofproto_dpif *ofproto)
2057 poll_timer_wait(1000);
2061 /* Returns true if STP should process 'flow'. */
2063 stp_should_process_flow(const struct flow *flow)
2065 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
2069 stp_process_packet(const struct ofport_dpif *ofport,
2070 const struct ofpbuf *packet)
2072 struct ofpbuf payload = *packet;
2073 struct eth_header *eth = payload.data;
2074 struct stp_port *sp = ofport->stp_port;
2076 /* Sink packets on ports that have STP disabled when the bridge has
2078 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
2082 /* Trim off padding on payload. */
2083 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
2084 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
2087 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
2088 stp_received_bpdu(sp, payload.data, payload.size);
2092 static struct priority_to_dscp *
2093 get_priority(const struct ofport_dpif *ofport, uint32_t priority)
2095 struct priority_to_dscp *pdscp;
2098 hash = hash_int(priority, 0);
2099 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) {
2100 if (pdscp->priority == priority) {
2108 ofport_clear_priorities(struct ofport_dpif *ofport)
2110 struct priority_to_dscp *pdscp, *next;
2112 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) {
2113 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
2119 set_queues(struct ofport *ofport_,
2120 const struct ofproto_port_queue *qdscp_list,
2123 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2124 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2125 struct hmap new = HMAP_INITIALIZER(&new);
2128 for (i = 0; i < n_qdscp; i++) {
2129 struct priority_to_dscp *pdscp;
2133 dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
2134 if (dpif_queue_to_priority(ofproto->backer->dpif, qdscp_list[i].queue,
2139 pdscp = get_priority(ofport, priority);
2141 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
2143 pdscp = xmalloc(sizeof *pdscp);
2144 pdscp->priority = priority;
2146 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2149 if (pdscp->dscp != dscp) {
2151 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2154 hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0));
2157 if (!hmap_is_empty(&ofport->priorities)) {
2158 ofport_clear_priorities(ofport);
2159 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2162 hmap_swap(&new, &ofport->priorities);
2170 /* Expires all MAC learning entries associated with 'bundle' and forces its
2171 * ofproto to revalidate every flow.
2173 * Normally MAC learning entries are removed only from the ofproto associated
2174 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2175 * are removed from every ofproto. When patch ports and SLB bonds are in use
2176 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2177 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2178 * with the host from which it migrated. */
2180 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2182 struct ofproto_dpif *ofproto = bundle->ofproto;
2183 struct mac_learning *ml = ofproto->ml;
2184 struct mac_entry *mac, *next_mac;
2186 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2187 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2188 if (mac->port.p == bundle) {
2190 struct ofproto_dpif *o;
2192 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2194 struct mac_entry *e;
2196 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan,
2199 mac_learning_expire(o->ml, e);
2205 mac_learning_expire(ml, mac);
2210 static struct ofbundle *
2211 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2213 struct ofbundle *bundle;
2215 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2216 &ofproto->bundles) {
2217 if (bundle->aux == aux) {
2224 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
2225 * ones that are found to 'bundles'. */
2227 bundle_lookup_multiple(struct ofproto_dpif *ofproto,
2228 void **auxes, size_t n_auxes,
2229 struct hmapx *bundles)
2233 hmapx_init(bundles);
2234 for (i = 0; i < n_auxes; i++) {
2235 struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
2237 hmapx_add(bundles, bundle);
2243 bundle_update(struct ofbundle *bundle)
2245 struct ofport_dpif *port;
2247 bundle->floodable = true;
2248 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2249 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2250 || !stp_forward_in_state(port->stp_state)) {
2251 bundle->floodable = false;
2258 bundle_del_port(struct ofport_dpif *port)
2260 struct ofbundle *bundle = port->bundle;
2262 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2264 list_remove(&port->bundle_node);
2265 port->bundle = NULL;
2268 lacp_slave_unregister(bundle->lacp, port);
2271 bond_slave_unregister(bundle->bond, port);
2274 bundle_update(bundle);
2278 bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port,
2279 struct lacp_slave_settings *lacp)
2281 struct ofport_dpif *port;
2283 port = get_ofp_port(bundle->ofproto, ofp_port);
2288 if (port->bundle != bundle) {
2289 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2291 bundle_del_port(port);
2294 port->bundle = bundle;
2295 list_push_back(&bundle->ports, &port->bundle_node);
2296 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2297 || !stp_forward_in_state(port->stp_state)) {
2298 bundle->floodable = false;
2302 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2303 lacp_slave_register(bundle->lacp, port, lacp);
2310 bundle_destroy(struct ofbundle *bundle)
2312 struct ofproto_dpif *ofproto;
2313 struct ofport_dpif *port, *next_port;
2320 ofproto = bundle->ofproto;
2321 for (i = 0; i < MAX_MIRRORS; i++) {
2322 struct ofmirror *m = ofproto->mirrors[i];
2324 if (m->out == bundle) {
2326 } else if (hmapx_find_and_delete(&m->srcs, bundle)
2327 || hmapx_find_and_delete(&m->dsts, bundle)) {
2328 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2333 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2334 bundle_del_port(port);
2337 bundle_flush_macs(bundle, true);
2338 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2340 free(bundle->trunks);
2341 lacp_destroy(bundle->lacp);
2342 bond_destroy(bundle->bond);
2347 bundle_set(struct ofproto *ofproto_, void *aux,
2348 const struct ofproto_bundle_settings *s)
2350 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2351 bool need_flush = false;
2352 struct ofport_dpif *port;
2353 struct ofbundle *bundle;
2354 unsigned long *trunks;
2360 bundle_destroy(bundle_lookup(ofproto, aux));
2364 ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2365 ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2367 bundle = bundle_lookup(ofproto, aux);
2369 bundle = xmalloc(sizeof *bundle);
2371 bundle->ofproto = ofproto;
2372 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2373 hash_pointer(aux, 0));
2375 bundle->name = NULL;
2377 list_init(&bundle->ports);
2378 bundle->vlan_mode = PORT_VLAN_TRUNK;
2380 bundle->trunks = NULL;
2381 bundle->use_priority_tags = s->use_priority_tags;
2382 bundle->lacp = NULL;
2383 bundle->bond = NULL;
2385 bundle->floodable = true;
2387 bundle->src_mirrors = 0;
2388 bundle->dst_mirrors = 0;
2389 bundle->mirror_out = 0;
2392 if (!bundle->name || strcmp(s->name, bundle->name)) {
2394 bundle->name = xstrdup(s->name);
2399 if (!bundle->lacp) {
2400 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2401 bundle->lacp = lacp_create();
2403 lacp_configure(bundle->lacp, s->lacp);
2405 lacp_destroy(bundle->lacp);
2406 bundle->lacp = NULL;
2409 /* Update set of ports. */
2411 for (i = 0; i < s->n_slaves; i++) {
2412 if (!bundle_add_port(bundle, s->slaves[i],
2413 s->lacp ? &s->lacp_slaves[i] : NULL)) {
2417 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2418 struct ofport_dpif *next_port;
2420 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2421 for (i = 0; i < s->n_slaves; i++) {
2422 if (s->slaves[i] == port->up.ofp_port) {
2427 bundle_del_port(port);
2431 ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2433 if (list_is_empty(&bundle->ports)) {
2434 bundle_destroy(bundle);
2438 /* Set VLAN tagging mode */
2439 if (s->vlan_mode != bundle->vlan_mode
2440 || s->use_priority_tags != bundle->use_priority_tags) {
2441 bundle->vlan_mode = s->vlan_mode;
2442 bundle->use_priority_tags = s->use_priority_tags;
2447 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2448 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2450 if (vlan != bundle->vlan) {
2451 bundle->vlan = vlan;
2455 /* Get trunked VLANs. */
2456 switch (s->vlan_mode) {
2457 case PORT_VLAN_ACCESS:
2461 case PORT_VLAN_TRUNK:
2462 trunks = CONST_CAST(unsigned long *, s->trunks);
2465 case PORT_VLAN_NATIVE_UNTAGGED:
2466 case PORT_VLAN_NATIVE_TAGGED:
2467 if (vlan != 0 && (!s->trunks
2468 || !bitmap_is_set(s->trunks, vlan)
2469 || bitmap_is_set(s->trunks, 0))) {
2470 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2472 trunks = bitmap_clone(s->trunks, 4096);
2474 trunks = bitmap_allocate1(4096);
2476 bitmap_set1(trunks, vlan);
2477 bitmap_set0(trunks, 0);
2479 trunks = CONST_CAST(unsigned long *, s->trunks);
2486 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2487 free(bundle->trunks);
2488 if (trunks == s->trunks) {
2489 bundle->trunks = vlan_bitmap_clone(trunks);
2491 bundle->trunks = trunks;
2496 if (trunks != s->trunks) {
2501 if (!list_is_short(&bundle->ports)) {
2502 bundle->ofproto->has_bonded_bundles = true;
2504 if (bond_reconfigure(bundle->bond, s->bond)) {
2505 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2508 bundle->bond = bond_create(s->bond);
2509 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2512 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2513 bond_slave_register(bundle->bond, port, port->up.netdev);
2516 bond_destroy(bundle->bond);
2517 bundle->bond = NULL;
2520 /* If we changed something that would affect MAC learning, un-learn
2521 * everything on this port and force flow revalidation. */
2523 bundle_flush_macs(bundle, false);
2530 bundle_remove(struct ofport *port_)
2532 struct ofport_dpif *port = ofport_dpif_cast(port_);
2533 struct ofbundle *bundle = port->bundle;
2536 bundle_del_port(port);
2537 if (list_is_empty(&bundle->ports)) {
2538 bundle_destroy(bundle);
2539 } else if (list_is_short(&bundle->ports)) {
2540 bond_destroy(bundle->bond);
2541 bundle->bond = NULL;
2547 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2549 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2550 struct ofport_dpif *port = port_;
2551 uint8_t ea[ETH_ADDR_LEN];
2554 error = netdev_get_etheraddr(port->up.netdev, ea);
2556 struct ofpbuf packet;
2559 ofpbuf_init(&packet, 0);
2560 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2562 memcpy(packet_pdu, pdu, pdu_size);
2564 send_packet(port, &packet);
2565 ofpbuf_uninit(&packet);
2567 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2568 "%s (%s)", port->bundle->name,
2569 netdev_get_name(port->up.netdev), strerror(error));
2574 bundle_send_learning_packets(struct ofbundle *bundle)
2576 struct ofproto_dpif *ofproto = bundle->ofproto;
2577 int error, n_packets, n_errors;
2578 struct mac_entry *e;
2580 error = n_packets = n_errors = 0;
2581 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2582 if (e->port.p != bundle) {
2583 struct ofpbuf *learning_packet;
2584 struct ofport_dpif *port;
2588 /* The assignment to "port" is unnecessary but makes "grep"ing for
2589 * struct ofport_dpif more effective. */
2590 learning_packet = bond_compose_learning_packet(bundle->bond,
2594 ret = send_packet(port, learning_packet);
2595 ofpbuf_delete(learning_packet);
2605 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2606 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2607 "packets, last error was: %s",
2608 bundle->name, n_errors, n_packets, strerror(error));
2610 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2611 bundle->name, n_packets);
2616 bundle_run(struct ofbundle *bundle)
2619 lacp_run(bundle->lacp, send_pdu_cb);
2622 struct ofport_dpif *port;
2624 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2625 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2628 bond_run(bundle->bond, &bundle->ofproto->backer->revalidate_set,
2629 lacp_status(bundle->lacp));
2630 if (bond_should_send_learning_packets(bundle->bond)) {
2631 bundle_send_learning_packets(bundle);
2637 bundle_wait(struct ofbundle *bundle)
2640 lacp_wait(bundle->lacp);
2643 bond_wait(bundle->bond);
2650 mirror_scan(struct ofproto_dpif *ofproto)
2654 for (idx = 0; idx < MAX_MIRRORS; idx++) {
2655 if (!ofproto->mirrors[idx]) {
2662 static struct ofmirror *
2663 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
2667 for (i = 0; i < MAX_MIRRORS; i++) {
2668 struct ofmirror *mirror = ofproto->mirrors[i];
2669 if (mirror && mirror->aux == aux) {
2677 /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */
2679 mirror_update_dups(struct ofproto_dpif *ofproto)
2683 for (i = 0; i < MAX_MIRRORS; i++) {
2684 struct ofmirror *m = ofproto->mirrors[i];
2687 m->dup_mirrors = MIRROR_MASK_C(1) << i;
2691 for (i = 0; i < MAX_MIRRORS; i++) {
2692 struct ofmirror *m1 = ofproto->mirrors[i];
2699 for (j = i + 1; j < MAX_MIRRORS; j++) {
2700 struct ofmirror *m2 = ofproto->mirrors[j];
2702 if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) {
2703 m1->dup_mirrors |= MIRROR_MASK_C(1) << j;
2704 m2->dup_mirrors |= m1->dup_mirrors;
2711 mirror_set(struct ofproto *ofproto_, void *aux,
2712 const struct ofproto_mirror_settings *s)
2714 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2715 mirror_mask_t mirror_bit;
2716 struct ofbundle *bundle;
2717 struct ofmirror *mirror;
2718 struct ofbundle *out;
2719 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
2720 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
2723 mirror = mirror_lookup(ofproto, aux);
2725 mirror_destroy(mirror);
2731 idx = mirror_scan(ofproto);
2733 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
2735 ofproto->up.name, MAX_MIRRORS, s->name);
2739 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
2740 mirror->ofproto = ofproto;
2743 mirror->out_vlan = -1;
2744 mirror->name = NULL;
2747 if (!mirror->name || strcmp(s->name, mirror->name)) {
2749 mirror->name = xstrdup(s->name);
2752 /* Get the new configuration. */
2753 if (s->out_bundle) {
2754 out = bundle_lookup(ofproto, s->out_bundle);
2756 mirror_destroy(mirror);
2762 out_vlan = s->out_vlan;
2764 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
2765 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
2767 /* If the configuration has not changed, do nothing. */
2768 if (hmapx_equals(&srcs, &mirror->srcs)
2769 && hmapx_equals(&dsts, &mirror->dsts)
2770 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
2771 && mirror->out == out
2772 && mirror->out_vlan == out_vlan)
2774 hmapx_destroy(&srcs);
2775 hmapx_destroy(&dsts);
2779 hmapx_swap(&srcs, &mirror->srcs);
2780 hmapx_destroy(&srcs);
2782 hmapx_swap(&dsts, &mirror->dsts);
2783 hmapx_destroy(&dsts);
2785 free(mirror->vlans);
2786 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
2789 mirror->out_vlan = out_vlan;
2791 /* Update bundles. */
2792 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2793 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
2794 if (hmapx_contains(&mirror->srcs, bundle)) {
2795 bundle->src_mirrors |= mirror_bit;
2797 bundle->src_mirrors &= ~mirror_bit;
2800 if (hmapx_contains(&mirror->dsts, bundle)) {
2801 bundle->dst_mirrors |= mirror_bit;
2803 bundle->dst_mirrors &= ~mirror_bit;
2806 if (mirror->out == bundle) {
2807 bundle->mirror_out |= mirror_bit;
2809 bundle->mirror_out &= ~mirror_bit;
2813 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2814 ofproto->has_mirrors = true;
2815 mac_learning_flush(ofproto->ml,
2816 &ofproto->backer->revalidate_set);
2817 mirror_update_dups(ofproto);
2823 mirror_destroy(struct ofmirror *mirror)
2825 struct ofproto_dpif *ofproto;
2826 mirror_mask_t mirror_bit;
2827 struct ofbundle *bundle;
2834 ofproto = mirror->ofproto;
2835 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2836 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
2838 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2839 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
2840 bundle->src_mirrors &= ~mirror_bit;
2841 bundle->dst_mirrors &= ~mirror_bit;
2842 bundle->mirror_out &= ~mirror_bit;
2845 hmapx_destroy(&mirror->srcs);
2846 hmapx_destroy(&mirror->dsts);
2847 free(mirror->vlans);
2849 ofproto->mirrors[mirror->idx] = NULL;
2853 mirror_update_dups(ofproto);
2855 ofproto->has_mirrors = false;
2856 for (i = 0; i < MAX_MIRRORS; i++) {
2857 if (ofproto->mirrors[i]) {
2858 ofproto->has_mirrors = true;
2865 mirror_get_stats(struct ofproto *ofproto_, void *aux,
2866 uint64_t *packets, uint64_t *bytes)
2868 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2869 struct ofmirror *mirror = mirror_lookup(ofproto, aux);
2872 *packets = *bytes = UINT64_MAX;
2876 *packets = mirror->packet_count;
2877 *bytes = mirror->byte_count;
2883 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2885 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2886 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2887 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
2893 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2895 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2896 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2897 return bundle && bundle->mirror_out != 0;
2901 forward_bpdu_changed(struct ofproto *ofproto_)
2903 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2904 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2908 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
2911 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2912 mac_learning_set_idle_time(ofproto->ml, idle_time);
2913 mac_learning_set_max_entries(ofproto->ml, max_entries);
2918 static struct ofport_dpif *
2919 get_ofp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
2921 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2922 return ofport ? ofport_dpif_cast(ofport) : NULL;
2925 static struct ofport_dpif *
2926 get_odp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
2928 struct ofport_dpif *port = odp_port_to_ofport(ofproto->backer, odp_port);
2929 return port && &ofproto->up == port->up.ofproto ? port : NULL;
2933 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
2934 struct ofproto_port *ofproto_port,
2935 struct dpif_port *dpif_port)
2937 ofproto_port->name = dpif_port->name;
2938 ofproto_port->type = dpif_port->type;
2939 ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
2942 static struct ofport_dpif *
2943 ofport_get_peer(const struct ofport_dpif *ofport_dpif)
2945 const struct ofproto_dpif *ofproto;
2948 peer = netdev_vport_patch_peer(ofport_dpif->up.netdev);
2953 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2954 struct ofport *ofport;
2956 ofport = shash_find_data(&ofproto->up.port_by_name, peer);
2957 if (ofport && ofport->ofproto->ofproto_class == &ofproto_dpif_class) {
2958 return ofport_dpif_cast(ofport);
2965 port_run_fast(struct ofport_dpif *ofport)
2967 if (ofport->cfm && cfm_should_send_ccm(ofport->cfm)) {
2968 struct ofpbuf packet;
2970 ofpbuf_init(&packet, 0);
2971 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
2972 send_packet(ofport, &packet);
2973 ofpbuf_uninit(&packet);
2978 port_run(struct ofport_dpif *ofport)
2980 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
2981 bool carrier_changed = carrier_seq != ofport->carrier_seq;
2982 bool enable = netdev_get_carrier(ofport->up.netdev);
2984 ofport->carrier_seq = carrier_seq;
2986 port_run_fast(ofport);
2988 if (ofport->tnl_port
2989 && tnl_port_reconfigure(&ofport->up, ofport->odp_port,
2990 &ofport->tnl_port)) {
2991 ofproto_dpif_cast(ofport->up.ofproto)->backer->need_revalidate = true;
2995 int cfm_opup = cfm_get_opup(ofport->cfm);
2997 cfm_run(ofport->cfm);
2998 enable = enable && !cfm_get_fault(ofport->cfm);
3000 if (cfm_opup >= 0) {
3001 enable = enable && cfm_opup;
3005 if (ofport->bundle) {
3006 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
3007 if (carrier_changed) {
3008 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
3012 if (ofport->may_enable != enable) {
3013 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3015 if (ofproto->has_bundle_action) {
3016 ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
3020 ofport->may_enable = enable;
3024 port_wait(struct ofport_dpif *ofport)
3027 cfm_wait(ofport->cfm);
3032 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
3033 struct ofproto_port *ofproto_port)
3035 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3036 struct dpif_port dpif_port;
3039 if (sset_contains(&ofproto->ghost_ports, devname)) {
3040 const char *type = netdev_get_type_from_name(devname);
3042 /* We may be called before ofproto->up.port_by_name is populated with
3043 * the appropriate ofport. For this reason, we must get the name and
3044 * type from the netdev layer directly. */
3046 const struct ofport *ofport;
3048 ofport = shash_find_data(&ofproto->up.port_by_name, devname);
3049 ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
3050 ofproto_port->name = xstrdup(devname);
3051 ofproto_port->type = xstrdup(type);
3057 if (!sset_contains(&ofproto->ports, devname)) {
3060 error = dpif_port_query_by_name(ofproto->backer->dpif,
3061 devname, &dpif_port);
3063 ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
3069 port_add(struct ofproto *ofproto_, struct netdev *netdev)
3071 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3072 const char *dp_port_name = netdev_vport_get_dpif_port(netdev);
3073 const char *devname = netdev_get_name(netdev);
3075 if (netdev_vport_is_patch(netdev)) {
3076 sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
3080 if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
3081 uint32_t port_no = UINT32_MAX;
3084 error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
3088 if (netdev_get_tunnel_config(netdev)) {
3089 simap_put(&ofproto->backer->tnl_backers, dp_port_name, port_no);
3093 if (netdev_get_tunnel_config(netdev)) {
3094 sset_add(&ofproto->ghost_ports, devname);
3096 sset_add(&ofproto->ports, devname);
3102 port_del(struct ofproto *ofproto_, uint16_t ofp_port)
3104 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3105 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
3112 sset_find_and_delete(&ofproto->ghost_ports,
3113 netdev_get_name(ofport->up.netdev));
3114 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3115 if (!ofport->tnl_port) {
3116 error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
3118 /* The caller is going to close ofport->up.netdev. If this is a
3119 * bonded port, then the bond is using that netdev, so remove it
3120 * from the bond. The client will need to reconfigure everything
3121 * after deleting ports, so then the slave will get re-added. */
3122 bundle_remove(&ofport->up);
3129 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
3131 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3134 error = netdev_get_stats(ofport->up.netdev, stats);
3136 if (!error && ofport_->ofp_port == OFPP_LOCAL) {
3137 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3139 /* ofproto->stats.tx_packets represents packets that we created
3140 * internally and sent to some port (e.g. packets sent with
3141 * send_packet()). Account for them as if they had come from
3142 * OFPP_LOCAL and got forwarded. */
3144 if (stats->rx_packets != UINT64_MAX) {
3145 stats->rx_packets += ofproto->stats.tx_packets;
3148 if (stats->rx_bytes != UINT64_MAX) {
3149 stats->rx_bytes += ofproto->stats.tx_bytes;
3152 /* ofproto->stats.rx_packets represents packets that were received on
3153 * some port and we processed internally and dropped (e.g. STP).
