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 struct ofproto_dpif {
669 struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
671 struct dpif_backer *backer;
673 /* Special OpenFlow rules. */
674 struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */
675 struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */
681 struct netflow *netflow;
682 struct dpif_sflow *sflow;
683 struct hmap bundles; /* Contains "struct ofbundle"s. */
684 struct mac_learning *ml;
685 struct ofmirror *mirrors[MAX_MIRRORS];
687 bool has_bonded_bundles;
691 struct hmap subfacets;
692 struct governor *governor;
695 struct table_dpif tables[N_TABLES];
697 /* Support for debugging async flow mods. */
698 struct list completions;
700 bool has_bundle_action; /* True when the first bundle action appears. */
701 struct netdev_stats stats; /* To account packets generated and consumed in
706 long long int stp_last_tick;
708 /* VLAN splinters. */
709 struct hmap realdev_vid_map; /* (realdev,vid) -> vlandev. */
710 struct hmap vlandev_map; /* vlandev -> (realdev,vid). */
713 struct sset ports; /* Set of standard port names. */
714 struct sset ghost_ports; /* Ports with no datapath port. */
715 struct sset port_poll_set; /* Queued names for port_poll() reply. */
716 int port_poll_errno; /* Last errno for port_poll() reply. */
719 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
720 * for debugging the asynchronous flow_mod implementation.) */
723 /* All existing ofproto_dpif instances, indexed by ->up.name. */
724 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
726 static void ofproto_dpif_unixctl_init(void);
728 static struct ofproto_dpif *
729 ofproto_dpif_cast(const struct ofproto *ofproto)
731 ovs_assert(ofproto->ofproto_class == &ofproto_dpif_class);
732 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
735 static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *,
737 static struct ofport_dpif *get_odp_port(const struct ofproto_dpif *,
739 static void ofproto_trace(struct ofproto_dpif *, const struct flow *,
740 const struct ofpbuf *,
741 const struct initial_vals *, struct ds *);
743 /* Packet processing. */
744 static void update_learning_table(struct ofproto_dpif *,
745 const struct flow *, int vlan,
748 #define FLOW_MISS_MAX_BATCH 50
749 static int handle_upcalls(struct dpif_backer *, unsigned int max_batch);
751 /* Flow expiration. */
752 static int expire(struct dpif_backer *);
755 static void send_netflow_active_timeouts(struct ofproto_dpif *);
758 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
759 static size_t compose_sflow_action(const struct ofproto_dpif *,
760 struct ofpbuf *odp_actions,
761 const struct flow *, uint32_t odp_port);
762 static void add_mirror_actions(struct action_xlate_ctx *ctx,
763 const struct flow *flow);
764 /* Global variables. */
765 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
767 /* Initial mappings of port to bridge mappings. */
768 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
770 /* Factory functions. */
773 init(const struct shash *iface_hints)
775 struct shash_node *node;
777 /* Make a local copy, since we don't own 'iface_hints' elements. */
778 SHASH_FOR_EACH(node, iface_hints) {
779 const struct iface_hint *orig_hint = node->data;
780 struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
782 new_hint->br_name = xstrdup(orig_hint->br_name);
783 new_hint->br_type = xstrdup(orig_hint->br_type);
784 new_hint->ofp_port = orig_hint->ofp_port;
786 shash_add(&init_ofp_ports, node->name, new_hint);
791 enumerate_types(struct sset *types)
793 dp_enumerate_types(types);
797 enumerate_names(const char *type, struct sset *names)
799 struct ofproto_dpif *ofproto;
802 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
803 if (strcmp(type, ofproto->up.type)) {
806 sset_add(names, ofproto->up.name);
813 del(const char *type, const char *name)
818 error = dpif_open(name, type, &dpif);
820 error = dpif_delete(dpif);
827 port_open_type(const char *datapath_type, const char *port_type)
829 return dpif_port_open_type(datapath_type, port_type);
832 /* Type functions. */
834 static struct ofproto_dpif *
835 lookup_ofproto_dpif_by_port_name(const char *name)
837 struct ofproto_dpif *ofproto;
839 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
840 if (sset_contains(&ofproto->ports, name)) {
849 type_run(const char *type)
851 struct dpif_backer *backer;
855 backer = shash_find_data(&all_dpif_backers, type);
857 /* This is not necessarily a problem, since backers are only
858 * created on demand. */
862 dpif_run(backer->dpif);
864 if (backer->need_revalidate
865 || !tag_set_is_empty(&backer->revalidate_set)) {
866 struct tag_set revalidate_set = backer->revalidate_set;
867 bool need_revalidate = backer->need_revalidate;
868 struct ofproto_dpif *ofproto;
869 struct simap_node *node;
870 struct simap tmp_backers;
872 /* Handle tunnel garbage collection. */
873 simap_init(&tmp_backers);
874 simap_swap(&backer->tnl_backers, &tmp_backers);
876 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
877 struct ofport_dpif *iter;
879 if (backer != ofproto->backer) {
883 HMAP_FOR_EACH (iter, up.hmap_node, &ofproto->up.ports) {
886 if (!iter->tnl_port) {
890 dp_port = netdev_vport_get_dpif_port(iter->up.netdev);
891 node = simap_find(&tmp_backers, dp_port);
893 simap_put(&backer->tnl_backers, dp_port, node->data);
894 simap_delete(&tmp_backers, node);
895 node = simap_find(&backer->tnl_backers, dp_port);
897 node = simap_find(&backer->tnl_backers, dp_port);
899 uint32_t odp_port = UINT32_MAX;
901 if (!dpif_port_add(backer->dpif, iter->up.netdev,
903 simap_put(&backer->tnl_backers, dp_port, odp_port);
904 node = simap_find(&backer->tnl_backers, dp_port);
909 iter->odp_port = node ? node->data : OVSP_NONE;
910 if (tnl_port_reconfigure(&iter->up, iter->odp_port,
912 backer->need_revalidate = REV_RECONFIGURE;
917 SIMAP_FOR_EACH (node, &tmp_backers) {
918 dpif_port_del(backer->dpif, node->data);
920 simap_destroy(&tmp_backers);
922 switch (backer->need_revalidate) {
923 case REV_RECONFIGURE: COVERAGE_INC(rev_reconfigure); break;
924 case REV_STP: COVERAGE_INC(rev_stp); break;
925 case REV_PORT_TOGGLED: COVERAGE_INC(rev_port_toggled); break;
926 case REV_FLOW_TABLE: COVERAGE_INC(rev_flow_table); break;
927 case REV_INCONSISTENCY: COVERAGE_INC(rev_inconsistency); break;
930 if (backer->need_revalidate) {
931 /* Clear the drop_keys in case we should now be accepting some
932 * formerly dropped flows. */
933 drop_key_clear(backer);
936 /* Clear the revalidation flags. */
937 tag_set_init(&backer->revalidate_set);
938 backer->need_revalidate = 0;
940 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
941 struct facet *facet, *next;
943 if (ofproto->backer != backer) {
947 HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &ofproto->facets) {
949 || tag_set_intersects(&revalidate_set, facet->tags)) {
950 facet_revalidate(facet);
956 if (timer_expired(&backer->next_expiration)) {
957 int delay = expire(backer);
958 timer_set_duration(&backer->next_expiration, delay);
961 /* Check for port changes in the dpif. */
962 while ((error = dpif_port_poll(backer->dpif, &devname)) == 0) {
963 struct ofproto_dpif *ofproto;
964 struct dpif_port port;
966 /* Don't report on the datapath's device. */
967 if (!strcmp(devname, dpif_base_name(backer->dpif))) {
971 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
972 &all_ofproto_dpifs) {
973 if (simap_contains(&ofproto->backer->tnl_backers, devname)) {
978 ofproto = lookup_ofproto_dpif_by_port_name(devname);
979 if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
980 /* The port was removed. If we know the datapath,
981 * report it through poll_set(). If we don't, it may be
982 * notifying us of a removal we initiated, so ignore it.
983 * If there's a pending ENOBUFS, let it stand, since
984 * everything will be reevaluated. */
985 if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
986 sset_add(&ofproto->port_poll_set, devname);
987 ofproto->port_poll_errno = 0;
989 } else if (!ofproto) {
990 /* The port was added, but we don't know with which
991 * ofproto we should associate it. Delete it. */
992 dpif_port_del(backer->dpif, port.port_no);
994 dpif_port_destroy(&port);
1000 if (error != EAGAIN) {
1001 struct ofproto_dpif *ofproto;
1003 /* There was some sort of error, so propagate it to all
1004 * ofprotos that use this backer. */
1005 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
1006 &all_ofproto_dpifs) {
1007 if (ofproto->backer == backer) {
1008 sset_clear(&ofproto->port_poll_set);
1009 ofproto->port_poll_errno = error;
1018 type_run_fast(const char *type)
1020 struct dpif_backer *backer;
1023 backer = shash_find_data(&all_dpif_backers, type);
1025 /* This is not necessarily a problem, since backers are only
1026 * created on demand. */
1030 /* Handle one or more batches of upcalls, until there's nothing left to do
1031 * or until we do a fixed total amount of work.
1033 * We do work in batches because it can be much cheaper to set up a number
1034 * of flows and fire off their patches all at once. We do multiple batches
1035 * because in some cases handling a packet can cause another packet to be
1036 * queued almost immediately as part of the return flow. Both
1037 * optimizations can make major improvements on some benchmarks and
1038 * presumably for real traffic as well. */
1040 while (work < FLOW_MISS_MAX_BATCH) {
1041 int retval = handle_upcalls(backer, FLOW_MISS_MAX_BATCH - work);
1052 type_wait(const char *type)
1054 struct dpif_backer *backer;
1056 backer = shash_find_data(&all_dpif_backers, type);
1058 /* This is not necessarily a problem, since backers are only
1059 * created on demand. */
1063 timer_wait(&backer->next_expiration);
1066 /* Basic life-cycle. */
1068 static int add_internal_flows(struct ofproto_dpif *);
1070 static struct ofproto *
1073 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
1074 return &ofproto->up;
1078 dealloc(struct ofproto *ofproto_)
1080 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1085 close_dpif_backer(struct dpif_backer *backer)
1087 struct shash_node *node;
1089 ovs_assert(backer->refcount > 0);
1091 if (--backer->refcount) {
1095 drop_key_clear(backer);
1096 hmap_destroy(&backer->drop_keys);
1098 simap_destroy(&backer->tnl_backers);
1099 hmap_destroy(&backer->odp_to_ofport_map);
1100 node = shash_find(&all_dpif_backers, backer->type);
1102 shash_delete(&all_dpif_backers, node);
1103 dpif_close(backer->dpif);
1108 /* Datapath port slated for removal from datapath. */
1109 struct odp_garbage {
1110 struct list list_node;
1115 open_dpif_backer(const char *type, struct dpif_backer **backerp)
1117 struct dpif_backer *backer;
1118 struct dpif_port_dump port_dump;
1119 struct dpif_port port;
1120 struct shash_node *node;
1121 struct list garbage_list;
1122 struct odp_garbage *garbage, *next;
1128 backer = shash_find_data(&all_dpif_backers, type);
1135 backer_name = xasprintf("ovs-%s", type);
1137 /* Remove any existing datapaths, since we assume we're the only
1138 * userspace controlling the datapath. */
1140 dp_enumerate_names(type, &names);
1141 SSET_FOR_EACH(name, &names) {
1142 struct dpif *old_dpif;
1144 /* Don't remove our backer if it exists. */
1145 if (!strcmp(name, backer_name)) {
1149 if (dpif_open(name, type, &old_dpif)) {
1150 VLOG_WARN("couldn't open old datapath %s to remove it", name);
1152 dpif_delete(old_dpif);
1153 dpif_close(old_dpif);
1156 sset_destroy(&names);
1158 backer = xmalloc(sizeof *backer);
1160 error = dpif_create_and_open(backer_name, type, &backer->dpif);
1163 VLOG_ERR("failed to open datapath of type %s: %s", type,
1169 backer->type = xstrdup(type);
1170 backer->refcount = 1;
1171 hmap_init(&backer->odp_to_ofport_map);
1172 hmap_init(&backer->drop_keys);
1173 timer_set_duration(&backer->next_expiration, 1000);
1174 backer->need_revalidate = 0;
1175 simap_init(&backer->tnl_backers);
1176 tag_set_init(&backer->revalidate_set);
1179 dpif_flow_flush(backer->dpif);
1181 /* Loop through the ports already on the datapath and remove any
1182 * that we don't need anymore. */
1183 list_init(&garbage_list);
1184 dpif_port_dump_start(&port_dump, backer->dpif);
1185 while (dpif_port_dump_next(&port_dump, &port)) {
1186 node = shash_find(&init_ofp_ports, port.name);
1187 if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
1188 garbage = xmalloc(sizeof *garbage);
1189 garbage->odp_port = port.port_no;
1190 list_push_front(&garbage_list, &garbage->list_node);
1193 dpif_port_dump_done(&port_dump);
1195 LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) {
1196 dpif_port_del(backer->dpif, garbage->odp_port);
1197 list_remove(&garbage->list_node);
1201 shash_add(&all_dpif_backers, type, backer);
1203 error = dpif_recv_set(backer->dpif, true);
1205 VLOG_ERR("failed to listen on datapath of type %s: %s",
1206 type, strerror(error));
1207 close_dpif_backer(backer);
1215 construct(struct ofproto *ofproto_)
1217 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1218 struct shash_node *node, *next;
1223 error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1228 max_ports = dpif_get_max_ports(ofproto->backer->dpif);
1229 ofproto_init_max_ports(ofproto_, MIN(max_ports, OFPP_MAX));
1231 ofproto->n_matches = 0;
1233 ofproto->netflow = NULL;
1234 ofproto->sflow = NULL;
1235 ofproto->stp = NULL;
1236 hmap_init(&ofproto->bundles);
1237 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1238 for (i = 0; i < MAX_MIRRORS; i++) {
1239 ofproto->mirrors[i] = NULL;
1241 ofproto->has_bonded_bundles = false;
1243 hmap_init(&ofproto->facets);
1244 hmap_init(&ofproto->subfacets);
1245 ofproto->governor = NULL;
1247 for (i = 0; i < N_TABLES; i++) {
1248 struct table_dpif *table = &ofproto->tables[i];
1250 table->catchall_table = NULL;
1251 table->other_table = NULL;
1252 table->basis = random_uint32();
1255 list_init(&ofproto->completions);
1257 ofproto_dpif_unixctl_init();
1259 ofproto->has_mirrors = false;
1260 ofproto->has_bundle_action = false;
1262 hmap_init(&ofproto->vlandev_map);
1263 hmap_init(&ofproto->realdev_vid_map);
1265 sset_init(&ofproto->ports);
1266 sset_init(&ofproto->ghost_ports);
1267 sset_init(&ofproto->port_poll_set);
1268 ofproto->port_poll_errno = 0;
1270 SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1271 struct iface_hint *iface_hint = node->data;
1273 if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1274 /* Check if the datapath already has this port. */
1275 if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1276 sset_add(&ofproto->ports, node->name);
1279 free(iface_hint->br_name);
1280 free(iface_hint->br_type);
1282 shash_delete(&init_ofp_ports, node);
1286 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1287 hash_string(ofproto->up.name, 0));
1288 memset(&ofproto->stats, 0, sizeof ofproto->stats);
1290 ofproto_init_tables(ofproto_, N_TABLES);
1291 error = add_internal_flows(ofproto);
1292 ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1298 add_internal_flow(struct ofproto_dpif *ofproto, int id,
1299 const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1301 struct ofputil_flow_mod fm;
1304 match_init_catchall(&fm.match);
1306 match_set_reg(&fm.match, 0, id);
1307 fm.new_cookie = htonll(0);
1308 fm.cookie = htonll(0);
1309 fm.cookie_mask = htonll(0);
1310 fm.table_id = TBL_INTERNAL;
1311 fm.command = OFPFC_ADD;
1312 fm.idle_timeout = 0;
1313 fm.hard_timeout = 0;
1317 fm.ofpacts = ofpacts->data;
1318 fm.ofpacts_len = ofpacts->size;
1320 error = ofproto_flow_mod(&ofproto->up, &fm);
1322 VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
1323 id, ofperr_to_string(error));
1327 *rulep = rule_dpif_lookup__(ofproto, &fm.match.flow, TBL_INTERNAL);
1328 ovs_assert(*rulep != NULL);
1334 add_internal_flows(struct ofproto_dpif *ofproto)
1336 struct ofpact_controller *controller;
1337 uint64_t ofpacts_stub[128 / 8];
1338 struct ofpbuf ofpacts;
1342 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1345 controller = ofpact_put_CONTROLLER(&ofpacts);
1346 controller->max_len = UINT16_MAX;
1347 controller->controller_id = 0;
1348 controller->reason = OFPR_NO_MATCH;
1349 ofpact_pad(&ofpacts);
1351 error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule);
1356 ofpbuf_clear(&ofpacts);
1357 error = add_internal_flow(ofproto, id++, &ofpacts,
1358 &ofproto->no_packet_in_rule);
1363 complete_operations(struct ofproto_dpif *ofproto)
1365 struct dpif_completion *c, *next;
1367 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
1368 ofoperation_complete(c->op, 0);
1369 list_remove(&c->list_node);
1375 destruct(struct ofproto *ofproto_)
1377 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1378 struct rule_dpif *rule, *next_rule;
1379 struct oftable *table;
1382 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1383 complete_operations(ofproto);
1385 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1386 struct cls_cursor cursor;
1388 cls_cursor_init(&cursor, &table->cls, NULL);
1389 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1390 ofproto_rule_destroy(&rule->up);
1394 for (i = 0; i < MAX_MIRRORS; i++) {
1395 mirror_destroy(ofproto->mirrors[i]);
1398 netflow_destroy(ofproto->netflow);
1399 dpif_sflow_destroy(ofproto->sflow);
1400 hmap_destroy(&ofproto->bundles);
1401 mac_learning_destroy(ofproto->ml);
1403 hmap_destroy(&ofproto->facets);
1404 hmap_destroy(&ofproto->subfacets);
1405 governor_destroy(ofproto->governor);
1407 hmap_destroy(&ofproto->vlandev_map);
1408 hmap_destroy(&ofproto->realdev_vid_map);
1410 sset_destroy(&ofproto->ports);
1411 sset_destroy(&ofproto->ghost_ports);
1412 sset_destroy(&ofproto->port_poll_set);
1414 close_dpif_backer(ofproto->backer);
1418 run_fast(struct ofproto *ofproto_)
1420 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1421 struct ofport_dpif *ofport;
1423 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1424 port_run_fast(ofport);
1431 run(struct ofproto *ofproto_)
1433 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1434 struct ofport_dpif *ofport;
1435 struct ofbundle *bundle;
1439 complete_operations(ofproto);
1442 error = run_fast(ofproto_);
1447 if (ofproto->netflow) {
1448 if (netflow_run(ofproto->netflow)) {
1449 send_netflow_active_timeouts(ofproto);
1452 if (ofproto->sflow) {
1453 dpif_sflow_run(ofproto->sflow);
1456 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1459 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1464 mac_learning_run(ofproto->ml, &ofproto->backer->revalidate_set);
1466 /* Check the consistency of a random facet, to aid debugging. */
1467 if (!hmap_is_empty(&ofproto->facets)
1468 && !ofproto->backer->need_revalidate) {
1469 struct facet *facet;
1471 facet = CONTAINER_OF(hmap_random_node(&ofproto->facets),
1472 struct facet, hmap_node);
1473 if (!tag_set_intersects(&ofproto->backer->revalidate_set,
1475 if (!facet_check_consistency(facet)) {
1476 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
1481 if (ofproto->governor) {
1484 governor_run(ofproto->governor);
1486 /* If the governor has shrunk to its minimum size and the number of
1487 * subfacets has dwindled, then drop the governor entirely.