3154 * Account for them as if they had been forwarded to OFPP_LOCAL. */
3156 if (stats->tx_packets != UINT64_MAX) {
3157 stats->tx_packets += ofproto->stats.rx_packets;
3160 if (stats->tx_bytes != UINT64_MAX) {
3161 stats->tx_bytes += ofproto->stats.rx_bytes;
3168 /* Account packets for LOCAL port. */
3170 ofproto_update_local_port_stats(const struct ofproto *ofproto_,
3171 size_t tx_size, size_t rx_size)
3173 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3176 ofproto->stats.rx_packets++;
3177 ofproto->stats.rx_bytes += rx_size;
3180 ofproto->stats.tx_packets++;
3181 ofproto->stats.tx_bytes += tx_size;
3185 struct port_dump_state {
3190 struct ofproto_port port;
3195 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
3197 *statep = xzalloc(sizeof(struct port_dump_state));
3202 port_dump_next(const struct ofproto *ofproto_, void *state_,
3203 struct ofproto_port *port)
3205 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3206 struct port_dump_state *state = state_;
3207 const struct sset *sset;
3208 struct sset_node *node;
3210 if (state->has_port) {
3211 ofproto_port_destroy(&state->port);
3212 state->has_port = false;
3214 sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
3215 while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
3218 error = port_query_by_name(ofproto_, node->name, &state->port);
3220 *port = state->port;
3221 state->has_port = true;
3223 } else if (error != ENODEV) {
3228 if (!state->ghost) {
3229 state->ghost = true;
3232 return port_dump_next(ofproto_, state_, port);
3239 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3241 struct port_dump_state *state = state_;
3243 if (state->has_port) {
3244 ofproto_port_destroy(&state->port);
3251 port_poll(const struct ofproto *ofproto_, char **devnamep)
3253 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3255 if (ofproto->port_poll_errno) {
3256 int error = ofproto->port_poll_errno;
3257 ofproto->port_poll_errno = 0;
3261 if (sset_is_empty(&ofproto->port_poll_set)) {
3265 *devnamep = sset_pop(&ofproto->port_poll_set);
3270 port_poll_wait(const struct ofproto *ofproto_)
3272 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3273 dpif_port_poll_wait(ofproto->backer->dpif);
3277 port_is_lacp_current(const struct ofport *ofport_)
3279 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3280 return (ofport->bundle && ofport->bundle->lacp
3281 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3285 /* Upcall handling. */
3287 /* Flow miss batching.
3289 * Some dpifs implement operations faster when you hand them off in a batch.
3290 * To allow batching, "struct flow_miss" queues the dpif-related work needed
3291 * for a given flow. Each "struct flow_miss" corresponds to sending one or
3292 * more packets, plus possibly installing the flow in the dpif.
3294 * So far we only batch the operations that affect flow setup time the most.
3295 * It's possible to batch more than that, but the benefit might be minimal. */
3297 struct hmap_node hmap_node;
3298 struct ofproto_dpif *ofproto;
3300 enum odp_key_fitness key_fitness;
3301 const struct nlattr *key;
3303 struct initial_vals initial_vals;
3304 struct list packets;
3305 enum dpif_upcall_type upcall_type;
3306 uint32_t odp_in_port;
3309 struct flow_miss_op {
3310 struct dpif_op dpif_op;
3311 void *garbage; /* Pointer to pass to free(), NULL if none. */
3312 uint64_t stub[1024 / 8]; /* Temporary buffer. */
3315 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
3316 * OpenFlow controller as necessary according to their individual
3317 * configurations. */
3319 send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet,
3320 const struct flow *flow)
3322 struct ofputil_packet_in pin;
3324 pin.packet = packet->data;
3325 pin.packet_len = packet->size;
3326 pin.reason = OFPR_NO_MATCH;
3327 pin.controller_id = 0;
3332 pin.send_len = 0; /* not used for flow table misses */
3334 flow_get_metadata(flow, &pin.fmd);
3336 connmgr_send_packet_in(ofproto->up.connmgr, &pin);
3339 static enum slow_path_reason
3340 process_special(struct ofproto_dpif *ofproto, const struct flow *flow,
3341 const struct ofport_dpif *ofport, const struct ofpbuf *packet)
3345 } else if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) {
3347 cfm_process_heartbeat(ofport->cfm, packet);
3350 } else if (ofport->bundle && ofport->bundle->lacp
3351 && flow->dl_type == htons(ETH_TYPE_LACP)) {
3353 lacp_process_packet(ofport->bundle->lacp, ofport, packet);
3356 } else if (ofproto->stp && stp_should_process_flow(flow)) {
3358 stp_process_packet(ofport, packet);
3366 static struct flow_miss *
3367 flow_miss_find(struct hmap *todo, const struct ofproto_dpif *ofproto,
3368 const struct flow *flow, uint32_t hash)
3370 struct flow_miss *miss;
3372 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
3373 if (miss->ofproto == ofproto && flow_equal(&miss->flow, flow)) {
3381 /* Partially Initializes 'op' as an "execute" operation for 'miss' and
3382 * 'packet'. The caller must initialize op->actions and op->actions_len. If
3383 * 'miss' is associated with a subfacet the caller must also initialize the
3384 * returned op->subfacet, and if anything needs to be freed after processing
3385 * the op, the caller must initialize op->garbage also. */
3387 init_flow_miss_execute_op(struct flow_miss *miss, struct ofpbuf *packet,
3388 struct flow_miss_op *op)
3390 if (miss->flow.vlan_tci != miss->initial_vals.vlan_tci) {
3391 /* This packet was received on a VLAN splinter port. We
3392 * added a VLAN to the packet to make the packet resemble
3393 * the flow, but the actions were composed assuming that
3394 * the packet contained no VLAN. So, we must remove the
3395 * VLAN header from the packet before trying to execute the
3397 eth_pop_vlan(packet);
3401 op->dpif_op.type = DPIF_OP_EXECUTE;
3402 op->dpif_op.u.execute.key = miss->key;
3403 op->dpif_op.u.execute.key_len = miss->key_len;
3404 op->dpif_op.u.execute.packet = packet;
3407 /* Helper for handle_flow_miss_without_facet() and
3408 * handle_flow_miss_with_facet(). */
3410 handle_flow_miss_common(struct rule_dpif *rule,
3411 struct ofpbuf *packet, const struct flow *flow)
3413 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3415 ofproto->n_matches++;
3417 if (rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
3419 * Extra-special case for fail-open mode.
3421 * We are in fail-open mode and the packet matched the fail-open
3422 * rule, but we are connected to a controller too. We should send
3423 * the packet up to the controller in the hope that it will try to
3424 * set up a flow and thereby allow us to exit fail-open.
3426 * See the top-level comment in fail-open.c for more information.
3428 send_packet_in_miss(ofproto, packet, flow);
3432 /* Figures out whether a flow that missed in 'ofproto', whose details are in
3433 * 'miss', is likely to be worth tracking in detail in userspace and (usually)
3434 * installing a datapath flow. The answer is usually "yes" (a return value of
3435 * true). However, for short flows the cost of bookkeeping is much higher than
3436 * the benefits, so when the datapath holds a large number of flows we impose
3437 * some heuristics to decide which flows are likely to be worth tracking. */
3439 flow_miss_should_make_facet(struct ofproto_dpif *ofproto,
3440 struct flow_miss *miss, uint32_t hash)
3442 if (!ofproto->governor) {
3445 n_subfacets = hmap_count(&ofproto->subfacets);
3446 if (n_subfacets * 2 <= ofproto->up.flow_eviction_threshold) {
3450 ofproto->governor = governor_create(ofproto->up.name);
3453 return governor_should_install_flow(ofproto->governor, hash,
3454 list_size(&miss->packets));
3457 /* Handles 'miss', which matches 'rule', without creating a facet or subfacet
3458 * or creating any datapath flow. May add an "execute" operation to 'ops' and
3459 * increment '*n_ops'. */
3461 handle_flow_miss_without_facet(struct flow_miss *miss,
3462 struct rule_dpif *rule,
3463 struct flow_miss_op *ops, size_t *n_ops)
3465 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3466 long long int now = time_msec();
3467 struct action_xlate_ctx ctx;
3468 struct ofpbuf *packet;
3470 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3471 struct flow_miss_op *op = &ops[*n_ops];
3472 struct dpif_flow_stats stats;
3473 struct ofpbuf odp_actions;
3475 COVERAGE_INC(facet_suppress);
3477 ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub);
3479 dpif_flow_stats_extract(&miss->flow, packet, now, &stats);
3480 rule_credit_stats(rule, &stats);
3482 action_xlate_ctx_init(&ctx, ofproto, &miss->flow,
3483 &miss->initial_vals, rule, 0, packet);
3484 ctx.resubmit_stats = &stats;
3485 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len,
3488 if (odp_actions.size) {
3489 struct dpif_execute *execute = &op->dpif_op.u.execute;
3491 init_flow_miss_execute_op(miss, packet, op);
3492 execute->actions = odp_actions.data;
3493 execute->actions_len = odp_actions.size;
3494 op->garbage = ofpbuf_get_uninit_pointer(&odp_actions);
3498 ofpbuf_uninit(&odp_actions);
3503 /* Handles 'miss', which matches 'facet'. May add any required datapath
3504 * operations to 'ops', incrementing '*n_ops' for each new op.
3506 * All of the packets in 'miss' are considered to have arrived at time 'now'.
3507 * This is really important only for new facets: if we just called time_msec()
3508 * here, then the new subfacet or its packets could look (occasionally) as
3509 * though it was used some time after the facet was used. That can make a
3510 * one-packet flow look like it has a nonzero duration, which looks odd in
3511 * e.g. NetFlow statistics. */
3513 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
3515 struct flow_miss_op *ops, size_t *n_ops)
3517 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3518 enum subfacet_path want_path;
3519 struct subfacet *subfacet;
3520 struct ofpbuf *packet;
3522 subfacet = subfacet_create(facet, miss, now);
3524 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3525 struct flow_miss_op *op = &ops[*n_ops];
3526 struct dpif_flow_stats stats;
3527 struct ofpbuf odp_actions;
3529 handle_flow_miss_common(facet->rule, packet, &miss->flow);
3531 ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub);
3532 if (!subfacet->actions || subfacet->slow) {
3533 subfacet_make_actions(subfacet, packet, &odp_actions);
3536 dpif_flow_stats_extract(&facet->flow, packet, now, &stats);
3537 subfacet_update_stats(subfacet, &stats);
3539 if (subfacet->actions_len) {
3540 struct dpif_execute *execute = &op->dpif_op.u.execute;
3542 init_flow_miss_execute_op(miss, packet, op);
3543 if (!subfacet->slow) {
3544 execute->actions = subfacet->actions;
3545 execute->actions_len = subfacet->actions_len;
3546 ofpbuf_uninit(&odp_actions);
3548 execute->actions = odp_actions.data;
3549 execute->actions_len = odp_actions.size;
3550 op->garbage = ofpbuf_get_uninit_pointer(&odp_actions);
3555 ofpbuf_uninit(&odp_actions);
3559 want_path = subfacet_want_path(subfacet->slow);
3560 if (miss->upcall_type == DPIF_UC_MISS || subfacet->path != want_path) {
3561 struct flow_miss_op *op = &ops[(*n_ops)++];
3562 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
3564 subfacet->path = want_path;
3567 op->dpif_op.type = DPIF_OP_FLOW_PUT;
3568 put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
3569 put->key = miss->key;
3570 put->key_len = miss->key_len;
3571 if (want_path == SF_FAST_PATH) {
3572 put->actions = subfacet->actions;
3573 put->actions_len = subfacet->actions_len;
3575 compose_slow_path(ofproto, &facet->flow, subfacet->slow,
3576 op->stub, sizeof op->stub,
3577 &put->actions, &put->actions_len);
3583 /* Handles flow miss 'miss'. May add any required datapath operations
3584 * to 'ops', incrementing '*n_ops' for each new op. */
3586 handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops,
3589 struct ofproto_dpif *ofproto = miss->ofproto;
3590 struct facet *facet;
3594 /* The caller must ensure that miss->hmap_node.hash contains
3595 * flow_hash(miss->flow, 0). */
3596 hash = miss->hmap_node.hash;
3598 facet = facet_lookup_valid(ofproto, &miss->flow, hash);
3600 struct rule_dpif *rule = rule_dpif_lookup(ofproto, &miss->flow);
3602 if (!flow_miss_should_make_facet(ofproto, miss, hash)) {
3603 handle_flow_miss_without_facet(miss, rule, ops, n_ops);
3607 facet = facet_create(rule, &miss->flow, hash);
3612 handle_flow_miss_with_facet(miss, facet, now, ops, n_ops);
3615 static struct drop_key *
3616 drop_key_lookup(const struct dpif_backer *backer, const struct nlattr *key,
3619 struct drop_key *drop_key;
3621 HMAP_FOR_EACH_WITH_HASH (drop_key, hmap_node, hash_bytes(key, key_len, 0),
3622 &backer->drop_keys) {
3623 if (drop_key->key_len == key_len
3624 && !memcmp(drop_key->key, key, key_len)) {
3632 drop_key_clear(struct dpif_backer *backer)
3634 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3635 struct drop_key *drop_key, *next;
3637 HMAP_FOR_EACH_SAFE (drop_key, next, hmap_node, &backer->drop_keys) {
3640 error = dpif_flow_del(backer->dpif, drop_key->key, drop_key->key_len,
3642 if (error && !VLOG_DROP_WARN(&rl)) {
3643 struct ds ds = DS_EMPTY_INITIALIZER;
3644 odp_flow_key_format(drop_key->key, drop_key->key_len, &ds);
3645 VLOG_WARN("Failed to delete drop key (%s) (%s)", strerror(error),
3650 hmap_remove(&backer->drop_keys, &drop_key->hmap_node);
3651 free(drop_key->key);
3656 /* Given a datpath, packet, and flow metadata ('backer', 'packet', and 'key'
3657 * respectively), populates 'flow' with the result of odp_flow_key_to_flow().
3658 * Optionally, if nonnull, populates 'fitnessp' with the fitness of 'flow' as
3659 * returned by odp_flow_key_to_flow(). Also, optionally populates 'ofproto'
3660 * with the ofproto_dpif, and 'odp_in_port' with the datapath in_port, that
3661 * 'packet' ingressed.
3663 * If 'ofproto' is nonnull, requires 'flow''s in_port to exist. Otherwise sets
3664 * 'flow''s in_port to OFPP_NONE.
3666 * This function does post-processing on data returned from
3667 * odp_flow_key_to_flow() to help make VLAN splinters transparent to the rest
3668 * of the upcall processing logic. In particular, if the extracted in_port is
3669 * a VLAN splinter port, it replaces flow->in_port by the "real" port, sets
3670 * flow->vlan_tci correctly for the VLAN of the VLAN splinter port, and pushes
3671 * a VLAN header onto 'packet' (if it is nonnull).
3673 * Optionally, if 'initial_vals' is nonnull, sets 'initial_vals->vlan_tci'
3674 * to the VLAN TCI with which the packet was really received, that is, the
3675 * actual VLAN TCI extracted by odp_flow_key_to_flow(). (This differs from
3676 * the value returned in flow->vlan_tci only for packets received on
3677 * VLAN splinters.) Also, if received on an IP tunnel, sets
3678 * 'initial_vals->tunnel_ip_tos' to the tunnel's IP TOS.
3680 * Similarly, this function also includes some logic to help with tunnels. It
3681 * may modify 'flow' as necessary to make the tunneling implementation
3682 * transparent to the upcall processing logic.
3684 * Returns 0 if successful, ENODEV if the parsed flow has no associated ofport,
3685 * or some other positive errno if there are other problems. */
3687 ofproto_receive(const struct dpif_backer *backer, struct ofpbuf *packet,
3688 const struct nlattr *key, size_t key_len,
3689 struct flow *flow, enum odp_key_fitness *fitnessp,
3690 struct ofproto_dpif **ofproto, uint32_t *odp_in_port,
3691 struct initial_vals *initial_vals)
3693 const struct ofport_dpif *port;
3694 enum odp_key_fitness fitness;
3697 fitness = odp_flow_key_to_flow(key, key_len, flow);
3698 if (fitness == ODP_FIT_ERROR) {
3704 initial_vals->vlan_tci = flow->vlan_tci;
3705 initial_vals->tunnel_ip_tos = flow->tunnel.ip_tos;
3709 *odp_in_port = flow->in_port;
3712 if (tnl_port_should_receive(flow)) {
3713 const struct ofport *ofport = tnl_port_receive(flow);
3715 flow->in_port = OFPP_NONE;
3718 port = ofport_dpif_cast(ofport);
3720 /* We can't reproduce 'key' from 'flow'. */
3721 fitness = fitness == ODP_FIT_PERFECT ? ODP_FIT_TOO_MUCH : fitness;
3723 /* XXX: Since the tunnel module is not scoped per backer, it's
3724 * theoretically possible that we'll receive an ofport belonging to an
3725 * entirely different datapath. In practice, this can't happen because
3726 * no platforms has two separate datapaths which each support
3728 ovs_assert(ofproto_dpif_cast(port->up.ofproto)->backer == backer);
3730 port = odp_port_to_ofport(backer, flow->in_port);
3732 flow->in_port = OFPP_NONE;
3736 flow->in_port = port->up.ofp_port;
3737 if (vsp_adjust_flow(ofproto_dpif_cast(port->up.ofproto), flow)) {
3739 /* Make the packet resemble the flow, so that it gets sent to
3740 * an OpenFlow controller properly, so that it looks correct
3741 * for sFlow, and so that flow_extract() will get the correct
3742 * vlan_tci if it is called on 'packet'.
3744 * The allocated space inside 'packet' probably also contains
3745 * 'key', that is, both 'packet' and 'key' are probably part of
3746 * a struct dpif_upcall (see the large comment on that
3747 * structure definition), so pushing data on 'packet' is in
3748 * general not a good idea since it could overwrite 'key' or
3749 * free it as a side effect. However, it's OK in this special
3750 * case because we know that 'packet' is inside a Netlink
3751 * attribute: pushing 4 bytes will just overwrite the 4-byte
3752 * "struct nlattr", which is fine since we don't need that
3753 * header anymore. */
3754 eth_push_vlan(packet, flow->vlan_tci);
3756 /* We can't reproduce 'key' from 'flow'. */
3757 fitness = fitness == ODP_FIT_PERFECT ? ODP_FIT_TOO_MUCH : fitness;
3763 *ofproto = ofproto_dpif_cast(port->up.ofproto);
3768 *fitnessp = fitness;
3774 handle_miss_upcalls(struct dpif_backer *backer, struct dpif_upcall *upcalls,
3777 struct dpif_upcall *upcall;
3778 struct flow_miss *miss;
3779 struct flow_miss misses[FLOW_MISS_MAX_BATCH];
3780 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
3781 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
3791 /* Construct the to-do list.
3793 * This just amounts to extracting the flow from each packet and sticking
3794 * the packets that have the same flow in the same "flow_miss" structure so
3795 * that we can process them together. */
3798 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
3799 struct flow_miss *miss = &misses[n_misses];
3800 struct flow_miss *existing_miss;
3801 struct ofproto_dpif *ofproto;
3802 uint32_t odp_in_port;
3807 error = ofproto_receive(backer, upcall->packet, upcall->key,
3808 upcall->key_len, &flow, &miss->key_fitness,
3809 &ofproto, &odp_in_port, &miss->initial_vals);
3810 if (error == ENODEV) {
3811 struct drop_key *drop_key;
3813 /* Received packet on port for which we couldn't associate
3814 * an ofproto. This can happen if a port is removed while
3815 * traffic is being received. Print a rate-limited message
3816 * in case it happens frequently. Install a drop flow so
3817 * that future packets of the flow are inexpensively dropped
3819 VLOG_INFO_RL(&rl, "received packet on unassociated port %"PRIu32,
3822 drop_key = drop_key_lookup(backer, upcall->key, upcall->key_len);
3824 drop_key = xmalloc(sizeof *drop_key);
3825 drop_key->key = xmemdup(upcall->key, upcall->key_len);
3826 drop_key->key_len = upcall->key_len;
3828 hmap_insert(&backer->drop_keys, &drop_key->hmap_node,
3829 hash_bytes(drop_key->key, drop_key->key_len, 0));
3830 dpif_flow_put(backer->dpif, DPIF_FP_CREATE | DPIF_FP_MODIFY,
3831 drop_key->key, drop_key->key_len, NULL, 0, NULL);
3839 ofproto->n_missed++;
3840 flow_extract(upcall->packet, flow.skb_priority, flow.skb_mark,
3841 &flow.tunnel, flow.in_port, &miss->flow);
3843 /* Add other packets to a to-do list. */
3844 hash = flow_hash(&miss->flow, 0);
3845 existing_miss = flow_miss_find(&todo, ofproto, &miss->flow, hash);
3846 if (!existing_miss) {
3847 hmap_insert(&todo, &miss->hmap_node, hash);
3848 miss->ofproto = ofproto;
3849 miss->key = upcall->key;
3850 miss->key_len = upcall->key_len;
3851 miss->upcall_type = upcall->type;
3852 miss->odp_in_port = odp_in_port;
3853 list_init(&miss->packets);
3857 miss = existing_miss;
3859 list_push_back(&miss->packets, &upcall->packet->list_node);
3862 /* Process each element in the to-do list, constructing the set of
3863 * operations to batch. */
3865 HMAP_FOR_EACH (miss, hmap_node, &todo) {
3866 handle_flow_miss(miss, flow_miss_ops, &n_ops);
3868 ovs_assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
3870 /* Execute batch. */
3871 for (i = 0; i < n_ops; i++) {
3872 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
3874 dpif_operate(backer->dpif, dpif_ops, n_ops);
3877 for (i = 0; i < n_ops; i++) {
3878 free(flow_miss_ops[i].garbage);
3880 hmap_destroy(&todo);
3883 static enum { SFLOW_UPCALL, MISS_UPCALL, BAD_UPCALL }
3884 classify_upcall(const struct dpif_upcall *upcall)
3886 union user_action_cookie cookie;
3888 /* First look at the upcall type. */
3889 switch (upcall->type) {
3890 case DPIF_UC_ACTION:
3896 case DPIF_N_UC_TYPES:
3898 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
3902 /* "action" upcalls need a closer look. */
3903 if (!upcall->userdata) {
3904 VLOG_WARN_RL(&rl, "action upcall missing cookie");
3907 if (nl_attr_get_size(upcall->userdata) != sizeof(cookie)) {
3908 VLOG_WARN_RL(&rl, "action upcall cookie has unexpected size %zu",
3909 nl_attr_get_size(upcall->userdata));
3912 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof(cookie));
3913 switch (cookie.type) {
3914 case USER_ACTION_COOKIE_SFLOW:
3915 return SFLOW_UPCALL;
3917 case USER_ACTION_COOKIE_SLOW_PATH:
3920 case USER_ACTION_COOKIE_UNSPEC:
3922 VLOG_WARN_RL(&rl, "invalid user cookie : 0x%"PRIx64,
3923 nl_attr_get_u64(upcall->userdata));
3929 handle_sflow_upcall(struct dpif_backer *backer,
3930 const struct dpif_upcall *upcall)
3932 struct ofproto_dpif *ofproto;
3933 union user_action_cookie cookie;
3935 uint32_t odp_in_port;
3937 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3938 &flow, NULL, &ofproto, &odp_in_port, NULL)
3939 || !ofproto->sflow) {
3943 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof(cookie));
3944 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
3945 odp_in_port, &cookie);
3949 handle_upcalls(struct dpif_backer *backer, unsigned int max_batch)
3951 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
3952 struct ofpbuf miss_bufs[FLOW_MISS_MAX_BATCH];
3953 uint64_t miss_buf_stubs[FLOW_MISS_MAX_BATCH][4096 / 8];
3958 ovs_assert(max_batch <= FLOW_MISS_MAX_BATCH);
3961 for (n_processed = 0; n_processed < max_batch; n_processed++) {
3962 struct dpif_upcall *upcall = &misses[n_misses];
3963 struct ofpbuf *buf = &miss_bufs[n_misses];
3966 ofpbuf_use_stub(buf, miss_buf_stubs[n_misses],
3967 sizeof miss_buf_stubs[n_misses]);
3968 error = dpif_recv(backer->dpif, upcall, buf);
3974 switch (classify_upcall(upcall)) {
3976 /* Handle it later. */
3981 handle_sflow_upcall(backer, upcall);
3991 /* Handle deferred MISS_UPCALL processing. */
3992 handle_miss_upcalls(backer, misses, n_misses);
3993 for (i = 0; i < n_misses; i++) {
3994 ofpbuf_uninit(&miss_bufs[i]);
4000 /* Flow expiration. */
4002 static int subfacet_max_idle(const struct ofproto_dpif *);
4003 static void update_stats(struct dpif_backer *);
4004 static void rule_expire(struct rule_dpif *);
4005 static void expire_subfacets(struct ofproto_dpif *, int dp_max_idle);
4007 /* This function is called periodically by run(). Its job is to collect
4008 * updates for the flows that have been installed into the datapath, most
4009 * importantly when they last were used, and then use that information to
4010 * expire flows that have not been used recently.
4012 * Returns the number of milliseconds after which it should be called again. */
4014 expire(struct dpif_backer *backer)
4016 struct ofproto_dpif *ofproto;
4017 int max_idle = INT32_MAX;
4019 /* Periodically clear out the drop keys in an effort to keep them
4020 * relatively few. */
4021 drop_key_clear(backer);
4023 /* Update stats for each flow in the backer. */
4024 update_stats(backer);
4026 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4027 struct rule *rule, *next_rule;
4030 if (ofproto->backer != backer) {
4034 /* Expire subfacets that have been idle too long. */
4035 dp_max_idle = subfacet_max_idle(ofproto);
4036 expire_subfacets(ofproto, dp_max_idle);
4038 max_idle = MIN(max_idle, dp_max_idle);
4040 /* Expire OpenFlow flows whose idle_timeout or hard_timeout
4042 LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
4043 &ofproto->up.expirable) {
4044 rule_expire(rule_dpif_cast(rule));
4047 /* All outstanding data in existing flows has been accounted, so it's a
4048 * good time to do bond rebalancing. */
4049 if (ofproto->has_bonded_bundles) {
4050 struct ofbundle *bundle;
4052 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
4054 bond_rebalance(bundle->bond, &backer->revalidate_set);
4060 return MIN(max_idle, 1000);
4063 /* Updates flow table statistics given that the datapath just reported 'stats'
4064 * as 'subfacet''s statistics. */
4066 update_subfacet_stats(struct subfacet *subfacet,
4067 const struct dpif_flow_stats *stats)
4069 struct facet *facet = subfacet->facet;
4071 if (stats->n_packets >= subfacet->dp_packet_count) {
4072 uint64_t extra = stats->n_packets - subfacet->dp_packet_count;
4073 facet->packet_count += extra;
4075 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
4078 if (stats->n_bytes >= subfacet->dp_byte_count) {
4079 facet->byte_count += stats->n_bytes - subfacet->dp_byte_count;
4081 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
4084 subfacet->dp_packet_count = stats->n_packets;
4085 subfacet->dp_byte_count = stats->n_bytes;
4087 facet->tcp_flags |= stats->tcp_flags;
4089 subfacet_update_time(subfacet, stats->used);
4090 if (facet->accounted_bytes < facet->byte_count) {
4092 facet_account(facet);
4093 facet->accounted_bytes = facet->byte_count;
4095 facet_push_stats(facet);
4098 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
4099 * about, or a flow that shouldn't be installed but was anyway. Delete it. */
4101 delete_unexpected_flow(struct ofproto_dpif *ofproto,
4102 const struct nlattr *key, size_t key_len)
4104 if (!VLOG_DROP_WARN(&rl)) {
4108 odp_flow_key_format(key, key_len, &s);
4109 VLOG_WARN("unexpected flow on %s: %s", ofproto->up.name, ds_cstr(&s));
4113 COVERAGE_INC(facet_unexpected);
4114 dpif_flow_del(ofproto->backer->dpif, key, key_len, NULL);
4117 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
4119 * This function also pushes statistics updates to rules which each facet
4120 * resubmits into. Generally these statistics will be accurate. However, if a
4121 * facet changes the rule it resubmits into at some time in between
4122 * update_stats() runs, it is possible that statistics accrued to the
4123 * old rule will be incorrectly attributed to the new rule. This could be
4124 * avoided by calling update_stats() whenever rules are created or
4125 * deleted. However, the performance impact of making so many calls to the
4126 * datapath do not justify the benefit of having perfectly accurate statistics.
4128 * In addition, this function maintains per ofproto flow hit counts. The patch
4129 * port is not treated specially. e.g. A packet ingress from br0 patched into
4130 * br1 will increase the hit count of br0 by 1, however, does not affect
4131 * the hit or miss counts of br1.
4134 update_stats(struct dpif_backer *backer)
4136 const struct dpif_flow_stats *stats;
4137 struct dpif_flow_dump dump;
4138 const struct nlattr *key;
4141 dpif_flow_dump_start(&dump, backer->dpif);
4142 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
4144 struct subfacet *subfacet;
4145 struct ofproto_dpif *ofproto;
4146 struct ofport_dpif *ofport;
4149 if (ofproto_receive(backer, NULL, key, key_len, &flow, NULL, &ofproto,
4154 ofport = get_ofp_port(ofproto, flow.in_port);
4155 if (ofport && ofport->tnl_port) {
4156 netdev_vport_inc_rx(ofport->up.netdev, stats);
4159 key_hash = odp_flow_key_hash(key, key_len);
4160 subfacet = subfacet_find(ofproto, key, key_len, key_hash);
4161 switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
4163 /* Update ofproto_dpif's hit count. */
4164 if (stats->n_packets > subfacet->dp_packet_count) {
4165 uint64_t delta = stats->n_packets - subfacet->dp_packet_count;
4166 dpif_stats_update_hit_count(ofproto, delta);
4169 update_subfacet_stats(subfacet, stats);
4173 /* Stats are updated per-packet. */
4176 case SF_NOT_INSTALLED:
4178 delete_unexpected_flow(ofproto, key, key_len);
4182 dpif_flow_dump_done(&dump);
4185 /* Calculates and returns the number of milliseconds of idle time after which
4186 * subfacets should expire from the datapath. When a subfacet expires, we fold
4187 * its statistics into its facet, and when a facet's last subfacet expires, we
4188 * fold its statistic into its rule. */
4190 subfacet_max_idle(const struct ofproto_dpif *ofproto)
4193 * Idle time histogram.
4195 * Most of the time a switch has a relatively small number of subfacets.
4196 * When this is the case we might as well keep statistics for all of them
4197 * in userspace and to cache them in the kernel datapath for performance as
4200 * As the number of subfacets increases, the memory required to maintain
4201 * statistics about them in userspace and in the kernel becomes
4202 * significant. However, with a large number of subfacets it is likely
4203 * that only a few of them are "heavy hitters" that consume a large amount
4204 * of bandwidth. At this point, only heavy hitters are worth caching in
4205 * the kernel and maintaining in userspaces; other subfacets we can
4208 * The technique used to compute the idle time is to build a histogram with
4209 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
4210 * that is installed in the kernel gets dropped in the appropriate bucket.
4211 * After the histogram has been built, we compute the cutoff so that only
4212 * the most-recently-used 1% of subfacets (but at least
4213 * ofproto->up.flow_eviction_threshold flows) are kept cached. At least
4214 * the most-recently-used bucket of subfacets is kept, so actually an
4215 * arbitrary number of subfacets can be kept in any given expiration run
4216 * (though the next run will delete most of those unless they receive
4219 * This requires a second pass through the subfacets, in addition to the
4220 * pass made by update_stats(), because the former function never looks at
4221 * uninstallable subfacets.
4223 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
4224 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
4225 int buckets[N_BUCKETS] = { 0 };
4226 int total, subtotal, bucket;
4227 struct subfacet *subfacet;
4231 total = hmap_count(&ofproto->subfacets);
4232 if (total <= ofproto->up.flow_eviction_threshold) {
4233 return N_BUCKETS * BUCKET_WIDTH;
4236 /* Build histogram. */
4238 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
4239 long long int idle = now - subfacet->used;
4240 int bucket = (idle <= 0 ? 0
4241 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
4242 : (unsigned int) idle / BUCKET_WIDTH);
4246 /* Find the first bucket whose flows should be expired. */
4247 subtotal = bucket = 0;
4249 subtotal += buckets[bucket++];
4250 } while (bucket < N_BUCKETS &&
4251 subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100));
4253 if (VLOG_IS_DBG_ENABLED()) {
4257 ds_put_cstr(&s, "keep");
4258 for (i = 0; i < N_BUCKETS; i++) {
4260 ds_put_cstr(&s, ", drop");
4263 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
4266 VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s));
4270 return bucket * BUCKET_WIDTH;
4274 expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle)
4276 /* Cutoff time for most flows. */
4277 long long int normal_cutoff = time_msec() - dp_max_idle;
4279 /* We really want to keep flows for special protocols around, so use a more
4280 * conservative cutoff. */
4281 long long int special_cutoff = time_msec() - 10000;
4283 struct subfacet *subfacet, *next_subfacet;
4284 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
4288 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
4289 &ofproto->subfacets) {
4290 long long int cutoff;
4292 cutoff = (subfacet->slow & (SLOW_CFM | SLOW_LACP | SLOW_STP)
4295 if (subfacet->used < cutoff) {
4296 if (subfacet->path != SF_NOT_INSTALLED) {
4297 batch[n_batch++] = subfacet;
4298 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
4299 subfacet_destroy_batch(ofproto, batch, n_batch);
4303 subfacet_destroy(subfacet);
4309 subfacet_destroy_batch(ofproto, batch, n_batch);
4313 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
4314 * then delete it entirely. */
4316 rule_expire(struct rule_dpif *rule)
4318 struct facet *facet, *next_facet;
4322 if (rule->up.pending) {
4323 /* We'll have to expire it later. */
4327 /* Has 'rule' expired? */
4329 if (rule->up.hard_timeout
4330 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
4331 reason = OFPRR_HARD_TIMEOUT;
4332 } else if (rule->up.idle_timeout
4333 && now > rule->up.used + rule->up.idle_timeout * 1000) {
4334 reason = OFPRR_IDLE_TIMEOUT;
4339 COVERAGE_INC(ofproto_dpif_expired);
4341 /* Update stats. (This is a no-op if the rule expired due to an idle
4342 * timeout, because that only happens when the rule has no facets left.) */
4343 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4344 facet_remove(facet);
4347 /* Get rid of the rule. */
4348 ofproto_rule_expire(&rule->up, reason);
4353 /* Creates and returns a new facet owned by 'rule', given a 'flow'.
4355 * The caller must already have determined that no facet with an identical
4356 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
4357 * the ofproto's classifier table.
4359 * 'hash' must be the return value of flow_hash(flow, 0).
4361 * The facet will initially have no subfacets. The caller should create (at
4362 * least) one subfacet with subfacet_create(). */
4363 static struct facet *
4364 facet_create(struct rule_dpif *rule, const struct flow *flow, uint32_t hash)
4366 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4367 struct facet *facet;
4369 facet = xzalloc(sizeof *facet);
4370 facet->used = time_msec();
4371 hmap_insert(&ofproto->facets, &facet->hmap_node, hash);
4372 list_push_back(&rule->facets, &facet->list_node);
4374 facet->flow = *flow;
4375 list_init(&facet->subfacets);
4376 netflow_flow_init(&facet->nf_flow);
4377 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
4383 facet_free(struct facet *facet)
4388 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
4389 * 'packet', which arrived on 'in_port'. */
4391 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
4392 const struct nlattr *odp_actions, size_t actions_len,
4393 struct ofpbuf *packet)
4395 struct odputil_keybuf keybuf;
4399 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4400 odp_flow_key_from_flow(&key, flow,
4401 ofp_port_to_odp_port(ofproto, flow->in_port));
4403 error = dpif_execute(ofproto->backer->dpif, key.data, key.size,
4404 odp_actions, actions_len, packet);
4408 /* Remove 'facet' from 'ofproto' and free up the associated memory:
4410 * - If 'facet' was installed in the datapath, uninstalls it and updates its
4411 * rule's statistics, via subfacet_uninstall().
4413 * - Removes 'facet' from its rule and from ofproto->facets.
4416 facet_remove(struct facet *facet)
4418 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4419 struct subfacet *subfacet, *next_subfacet;
4421 ovs_assert(!list_is_empty(&facet->subfacets));
4423 /* First uninstall all of the subfacets to get final statistics. */
4424 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4425 subfacet_uninstall(subfacet);
4428 /* Flush the final stats to the rule.
4430 * This might require us to have at least one subfacet around so that we
4431 * can use its actions for accounting in facet_account(), which is why we
4432 * have uninstalled but not yet destroyed the subfacets. */
4433 facet_flush_stats(facet);
4435 /* Now we're really all done so destroy everything. */
4436 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
4437 &facet->subfacets) {
4438 subfacet_destroy__(subfacet);
4440 hmap_remove(&ofproto->facets, &facet->hmap_node);
4441 list_remove(&facet->list_node);
4445 /* Feed information from 'facet' back into the learning table to keep it in
4446 * sync with what is actually flowing through the datapath. */
4448 facet_learn(struct facet *facet)
4450 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4451 struct subfacet *subfacet= CONTAINER_OF(list_front(&facet->subfacets),
4452 struct subfacet, list_node);
4453 struct action_xlate_ctx ctx;
4455 if (!facet->has_learn
4456 && !facet->has_normal
4457 && (!facet->has_fin_timeout
4458 || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
4462 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
4463 &subfacet->initial_vals,
4464 facet->rule, facet->tcp_flags, NULL);
4465 ctx.may_learn = true;
4466 xlate_actions_for_side_effects(&ctx, facet->rule->up.ofpacts,
4467 facet->rule->up.ofpacts_len);
4471 facet_account(struct facet *facet)
4473 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4474 struct subfacet *subfacet = facet_get_subfacet(facet);
4475 const struct nlattr *a;
4480 if (!facet->has_normal || !ofproto->has_bonded_bundles) {
4483 n_bytes = facet->byte_count - facet->accounted_bytes;
4485 /* This loop feeds byte counters to bond_account() for rebalancing to use
4486 * as a basis. We also need to track the actual VLAN on which the packet
4487 * is going to be sent to ensure that it matches the one passed to
4488 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
4491 * We use the actions from an arbitrary subfacet because they should all
4492 * be equally valid for our purpose. */
4493 vlan_tci = facet->flow.vlan_tci;
4494 NL_ATTR_FOR_EACH_UNSAFE (a, left,
4495 subfacet->actions, subfacet->actions_len) {
4496 const struct ovs_action_push_vlan *vlan;
4497 struct ofport_dpif *port;
4499 switch (nl_attr_type(a)) {
4500 case OVS_ACTION_ATTR_OUTPUT:
4501 port = get_odp_port(ofproto, nl_attr_get_u32(a));
4502 if (port && port->bundle && port->bundle->bond) {
4503 bond_account(port->bundle->bond, &facet->flow,
4504 vlan_tci_to_vid(vlan_tci), n_bytes);
4508 case OVS_ACTION_ATTR_POP_VLAN:
4509 vlan_tci = htons(0);
4512 case OVS_ACTION_ATTR_PUSH_VLAN:
4513 vlan = nl_attr_get(a);
4514 vlan_tci = vlan->vlan_tci;
4520 /* Returns true if the only action for 'facet' is to send to the controller.
4521 * (We don't report NetFlow expiration messages for such facets because they
4522 * are just part of the control logic for the network, not real traffic). */
4524 facet_is_controller_flow(struct facet *facet)
4527 const struct rule *rule = &facet->rule->up;
4528 const struct ofpact *ofpacts = rule->ofpacts;
4529 size_t ofpacts_len = rule->ofpacts_len;
4531 if (ofpacts_len > 0 &&
4532 ofpacts->type == OFPACT_CONTROLLER &&
4533 ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len)) {
4540 /* Folds all of 'facet''s statistics into its rule. Also updates the
4541 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
4542 * 'facet''s statistics in the datapath should have been zeroed and folded into
4543 * its packet and byte counts before this function is called. */
4545 facet_flush_stats(struct facet *facet)
4547 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4548 struct subfacet *subfacet;
4550 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4551 ovs_assert(!subfacet->dp_byte_count);
4552 ovs_assert(!subfacet->dp_packet_count);
4555 facet_push_stats(facet);
4556 if (facet->accounted_bytes < facet->byte_count) {
4557 facet_account(facet);
4558 facet->accounted_bytes = facet->byte_count;
4561 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
4562 struct ofexpired expired;
4563 expired.flow = facet->flow;
4564 expired.packet_count = facet->packet_count;
4565 expired.byte_count = facet->byte_count;
4566 expired.used = facet->used;
4567 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4570 facet->rule->packet_count += facet->packet_count;
4571 facet->rule->byte_count += facet->byte_count;
4573 /* Reset counters to prevent double counting if 'facet' ever gets
4575 facet_reset_counters(facet);
4577 netflow_flow_clear(&facet->nf_flow);
4578 facet->tcp_flags = 0;
4581 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
4582 * Returns it if found, otherwise a null pointer.
4584 * 'hash' must be the return value of flow_hash(flow, 0).
4586 * The returned facet might need revalidation; use facet_lookup_valid()
4587 * instead if that is important. */
4588 static struct facet *
4589 facet_find(struct ofproto_dpif *ofproto,
4590 const struct flow *flow, uint32_t hash)
4592 struct facet *facet;
4594 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, hash, &ofproto->facets) {
4595 if (flow_equal(flow, &facet->flow)) {
4603 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
4604 * Returns it if found, otherwise a null pointer.
4606 * 'hash' must be the return value of flow_hash(flow, 0).