1489 * For hysteresis, the number of subfacets to drop the governor is
1490 * smaller than the number needed to trigger its creation. */
1491 n_subfacets = hmap_count(&ofproto->subfacets);
1492 if (n_subfacets * 4 < ofproto->up.flow_eviction_threshold
1493 && governor_is_idle(ofproto->governor)) {
1494 governor_destroy(ofproto->governor);
1495 ofproto->governor = NULL;
1503 wait(struct ofproto *ofproto_)
1505 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1506 struct ofport_dpif *ofport;
1507 struct ofbundle *bundle;
1509 if (!clogged && !list_is_empty(&ofproto->completions)) {
1510 poll_immediate_wake();
1513 dpif_wait(ofproto->backer->dpif);
1514 dpif_recv_wait(ofproto->backer->dpif);
1515 if (ofproto->sflow) {
1516 dpif_sflow_wait(ofproto->sflow);
1518 if (!tag_set_is_empty(&ofproto->backer->revalidate_set)) {
1519 poll_immediate_wake();
1521 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1524 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1525 bundle_wait(bundle);
1527 if (ofproto->netflow) {
1528 netflow_wait(ofproto->netflow);
1530 mac_learning_wait(ofproto->ml);
1532 if (ofproto->backer->need_revalidate) {
1533 /* Shouldn't happen, but if it does just go around again. */
1534 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1535 poll_immediate_wake();
1537 if (ofproto->governor) {
1538 governor_wait(ofproto->governor);
1543 get_memory_usage(const struct ofproto *ofproto_, struct simap *usage)
1545 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1547 simap_increase(usage, "facets", hmap_count(&ofproto->facets));
1548 simap_increase(usage, "subfacets", hmap_count(&ofproto->subfacets));
1552 flush(struct ofproto *ofproto_)
1554 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1555 struct subfacet *subfacet, *next_subfacet;
1556 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
1560 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
1561 &ofproto->subfacets) {
1562 if (subfacet->path != SF_NOT_INSTALLED) {
1563 batch[n_batch++] = subfacet;
1564 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
1565 subfacet_destroy_batch(ofproto, batch, n_batch);
1569 subfacet_destroy(subfacet);
1574 subfacet_destroy_batch(ofproto, batch, n_batch);
1579 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1580 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1582 *arp_match_ip = true;
1583 *actions = (OFPUTIL_A_OUTPUT |
1584 OFPUTIL_A_SET_VLAN_VID |
1585 OFPUTIL_A_SET_VLAN_PCP |
1586 OFPUTIL_A_STRIP_VLAN |
1587 OFPUTIL_A_SET_DL_SRC |
1588 OFPUTIL_A_SET_DL_DST |
1589 OFPUTIL_A_SET_NW_SRC |
1590 OFPUTIL_A_SET_NW_DST |
1591 OFPUTIL_A_SET_NW_TOS |
1592 OFPUTIL_A_SET_TP_SRC |
1593 OFPUTIL_A_SET_TP_DST |
1598 get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots)
1600 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1601 struct dpif_dp_stats s;
1603 strcpy(ots->name, "classifier");
1605 dpif_get_dp_stats(ofproto->backer->dpif, &s);
1607 ots->lookup_count = htonll(s.n_hit + s.n_missed);
1608 ots->matched_count = htonll(s.n_hit + ofproto->n_matches);
1611 static struct ofport *
1614 struct ofport_dpif *port = xmalloc(sizeof *port);
1619 port_dealloc(struct ofport *port_)
1621 struct ofport_dpif *port = ofport_dpif_cast(port_);
1626 port_construct(struct ofport *port_)
1628 struct ofport_dpif *port = ofport_dpif_cast(port_);
1629 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1630 const struct netdev *netdev = port->up.netdev;
1631 struct dpif_port dpif_port;
1634 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1635 port->bundle = NULL;
1637 port->tag = tag_create_random();
1638 port->may_enable = true;
1639 port->stp_port = NULL;
1640 port->stp_state = STP_DISABLED;
1641 port->tnl_port = NULL;
1642 hmap_init(&port->priorities);
1643 port->realdev_ofp_port = 0;
1644 port->vlandev_vid = 0;
1645 port->carrier_seq = netdev_get_carrier_resets(netdev);
1647 if (netdev_vport_is_patch(netdev)) {
1648 /* XXX By bailing out here, we don't do required sFlow work. */
1649 port->odp_port = OVSP_NONE;
1653 error = dpif_port_query_by_name(ofproto->backer->dpif,
1654 netdev_vport_get_dpif_port(netdev),
1660 port->odp_port = dpif_port.port_no;
1662 if (netdev_get_tunnel_config(netdev)) {
1663 port->tnl_port = tnl_port_add(&port->up, port->odp_port);
1665 /* Sanity-check that a mapping doesn't already exist. This
1666 * shouldn't happen for non-tunnel ports. */
1667 if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1668 VLOG_ERR("port %s already has an OpenFlow port number",
1670 dpif_port_destroy(&dpif_port);
1674 hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1675 hash_int(port->odp_port, 0));
1677 dpif_port_destroy(&dpif_port);
1679 if (ofproto->sflow) {
1680 dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1687 port_destruct(struct ofport *port_)
1689 struct ofport_dpif *port = ofport_dpif_cast(port_);
1690 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1691 const char *dp_port_name = netdev_vport_get_dpif_port(port->up.netdev);
1692 const char *devname = netdev_get_name(port->up.netdev);
1694 if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1695 /* The underlying device is still there, so delete it. This
1696 * happens when the ofproto is being destroyed, since the caller
1697 * assumes that removal of attached ports will happen as part of
1699 if (!port->tnl_port) {
1700 dpif_port_del(ofproto->backer->dpif, port->odp_port);
1702 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1705 if (port->odp_port != OVSP_NONE && !port->tnl_port) {
1706 hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1709 tnl_port_del(port->tnl_port);
1710 sset_find_and_delete(&ofproto->ports, devname);
1711 sset_find_and_delete(&ofproto->ghost_ports, devname);
1712 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1713 bundle_remove(port_);
1714 set_cfm(port_, NULL);
1715 if (ofproto->sflow) {
1716 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1719 ofport_clear_priorities(port);
1720 hmap_destroy(&port->priorities);
1724 port_modified(struct ofport *port_)
1726 struct ofport_dpif *port = ofport_dpif_cast(port_);
1728 if (port->bundle && port->bundle->bond) {
1729 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1734 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1736 struct ofport_dpif *port = ofport_dpif_cast(port_);
1737 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1738 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1740 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1741 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1742 OFPUTIL_PC_NO_PACKET_IN)) {
1743 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1745 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1746 bundle_update(port->bundle);
1752 set_sflow(struct ofproto *ofproto_,
1753 const struct ofproto_sflow_options *sflow_options)
1755 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1756 struct dpif_sflow *ds = ofproto->sflow;
1758 if (sflow_options) {
1760 struct ofport_dpif *ofport;
1762 ds = ofproto->sflow = dpif_sflow_create();
1763 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1764 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1766 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1768 dpif_sflow_set_options(ds, sflow_options);
1771 dpif_sflow_destroy(ds);
1772 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1773 ofproto->sflow = NULL;
1780 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1782 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1789 struct ofproto_dpif *ofproto;
1791 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1792 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1793 ofport->cfm = cfm_create(netdev_get_name(ofport->up.netdev));
1796 if (cfm_configure(ofport->cfm, s)) {
1802 cfm_destroy(ofport->cfm);
1808 get_cfm_status(const struct ofport *ofport_,
1809 struct ofproto_cfm_status *status)
1811 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1814 status->faults = cfm_get_fault(ofport->cfm);
1815 status->remote_opstate = cfm_get_opup(ofport->cfm);
1816 status->health = cfm_get_health(ofport->cfm);
1817 cfm_get_remote_mpids(ofport->cfm, &status->rmps, &status->n_rmps);
1824 /* Spanning Tree. */
1827 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
1829 struct ofproto_dpif *ofproto = ofproto_;
1830 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
1831 struct ofport_dpif *ofport;
1833 ofport = stp_port_get_aux(sp);
1835 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
1836 ofproto->up.name, port_num);
1838 struct eth_header *eth = pkt->l2;
1840 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
1841 if (eth_addr_is_zero(eth->eth_src)) {
1842 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
1843 "with unknown MAC", ofproto->up.name, port_num);
1845 send_packet(ofport, pkt);
1851 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
1853 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
1855 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1857 /* Only revalidate flows if the configuration changed. */
1858 if (!s != !ofproto->stp) {
1859 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1863 if (!ofproto->stp) {
1864 ofproto->stp = stp_create(ofproto_->name, s->system_id,
1865 send_bpdu_cb, ofproto);
1866 ofproto->stp_last_tick = time_msec();
1869 stp_set_bridge_id(ofproto->stp, s->system_id);
1870 stp_set_bridge_priority(ofproto->stp, s->priority);
1871 stp_set_hello_time(ofproto->stp, s->hello_time);
1872 stp_set_max_age(ofproto->stp, s->max_age);
1873 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
1875 struct ofport *ofport;
1877 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
1878 set_stp_port(ofport, NULL);
1881 stp_destroy(ofproto->stp);
1882 ofproto->stp = NULL;
1889 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
1891 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1895 s->bridge_id = stp_get_bridge_id(ofproto->stp);
1896 s->designated_root = stp_get_designated_root(ofproto->stp);
1897 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
1906 update_stp_port_state(struct ofport_dpif *ofport)
1908 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1909 enum stp_state state;
1911 /* Figure out new state. */
1912 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
1916 if (ofport->stp_state != state) {
1917 enum ofputil_port_state of_state;
1920 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
1921 netdev_get_name(ofport->up.netdev),
1922 stp_state_name(ofport->stp_state),
1923 stp_state_name(state));
1924 if (stp_learn_in_state(ofport->stp_state)
1925 != stp_learn_in_state(state)) {
1926 /* xxx Learning action flows should also be flushed. */
1927 mac_learning_flush(ofproto->ml,
1928 &ofproto->backer->revalidate_set);
1930 fwd_change = stp_forward_in_state(ofport->stp_state)
1931 != stp_forward_in_state(state);
1933 ofproto->backer->need_revalidate = REV_STP;
1934 ofport->stp_state = state;
1935 ofport->stp_state_entered = time_msec();
1937 if (fwd_change && ofport->bundle) {
1938 bundle_update(ofport->bundle);
1941 /* Update the STP state bits in the OpenFlow port description. */
1942 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
1943 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
1944 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
1945 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
1946 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
1948 ofproto_port_set_state(&ofport->up, of_state);
1952 /* Configures STP on 'ofport_' using the settings defined in 's'. The
1953 * caller is responsible for assigning STP port numbers and ensuring
1954 * there are no duplicates. */
1956 set_stp_port(struct ofport *ofport_,
1957 const struct ofproto_port_stp_settings *s)
1959 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1960 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1961 struct stp_port *sp = ofport->stp_port;
1963 if (!s || !s->enable) {
1965 ofport->stp_port = NULL;
1966 stp_port_disable(sp);
1967 update_stp_port_state(ofport);
1970 } else if (sp && stp_port_no(sp) != s->port_num
1971 && ofport == stp_port_get_aux(sp)) {
1972 /* The port-id changed, so disable the old one if it's not
1973 * already in use by another port. */
1974 stp_port_disable(sp);
1977 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
1978 stp_port_enable(sp);
1980 stp_port_set_aux(sp, ofport);
1981 stp_port_set_priority(sp, s->priority);
1982 stp_port_set_path_cost(sp, s->path_cost);
1984 update_stp_port_state(ofport);
1990 get_stp_port_status(struct ofport *ofport_,
1991 struct ofproto_port_stp_status *s)
1993 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1994 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1995 struct stp_port *sp = ofport->stp_port;
1997 if (!ofproto->stp || !sp) {
2003 s->port_id = stp_port_get_id(sp);
2004 s->state = stp_port_get_state(sp);
2005 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
2006 s->role = stp_port_get_role(sp);
2007 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
2013 stp_run(struct ofproto_dpif *ofproto)
2016 long long int now = time_msec();
2017 long long int elapsed = now - ofproto->stp_last_tick;
2018 struct stp_port *sp;
2021 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
2022 ofproto->stp_last_tick = now;
2024 while (stp_get_changed_port(ofproto->stp, &sp)) {
2025 struct ofport_dpif *ofport = stp_port_get_aux(sp);
2028 update_stp_port_state(ofport);
2032 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
2033 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
2039 stp_wait(struct ofproto_dpif *ofproto)
2042 poll_timer_wait(1000);
2046 /* Returns true if STP should process 'flow'. */
2048 stp_should_process_flow(const struct flow *flow)
2050 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
2054 stp_process_packet(const struct ofport_dpif *ofport,
2055 const struct ofpbuf *packet)
2057 struct ofpbuf payload = *packet;
2058 struct eth_header *eth = payload.data;
2059 struct stp_port *sp = ofport->stp_port;
2061 /* Sink packets on ports that have STP disabled when the bridge has
2063 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
2067 /* Trim off padding on payload. */
2068 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
2069 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
2072 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
2073 stp_received_bpdu(sp, payload.data, payload.size);
2077 static struct priority_to_dscp *
2078 get_priority(const struct ofport_dpif *ofport, uint32_t priority)
2080 struct priority_to_dscp *pdscp;
2083 hash = hash_int(priority, 0);
2084 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) {
2085 if (pdscp->priority == priority) {
2093 ofport_clear_priorities(struct ofport_dpif *ofport)
2095 struct priority_to_dscp *pdscp, *next;
2097 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) {
2098 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
2104 set_queues(struct ofport *ofport_,
2105 const struct ofproto_port_queue *qdscp_list,
2108 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2109 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2110 struct hmap new = HMAP_INITIALIZER(&new);
2113 for (i = 0; i < n_qdscp; i++) {
2114 struct priority_to_dscp *pdscp;
2118 dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
2119 if (dpif_queue_to_priority(ofproto->backer->dpif, qdscp_list[i].queue,
2124 pdscp = get_priority(ofport, priority);
2126 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
2128 pdscp = xmalloc(sizeof *pdscp);
2129 pdscp->priority = priority;
2131 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2134 if (pdscp->dscp != dscp) {
2136 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2139 hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0));
2142 if (!hmap_is_empty(&ofport->priorities)) {
2143 ofport_clear_priorities(ofport);
2144 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2147 hmap_swap(&new, &ofport->priorities);
2155 /* Expires all MAC learning entries associated with 'bundle' and forces its
2156 * ofproto to revalidate every flow.
2158 * Normally MAC learning entries are removed only from the ofproto associated
2159 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2160 * are removed from every ofproto. When patch ports and SLB bonds are in use
2161 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2162 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2163 * with the host from which it migrated. */
2165 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2167 struct ofproto_dpif *ofproto = bundle->ofproto;
2168 struct mac_learning *ml = ofproto->ml;
2169 struct mac_entry *mac, *next_mac;
2171 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2172 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2173 if (mac->port.p == bundle) {
2175 struct ofproto_dpif *o;
2177 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2179 struct mac_entry *e;
2181 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan,
2184 mac_learning_expire(o->ml, e);
2190 mac_learning_expire(ml, mac);
2195 static struct ofbundle *
2196 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2198 struct ofbundle *bundle;
2200 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2201 &ofproto->bundles) {
2202 if (bundle->aux == aux) {
2209 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
2210 * ones that are found to 'bundles'. */
2212 bundle_lookup_multiple(struct ofproto_dpif *ofproto,
2213 void **auxes, size_t n_auxes,
2214 struct hmapx *bundles)
2218 hmapx_init(bundles);
2219 for (i = 0; i < n_auxes; i++) {
2220 struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
2222 hmapx_add(bundles, bundle);
2228 bundle_update(struct ofbundle *bundle)
2230 struct ofport_dpif *port;
2232 bundle->floodable = true;
2233 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2234 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2235 || !stp_forward_in_state(port->stp_state)) {
2236 bundle->floodable = false;
2243 bundle_del_port(struct ofport_dpif *port)
2245 struct ofbundle *bundle = port->bundle;
2247 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2249 list_remove(&port->bundle_node);
2250 port->bundle = NULL;
2253 lacp_slave_unregister(bundle->lacp, port);
2256 bond_slave_unregister(bundle->bond, port);
2259 bundle_update(bundle);
2263 bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port,
2264 struct lacp_slave_settings *lacp)
2266 struct ofport_dpif *port;
2268 port = get_ofp_port(bundle->ofproto, ofp_port);
2273 if (port->bundle != bundle) {
2274 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2276 bundle_del_port(port);
2279 port->bundle = bundle;
2280 list_push_back(&bundle->ports, &port->bundle_node);
2281 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2282 || !stp_forward_in_state(port->stp_state)) {
2283 bundle->floodable = false;
2287 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2288 lacp_slave_register(bundle->lacp, port, lacp);
2295 bundle_destroy(struct ofbundle *bundle)
2297 struct ofproto_dpif *ofproto;
2298 struct ofport_dpif *port, *next_port;
2305 ofproto = bundle->ofproto;
2306 for (i = 0; i < MAX_MIRRORS; i++) {
2307 struct ofmirror *m = ofproto->mirrors[i];
2309 if (m->out == bundle) {
2311 } else if (hmapx_find_and_delete(&m->srcs, bundle)
2312 || hmapx_find_and_delete(&m->dsts, bundle)) {
2313 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2318 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2319 bundle_del_port(port);
2322 bundle_flush_macs(bundle, true);
2323 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2325 free(bundle->trunks);
2326 lacp_destroy(bundle->lacp);
2327 bond_destroy(bundle->bond);
2332 bundle_set(struct ofproto *ofproto_, void *aux,
2333 const struct ofproto_bundle_settings *s)
2335 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2336 bool need_flush = false;
2337 struct ofport_dpif *port;
2338 struct ofbundle *bundle;
2339 unsigned long *trunks;
2345 bundle_destroy(bundle_lookup(ofproto, aux));
2349 ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2350 ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2352 bundle = bundle_lookup(ofproto, aux);
2354 bundle = xmalloc(sizeof *bundle);
2356 bundle->ofproto = ofproto;
2357 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2358 hash_pointer(aux, 0));
2360 bundle->name = NULL;
2362 list_init(&bundle->ports);
2363 bundle->vlan_mode = PORT_VLAN_TRUNK;
2365 bundle->trunks = NULL;
2366 bundle->use_priority_tags = s->use_priority_tags;
2367 bundle->lacp = NULL;
2368 bundle->bond = NULL;
2370 bundle->floodable = true;
2372 bundle->src_mirrors = 0;
2373 bundle->dst_mirrors = 0;
2374 bundle->mirror_out = 0;
2377 if (!bundle->name || strcmp(s->name, bundle->name)) {
2379 bundle->name = xstrdup(s->name);
2384 if (!bundle->lacp) {
2385 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2386 bundle->lacp = lacp_create();
2388 lacp_configure(bundle->lacp, s->lacp);
2390 lacp_destroy(bundle->lacp);
2391 bundle->lacp = NULL;
2394 /* Update set of ports. */
2396 for (i = 0; i < s->n_slaves; i++) {
2397 if (!bundle_add_port(bundle, s->slaves[i],
2398 s->lacp ? &s->lacp_slaves[i] : NULL)) {
2402 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2403 struct ofport_dpif *next_port;
2405 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2406 for (i = 0; i < s->n_slaves; i++) {
2407 if (s->slaves[i] == port->up.ofp_port) {
2412 bundle_del_port(port);
2416 ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2418 if (list_is_empty(&bundle->ports)) {
2419 bundle_destroy(bundle);
2423 /* Set VLAN tagging mode */
2424 if (s->vlan_mode != bundle->vlan_mode
2425 || s->use_priority_tags != bundle->use_priority_tags) {
2426 bundle->vlan_mode = s->vlan_mode;
2427 bundle->use_priority_tags = s->use_priority_tags;
2432 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2433 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2435 if (vlan != bundle->vlan) {
2436 bundle->vlan = vlan;
2440 /* Get trunked VLANs. */
2441 switch (s->vlan_mode) {
2442 case PORT_VLAN_ACCESS:
2446 case PORT_VLAN_TRUNK:
2447 trunks = CONST_CAST(unsigned long *, s->trunks);
2450 case PORT_VLAN_NATIVE_UNTAGGED:
2451 case PORT_VLAN_NATIVE_TAGGED:
2452 if (vlan != 0 && (!s->trunks
2453 || !bitmap_is_set(s->trunks, vlan)
2454 || bitmap_is_set(s->trunks, 0))) {
2455 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2457 trunks = bitmap_clone(s->trunks, 4096);
2459 trunks = bitmap_allocate1(4096);
2461 bitmap_set1(trunks, vlan);
2462 bitmap_set0(trunks, 0);
2464 trunks = CONST_CAST(unsigned long *, s->trunks);
2471 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2472 free(bundle->trunks);
2473 if (trunks == s->trunks) {
2474 bundle->trunks = vlan_bitmap_clone(trunks);
2476 bundle->trunks = trunks;
2481 if (trunks != s->trunks) {
2486 if (!list_is_short(&bundle->ports)) {
2487 bundle->ofproto->has_bonded_bundles = true;
2489 if (bond_reconfigure(bundle->bond, s->bond)) {
2490 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2493 bundle->bond = bond_create(s->bond);
2494 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2497 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2498 bond_slave_register(bundle->bond, port, port->up.netdev);
2501 bond_destroy(bundle->bond);
2502 bundle->bond = NULL;
2505 /* If we changed something that would affect MAC learning, un-learn
2506 * everything on this port and force flow revalidation. */
2508 bundle_flush_macs(bundle, false);
2515 bundle_remove(struct ofport *port_)
2517 struct ofport_dpif *port = ofport_dpif_cast(port_);
2518 struct ofbundle *bundle = port->bundle;
2521 bundle_del_port(port);
2522 if (list_is_empty(&bundle->ports)) {
2523 bundle_destroy(bundle);
2524 } else if (list_is_short(&bundle->ports)) {
2525 bond_destroy(bundle->bond);
2526 bundle->bond = NULL;
2532 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2534 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2535 struct ofport_dpif *port = port_;
2536 uint8_t ea[ETH_ADDR_LEN];
2539 error = netdev_get_etheraddr(port->up.netdev, ea);
2541 struct ofpbuf packet;
2544 ofpbuf_init(&packet, 0);
2545 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2547 memcpy(packet_pdu, pdu, pdu_size);
2549 send_packet(port, &packet);
2550 ofpbuf_uninit(&packet);
2552 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2553 "%s (%s)", port->bundle->name,
2554 netdev_get_name(port->up.netdev), strerror(error));
2559 bundle_send_learning_packets(struct ofbundle *bundle)
2561 struct ofproto_dpif *ofproto = bundle->ofproto;
2562 int error, n_packets, n_errors;
2563 struct mac_entry *e;
2565 error = n_packets = n_errors = 0;
2566 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2567 if (e->port.p != bundle) {
2568 struct ofpbuf *learning_packet;
2569 struct ofport_dpif *port;
2573 /* The assignment to "port" is unnecessary but makes "grep"ing for
2574 * struct ofport_dpif more effective. */
2575 learning_packet = bond_compose_learning_packet(bundle->bond,
2579 ret = send_packet(port, learning_packet);
2580 ofpbuf_delete(learning_packet);
2590 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2591 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2592 "packets, last error was: %s",
2593 bundle->name, n_errors, n_packets, strerror(error));
2595 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2596 bundle->name, n_packets);
2601 bundle_run(struct ofbundle *bundle)
2604 lacp_run(bundle->lacp, send_pdu_cb);
2607 struct ofport_dpif *port;
2609 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2610 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2613 bond_run(bundle->bond, &bundle->ofproto->backer->revalidate_set,
2614 lacp_status(bundle->lacp));
2615 if (bond_should_send_learning_packets(bundle->bond)) {
2616 bundle_send_learning_packets(bundle);
2622 bundle_wait(struct ofbundle *bundle)
2625 lacp_wait(bundle->lacp);
2628 bond_wait(bundle->bond);
2635 mirror_scan(struct ofproto_dpif *ofproto)
2639 for (idx = 0; idx < MAX_MIRRORS; idx++) {
2640 if (!ofproto->mirrors[idx]) {
2647 static struct ofmirror *
2648 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
2652 for (i = 0; i < MAX_MIRRORS; i++) {
2653 struct ofmirror *mirror = ofproto->mirrors[i];
2654 if (mirror && mirror->aux == aux) {
2662 /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */
2664 mirror_update_dups(struct ofproto_dpif *ofproto)
2668 for (i = 0; i < MAX_MIRRORS; i++) {
2669 struct ofmirror *m = ofproto->mirrors[i];
2672 m->dup_mirrors = MIRROR_MASK_C(1) << i;
2676 for (i = 0; i < MAX_MIRRORS; i++) {
2677 struct ofmirror *m1 = ofproto->mirrors[i];
2684 for (j = i + 1; j < MAX_MIRRORS; j++) {
2685 struct ofmirror *m2 = ofproto->mirrors[j];
2687 if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) {
2688 m1->dup_mirrors |= MIRROR_MASK_C(1) << j;
2689 m2->dup_mirrors |= m1->dup_mirrors;
2696 mirror_set(struct ofproto *ofproto_, void *aux,
2697 const struct ofproto_mirror_settings *s)
2699 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2700 mirror_mask_t mirror_bit;
2701 struct ofbundle *bundle;
2702 struct ofmirror *mirror;
2703 struct ofbundle *out;
2704 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
2705 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
2708 mirror = mirror_lookup(ofproto, aux);
2710 mirror_destroy(mirror);
2716 idx = mirror_scan(ofproto);
2718 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
2720 ofproto->up.name, MAX_MIRRORS, s->name);
2724 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
2725 mirror->ofproto = ofproto;
2728 mirror->out_vlan = -1;
2729 mirror->name = NULL;
2732 if (!mirror->name || strcmp(s->name, mirror->name)) {
2734 mirror->name = xstrdup(s->name);
2737 /* Get the new configuration. */
2738 if (s->out_bundle) {
2739 out = bundle_lookup(ofproto, s->out_bundle);
2741 mirror_destroy(mirror);
2747 out_vlan = s->out_vlan;
2749 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
2750 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
2752 /* If the configuration has not changed, do nothing. */
2753 if (hmapx_equals(&srcs, &mirror->srcs)
2754 && hmapx_equals(&dsts, &mirror->dsts)
2755 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
2756 && mirror->out == out
2757 && mirror->out_vlan == out_vlan)
2759 hmapx_destroy(&srcs);
2760 hmapx_destroy(&dsts);
2764 hmapx_swap(&srcs, &mirror->srcs);
2765 hmapx_destroy(&srcs);
2767 hmapx_swap(&dsts, &mirror->dsts);
2768 hmapx_destroy(&dsts);
2770 free(mirror->vlans);
2771 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
2774 mirror->out_vlan = out_vlan;
2776 /* Update bundles. */
2777 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2778 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
2779 if (hmapx_contains(&mirror->srcs, bundle)) {
2780 bundle->src_mirrors |= mirror_bit;
2782 bundle->src_mirrors &= ~mirror_bit;
2785 if (hmapx_contains(&mirror->dsts, bundle)) {
2786 bundle->dst_mirrors |= mirror_bit;
2788 bundle->dst_mirrors &= ~mirror_bit;
2791 if (mirror->out == bundle) {
2792 bundle->mirror_out |= mirror_bit;
2794 bundle->mirror_out &= ~mirror_bit;
2798 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2799 ofproto->has_mirrors = true;
2800 mac_learning_flush(ofproto->ml,
2801 &ofproto->backer->revalidate_set);
2802 mirror_update_dups(ofproto);
2808 mirror_destroy(struct ofmirror *mirror)
2810 struct ofproto_dpif *ofproto;
2811 mirror_mask_t mirror_bit;
2812 struct ofbundle *bundle;
2819 ofproto = mirror->ofproto;
2820 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2821 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
2823 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2824 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
2825 bundle->src_mirrors &= ~mirror_bit;
2826 bundle->dst_mirrors &= ~mirror_bit;
2827 bundle->mirror_out &= ~mirror_bit;
2830 hmapx_destroy(&mirror->srcs);
2831 hmapx_destroy(&mirror->dsts);
2832 free(mirror->vlans);
2834 ofproto->mirrors[mirror->idx] = NULL;
2838 mirror_update_dups(ofproto);
2840 ofproto->has_mirrors = false;
2841 for (i = 0; i < MAX_MIRRORS; i++) {
2842 if (ofproto->mirrors[i]) {
2843 ofproto->has_mirrors = true;
2850 mirror_get_stats(struct ofproto *ofproto_, void *aux,
2851 uint64_t *packets, uint64_t *bytes)
2853 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2854 struct ofmirror *mirror = mirror_lookup(ofproto, aux);
2857 *packets = *bytes = UINT64_MAX;
2861 *packets = mirror->packet_count;
2862 *bytes = mirror->byte_count;
2868 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2870 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2871 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2872 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
2878 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2880 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2881 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2882 return bundle && bundle->mirror_out != 0;
2886 forward_bpdu_changed(struct ofproto *ofproto_)
2888 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2889 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2893 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
2896 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2897 mac_learning_set_idle_time(ofproto->ml, idle_time);
2898 mac_learning_set_max_entries(ofproto->ml, max_entries);
2903 static struct ofport_dpif *
2904 get_ofp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
2906 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2907 return ofport ? ofport_dpif_cast(ofport) : NULL;
2910 static struct ofport_dpif *
2911 get_odp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
2913 struct ofport_dpif *port = odp_port_to_ofport(ofproto->backer, odp_port);
2914 return port && &ofproto->up == port->up.ofproto ? port : NULL;
2918 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
2919 struct ofproto_port *ofproto_port,
2920 struct dpif_port *dpif_port)
2922 ofproto_port->name = dpif_port->name;
2923 ofproto_port->type = dpif_port->type;
2924 ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
2927 static struct ofport_dpif *
2928 ofport_get_peer(const struct ofport_dpif *ofport_dpif)
2930 const struct ofproto_dpif *ofproto;
2933 peer = netdev_vport_patch_peer(ofport_dpif->up.netdev);
2938 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2939 struct ofport *ofport;
2941 ofport = shash_find_data(&ofproto->up.port_by_name, peer);
2942 if (ofport && ofport->ofproto->ofproto_class == &ofproto_dpif_class) {
2943 return ofport_dpif_cast(ofport);
2950 port_run_fast(struct ofport_dpif *ofport)
2952 if (ofport->cfm && cfm_should_send_ccm(ofport->cfm)) {
2953 struct ofpbuf packet;
2955 ofpbuf_init(&packet, 0);
2956 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
2957 send_packet(ofport, &packet);
2958 ofpbuf_uninit(&packet);
2963 port_run(struct ofport_dpif *ofport)
2965 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
2966 bool carrier_changed = carrier_seq != ofport->carrier_seq;
2967 bool enable = netdev_get_carrier(ofport->up.netdev);
2969 ofport->carrier_seq = carrier_seq;
2971 port_run_fast(ofport);
2973 if (ofport->tnl_port
2974 && tnl_port_reconfigure(&ofport->up, ofport->odp_port,
2975 &ofport->tnl_port)) {
2976 ofproto_dpif_cast(ofport->up.ofproto)->backer->need_revalidate = true;
2980 int cfm_opup = cfm_get_opup(ofport->cfm);
2982 cfm_run(ofport->cfm);
2983 enable = enable && !cfm_get_fault(ofport->cfm);
2985 if (cfm_opup >= 0) {
2986 enable = enable && cfm_opup;
2990 if (ofport->bundle) {
2991 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
2992 if (carrier_changed) {
2993 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
2997 if (ofport->may_enable != enable) {
2998 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3000 if (ofproto->has_bundle_action) {
3001 ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
3005 ofport->may_enable = enable;
3009 port_wait(struct ofport_dpif *ofport)
3012 cfm_wait(ofport->cfm);
3017 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
3018 struct ofproto_port *ofproto_port)
3020 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3021 struct dpif_port dpif_port;
3024 if (sset_contains(&ofproto->ghost_ports, devname)) {
3025 const char *type = netdev_get_type_from_name(devname);
3027 /* We may be called before ofproto->up.port_by_name is populated with
3028 * the appropriate ofport. For this reason, we must get the name and
3029 * type from the netdev layer directly. */
3031 const struct ofport *ofport;
3033 ofport = shash_find_data(&ofproto->up.port_by_name, devname);
3034 ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
3035 ofproto_port->name = xstrdup(devname);
3036 ofproto_port->type = xstrdup(type);
3042 if (!sset_contains(&ofproto->ports, devname)) {
3045 error = dpif_port_query_by_name(ofproto->backer->dpif,
3046 devname, &dpif_port);
3048 ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
3054 port_add(struct ofproto *ofproto_, struct netdev *netdev)
3056 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3057 const char *dp_port_name = netdev_vport_get_dpif_port(netdev);
3058 const char *devname = netdev_get_name(netdev);
3060 if (netdev_vport_is_patch(netdev)) {
3061 sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
3065 if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
3066 uint32_t port_no = UINT32_MAX;
3069 error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
3073 if (netdev_get_tunnel_config(netdev)) {
3074 simap_put(&ofproto->backer->tnl_backers, dp_port_name, port_no);
3078 if (netdev_get_tunnel_config(netdev)) {
3079 sset_add(&ofproto->ghost_ports, devname);
3081 sset_add(&ofproto->ports, devname);
3087 port_del(struct ofproto *ofproto_, uint16_t ofp_port)
3089 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3090 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
3097 sset_find_and_delete(&ofproto->ghost_ports,
3098 netdev_get_name(ofport->up.netdev));
3099 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3100 if (!ofport->tnl_port) {
3101 error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
3103 /* The caller is going to close ofport->up.netdev. If this is a
3104 * bonded port, then the bond is using that netdev, so remove it
3105 * from the bond. The client will need to reconfigure everything
3106 * after deleting ports, so then the slave will get re-added. */
3107 bundle_remove(&ofport->up);
3114 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
3116 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3119 error = netdev_get_stats(ofport->up.netdev, stats);
3121 if (!error && ofport_->ofp_port == OFPP_LOCAL) {
3122 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3124 /* ofproto->stats.tx_packets represents packets that we created
3125 * internally and sent to some port (e.g. packets sent with
3126 * send_packet()). Account for them as if they had come from
3127 * OFPP_LOCAL and got forwarded. */
3129 if (stats->rx_packets != UINT64_MAX) {
3130 stats->rx_packets += ofproto->stats.tx_packets;
3133 if (stats->rx_bytes != UINT64_MAX) {
3134 stats->rx_bytes += ofproto->stats.tx_bytes;
3137 /* ofproto->stats.rx_packets represents packets that were received on
3138 * some port and we processed internally and dropped (e.g. STP).
3139 * Account for them as if they had been forwarded to OFPP_LOCAL. */
3141 if (stats->tx_packets != UINT64_MAX) {
3142 stats->tx_packets += ofproto->stats.rx_packets;
3145 if (stats->tx_bytes != UINT64_MAX) {
3146 stats->tx_bytes += ofproto->stats.rx_bytes;
3153 /* Account packets for LOCAL port. */
3155 ofproto_update_local_port_stats(const struct ofproto *ofproto_,
3156 size_t tx_size, size_t rx_size)
3158 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3161 ofproto->stats.rx_packets++;
3162 ofproto->stats.rx_bytes += rx_size;
3165 ofproto->stats.tx_packets++;
3166 ofproto->stats.tx_bytes += tx_size;
3170 struct port_dump_state {
3175 struct ofproto_port port;
3180 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
3182 *statep = xzalloc(sizeof(struct port_dump_state));
3187 port_dump_next(const struct ofproto *ofproto_, void *state_,
3188 struct ofproto_port *port)
3190 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3191 struct port_dump_state *state = state_;
3192 const struct sset *sset;
3193 struct sset_node *node;
3195 if (state->has_port) {
3196 ofproto_port_destroy(&state->port);
3197 state->has_port = false;
3199 sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
3200 while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
3203 error = port_query_by_name(ofproto_, node->name, &state->port);
3205 *port = state->port;
3206 state->has_port = true;
3208 } else if (error != ENODEV) {
3213 if (!state->ghost) {
3214 state->ghost = true;
3217 return port_dump_next(ofproto_, state_, port);
3224 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3226 struct port_dump_state *state = state_;
3228 if (state->has_port) {
3229 ofproto_port_destroy(&state->port);
3236 port_poll(const struct ofproto *ofproto_, char **devnamep)
3238 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3240 if (ofproto->port_poll_errno) {
3241 int error = ofproto->port_poll_errno;
3242 ofproto->port_poll_errno = 0;
3246 if (sset_is_empty(&ofproto->port_poll_set)) {
3250 *devnamep = sset_pop(&ofproto->port_poll_set);
3255 port_poll_wait(const struct ofproto *ofproto_)
3257 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3258 dpif_port_poll_wait(ofproto->backer->dpif);
3262 port_is_lacp_current(const struct ofport *ofport_)
3264 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3265 return (ofport->bundle && ofport->bundle->lacp
3266 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3270 /* Upcall handling. */
3272 /* Flow miss batching.
3274 * Some dpifs implement operations faster when you hand them off in a batch.
3275 * To allow batching, "struct flow_miss" queues the dpif-related work needed
3276 * for a given flow. Each "struct flow_miss" corresponds to sending one or
3277 * more packets, plus possibly installing the flow in the dpif.
3279 * So far we only batch the operations that affect flow setup time the most.
3280 * It's possible to batch more than that, but the benefit might be minimal. */
3282 struct hmap_node hmap_node;
3283 struct ofproto_dpif *ofproto;
3285 enum odp_key_fitness key_fitness;
3286 const struct nlattr *key;
3288 struct initial_vals initial_vals;
3289 struct list packets;
3290 enum dpif_upcall_type upcall_type;
3291 uint32_t odp_in_port;
3294 struct flow_miss_op {
3295 struct dpif_op dpif_op;
3296 void *garbage; /* Pointer to pass to free(), NULL if none. */
3297 uint64_t stub[1024 / 8]; /* Temporary buffer. */
3300 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
3301 * OpenFlow controller as necessary according to their individual
3302 * configurations. */
3304 send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet,
3305 const struct flow *flow)
3307 struct ofputil_packet_in pin;
3309 pin.packet = packet->data;
3310 pin.packet_len = packet->size;
3311 pin.reason = OFPR_NO_MATCH;
3312 pin.controller_id = 0;
3317 pin.send_len = 0; /* not used for flow table misses */
3319 flow_get_metadata(flow, &pin.fmd);
3321 connmgr_send_packet_in(ofproto->up.connmgr, &pin);
3324 static enum slow_path_reason
3325 process_special(struct ofproto_dpif *ofproto, const struct flow *flow,
3326 const struct ofport_dpif *ofport, const struct ofpbuf *packet)
3330 } else if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) {
3332 cfm_process_heartbeat(ofport->cfm, packet);
3335 } else if (ofport->bundle && ofport->bundle->lacp
3336 && flow->dl_type == htons(ETH_TYPE_LACP)) {
3338 lacp_process_packet(ofport->bundle->lacp, ofport, packet);
3341 } else if (ofproto->stp && stp_should_process_flow(flow)) {
3343 stp_process_packet(ofport, packet);
3351 static struct flow_miss *
3352 flow_miss_find(struct hmap *todo, const struct ofproto_dpif *ofproto,
3353 const struct flow *flow, uint32_t hash)
3355 struct flow_miss *miss;
3357 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
3358 if (miss->ofproto == ofproto && flow_equal(&miss->flow, flow)) {
3366 /* Partially Initializes 'op' as an "execute" operation for 'miss' and
3367 * 'packet'. The caller must initialize op->actions and op->actions_len. If
3368 * 'miss' is associated with a subfacet the caller must also initialize the
3369 * returned op->subfacet, and if anything needs to be freed after processing
3370 * the op, the caller must initialize op->garbage also. */
3372 init_flow_miss_execute_op(struct flow_miss *miss, struct ofpbuf *packet,
3373 struct flow_miss_op *op)
3375 if (miss->flow.vlan_tci != miss->initial_vals.vlan_tci) {
3376 /* This packet was received on a VLAN splinter port. We
3377 * added a VLAN to the packet to make the packet resemble
3378 * the flow, but the actions were composed assuming that
3379 * the packet contained no VLAN. So, we must remove the
3380 * VLAN header from the packet before trying to execute the
3382 eth_pop_vlan(packet);
3386 op->dpif_op.type = DPIF_OP_EXECUTE;
3387 op->dpif_op.u.execute.key = miss->key;
3388 op->dpif_op.u.execute.key_len = miss->key_len;
3389 op->dpif_op.u.execute.packet = packet;
3392 /* Helper for handle_flow_miss_without_facet() and
3393 * handle_flow_miss_with_facet(). */
3395 handle_flow_miss_common(struct rule_dpif *rule,
3396 struct ofpbuf *packet, const struct flow *flow)
3398 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3400 ofproto->n_matches++;
3402 if (rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
3404 * Extra-special case for fail-open mode.
3406 * We are in fail-open mode and the packet matched the fail-open
3407 * rule, but we are connected to a controller too. We should send
3408 * the packet up to the controller in the hope that it will try to
3409 * set up a flow and thereby allow us to exit fail-open.
3411 * See the top-level comment in fail-open.c for more information.
3413 send_packet_in_miss(ofproto, packet, flow);
3417 /* Figures out whether a flow that missed in 'ofproto', whose details are in
3418 * 'miss', is likely to be worth tracking in detail in userspace and (usually)
3419 * installing a datapath flow. The answer is usually "yes" (a return value of
3420 * true). However, for short flows the cost of bookkeeping is much higher than
3421 * the benefits, so when the datapath holds a large number of flows we impose
3422 * some heuristics to decide which flows are likely to be worth tracking. */
3424 flow_miss_should_make_facet(struct ofproto_dpif *ofproto,
3425 struct flow_miss *miss, uint32_t hash)
3427 if (!ofproto->governor) {
3430 n_subfacets = hmap_count(&ofproto->subfacets);
3431 if (n_subfacets * 2 <= ofproto->up.flow_eviction_threshold) {
3435 ofproto->governor = governor_create(ofproto->up.name);
3438 return governor_should_install_flow(ofproto->governor, hash,
3439 list_size(&miss->packets));
3442 /* Handles 'miss', which matches 'rule', without creating a facet or subfacet
3443 * or creating any datapath flow. May add an "execute" operation to 'ops' and
3444 * increment '*n_ops'. */
3446 handle_flow_miss_without_facet(struct flow_miss *miss,
3447 struct rule_dpif *rule,
3448 struct flow_miss_op *ops, size_t *n_ops)
3450 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3451 long long int now = time_msec();
3452 struct action_xlate_ctx ctx;
3453 struct ofpbuf *packet;
3455 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3456 struct flow_miss_op *op = &ops[*n_ops];
3457 struct dpif_flow_stats stats;
3458 struct ofpbuf odp_actions;
3460 COVERAGE_INC(facet_suppress);
3462 ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub);
3464 dpif_flow_stats_extract(&miss->flow, packet, now, &stats);
3465 rule_credit_stats(rule, &stats);
3467 action_xlate_ctx_init(&ctx, ofproto, &miss->flow,
3468 &miss->initial_vals, rule, 0, packet);
3469 ctx.resubmit_stats = &stats;
3470 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len,
3473 if (odp_actions.size) {
3474 struct dpif_execute *execute = &op->dpif_op.u.execute;
3476 init_flow_miss_execute_op(miss, packet, op);
3477 execute->actions = odp_actions.data;
3478 execute->actions_len = odp_actions.size;
3479 op->garbage = ofpbuf_get_uninit_pointer(&odp_actions);
3483 ofpbuf_uninit(&odp_actions);
3488 /* Handles 'miss', which matches 'facet'. May add any required datapath
3489 * operations to 'ops', incrementing '*n_ops' for each new op.
3491 * All of the packets in 'miss' are considered to have arrived at time 'now'.
3492 * This is really important only for new facets: if we just called time_msec()
3493 * here, then the new subfacet or its packets could look (occasionally) as
3494 * though it was used some time after the facet was used. That can make a
3495 * one-packet flow look like it has a nonzero duration, which looks odd in
3496 * e.g. NetFlow statistics. */
3498 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
3500 struct flow_miss_op *ops, size_t *n_ops)
3502 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3503 enum subfacet_path want_path;
3504 struct subfacet *subfacet;
3505 struct ofpbuf *packet;
3507 subfacet = subfacet_create(facet, miss, now);
3509 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3510 struct flow_miss_op *op = &ops[*n_ops];
3511 struct dpif_flow_stats stats;
3512 struct ofpbuf odp_actions;
3514 handle_flow_miss_common(facet->rule, packet, &miss->flow);
3516 ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub);
3517 if (!subfacet->actions || subfacet->slow) {
3518 subfacet_make_actions(subfacet, packet, &odp_actions);
3521 dpif_flow_stats_extract(&facet->flow, packet, now, &stats);
3522 subfacet_update_stats(subfacet, &stats);
3524 if (subfacet->actions_len) {
3525 struct dpif_execute *execute = &op->dpif_op.u.execute;
3527 init_flow_miss_execute_op(miss, packet, op);
3528 if (!subfacet->slow) {
3529 execute->actions = subfacet->actions;
3530 execute->actions_len = subfacet->actions_len;
3531 ofpbuf_uninit(&odp_actions);
3533 execute->actions = odp_actions.data;
3534 execute->actions_len = odp_actions.size;
3535 op->garbage = ofpbuf_get_uninit_pointer(&odp_actions);
3540 ofpbuf_uninit(&odp_actions);
3544 want_path = subfacet_want_path(subfacet->slow);
3545 if (miss->upcall_type == DPIF_UC_MISS || subfacet->path != want_path) {
3546 struct flow_miss_op *op = &ops[(*n_ops)++];
3547 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
3549 subfacet->path = want_path;
3552 op->dpif_op.type = DPIF_OP_FLOW_PUT;
3553 put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
3554 put->key = miss->key;
3555 put->key_len = miss->key_len;
3556 if (want_path == SF_FAST_PATH) {
3557 put->actions = subfacet->actions;
3558 put->actions_len = subfacet->actions_len;
3560 compose_slow_path(ofproto, &facet->flow, subfacet->slow,
3561 op->stub, sizeof op->stub,
3562 &put->actions, &put->actions_len);
3568 /* Handles flow miss 'miss'. May add any required datapath operations
3569 * to 'ops', incrementing '*n_ops' for each new op. */
3571 handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops,
3574 struct ofproto_dpif *ofproto = miss->ofproto;
3575 struct facet *facet;
3579 /* The caller must ensure that miss->hmap_node.hash contains
3580 * flow_hash(miss->flow, 0). */
3581 hash = miss->hmap_node.hash;
3583 facet = facet_lookup_valid(ofproto, &miss->flow, hash);
3585 struct rule_dpif *rule = rule_dpif_lookup(ofproto, &miss->flow);
3587 if (!flow_miss_should_make_facet(ofproto, miss, hash)) {
3588 handle_flow_miss_without_facet(miss, rule, ops, n_ops);
3592 facet = facet_create(rule, &miss->flow, hash);
3597 handle_flow_miss_with_facet(miss, facet, now, ops, n_ops);
3600 static struct drop_key *
3601 drop_key_lookup(const struct dpif_backer *backer, const struct nlattr *key,
3604 struct drop_key *drop_key;
3606 HMAP_FOR_EACH_WITH_HASH (drop_key, hmap_node, hash_bytes(key, key_len, 0),
3607 &backer->drop_keys) {
3608 if (drop_key->key_len == key_len
3609 && !memcmp(drop_key->key, key, key_len)) {
3617 drop_key_clear(struct dpif_backer *backer)
3619 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3620 struct drop_key *drop_key, *next;
3622 HMAP_FOR_EACH_SAFE (drop_key, next, hmap_node, &backer->drop_keys) {
3625 error = dpif_flow_del(backer->dpif, drop_key->key, drop_key->key_len,
3627 if (error && !VLOG_DROP_WARN(&rl)) {
3628 struct ds ds = DS_EMPTY_INITIALIZER;
3629 odp_flow_key_format(drop_key->key, drop_key->key_len, &ds);
3630 VLOG_WARN("Failed to delete drop key (%s) (%s)", strerror(error),
3635 hmap_remove(&backer->drop_keys, &drop_key->hmap_node);
3636 free(drop_key->key);
3641 /* Given a datpath, packet, and flow metadata ('backer', 'packet', and 'key'
3642 * respectively), populates 'flow' with the result of odp_flow_key_to_flow().
3643 * Optionally, if nonnull, populates 'fitnessp' with the fitness of 'flow' as
3644 * returned by odp_flow_key_to_flow(). Also, optionally populates 'ofproto'
3645 * with the ofproto_dpif, and 'odp_in_port' with the datapath in_port, that
3646 * 'packet' ingressed.
3648 * If 'ofproto' is nonnull, requires 'flow''s in_port to exist. Otherwise sets
3649 * 'flow''s in_port to OFPP_NONE.
3651 * This function does post-processing on data returned from
3652 * odp_flow_key_to_flow() to help make VLAN splinters transparent to the rest
3653 * of the upcall processing logic. In particular, if the extracted in_port is
3654 * a VLAN splinter port, it replaces flow->in_port by the "real" port, sets
3655 * flow->vlan_tci correctly for the VLAN of the VLAN splinter port, and pushes
3656 * a VLAN header onto 'packet' (if it is nonnull).
3658 * Optionally, if 'initial_vals' is nonnull, sets 'initial_vals->vlan_tci'
3659 * to the VLAN TCI with which the packet was really received, that is, the
3660 * actual VLAN TCI extracted by odp_flow_key_to_flow(). (This differs from
3661 * the value returned in flow->vlan_tci only for packets received on
3662 * VLAN splinters.) Also, if received on an IP tunnel, sets
3663 * 'initial_vals->tunnel_ip_tos' to the tunnel's IP TOS.
3665 * Similarly, this function also includes some logic to help with tunnels. It
3666 * may modify 'flow' as necessary to make the tunneling implementation
3667 * transparent to the upcall processing logic.
3669 * Returns 0 if successful, ENODEV if the parsed flow has no associated ofport,
3670 * or some other positive errno if there are other problems. */
3672 ofproto_receive(const struct dpif_backer *backer, struct ofpbuf *packet,
3673 const struct nlattr *key, size_t key_len,
3674 struct flow *flow, enum odp_key_fitness *fitnessp,
3675 struct ofproto_dpif **ofproto, uint32_t *odp_in_port,
3676 struct initial_vals *initial_vals)
3678 const struct ofport_dpif *port;
3679 enum odp_key_fitness fitness;
3682 fitness = odp_flow_key_to_flow(key, key_len, flow);
3683 if (fitness == ODP_FIT_ERROR) {
3689 initial_vals->vlan_tci = flow->vlan_tci;
3690 initial_vals->tunnel_ip_tos = flow->tunnel.ip_tos;
3694 *odp_in_port = flow->in_port;
3697 if (tnl_port_should_receive(flow)) {
3698 const struct ofport *ofport = tnl_port_receive(flow);
3700 flow->in_port = OFPP_NONE;
3703 port = ofport_dpif_cast(ofport);
3705 /* We can't reproduce 'key' from 'flow'. */
3706 fitness = fitness == ODP_FIT_PERFECT ? ODP_FIT_TOO_MUCH : fitness;
3708 /* XXX: Since the tunnel module is not scoped per backer, it's
3709 * theoretically possible that we'll receive an ofport belonging to an
3710 * entirely different datapath. In practice, this can't happen because
3711 * no platforms has two separate datapaths which each support
3713 ovs_assert(ofproto_dpif_cast(port->up.ofproto)->backer == backer);
3715 port = odp_port_to_ofport(backer, flow->in_port);
3717 flow->in_port = OFPP_NONE;
3721 flow->in_port = port->up.ofp_port;
3722 if (vsp_adjust_flow(ofproto_dpif_cast(port->up.ofproto), flow)) {
3724 /* Make the packet resemble the flow, so that it gets sent to
3725 * an OpenFlow controller properly, so that it looks correct
3726 * for sFlow, and so that flow_extract() will get the correct
3727 * vlan_tci if it is called on 'packet'.
3729 * The allocated space inside 'packet' probably also contains
3730 * 'key', that is, both 'packet' and 'key' are probably part of
3731 * a struct dpif_upcall (see the large comment on that
3732 * structure definition), so pushing data on 'packet' is in
3733 * general not a good idea since it could overwrite 'key' or
3734 * free it as a side effect. However, it's OK in this special
3735 * case because we know that 'packet' is inside a Netlink
3736 * attribute: pushing 4 bytes will just overwrite the 4-byte
3737 * "struct nlattr", which is fine since we don't need that
3738 * header anymore. */
3739 eth_push_vlan(packet, flow->vlan_tci);
3741 /* We can't reproduce 'key' from 'flow'. */
3742 fitness = fitness == ODP_FIT_PERFECT ? ODP_FIT_TOO_MUCH : fitness;
3748 *ofproto = ofproto_dpif_cast(port->up.ofproto);
3753 *fitnessp = fitness;
3759 handle_miss_upcalls(struct dpif_backer *backer, struct dpif_upcall *upcalls,
3762 struct dpif_upcall *upcall;
3763 struct flow_miss *miss;
3764 struct flow_miss misses[FLOW_MISS_MAX_BATCH];
3765 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
3766 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
3776 /* Construct the to-do list.