4608 * The returned facet is guaranteed to be valid. */
4609 static struct facet *
4610 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow,
4613 struct facet *facet;
4615 facet = facet_find(ofproto, flow, hash);
4617 && (ofproto->backer->need_revalidate
4618 || tag_set_intersects(&ofproto->backer->revalidate_set,
4620 facet_revalidate(facet);
4622 /* facet_revalidate() may have destroyed 'facet'. */
4623 facet = facet_find(ofproto, flow, hash);
4629 /* Return a subfacet from 'facet'. A facet consists of one or more
4630 * subfacets, and this function returns one of them. */
4631 static struct subfacet *facet_get_subfacet(struct facet *facet)
4633 return CONTAINER_OF(list_front(&facet->subfacets), struct subfacet,
4638 subfacet_path_to_string(enum subfacet_path path)
4641 case SF_NOT_INSTALLED:
4642 return "not installed";
4644 return "in fast path";
4646 return "in slow path";
4652 /* Returns the path in which a subfacet should be installed if its 'slow'
4653 * member has the specified value. */
4654 static enum subfacet_path
4655 subfacet_want_path(enum slow_path_reason slow)
4657 return slow ? SF_SLOW_PATH : SF_FAST_PATH;
4660 /* Returns true if 'subfacet' needs to have its datapath flow updated,
4661 * supposing that its actions have been recalculated as 'want_actions' and that
4662 * 'slow' is nonzero iff 'subfacet' should be in the slow path. */
4664 subfacet_should_install(struct subfacet *subfacet, enum slow_path_reason slow,
4665 const struct ofpbuf *want_actions)
4667 enum subfacet_path want_path = subfacet_want_path(slow);
4668 return (want_path != subfacet->path
4669 || (want_path == SF_FAST_PATH
4670 && (subfacet->actions_len != want_actions->size
4671 || memcmp(subfacet->actions, want_actions->data,
4672 subfacet->actions_len))));
4676 facet_check_consistency(struct facet *facet)
4678 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
4680 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4682 uint64_t odp_actions_stub[1024 / 8];
4683 struct ofpbuf odp_actions;
4685 struct rule_dpif *rule;
4686 struct subfacet *subfacet;
4687 bool may_log = false;
4690 /* Check the rule for consistency. */
4691 rule = rule_dpif_lookup(ofproto, &facet->flow);
4692 ok = rule == facet->rule;
4694 may_log = !VLOG_DROP_WARN(&rl);
4699 flow_format(&s, &facet->flow);
4700 ds_put_format(&s, ": facet associated with wrong rule (was "
4701 "table=%"PRIu8",", facet->rule->up.table_id);
4702 cls_rule_format(&facet->rule->up.cr, &s);
4703 ds_put_format(&s, ") (should have been table=%"PRIu8",",
4705 cls_rule_format(&rule->up.cr, &s);
4706 ds_put_char(&s, ')');
4708 VLOG_WARN("%s", ds_cstr(&s));
4713 /* Check the datapath actions for consistency. */
4714 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
4715 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4716 enum subfacet_path want_path;
4717 struct action_xlate_ctx ctx;
4720 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
4721 &subfacet->initial_vals, rule, 0, NULL);
4722 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len,
4725 if (subfacet->path == SF_NOT_INSTALLED) {
4726 /* This only happens if the datapath reported an error when we
4727 * tried to install the flow. Don't flag another error here. */
4731 want_path = subfacet_want_path(subfacet->slow);
4732 if (want_path == SF_SLOW_PATH && subfacet->path == SF_SLOW_PATH) {
4733 /* The actions for slow-path flows may legitimately vary from one
4734 * packet to the next. We're done. */
4738 if (!subfacet_should_install(subfacet, subfacet->slow, &odp_actions)) {
4742 /* Inconsistency! */
4744 may_log = !VLOG_DROP_WARN(&rl);
4748 /* Rate-limited, skip reporting. */
4753 odp_flow_key_format(subfacet->key, subfacet->key_len, &s);
4755 ds_put_cstr(&s, ": inconsistency in subfacet");
4756 if (want_path != subfacet->path) {
4757 enum odp_key_fitness fitness = subfacet->key_fitness;
4759 ds_put_format(&s, " (%s, fitness=%s)",
4760 subfacet_path_to_string(subfacet->path),
4761 odp_key_fitness_to_string(fitness));
4762 ds_put_format(&s, " (should have been %s)",
4763 subfacet_path_to_string(want_path));
4764 } else if (want_path == SF_FAST_PATH) {
4765 ds_put_cstr(&s, " (actions were: ");
4766 format_odp_actions(&s, subfacet->actions,
4767 subfacet->actions_len);
4768 ds_put_cstr(&s, ") (correct actions: ");
4769 format_odp_actions(&s, odp_actions.data, odp_actions.size);
4770 ds_put_char(&s, ')');
4772 ds_put_cstr(&s, " (actions: ");
4773 format_odp_actions(&s, subfacet->actions,
4774 subfacet->actions_len);
4775 ds_put_char(&s, ')');
4777 VLOG_WARN("%s", ds_cstr(&s));
4780 ofpbuf_uninit(&odp_actions);
4785 /* Re-searches the classifier for 'facet':
4787 * - If the rule found is different from 'facet''s current rule, moves
4788 * 'facet' to the new rule and recompiles its actions.
4790 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
4791 * where it is and recompiles its actions anyway.
4793 * - If any of 'facet''s subfacets correspond to a new flow according to
4794 * ofproto_receive(), 'facet' is removed. */
4796 facet_revalidate(struct facet *facet)
4798 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4800 struct nlattr *odp_actions;
4803 struct actions *new_actions;
4805 struct action_xlate_ctx ctx;
4806 uint64_t odp_actions_stub[1024 / 8];
4807 struct ofpbuf odp_actions;
4809 struct rule_dpif *new_rule;
4810 struct subfacet *subfacet;
4813 COVERAGE_INC(facet_revalidate);
4815 /* Check that child subfacets still correspond to this facet. Tunnel
4816 * configuration changes could cause a subfacet's OpenFlow in_port to
4818 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4819 struct ofproto_dpif *recv_ofproto;
4820 struct flow recv_flow;
4823 error = ofproto_receive(ofproto->backer, NULL, subfacet->key,
4824 subfacet->key_len, &recv_flow, NULL,
4825 &recv_ofproto, NULL, NULL);
4827 || recv_ofproto != ofproto
4828 || memcmp(&recv_flow, &facet->flow, sizeof recv_flow)) {
4829 facet_remove(facet);
4834 new_rule = rule_dpif_lookup(ofproto, &facet->flow);
4836 /* Calculate new datapath actions.
4838 * We do not modify any 'facet' state yet, because we might need to, e.g.,
4839 * emit a NetFlow expiration and, if so, we need to have the old state
4840 * around to properly compose it. */
4842 /* If the datapath actions changed or the installability changed,
4843 * then we need to talk to the datapath. */
4846 memset(&ctx, 0, sizeof ctx);
4847 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
4848 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4849 enum slow_path_reason slow;
4851 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
4852 &subfacet->initial_vals, new_rule, 0, NULL);
4853 xlate_actions(&ctx, new_rule->up.ofpacts, new_rule->up.ofpacts_len,
4856 slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
4857 if (subfacet_should_install(subfacet, slow, &odp_actions)) {
4858 struct dpif_flow_stats stats;
4860 subfacet_install(subfacet,
4861 odp_actions.data, odp_actions.size, &stats, slow);
4862 subfacet_update_stats(subfacet, &stats);
4865 new_actions = xcalloc(list_size(&facet->subfacets),
4866 sizeof *new_actions);
4868 new_actions[i].odp_actions = xmemdup(odp_actions.data,
4870 new_actions[i].actions_len = odp_actions.size;
4875 ofpbuf_uninit(&odp_actions);
4878 facet_flush_stats(facet);
4881 /* Update 'facet' now that we've taken care of all the old state. */
4882 facet->tags = ctx.tags;
4883 facet->nf_flow.output_iface = ctx.nf_output_iface;
4884 facet->has_learn = ctx.has_learn;
4885 facet->has_normal = ctx.has_normal;
4886 facet->has_fin_timeout = ctx.has_fin_timeout;
4887 facet->mirrors = ctx.mirrors;
4890 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4891 subfacet->slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
4893 if (new_actions && new_actions[i].odp_actions) {
4894 free(subfacet->actions);
4895 subfacet->actions = new_actions[i].odp_actions;
4896 subfacet->actions_len = new_actions[i].actions_len;
4902 if (facet->rule != new_rule) {
4903 COVERAGE_INC(facet_changed_rule);
4904 list_remove(&facet->list_node);
4905 list_push_back(&new_rule->facets, &facet->list_node);
4906 facet->rule = new_rule;
4907 facet->used = new_rule->up.created;
4908 facet->prev_used = facet->used;
4912 /* Updates 'facet''s used time. Caller is responsible for calling
4913 * facet_push_stats() to update the flows which 'facet' resubmits into. */
4915 facet_update_time(struct facet *facet, long long int used)
4917 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4918 if (used > facet->used) {
4920 ofproto_rule_update_used(&facet->rule->up, used);
4921 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
4926 facet_reset_counters(struct facet *facet)
4928 facet->packet_count = 0;
4929 facet->byte_count = 0;
4930 facet->prev_packet_count = 0;
4931 facet->prev_byte_count = 0;
4932 facet->accounted_bytes = 0;
4936 facet_push_stats(struct facet *facet)
4938 struct dpif_flow_stats stats;
4940 ovs_assert(facet->packet_count >= facet->prev_packet_count);
4941 ovs_assert(facet->byte_count >= facet->prev_byte_count);
4942 ovs_assert(facet->used >= facet->prev_used);
4944 stats.n_packets = facet->packet_count - facet->prev_packet_count;
4945 stats.n_bytes = facet->byte_count - facet->prev_byte_count;
4946 stats.used = facet->used;
4947 stats.tcp_flags = 0;
4949 if (stats.n_packets || stats.n_bytes || facet->used > facet->prev_used) {
4950 facet->prev_packet_count = facet->packet_count;
4951 facet->prev_byte_count = facet->byte_count;
4952 facet->prev_used = facet->used;
4954 flow_push_stats(facet, &stats);
4956 update_mirror_stats(ofproto_dpif_cast(facet->rule->up.ofproto),
4957 facet->mirrors, stats.n_packets, stats.n_bytes);
4962 rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats)
4964 rule->packet_count += stats->n_packets;
4965 rule->byte_count += stats->n_bytes;
4966 ofproto_rule_update_used(&rule->up, stats->used);
4969 /* Pushes flow statistics to the rules which 'facet->flow' resubmits
4970 * into given 'facet->rule''s actions and mirrors. */
4972 flow_push_stats(struct facet *facet, const struct dpif_flow_stats *stats)
4974 struct rule_dpif *rule = facet->rule;
4975 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4976 struct subfacet *subfacet = facet_get_subfacet(facet);
4977 struct action_xlate_ctx ctx;
4979 ofproto_rule_update_used(&rule->up, stats->used);
4981 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
4982 &subfacet->initial_vals, rule, 0, NULL);
4983 ctx.resubmit_stats = stats;
4984 xlate_actions_for_side_effects(&ctx, rule->up.ofpacts,
4985 rule->up.ofpacts_len);
4990 static struct subfacet *
4991 subfacet_find(struct ofproto_dpif *ofproto,
4992 const struct nlattr *key, size_t key_len, uint32_t key_hash)
4994 struct subfacet *subfacet;
4996 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
4997 &ofproto->subfacets) {
4998 if (subfacet->key_len == key_len
4999 && !memcmp(key, subfacet->key, key_len)) {
5007 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
5008 * 'key_fitness', 'key', and 'key_len' members in 'miss'. Returns the
5009 * existing subfacet if there is one, otherwise creates and returns a
5012 * If the returned subfacet is new, then subfacet->actions will be NULL, in
5013 * which case the caller must populate the actions with
5014 * subfacet_make_actions(). */
5015 static struct subfacet *
5016 subfacet_create(struct facet *facet, struct flow_miss *miss,
5019 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5020 enum odp_key_fitness key_fitness = miss->key_fitness;
5021 const struct nlattr *key = miss->key;
5022 size_t key_len = miss->key_len;
5024 struct subfacet *subfacet;
5026 key_hash = odp_flow_key_hash(key, key_len);
5028 if (list_is_empty(&facet->subfacets)) {
5029 subfacet = &facet->one_subfacet;
5031 subfacet = subfacet_find(ofproto, key, key_len, key_hash);
5033 if (subfacet->facet == facet) {
5037 /* This shouldn't happen. */
5038 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
5039 subfacet_destroy(subfacet);
5042 subfacet = xmalloc(sizeof *subfacet);
5045 hmap_insert(&ofproto->subfacets, &subfacet->hmap_node, key_hash);
5046 list_push_back(&facet->subfacets, &subfacet->list_node);
5047 subfacet->facet = facet;
5048 subfacet->key_fitness = key_fitness;
5049 subfacet->key = xmemdup(key, key_len);
5050 subfacet->key_len = key_len;
5051 subfacet->used = now;
5052 subfacet->dp_packet_count = 0;
5053 subfacet->dp_byte_count = 0;
5054 subfacet->actions_len = 0;
5055 subfacet->actions = NULL;
5056 subfacet->slow = (subfacet->key_fitness == ODP_FIT_TOO_LITTLE
5059 subfacet->path = SF_NOT_INSTALLED;
5060 subfacet->initial_vals = miss->initial_vals;
5061 subfacet->odp_in_port = miss->odp_in_port;
5066 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
5067 * its facet within 'ofproto', and frees it. */
5069 subfacet_destroy__(struct subfacet *subfacet)
5071 struct facet *facet = subfacet->facet;
5072 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5074 subfacet_uninstall(subfacet);
5075 hmap_remove(&ofproto->subfacets, &subfacet->hmap_node);
5076 list_remove(&subfacet->list_node);
5077 free(subfacet->key);
5078 free(subfacet->actions);
5079 if (subfacet != &facet->one_subfacet) {
5084 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
5085 * last remaining subfacet in its facet destroys the facet too. */
5087 subfacet_destroy(struct subfacet *subfacet)
5089 struct facet *facet = subfacet->facet;
5091 if (list_is_singleton(&facet->subfacets)) {
5092 /* facet_remove() needs at least one subfacet (it will remove it). */
5093 facet_remove(facet);
5095 subfacet_destroy__(subfacet);
5100 subfacet_destroy_batch(struct ofproto_dpif *ofproto,
5101 struct subfacet **subfacets, int n)
5103 struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH];
5104 struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH];
5105 struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH];
5108 for (i = 0; i < n; i++) {
5109 ops[i].type = DPIF_OP_FLOW_DEL;
5110 ops[i].u.flow_del.key = subfacets[i]->key;
5111 ops[i].u.flow_del.key_len = subfacets[i]->key_len;
5112 ops[i].u.flow_del.stats = &stats[i];
5116 dpif_operate(ofproto->backer->dpif, opsp, n);
5117 for (i = 0; i < n; i++) {
5118 subfacet_reset_dp_stats(subfacets[i], &stats[i]);
5119 subfacets[i]->path = SF_NOT_INSTALLED;
5120 subfacet_destroy(subfacets[i]);
5124 /* Composes the datapath actions for 'subfacet' based on its rule's actions.
5125 * Translates the actions into 'odp_actions', which the caller must have
5126 * initialized and is responsible for uninitializing. */
5128 subfacet_make_actions(struct subfacet *subfacet, const struct ofpbuf *packet,
5129 struct ofpbuf *odp_actions)
5131 struct facet *facet = subfacet->facet;
5132 struct rule_dpif *rule = facet->rule;
5133 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5135 struct action_xlate_ctx ctx;
5137 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
5138 &subfacet->initial_vals, rule, 0, packet);
5139 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, odp_actions);
5140 facet->tags = ctx.tags;
5141 facet->has_learn = ctx.has_learn;
5142 facet->has_normal = ctx.has_normal;
5143 facet->has_fin_timeout = ctx.has_fin_timeout;
5144 facet->nf_flow.output_iface = ctx.nf_output_iface;
5145 facet->mirrors = ctx.mirrors;
5147 subfacet->slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
5148 if (subfacet->actions_len != odp_actions->size
5149 || memcmp(subfacet->actions, odp_actions->data, odp_actions->size)) {
5150 free(subfacet->actions);
5151 subfacet->actions_len = odp_actions->size;
5152 subfacet->actions = xmemdup(odp_actions->data, odp_actions->size);
5156 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
5157 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
5158 * in the datapath will be zeroed and 'stats' will be updated with traffic new
5159 * since 'subfacet' was last updated.
5161 * Returns 0 if successful, otherwise a positive errno value. */
5163 subfacet_install(struct subfacet *subfacet,
5164 const struct nlattr *actions, size_t actions_len,
5165 struct dpif_flow_stats *stats,
5166 enum slow_path_reason slow)
5168 struct facet *facet = subfacet->facet;
5169 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5170 enum subfacet_path path = subfacet_want_path(slow);
5171 uint64_t slow_path_stub[128 / 8];
5172 enum dpif_flow_put_flags flags;
5175 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
5177 flags |= DPIF_FP_ZERO_STATS;
5180 if (path == SF_SLOW_PATH) {
5181 compose_slow_path(ofproto, &facet->flow, slow,
5182 slow_path_stub, sizeof slow_path_stub,
5183 &actions, &actions_len);
5186 ret = dpif_flow_put(ofproto->backer->dpif, flags, subfacet->key,
5187 subfacet->key_len, actions, actions_len, stats);
5190 subfacet_reset_dp_stats(subfacet, stats);
5194 subfacet->path = path;
5200 subfacet_reinstall(struct subfacet *subfacet, struct dpif_flow_stats *stats)
5202 return subfacet_install(subfacet, subfacet->actions, subfacet->actions_len,
5203 stats, subfacet->slow);
5206 /* If 'subfacet' is installed in the datapath, uninstalls it. */
5208 subfacet_uninstall(struct subfacet *subfacet)
5210 if (subfacet->path != SF_NOT_INSTALLED) {
5211 struct rule_dpif *rule = subfacet->facet->rule;
5212 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5213 struct dpif_flow_stats stats;
5216 error = dpif_flow_del(ofproto->backer->dpif, subfacet->key,
5217 subfacet->key_len, &stats);
5218 subfacet_reset_dp_stats(subfacet, &stats);
5220 subfacet_update_stats(subfacet, &stats);
5222 subfacet->path = SF_NOT_INSTALLED;
5224 ovs_assert(subfacet->dp_packet_count == 0);
5225 ovs_assert(subfacet->dp_byte_count == 0);
5229 /* Resets 'subfacet''s datapath statistics counters. This should be called
5230 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
5231 * non-null, it should contain the statistics returned by dpif when 'subfacet'
5232 * was reset in the datapath. 'stats' will be modified to include only
5233 * statistics new since 'subfacet' was last updated. */
5235 subfacet_reset_dp_stats(struct subfacet *subfacet,
5236 struct dpif_flow_stats *stats)
5239 && subfacet->dp_packet_count <= stats->n_packets
5240 && subfacet->dp_byte_count <= stats->n_bytes) {
5241 stats->n_packets -= subfacet->dp_packet_count;
5242 stats->n_bytes -= subfacet->dp_byte_count;
5245 subfacet->dp_packet_count = 0;
5246 subfacet->dp_byte_count = 0;
5249 /* Updates 'subfacet''s used time. The caller is responsible for calling
5250 * facet_push_stats() to update the flows which 'subfacet' resubmits into. */
5252 subfacet_update_time(struct subfacet *subfacet, long long int used)
5254 if (used > subfacet->used) {
5255 subfacet->used = used;
5256 facet_update_time(subfacet->facet, used);
5260 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
5262 * Because of the meaning of a subfacet's counters, it only makes sense to do
5263 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
5264 * represents a packet that was sent by hand or if it represents statistics
5265 * that have been cleared out of the datapath. */
5267 subfacet_update_stats(struct subfacet *subfacet,
5268 const struct dpif_flow_stats *stats)
5270 if (stats->n_packets || stats->used > subfacet->used) {
5271 struct facet *facet = subfacet->facet;
5273 subfacet_update_time(subfacet, stats->used);
5274 facet->packet_count += stats->n_packets;
5275 facet->byte_count += stats->n_bytes;
5276 facet->tcp_flags |= stats->tcp_flags;
5277 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
5283 static struct rule_dpif *
5284 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow)
5286 struct rule_dpif *rule;
5288 rule = rule_dpif_lookup__(ofproto, flow, 0);
5293 return rule_dpif_miss_rule(ofproto, flow);
5296 static struct rule_dpif *
5297 rule_dpif_lookup__(struct ofproto_dpif *ofproto, const struct flow *flow,
5300 struct cls_rule *cls_rule;
5301 struct classifier *cls;
5303 if (table_id >= N_TABLES) {
5307 cls = &ofproto->up.tables[table_id].cls;
5308 if (flow->nw_frag & FLOW_NW_FRAG_ANY
5309 && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
5310 /* For OFPC_NORMAL frag_handling, we must pretend that transport ports
5311 * are unavailable. */
5312 struct flow ofpc_normal_flow = *flow;
5313 ofpc_normal_flow.tp_src = htons(0);
5314 ofpc_normal_flow.tp_dst = htons(0);
5315 cls_rule = classifier_lookup(cls, &ofpc_normal_flow);
5317 cls_rule = classifier_lookup(cls, flow);
5319 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
5322 static struct rule_dpif *
5323 rule_dpif_miss_rule(struct ofproto_dpif *ofproto, const struct flow *flow)
5325 struct ofport_dpif *port;
5327 port = get_ofp_port(ofproto, flow->in_port);
5329 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, flow->in_port);
5330 return ofproto->miss_rule;
5333 if (port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN) {
5334 return ofproto->no_packet_in_rule;
5336 return ofproto->miss_rule;
5340 complete_operation(struct rule_dpif *rule)
5342 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5344 rule_invalidate(rule);
5346 struct dpif_completion *c = xmalloc(sizeof *c);
5347 c->op = rule->up.pending;
5348 list_push_back(&ofproto->completions, &c->list_node);
5350 ofoperation_complete(rule->up.pending, 0);
5354 static struct rule *
5357 struct rule_dpif *rule = xmalloc(sizeof *rule);
5362 rule_dealloc(struct rule *rule_)
5364 struct rule_dpif *rule = rule_dpif_cast(rule_);
5369 rule_construct(struct rule *rule_)
5371 struct rule_dpif *rule = rule_dpif_cast(rule_);
5372 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5373 struct rule_dpif *victim;
5376 rule->packet_count = 0;
5377 rule->byte_count = 0;
5379 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
5380 if (victim && !list_is_empty(&victim->facets)) {
5381 struct facet *facet;
5383 rule->facets = victim->facets;
5384 list_moved(&rule->facets);
5385 LIST_FOR_EACH (facet, list_node, &rule->facets) {
5386 /* XXX: We're only clearing our local counters here. It's possible
5387 * that quite a few packets are unaccounted for in the datapath
5388 * statistics. These will be accounted to the new rule instead of
5389 * cleared as required. This could be fixed by clearing out the
5390 * datapath statistics for this facet, but currently it doesn't
5392 facet_reset_counters(facet);
5396 /* Must avoid list_moved() in this case. */
5397 list_init(&rule->facets);
5400 table_id = rule->up.table_id;
5402 rule->tag = victim->tag;
5403 } else if (table_id == 0) {
5408 miniflow_expand(&rule->up.cr.match.flow, &flow);
5409 rule->tag = rule_calculate_tag(&flow, &rule->up.cr.match.mask,
5410 ofproto->tables[table_id].basis);
5413 complete_operation(rule);
5418 rule_destruct(struct rule *rule_)
5420 struct rule_dpif *rule = rule_dpif_cast(rule_);
5421 struct facet *facet, *next_facet;
5423 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
5424 facet_revalidate(facet);
5427 complete_operation(rule);
5431 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
5433 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule_->ofproto);
5434 struct rule_dpif *rule = rule_dpif_cast(rule_);
5435 struct facet *facet;
5437 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
5438 facet_push_stats(facet);
5441 /* Start from historical data for 'rule' itself that are no longer tracked
5442 * in facets. This counts, for example, facets that have expired. */
5443 *packets = rule->packet_count;
5444 *bytes = rule->byte_count;
5446 /* Add any statistics that are tracked by facets. This includes
5447 * statistical data recently updated by ofproto_update_stats() as well as
5448 * stats for packets that were executed "by hand" via dpif_execute(). */
5449 LIST_FOR_EACH (facet, list_node, &rule->facets) {
5450 *packets += facet->packet_count;
5451 *bytes += facet->byte_count;
5456 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
5457 struct ofpbuf *packet)
5459 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5460 struct initial_vals initial_vals;
5461 struct dpif_flow_stats stats;
5462 struct action_xlate_ctx ctx;
5463 uint64_t odp_actions_stub[1024 / 8];
5464 struct ofpbuf odp_actions;
5466 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
5467 rule_credit_stats(rule, &stats);
5469 initial_vals.vlan_tci = flow->vlan_tci;
5470 initial_vals.tunnel_ip_tos = flow->tunnel.ip_tos;
5471 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5472 action_xlate_ctx_init(&ctx, ofproto, flow, &initial_vals,
5473 rule, stats.tcp_flags, packet);
5474 ctx.resubmit_stats = &stats;
5475 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, &odp_actions);
5477 execute_odp_actions(ofproto, flow, odp_actions.data,
5478 odp_actions.size, packet);
5480 ofpbuf_uninit(&odp_actions);
5484 rule_execute(struct rule *rule, const struct flow *flow,
5485 struct ofpbuf *packet)
5487 rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
5488 ofpbuf_delete(packet);
5493 rule_modify_actions(struct rule *rule_)
5495 struct rule_dpif *rule = rule_dpif_cast(rule_);
5497 complete_operation(rule);
5500 /* Sends 'packet' out 'ofport'.