3778 * This just amounts to extracting the flow from each packet and sticking
3779 * the packets that have the same flow in the same "flow_miss" structure so
3780 * that we can process them together. */
3783 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
3784 struct flow_miss *miss = &misses[n_misses];
3785 struct flow_miss *existing_miss;
3786 struct ofproto_dpif *ofproto;
3787 uint32_t odp_in_port;
3792 error = ofproto_receive(backer, upcall->packet, upcall->key,
3793 upcall->key_len, &flow, &miss->key_fitness,
3794 &ofproto, &odp_in_port, &miss->initial_vals);
3795 if (error == ENODEV) {
3796 struct drop_key *drop_key;
3798 /* Received packet on port for which we couldn't associate
3799 * an ofproto. This can happen if a port is removed while
3800 * traffic is being received. Print a rate-limited message
3801 * in case it happens frequently. Install a drop flow so
3802 * that future packets of the flow are inexpensively dropped
3804 VLOG_INFO_RL(&rl, "received packet on unassociated port %"PRIu32,
3807 drop_key = drop_key_lookup(backer, upcall->key, upcall->key_len);
3809 drop_key = xmalloc(sizeof *drop_key);
3810 drop_key->key = xmemdup(upcall->key, upcall->key_len);
3811 drop_key->key_len = upcall->key_len;
3813 hmap_insert(&backer->drop_keys, &drop_key->hmap_node,
3814 hash_bytes(drop_key->key, drop_key->key_len, 0));
3815 dpif_flow_put(backer->dpif, DPIF_FP_CREATE | DPIF_FP_MODIFY,
3816 drop_key->key, drop_key->key_len, NULL, 0, NULL);
3823 flow_extract(upcall->packet, flow.skb_priority, flow.skb_mark,
3824 &flow.tunnel, flow.in_port, &miss->flow);
3826 /* Add other packets to a to-do list. */
3827 hash = flow_hash(&miss->flow, 0);
3828 existing_miss = flow_miss_find(&todo, ofproto, &miss->flow, hash);
3829 if (!existing_miss) {
3830 hmap_insert(&todo, &miss->hmap_node, hash);
3831 miss->ofproto = ofproto;
3832 miss->key = upcall->key;
3833 miss->key_len = upcall->key_len;
3834 miss->upcall_type = upcall->type;
3835 miss->odp_in_port = odp_in_port;
3836 list_init(&miss->packets);
3840 miss = existing_miss;
3842 list_push_back(&miss->packets, &upcall->packet->list_node);
3845 /* Process each element in the to-do list, constructing the set of
3846 * operations to batch. */
3848 HMAP_FOR_EACH (miss, hmap_node, &todo) {
3849 handle_flow_miss(miss, flow_miss_ops, &n_ops);
3851 ovs_assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
3853 /* Execute batch. */
3854 for (i = 0; i < n_ops; i++) {
3855 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
3857 dpif_operate(backer->dpif, dpif_ops, n_ops);
3860 for (i = 0; i < n_ops; i++) {
3861 free(flow_miss_ops[i].garbage);
3863 hmap_destroy(&todo);
3866 static enum { SFLOW_UPCALL, MISS_UPCALL, BAD_UPCALL }
3867 classify_upcall(const struct dpif_upcall *upcall)
3869 union user_action_cookie cookie;
3871 /* First look at the upcall type. */
3872 switch (upcall->type) {
3873 case DPIF_UC_ACTION:
3879 case DPIF_N_UC_TYPES:
3881 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
3885 /* "action" upcalls need a closer look. */
3886 if (!upcall->userdata) {
3887 VLOG_WARN_RL(&rl, "action upcall missing cookie");
3890 if (nl_attr_get_size(upcall->userdata) != sizeof(cookie)) {
3891 VLOG_WARN_RL(&rl, "action upcall cookie has unexpected size %zu",
3892 nl_attr_get_size(upcall->userdata));
3895 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof(cookie));
3896 switch (cookie.type) {
3897 case USER_ACTION_COOKIE_SFLOW:
3898 return SFLOW_UPCALL;
3900 case USER_ACTION_COOKIE_SLOW_PATH:
3903 case USER_ACTION_COOKIE_UNSPEC:
3905 VLOG_WARN_RL(&rl, "invalid user cookie : 0x%"PRIx64,
3906 nl_attr_get_u64(upcall->userdata));
3912 handle_sflow_upcall(struct dpif_backer *backer,
3913 const struct dpif_upcall *upcall)
3915 struct ofproto_dpif *ofproto;
3916 union user_action_cookie cookie;
3918 uint32_t odp_in_port;
3920 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3921 &flow, NULL, &ofproto, &odp_in_port, NULL)
3922 || !ofproto->sflow) {
3926 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof(cookie));
3927 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
3928 odp_in_port, &cookie);
3932 handle_upcalls(struct dpif_backer *backer, unsigned int max_batch)
3934 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
3935 struct ofpbuf miss_bufs[FLOW_MISS_MAX_BATCH];
3936 uint64_t miss_buf_stubs[FLOW_MISS_MAX_BATCH][4096 / 8];
3941 ovs_assert(max_batch <= FLOW_MISS_MAX_BATCH);
3944 for (n_processed = 0; n_processed < max_batch; n_processed++) {
3945 struct dpif_upcall *upcall = &misses[n_misses];
3946 struct ofpbuf *buf = &miss_bufs[n_misses];
3949 ofpbuf_use_stub(buf, miss_buf_stubs[n_misses],
3950 sizeof miss_buf_stubs[n_misses]);
3951 error = dpif_recv(backer->dpif, upcall, buf);
3957 switch (classify_upcall(upcall)) {
3959 /* Handle it later. */
3964 handle_sflow_upcall(backer, upcall);
3974 /* Handle deferred MISS_UPCALL processing. */
3975 handle_miss_upcalls(backer, misses, n_misses);
3976 for (i = 0; i < n_misses; i++) {
3977 ofpbuf_uninit(&miss_bufs[i]);
3983 /* Flow expiration. */
3985 static int subfacet_max_idle(const struct ofproto_dpif *);
3986 static void update_stats(struct dpif_backer *);
3987 static void rule_expire(struct rule_dpif *);
3988 static void expire_subfacets(struct ofproto_dpif *, int dp_max_idle);
3990 /* This function is called periodically by run(). Its job is to collect
3991 * updates for the flows that have been installed into the datapath, most
3992 * importantly when they last were used, and then use that information to
3993 * expire flows that have not been used recently.
3995 * Returns the number of milliseconds after which it should be called again. */
3997 expire(struct dpif_backer *backer)
3999 struct ofproto_dpif *ofproto;
4000 int max_idle = INT32_MAX;
4002 /* Periodically clear out the drop keys in an effort to keep them
4003 * relatively few. */
4004 drop_key_clear(backer);
4006 /* Update stats for each flow in the backer. */
4007 update_stats(backer);
4009 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4010 struct rule *rule, *next_rule;
4013 if (ofproto->backer != backer) {
4017 /* Expire subfacets that have been idle too long. */
4018 dp_max_idle = subfacet_max_idle(ofproto);
4019 expire_subfacets(ofproto, dp_max_idle);
4021 max_idle = MIN(max_idle, dp_max_idle);
4023 /* Expire OpenFlow flows whose idle_timeout or hard_timeout
4025 LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
4026 &ofproto->up.expirable) {
4027 rule_expire(rule_dpif_cast(rule));
4030 /* All outstanding data in existing flows has been accounted, so it's a
4031 * good time to do bond rebalancing. */
4032 if (ofproto->has_bonded_bundles) {
4033 struct ofbundle *bundle;
4035 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
4037 bond_rebalance(bundle->bond, &backer->revalidate_set);
4043 return MIN(max_idle, 1000);
4046 /* Updates flow table statistics given that the datapath just reported 'stats'
4047 * as 'subfacet''s statistics. */
4049 update_subfacet_stats(struct subfacet *subfacet,
4050 const struct dpif_flow_stats *stats)
4052 struct facet *facet = subfacet->facet;
4054 if (stats->n_packets >= subfacet->dp_packet_count) {
4055 uint64_t extra = stats->n_packets - subfacet->dp_packet_count;
4056 facet->packet_count += extra;
4058 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
4061 if (stats->n_bytes >= subfacet->dp_byte_count) {
4062 facet->byte_count += stats->n_bytes - subfacet->dp_byte_count;
4064 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
4067 subfacet->dp_packet_count = stats->n_packets;
4068 subfacet->dp_byte_count = stats->n_bytes;
4070 facet->tcp_flags |= stats->tcp_flags;
4072 subfacet_update_time(subfacet, stats->used);
4073 if (facet->accounted_bytes < facet->byte_count) {
4075 facet_account(facet);
4076 facet->accounted_bytes = facet->byte_count;
4078 facet_push_stats(facet);
4081 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
4082 * about, or a flow that shouldn't be installed but was anyway. Delete it. */
4084 delete_unexpected_flow(struct ofproto_dpif *ofproto,
4085 const struct nlattr *key, size_t key_len)
4087 if (!VLOG_DROP_WARN(&rl)) {
4091 odp_flow_key_format(key, key_len, &s);
4092 VLOG_WARN("unexpected flow on %s: %s", ofproto->up.name, ds_cstr(&s));
4096 COVERAGE_INC(facet_unexpected);
4097 dpif_flow_del(ofproto->backer->dpif, key, key_len, NULL);
4100 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
4102 * This function also pushes statistics updates to rules which each facet
4103 * resubmits into. Generally these statistics will be accurate. However, if a
4104 * facet changes the rule it resubmits into at some time in between
4105 * update_stats() runs, it is possible that statistics accrued to the
4106 * old rule will be incorrectly attributed to the new rule. This could be
4107 * avoided by calling update_stats() whenever rules are created or
4108 * deleted. However, the performance impact of making so many calls to the
4109 * datapath do not justify the benefit of having perfectly accurate statistics.
4112 update_stats(struct dpif_backer *backer)
4114 const struct dpif_flow_stats *stats;
4115 struct dpif_flow_dump dump;
4116 const struct nlattr *key;
4119 dpif_flow_dump_start(&dump, backer->dpif);
4120 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
4122 struct subfacet *subfacet;
4123 struct ofproto_dpif *ofproto;
4124 struct ofport_dpif *ofport;
4127 if (ofproto_receive(backer, NULL, key, key_len, &flow, NULL, &ofproto,
4132 ofport = get_ofp_port(ofproto, flow.in_port);
4133 if (ofport && ofport->tnl_port) {
4134 netdev_vport_inc_rx(ofport->up.netdev, stats);
4137 key_hash = odp_flow_key_hash(key, key_len);
4138 subfacet = subfacet_find(ofproto, key, key_len, key_hash);
4139 switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
4141 update_subfacet_stats(subfacet, stats);
4145 /* Stats are updated per-packet. */
4148 case SF_NOT_INSTALLED:
4150 delete_unexpected_flow(ofproto, key, key_len);
4154 dpif_flow_dump_done(&dump);
4157 /* Calculates and returns the number of milliseconds of idle time after which
4158 * subfacets should expire from the datapath. When a subfacet expires, we fold
4159 * its statistics into its facet, and when a facet's last subfacet expires, we
4160 * fold its statistic into its rule. */
4162 subfacet_max_idle(const struct ofproto_dpif *ofproto)
4165 * Idle time histogram.
4167 * Most of the time a switch has a relatively small number of subfacets.
4168 * When this is the case we might as well keep statistics for all of them
4169 * in userspace and to cache them in the kernel datapath for performance as
4172 * As the number of subfacets increases, the memory required to maintain
4173 * statistics about them in userspace and in the kernel becomes
4174 * significant. However, with a large number of subfacets it is likely
4175 * that only a few of them are "heavy hitters" that consume a large amount
4176 * of bandwidth. At this point, only heavy hitters are worth caching in
4177 * the kernel and maintaining in userspaces; other subfacets we can
4180 * The technique used to compute the idle time is to build a histogram with
4181 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
4182 * that is installed in the kernel gets dropped in the appropriate bucket.
4183 * After the histogram has been built, we compute the cutoff so that only
4184 * the most-recently-used 1% of subfacets (but at least
4185 * ofproto->up.flow_eviction_threshold flows) are kept cached. At least
4186 * the most-recently-used bucket of subfacets is kept, so actually an
4187 * arbitrary number of subfacets can be kept in any given expiration run
4188 * (though the next run will delete most of those unless they receive
4191 * This requires a second pass through the subfacets, in addition to the
4192 * pass made by update_stats(), because the former function never looks at
4193 * uninstallable subfacets.
4195 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
4196 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
4197 int buckets[N_BUCKETS] = { 0 };
4198 int total, subtotal, bucket;
4199 struct subfacet *subfacet;
4203 total = hmap_count(&ofproto->subfacets);
4204 if (total <= ofproto->up.flow_eviction_threshold) {
4205 return N_BUCKETS * BUCKET_WIDTH;
4208 /* Build histogram. */
4210 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
4211 long long int idle = now - subfacet->used;
4212 int bucket = (idle <= 0 ? 0
4213 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
4214 : (unsigned int) idle / BUCKET_WIDTH);
4218 /* Find the first bucket whose flows should be expired. */
4219 subtotal = bucket = 0;
4221 subtotal += buckets[bucket++];
4222 } while (bucket < N_BUCKETS &&
4223 subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100));
4225 if (VLOG_IS_DBG_ENABLED()) {
4229 ds_put_cstr(&s, "keep");
4230 for (i = 0; i < N_BUCKETS; i++) {
4232 ds_put_cstr(&s, ", drop");
4235 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
4238 VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s));
4242 return bucket * BUCKET_WIDTH;
4246 expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle)
4248 /* Cutoff time for most flows. */
4249 long long int normal_cutoff = time_msec() - dp_max_idle;
4251 /* We really want to keep flows for special protocols around, so use a more
4252 * conservative cutoff. */
4253 long long int special_cutoff = time_msec() - 10000;
4255 struct subfacet *subfacet, *next_subfacet;
4256 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
4260 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
4261 &ofproto->subfacets) {
4262 long long int cutoff;
4264 cutoff = (subfacet->slow & (SLOW_CFM | SLOW_LACP | SLOW_STP)
4267 if (subfacet->used < cutoff) {
4268 if (subfacet->path != SF_NOT_INSTALLED) {
4269 batch[n_batch++] = subfacet;
4270 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
4271 subfacet_destroy_batch(ofproto, batch, n_batch);
4275 subfacet_destroy(subfacet);
4281 subfacet_destroy_batch(ofproto, batch, n_batch);
4285 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
4286 * then delete it entirely. */
4288 rule_expire(struct rule_dpif *rule)
4290 struct facet *facet, *next_facet;
4294 if (rule->up.pending) {
4295 /* We'll have to expire it later. */
4299 /* Has 'rule' expired? */
4301 if (rule->up.hard_timeout
4302 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
4303 reason = OFPRR_HARD_TIMEOUT;
4304 } else if (rule->up.idle_timeout
4305 && now > rule->up.used + rule->up.idle_timeout * 1000) {
4306 reason = OFPRR_IDLE_TIMEOUT;
4311 COVERAGE_INC(ofproto_dpif_expired);
4313 /* Update stats. (This is a no-op if the rule expired due to an idle
4314 * timeout, because that only happens when the rule has no facets left.) */
4315 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4316 facet_remove(facet);
4319 /* Get rid of the rule. */
4320 ofproto_rule_expire(&rule->up, reason);
4325 /* Creates and returns a new facet owned by 'rule', given a 'flow'.
4327 * The caller must already have determined that no facet with an identical
4328 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
4329 * the ofproto's classifier table.
4331 * 'hash' must be the return value of flow_hash(flow, 0).
4333 * The facet will initially have no subfacets. The caller should create (at
4334 * least) one subfacet with subfacet_create(). */
4335 static struct facet *
4336 facet_create(struct rule_dpif *rule, const struct flow *flow, uint32_t hash)
4338 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4339 struct facet *facet;
4341 facet = xzalloc(sizeof *facet);
4342 facet->used = time_msec();
4343 hmap_insert(&ofproto->facets, &facet->hmap_node, hash);
4344 list_push_back(&rule->facets, &facet->list_node);
4346 facet->flow = *flow;
4347 list_init(&facet->subfacets);
4348 netflow_flow_init(&facet->nf_flow);
4349 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
4355 facet_free(struct facet *facet)
4360 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
4361 * 'packet', which arrived on 'in_port'. */
4363 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
4364 const struct nlattr *odp_actions, size_t actions_len,
4365 struct ofpbuf *packet)
4367 struct odputil_keybuf keybuf;
4371 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4372 odp_flow_key_from_flow(&key, flow,
4373 ofp_port_to_odp_port(ofproto, flow->in_port));
4375 error = dpif_execute(ofproto->backer->dpif, key.data, key.size,
4376 odp_actions, actions_len, packet);
4380 /* Remove 'facet' from 'ofproto' and free up the associated memory:
4382 * - If 'facet' was installed in the datapath, uninstalls it and updates its
4383 * rule's statistics, via subfacet_uninstall().
4385 * - Removes 'facet' from its rule and from ofproto->facets.
4388 facet_remove(struct facet *facet)
4390 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4391 struct subfacet *subfacet, *next_subfacet;
4393 ovs_assert(!list_is_empty(&facet->subfacets));
4395 /* First uninstall all of the subfacets to get final statistics. */
4396 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4397 subfacet_uninstall(subfacet);
4400 /* Flush the final stats to the rule.
4402 * This might require us to have at least one subfacet around so that we
4403 * can use its actions for accounting in facet_account(), which is why we
4404 * have uninstalled but not yet destroyed the subfacets. */
4405 facet_flush_stats(facet);
4407 /* Now we're really all done so destroy everything. */
4408 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
4409 &facet->subfacets) {
4410 subfacet_destroy__(subfacet);
4412 hmap_remove(&ofproto->facets, &facet->hmap_node);
4413 list_remove(&facet->list_node);
4417 /* Feed information from 'facet' back into the learning table to keep it in
4418 * sync with what is actually flowing through the datapath. */
4420 facet_learn(struct facet *facet)
4422 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4423 struct subfacet *subfacet= CONTAINER_OF(list_front(&facet->subfacets),
4424 struct subfacet, list_node);
4425 struct action_xlate_ctx ctx;
4427 if (!facet->has_learn
4428 && !facet->has_normal
4429 && (!facet->has_fin_timeout
4430 || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
4434 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
4435 &subfacet->initial_vals,
4436 facet->rule, facet->tcp_flags, NULL);
4437 ctx.may_learn = true;
4438 xlate_actions_for_side_effects(&ctx, facet->rule->up.ofpacts,
4439 facet->rule->up.ofpacts_len);
4443 facet_account(struct facet *facet)
4445 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4446 struct subfacet *subfacet = facet_get_subfacet(facet);
4447 const struct nlattr *a;
4452 if (!facet->has_normal || !ofproto->has_bonded_bundles) {
4455 n_bytes = facet->byte_count - facet->accounted_bytes;
4457 /* This loop feeds byte counters to bond_account() for rebalancing to use
4458 * as a basis. We also need to track the actual VLAN on which the packet
4459 * is going to be sent to ensure that it matches the one passed to
4460 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
4463 * We use the actions from an arbitrary subfacet because they should all
4464 * be equally valid for our purpose. */
4465 vlan_tci = facet->flow.vlan_tci;
4466 NL_ATTR_FOR_EACH_UNSAFE (a, left,
4467 subfacet->actions, subfacet->actions_len) {
4468 const struct ovs_action_push_vlan *vlan;
4469 struct ofport_dpif *port;
4471 switch (nl_attr_type(a)) {
4472 case OVS_ACTION_ATTR_OUTPUT:
4473 port = get_odp_port(ofproto, nl_attr_get_u32(a));
4474 if (port && port->bundle && port->bundle->bond) {
4475 bond_account(port->bundle->bond, &facet->flow,
4476 vlan_tci_to_vid(vlan_tci), n_bytes);
4480 case OVS_ACTION_ATTR_POP_VLAN:
4481 vlan_tci = htons(0);
4484 case OVS_ACTION_ATTR_PUSH_VLAN:
4485 vlan = nl_attr_get(a);
4486 vlan_tci = vlan->vlan_tci;
4492 /* Returns true if the only action for 'facet' is to send to the controller.
4493 * (We don't report NetFlow expiration messages for such facets because they
4494 * are just part of the control logic for the network, not real traffic). */
4496 facet_is_controller_flow(struct facet *facet)
4499 const struct rule *rule = &facet->rule->up;
4500 const struct ofpact *ofpacts = rule->ofpacts;
4501 size_t ofpacts_len = rule->ofpacts_len;
4503 if (ofpacts_len > 0 &&
4504 ofpacts->type == OFPACT_CONTROLLER &&
4505 ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len)) {
4512 /* Folds all of 'facet''s statistics into its rule. Also updates the
4513 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
4514 * 'facet''s statistics in the datapath should have been zeroed and folded into
4515 * its packet and byte counts before this function is called. */
4517 facet_flush_stats(struct facet *facet)
4519 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4520 struct subfacet *subfacet;
4522 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4523 ovs_assert(!subfacet->dp_byte_count);
4524 ovs_assert(!subfacet->dp_packet_count);
4527 facet_push_stats(facet);
4528 if (facet->accounted_bytes < facet->byte_count) {
4529 facet_account(facet);
4530 facet->accounted_bytes = facet->byte_count;
4533 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
4534 struct ofexpired expired;
4535 expired.flow = facet->flow;
4536 expired.packet_count = facet->packet_count;
4537 expired.byte_count = facet->byte_count;
4538 expired.used = facet->used;
4539 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4542 facet->rule->packet_count += facet->packet_count;
4543 facet->rule->byte_count += facet->byte_count;
4545 /* Reset counters to prevent double counting if 'facet' ever gets
4547 facet_reset_counters(facet);
4549 netflow_flow_clear(&facet->nf_flow);
4550 facet->tcp_flags = 0;
4553 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
4554 * Returns it if found, otherwise a null pointer.
4556 * 'hash' must be the return value of flow_hash(flow, 0).
4558 * The returned facet might need revalidation; use facet_lookup_valid()
4559 * instead if that is important. */
4560 static struct facet *
4561 facet_find(struct ofproto_dpif *ofproto,
4562 const struct flow *flow, uint32_t hash)
4564 struct facet *facet;
4566 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, hash, &ofproto->facets) {
4567 if (flow_equal(flow, &facet->flow)) {
4575 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
4576 * Returns it if found, otherwise a null pointer.
4578 * 'hash' must be the return value of flow_hash(flow, 0).
4580 * The returned facet is guaranteed to be valid. */
4581 static struct facet *
4582 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow,
4585 struct facet *facet;
4587 facet = facet_find(ofproto, flow, hash);
4589 && (ofproto->backer->need_revalidate
4590 || tag_set_intersects(&ofproto->backer->revalidate_set,
4592 facet_revalidate(facet);
4594 /* facet_revalidate() may have destroyed 'facet'. */
4595 facet = facet_find(ofproto, flow, hash);
4601 /* Return a subfacet from 'facet'. A facet consists of one or more
4602 * subfacets, and this function returns one of them. */
4603 static struct subfacet *facet_get_subfacet(struct facet *facet)
4605 return CONTAINER_OF(list_front(&facet->subfacets), struct subfacet,
4610 subfacet_path_to_string(enum subfacet_path path)
4613 case SF_NOT_INSTALLED:
4614 return "not installed";
4616 return "in fast path";
4618 return "in slow path";
4624 /* Returns the path in which a subfacet should be installed if its 'slow'
4625 * member has the specified value. */
4626 static enum subfacet_path
4627 subfacet_want_path(enum slow_path_reason slow)
4629 return slow ? SF_SLOW_PATH : SF_FAST_PATH;
4632 /* Returns true if 'subfacet' needs to have its datapath flow updated,
4633 * supposing that its actions have been recalculated as 'want_actions' and that
4634 * 'slow' is nonzero iff 'subfacet' should be in the slow path. */
4636 subfacet_should_install(struct subfacet *subfacet, enum slow_path_reason slow,
4637 const struct ofpbuf *want_actions)
4639 enum subfacet_path want_path = subfacet_want_path(slow);
4640 return (want_path != subfacet->path
4641 || (want_path == SF_FAST_PATH
4642 && (subfacet->actions_len != want_actions->size
4643 || memcmp(subfacet->actions, want_actions->data,
4644 subfacet->actions_len))));
4648 facet_check_consistency(struct facet *facet)
4650 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
4652 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4654 uint64_t odp_actions_stub[1024 / 8];
4655 struct ofpbuf odp_actions;
4657 struct rule_dpif *rule;
4658 struct subfacet *subfacet;
4659 bool may_log = false;
4662 /* Check the rule for consistency. */
4663 rule = rule_dpif_lookup(ofproto, &facet->flow);
4664 ok = rule == facet->rule;
4666 may_log = !VLOG_DROP_WARN(&rl);
4671 flow_format(&s, &facet->flow);
4672 ds_put_format(&s, ": facet associated with wrong rule (was "
4673 "table=%"PRIu8",", facet->rule->up.table_id);
4674 cls_rule_format(&facet->rule->up.cr, &s);
4675 ds_put_format(&s, ") (should have been table=%"PRIu8",",
4677 cls_rule_format(&rule->up.cr, &s);
4678 ds_put_char(&s, ')');
4680 VLOG_WARN("%s", ds_cstr(&s));
4685 /* Check the datapath actions for consistency. */
4686 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
4687 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4688 enum subfacet_path want_path;
4689 struct action_xlate_ctx ctx;
4692 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
4693 &subfacet->initial_vals, rule, 0, NULL);
4694 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len,
4697 if (subfacet->path == SF_NOT_INSTALLED) {
4698 /* This only happens if the datapath reported an error when we
4699 * tried to install the flow. Don't flag another error here. */
4703 want_path = subfacet_want_path(subfacet->slow);
4704 if (want_path == SF_SLOW_PATH && subfacet->path == SF_SLOW_PATH) {
4705 /* The actions for slow-path flows may legitimately vary from one
4706 * packet to the next. We're done. */
4710 if (!subfacet_should_install(subfacet, subfacet->slow, &odp_actions)) {
4714 /* Inconsistency! */
4716 may_log = !VLOG_DROP_WARN(&rl);
4720 /* Rate-limited, skip reporting. */
4725 odp_flow_key_format(subfacet->key, subfacet->key_len, &s);
4727 ds_put_cstr(&s, ": inconsistency in subfacet");
4728 if (want_path != subfacet->path) {
4729 enum odp_key_fitness fitness = subfacet->key_fitness;
4731 ds_put_format(&s, " (%s, fitness=%s)",
4732 subfacet_path_to_string(subfacet->path),
4733 odp_key_fitness_to_string(fitness));
4734 ds_put_format(&s, " (should have been %s)",
4735 subfacet_path_to_string(want_path));
4736 } else if (want_path == SF_FAST_PATH) {
4737 ds_put_cstr(&s, " (actions were: ");
4738 format_odp_actions(&s, subfacet->actions,
4739 subfacet->actions_len);
4740 ds_put_cstr(&s, ") (correct actions: ");
4741 format_odp_actions(&s, odp_actions.data, odp_actions.size);
4742 ds_put_char(&s, ')');
4744 ds_put_cstr(&s, " (actions: ");
4745 format_odp_actions(&s, subfacet->actions,
4746 subfacet->actions_len);
4747 ds_put_char(&s, ')');
4749 VLOG_WARN("%s", ds_cstr(&s));
4752 ofpbuf_uninit(&odp_actions);
4757 /* Re-searches the classifier for 'facet':
4759 * - If the rule found is different from 'facet''s current rule, moves
4760 * 'facet' to the new rule and recompiles its actions.