5501 * May modify 'packet'.
5502 * Returns 0 if successful, otherwise a positive errno value. */
5504 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
5506 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
5507 uint64_t odp_actions_stub[1024 / 8];
5508 struct ofpbuf key, odp_actions;
5509 struct odputil_keybuf keybuf;
5514 flow_extract(packet, 0, 0, NULL, OFPP_LOCAL, &flow);
5515 if (netdev_vport_is_patch(ofport->up.netdev)) {
5516 struct ofproto_dpif *peer_ofproto;
5517 struct dpif_flow_stats stats;
5518 struct ofport_dpif *peer;
5519 struct rule_dpif *rule;
5521 peer = ofport_get_peer(ofport);
5526 dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
5527 netdev_vport_inc_tx(ofport->up.netdev, &stats);
5528 netdev_vport_inc_rx(peer->up.netdev, &stats);
5530 flow.in_port = peer->up.ofp_port;
5531 peer_ofproto = ofproto_dpif_cast(peer->up.ofproto);
5532 rule = rule_dpif_lookup(peer_ofproto, &flow);
5533 rule_dpif_execute(rule, &flow, packet);
5538 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5540 if (ofport->tnl_port) {
5541 struct dpif_flow_stats stats;
5543 odp_port = tnl_port_send(ofport->tnl_port, &flow);
5544 if (odp_port == OVSP_NONE) {
5548 dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
5549 netdev_vport_inc_tx(ofport->up.netdev, &stats);
5550 odp_put_tunnel_action(&flow.tunnel, &odp_actions);
5551 odp_put_skb_mark_action(flow.skb_mark, &odp_actions);
5553 odp_port = vsp_realdev_to_vlandev(ofproto, ofport->odp_port,
5555 if (odp_port != ofport->odp_port) {
5556 eth_pop_vlan(packet);
5557 flow.vlan_tci = htons(0);
5561 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5562 odp_flow_key_from_flow(&key, &flow,
5563 ofp_port_to_odp_port(ofproto, flow.in_port));
5565 compose_sflow_action(ofproto, &odp_actions, &flow, odp_port);
5567 nl_msg_put_u32(&odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port);
5568 error = dpif_execute(ofproto->backer->dpif,
5570 odp_actions.data, odp_actions.size,
5572 ofpbuf_uninit(&odp_actions);
5575 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)",
5576 ofproto->up.name, odp_port, strerror(error));
5578 ofproto_update_local_port_stats(ofport->up.ofproto, packet->size, 0);
5582 /* OpenFlow to datapath action translation. */
5584 static bool may_receive(const struct ofport_dpif *, struct action_xlate_ctx *);
5585 static void do_xlate_actions(const struct ofpact *, size_t ofpacts_len,
5586 struct action_xlate_ctx *);
5587 static void xlate_normal(struct action_xlate_ctx *);
5589 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
5590 * The action will state 'slow' as the reason that the action is in the slow
5591 * path. (This is purely informational: it allows a human viewing "ovs-dpctl
5592 * dump-flows" output to see why a flow is in the slow path.)
5594 * The 'stub_size' bytes in 'stub' will be used to store the action.
5595 * 'stub_size' must be large enough for the action.
5597 * The action and its size will be stored in '*actionsp' and '*actions_lenp',
5600 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
5601 enum slow_path_reason slow,
5602 uint64_t *stub, size_t stub_size,
5603 const struct nlattr **actionsp, size_t *actions_lenp)
5605 union user_action_cookie cookie;
5608 cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
5609 cookie.slow_path.unused = 0;
5610 cookie.slow_path.reason = slow;
5612 ofpbuf_use_stack(&buf, stub, stub_size);
5613 if (slow & (SLOW_CFM | SLOW_LACP | SLOW_STP)) {
5614 uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif, UINT32_MAX);
5615 odp_put_userspace_action(pid, &cookie, sizeof cookie, &buf);
5617 put_userspace_action(ofproto, &buf, flow, &cookie);
5619 *actionsp = buf.data;
5620 *actions_lenp = buf.size;
5624 put_userspace_action(const struct ofproto_dpif *ofproto,
5625 struct ofpbuf *odp_actions,
5626 const struct flow *flow,
5627 const union user_action_cookie *cookie)
5631 pid = dpif_port_get_pid(ofproto->backer->dpif,
5632 ofp_port_to_odp_port(ofproto, flow->in_port));
5634 return odp_put_userspace_action(pid, cookie, sizeof *cookie, odp_actions);
5638 compose_sflow_cookie(const struct ofproto_dpif *ofproto,
5639 ovs_be16 vlan_tci, uint32_t odp_port,
5640 unsigned int n_outputs, union user_action_cookie *cookie)
5644 cookie->type = USER_ACTION_COOKIE_SFLOW;
5645 cookie->sflow.vlan_tci = vlan_tci;
5647 /* See http://www.sflow.org/sflow_version_5.txt (search for "Input/output
5648 * port information") for the interpretation of cookie->output. */
5649 switch (n_outputs) {
5651 /* 0x40000000 | 256 means "packet dropped for unknown reason". */
5652 cookie->sflow.output = 0x40000000 | 256;
5656 ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port);
5658 cookie->sflow.output = ifindex;
5663 /* 0x80000000 means "multiple output ports. */
5664 cookie->sflow.output = 0x80000000 | n_outputs;
5669 /* Compose SAMPLE action for sFlow. */
5671 compose_sflow_action(const struct ofproto_dpif *ofproto,
5672 struct ofpbuf *odp_actions,
5673 const struct flow *flow,
5676 uint32_t probability;
5677 union user_action_cookie cookie;
5678 size_t sample_offset, actions_offset;
5681 if (!ofproto->sflow || flow->in_port == OFPP_NONE) {
5685 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
5687 /* Number of packets out of UINT_MAX to sample. */
5688 probability = dpif_sflow_get_probability(ofproto->sflow);
5689 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
5691 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
5692 compose_sflow_cookie(ofproto, htons(0), odp_port,
5693 odp_port == OVSP_NONE ? 0 : 1, &cookie);
5694 cookie_offset = put_userspace_action(ofproto, odp_actions, flow, &cookie);
5696 nl_msg_end_nested(odp_actions, actions_offset);
5697 nl_msg_end_nested(odp_actions, sample_offset);
5698 return cookie_offset;
5701 /* SAMPLE action must be first action in any given list of actions.
5702 * At this point we do not have all information required to build it. So try to
5703 * build sample action as complete as possible. */
5705 add_sflow_action(struct action_xlate_ctx *ctx)
5707 ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto,
5709 &ctx->flow, OVSP_NONE);
5710 ctx->sflow_odp_port = 0;
5711 ctx->sflow_n_outputs = 0;
5714 /* Fix SAMPLE action according to data collected while composing ODP actions.
5715 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
5716 * USERSPACE action's user-cookie which is required for sflow. */
5718 fix_sflow_action(struct action_xlate_ctx *ctx)
5720 const struct flow *base = &ctx->base_flow;
5721 union user_action_cookie *cookie;
5723 if (!ctx->user_cookie_offset) {
5727 cookie = ofpbuf_at(ctx->odp_actions, ctx->user_cookie_offset,
5729 ovs_assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
5731 compose_sflow_cookie(ctx->ofproto, base->vlan_tci,
5732 ctx->sflow_odp_port, ctx->sflow_n_outputs, cookie);
5736 compose_output_action__(struct action_xlate_ctx *ctx, uint16_t ofp_port,
5739 const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
5740 ovs_be16 flow_vlan_tci = ctx->flow.vlan_tci;
5741 ovs_be64 flow_tun_id = ctx->flow.tunnel.tun_id;
5742 uint8_t flow_nw_tos = ctx->flow.nw_tos;
5743 struct priority_to_dscp *pdscp;
5744 uint32_t out_port, odp_port;
5746 /* If 'struct flow' gets additional metadata, we'll need to zero it out
5747 * before traversing a patch port. */
5748 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 20);
5751 xlate_report(ctx, "Nonexistent output port");
5753 } else if (ofport->up.pp.config & OFPUTIL_PC_NO_FWD) {
5754 xlate_report(ctx, "OFPPC_NO_FWD set, skipping output");
5756 } else if (check_stp && !stp_forward_in_state(ofport->stp_state)) {
5757 xlate_report(ctx, "STP not in forwarding state, skipping output");
5761 if (netdev_vport_is_patch(ofport->up.netdev)) {
5762 struct ofport_dpif *peer = ofport_get_peer(ofport);
5763 struct flow old_flow = ctx->flow;
5764 const struct ofproto_dpif *peer_ofproto;
5765 enum slow_path_reason special;
5766 struct ofport_dpif *in_port;
5769 xlate_report(ctx, "Nonexistent patch port peer");
5773 peer_ofproto = ofproto_dpif_cast(peer->up.ofproto);
5774 if (peer_ofproto->backer != ctx->ofproto->backer) {
5775 xlate_report(ctx, "Patch port peer on a different datapath");
5779 ctx->ofproto = ofproto_dpif_cast(peer->up.ofproto);
5780 ctx->flow.in_port = peer->up.ofp_port;
5781 ctx->flow.metadata = htonll(0);
5782 memset(&ctx->flow.tunnel, 0, sizeof ctx->flow.tunnel);
5783 memset(ctx->flow.regs, 0, sizeof ctx->flow.regs);
5785 in_port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
5786 special = process_special(ctx->ofproto, &ctx->flow, in_port,
5789 ctx->slow |= special;
5790 } else if (!in_port || may_receive(in_port, ctx)) {
5791 if (!in_port || stp_forward_in_state(in_port->stp_state)) {
5792 xlate_table_action(ctx, ctx->flow.in_port, 0, true);
5794 /* Forwarding is disabled by STP. Let OFPP_NORMAL and the
5795 * learning action look at the packet, then drop it. */
5796 struct flow old_base_flow = ctx->base_flow;
5797 size_t old_size = ctx->odp_actions->size;
5798 xlate_table_action(ctx, ctx->flow.in_port, 0, true);
5799 ctx->base_flow = old_base_flow;
5800 ctx->odp_actions->size = old_size;
5804 ctx->flow = old_flow;
5805 ctx->ofproto = ofproto_dpif_cast(ofport->up.ofproto);
5807 if (ctx->resubmit_stats) {
5808 netdev_vport_inc_tx(ofport->up.netdev, ctx->resubmit_stats);
5809 netdev_vport_inc_rx(peer->up.netdev, ctx->resubmit_stats);
5815 pdscp = get_priority(ofport, ctx->flow.skb_priority);
5817 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
5818 ctx->flow.nw_tos |= pdscp->dscp;
5821 if (ofport->tnl_port) {
5822 odp_port = tnl_port_send(ofport->tnl_port, &ctx->flow);
5823 if (odp_port == OVSP_NONE) {
5824 xlate_report(ctx, "Tunneling decided against output");
5828 if (ctx->resubmit_stats) {
5829 netdev_vport_inc_tx(ofport->up.netdev, ctx->resubmit_stats);
5831 out_port = odp_port;
5832 commit_odp_tunnel_action(&ctx->flow, &ctx->base_flow,
5835 odp_port = ofport->odp_port;
5836 out_port = vsp_realdev_to_vlandev(ctx->ofproto, odp_port,
5837 ctx->flow.vlan_tci);
5838 if (out_port != odp_port) {
5839 ctx->flow.vlan_tci = htons(0);
5841 ctx->flow.skb_mark &= ~IPSEC_MARK;
5843 commit_odp_actions(&ctx->flow, &ctx->base_flow, ctx->odp_actions);
5844 nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_OUTPUT, out_port);
5846 ctx->sflow_odp_port = odp_port;
5847 ctx->sflow_n_outputs++;
5848 ctx->nf_output_iface = ofp_port;
5849 ctx->flow.tunnel.tun_id = flow_tun_id;
5850 ctx->flow.vlan_tci = flow_vlan_tci;
5851 ctx->flow.nw_tos = flow_nw_tos;
5855 compose_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port)
5857 compose_output_action__(ctx, ofp_port, true);
5861 tag_the_flow(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
5863 struct ofproto_dpif *ofproto = ctx->ofproto;
5864 uint8_t table_id = ctx->table_id;
5866 if (table_id > 0 && table_id < N_TABLES) {
5867 struct table_dpif *table = &ofproto->tables[table_id];
5868 if (table->other_table) {
5869 ctx->tags |= (rule && rule->tag
5871 : rule_calculate_tag(&ctx->flow,
5872 &table->other_table->mask,
5878 /* Common rule processing in one place to avoid duplicating code. */
5879 static struct rule_dpif *
5880 ctx_rule_hooks(struct action_xlate_ctx *ctx, struct rule_dpif *rule,
5883 if (ctx->resubmit_hook) {
5884 ctx->resubmit_hook(ctx, rule);
5886 if (rule == NULL && may_packet_in) {
5888 * check if table configuration flags
5889 * OFPTC_TABLE_MISS_CONTROLLER, default.
5890 * OFPTC_TABLE_MISS_CONTINUE,
5891 * OFPTC_TABLE_MISS_DROP
5892 * When OF1.0, OFPTC_TABLE_MISS_CONTINUE is used. What to do?
5894 rule = rule_dpif_miss_rule(ctx->ofproto, &ctx->flow);
5896 if (rule && ctx->resubmit_stats) {
5897 rule_credit_stats(rule, ctx->resubmit_stats);
5903 xlate_table_action(struct action_xlate_ctx *ctx,
5904 uint16_t in_port, uint8_t table_id, bool may_packet_in)
5906 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
5907 struct rule_dpif *rule;
5908 uint16_t old_in_port = ctx->flow.in_port;
5909 uint8_t old_table_id = ctx->table_id;
5911 ctx->table_id = table_id;
5913 /* Look up a flow with 'in_port' as the input port. */
5914 ctx->flow.in_port = in_port;
5915 rule = rule_dpif_lookup__(ctx->ofproto, &ctx->flow, table_id);
5917 tag_the_flow(ctx, rule);
5919 /* Restore the original input port. Otherwise OFPP_NORMAL and
5920 * OFPP_IN_PORT will have surprising behavior. */
5921 ctx->flow.in_port = old_in_port;
5923 rule = ctx_rule_hooks(ctx, rule, may_packet_in);
5926 struct rule_dpif *old_rule = ctx->rule;
5930 do_xlate_actions(rule->up.ofpacts, rule->up.ofpacts_len, ctx);
5931 ctx->rule = old_rule;
5935 ctx->table_id = old_table_id;
5937 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
5939 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
5940 MAX_RESUBMIT_RECURSION);
5941 ctx->max_resubmit_trigger = true;
5946 xlate_ofpact_resubmit(struct action_xlate_ctx *ctx,
5947 const struct ofpact_resubmit *resubmit)
5952 in_port = resubmit->in_port;
5953 if (in_port == OFPP_IN_PORT) {
5954 in_port = ctx->flow.in_port;
5957 table_id = resubmit->table_id;
5958 if (table_id == 255) {
5959 table_id = ctx->table_id;
5962 xlate_table_action(ctx, in_port, table_id, false);
5966 flood_packets(struct action_xlate_ctx *ctx, bool all)
5968 struct ofport_dpif *ofport;
5970 HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) {
5971 uint16_t ofp_port = ofport->up.ofp_port;
5973 if (ofp_port == ctx->flow.in_port) {
5978 compose_output_action__(ctx, ofp_port, false);
5979 } else if (!(ofport->up.pp.config & OFPUTIL_PC_NO_FLOOD)) {
5980 compose_output_action(ctx, ofp_port);
5984 ctx->nf_output_iface = NF_OUT_FLOOD;
5988 execute_controller_action(struct action_xlate_ctx *ctx, int len,
5989 enum ofp_packet_in_reason reason,
5990 uint16_t controller_id)
5992 struct ofputil_packet_in pin;
5993 struct ofpbuf *packet;
5995 ctx->slow |= SLOW_CONTROLLER;
6000 packet = ofpbuf_clone(ctx->packet);
6002 if (packet->l2 && packet->l3) {
6003 struct eth_header *eh;
6004 uint16_t mpls_depth;
6006 eth_pop_vlan(packet);
6009 memcpy(eh->eth_src, ctx->flow.dl_src, sizeof eh->eth_src);
6010 memcpy(eh->eth_dst, ctx->flow.dl_dst, sizeof eh->eth_dst);
6012 if (ctx->flow.vlan_tci & htons(VLAN_CFI)) {
6013 eth_push_vlan(packet, ctx->flow.vlan_tci);
6016 mpls_depth = eth_mpls_depth(packet);
6018 if (mpls_depth < ctx->flow.mpls_depth) {
6019 push_mpls(packet, ctx->flow.dl_type, ctx->flow.mpls_lse);
6020 } else if (mpls_depth > ctx->flow.mpls_depth) {
6021 pop_mpls(packet, ctx->flow.dl_type);
6022 } else if (mpls_depth) {
6023 set_mpls_lse(packet, ctx->flow.mpls_lse);
6027 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
6028 packet_set_ipv4(packet, ctx->flow.nw_src, ctx->flow.nw_dst,
6029 ctx->flow.nw_tos, ctx->flow.nw_ttl);
6033 if (ctx->flow.nw_proto == IPPROTO_TCP) {
6034 packet_set_tcp_port(packet, ctx->flow.tp_src,
6036 } else if (ctx->flow.nw_proto == IPPROTO_UDP) {
6037 packet_set_udp_port(packet, ctx->flow.tp_src,
6044 pin.packet = packet->data;
6045 pin.packet_len = packet->size;
6046 pin.reason = reason;
6047 pin.controller_id = controller_id;
6048 pin.table_id = ctx->table_id;
6049 pin.cookie = ctx->rule ? ctx->rule->up.flow_cookie : 0;
6052 flow_get_metadata(&ctx->flow, &pin.fmd);
6054 connmgr_send_packet_in(ctx->ofproto->up.connmgr, &pin);
6055 ofpbuf_delete(packet);
6059 execute_mpls_push_action(struct action_xlate_ctx *ctx, ovs_be16 eth_type)
6061 ovs_assert(eth_type_mpls(eth_type));
6063 if (ctx->base_flow.mpls_depth) {
6064 ctx->flow.mpls_lse &= ~htonl(MPLS_BOS_MASK);
6065 ctx->flow.mpls_depth++;
6070 if (ctx->flow.dl_type == htons(ETH_TYPE_IPV6)) {
6071 label = htonl(0x2); /* IPV6 Explicit Null. */
6073 label = htonl(0x0); /* IPV4 Explicit Null. */
6075 tc = (ctx->flow.nw_tos & IP_DSCP_MASK) >> 2;
6076 ttl = ctx->flow.nw_ttl ? ctx->flow.nw_ttl : 0x40;
6077 ctx->flow.mpls_lse = set_mpls_lse_values(ttl, tc, 1, label);
6078 ctx->flow.mpls_depth = 1;
6080 ctx->flow.dl_type = eth_type;
6084 execute_mpls_pop_action(struct action_xlate_ctx *ctx, ovs_be16 eth_type)
6086 ovs_assert(eth_type_mpls(ctx->flow.dl_type));
6087 ovs_assert(!eth_type_mpls(eth_type));
6089 if (ctx->flow.mpls_depth) {
6090 ctx->flow.mpls_depth--;
6091 ctx->flow.mpls_lse = htonl(0);
6092 if (!ctx->flow.mpls_depth) {
6093 ctx->flow.dl_type = eth_type;
6099 compose_dec_ttl(struct action_xlate_ctx *ctx, struct ofpact_cnt_ids *ids)
6101 if (ctx->flow.dl_type != htons(ETH_TYPE_IP) &&
6102 ctx->flow.dl_type != htons(ETH_TYPE_IPV6)) {
6106 if (ctx->flow.nw_ttl > 1) {
6112 for (i = 0; i < ids->n_controllers; i++) {
6113 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL,
6117 /* Stop processing for current table. */
6123 execute_set_mpls_ttl_action(struct action_xlate_ctx *ctx, uint8_t ttl)
6125 if (!eth_type_mpls(ctx->flow.dl_type)) {
6129 set_mpls_lse_ttl(&ctx->flow.mpls_lse, ttl);
6134 execute_dec_mpls_ttl_action(struct action_xlate_ctx *ctx)
6136 uint8_t ttl = mpls_lse_to_ttl(ctx->flow.mpls_lse);
6138 if (!eth_type_mpls(ctx->flow.dl_type)) {
6144 set_mpls_lse_ttl(&ctx->flow.mpls_lse, ttl);
6147 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL, 0);
6149 /* Stop processing for current table. */
6155 xlate_output_action(struct action_xlate_ctx *ctx,
6156 uint16_t port, uint16_t max_len, bool may_packet_in)
6158 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
6160 ctx->nf_output_iface = NF_OUT_DROP;
6164 compose_output_action(ctx, ctx->flow.in_port);
6167 xlate_table_action(ctx, ctx->flow.in_port, 0, may_packet_in);
6173 flood_packets(ctx, false);
6176 flood_packets(ctx, true);
6178 case OFPP_CONTROLLER:
6179 execute_controller_action(ctx, max_len, OFPR_ACTION, 0);
6185 if (port != ctx->flow.in_port) {
6186 compose_output_action(ctx, port);
6188 xlate_report(ctx, "skipping output to input port");
6193 if (prev_nf_output_iface == NF_OUT_FLOOD) {
6194 ctx->nf_output_iface = NF_OUT_FLOOD;
6195 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
6196 ctx->nf_output_iface = prev_nf_output_iface;
6197 } else if (prev_nf_output_iface != NF_OUT_DROP &&
6198 ctx->nf_output_iface != NF_OUT_FLOOD) {
6199 ctx->nf_output_iface = NF_OUT_MULTI;
6204 xlate_output_reg_action(struct action_xlate_ctx *ctx,
6205 const struct ofpact_output_reg *or)
6207 uint64_t port = mf_get_subfield(&or->src, &ctx->flow);
6208 if (port <= UINT16_MAX) {
6209 xlate_output_action(ctx, port, or->max_len, false);
6214 xlate_enqueue_action(struct action_xlate_ctx *ctx,
6215 const struct ofpact_enqueue *enqueue)
6217 uint16_t ofp_port = enqueue->port;
6218 uint32_t queue_id = enqueue->queue;
6219 uint32_t flow_priority, priority;
6222 /* Translate queue to priority. */
6223 error = dpif_queue_to_priority(ctx->ofproto->backer->dpif,
6224 queue_id, &priority);
6226 /* Fall back to ordinary output action. */
6227 xlate_output_action(ctx, enqueue->port, 0, false);
6231 /* Check output port. */
6232 if (ofp_port == OFPP_IN_PORT) {
6233 ofp_port = ctx->flow.in_port;
6234 } else if (ofp_port == ctx->flow.in_port) {
6238 /* Add datapath actions. */
6239 flow_priority = ctx->flow.skb_priority;
6240 ctx->flow.skb_priority = priority;
6241 compose_output_action(ctx, ofp_port);
6242 ctx->flow.skb_priority = flow_priority;
6244 /* Update NetFlow output port. */
6245 if (ctx->nf_output_iface == NF_OUT_DROP) {
6246 ctx->nf_output_iface = ofp_port;
6247 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
6248 ctx->nf_output_iface = NF_OUT_MULTI;
6253 xlate_set_queue_action(struct action_xlate_ctx *ctx, uint32_t queue_id)
6255 uint32_t skb_priority;
6257 if (!dpif_queue_to_priority(ctx->ofproto->backer->dpif,
6258 queue_id, &skb_priority)) {
6259 ctx->flow.skb_priority = skb_priority;
6261 /* Couldn't translate queue to a priority. Nothing to do. A warning
6262 * has already been logged. */
6266 struct xlate_reg_state {
6272 slave_enabled_cb(uint16_t ofp_port, void *ofproto_)
6274 struct ofproto_dpif *ofproto = ofproto_;
6275 struct ofport_dpif *port;
6285 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
6288 port = get_ofp_port(ofproto, ofp_port);
6289 return port ? port->may_enable : false;
6294 xlate_bundle_action(struct action_xlate_ctx *ctx,
6295 const struct ofpact_bundle *bundle)
6299 port = bundle_execute(bundle, &ctx->flow, slave_enabled_cb, ctx->ofproto);
6300 if (bundle->dst.field) {
6301 nxm_reg_load(&bundle->dst, port, &ctx->flow);
6303 xlate_output_action(ctx, port, 0, false);
6308 xlate_learn_action(struct action_xlate_ctx *ctx,
6309 const struct ofpact_learn *learn)
6311 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
6312 struct ofputil_flow_mod fm;
6313 uint64_t ofpacts_stub[1024 / 8];
6314 struct ofpbuf ofpacts;
6317 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
6318 learn_execute(learn, &ctx->flow, &fm, &ofpacts);
6320 error = ofproto_flow_mod(&ctx->ofproto->up, &fm);
6321 if (error && !VLOG_DROP_WARN(&rl)) {
6322 VLOG_WARN("learning action failed to modify flow table (%s)",
6323 ofperr_get_name(error));
6326 ofpbuf_uninit(&ofpacts);
6329 /* Reduces '*timeout' to no more than 'max'. A value of zero in either case
6330 * means "infinite". */
6332 reduce_timeout(uint16_t max, uint16_t *timeout)
6334 if (max && (!*timeout || *timeout > max)) {
6340 xlate_fin_timeout(struct action_xlate_ctx *ctx,
6341 const struct ofpact_fin_timeout *oft)
6343 if (ctx->tcp_flags & (TCP_FIN | TCP_RST) && ctx->rule) {
6344 struct rule_dpif *rule = ctx->rule;
6346 reduce_timeout(oft->fin_idle_timeout, &rule->up.idle_timeout);
6347 reduce_timeout(oft->fin_hard_timeout, &rule->up.hard_timeout);
6352 may_receive(const struct ofport_dpif *port, struct action_xlate_ctx *ctx)
6354 if (port->up.pp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
6355 ? OFPUTIL_PC_NO_RECV_STP
6356 : OFPUTIL_PC_NO_RECV)) {
6360 /* Only drop packets here if both forwarding and learning are
6361 * disabled. If just learning is enabled, we need to have
6362 * OFPP_NORMAL and the learning action have a look at the packet
6363 * before we can drop it. */
6364 if (!stp_forward_in_state(port->stp_state)
6365 && !stp_learn_in_state(port->stp_state)) {
6373 tunnel_ecn_ok(struct action_xlate_ctx *ctx)
6375 if (is_ip_any(&ctx->base_flow)
6376 && (ctx->base_flow.tunnel.ip_tos & IP_ECN_MASK) == IP_ECN_CE) {
6377 if ((ctx->base_flow.nw_tos & IP_ECN_MASK) == IP_ECN_NOT_ECT) {
6378 VLOG_WARN_RL(&rl, "dropping tunnel packet marked ECN CE"
6379 " but is not ECN capable");
6382 /* Set the ECN CE value in the tunneled packet. */
6383 ctx->flow.nw_tos |= IP_ECN_CE;
6391 do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len,
6392 struct action_xlate_ctx *ctx)
6394 bool was_evictable = true;
6395 const struct ofpact *a;
6398 /* Don't let the rule we're working on get evicted underneath us. */
6399 was_evictable = ctx->rule->up.evictable;
6400 ctx->rule->up.evictable = false;
6403 do_xlate_actions_again:
6404 OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) {
6405 struct ofpact_controller *controller;
6406 const struct ofpact_metadata *metadata;
6414 xlate_output_action(ctx, ofpact_get_OUTPUT(a)->port,
6415 ofpact_get_OUTPUT(a)->max_len, true);
6418 case OFPACT_CONTROLLER:
6419 controller = ofpact_get_CONTROLLER(a);
6420 execute_controller_action(ctx, controller->max_len,
6422 controller->controller_id);
6425 case OFPACT_ENQUEUE:
6426 xlate_enqueue_action(ctx, ofpact_get_ENQUEUE(a));
6429 case OFPACT_SET_VLAN_VID:
6430 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
6431 ctx->flow.vlan_tci |= (htons(ofpact_get_SET_VLAN_VID(a)->vlan_vid)
6435 case OFPACT_SET_VLAN_PCP:
6436 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
6437 ctx->flow.vlan_tci |= htons((ofpact_get_SET_VLAN_PCP(a)->vlan_pcp
6442 case OFPACT_STRIP_VLAN:
6443 ctx->flow.vlan_tci = htons(0);
6446 case OFPACT_PUSH_VLAN:
6447 /* XXX 802.1AD(QinQ) */
6448 ctx->flow.vlan_tci = htons(VLAN_CFI);
6451 case OFPACT_SET_ETH_SRC:
6452 memcpy(ctx->flow.dl_src, ofpact_get_SET_ETH_SRC(a)->mac,
6456 case OFPACT_SET_ETH_DST:
6457 memcpy(ctx->flow.dl_dst, ofpact_get_SET_ETH_DST(a)->mac,
6461 case OFPACT_SET_IPV4_SRC:
6462 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
6463 ctx->flow.nw_src = ofpact_get_SET_IPV4_SRC(a)->ipv4;
6467 case OFPACT_SET_IPV4_DST:
6468 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
6469 ctx->flow.nw_dst = ofpact_get_SET_IPV4_DST(a)->ipv4;
6473 case OFPACT_SET_IPV4_DSCP:
6474 /* OpenFlow 1.0 only supports IPv4. */
6475 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
6476 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
6477 ctx->flow.nw_tos |= ofpact_get_SET_IPV4_DSCP(a)->dscp;
6481 case OFPACT_SET_L4_SRC_PORT:
6482 if (is_ip_any(&ctx->flow)) {
6483 ctx->flow.tp_src = htons(ofpact_get_SET_L4_SRC_PORT(a)->port);
6487 case OFPACT_SET_L4_DST_PORT:
6488 if (is_ip_any(&ctx->flow)) {
6489 ctx->flow.tp_dst = htons(ofpact_get_SET_L4_DST_PORT(a)->port);
6493 case OFPACT_RESUBMIT:
6494 xlate_ofpact_resubmit(ctx, ofpact_get_RESUBMIT(a));
6497 case OFPACT_SET_TUNNEL:
6498 ctx->flow.tunnel.tun_id = htonll(ofpact_get_SET_TUNNEL(a)->tun_id);
6501 case OFPACT_SET_QUEUE:
6502 xlate_set_queue_action(ctx, ofpact_get_SET_QUEUE(a)->queue_id);
6505 case OFPACT_POP_QUEUE:
6506 ctx->flow.skb_priority = ctx->orig_skb_priority;
6509 case OFPACT_REG_MOVE:
6510 nxm_execute_reg_move(ofpact_get_REG_MOVE(a), &ctx->flow);
6513 case OFPACT_REG_LOAD:
6514 nxm_execute_reg_load(ofpact_get_REG_LOAD(a), &ctx->flow);
6517 case OFPACT_STACK_PUSH:
6518 nxm_execute_stack_push(ofpact_get_STACK_PUSH(a), &ctx->flow,
6522 case OFPACT_STACK_POP:
6523 nxm_execute_stack_pop(ofpact_get_STACK_POP(a), &ctx->flow,
6527 case OFPACT_PUSH_MPLS:
6528 execute_mpls_push_action(ctx, ofpact_get_PUSH_MPLS(a)->ethertype);
6531 case OFPACT_POP_MPLS:
6532 execute_mpls_pop_action(ctx, ofpact_get_POP_MPLS(a)->ethertype);
6535 case OFPACT_SET_MPLS_TTL:
6536 if (execute_set_mpls_ttl_action(ctx, ofpact_get_SET_MPLS_TTL(a)->ttl)) {
6541 case OFPACT_DEC_MPLS_TTL:
6542 if (execute_dec_mpls_ttl_action(ctx)) {
6547 case OFPACT_DEC_TTL:
6548 if (compose_dec_ttl(ctx, ofpact_get_DEC_TTL(a))) {
6554 /* Nothing to do. */
6557 case OFPACT_MULTIPATH:
6558 multipath_execute(ofpact_get_MULTIPATH(a), &ctx->flow);
6562 ctx->ofproto->has_bundle_action = true;
6563 xlate_bundle_action(ctx, ofpact_get_BUNDLE(a));
6566 case OFPACT_OUTPUT_REG:
6567 xlate_output_reg_action(ctx, ofpact_get_OUTPUT_REG(a));
6571 ctx->has_learn = true;
6572 if (ctx->may_learn) {
6573 xlate_learn_action(ctx, ofpact_get_LEARN(a));
6581 case OFPACT_FIN_TIMEOUT:
6582 ctx->has_fin_timeout = true;
6583 xlate_fin_timeout(ctx, ofpact_get_FIN_TIMEOUT(a));
6586 case OFPACT_CLEAR_ACTIONS:
6588 * Nothing to do because writa-actions is not supported for now.
6589 * When writa-actions is supported, clear-actions also must
6590 * be supported at the same time.
6594 case OFPACT_WRITE_METADATA:
6595 metadata = ofpact_get_WRITE_METADATA(a);
6596 ctx->flow.metadata &= ~metadata->mask;
6597 ctx->flow.metadata |= metadata->metadata & metadata->mask;
6600 case OFPACT_GOTO_TABLE: {
6601 /* It is assumed that goto-table is the last action. */
6602 struct ofpact_goto_table *ogt = ofpact_get_GOTO_TABLE(a);
6603 struct rule_dpif *rule;
6605 ovs_assert(ctx->table_id < ogt->table_id);
6607 ctx->table_id = ogt->table_id;
6609 /* Look up a flow from the new table. */
6610 rule = rule_dpif_lookup__(ctx->ofproto, &ctx->flow, ctx->table_id);
6612 tag_the_flow(ctx, rule);
6614 rule = ctx_rule_hooks(ctx, rule, true);
6618 ctx->rule->up.evictable = was_evictable;
6621 was_evictable = rule->up.evictable;
6622 rule->up.evictable = false;
6624 /* Tail recursion removal. */
6625 ofpacts = rule->up.ofpacts;
6626 ofpacts_len = rule->up.ofpacts_len;
6627 goto do_xlate_actions_again;
6636 ctx->rule->up.evictable = was_evictable;
6641 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
6642 struct ofproto_dpif *ofproto, const struct flow *flow,
6643 const struct initial_vals *initial_vals,
6644 struct rule_dpif *rule,
6645 uint8_t tcp_flags, const struct ofpbuf *packet)
6647 ovs_be64 initial_tun_id = flow->tunnel.tun_id;
6649 /* Flow initialization rules:
6650 * - 'base_flow' must match the kernel's view of the packet at the
6651 * time that action processing starts. 'flow' represents any
6652 * transformations we wish to make through actions.
6653 * - By default 'base_flow' and 'flow' are the same since the input
6654 * packet matches the output before any actions are applied.
6655 * - When using VLAN splinters, 'base_flow''s VLAN is set to the value
6656 * of the received packet as seen by the kernel. If we later output
6657 * to another device without any modifications this will cause us to
6658 * insert a new tag since the original one was stripped off by the
6660 * - Tunnel 'flow' is largely cleared when transitioning between
6661 * the input and output stages since it does not make sense to output
6662 * a packet with the exact headers that it was received with (i.e.
6663 * the destination IP is us). The one exception is the tun_id, which
6664 * is preserved to allow use in later resubmit lookups and loads into
6666 * - Tunnel 'base_flow' is completely cleared since that is what the
6667 * kernel does. If we wish to maintain the original values an action
6668 * needs to be generated. */
6670 ctx->ofproto = ofproto;
6672 memset(&ctx->flow.tunnel, 0, sizeof ctx->flow.tunnel);
6673 ctx->base_flow = ctx->flow;
6674 ctx->base_flow.vlan_tci = initial_vals->vlan_tci;
6675 ctx->base_flow.tunnel.ip_tos = initial_vals->tunnel_ip_tos;
6676 ctx->flow.tunnel.tun_id = initial_tun_id;
6678 ctx->packet = packet;
6679 ctx->may_learn = packet != NULL;
6680 ctx->tcp_flags = tcp_flags;
6681 ctx->resubmit_hook = NULL;
6682 ctx->report_hook = NULL;
6683 ctx->resubmit_stats = NULL;
6686 /* Translates the 'ofpacts_len' bytes of "struct ofpacts" starting at 'ofpacts'
6687 * into datapath actions in 'odp_actions', using 'ctx'. */
6689 xlate_actions(struct action_xlate_ctx *ctx,
6690 const struct ofpact *ofpacts, size_t ofpacts_len,
6691 struct ofpbuf *odp_actions)
6693 /* Normally false. Set to true if we ever hit MAX_RESUBMIT_RECURSION, so
6694 * that in the future we always keep a copy of the original flow for
6695 * tracing purposes. */
6696 static bool hit_resubmit_limit;
6698 enum slow_path_reason special;
6699 struct ofport_dpif *in_port;
6700 struct flow orig_flow;
6702 COVERAGE_INC(ofproto_dpif_xlate);
6704 ofpbuf_clear(odp_actions);
6705 ofpbuf_reserve(odp_actions, NL_A_U32_SIZE);
6707 ctx->odp_actions = odp_actions;
6710 ctx->has_learn = false;
6711 ctx->has_normal = false;
6712 ctx->has_fin_timeout = false;
6713 ctx->nf_output_iface = NF_OUT_DROP;
6716 ctx->max_resubmit_trigger = false;
6717 ctx->orig_skb_priority = ctx->flow.skb_priority;
6721 ofpbuf_use_stub(&ctx->stack, ctx->init_stack, sizeof ctx->init_stack);
6723 if (ctx->ofproto->has_mirrors || hit_resubmit_limit) {
6724 /* Do this conditionally because the copy is expensive enough that it
6725 * shows up in profiles. */
6726 orig_flow = ctx->flow;
6729 if (ctx->flow.nw_frag & FLOW_NW_FRAG_ANY) {
6730 switch (ctx->ofproto->up.frag_handling) {
6731 case OFPC_FRAG_NORMAL:
6732 /* We must pretend that transport ports are unavailable. */
6733 ctx->flow.tp_src = ctx->base_flow.tp_src = htons(0);
6734 ctx->flow.tp_dst = ctx->base_flow.tp_dst = htons(0);
6737 case OFPC_FRAG_DROP:
6740 case OFPC_FRAG_REASM:
6743 case OFPC_FRAG_NX_MATCH:
6744 /* Nothing to do. */
6747 case OFPC_INVALID_TTL_TO_CONTROLLER:
6752 in_port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
6753 special = process_special(ctx->ofproto, &ctx->flow, in_port, ctx->packet);
6755 ctx->slow |= special;
6757 static struct vlog_rate_limit trace_rl = VLOG_RATE_LIMIT_INIT(1, 1);
6758 struct initial_vals initial_vals;
6759 uint32_t local_odp_port;
6761 initial_vals.vlan_tci = ctx->base_flow.vlan_tci;
6762 initial_vals.tunnel_ip_tos = ctx->base_flow.tunnel.ip_tos;
6764 add_sflow_action(ctx);
6766 if (tunnel_ecn_ok(ctx) && (!in_port || may_receive(in_port, ctx))) {
6767 do_xlate_actions(ofpacts, ofpacts_len, ctx);
6769 /* We've let OFPP_NORMAL and the learning action look at the
6770 * packet, so drop it now if forwarding is disabled. */
6771 if (in_port && !stp_forward_in_state(in_port->stp_state)) {
6772 ofpbuf_clear(ctx->odp_actions);
6773 add_sflow_action(ctx);
6777 if (ctx->max_resubmit_trigger && !ctx->resubmit_hook) {
6778 if (!hit_resubmit_limit) {
6779 /* We didn't record the original flow. Make sure we do from
6781 hit_resubmit_limit = true;
6782 } else if (!VLOG_DROP_ERR(&trace_rl)) {
6783 struct ds ds = DS_EMPTY_INITIALIZER;
6785 ofproto_trace(ctx->ofproto, &orig_flow, ctx->packet,
6786 &initial_vals, &ds);
6787 VLOG_ERR("Trace triggered by excessive resubmit "
6788 "recursion:\n%s", ds_cstr(&ds));
6793 local_odp_port = ofp_port_to_odp_port(ctx->ofproto, OFPP_LOCAL);
6794 if (!connmgr_may_set_up_flow(ctx->ofproto->up.connmgr, &ctx->flow,
6796 ctx->odp_actions->data,
6797 ctx->odp_actions->size)) {
6798 ctx->slow |= SLOW_IN_BAND;
6800 && connmgr_msg_in_hook(ctx->ofproto->up.connmgr, &ctx->flow,
6802 compose_output_action(ctx, OFPP_LOCAL);
6805 if (ctx->ofproto->has_mirrors) {
6806 add_mirror_actions(ctx, &orig_flow);
6808 fix_sflow_action(ctx);
6811 ofpbuf_uninit(&ctx->stack);
6814 /* Translates the 'ofpacts_len' bytes of "struct ofpact"s starting at 'ofpacts'
6815 * into datapath actions, using 'ctx', and discards the datapath actions. */
6817 xlate_actions_for_side_effects(struct action_xlate_ctx *ctx,
6818 const struct ofpact *ofpacts,
6821 uint64_t odp_actions_stub[1024 / 8];
6822 struct ofpbuf odp_actions;
6824 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
6825 xlate_actions(ctx, ofpacts, ofpacts_len, &odp_actions);
6826 ofpbuf_uninit(&odp_actions);
6830 xlate_report(struct action_xlate_ctx *ctx, const char *s)
6832 if (ctx->report_hook) {
6833 ctx->report_hook(ctx, s);
6837 /* OFPP_NORMAL implementation. */
6839 static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *);
6841 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
6842 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle',
6843 * the bundle on which the packet was received, returns the VLAN to which the
6846 * Both 'vid' and the return value are in the range 0...4095. */
6848 input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid)
6850 switch (in_bundle->vlan_mode) {
6851 case PORT_VLAN_ACCESS:
6852 return in_bundle->vlan;
6855 case PORT_VLAN_TRUNK:
6858 case PORT_VLAN_NATIVE_UNTAGGED:
6859 case PORT_VLAN_NATIVE_TAGGED:
6860 return vid ? vid : in_bundle->vlan;
6867 /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'.