4762 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
4763 * where it is and recompiles its actions anyway.
4765 * - If any of 'facet''s subfacets correspond to a new flow according to
4766 * ofproto_receive(), 'facet' is removed. */
4768 facet_revalidate(struct facet *facet)
4770 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4772 struct nlattr *odp_actions;
4775 struct actions *new_actions;
4777 struct action_xlate_ctx ctx;
4778 uint64_t odp_actions_stub[1024 / 8];
4779 struct ofpbuf odp_actions;
4781 struct rule_dpif *new_rule;
4782 struct subfacet *subfacet;
4785 COVERAGE_INC(facet_revalidate);
4787 /* Check that child subfacets still correspond to this facet. Tunnel
4788 * configuration changes could cause a subfacet's OpenFlow in_port to
4790 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4791 struct ofproto_dpif *recv_ofproto;
4792 struct flow recv_flow;
4795 error = ofproto_receive(ofproto->backer, NULL, subfacet->key,
4796 subfacet->key_len, &recv_flow, NULL,
4797 &recv_ofproto, NULL, NULL);
4799 || recv_ofproto != ofproto
4800 || memcmp(&recv_flow, &facet->flow, sizeof recv_flow)) {
4801 facet_remove(facet);
4806 new_rule = rule_dpif_lookup(ofproto, &facet->flow);
4808 /* Calculate new datapath actions.
4810 * We do not modify any 'facet' state yet, because we might need to, e.g.,
4811 * emit a NetFlow expiration and, if so, we need to have the old state
4812 * around to properly compose it. */
4814 /* If the datapath actions changed or the installability changed,
4815 * then we need to talk to the datapath. */
4818 memset(&ctx, 0, sizeof ctx);
4819 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
4820 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4821 enum slow_path_reason slow;
4823 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
4824 &subfacet->initial_vals, new_rule, 0, NULL);
4825 xlate_actions(&ctx, new_rule->up.ofpacts, new_rule->up.ofpacts_len,
4828 slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
4829 if (subfacet_should_install(subfacet, slow, &odp_actions)) {
4830 struct dpif_flow_stats stats;
4832 subfacet_install(subfacet,
4833 odp_actions.data, odp_actions.size, &stats, slow);
4834 subfacet_update_stats(subfacet, &stats);
4837 new_actions = xcalloc(list_size(&facet->subfacets),
4838 sizeof *new_actions);
4840 new_actions[i].odp_actions = xmemdup(odp_actions.data,
4842 new_actions[i].actions_len = odp_actions.size;
4847 ofpbuf_uninit(&odp_actions);
4850 facet_flush_stats(facet);
4853 /* Update 'facet' now that we've taken care of all the old state. */
4854 facet->tags = ctx.tags;
4855 facet->nf_flow.output_iface = ctx.nf_output_iface;
4856 facet->has_learn = ctx.has_learn;
4857 facet->has_normal = ctx.has_normal;
4858 facet->has_fin_timeout = ctx.has_fin_timeout;
4859 facet->mirrors = ctx.mirrors;
4862 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4863 subfacet->slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
4865 if (new_actions && new_actions[i].odp_actions) {
4866 free(subfacet->actions);
4867 subfacet->actions = new_actions[i].odp_actions;
4868 subfacet->actions_len = new_actions[i].actions_len;
4874 if (facet->rule != new_rule) {
4875 COVERAGE_INC(facet_changed_rule);
4876 list_remove(&facet->list_node);
4877 list_push_back(&new_rule->facets, &facet->list_node);
4878 facet->rule = new_rule;
4879 facet->used = new_rule->up.created;
4880 facet->prev_used = facet->used;
4884 /* Updates 'facet''s used time. Caller is responsible for calling
4885 * facet_push_stats() to update the flows which 'facet' resubmits into. */
4887 facet_update_time(struct facet *facet, long long int used)
4889 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4890 if (used > facet->used) {
4892 ofproto_rule_update_used(&facet->rule->up, used);
4893 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
4898 facet_reset_counters(struct facet *facet)
4900 facet->packet_count = 0;
4901 facet->byte_count = 0;
4902 facet->prev_packet_count = 0;
4903 facet->prev_byte_count = 0;
4904 facet->accounted_bytes = 0;
4908 facet_push_stats(struct facet *facet)
4910 struct dpif_flow_stats stats;
4912 ovs_assert(facet->packet_count >= facet->prev_packet_count);
4913 ovs_assert(facet->byte_count >= facet->prev_byte_count);
4914 ovs_assert(facet->used >= facet->prev_used);
4916 stats.n_packets = facet->packet_count - facet->prev_packet_count;
4917 stats.n_bytes = facet->byte_count - facet->prev_byte_count;
4918 stats.used = facet->used;
4919 stats.tcp_flags = 0;
4921 if (stats.n_packets || stats.n_bytes || facet->used > facet->prev_used) {
4922 facet->prev_packet_count = facet->packet_count;
4923 facet->prev_byte_count = facet->byte_count;
4924 facet->prev_used = facet->used;
4926 flow_push_stats(facet, &stats);
4928 update_mirror_stats(ofproto_dpif_cast(facet->rule->up.ofproto),
4929 facet->mirrors, stats.n_packets, stats.n_bytes);
4934 rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats)
4936 rule->packet_count += stats->n_packets;
4937 rule->byte_count += stats->n_bytes;
4938 ofproto_rule_update_used(&rule->up, stats->used);
4941 /* Pushes flow statistics to the rules which 'facet->flow' resubmits
4942 * into given 'facet->rule''s actions and mirrors. */
4944 flow_push_stats(struct facet *facet, const struct dpif_flow_stats *stats)
4946 struct rule_dpif *rule = facet->rule;
4947 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4948 struct subfacet *subfacet = facet_get_subfacet(facet);
4949 struct action_xlate_ctx ctx;
4951 ofproto_rule_update_used(&rule->up, stats->used);
4953 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
4954 &subfacet->initial_vals, rule, 0, NULL);
4955 ctx.resubmit_stats = stats;
4956 xlate_actions_for_side_effects(&ctx, rule->up.ofpacts,
4957 rule->up.ofpacts_len);
4962 static struct subfacet *
4963 subfacet_find(struct ofproto_dpif *ofproto,
4964 const struct nlattr *key, size_t key_len, uint32_t key_hash)
4966 struct subfacet *subfacet;
4968 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
4969 &ofproto->subfacets) {
4970 if (subfacet->key_len == key_len
4971 && !memcmp(key, subfacet->key, key_len)) {
4979 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
4980 * 'key_fitness', 'key', and 'key_len' members in 'miss'. Returns the
4981 * existing subfacet if there is one, otherwise creates and returns a
4984 * If the returned subfacet is new, then subfacet->actions will be NULL, in
4985 * which case the caller must populate the actions with
4986 * subfacet_make_actions(). */
4987 static struct subfacet *
4988 subfacet_create(struct facet *facet, struct flow_miss *miss,
4991 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4992 enum odp_key_fitness key_fitness = miss->key_fitness;
4993 const struct nlattr *key = miss->key;
4994 size_t key_len = miss->key_len;
4996 struct subfacet *subfacet;
4998 key_hash = odp_flow_key_hash(key, key_len);
5000 if (list_is_empty(&facet->subfacets)) {
5001 subfacet = &facet->one_subfacet;
5003 subfacet = subfacet_find(ofproto, key, key_len, key_hash);
5005 if (subfacet->facet == facet) {
5009 /* This shouldn't happen. */
5010 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
5011 subfacet_destroy(subfacet);
5014 subfacet = xmalloc(sizeof *subfacet);
5017 hmap_insert(&ofproto->subfacets, &subfacet->hmap_node, key_hash);
5018 list_push_back(&facet->subfacets, &subfacet->list_node);
5019 subfacet->facet = facet;
5020 subfacet->key_fitness = key_fitness;
5021 subfacet->key = xmemdup(key, key_len);
5022 subfacet->key_len = key_len;
5023 subfacet->used = now;
5024 subfacet->dp_packet_count = 0;
5025 subfacet->dp_byte_count = 0;
5026 subfacet->actions_len = 0;
5027 subfacet->actions = NULL;
5028 subfacet->slow = (subfacet->key_fitness == ODP_FIT_TOO_LITTLE
5031 subfacet->path = SF_NOT_INSTALLED;
5032 subfacet->initial_vals = miss->initial_vals;
5033 subfacet->odp_in_port = miss->odp_in_port;
5038 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
5039 * its facet within 'ofproto', and frees it. */
5041 subfacet_destroy__(struct subfacet *subfacet)
5043 struct facet *facet = subfacet->facet;
5044 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5046 subfacet_uninstall(subfacet);
5047 hmap_remove(&ofproto->subfacets, &subfacet->hmap_node);
5048 list_remove(&subfacet->list_node);
5049 free(subfacet->key);
5050 free(subfacet->actions);
5051 if (subfacet != &facet->one_subfacet) {
5056 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
5057 * last remaining subfacet in its facet destroys the facet too. */
5059 subfacet_destroy(struct subfacet *subfacet)
5061 struct facet *facet = subfacet->facet;
5063 if (list_is_singleton(&facet->subfacets)) {
5064 /* facet_remove() needs at least one subfacet (it will remove it). */
5065 facet_remove(facet);
5067 subfacet_destroy__(subfacet);
5072 subfacet_destroy_batch(struct ofproto_dpif *ofproto,
5073 struct subfacet **subfacets, int n)
5075 struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH];
5076 struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH];
5077 struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH];
5080 for (i = 0; i < n; i++) {
5081 ops[i].type = DPIF_OP_FLOW_DEL;
5082 ops[i].u.flow_del.key = subfacets[i]->key;
5083 ops[i].u.flow_del.key_len = subfacets[i]->key_len;
5084 ops[i].u.flow_del.stats = &stats[i];
5088 dpif_operate(ofproto->backer->dpif, opsp, n);
5089 for (i = 0; i < n; i++) {
5090 subfacet_reset_dp_stats(subfacets[i], &stats[i]);
5091 subfacets[i]->path = SF_NOT_INSTALLED;
5092 subfacet_destroy(subfacets[i]);
5096 /* Composes the datapath actions for 'subfacet' based on its rule's actions.
5097 * Translates the actions into 'odp_actions', which the caller must have
5098 * initialized and is responsible for uninitializing. */
5100 subfacet_make_actions(struct subfacet *subfacet, const struct ofpbuf *packet,
5101 struct ofpbuf *odp_actions)
5103 struct facet *facet = subfacet->facet;
5104 struct rule_dpif *rule = facet->rule;
5105 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5107 struct action_xlate_ctx ctx;
5109 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
5110 &subfacet->initial_vals, rule, 0, packet);
5111 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, odp_actions);
5112 facet->tags = ctx.tags;
5113 facet->has_learn = ctx.has_learn;
5114 facet->has_normal = ctx.has_normal;
5115 facet->has_fin_timeout = ctx.has_fin_timeout;
5116 facet->nf_flow.output_iface = ctx.nf_output_iface;
5117 facet->mirrors = ctx.mirrors;
5119 subfacet->slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
5120 if (subfacet->actions_len != odp_actions->size
5121 || memcmp(subfacet->actions, odp_actions->data, odp_actions->size)) {
5122 free(subfacet->actions);
5123 subfacet->actions_len = odp_actions->size;
5124 subfacet->actions = xmemdup(odp_actions->data, odp_actions->size);
5128 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
5129 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
5130 * in the datapath will be zeroed and 'stats' will be updated with traffic new
5131 * since 'subfacet' was last updated.
5133 * Returns 0 if successful, otherwise a positive errno value. */
5135 subfacet_install(struct subfacet *subfacet,
5136 const struct nlattr *actions, size_t actions_len,
5137 struct dpif_flow_stats *stats,
5138 enum slow_path_reason slow)
5140 struct facet *facet = subfacet->facet;
5141 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5142 enum subfacet_path path = subfacet_want_path(slow);
5143 uint64_t slow_path_stub[128 / 8];
5144 enum dpif_flow_put_flags flags;
5147 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
5149 flags |= DPIF_FP_ZERO_STATS;
5152 if (path == SF_SLOW_PATH) {
5153 compose_slow_path(ofproto, &facet->flow, slow,
5154 slow_path_stub, sizeof slow_path_stub,
5155 &actions, &actions_len);
5158 ret = dpif_flow_put(ofproto->backer->dpif, flags, subfacet->key,
5159 subfacet->key_len, actions, actions_len, stats);
5162 subfacet_reset_dp_stats(subfacet, stats);
5166 subfacet->path = path;
5172 subfacet_reinstall(struct subfacet *subfacet, struct dpif_flow_stats *stats)
5174 return subfacet_install(subfacet, subfacet->actions, subfacet->actions_len,
5175 stats, subfacet->slow);
5178 /* If 'subfacet' is installed in the datapath, uninstalls it. */
5180 subfacet_uninstall(struct subfacet *subfacet)
5182 if (subfacet->path != SF_NOT_INSTALLED) {
5183 struct rule_dpif *rule = subfacet->facet->rule;
5184 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5185 struct dpif_flow_stats stats;
5188 error = dpif_flow_del(ofproto->backer->dpif, subfacet->key,
5189 subfacet->key_len, &stats);
5190 subfacet_reset_dp_stats(subfacet, &stats);
5192 subfacet_update_stats(subfacet, &stats);
5194 subfacet->path = SF_NOT_INSTALLED;
5196 ovs_assert(subfacet->dp_packet_count == 0);
5197 ovs_assert(subfacet->dp_byte_count == 0);
5201 /* Resets 'subfacet''s datapath statistics counters. This should be called
5202 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
5203 * non-null, it should contain the statistics returned by dpif when 'subfacet'
5204 * was reset in the datapath. 'stats' will be modified to include only
5205 * statistics new since 'subfacet' was last updated. */
5207 subfacet_reset_dp_stats(struct subfacet *subfacet,
5208 struct dpif_flow_stats *stats)
5211 && subfacet->dp_packet_count <= stats->n_packets
5212 && subfacet->dp_byte_count <= stats->n_bytes) {
5213 stats->n_packets -= subfacet->dp_packet_count;
5214 stats->n_bytes -= subfacet->dp_byte_count;
5217 subfacet->dp_packet_count = 0;
5218 subfacet->dp_byte_count = 0;
5221 /* Updates 'subfacet''s used time. The caller is responsible for calling
5222 * facet_push_stats() to update the flows which 'subfacet' resubmits into. */
5224 subfacet_update_time(struct subfacet *subfacet, long long int used)
5226 if (used > subfacet->used) {
5227 subfacet->used = used;
5228 facet_update_time(subfacet->facet, used);
5232 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
5234 * Because of the meaning of a subfacet's counters, it only makes sense to do
5235 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
5236 * represents a packet that was sent by hand or if it represents statistics
5237 * that have been cleared out of the datapath. */
5239 subfacet_update_stats(struct subfacet *subfacet,
5240 const struct dpif_flow_stats *stats)
5242 if (stats->n_packets || stats->used > subfacet->used) {
5243 struct facet *facet = subfacet->facet;
5245 subfacet_update_time(subfacet, stats->used);
5246 facet->packet_count += stats->n_packets;
5247 facet->byte_count += stats->n_bytes;
5248 facet->tcp_flags |= stats->tcp_flags;
5249 facet_push_stats(facet);
5250 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
5256 static struct rule_dpif *
5257 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow)
5259 struct rule_dpif *rule;
5261 rule = rule_dpif_lookup__(ofproto, flow, 0);
5266 return rule_dpif_miss_rule(ofproto, flow);
5269 static struct rule_dpif *
5270 rule_dpif_lookup__(struct ofproto_dpif *ofproto, const struct flow *flow,
5273 struct cls_rule *cls_rule;
5274 struct classifier *cls;
5276 if (table_id >= N_TABLES) {
5280 cls = &ofproto->up.tables[table_id].cls;
5281 if (flow->nw_frag & FLOW_NW_FRAG_ANY
5282 && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
5283 /* For OFPC_NORMAL frag_handling, we must pretend that transport ports
5284 * are unavailable. */
5285 struct flow ofpc_normal_flow = *flow;
5286 ofpc_normal_flow.tp_src = htons(0);
5287 ofpc_normal_flow.tp_dst = htons(0);
5288 cls_rule = classifier_lookup(cls, &ofpc_normal_flow);
5290 cls_rule = classifier_lookup(cls, flow);
5292 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
5295 static struct rule_dpif *
5296 rule_dpif_miss_rule(struct ofproto_dpif *ofproto, const struct flow *flow)
5298 struct ofport_dpif *port;
5300 port = get_ofp_port(ofproto, flow->in_port);
5302 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, flow->in_port);
5303 return ofproto->miss_rule;
5306 if (port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN) {
5307 return ofproto->no_packet_in_rule;
5309 return ofproto->miss_rule;
5313 complete_operation(struct rule_dpif *rule)
5315 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5317 rule_invalidate(rule);
5319 struct dpif_completion *c = xmalloc(sizeof *c);
5320 c->op = rule->up.pending;
5321 list_push_back(&ofproto->completions, &c->list_node);
5323 ofoperation_complete(rule->up.pending, 0);
5327 static struct rule *
5330 struct rule_dpif *rule = xmalloc(sizeof *rule);
5335 rule_dealloc(struct rule *rule_)
5337 struct rule_dpif *rule = rule_dpif_cast(rule_);
5342 rule_construct(struct rule *rule_)
5344 struct rule_dpif *rule = rule_dpif_cast(rule_);
5345 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5346 struct rule_dpif *victim;
5349 rule->packet_count = 0;
5350 rule->byte_count = 0;
5352 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
5353 if (victim && !list_is_empty(&victim->facets)) {
5354 struct facet *facet;
5356 rule->facets = victim->facets;
5357 list_moved(&rule->facets);
5358 LIST_FOR_EACH (facet, list_node, &rule->facets) {
5359 /* XXX: We're only clearing our local counters here. It's possible
5360 * that quite a few packets are unaccounted for in the datapath
5361 * statistics. These will be accounted to the new rule instead of
5362 * cleared as required. This could be fixed by clearing out the
5363 * datapath statistics for this facet, but currently it doesn't
5365 facet_reset_counters(facet);
5369 /* Must avoid list_moved() in this case. */
5370 list_init(&rule->facets);
5373 table_id = rule->up.table_id;
5375 rule->tag = victim->tag;
5376 } else if (table_id == 0) {
5381 miniflow_expand(&rule->up.cr.match.flow, &flow);
5382 rule->tag = rule_calculate_tag(&flow, &rule->up.cr.match.mask,
5383 ofproto->tables[table_id].basis);
5386 complete_operation(rule);
5391 rule_destruct(struct rule *rule_)
5393 struct rule_dpif *rule = rule_dpif_cast(rule_);
5394 struct facet *facet, *next_facet;
5396 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
5397 facet_revalidate(facet);
5400 complete_operation(rule);
5404 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
5406 struct rule_dpif *rule = rule_dpif_cast(rule_);
5407 struct facet *facet;
5409 /* Start from historical data for 'rule' itself that are no longer tracked
5410 * in facets. This counts, for example, facets that have expired. */
5411 *packets = rule->packet_count;
5412 *bytes = rule->byte_count;
5414 /* Add any statistics that are tracked by facets. This includes
5415 * statistical data recently updated by ofproto_update_stats() as well as
5416 * stats for packets that were executed "by hand" via dpif_execute(). */
5417 LIST_FOR_EACH (facet, list_node, &rule->facets) {
5418 *packets += facet->packet_count;
5419 *bytes += facet->byte_count;
5424 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
5425 struct ofpbuf *packet)
5427 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5428 struct initial_vals initial_vals;
5429 struct dpif_flow_stats stats;
5430 struct action_xlate_ctx ctx;
5431 uint64_t odp_actions_stub[1024 / 8];
5432 struct ofpbuf odp_actions;
5434 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
5435 rule_credit_stats(rule, &stats);
5437 initial_vals.vlan_tci = flow->vlan_tci;
5438 initial_vals.tunnel_ip_tos = flow->tunnel.ip_tos;
5439 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5440 action_xlate_ctx_init(&ctx, ofproto, flow, &initial_vals,
5441 rule, stats.tcp_flags, packet);
5442 ctx.resubmit_stats = &stats;
5443 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, &odp_actions);
5445 execute_odp_actions(ofproto, flow, odp_actions.data,
5446 odp_actions.size, packet);
5448 ofpbuf_uninit(&odp_actions);
5452 rule_execute(struct rule *rule, const struct flow *flow,
5453 struct ofpbuf *packet)
5455 rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
5456 ofpbuf_delete(packet);
5461 rule_modify_actions(struct rule *rule_)
5463 struct rule_dpif *rule = rule_dpif_cast(rule_);
5465 complete_operation(rule);
5468 /* Sends 'packet' out 'ofport'.
5469 * May modify 'packet'.
5470 * Returns 0 if successful, otherwise a positive errno value. */
5472 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
5474 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
5475 uint64_t odp_actions_stub[1024 / 8];
5476 struct ofpbuf key, odp_actions;
5477 struct odputil_keybuf keybuf;
5482 flow_extract(packet, 0, 0, NULL, OFPP_LOCAL, &flow);
5483 if (netdev_vport_is_patch(ofport->up.netdev)) {
5484 struct ofproto_dpif *peer_ofproto;
5485 struct dpif_flow_stats stats;
5486 struct ofport_dpif *peer;
5487 struct rule_dpif *rule;
5489 peer = ofport_get_peer(ofport);
5494 dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
5495 netdev_vport_inc_tx(ofport->up.netdev, &stats);
5496 netdev_vport_inc_rx(peer->up.netdev, &stats);
5498 flow.in_port = peer->up.ofp_port;
5499 peer_ofproto = ofproto_dpif_cast(peer->up.ofproto);
5500 rule = rule_dpif_lookup(peer_ofproto, &flow);
5501 rule_dpif_execute(rule, &flow, packet);
5506 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5508 if (ofport->tnl_port) {
5509 struct dpif_flow_stats stats;
5511 odp_port = tnl_port_send(ofport->tnl_port, &flow);
5512 if (odp_port == OVSP_NONE) {
5516 dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
5517 netdev_vport_inc_tx(ofport->up.netdev, &stats);
5518 odp_put_tunnel_action(&flow.tunnel, &odp_actions);
5519 odp_put_skb_mark_action(flow.skb_mark, &odp_actions);
5521 odp_port = vsp_realdev_to_vlandev(ofproto, ofport->odp_port,
5523 if (odp_port != ofport->odp_port) {
5524 eth_pop_vlan(packet);
5525 flow.vlan_tci = htons(0);
5529 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5530 odp_flow_key_from_flow(&key, &flow,
5531 ofp_port_to_odp_port(ofproto, flow.in_port));
5533 compose_sflow_action(ofproto, &odp_actions, &flow, odp_port);
5535 nl_msg_put_u32(&odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port);
5536 error = dpif_execute(ofproto->backer->dpif,
5538 odp_actions.data, odp_actions.size,
5540 ofpbuf_uninit(&odp_actions);
5543 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)",
5544 ofproto->up.name, odp_port, strerror(error));
5546 ofproto_update_local_port_stats(ofport->up.ofproto, packet->size, 0);
5550 /* OpenFlow to datapath action translation. */
5552 static bool may_receive(const struct ofport_dpif *, struct action_xlate_ctx *);
5553 static void do_xlate_actions(const struct ofpact *, size_t ofpacts_len,
5554 struct action_xlate_ctx *);
5555 static void xlate_normal(struct action_xlate_ctx *);
5557 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
5558 * The action will state 'slow' as the reason that the action is in the slow
5559 * path. (This is purely informational: it allows a human viewing "ovs-dpctl
5560 * dump-flows" output to see why a flow is in the slow path.)
5562 * The 'stub_size' bytes in 'stub' will be used to store the action.
5563 * 'stub_size' must be large enough for the action.
5565 * The action and its size will be stored in '*actionsp' and '*actions_lenp',
5568 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
5569 enum slow_path_reason slow,
5570 uint64_t *stub, size_t stub_size,
5571 const struct nlattr **actionsp, size_t *actions_lenp)
5573 union user_action_cookie cookie;
5576 cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
5577 cookie.slow_path.unused = 0;
5578 cookie.slow_path.reason = slow;
5580 ofpbuf_use_stack(&buf, stub, stub_size);
5581 if (slow & (SLOW_CFM | SLOW_LACP | SLOW_STP)) {
5582 uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif, UINT32_MAX);
5583 odp_put_userspace_action(pid, &cookie, sizeof cookie, &buf);
5585 put_userspace_action(ofproto, &buf, flow, &cookie);
5587 *actionsp = buf.data;
5588 *actions_lenp = buf.size;
5592 put_userspace_action(const struct ofproto_dpif *ofproto,
5593 struct ofpbuf *odp_actions,
5594 const struct flow *flow,
5595 const union user_action_cookie *cookie)
5599 pid = dpif_port_get_pid(ofproto->backer->dpif,
5600 ofp_port_to_odp_port(ofproto, flow->in_port));
5602 return odp_put_userspace_action(pid, cookie, sizeof *cookie, odp_actions);
5606 compose_sflow_cookie(const struct ofproto_dpif *ofproto,
5607 ovs_be16 vlan_tci, uint32_t odp_port,
5608 unsigned int n_outputs, union user_action_cookie *cookie)
5612 cookie->type = USER_ACTION_COOKIE_SFLOW;
5613 cookie->sflow.vlan_tci = vlan_tci;
5615 /* See http://www.sflow.org/sflow_version_5.txt (search for "Input/output
5616 * port information") for the interpretation of cookie->output. */
5617 switch (n_outputs) {
5619 /* 0x40000000 | 256 means "packet dropped for unknown reason". */
5620 cookie->sflow.output = 0x40000000 | 256;
5624 ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port);
5626 cookie->sflow.output = ifindex;
5631 /* 0x80000000 means "multiple output ports. */
5632 cookie->sflow.output = 0x80000000 | n_outputs;
5637 /* Compose SAMPLE action for sFlow. */
5639 compose_sflow_action(const struct ofproto_dpif *ofproto,
5640 struct ofpbuf *odp_actions,
5641 const struct flow *flow,
5644 uint32_t probability;
5645 union user_action_cookie cookie;
5646 size_t sample_offset, actions_offset;
5649 if (!ofproto->sflow || flow->in_port == OFPP_NONE) {
5653 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
5655 /* Number of packets out of UINT_MAX to sample. */
5656 probability = dpif_sflow_get_probability(ofproto->sflow);
5657 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
5659 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
5660 compose_sflow_cookie(ofproto, htons(0), odp_port,
5661 odp_port == OVSP_NONE ? 0 : 1, &cookie);
5662 cookie_offset = put_userspace_action(ofproto, odp_actions, flow, &cookie);
5664 nl_msg_end_nested(odp_actions, actions_offset);
5665 nl_msg_end_nested(odp_actions, sample_offset);
5666 return cookie_offset;
5669 /* SAMPLE action must be first action in any given list of actions.
5670 * At this point we do not have all information required to build it. So try to
5671 * build sample action as complete as possible. */
5673 add_sflow_action(struct action_xlate_ctx *ctx)
5675 ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto,
5677 &ctx->flow, OVSP_NONE);
5678 ctx->sflow_odp_port = 0;
5679 ctx->sflow_n_outputs = 0;
5682 /* Fix SAMPLE action according to data collected while composing ODP actions.
5683 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
5684 * USERSPACE action's user-cookie which is required for sflow. */
5686 fix_sflow_action(struct action_xlate_ctx *ctx)
5688 const struct flow *base = &ctx->base_flow;
5689 union user_action_cookie *cookie;
5691 if (!ctx->user_cookie_offset) {
5695 cookie = ofpbuf_at(ctx->odp_actions, ctx->user_cookie_offset,
5697 ovs_assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
5699 compose_sflow_cookie(ctx->ofproto, base->vlan_tci,
5700 ctx->sflow_odp_port, ctx->sflow_n_outputs, cookie);
5704 compose_output_action__(struct action_xlate_ctx *ctx, uint16_t ofp_port,
5707 const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
5708 ovs_be16 flow_vlan_tci = ctx->flow.vlan_tci;
5709 ovs_be64 flow_tun_id = ctx->flow.tunnel.tun_id;
5710 uint8_t flow_nw_tos = ctx->flow.nw_tos;
5711 struct priority_to_dscp *pdscp;
5712 uint32_t out_port, odp_port;
5714 /* If 'struct flow' gets additional metadata, we'll need to zero it out
5715 * before traversing a patch port. */
5716 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 19);
5719 xlate_report(ctx, "Nonexistent output port");
5721 } else if (ofport->up.pp.config & OFPUTIL_PC_NO_FWD) {
5722 xlate_report(ctx, "OFPPC_NO_FWD set, skipping output");
5724 } else if (check_stp && !stp_forward_in_state(ofport->stp_state)) {
5725 xlate_report(ctx, "STP not in forwarding state, skipping output");
5729 if (netdev_vport_is_patch(ofport->up.netdev)) {
5730 struct ofport_dpif *peer = ofport_get_peer(ofport);
5731 struct flow old_flow = ctx->flow;
5732 const struct ofproto_dpif *peer_ofproto;
5733 enum slow_path_reason special;
5734 struct ofport_dpif *in_port;
5737 xlate_report(ctx, "Nonexistent patch port peer");
5741 peer_ofproto = ofproto_dpif_cast(peer->up.ofproto);
5742 if (peer_ofproto->backer != ctx->ofproto->backer) {
5743 xlate_report(ctx, "Patch port peer on a different datapath");
5747 ctx->ofproto = ofproto_dpif_cast(peer->up.ofproto);
5748 ctx->flow.in_port = peer->up.ofp_port;
5749 ctx->flow.metadata = htonll(0);
5750 memset(&ctx->flow.tunnel, 0, sizeof ctx->flow.tunnel);
5751 memset(ctx->flow.regs, 0, sizeof ctx->flow.regs);
5753 in_port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
5754 special = process_special(ctx->ofproto, &ctx->flow, in_port,
5757 ctx->slow |= special;
5758 } else if (!in_port || may_receive(in_port, ctx)) {
5759 if (!in_port || stp_forward_in_state(in_port->stp_state)) {
5760 xlate_table_action(ctx, ctx->flow.in_port, 0, true);
5762 /* Forwarding is disabled by STP. Let OFPP_NORMAL and the
5763 * learning action look at the packet, then drop it. */
5764 struct flow old_base_flow = ctx->base_flow;
5765 size_t old_size = ctx->odp_actions->size;
5766 xlate_table_action(ctx, ctx->flow.in_port, 0, true);
5767 ctx->base_flow = old_base_flow;
5768 ctx->odp_actions->size = old_size;
5772 ctx->flow = old_flow;
5773 ctx->ofproto = ofproto_dpif_cast(ofport->up.ofproto);
5775 if (ctx->resubmit_stats) {
5776 netdev_vport_inc_tx(ofport->up.netdev, ctx->resubmit_stats);
5777 netdev_vport_inc_rx(peer->up.netdev, ctx->resubmit_stats);
5783 pdscp = get_priority(ofport, ctx->flow.skb_priority);
5785 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
5786 ctx->flow.nw_tos |= pdscp->dscp;
5789 if (ofport->tnl_port) {
5790 odp_port = tnl_port_send(ofport->tnl_port, &ctx->flow);
5791 if (odp_port == OVSP_NONE) {
5792 xlate_report(ctx, "Tunneling decided against output");
5796 if (ctx->resubmit_stats) {
5797 netdev_vport_inc_tx(ofport->up.netdev, ctx->resubmit_stats);
5799 out_port = odp_port;
5800 commit_odp_tunnel_action(&ctx->flow, &ctx->base_flow,
5803 odp_port = ofport->odp_port;
5804 out_port = vsp_realdev_to_vlandev(ctx->ofproto, odp_port,
5805 ctx->flow.vlan_tci);
5806 if (out_port != odp_port) {
5807 ctx->flow.vlan_tci = htons(0);
5810 commit_odp_actions(&ctx->flow, &ctx->base_flow, ctx->odp_actions);
5811 nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_OUTPUT, out_port);
5813 ctx->sflow_odp_port = odp_port;
5814 ctx->sflow_n_outputs++;
5815 ctx->nf_output_iface = ofp_port;
5816 ctx->flow.tunnel.tun_id = flow_tun_id;
5817 ctx->flow.vlan_tci = flow_vlan_tci;
5818 ctx->flow.nw_tos = flow_nw_tos;
5822 compose_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port)
5824 compose_output_action__(ctx, ofp_port, true);
5828 tag_the_flow(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
5830 struct ofproto_dpif *ofproto = ctx->ofproto;
5831 uint8_t table_id = ctx->table_id;
5833 if (table_id > 0 && table_id < N_TABLES) {
5834 struct table_dpif *table = &ofproto->tables[table_id];
5835 if (table->other_table) {
5836 ctx->tags |= (rule && rule->tag
5838 : rule_calculate_tag(&ctx->flow,
5839 &table->other_table->mask,
5845 /* Common rule processing in one place to avoid duplicating code. */
5846 static struct rule_dpif *
5847 ctx_rule_hooks(struct action_xlate_ctx *ctx, struct rule_dpif *rule,
5850 if (ctx->resubmit_hook) {
5851 ctx->resubmit_hook(ctx, rule);
5853 if (rule == NULL && may_packet_in) {
5855 * check if table configuration flags
5856 * OFPTC_TABLE_MISS_CONTROLLER, default.
5857 * OFPTC_TABLE_MISS_CONTINUE,
5858 * OFPTC_TABLE_MISS_DROP
5859 * When OF1.0, OFPTC_TABLE_MISS_CONTINUE is used. What to do?
5861 rule = rule_dpif_miss_rule(ctx->ofproto, &ctx->flow);
5863 if (rule && ctx->resubmit_stats) {
5864 rule_credit_stats(rule, ctx->resubmit_stats);
5870 xlate_table_action(struct action_xlate_ctx *ctx,
5871 uint16_t in_port, uint8_t table_id, bool may_packet_in)
5873 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
5874 struct rule_dpif *rule;
5875 uint16_t old_in_port = ctx->flow.in_port;
5876 uint8_t old_table_id = ctx->table_id;
5878 ctx->table_id = table_id;
5880 /* Look up a flow with 'in_port' as the input port. */
5881 ctx->flow.in_port = in_port;
5882 rule = rule_dpif_lookup__(ctx->ofproto, &ctx->flow, table_id);
5884 tag_the_flow(ctx, rule);
5886 /* Restore the original input port. Otherwise OFPP_NORMAL and
5887 * OFPP_IN_PORT will have surprising behavior. */
5888 ctx->flow.in_port = old_in_port;
5890 rule = ctx_rule_hooks(ctx, rule, may_packet_in);
5893 struct rule_dpif *old_rule = ctx->rule;
5897 do_xlate_actions(rule->up.ofpacts, rule->up.ofpacts_len, ctx);
5898 ctx->rule = old_rule;
5902 ctx->table_id = old_table_id;
5904 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
5906 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
5907 MAX_RESUBMIT_RECURSION);
5908 ctx->max_resubmit_trigger = true;
5913 xlate_ofpact_resubmit(struct action_xlate_ctx *ctx,
5914 const struct ofpact_resubmit *resubmit)
5919 in_port = resubmit->in_port;
5920 if (in_port == OFPP_IN_PORT) {
5921 in_port = ctx->flow.in_port;
5924 table_id = resubmit->table_id;
5925 if (table_id == 255) {
5926 table_id = ctx->table_id;
5929 xlate_table_action(ctx, in_port, table_id, false);
5933 flood_packets(struct action_xlate_ctx *ctx, bool all)
5935 struct ofport_dpif *ofport;
5937 HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) {
5938 uint16_t ofp_port = ofport->up.ofp_port;
5940 if (ofp_port == ctx->flow.in_port) {
5945 compose_output_action__(ctx, ofp_port, false);
5946 } else if (!(ofport->up.pp.config & OFPUTIL_PC_NO_FLOOD)) {
5947 compose_output_action(ctx, ofp_port);
5951 ctx->nf_output_iface = NF_OUT_FLOOD;
5955 execute_controller_action(struct action_xlate_ctx *ctx, int len,
5956 enum ofp_packet_in_reason reason,
5957 uint16_t controller_id)
5959 struct ofputil_packet_in pin;
5960 struct ofpbuf *packet;
5962 ctx->slow |= SLOW_CONTROLLER;
5967 packet = ofpbuf_clone(ctx->packet);
5969 if (packet->l2 && packet->l3) {
5970 struct eth_header *eh;
5971 uint16_t mpls_depth;
5973 eth_pop_vlan(packet);
5976 memcpy(eh->eth_src, ctx->flow.dl_src, sizeof eh->eth_src);
5977 memcpy(eh->eth_dst, ctx->flow.dl_dst, sizeof eh->eth_dst);
5979 if (ctx->flow.vlan_tci & htons(VLAN_CFI)) {
5980 eth_push_vlan(packet, ctx->flow.vlan_tci);
5983 mpls_depth = eth_mpls_depth(packet);
5985 if (mpls_depth < ctx->flow.mpls_depth) {
5986 push_mpls(packet, ctx->flow.dl_type, ctx->flow.mpls_lse);
5987 } else if (mpls_depth > ctx->flow.mpls_depth) {
5988 pop_mpls(packet, ctx->flow.dl_type);
5989 } else if (mpls_depth) {
5990 set_mpls_lse(packet, ctx->flow.mpls_lse);
5994 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
5995 packet_set_ipv4(packet, ctx->flow.nw_src, ctx->flow.nw_dst,
5996 ctx->flow.nw_tos, ctx->flow.nw_ttl);
6000 if (ctx->flow.nw_proto == IPPROTO_TCP) {
6001 packet_set_tcp_port(packet, ctx->flow.tp_src,
6003 } else if (ctx->flow.nw_proto == IPPROTO_UDP) {
6004 packet_set_udp_port(packet, ctx->flow.tp_src,
6011 pin.packet = packet->data;
6012 pin.packet_len = packet->size;
6013 pin.reason = reason;
6014 pin.controller_id = controller_id;
6015 pin.table_id = ctx->table_id;
6016 pin.cookie = ctx->rule ? ctx->rule->up.flow_cookie : 0;
6019 flow_get_metadata(&ctx->flow, &pin.fmd);
6021 connmgr_send_packet_in(ctx->ofproto->up.connmgr, &pin);
6022 ofpbuf_delete(packet);
6026 execute_mpls_push_action(struct action_xlate_ctx *ctx, ovs_be16 eth_type)
6028 ovs_assert(eth_type_mpls(eth_type));
6030 if (ctx->base_flow.mpls_depth) {
6031 ctx->flow.mpls_lse &= ~htonl(MPLS_BOS_MASK);
6032 ctx->flow.mpls_depth++;
6037 if (ctx->flow.dl_type == htons(ETH_TYPE_IPV6)) {
6038 label = htonl(0x2); /* IPV6 Explicit Null. */
6040 label = htonl(0x0); /* IPV4 Explicit Null. */
6042 tc = (ctx->flow.nw_tos & IP_DSCP_MASK) >> 2;
6043 ttl = ctx->flow.nw_ttl ? ctx->flow.nw_ttl : 0x40;
6044 ctx->flow.mpls_lse = set_mpls_lse_values(ttl, tc, 1, label);
6045 ctx->flow.encap_dl_type = ctx->flow.dl_type;
6046 ctx->flow.mpls_depth = 1;
6048 ctx->flow.dl_type = eth_type;
6052 execute_mpls_pop_action(struct action_xlate_ctx *ctx, ovs_be16 eth_type)
6054 ovs_assert(eth_type_mpls(ctx->flow.dl_type));
6055 ovs_assert(!eth_type_mpls(eth_type));
6057 if (ctx->flow.mpls_depth) {
6058 ctx->flow.mpls_depth--;
6059 ctx->flow.mpls_lse = htonl(0);
6060 if (!ctx->flow.mpls_depth) {
6061 ctx->flow.dl_type = eth_type;
6062 ctx->flow.encap_dl_type = htons(0);
6068 compose_dec_ttl(struct action_xlate_ctx *ctx, struct ofpact_cnt_ids *ids)
6070 if (ctx->flow.dl_type != htons(ETH_TYPE_IP) &&
6071 ctx->flow.dl_type != htons(ETH_TYPE_IPV6)) {
6075 if (ctx->flow.nw_ttl > 1) {
6081 for (i = 0; i < ids->n_controllers; i++) {
6082 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL,
6086 /* Stop processing for current table. */
6092 execute_set_mpls_ttl_action(struct action_xlate_ctx *ctx, uint8_t ttl)
6094 if (!eth_type_mpls(ctx->flow.dl_type)) {
6098 set_mpls_lse_ttl(&ctx->flow.mpls_lse, ttl);
6103 execute_dec_mpls_ttl_action(struct action_xlate_ctx *ctx)
6105 uint8_t ttl = mpls_lse_to_ttl(ctx->flow.mpls_lse);
6107 if (!eth_type_mpls(ctx->flow.dl_type)) {
6113 set_mpls_lse_ttl(&ctx->flow.mpls_lse, ttl);
6116 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL, 0);
6118 /* Stop processing for current table. */
6124 xlate_output_action(struct action_xlate_ctx *ctx,
6125 uint16_t port, uint16_t max_len, bool may_packet_in)
6127 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
6129 ctx->nf_output_iface = NF_OUT_DROP;
6133 compose_output_action(ctx, ctx->flow.in_port);
6136 xlate_table_action(ctx, ctx->flow.in_port, 0, may_packet_in);
6142 flood_packets(ctx, false);
6145 flood_packets(ctx, true);
6147 case OFPP_CONTROLLER:
6148 execute_controller_action(ctx, max_len, OFPR_ACTION, 0);
6154 if (port != ctx->flow.in_port) {
6155 compose_output_action(ctx, port);
6157 xlate_report(ctx, "skipping output to input port");
6162 if (prev_nf_output_iface == NF_OUT_FLOOD) {
6163 ctx->nf_output_iface = NF_OUT_FLOOD;
6164 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
6165 ctx->nf_output_iface = prev_nf_output_iface;
6166 } else if (prev_nf_output_iface != NF_OUT_DROP &&
6167 ctx->nf_output_iface != NF_OUT_FLOOD) {
6168 ctx->nf_output_iface = NF_OUT_MULTI;
6173 xlate_output_reg_action(struct action_xlate_ctx *ctx,
6174 const struct ofpact_output_reg *or)
6176 uint64_t port = mf_get_subfield(&or->src, &ctx->flow);
6177 if (port <= UINT16_MAX) {
6178 xlate_output_action(ctx, port, or->max_len, false);
6183 xlate_enqueue_action(struct action_xlate_ctx *ctx,
6184 const struct ofpact_enqueue *enqueue)
6186 uint16_t ofp_port = enqueue->port;
6187 uint32_t queue_id = enqueue->queue;
6188 uint32_t flow_priority, priority;
6191 /* Translate queue to priority. */
6192 error = dpif_queue_to_priority(ctx->ofproto->backer->dpif,
6193 queue_id, &priority);
6195 /* Fall back to ordinary output action. */
6196 xlate_output_action(ctx, enqueue->port, 0, false);
6200 /* Check output port. */
6201 if (ofp_port == OFPP_IN_PORT) {
6202 ofp_port = ctx->flow.in_port;
6203 } else if (ofp_port == ctx->flow.in_port) {
6207 /* Add datapath actions. */
6208 flow_priority = ctx->flow.skb_priority;
6209 ctx->flow.skb_priority = priority;
6210 compose_output_action(ctx, ofp_port);
6211 ctx->flow.skb_priority = flow_priority;
6213 /* Update NetFlow output port. */
6214 if (ctx->nf_output_iface == NF_OUT_DROP) {
6215 ctx->nf_output_iface = ofp_port;
6216 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
6217 ctx->nf_output_iface = NF_OUT_MULTI;
6222 xlate_set_queue_action(struct action_xlate_ctx *ctx, uint32_t queue_id)
6224 uint32_t skb_priority;
6226 if (!dpif_queue_to_priority(ctx->ofproto->backer->dpif,
6227 queue_id, &skb_priority)) {
6228 ctx->flow.skb_priority = skb_priority;
6230 /* Couldn't translate queue to a priority. Nothing to do. A warning
6231 * has already been logged. */
6235 struct xlate_reg_state {
6241 slave_enabled_cb(uint16_t ofp_port, void *ofproto_)
6243 struct ofproto_dpif *ofproto = ofproto_;
6244 struct ofport_dpif *port;
6254 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
6257 port = get_ofp_port(ofproto, ofp_port);
6258 return port ? port->may_enable : false;
6263 xlate_bundle_action(struct action_xlate_ctx *ctx,
6264 const struct ofpact_bundle *bundle)
6268 port = bundle_execute(bundle, &ctx->flow, slave_enabled_cb, ctx->ofproto);
6269 if (bundle->dst.field) {
6270 nxm_reg_load(&bundle->dst, port, &ctx->flow);
6272 xlate_output_action(ctx, port, 0, false);
6277 xlate_learn_action(struct action_xlate_ctx *ctx,
6278 const struct ofpact_learn *learn)
6280 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
6281 struct ofputil_flow_mod fm;
6282 uint64_t ofpacts_stub[1024 / 8];
6283 struct ofpbuf ofpacts;
6286 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
6287 learn_execute(learn, &ctx->flow, &fm, &ofpacts);
6289 error = ofproto_flow_mod(&ctx->ofproto->up, &fm);
6290 if (error && !VLOG_DROP_WARN(&rl)) {
6291 VLOG_WARN("learning action failed to modify flow table (%s)",
6292 ofperr_get_name(error));
6295 ofpbuf_uninit(&ofpacts);
6298 /* Reduces '*timeout' to no more than 'max'. A value of zero in either case
6299 * means "infinite". */
6301 reduce_timeout(uint16_t max, uint16_t *timeout)
6303 if (max && (!*timeout || *timeout > max)) {
6309 xlate_fin_timeout(struct action_xlate_ctx *ctx,
6310 const struct ofpact_fin_timeout *oft)
6312 if (ctx->tcp_flags & (TCP_FIN | TCP_RST) && ctx->rule) {
6313 struct rule_dpif *rule = ctx->rule;
6315 reduce_timeout(oft->fin_idle_timeout, &rule->up.idle_timeout);
6316 reduce_timeout(oft->fin_hard_timeout, &rule->up.hard_timeout);
6321 may_receive(const struct ofport_dpif *port, struct action_xlate_ctx *ctx)
6323 if (port->up.pp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
6324 ? OFPUTIL_PC_NO_RECV_STP
6325 : OFPUTIL_PC_NO_RECV)) {
6329 /* Only drop packets here if both forwarding and learning are
6330 * disabled. If just learning is enabled, we need to have
6331 * OFPP_NORMAL and the learning action have a look at the packet
6332 * before we can drop it. */
6333 if (!stp_forward_in_state(port->stp_state)
6334 && !stp_learn_in_state(port->stp_state)) {
6342 tunnel_ecn_ok(struct action_xlate_ctx *ctx)
6344 if (is_ip_any(&ctx->base_flow)
6345 && (ctx->base_flow.tunnel.ip_tos & IP_ECN_MASK) == IP_ECN_CE) {
6346 if ((ctx->base_flow.nw_tos & IP_ECN_MASK) == IP_ECN_NOT_ECT) {
6347 VLOG_WARN_RL(&rl, "dropping tunnel packet marked ECN CE"
6348 " but is not ECN capable");
6351 /* Set the ECN CE value in the tunneled packet. */
6352 ctx->flow.nw_tos |= IP_ECN_CE;
6360 do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len,
6361 struct action_xlate_ctx *ctx)
6363 bool was_evictable = true;
6364 const struct ofpact *a;
6367 /* Don't let the rule we're working on get evicted underneath us. */
6368 was_evictable = ctx->rule->up.evictable;
6369 ctx->rule->up.evictable = false;
6372 do_xlate_actions_again:
6373 OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) {
6374 struct ofpact_controller *controller;
6375 const struct ofpact_metadata *metadata;
6383 xlate_output_action(ctx, ofpact_get_OUTPUT(a)->port,
6384 ofpact_get_OUTPUT(a)->max_len, true);
6387 case OFPACT_CONTROLLER:
6388 controller = ofpact_get_CONTROLLER(a);
6389 execute_controller_action(ctx, controller->max_len,
6391 controller->controller_id);
6394 case OFPACT_ENQUEUE:
6395 xlate_enqueue_action(ctx, ofpact_get_ENQUEUE(a));
6398 case OFPACT_SET_VLAN_VID:
6399 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
6400 ctx->flow.vlan_tci |= (htons(ofpact_get_SET_VLAN_VID(a)->vlan_vid)
6404 case OFPACT_SET_VLAN_PCP:
6405 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
6406 ctx->flow.vlan_tci |= htons((ofpact_get_SET_VLAN_PCP(a)->vlan_pcp
6411 case OFPACT_STRIP_VLAN:
6412 ctx->flow.vlan_tci = htons(0);
6415 case OFPACT_PUSH_VLAN:
6416 /* XXX 802.1AD(QinQ) */
6417 ctx->flow.vlan_tci = htons(VLAN_CFI);
6420 case OFPACT_SET_ETH_SRC:
6421 memcpy(ctx->flow.dl_src, ofpact_get_SET_ETH_SRC(a)->mac,
6425 case OFPACT_SET_ETH_DST:
6426 memcpy(ctx->flow.dl_dst, ofpact_get_SET_ETH_DST(a)->mac,
6430 case OFPACT_SET_IPV4_SRC:
6431 ctx->flow.nw_src = ofpact_get_SET_IPV4_SRC(a)->ipv4;
6434 case OFPACT_SET_IPV4_DST:
6435 ctx->flow.nw_dst = ofpact_get_SET_IPV4_DST(a)->ipv4;
6438 case OFPACT_SET_IPV4_DSCP:
6439 /* OpenFlow 1.0 only supports IPv4. */
6440 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
6441 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
6442 ctx->flow.nw_tos |= ofpact_get_SET_IPV4_DSCP(a)->dscp;
6446 case OFPACT_SET_L4_SRC_PORT:
6447 ctx->flow.tp_src = htons(ofpact_get_SET_L4_SRC_PORT(a)->port);
6450 case OFPACT_SET_L4_DST_PORT:
6451 ctx->flow.tp_dst = htons(ofpact_get_SET_L4_DST_PORT(a)->port);
6454 case OFPACT_RESUBMIT:
6455 xlate_ofpact_resubmit(ctx, ofpact_get_RESUBMIT(a));
6458 case OFPACT_SET_TUNNEL:
6459 ctx->flow.tunnel.tun_id = htonll(ofpact_get_SET_TUNNEL(a)->tun_id);
6462 case OFPACT_SET_QUEUE:
6463 xlate_set_queue_action(ctx, ofpact_get_SET_QUEUE(a)->queue_id);
6466 case OFPACT_POP_QUEUE:
6467 ctx->flow.skb_priority = ctx->orig_skb_priority;
6470 case OFPACT_REG_MOVE:
6471 nxm_execute_reg_move(ofpact_get_REG_MOVE(a), &ctx->flow);
6474 case OFPACT_REG_LOAD:
6475 nxm_execute_reg_load(ofpact_get_REG_LOAD(a), &ctx->flow);
6478 case OFPACT_STACK_PUSH:
6479 nxm_execute_stack_push(ofpact_get_STACK_PUSH(a), &ctx->flow,
6483 case OFPACT_STACK_POP:
6484 nxm_execute_stack_pop(ofpact_get_STACK_POP(a), &ctx->flow,
6488 case OFPACT_PUSH_MPLS:
6489 execute_mpls_push_action(ctx, ofpact_get_PUSH_MPLS(a)->ethertype);
6492 case OFPACT_POP_MPLS:
6493 execute_mpls_pop_action(ctx, ofpact_get_POP_MPLS(a)->ethertype);
6496 case OFPACT_SET_MPLS_TTL:
6497 if (execute_set_mpls_ttl_action(ctx, ofpact_get_SET_MPLS_TTL(a)->ttl)) {
6502 case OFPACT_DEC_MPLS_TTL:
6503 if (execute_dec_mpls_ttl_action(ctx)) {
6508 case OFPACT_DEC_TTL:
6509 if (compose_dec_ttl(ctx, ofpact_get_DEC_TTL(a))) {
6515 /* Nothing to do. */
6518 case OFPACT_MULTIPATH:
6519 multipath_execute(ofpact_get_MULTIPATH(a), &ctx->flow);
6523 ctx->ofproto->has_bundle_action = true;
6524 xlate_bundle_action(ctx, ofpact_get_BUNDLE(a));
6527 case OFPACT_OUTPUT_REG:
6528 xlate_output_reg_action(ctx, ofpact_get_OUTPUT_REG(a));
6532 ctx->has_learn = true;
6533 if (ctx->may_learn) {
6534 xlate_learn_action(ctx, ofpact_get_LEARN(a));
6542 case OFPACT_FIN_TIMEOUT:
6543 ctx->has_fin_timeout = true;
6544 xlate_fin_timeout(ctx, ofpact_get_FIN_TIMEOUT(a));
6547 case OFPACT_CLEAR_ACTIONS:
6549 * Nothing to do because writa-actions is not supported for now.