6868 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
6871 * 'vid' should be the VID obtained from the 802.1Q header that was received as
6872 * part of a packet (specify 0 if there was no 802.1Q header), in the range
6875 input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn)
6877 /* Allow any VID on the OFPP_NONE port. */
6878 if (in_bundle == &ofpp_none_bundle) {
6882 switch (in_bundle->vlan_mode) {
6883 case PORT_VLAN_ACCESS:
6886 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6887 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
6888 "packet received on port %s configured as VLAN "
6889 "%"PRIu16" access port",
6890 in_bundle->ofproto->up.name, vid,
6891 in_bundle->name, in_bundle->vlan);
6897 case PORT_VLAN_NATIVE_UNTAGGED:
6898 case PORT_VLAN_NATIVE_TAGGED:
6900 /* Port must always carry its native VLAN. */
6904 case PORT_VLAN_TRUNK:
6905 if (!ofbundle_includes_vlan(in_bundle, vid)) {
6907 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6908 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet "
6909 "received on port %s not configured for trunking "
6911 in_bundle->ofproto->up.name, vid,
6912 in_bundle->name, vid);
6924 /* Given 'vlan', the VLAN that a packet belongs to, and
6925 * 'out_bundle', a bundle on which the packet is to be output, returns the VID
6926 * that should be included in the 802.1Q header. (If the return value is 0,
6927 * then the 802.1Q header should only be included in the packet if there is a
6930 * Both 'vlan' and the return value are in the range 0...4095. */
6932 output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan)
6934 switch (out_bundle->vlan_mode) {
6935 case PORT_VLAN_ACCESS:
6938 case PORT_VLAN_TRUNK:
6939 case PORT_VLAN_NATIVE_TAGGED:
6942 case PORT_VLAN_NATIVE_UNTAGGED:
6943 return vlan == out_bundle->vlan ? 0 : vlan;
6951 output_normal(struct action_xlate_ctx *ctx, const struct ofbundle *out_bundle,
6954 struct ofport_dpif *port;
6956 ovs_be16 tci, old_tci;
6958 vid = output_vlan_to_vid(out_bundle, vlan);
6959 if (!out_bundle->bond) {
6960 port = ofbundle_get_a_port(out_bundle);
6962 port = bond_choose_output_slave(out_bundle->bond, &ctx->flow,
6965 /* No slaves enabled, so drop packet. */
6970 old_tci = ctx->flow.vlan_tci;
6972 if (tci || out_bundle->use_priority_tags) {
6973 tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK);
6975 tci |= htons(VLAN_CFI);
6978 ctx->flow.vlan_tci = tci;
6980 compose_output_action(ctx, port->up.ofp_port);
6981 ctx->flow.vlan_tci = old_tci;
6985 mirror_mask_ffs(mirror_mask_t mask)
6987 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
6992 ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
6994 return (bundle->vlan_mode != PORT_VLAN_ACCESS
6995 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
6999 ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan)
7001 return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan);
7004 /* Returns an arbitrary interface within 'bundle'. */
7005 static struct ofport_dpif *
7006 ofbundle_get_a_port(const struct ofbundle *bundle)
7008 return CONTAINER_OF(list_front(&bundle->ports),
7009 struct ofport_dpif, bundle_node);
7013 vlan_is_mirrored(const struct ofmirror *m, int vlan)
7015 return !m->vlans || bitmap_is_set(m->vlans, vlan);
7019 add_mirror_actions(struct action_xlate_ctx *ctx, const struct flow *orig_flow)
7021 struct ofproto_dpif *ofproto = ctx->ofproto;
7022 mirror_mask_t mirrors;
7023 struct ofbundle *in_bundle;
7026 const struct nlattr *a;
7029 in_bundle = lookup_input_bundle(ctx->ofproto, orig_flow->in_port,
7030 ctx->packet != NULL, NULL);
7034 mirrors = in_bundle->src_mirrors;
7036 /* Drop frames on bundles reserved for mirroring. */
7037 if (in_bundle->mirror_out) {
7038 if (ctx->packet != NULL) {
7039 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7040 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
7041 "%s, which is reserved exclusively for mirroring",
7042 ctx->ofproto->up.name, in_bundle->name);
7048 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
7049 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
7052 vlan = input_vid_to_vlan(in_bundle, vid);
7054 /* Look at the output ports to check for destination selections. */
7056 NL_ATTR_FOR_EACH (a, left, ctx->odp_actions->data,
7057 ctx->odp_actions->size) {
7058 enum ovs_action_attr type = nl_attr_type(a);
7059 struct ofport_dpif *ofport;
7061 if (type != OVS_ACTION_ATTR_OUTPUT) {
7065 ofport = get_odp_port(ofproto, nl_attr_get_u32(a));
7066 if (ofport && ofport->bundle) {
7067 mirrors |= ofport->bundle->dst_mirrors;
7075 /* Restore the original packet before adding the mirror actions. */
7076 ctx->flow = *orig_flow;
7081 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
7083 if (!vlan_is_mirrored(m, vlan)) {
7084 mirrors = zero_rightmost_1bit(mirrors);
7088 mirrors &= ~m->dup_mirrors;
7089 ctx->mirrors |= m->dup_mirrors;
7091 output_normal(ctx, m->out, vlan);
7092 } else if (vlan != m->out_vlan
7093 && !eth_addr_is_reserved(orig_flow->dl_dst)) {
7094 struct ofbundle *bundle;
7096 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
7097 if (ofbundle_includes_vlan(bundle, m->out_vlan)
7098 && !bundle->mirror_out) {
7099 output_normal(ctx, bundle, m->out_vlan);
7107 update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors,
7108 uint64_t packets, uint64_t bytes)
7114 for (; mirrors; mirrors = zero_rightmost_1bit(mirrors)) {
7117 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
7120 /* In normal circumstances 'm' will not be NULL. However,
7121 * if mirrors are reconfigured, we can temporarily get out
7122 * of sync in facet_revalidate(). We could "correct" the
7123 * mirror list before reaching here, but doing that would
7124 * not properly account the traffic stats we've currently
7125 * accumulated for previous mirror configuration. */
7129 m->packet_count += packets;
7130 m->byte_count += bytes;
7134 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
7135 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
7136 * indicate this; newer upstream kernels use gratuitous ARP requests. */
7138 is_gratuitous_arp(const struct flow *flow)
7140 return (flow->dl_type == htons(ETH_TYPE_ARP)
7141 && eth_addr_is_broadcast(flow->dl_dst)
7142 && (flow->nw_proto == ARP_OP_REPLY
7143 || (flow->nw_proto == ARP_OP_REQUEST
7144 && flow->nw_src == flow->nw_dst)));
7148 update_learning_table(struct ofproto_dpif *ofproto,
7149 const struct flow *flow, int vlan,
7150 struct ofbundle *in_bundle)
7152 struct mac_entry *mac;
7154 /* Don't learn the OFPP_NONE port. */
7155 if (in_bundle == &ofpp_none_bundle) {
7159 if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
7163 mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan);
7164 if (is_gratuitous_arp(flow)) {
7165 /* We don't want to learn from gratuitous ARP packets that are
7166 * reflected back over bond slaves so we lock the learning table. */
7167 if (!in_bundle->bond) {
7168 mac_entry_set_grat_arp_lock(mac);
7169 } else if (mac_entry_is_grat_arp_locked(mac)) {
7174 if (mac_entry_is_new(mac) || mac->port.p != in_bundle) {
7175 /* The log messages here could actually be useful in debugging,
7176 * so keep the rate limit relatively high. */
7177 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
7178 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
7179 "on port %s in VLAN %d",
7180 ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src),
7181 in_bundle->name, vlan);
7183 mac->port.p = in_bundle;
7184 tag_set_add(&ofproto->backer->revalidate_set,
7185 mac_learning_changed(ofproto->ml, mac));
7189 static struct ofbundle *
7190 lookup_input_bundle(const struct ofproto_dpif *ofproto, uint16_t in_port,
7191 bool warn, struct ofport_dpif **in_ofportp)
7193 struct ofport_dpif *ofport;
7195 /* Find the port and bundle for the received packet. */
7196 ofport = get_ofp_port(ofproto, in_port);
7198 *in_ofportp = ofport;
7200 if (ofport && ofport->bundle) {
7201 return ofport->bundle;
7204 /* Special-case OFPP_NONE, which a controller may use as the ingress
7205 * port for traffic that it is sourcing. */
7206 if (in_port == OFPP_NONE) {
7207 return &ofpp_none_bundle;
7210 /* Odd. A few possible reasons here:
7212 * - We deleted a port but there are still a few packets queued up
7215 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
7216 * we don't know about.
7218 * - The ofproto client didn't configure the port as part of a bundle.
7219 * This is particularly likely to happen if a packet was received on the
7220 * port after it was created, but before the client had a chance to
7221 * configure its bundle.
7224 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7226 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
7227 "port %"PRIu16, ofproto->up.name, in_port);
7232 /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or
7233 * dropped. Returns true if they may be forwarded, false if they should be
7236 * 'in_port' must be the ofport_dpif that corresponds to flow->in_port.
7237 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
7239 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
7240 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
7241 * checked by input_vid_is_valid().
7243 * May also add tags to '*tags', although the current implementation only does
7244 * so in one special case.
7247 is_admissible(struct action_xlate_ctx *ctx, struct ofport_dpif *in_port,
7250 struct ofproto_dpif *ofproto = ctx->ofproto;
7251 struct flow *flow = &ctx->flow;
7252 struct ofbundle *in_bundle = in_port->bundle;
7254 /* Drop frames for reserved multicast addresses
7255 * only if forward_bpdu option is absent. */
7256 if (!ofproto->up.forward_bpdu && eth_addr_is_reserved(flow->dl_dst)) {
7257 xlate_report(ctx, "packet has reserved destination MAC, dropping");
7261 if (in_bundle->bond) {
7262 struct mac_entry *mac;
7264 switch (bond_check_admissibility(in_bundle->bond, in_port,
7265 flow->dl_dst, &ctx->tags)) {
7270 xlate_report(ctx, "bonding refused admissibility, dropping");
7273 case BV_DROP_IF_MOVED:
7274 mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL);
7275 if (mac && mac->port.p != in_bundle &&
7276 (!is_gratuitous_arp(flow)
7277 || mac_entry_is_grat_arp_locked(mac))) {
7278 xlate_report(ctx, "SLB bond thinks this packet looped back, "
7290 xlate_normal(struct action_xlate_ctx *ctx)
7292 struct ofport_dpif *in_port;
7293 struct ofbundle *in_bundle;
7294 struct mac_entry *mac;
7298 ctx->has_normal = true;
7300 in_bundle = lookup_input_bundle(ctx->ofproto, ctx->flow.in_port,
7301 ctx->packet != NULL, &in_port);
7303 xlate_report(ctx, "no input bundle, dropping");
7307 /* Drop malformed frames. */
7308 if (ctx->flow.dl_type == htons(ETH_TYPE_VLAN) &&
7309 !(ctx->flow.vlan_tci & htons(VLAN_CFI))) {
7310 if (ctx->packet != NULL) {
7311 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7312 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
7313 "VLAN tag received on port %s",
7314 ctx->ofproto->up.name, in_bundle->name);
7316 xlate_report(ctx, "partial VLAN tag, dropping");
7320 /* Drop frames on bundles reserved for mirroring. */
7321 if (in_bundle->mirror_out) {
7322 if (ctx->packet != NULL) {
7323 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7324 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
7325 "%s, which is reserved exclusively for mirroring",
7326 ctx->ofproto->up.name, in_bundle->name);
7328 xlate_report(ctx, "input port is mirror output port, dropping");
7333 vid = vlan_tci_to_vid(ctx->flow.vlan_tci);
7334 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
7335 xlate_report(ctx, "disallowed VLAN VID for this input port, dropping");
7338 vlan = input_vid_to_vlan(in_bundle, vid);
7340 /* Check other admissibility requirements. */
7341 if (in_port && !is_admissible(ctx, in_port, vlan)) {
7345 /* Learn source MAC. */
7346 if (ctx->may_learn) {
7347 update_learning_table(ctx->ofproto, &ctx->flow, vlan, in_bundle);
7350 /* Determine output bundle. */
7351 mac = mac_learning_lookup(ctx->ofproto->ml, ctx->flow.dl_dst, vlan,
7354 if (mac->port.p != in_bundle) {
7355 xlate_report(ctx, "forwarding to learned port");
7356 output_normal(ctx, mac->port.p, vlan);
7358 xlate_report(ctx, "learned port is input port, dropping");
7361 struct ofbundle *bundle;
7363 xlate_report(ctx, "no learned MAC for destination, flooding");
7364 HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
7365 if (bundle != in_bundle
7366 && ofbundle_includes_vlan(bundle, vlan)
7367 && bundle->floodable
7368 && !bundle->mirror_out) {
7369 output_normal(ctx, bundle, vlan);
7372 ctx->nf_output_iface = NF_OUT_FLOOD;
7376 /* Optimized flow revalidation.
7378 * It's a difficult problem, in general, to tell which facets need to have
7379 * their actions recalculated whenever the OpenFlow flow table changes. We
7380 * don't try to solve that general problem: for most kinds of OpenFlow flow
7381 * table changes, we recalculate the actions for every facet. This is
7382 * relatively expensive, but it's good enough if the OpenFlow flow table
7383 * doesn't change very often.
7385 * However, we can expect one particular kind of OpenFlow flow table change to
7386 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
7387 * of CPU on revalidating every facet whenever MAC learning modifies the flow
7388 * table, we add a special case that applies to flow tables in which every rule
7389 * has the same form (that is, the same wildcards), except that the table is
7390 * also allowed to have a single "catch-all" flow that matches all packets. We
7391 * optimize this case by tagging all of the facets that resubmit into the table
7392 * and invalidating the same tag whenever a flow changes in that table. The
7393 * end result is that we revalidate just the facets that need it (and sometimes
7394 * a few more, but not all of the facets or even all of the facets that
7395 * resubmit to the table modified by MAC learning). */
7397 /* Calculates the tag to use for 'flow' and mask 'mask' when it is inserted
7398 * into an OpenFlow table with the given 'basis'. */
7400 rule_calculate_tag(const struct flow *flow, const struct minimask *mask,
7403 if (minimask_is_catchall(mask)) {
7406 uint32_t hash = flow_hash_in_minimask(flow, mask, secret);
7407 return tag_create_deterministic(hash);
7411 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
7412 * taggability of that table.
7414 * This function must be called after *each* change to a flow table. If you
7415 * skip calling it on some changes then the pointer comparisons at the end can
7416 * be invalid if you get unlucky. For example, if a flow removal causes a
7417 * cls_table to be destroyed and then a flow insertion causes a cls_table with
7418 * different wildcards to be created with the same address, then this function
7419 * will incorrectly skip revalidation. */
7421 table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
7423 struct table_dpif *table = &ofproto->tables[table_id];
7424 const struct oftable *oftable = &ofproto->up.tables[table_id];
7425 struct cls_table *catchall, *other;
7426 struct cls_table *t;
7428 catchall = other = NULL;
7430 switch (hmap_count(&oftable->cls.tables)) {
7432 /* We could tag this OpenFlow table but it would make the logic a
7433 * little harder and it's a corner case that doesn't seem worth it
7439 HMAP_FOR_EACH (t, hmap_node, &oftable->cls.tables) {
7440 if (cls_table_is_catchall(t)) {
7442 } else if (!other) {
7445 /* Indicate that we can't tag this by setting both tables to
7446 * NULL. (We know that 'catchall' is already NULL.) */
7453 /* Can't tag this table. */
7457 if (table->catchall_table != catchall || table->other_table != other) {
7458 table->catchall_table = catchall;
7459 table->other_table = other;
7460 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
7464 /* Given 'rule' that has changed in some way (either it is a rule being
7465 * inserted, a rule being deleted, or a rule whose actions are being
7466 * modified), marks facets for revalidation to ensure that packets will be
7467 * forwarded correctly according to the new state of the flow table.
7469 * This function must be called after *each* change to a flow table. See
7470 * the comment on table_update_taggable() for more information. */
7472 rule_invalidate(const struct rule_dpif *rule)
7474 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
7476 table_update_taggable(ofproto, rule->up.table_id);
7478 if (!ofproto->backer->need_revalidate) {
7479 struct table_dpif *table = &ofproto->tables[rule->up.table_id];
7481 if (table->other_table && rule->tag) {
7482 tag_set_add(&ofproto->backer->revalidate_set, rule->tag);
7484 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
7490 set_frag_handling(struct ofproto *ofproto_,
7491 enum ofp_config_flags frag_handling)
7493 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7494 if (frag_handling != OFPC_FRAG_REASM) {
7495 ofproto->backer->need_revalidate = REV_RECONFIGURE;
7503 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
7504 const struct flow *flow,
7505 const struct ofpact *ofpacts, size_t ofpacts_len)
7507 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7508 struct initial_vals initial_vals;
7509 struct odputil_keybuf keybuf;
7510 struct dpif_flow_stats stats;
7514 struct action_xlate_ctx ctx;
7515 uint64_t odp_actions_stub[1024 / 8];
7516 struct ofpbuf odp_actions;
7518 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
7519 odp_flow_key_from_flow(&key, flow,
7520 ofp_port_to_odp_port(ofproto, flow->in_port));
7522 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
7524 initial_vals.vlan_tci = flow->vlan_tci;
7525 initial_vals.tunnel_ip_tos = 0;
7526 action_xlate_ctx_init(&ctx, ofproto, flow, &initial_vals, NULL,
7527 packet_get_tcp_flags(packet, flow), packet);
7528 ctx.resubmit_stats = &stats;
7530 ofpbuf_use_stub(&odp_actions,
7531 odp_actions_stub, sizeof odp_actions_stub);
7532 xlate_actions(&ctx, ofpacts, ofpacts_len, &odp_actions);
7533 dpif_execute(ofproto->backer->dpif, key.data, key.size,
7534 odp_actions.data, odp_actions.size, packet);
7535 ofpbuf_uninit(&odp_actions);
7543 set_netflow(struct ofproto *ofproto_,
7544 const struct netflow_options *netflow_options)
7546 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7548 if (netflow_options) {
7549 if (!ofproto->netflow) {
7550 ofproto->netflow = netflow_create();
7552 return netflow_set_options(ofproto->netflow, netflow_options);
7554 netflow_destroy(ofproto->netflow);
7555 ofproto->netflow = NULL;
7561 get_netflow_ids(const struct ofproto *ofproto_,
7562 uint8_t *engine_type, uint8_t *engine_id)
7564 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7566 dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
7570 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
7572 if (!facet_is_controller_flow(facet) &&
7573 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
7574 struct subfacet *subfacet;
7575 struct ofexpired expired;
7577 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
7578 if (subfacet->path == SF_FAST_PATH) {
7579 struct dpif_flow_stats stats;
7581 subfacet_reinstall(subfacet, &stats);
7582 subfacet_update_stats(subfacet, &stats);
7586 expired.flow = facet->flow;
7587 expired.packet_count = facet->packet_count;
7588 expired.byte_count = facet->byte_count;
7589 expired.used = facet->used;
7590 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
7595 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
7597 struct facet *facet;
7599 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
7600 send_active_timeout(ofproto, facet);
7604 static struct ofproto_dpif *
7605 ofproto_dpif_lookup(const char *name)
7607 struct ofproto_dpif *ofproto;
7609 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
7610 hash_string(name, 0), &all_ofproto_dpifs) {
7611 if (!strcmp(ofproto->up.name, name)) {
7619 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
7620 const char *argv[], void *aux OVS_UNUSED)
7622 struct ofproto_dpif *ofproto;
7625 ofproto = ofproto_dpif_lookup(argv[1]);
7627 unixctl_command_reply_error(conn, "no such bridge");
7630 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
7632 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
7633 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
7637 unixctl_command_reply(conn, "table successfully flushed");
7641 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
7642 const char *argv[], void *aux OVS_UNUSED)
7644 struct ds ds = DS_EMPTY_INITIALIZER;
7645 const struct ofproto_dpif *ofproto;
7646 const struct mac_entry *e;
7648 ofproto = ofproto_dpif_lookup(argv[1]);
7650 unixctl_command_reply_error(conn, "no such bridge");
7654 ds_put_cstr(&ds, " port VLAN MAC Age\n");
7655 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
7656 struct ofbundle *bundle = e->port.p;
7657 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
7658 ofbundle_get_a_port(bundle)->odp_port,
7659 e->vlan, ETH_ADDR_ARGS(e->mac),
7660 mac_entry_age(ofproto->ml, e));
7662 unixctl_command_reply(conn, ds_cstr(&ds));
7667 struct action_xlate_ctx ctx;
7673 trace_format_rule(struct ds *result, uint8_t table_id, int level,
7674 const struct rule_dpif *rule)
7676 ds_put_char_multiple(result, '\t', level);
7678 ds_put_cstr(result, "No match\n");
7682 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
7683 table_id, ntohll(rule->up.flow_cookie));
7684 cls_rule_format(&rule->up.cr, result);
7685 ds_put_char(result, '\n');
7687 ds_put_char_multiple(result, '\t', level);
7688 ds_put_cstr(result, "OpenFlow ");
7689 ofpacts_format(rule->up.ofpacts, rule->up.ofpacts_len, result);
7690 ds_put_char(result, '\n');
7694 trace_format_flow(struct ds *result, int level, const char *title,
7695 struct trace_ctx *trace)
7697 ds_put_char_multiple(result, '\t', level);
7698 ds_put_format(result, "%s: ", title);
7699 if (flow_equal(&trace->ctx.flow, &trace->flow)) {
7700 ds_put_cstr(result, "unchanged");
7702 flow_format(result, &trace->ctx.flow);
7703 trace->flow = trace->ctx.flow;
7705 ds_put_char(result, '\n');
7709 trace_format_regs(struct ds *result, int level, const char *title,
7710 struct trace_ctx *trace)
7714 ds_put_char_multiple(result, '\t', level);
7715 ds_put_format(result, "%s:", title);
7716 for (i = 0; i < FLOW_N_REGS; i++) {
7717 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
7719 ds_put_char(result, '\n');
7723 trace_format_odp(struct ds *result, int level, const char *title,
7724 struct trace_ctx *trace)
7726 struct ofpbuf *odp_actions = trace->ctx.odp_actions;
7728 ds_put_char_multiple(result, '\t', level);
7729 ds_put_format(result, "%s: ", title);
7730 format_odp_actions(result, odp_actions->data, odp_actions->size);
7731 ds_put_char(result, '\n');
7735 trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
7737 struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx);
7738 struct ds *result = trace->result;
7740 ds_put_char(result, '\n');
7741 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
7742 trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace);
7743 trace_format_odp(result, ctx->recurse + 1, "Resubmitted odp", trace);
7744 trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule);
7748 trace_report(struct action_xlate_ctx *ctx, const char *s)
7750 struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx);
7751 struct ds *result = trace->result;
7753 ds_put_char_multiple(result, '\t', ctx->recurse);
7754 ds_put_cstr(result, s);
7755 ds_put_char(result, '\n');
7759 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
7760 void *aux OVS_UNUSED)
7762 const char *dpname = argv[1];
7763 struct ofproto_dpif *ofproto;
7764 struct ofpbuf odp_key;
7765 struct ofpbuf *packet;
7766 struct initial_vals initial_vals;
7772 ofpbuf_init(&odp_key, 0);
7775 ofproto = ofproto_dpif_lookup(dpname);
7777 unixctl_command_reply_error(conn, "Unknown ofproto (use ofproto/list "
7781 if (argc == 3 || (argc == 4 && !strcmp(argv[3], "-generate"))) {
7782 /* ofproto/trace dpname flow [-generate] */
7783 const char *flow_s = argv[2];
7784 const char *generate_s = argv[3];
7786 /* Allow 'flow_s' to be either a datapath flow or an OpenFlow-like
7787 * flow. We guess which type it is based on whether 'flow_s' contains
7788 * an '(', since a datapath flow always contains '(') but an
7789 * OpenFlow-like flow should not (in fact it's allowed but I believe
7790 * that's not documented anywhere).