6550 * When writa-actions is supported, clear-actions also must
6551 * be supported at the same time.
6555 case OFPACT_WRITE_METADATA:
6556 metadata = ofpact_get_WRITE_METADATA(a);
6557 ctx->flow.metadata &= ~metadata->mask;
6558 ctx->flow.metadata |= metadata->metadata & metadata->mask;
6561 case OFPACT_GOTO_TABLE: {
6562 /* It is assumed that goto-table is the last action. */
6563 struct ofpact_goto_table *ogt = ofpact_get_GOTO_TABLE(a);
6564 struct rule_dpif *rule;
6566 ovs_assert(ctx->table_id < ogt->table_id);
6568 ctx->table_id = ogt->table_id;
6570 /* Look up a flow from the new table. */
6571 rule = rule_dpif_lookup__(ctx->ofproto, &ctx->flow, ctx->table_id);
6573 tag_the_flow(ctx, rule);
6575 rule = ctx_rule_hooks(ctx, rule, true);
6579 ctx->rule->up.evictable = was_evictable;
6582 was_evictable = rule->up.evictable;
6583 rule->up.evictable = false;
6585 /* Tail recursion removal. */
6586 ofpacts = rule->up.ofpacts;
6587 ofpacts_len = rule->up.ofpacts_len;
6588 goto do_xlate_actions_again;
6597 ctx->rule->up.evictable = was_evictable;
6602 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
6603 struct ofproto_dpif *ofproto, const struct flow *flow,
6604 const struct initial_vals *initial_vals,
6605 struct rule_dpif *rule,
6606 uint8_t tcp_flags, const struct ofpbuf *packet)
6608 ovs_be64 initial_tun_id = flow->tunnel.tun_id;
6610 /* Flow initialization rules:
6611 * - 'base_flow' must match the kernel's view of the packet at the
6612 * time that action processing starts. 'flow' represents any
6613 * transformations we wish to make through actions.
6614 * - By default 'base_flow' and 'flow' are the same since the input
6615 * packet matches the output before any actions are applied.
6616 * - When using VLAN splinters, 'base_flow''s VLAN is set to the value
6617 * of the received packet as seen by the kernel. If we later output
6618 * to another device without any modifications this will cause us to
6619 * insert a new tag since the original one was stripped off by the
6621 * - Tunnel 'flow' is largely cleared when transitioning between
6622 * the input and output stages since it does not make sense to output
6623 * a packet with the exact headers that it was received with (i.e.
6624 * the destination IP is us). The one exception is the tun_id, which
6625 * is preserved to allow use in later resubmit lookups and loads into
6627 * - Tunnel 'base_flow' is completely cleared since that is what the
6628 * kernel does. If we wish to maintain the original values an action
6629 * needs to be generated. */
6631 ctx->ofproto = ofproto;
6633 memset(&ctx->flow.tunnel, 0, sizeof ctx->flow.tunnel);
6634 ctx->base_flow = ctx->flow;
6635 ctx->base_flow.vlan_tci = initial_vals->vlan_tci;
6636 ctx->base_flow.tunnel.ip_tos = initial_vals->tunnel_ip_tos;
6637 ctx->flow.tunnel.tun_id = initial_tun_id;
6639 ctx->packet = packet;
6640 ctx->may_learn = packet != NULL;
6641 ctx->tcp_flags = tcp_flags;
6642 ctx->resubmit_hook = NULL;
6643 ctx->report_hook = NULL;
6644 ctx->resubmit_stats = NULL;
6647 /* Translates the 'ofpacts_len' bytes of "struct ofpacts" starting at 'ofpacts'
6648 * into datapath actions in 'odp_actions', using 'ctx'. */
6650 xlate_actions(struct action_xlate_ctx *ctx,
6651 const struct ofpact *ofpacts, size_t ofpacts_len,
6652 struct ofpbuf *odp_actions)
6654 /* Normally false. Set to true if we ever hit MAX_RESUBMIT_RECURSION, so
6655 * that in the future we always keep a copy of the original flow for
6656 * tracing purposes. */
6657 static bool hit_resubmit_limit;
6659 enum slow_path_reason special;
6660 struct ofport_dpif *in_port;
6661 struct flow orig_flow;
6663 COVERAGE_INC(ofproto_dpif_xlate);
6665 ofpbuf_clear(odp_actions);
6666 ofpbuf_reserve(odp_actions, NL_A_U32_SIZE);
6668 ctx->odp_actions = odp_actions;
6671 ctx->has_learn = false;
6672 ctx->has_normal = false;
6673 ctx->has_fin_timeout = false;
6674 ctx->nf_output_iface = NF_OUT_DROP;
6677 ctx->max_resubmit_trigger = false;
6678 ctx->orig_skb_priority = ctx->flow.skb_priority;
6682 ofpbuf_use_stub(&ctx->stack, ctx->init_stack, sizeof ctx->init_stack);
6684 if (ctx->ofproto->has_mirrors || hit_resubmit_limit) {
6685 /* Do this conditionally because the copy is expensive enough that it
6686 * shows up in profiles. */
6687 orig_flow = ctx->flow;
6690 if (ctx->flow.nw_frag & FLOW_NW_FRAG_ANY) {
6691 switch (ctx->ofproto->up.frag_handling) {
6692 case OFPC_FRAG_NORMAL:
6693 /* We must pretend that transport ports are unavailable. */
6694 ctx->flow.tp_src = ctx->base_flow.tp_src = htons(0);
6695 ctx->flow.tp_dst = ctx->base_flow.tp_dst = htons(0);
6698 case OFPC_FRAG_DROP:
6701 case OFPC_FRAG_REASM:
6704 case OFPC_FRAG_NX_MATCH:
6705 /* Nothing to do. */
6708 case OFPC_INVALID_TTL_TO_CONTROLLER:
6713 in_port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
6714 special = process_special(ctx->ofproto, &ctx->flow, in_port, ctx->packet);
6716 ctx->slow |= special;
6718 static struct vlog_rate_limit trace_rl = VLOG_RATE_LIMIT_INIT(1, 1);
6719 struct initial_vals initial_vals;
6720 uint32_t local_odp_port;
6722 initial_vals.vlan_tci = ctx->base_flow.vlan_tci;
6723 initial_vals.tunnel_ip_tos = ctx->base_flow.tunnel.ip_tos;
6725 add_sflow_action(ctx);
6727 if (tunnel_ecn_ok(ctx) && (!in_port || may_receive(in_port, ctx))) {
6728 do_xlate_actions(ofpacts, ofpacts_len, ctx);
6730 /* We've let OFPP_NORMAL and the learning action look at the
6731 * packet, so drop it now if forwarding is disabled. */
6732 if (in_port && !stp_forward_in_state(in_port->stp_state)) {
6733 ofpbuf_clear(ctx->odp_actions);
6734 add_sflow_action(ctx);
6738 if (ctx->max_resubmit_trigger && !ctx->resubmit_hook) {
6739 if (!hit_resubmit_limit) {
6740 /* We didn't record the original flow. Make sure we do from
6742 hit_resubmit_limit = true;
6743 } else if (!VLOG_DROP_ERR(&trace_rl)) {
6744 struct ds ds = DS_EMPTY_INITIALIZER;
6746 ofproto_trace(ctx->ofproto, &orig_flow, ctx->packet,
6747 &initial_vals, &ds);
6748 VLOG_ERR("Trace triggered by excessive resubmit "
6749 "recursion:\n%s", ds_cstr(&ds));
6754 local_odp_port = ofp_port_to_odp_port(ctx->ofproto, OFPP_LOCAL);
6755 if (!connmgr_may_set_up_flow(ctx->ofproto->up.connmgr, &ctx->flow,
6757 ctx->odp_actions->data,
6758 ctx->odp_actions->size)) {
6759 ctx->slow |= SLOW_IN_BAND;
6761 && connmgr_msg_in_hook(ctx->ofproto->up.connmgr, &ctx->flow,
6763 compose_output_action(ctx, OFPP_LOCAL);
6766 if (ctx->ofproto->has_mirrors) {
6767 add_mirror_actions(ctx, &orig_flow);
6769 fix_sflow_action(ctx);
6772 ofpbuf_uninit(&ctx->stack);
6775 /* Translates the 'ofpacts_len' bytes of "struct ofpact"s starting at 'ofpacts'
6776 * into datapath actions, using 'ctx', and discards the datapath actions. */
6778 xlate_actions_for_side_effects(struct action_xlate_ctx *ctx,
6779 const struct ofpact *ofpacts,
6782 uint64_t odp_actions_stub[1024 / 8];
6783 struct ofpbuf odp_actions;
6785 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
6786 xlate_actions(ctx, ofpacts, ofpacts_len, &odp_actions);
6787 ofpbuf_uninit(&odp_actions);
6791 xlate_report(struct action_xlate_ctx *ctx, const char *s)
6793 if (ctx->report_hook) {
6794 ctx->report_hook(ctx, s);
6798 /* OFPP_NORMAL implementation. */
6800 static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *);
6802 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
6803 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle',
6804 * the bundle on which the packet was received, returns the VLAN to which the
6807 * Both 'vid' and the return value are in the range 0...4095. */
6809 input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid)
6811 switch (in_bundle->vlan_mode) {
6812 case PORT_VLAN_ACCESS:
6813 return in_bundle->vlan;
6816 case PORT_VLAN_TRUNK:
6819 case PORT_VLAN_NATIVE_UNTAGGED:
6820 case PORT_VLAN_NATIVE_TAGGED:
6821 return vid ? vid : in_bundle->vlan;
6828 /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'.
6829 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
6832 * 'vid' should be the VID obtained from the 802.1Q header that was received as
6833 * part of a packet (specify 0 if there was no 802.1Q header), in the range
6836 input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn)
6838 /* Allow any VID on the OFPP_NONE port. */
6839 if (in_bundle == &ofpp_none_bundle) {
6843 switch (in_bundle->vlan_mode) {
6844 case PORT_VLAN_ACCESS:
6847 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6848 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
6849 "packet received on port %s configured as VLAN "
6850 "%"PRIu16" access port",
6851 in_bundle->ofproto->up.name, vid,
6852 in_bundle->name, in_bundle->vlan);
6858 case PORT_VLAN_NATIVE_UNTAGGED:
6859 case PORT_VLAN_NATIVE_TAGGED:
6861 /* Port must always carry its native VLAN. */
6865 case PORT_VLAN_TRUNK:
6866 if (!ofbundle_includes_vlan(in_bundle, vid)) {
6868 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6869 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet "
6870 "received on port %s not configured for trunking "
6872 in_bundle->ofproto->up.name, vid,
6873 in_bundle->name, vid);
6885 /* Given 'vlan', the VLAN that a packet belongs to, and
6886 * 'out_bundle', a bundle on which the packet is to be output, returns the VID
6887 * that should be included in the 802.1Q header. (If the return value is 0,
6888 * then the 802.1Q header should only be included in the packet if there is a
6891 * Both 'vlan' and the return value are in the range 0...4095. */
6893 output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan)
6895 switch (out_bundle->vlan_mode) {
6896 case PORT_VLAN_ACCESS:
6899 case PORT_VLAN_TRUNK:
6900 case PORT_VLAN_NATIVE_TAGGED:
6903 case PORT_VLAN_NATIVE_UNTAGGED:
6904 return vlan == out_bundle->vlan ? 0 : vlan;
6912 output_normal(struct action_xlate_ctx *ctx, const struct ofbundle *out_bundle,
6915 struct ofport_dpif *port;
6917 ovs_be16 tci, old_tci;
6919 vid = output_vlan_to_vid(out_bundle, vlan);
6920 if (!out_bundle->bond) {
6921 port = ofbundle_get_a_port(out_bundle);
6923 port = bond_choose_output_slave(out_bundle->bond, &ctx->flow,
6926 /* No slaves enabled, so drop packet. */
6931 old_tci = ctx->flow.vlan_tci;
6933 if (tci || out_bundle->use_priority_tags) {
6934 tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK);
6936 tci |= htons(VLAN_CFI);
6939 ctx->flow.vlan_tci = tci;
6941 compose_output_action(ctx, port->up.ofp_port);
6942 ctx->flow.vlan_tci = old_tci;
6946 mirror_mask_ffs(mirror_mask_t mask)
6948 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
6953 ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
6955 return (bundle->vlan_mode != PORT_VLAN_ACCESS
6956 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
6960 ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan)
6962 return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan);
6965 /* Returns an arbitrary interface within 'bundle'. */
6966 static struct ofport_dpif *
6967 ofbundle_get_a_port(const struct ofbundle *bundle)
6969 return CONTAINER_OF(list_front(&bundle->ports),
6970 struct ofport_dpif, bundle_node);
6974 vlan_is_mirrored(const struct ofmirror *m, int vlan)
6976 return !m->vlans || bitmap_is_set(m->vlans, vlan);
6980 add_mirror_actions(struct action_xlate_ctx *ctx, const struct flow *orig_flow)
6982 struct ofproto_dpif *ofproto = ctx->ofproto;
6983 mirror_mask_t mirrors;
6984 struct ofbundle *in_bundle;
6987 const struct nlattr *a;
6990 in_bundle = lookup_input_bundle(ctx->ofproto, orig_flow->in_port,
6991 ctx->packet != NULL, NULL);
6995 mirrors = in_bundle->src_mirrors;
6997 /* Drop frames on bundles reserved for mirroring. */
6998 if (in_bundle->mirror_out) {
6999 if (ctx->packet != NULL) {
7000 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7001 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
7002 "%s, which is reserved exclusively for mirroring",
7003 ctx->ofproto->up.name, in_bundle->name);
7009 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
7010 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
7013 vlan = input_vid_to_vlan(in_bundle, vid);
7015 /* Look at the output ports to check for destination selections. */
7017 NL_ATTR_FOR_EACH (a, left, ctx->odp_actions->data,
7018 ctx->odp_actions->size) {
7019 enum ovs_action_attr type = nl_attr_type(a);
7020 struct ofport_dpif *ofport;
7022 if (type != OVS_ACTION_ATTR_OUTPUT) {
7026 ofport = get_odp_port(ofproto, nl_attr_get_u32(a));
7027 if (ofport && ofport->bundle) {
7028 mirrors |= ofport->bundle->dst_mirrors;
7036 /* Restore the original packet before adding the mirror actions. */
7037 ctx->flow = *orig_flow;
7042 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
7044 if (!vlan_is_mirrored(m, vlan)) {
7045 mirrors = zero_rightmost_1bit(mirrors);
7049 mirrors &= ~m->dup_mirrors;
7050 ctx->mirrors |= m->dup_mirrors;
7052 output_normal(ctx, m->out, vlan);
7053 } else if (vlan != m->out_vlan
7054 && !eth_addr_is_reserved(orig_flow->dl_dst)) {
7055 struct ofbundle *bundle;
7057 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
7058 if (ofbundle_includes_vlan(bundle, m->out_vlan)
7059 && !bundle->mirror_out) {
7060 output_normal(ctx, bundle, m->out_vlan);
7068 update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors,
7069 uint64_t packets, uint64_t bytes)
7075 for (; mirrors; mirrors = zero_rightmost_1bit(mirrors)) {
7078 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
7081 /* In normal circumstances 'm' will not be NULL. However,
7082 * if mirrors are reconfigured, we can temporarily get out
7083 * of sync in facet_revalidate(). We could "correct" the
7084 * mirror list before reaching here, but doing that would
7085 * not properly account the traffic stats we've currently
7086 * accumulated for previous mirror configuration. */
7090 m->packet_count += packets;
7091 m->byte_count += bytes;
7095 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
7096 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
7097 * indicate this; newer upstream kernels use gratuitous ARP requests. */
7099 is_gratuitous_arp(const struct flow *flow)
7101 return (flow->dl_type == htons(ETH_TYPE_ARP)
7102 && eth_addr_is_broadcast(flow->dl_dst)
7103 && (flow->nw_proto == ARP_OP_REPLY
7104 || (flow->nw_proto == ARP_OP_REQUEST
7105 && flow->nw_src == flow->nw_dst)));
7109 update_learning_table(struct ofproto_dpif *ofproto,
7110 const struct flow *flow, int vlan,
7111 struct ofbundle *in_bundle)
7113 struct mac_entry *mac;
7115 /* Don't learn the OFPP_NONE port. */
7116 if (in_bundle == &ofpp_none_bundle) {
7120 if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
7124 mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan);
7125 if (is_gratuitous_arp(flow)) {
7126 /* We don't want to learn from gratuitous ARP packets that are
7127 * reflected back over bond slaves so we lock the learning table. */
7128 if (!in_bundle->bond) {
7129 mac_entry_set_grat_arp_lock(mac);
7130 } else if (mac_entry_is_grat_arp_locked(mac)) {
7135 if (mac_entry_is_new(mac) || mac->port.p != in_bundle) {
7136 /* The log messages here could actually be useful in debugging,
7137 * so keep the rate limit relatively high. */
7138 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
7139 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
7140 "on port %s in VLAN %d",
7141 ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src),
7142 in_bundle->name, vlan);
7144 mac->port.p = in_bundle;
7145 tag_set_add(&ofproto->backer->revalidate_set,
7146 mac_learning_changed(ofproto->ml, mac));
7150 static struct ofbundle *
7151 lookup_input_bundle(const struct ofproto_dpif *ofproto, uint16_t in_port,
7152 bool warn, struct ofport_dpif **in_ofportp)
7154 struct ofport_dpif *ofport;
7156 /* Find the port and bundle for the received packet. */
7157 ofport = get_ofp_port(ofproto, in_port);
7159 *in_ofportp = ofport;
7161 if (ofport && ofport->bundle) {
7162 return ofport->bundle;
7165 /* Special-case OFPP_NONE, which a controller may use as the ingress
7166 * port for traffic that it is sourcing. */
7167 if (in_port == OFPP_NONE) {
7168 return &ofpp_none_bundle;
7171 /* Odd. A few possible reasons here:
7173 * - We deleted a port but there are still a few packets queued up
7176 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
7177 * we don't know about.
7179 * - The ofproto client didn't configure the port as part of a bundle.
7180 * This is particularly likely to happen if a packet was received on the
7181 * port after it was created, but before the client had a chance to
7182 * configure its bundle.
7185 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7187 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
7188 "port %"PRIu16, ofproto->up.name, in_port);
7193 /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or
7194 * dropped. Returns true if they may be forwarded, false if they should be
7197 * 'in_port' must be the ofport_dpif that corresponds to flow->in_port.
7198 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
7200 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
7201 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
7202 * checked by input_vid_is_valid().
7204 * May also add tags to '*tags', although the current implementation only does
7205 * so in one special case.
7208 is_admissible(struct action_xlate_ctx *ctx, struct ofport_dpif *in_port,
7211 struct ofproto_dpif *ofproto = ctx->ofproto;
7212 struct flow *flow = &ctx->flow;
7213 struct ofbundle *in_bundle = in_port->bundle;
7215 /* Drop frames for reserved multicast addresses
7216 * only if forward_bpdu option is absent. */
7217 if (!ofproto->up.forward_bpdu && eth_addr_is_reserved(flow->dl_dst)) {
7218 xlate_report(ctx, "packet has reserved destination MAC, dropping");
7222 if (in_bundle->bond) {
7223 struct mac_entry *mac;
7225 switch (bond_check_admissibility(in_bundle->bond, in_port,
7226 flow->dl_dst, &ctx->tags)) {
7231 xlate_report(ctx, "bonding refused admissibility, dropping");
7234 case BV_DROP_IF_MOVED:
7235 mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL);
7236 if (mac && mac->port.p != in_bundle &&
7237 (!is_gratuitous_arp(flow)
7238 || mac_entry_is_grat_arp_locked(mac))) {
7239 xlate_report(ctx, "SLB bond thinks this packet looped back, "
7251 xlate_normal(struct action_xlate_ctx *ctx)
7253 struct ofport_dpif *in_port;
7254 struct ofbundle *in_bundle;
7255 struct mac_entry *mac;
7259 ctx->has_normal = true;
7261 in_bundle = lookup_input_bundle(ctx->ofproto, ctx->flow.in_port,
7262 ctx->packet != NULL, &in_port);
7264 xlate_report(ctx, "no input bundle, dropping");
7268 /* Drop malformed frames. */
7269 if (ctx->flow.dl_type == htons(ETH_TYPE_VLAN) &&
7270 !(ctx->flow.vlan_tci & htons(VLAN_CFI))) {
7271 if (ctx->packet != NULL) {
7272 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7273 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
7274 "VLAN tag received on port %s",
7275 ctx->ofproto->up.name, in_bundle->name);
7277 xlate_report(ctx, "partial VLAN tag, dropping");
7281 /* Drop frames on bundles reserved for mirroring. */
7282 if (in_bundle->mirror_out) {
7283 if (ctx->packet != NULL) {
7284 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7285 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
7286 "%s, which is reserved exclusively for mirroring",
7287 ctx->ofproto->up.name, in_bundle->name);
7289 xlate_report(ctx, "input port is mirror output port, dropping");
7294 vid = vlan_tci_to_vid(ctx->flow.vlan_tci);
7295 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
7296 xlate_report(ctx, "disallowed VLAN VID for this input port, dropping");
7299 vlan = input_vid_to_vlan(in_bundle, vid);
7301 /* Check other admissibility requirements. */
7302 if (in_port && !is_admissible(ctx, in_port, vlan)) {
7306 /* Learn source MAC. */
7307 if (ctx->may_learn) {
7308 update_learning_table(ctx->ofproto, &ctx->flow, vlan, in_bundle);
7311 /* Determine output bundle. */
7312 mac = mac_learning_lookup(ctx->ofproto->ml, ctx->flow.dl_dst, vlan,
7315 if (mac->port.p != in_bundle) {
7316 xlate_report(ctx, "forwarding to learned port");
7317 output_normal(ctx, mac->port.p, vlan);
7319 xlate_report(ctx, "learned port is input port, dropping");
7322 struct ofbundle *bundle;
7324 xlate_report(ctx, "no learned MAC for destination, flooding");
7325 HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
7326 if (bundle != in_bundle
7327 && ofbundle_includes_vlan(bundle, vlan)
7328 && bundle->floodable
7329 && !bundle->mirror_out) {
7330 output_normal(ctx, bundle, vlan);
7333 ctx->nf_output_iface = NF_OUT_FLOOD;
7337 /* Optimized flow revalidation.