7792 * An alternative would be to try to parse 'flow_s' both ways, but then
7793 * it would be tricky giving a sensible error message. After all, do
7794 * you just say "syntax error" or do you present both error messages?
7795 * Both choices seem lousy. */
7796 if (strchr(flow_s, '(')) {
7799 /* Convert string to datapath key. */
7800 ofpbuf_init(&odp_key, 0);
7801 error = odp_flow_key_from_string(flow_s, NULL, &odp_key);
7803 unixctl_command_reply_error(conn, "Bad flow syntax");
7807 /* The user might have specified the wrong ofproto but within the
7808 * same backer. That's OK, ofproto_receive() can find the right
7810 if (ofproto_receive(ofproto->backer, NULL, odp_key.data,
7811 odp_key.size, &flow, NULL, &ofproto, NULL,
7813 unixctl_command_reply_error(conn, "Invalid flow");
7816 ds_put_format(&result, "Bridge: %s\n", ofproto->up.name);
7820 error_s = parse_ofp_exact_flow(&flow, argv[2]);
7822 unixctl_command_reply_error(conn, error_s);
7827 initial_vals.vlan_tci = flow.vlan_tci;
7828 initial_vals.tunnel_ip_tos = flow.tunnel.ip_tos;
7831 /* Generate a packet, if requested. */
7833 packet = ofpbuf_new(0);
7834 flow_compose(packet, &flow);
7836 } else if (argc == 7) {
7837 /* ofproto/trace dpname priority tun_id in_port mark packet */
7838 const char *priority_s = argv[2];
7839 const char *tun_id_s = argv[3];
7840 const char *in_port_s = argv[4];
7841 const char *mark_s = argv[5];
7842 const char *packet_s = argv[6];
7843 uint32_t in_port = atoi(in_port_s);
7844 ovs_be64 tun_id = htonll(strtoull(tun_id_s, NULL, 0));
7845 uint32_t priority = atoi(priority_s);
7846 uint32_t mark = atoi(mark_s);
7849 msg = eth_from_hex(packet_s, &packet);
7851 unixctl_command_reply_error(conn, msg);
7855 ds_put_cstr(&result, "Packet: ");
7856 s = ofp_packet_to_string(packet->data, packet->size);
7857 ds_put_cstr(&result, s);
7860 flow_extract(packet, priority, mark, NULL, in_port, &flow);
7861 flow.tunnel.tun_id = tun_id;
7862 initial_vals.vlan_tci = flow.vlan_tci;
7863 initial_vals.tunnel_ip_tos = flow.tunnel.ip_tos;
7865 unixctl_command_reply_error(conn, "Bad command syntax");
7869 ofproto_trace(ofproto, &flow, packet, &initial_vals, &result);
7870 unixctl_command_reply(conn, ds_cstr(&result));
7873 ds_destroy(&result);
7874 ofpbuf_delete(packet);
7875 ofpbuf_uninit(&odp_key);
7879 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
7880 const struct ofpbuf *packet,
7881 const struct initial_vals *initial_vals, struct ds *ds)
7883 struct rule_dpif *rule;
7885 ds_put_cstr(ds, "Flow: ");
7886 flow_format(ds, flow);
7887 ds_put_char(ds, '\n');
7889 rule = rule_dpif_lookup(ofproto, flow);
7891 trace_format_rule(ds, 0, 0, rule);
7892 if (rule == ofproto->miss_rule) {
7893 ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
7894 } else if (rule == ofproto->no_packet_in_rule) {
7895 ds_put_cstr(ds, "\nNo match, packets dropped because "
7896 "OFPPC_NO_PACKET_IN is set on in_port.\n");
7900 uint64_t odp_actions_stub[1024 / 8];
7901 struct ofpbuf odp_actions;
7903 struct trace_ctx trace;
7906 tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
7909 ofpbuf_use_stub(&odp_actions,
7910 odp_actions_stub, sizeof odp_actions_stub);
7911 action_xlate_ctx_init(&trace.ctx, ofproto, flow, initial_vals,
7912 rule, tcp_flags, packet);
7913 trace.ctx.resubmit_hook = trace_resubmit;
7914 trace.ctx.report_hook = trace_report;
7915 xlate_actions(&trace.ctx, rule->up.ofpacts, rule->up.ofpacts_len,
7918 ds_put_char(ds, '\n');
7919 trace_format_flow(ds, 0, "Final flow", &trace);
7920 ds_put_cstr(ds, "Datapath actions: ");
7921 format_odp_actions(ds, odp_actions.data, odp_actions.size);
7922 ofpbuf_uninit(&odp_actions);
7924 if (trace.ctx.slow) {
7925 enum slow_path_reason slow;
7927 ds_put_cstr(ds, "\nThis flow is handled by the userspace "
7928 "slow path because it:");
7929 for (slow = trace.ctx.slow; slow; ) {
7930 enum slow_path_reason bit = rightmost_1bit(slow);
7934 ds_put_cstr(ds, "\n\t- Consists of CFM packets.");
7937 ds_put_cstr(ds, "\n\t- Consists of LACP packets.");
7940 ds_put_cstr(ds, "\n\t- Consists of STP packets.");
7943 ds_put_cstr(ds, "\n\t- Needs in-band special case "
7946 ds_put_cstr(ds, "\n\t (The datapath actions are "
7947 "incomplete--for complete actions, "
7948 "please supply a packet.)");
7951 case SLOW_CONTROLLER:
7952 ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages "
7953 "to the OpenFlow controller.");
7956 ds_put_cstr(ds, "\n\t- Needs more specific matching "
7957 "than the datapath supports.");
7964 if (slow & ~SLOW_MATCH) {
7965 ds_put_cstr(ds, "\nThe datapath actions above do not reflect "
7966 "the special slow-path processing.");
7973 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
7974 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
7977 unixctl_command_reply(conn, NULL);
7981 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
7982 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
7985 unixctl_command_reply(conn, NULL);
7988 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
7989 * 'reply' describing the results. */
7991 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
7993 struct facet *facet;
7997 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
7998 if (!facet_check_consistency(facet)) {
8003 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
8007 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
8008 ofproto->up.name, errors);
8010 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
8015 ofproto_dpif_self_check(struct unixctl_conn *conn,
8016 int argc, const char *argv[], void *aux OVS_UNUSED)
8018 struct ds reply = DS_EMPTY_INITIALIZER;
8019 struct ofproto_dpif *ofproto;
8022 ofproto = ofproto_dpif_lookup(argv[1]);
8024 unixctl_command_reply_error(conn, "Unknown ofproto (use "
8025 "ofproto/list for help)");
8028 ofproto_dpif_self_check__(ofproto, &reply);
8030 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
8031 ofproto_dpif_self_check__(ofproto, &reply);
8035 unixctl_command_reply(conn, ds_cstr(&reply));
8039 /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list
8040 * of the 'ofproto_shash' nodes. It is the responsibility of the caller
8041 * to destroy 'ofproto_shash' and free the returned value. */
8042 static const struct shash_node **
8043 get_ofprotos(struct shash *ofproto_shash)
8045 const struct ofproto_dpif *ofproto;
8047 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
8048 char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
8049 shash_add_nocopy(ofproto_shash, name, ofproto);
8052 return shash_sort(ofproto_shash);
8056 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
8057 const char *argv[] OVS_UNUSED,
8058 void *aux OVS_UNUSED)
8060 struct ds ds = DS_EMPTY_INITIALIZER;
8061 struct shash ofproto_shash;
8062 const struct shash_node **sorted_ofprotos;
8065 shash_init(&ofproto_shash);
8066 sorted_ofprotos = get_ofprotos(&ofproto_shash);
8067 for (i = 0; i < shash_count(&ofproto_shash); i++) {
8068 const struct shash_node *node = sorted_ofprotos[i];
8069 ds_put_format(&ds, "%s\n", node->name);
8072 shash_destroy(&ofproto_shash);
8073 free(sorted_ofprotos);
8075 unixctl_command_reply(conn, ds_cstr(&ds));
8080 show_dp_format(const struct ofproto_dpif *ofproto, struct ds *ds)
8082 const struct shash_node **ports;
8085 ds_put_format(ds, "%s (%s):\n", ofproto->up.name,
8086 dpif_name(ofproto->backer->dpif));
8088 "\tlookups: hit:%"PRIu64" missed:%"PRIu64"\n",
8089 ofproto->n_hit, ofproto->n_missed);
8090 ds_put_format(ds, "\tflows: %zu\n",
8091 hmap_count(&ofproto->subfacets));
8093 ports = shash_sort(&ofproto->up.port_by_name);
8094 for (i = 0; i < shash_count(&ofproto->up.port_by_name); i++) {
8095 const struct shash_node *node = ports[i];
8096 struct ofport *ofport = node->data;
8097 const char *name = netdev_get_name(ofport->netdev);
8098 const char *type = netdev_get_type(ofport->netdev);
8101 ds_put_format(ds, "\t%s %u/", name, ofport->ofp_port);
8103 odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
8104 if (odp_port != OVSP_NONE) {
8105 ds_put_format(ds, "%"PRIu32":", odp_port);
8107 ds_put_cstr(ds, "none:");
8110 if (strcmp(type, "system")) {
8111 struct netdev *netdev;
8114 ds_put_format(ds, " (%s", type);
8116 error = netdev_open(name, type, &netdev);
8121 error = netdev_get_config(netdev, &config);
8123 const struct smap_node **nodes;
8126 nodes = smap_sort(&config);
8127 for (i = 0; i < smap_count(&config); i++) {
8128 const struct smap_node *node = nodes[i];
8129 ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
8130 node->key, node->value);
8134 smap_destroy(&config);
8136 netdev_close(netdev);
8138 ds_put_char(ds, ')');
8140 ds_put_char(ds, '\n');
8146 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc,
8147 const char *argv[], void *aux OVS_UNUSED)
8149 struct ds ds = DS_EMPTY_INITIALIZER;
8150 const struct ofproto_dpif *ofproto;
8154 for (i = 1; i < argc; i++) {
8155 ofproto = ofproto_dpif_lookup(argv[i]);
8157 ds_put_format(&ds, "Unknown bridge %s (use dpif/dump-dps "
8158 "for help)", argv[i]);
8159 unixctl_command_reply_error(conn, ds_cstr(&ds));
8162 show_dp_format(ofproto, &ds);
8165 struct shash ofproto_shash;
8166 const struct shash_node **sorted_ofprotos;
8169 shash_init(&ofproto_shash);
8170 sorted_ofprotos = get_ofprotos(&ofproto_shash);
8171 for (i = 0; i < shash_count(&ofproto_shash); i++) {
8172 const struct shash_node *node = sorted_ofprotos[i];
8173 show_dp_format(node->data, &ds);
8176 shash_destroy(&ofproto_shash);
8177 free(sorted_ofprotos);
8180 unixctl_command_reply(conn, ds_cstr(&ds));
8185 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
8186 int argc OVS_UNUSED, const char *argv[],
8187 void *aux OVS_UNUSED)
8189 struct ds ds = DS_EMPTY_INITIALIZER;
8190 const struct ofproto_dpif *ofproto;
8191 struct subfacet *subfacet;
8193 ofproto = ofproto_dpif_lookup(argv[1]);
8195 unixctl_command_reply_error(conn, "no such bridge");
8199 update_stats(ofproto->backer);
8201 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
8202 odp_flow_key_format(subfacet->key, subfacet->key_len, &ds);
8204 ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:",
8205 subfacet->dp_packet_count, subfacet->dp_byte_count);
8206 if (subfacet->used) {
8207 ds_put_format(&ds, "%.3fs",
8208 (time_msec() - subfacet->used) / 1000.0);
8210 ds_put_format(&ds, "never");
8212 if (subfacet->facet->tcp_flags) {
8213 ds_put_cstr(&ds, ", flags:");
8214 packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags);
8217 ds_put_cstr(&ds, ", actions:");
8218 if (subfacet->slow) {
8219 uint64_t slow_path_stub[128 / 8];
8220 const struct nlattr *actions;
8223 compose_slow_path(ofproto, &subfacet->facet->flow, subfacet->slow,
8224 slow_path_stub, sizeof slow_path_stub,
8225 &actions, &actions_len);
8226 format_odp_actions(&ds, actions, actions_len);
8228 format_odp_actions(&ds, subfacet->actions, subfacet->actions_len);
8230 ds_put_char(&ds, '\n');
8233 unixctl_command_reply(conn, ds_cstr(&ds));
8238 ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn,
8239 int argc OVS_UNUSED, const char *argv[],
8240 void *aux OVS_UNUSED)
8242 struct ds ds = DS_EMPTY_INITIALIZER;
8243 struct ofproto_dpif *ofproto;
8245 ofproto = ofproto_dpif_lookup(argv[1]);
8247 unixctl_command_reply_error(conn, "no such bridge");
8251 flush(&ofproto->up);
8253 unixctl_command_reply(conn, ds_cstr(&ds));
8258 ofproto_dpif_unixctl_init(void)
8260 static bool registered;
8266 unixctl_command_register(
8268 "bridge {priority tun_id in_port mark packet | odp_flow [-generate]}",
8269 2, 6, ofproto_unixctl_trace, NULL);
8270 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
8271 ofproto_unixctl_fdb_flush, NULL);
8272 unixctl_command_register("fdb/show", "bridge", 1, 1,
8273 ofproto_unixctl_fdb_show, NULL);
8274 unixctl_command_register("ofproto/clog", "", 0, 0,
8275 ofproto_dpif_clog, NULL);
8276 unixctl_command_register("ofproto/unclog", "", 0, 0,
8277 ofproto_dpif_unclog, NULL);
8278 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
8279 ofproto_dpif_self_check, NULL);
8280 unixctl_command_register("dpif/dump-dps", "", 0, 0,
8281 ofproto_unixctl_dpif_dump_dps, NULL);
8282 unixctl_command_register("dpif/show", "[bridge]", 0, INT_MAX,
8283 ofproto_unixctl_dpif_show, NULL);
8284 unixctl_command_register("dpif/dump-flows", "bridge", 1, 1,
8285 ofproto_unixctl_dpif_dump_flows, NULL);
8286 unixctl_command_register("dpif/del-flows", "bridge", 1, 1,
8287 ofproto_unixctl_dpif_del_flows, NULL);
8290 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
8292 * This is deprecated. It is only for compatibility with broken device drivers
8293 * in old versions of Linux that do not properly support VLANs when VLAN
8294 * devices are not used. When broken device drivers are no longer in
8295 * widespread use, we will delete these interfaces. */
8298 set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid)
8300 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
8301 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
8303 if (realdev_ofp_port == ofport->realdev_ofp_port
8304 && vid == ofport->vlandev_vid) {
8308 ofproto->backer->need_revalidate = REV_RECONFIGURE;
8310 if (ofport->realdev_ofp_port) {
8313 if (realdev_ofp_port && ofport->bundle) {
8314 /* vlandevs are enslaved to their realdevs, so they are not allowed to
8315 * themselves be part of a bundle. */
8316 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
8319 ofport->realdev_ofp_port = realdev_ofp_port;
8320 ofport->vlandev_vid = vid;
8322 if (realdev_ofp_port) {
8323 vsp_add(ofport, realdev_ofp_port, vid);
8330 hash_realdev_vid(uint16_t realdev_ofp_port, int vid)
8332 return hash_2words(realdev_ofp_port, vid);
8335 /* Returns the ODP port number of the Linux VLAN device that corresponds to
8336 * 'vlan_tci' on the network device with port number 'realdev_odp_port' in
8337 * 'ofproto'. For example, given 'realdev_odp_port' of eth0 and 'vlan_tci' 9,
8338 * it would return the port number of eth0.9.
8340 * Unless VLAN splinters are enabled for port 'realdev_odp_port', this
8341 * function just returns its 'realdev_odp_port' argument. */
8343 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
8344 uint32_t realdev_odp_port, ovs_be16 vlan_tci)
8346 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
8347 uint16_t realdev_ofp_port;
8348 int vid = vlan_tci_to_vid(vlan_tci);
8349 const struct vlan_splinter *vsp;
8351 realdev_ofp_port = odp_port_to_ofp_port(ofproto, realdev_odp_port);
8352 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
8353 hash_realdev_vid(realdev_ofp_port, vid),
8354 &ofproto->realdev_vid_map) {
8355 if (vsp->realdev_ofp_port == realdev_ofp_port
8356 && vsp->vid == vid) {
8357 return ofp_port_to_odp_port(ofproto, vsp->vlandev_ofp_port);
8361 return realdev_odp_port;
8364 static struct vlan_splinter *
8365 vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port)
8367 struct vlan_splinter *vsp;
8369 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, hash_int(vlandev_ofp_port, 0),
8370 &ofproto->vlandev_map) {
8371 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
8379 /* Returns the OpenFlow port number of the "real" device underlying the Linux
8380 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
8381 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
8382 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
8383 * eth0 and store 9 in '*vid'.
8385 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
8386 * VLAN device. Unless VLAN splinters are enabled, this is what this function
8389 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
8390 uint16_t vlandev_ofp_port, int *vid)
8392 if (!hmap_is_empty(&ofproto->vlandev_map)) {
8393 const struct vlan_splinter *vsp;
8395 vsp = vlandev_find(ofproto, vlandev_ofp_port);
8400 return vsp->realdev_ofp_port;
8406 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
8407 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
8408 * 'flow->in_port' to the "real" device backing the VLAN device, sets
8409 * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is
8410 * always the case unless VLAN splinters are enabled), returns false without
8411 * making any changes. */
8413 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
8418 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port, &vid);
8423 /* Cause the flow to be processed as if it came in on the real device with
8424 * the VLAN device's VLAN ID. */
8425 flow->in_port = realdev;
8426 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
8431 vsp_remove(struct ofport_dpif *port)
8433 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
8434 struct vlan_splinter *vsp;
8436 vsp = vlandev_find(ofproto, port->up.ofp_port);
8438 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
8439 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
8442 port->realdev_ofp_port = 0;
8444 VLOG_ERR("missing vlan device record");
8449 vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid)
8451 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
8453 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
8454 && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid))
8455 == realdev_ofp_port)) {
8456 struct vlan_splinter *vsp;
8458 vsp = xmalloc(sizeof *vsp);
8459 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
8460 hash_int(port->up.ofp_port, 0));
8461 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
8462 hash_realdev_vid(realdev_ofp_port, vid));
8463 vsp->realdev_ofp_port = realdev_ofp_port;
8464 vsp->vlandev_ofp_port = port->up.ofp_port;
8467 port->realdev_ofp_port = realdev_ofp_port;
8469 VLOG_ERR("duplicate vlan device record");
8474 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
8476 const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
8477 return ofport ? ofport->odp_port : OVSP_NONE;
8480 static struct ofport_dpif *
8481 odp_port_to_ofport(const struct dpif_backer *backer, uint32_t odp_port)
8483 struct ofport_dpif *port;
8485 HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node,
8486 hash_int(odp_port, 0),
8487 &backer->odp_to_ofport_map) {
8488 if (port->odp_port == odp_port) {
8497 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
8499 struct ofport_dpif *port;
8501 port = odp_port_to_ofport(ofproto->backer, odp_port);
8502 if (port && &ofproto->up == port->up.ofproto) {
8503 return port->up.ofp_port;
8510 dpif_stats_update_hit_count(struct ofproto_dpif *ofproto, uint64_t delta)
8512 ofproto->n_hit += delta;
8515 const struct ofproto_class ofproto_dpif_class = {
8550 port_is_lacp_current,
8551 NULL, /* rule_choose_table */
8558 rule_modify_actions,
8569 get_stp_port_status,
8576 is_mirror_output_bundle,
8577 forward_bpdu_changed,
8578 set_mac_table_config,