7339 * It's a difficult problem, in general, to tell which facets need to have
7340 * their actions recalculated whenever the OpenFlow flow table changes. We
7341 * don't try to solve that general problem: for most kinds of OpenFlow flow
7342 * table changes, we recalculate the actions for every facet. This is
7343 * relatively expensive, but it's good enough if the OpenFlow flow table
7344 * doesn't change very often.
7346 * However, we can expect one particular kind of OpenFlow flow table change to
7347 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
7348 * of CPU on revalidating every facet whenever MAC learning modifies the flow
7349 * table, we add a special case that applies to flow tables in which every rule
7350 * has the same form (that is, the same wildcards), except that the table is
7351 * also allowed to have a single "catch-all" flow that matches all packets. We
7352 * optimize this case by tagging all of the facets that resubmit into the table
7353 * and invalidating the same tag whenever a flow changes in that table. The
7354 * end result is that we revalidate just the facets that need it (and sometimes
7355 * a few more, but not all of the facets or even all of the facets that
7356 * resubmit to the table modified by MAC learning). */
7358 /* Calculates the tag to use for 'flow' and mask 'mask' when it is inserted
7359 * into an OpenFlow table with the given 'basis'. */
7361 rule_calculate_tag(const struct flow *flow, const struct minimask *mask,
7364 if (minimask_is_catchall(mask)) {
7367 uint32_t hash = flow_hash_in_minimask(flow, mask, secret);
7368 return tag_create_deterministic(hash);
7372 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
7373 * taggability of that table.
7375 * This function must be called after *each* change to a flow table. If you
7376 * skip calling it on some changes then the pointer comparisons at the end can
7377 * be invalid if you get unlucky. For example, if a flow removal causes a
7378 * cls_table to be destroyed and then a flow insertion causes a cls_table with
7379 * different wildcards to be created with the same address, then this function
7380 * will incorrectly skip revalidation. */
7382 table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
7384 struct table_dpif *table = &ofproto->tables[table_id];
7385 const struct oftable *oftable = &ofproto->up.tables[table_id];
7386 struct cls_table *catchall, *other;
7387 struct cls_table *t;
7389 catchall = other = NULL;
7391 switch (hmap_count(&oftable->cls.tables)) {
7393 /* We could tag this OpenFlow table but it would make the logic a
7394 * little harder and it's a corner case that doesn't seem worth it
7400 HMAP_FOR_EACH (t, hmap_node, &oftable->cls.tables) {
7401 if (cls_table_is_catchall(t)) {
7403 } else if (!other) {
7406 /* Indicate that we can't tag this by setting both tables to
7407 * NULL. (We know that 'catchall' is already NULL.) */
7414 /* Can't tag this table. */
7418 if (table->catchall_table != catchall || table->other_table != other) {
7419 table->catchall_table = catchall;
7420 table->other_table = other;
7421 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
7425 /* Given 'rule' that has changed in some way (either it is a rule being
7426 * inserted, a rule being deleted, or a rule whose actions are being
7427 * modified), marks facets for revalidation to ensure that packets will be
7428 * forwarded correctly according to the new state of the flow table.
7430 * This function must be called after *each* change to a flow table. See
7431 * the comment on table_update_taggable() for more information. */
7433 rule_invalidate(const struct rule_dpif *rule)
7435 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
7437 table_update_taggable(ofproto, rule->up.table_id);
7439 if (!ofproto->backer->need_revalidate) {
7440 struct table_dpif *table = &ofproto->tables[rule->up.table_id];
7442 if (table->other_table && rule->tag) {
7443 tag_set_add(&ofproto->backer->revalidate_set, rule->tag);
7445 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
7451 set_frag_handling(struct ofproto *ofproto_,
7452 enum ofp_config_flags frag_handling)
7454 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7455 if (frag_handling != OFPC_FRAG_REASM) {
7456 ofproto->backer->need_revalidate = REV_RECONFIGURE;
7464 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
7465 const struct flow *flow,
7466 const struct ofpact *ofpacts, size_t ofpacts_len)
7468 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7469 struct initial_vals initial_vals;
7470 struct odputil_keybuf keybuf;
7471 struct dpif_flow_stats stats;
7475 struct action_xlate_ctx ctx;
7476 uint64_t odp_actions_stub[1024 / 8];
7477 struct ofpbuf odp_actions;
7479 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
7480 odp_flow_key_from_flow(&key, flow,
7481 ofp_port_to_odp_port(ofproto, flow->in_port));
7483 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
7485 initial_vals.vlan_tci = flow->vlan_tci;
7486 initial_vals.tunnel_ip_tos = 0;
7487 action_xlate_ctx_init(&ctx, ofproto, flow, &initial_vals, NULL,
7488 packet_get_tcp_flags(packet, flow), packet);
7489 ctx.resubmit_stats = &stats;
7491 ofpbuf_use_stub(&odp_actions,
7492 odp_actions_stub, sizeof odp_actions_stub);
7493 xlate_actions(&ctx, ofpacts, ofpacts_len, &odp_actions);
7494 dpif_execute(ofproto->backer->dpif, key.data, key.size,
7495 odp_actions.data, odp_actions.size, packet);
7496 ofpbuf_uninit(&odp_actions);
7504 set_netflow(struct ofproto *ofproto_,
7505 const struct netflow_options *netflow_options)
7507 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7509 if (netflow_options) {
7510 if (!ofproto->netflow) {
7511 ofproto->netflow = netflow_create();
7513 return netflow_set_options(ofproto->netflow, netflow_options);
7515 netflow_destroy(ofproto->netflow);
7516 ofproto->netflow = NULL;
7522 get_netflow_ids(const struct ofproto *ofproto_,
7523 uint8_t *engine_type, uint8_t *engine_id)
7525 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7527 dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
7531 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
7533 if (!facet_is_controller_flow(facet) &&
7534 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
7535 struct subfacet *subfacet;
7536 struct ofexpired expired;
7538 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
7539 if (subfacet->path == SF_FAST_PATH) {
7540 struct dpif_flow_stats stats;
7542 subfacet_reinstall(subfacet, &stats);
7543 subfacet_update_stats(subfacet, &stats);
7547 expired.flow = facet->flow;
7548 expired.packet_count = facet->packet_count;
7549 expired.byte_count = facet->byte_count;
7550 expired.used = facet->used;
7551 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
7556 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
7558 struct facet *facet;
7560 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
7561 send_active_timeout(ofproto, facet);
7565 static struct ofproto_dpif *
7566 ofproto_dpif_lookup(const char *name)
7568 struct ofproto_dpif *ofproto;
7570 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
7571 hash_string(name, 0), &all_ofproto_dpifs) {
7572 if (!strcmp(ofproto->up.name, name)) {
7580 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
7581 const char *argv[], void *aux OVS_UNUSED)
7583 struct ofproto_dpif *ofproto;
7586 ofproto = ofproto_dpif_lookup(argv[1]);
7588 unixctl_command_reply_error(conn, "no such bridge");
7591 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
7593 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
7594 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
7598 unixctl_command_reply(conn, "table successfully flushed");
7602 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
7603 const char *argv[], void *aux OVS_UNUSED)
7605 struct ds ds = DS_EMPTY_INITIALIZER;
7606 const struct ofproto_dpif *ofproto;
7607 const struct mac_entry *e;
7609 ofproto = ofproto_dpif_lookup(argv[1]);
7611 unixctl_command_reply_error(conn, "no such bridge");
7615 ds_put_cstr(&ds, " port VLAN MAC Age\n");
7616 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
7617 struct ofbundle *bundle = e->port.p;
7618 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
7619 ofbundle_get_a_port(bundle)->odp_port,
7620 e->vlan, ETH_ADDR_ARGS(e->mac),
7621 mac_entry_age(ofproto->ml, e));
7623 unixctl_command_reply(conn, ds_cstr(&ds));
7628 struct action_xlate_ctx ctx;
7634 trace_format_rule(struct ds *result, uint8_t table_id, int level,
7635 const struct rule_dpif *rule)
7637 ds_put_char_multiple(result, '\t', level);
7639 ds_put_cstr(result, "No match\n");
7643 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
7644 table_id, ntohll(rule->up.flow_cookie));
7645 cls_rule_format(&rule->up.cr, result);
7646 ds_put_char(result, '\n');
7648 ds_put_char_multiple(result, '\t', level);
7649 ds_put_cstr(result, "OpenFlow ");
7650 ofpacts_format(rule->up.ofpacts, rule->up.ofpacts_len, result);
7651 ds_put_char(result, '\n');
7655 trace_format_flow(struct ds *result, int level, const char *title,
7656 struct trace_ctx *trace)
7658 ds_put_char_multiple(result, '\t', level);
7659 ds_put_format(result, "%s: ", title);
7660 if (flow_equal(&trace->ctx.flow, &trace->flow)) {
7661 ds_put_cstr(result, "unchanged");
7663 flow_format(result, &trace->ctx.flow);
7664 trace->flow = trace->ctx.flow;
7666 ds_put_char(result, '\n');
7670 trace_format_regs(struct ds *result, int level, const char *title,
7671 struct trace_ctx *trace)
7675 ds_put_char_multiple(result, '\t', level);
7676 ds_put_format(result, "%s:", title);
7677 for (i = 0; i < FLOW_N_REGS; i++) {
7678 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
7680 ds_put_char(result, '\n');
7684 trace_format_odp(struct ds *result, int level, const char *title,
7685 struct trace_ctx *trace)
7687 struct ofpbuf *odp_actions = trace->ctx.odp_actions;
7689 ds_put_char_multiple(result, '\t', level);
7690 ds_put_format(result, "%s: ", title);
7691 format_odp_actions(result, odp_actions->data, odp_actions->size);
7692 ds_put_char(result, '\n');
7696 trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
7698 struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx);
7699 struct ds *result = trace->result;
7701 ds_put_char(result, '\n');
7702 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
7703 trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace);
7704 trace_format_odp(result, ctx->recurse + 1, "Resubmitted odp", trace);
7705 trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule);
7709 trace_report(struct action_xlate_ctx *ctx, const char *s)
7711 struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx);
7712 struct ds *result = trace->result;
7714 ds_put_char_multiple(result, '\t', ctx->recurse);
7715 ds_put_cstr(result, s);
7716 ds_put_char(result, '\n');
7720 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
7721 void *aux OVS_UNUSED)
7723 const char *dpname = argv[1];
7724 struct ofproto_dpif *ofproto;
7725 struct ofpbuf odp_key;
7726 struct ofpbuf *packet;
7727 struct initial_vals initial_vals;
7733 ofpbuf_init(&odp_key, 0);
7736 ofproto = ofproto_dpif_lookup(dpname);
7738 unixctl_command_reply_error(conn, "Unknown ofproto (use ofproto/list "
7742 if (argc == 3 || (argc == 4 && !strcmp(argv[3], "-generate"))) {
7743 /* ofproto/trace dpname flow [-generate] */
7744 const char *flow_s = argv[2];
7745 const char *generate_s = argv[3];
7747 /* Allow 'flow_s' to be either a datapath flow or an OpenFlow-like
7748 * flow. We guess which type it is based on whether 'flow_s' contains
7749 * an '(', since a datapath flow always contains '(') but an
7750 * OpenFlow-like flow should not (in fact it's allowed but I believe
7751 * that's not documented anywhere).
7753 * An alternative would be to try to parse 'flow_s' both ways, but then
7754 * it would be tricky giving a sensible error message. After all, do
7755 * you just say "syntax error" or do you present both error messages?
7756 * Both choices seem lousy. */
7757 if (strchr(flow_s, '(')) {
7760 /* Convert string to datapath key. */
7761 ofpbuf_init(&odp_key, 0);
7762 error = odp_flow_key_from_string(flow_s, NULL, &odp_key);
7764 unixctl_command_reply_error(conn, "Bad flow syntax");
7768 /* The user might have specified the wrong ofproto but within the
7769 * same backer. That's OK, ofproto_receive() can find the right
7771 if (ofproto_receive(ofproto->backer, NULL, odp_key.data,
7772 odp_key.size, &flow, NULL, &ofproto, NULL,
7774 unixctl_command_reply_error(conn, "Invalid flow");
7777 ds_put_format(&result, "Bridge: %s\n", ofproto->up.name);
7781 error_s = parse_ofp_exact_flow(&flow, argv[2]);
7783 unixctl_command_reply_error(conn, error_s);
7788 initial_vals.vlan_tci = flow.vlan_tci;
7789 initial_vals.tunnel_ip_tos = flow.tunnel.ip_tos;
7792 /* Generate a packet, if requested. */
7794 packet = ofpbuf_new(0);
7795 flow_compose(packet, &flow);
7797 } else if (argc == 7) {
7798 /* ofproto/trace dpname priority tun_id in_port mark packet */
7799 const char *priority_s = argv[2];
7800 const char *tun_id_s = argv[3];
7801 const char *in_port_s = argv[4];
7802 const char *mark_s = argv[5];
7803 const char *packet_s = argv[6];
7804 uint32_t in_port = atoi(in_port_s);
7805 ovs_be64 tun_id = htonll(strtoull(tun_id_s, NULL, 0));
7806 uint32_t priority = atoi(priority_s);
7807 uint32_t mark = atoi(mark_s);
7810 msg = eth_from_hex(packet_s, &packet);
7812 unixctl_command_reply_error(conn, msg);
7816 ds_put_cstr(&result, "Packet: ");
7817 s = ofp_packet_to_string(packet->data, packet->size);
7818 ds_put_cstr(&result, s);
7821 flow_extract(packet, priority, mark, NULL, in_port, &flow);
7822 flow.tunnel.tun_id = tun_id;
7823 initial_vals.vlan_tci = flow.vlan_tci;
7824 initial_vals.tunnel_ip_tos = flow.tunnel.ip_tos;
7826 unixctl_command_reply_error(conn, "Bad command syntax");
7830 ofproto_trace(ofproto, &flow, packet, &initial_vals, &result);
7831 unixctl_command_reply(conn, ds_cstr(&result));
7834 ds_destroy(&result);
7835 ofpbuf_delete(packet);
7836 ofpbuf_uninit(&odp_key);
7840 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
7841 const struct ofpbuf *packet,
7842 const struct initial_vals *initial_vals, struct ds *ds)
7844 struct rule_dpif *rule;
7846 ds_put_cstr(ds, "Flow: ");
7847 flow_format(ds, flow);
7848 ds_put_char(ds, '\n');
7850 rule = rule_dpif_lookup(ofproto, flow);
7852 trace_format_rule(ds, 0, 0, rule);
7853 if (rule == ofproto->miss_rule) {
7854 ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
7855 } else if (rule == ofproto->no_packet_in_rule) {
7856 ds_put_cstr(ds, "\nNo match, packets dropped because "
7857 "OFPPC_NO_PACKET_IN is set on in_port.\n");
7861 uint64_t odp_actions_stub[1024 / 8];
7862 struct ofpbuf odp_actions;
7864 struct trace_ctx trace;
7867 tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
7870 ofpbuf_use_stub(&odp_actions,
7871 odp_actions_stub, sizeof odp_actions_stub);
7872 action_xlate_ctx_init(&trace.ctx, ofproto, flow, initial_vals,
7873 rule, tcp_flags, packet);
7874 trace.ctx.resubmit_hook = trace_resubmit;
7875 trace.ctx.report_hook = trace_report;
7876 xlate_actions(&trace.ctx, rule->up.ofpacts, rule->up.ofpacts_len,
7879 ds_put_char(ds, '\n');
7880 trace_format_flow(ds, 0, "Final flow", &trace);
7881 ds_put_cstr(ds, "Datapath actions: ");
7882 format_odp_actions(ds, odp_actions.data, odp_actions.size);
7883 ofpbuf_uninit(&odp_actions);
7885 if (trace.ctx.slow) {
7886 enum slow_path_reason slow;
7888 ds_put_cstr(ds, "\nThis flow is handled by the userspace "
7889 "slow path because it:");
7890 for (slow = trace.ctx.slow; slow; ) {
7891 enum slow_path_reason bit = rightmost_1bit(slow);
7895 ds_put_cstr(ds, "\n\t- Consists of CFM packets.");
7898 ds_put_cstr(ds, "\n\t- Consists of LACP packets.");
7901 ds_put_cstr(ds, "\n\t- Consists of STP packets.");
7904 ds_put_cstr(ds, "\n\t- Needs in-band special case "
7907 ds_put_cstr(ds, "\n\t (The datapath actions are "
7908 "incomplete--for complete actions, "
7909 "please supply a packet.)");
7912 case SLOW_CONTROLLER:
7913 ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages "
7914 "to the OpenFlow controller.");
7917 ds_put_cstr(ds, "\n\t- Needs more specific matching "
7918 "than the datapath supports.");
7925 if (slow & ~SLOW_MATCH) {
7926 ds_put_cstr(ds, "\nThe datapath actions above do not reflect "
7927 "the special slow-path processing.");
7934 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
7935 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
7938 unixctl_command_reply(conn, NULL);
7942 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
7943 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
7946 unixctl_command_reply(conn, NULL);
7949 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
7950 * 'reply' describing the results. */
7952 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
7954 struct facet *facet;
7958 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
7959 if (!facet_check_consistency(facet)) {
7964 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
7968 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
7969 ofproto->up.name, errors);
7971 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
7976 ofproto_dpif_self_check(struct unixctl_conn *conn,
7977 int argc, const char *argv[], void *aux OVS_UNUSED)
7979 struct ds reply = DS_EMPTY_INITIALIZER;
7980 struct ofproto_dpif *ofproto;
7983 ofproto = ofproto_dpif_lookup(argv[1]);
7985 unixctl_command_reply_error(conn, "Unknown ofproto (use "
7986 "ofproto/list for help)");
7989 ofproto_dpif_self_check__(ofproto, &reply);
7991 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
7992 ofproto_dpif_self_check__(ofproto, &reply);
7996 unixctl_command_reply(conn, ds_cstr(&reply));
8000 /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list
8001 * of the 'ofproto_shash' nodes. It is the responsibility of the caller
8002 * to destroy 'ofproto_shash' and free the returned value. */
8003 static const struct shash_node **
8004 get_ofprotos(struct shash *ofproto_shash)
8006 const struct ofproto_dpif *ofproto;
8008 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
8009 char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
8010 shash_add_nocopy(ofproto_shash, name, ofproto);
8013 return shash_sort(ofproto_shash);
8017 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
8018 const char *argv[] OVS_UNUSED,
8019 void *aux OVS_UNUSED)
8021 struct ds ds = DS_EMPTY_INITIALIZER;
8022 struct shash ofproto_shash;
8023 const struct shash_node **sorted_ofprotos;
8026 shash_init(&ofproto_shash);
8027 sorted_ofprotos = get_ofprotos(&ofproto_shash);
8028 for (i = 0; i < shash_count(&ofproto_shash); i++) {
8029 const struct shash_node *node = sorted_ofprotos[i];
8030 ds_put_format(&ds, "%s\n", node->name);
8033 shash_destroy(&ofproto_shash);
8034 free(sorted_ofprotos);
8036 unixctl_command_reply(conn, ds_cstr(&ds));
8041 show_dp_format(const struct ofproto_dpif *ofproto, struct ds *ds)
8043 struct dpif_dp_stats s;
8044 const struct shash_node **ports;
8047 dpif_get_dp_stats(ofproto->backer->dpif, &s);
8049 ds_put_format(ds, "%s (%s):\n", ofproto->up.name,
8050 dpif_name(ofproto->backer->dpif));
8051 /* xxx It would be better to show bridge-specific stats instead
8052 * xxx of dp ones. */
8054 "\tlookups: hit:%"PRIu64" missed:%"PRIu64" lost:%"PRIu64"\n",
8055 s.n_hit, s.n_missed, s.n_lost);
8056 ds_put_format(ds, "\tflows: %zu\n",
8057 hmap_count(&ofproto->subfacets));
8059 ports = shash_sort(&ofproto->up.port_by_name);
8060 for (i = 0; i < shash_count(&ofproto->up.port_by_name); i++) {
8061 const struct shash_node *node = ports[i];
8062 struct ofport *ofport = node->data;
8063 const char *name = netdev_get_name(ofport->netdev);
8064 const char *type = netdev_get_type(ofport->netdev);
8067 ds_put_format(ds, "\t%s %u/", name, ofport->ofp_port);
8069 odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
8070 if (odp_port != OVSP_NONE) {
8071 ds_put_format(ds, "%"PRIu32":", odp_port);
8073 ds_put_cstr(ds, "none:");
8076 if (strcmp(type, "system")) {
8077 struct netdev *netdev;
8080 ds_put_format(ds, " (%s", type);
8082 error = netdev_open(name, type, &netdev);
8087 error = netdev_get_config(netdev, &config);
8089 const struct smap_node **nodes;
8092 nodes = smap_sort(&config);
8093 for (i = 0; i < smap_count(&config); i++) {
8094 const struct smap_node *node = nodes[i];
8095 ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
8096 node->key, node->value);
8100 smap_destroy(&config);
8102 netdev_close(netdev);
8104 ds_put_char(ds, ')');
8106 ds_put_char(ds, '\n');
8112 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc,
8113 const char *argv[], void *aux OVS_UNUSED)
8115 struct ds ds = DS_EMPTY_INITIALIZER;
8116 const struct ofproto_dpif *ofproto;
8120 for (i = 1; i < argc; i++) {
8121 ofproto = ofproto_dpif_lookup(argv[i]);
8123 ds_put_format(&ds, "Unknown bridge %s (use dpif/dump-dps "
8124 "for help)", argv[i]);
8125 unixctl_command_reply_error(conn, ds_cstr(&ds));
8128 show_dp_format(ofproto, &ds);
8131 struct shash ofproto_shash;
8132 const struct shash_node **sorted_ofprotos;
8135 shash_init(&ofproto_shash);
8136 sorted_ofprotos = get_ofprotos(&ofproto_shash);
8137 for (i = 0; i < shash_count(&ofproto_shash); i++) {
8138 const struct shash_node *node = sorted_ofprotos[i];
8139 show_dp_format(node->data, &ds);
8142 shash_destroy(&ofproto_shash);
8143 free(sorted_ofprotos);
8146 unixctl_command_reply(conn, ds_cstr(&ds));
8151 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
8152 int argc OVS_UNUSED, const char *argv[],
8153 void *aux OVS_UNUSED)
8155 struct ds ds = DS_EMPTY_INITIALIZER;
8156 const struct ofproto_dpif *ofproto;
8157 struct subfacet *subfacet;
8159 ofproto = ofproto_dpif_lookup(argv[1]);
8161 unixctl_command_reply_error(conn, "no such bridge");
8165 update_stats(ofproto->backer);
8167 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
8168 odp_flow_key_format(subfacet->key, subfacet->key_len, &ds);
8170 ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:",
8171 subfacet->dp_packet_count, subfacet->dp_byte_count);
8172 if (subfacet->used) {
8173 ds_put_format(&ds, "%.3fs",
8174 (time_msec() - subfacet->used) / 1000.0);
8176 ds_put_format(&ds, "never");
8178 if (subfacet->facet->tcp_flags) {
8179 ds_put_cstr(&ds, ", flags:");
8180 packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags);
8183 ds_put_cstr(&ds, ", actions:");
8184 if (subfacet->slow) {
8185 uint64_t slow_path_stub[128 / 8];
8186 const struct nlattr *actions;
8189 compose_slow_path(ofproto, &subfacet->facet->flow, subfacet->slow,
8190 slow_path_stub, sizeof slow_path_stub,
8191 &actions, &actions_len);
8192 format_odp_actions(&ds, actions, actions_len);
8194 format_odp_actions(&ds, subfacet->actions, subfacet->actions_len);
8196 ds_put_char(&ds, '\n');
8199 unixctl_command_reply(conn, ds_cstr(&ds));
8204 ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn,
8205 int argc OVS_UNUSED, const char *argv[],
8206 void *aux OVS_UNUSED)
8208 struct ds ds = DS_EMPTY_INITIALIZER;
8209 struct ofproto_dpif *ofproto;
8211 ofproto = ofproto_dpif_lookup(argv[1]);
8213 unixctl_command_reply_error(conn, "no such bridge");
8217 flush(&ofproto->up);
8219 unixctl_command_reply(conn, ds_cstr(&ds));
8224 ofproto_dpif_unixctl_init(void)
8226 static bool registered;
8232 unixctl_command_register(
8234 "bridge {priority tun_id in_port mark packet | odp_flow [-generate]}",
8235 2, 6, ofproto_unixctl_trace, NULL);
8236 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
8237 ofproto_unixctl_fdb_flush, NULL);
8238 unixctl_command_register("fdb/show", "bridge", 1, 1,
8239 ofproto_unixctl_fdb_show, NULL);
8240 unixctl_command_register("ofproto/clog", "", 0, 0,
8241 ofproto_dpif_clog, NULL);
8242 unixctl_command_register("ofproto/unclog", "", 0, 0,
8243 ofproto_dpif_unclog, NULL);
8244 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
8245 ofproto_dpif_self_check, NULL);
8246 unixctl_command_register("dpif/dump-dps", "", 0, 0,
8247 ofproto_unixctl_dpif_dump_dps, NULL);
8248 unixctl_command_register("dpif/show", "[bridge]", 0, INT_MAX,
8249 ofproto_unixctl_dpif_show, NULL);
8250 unixctl_command_register("dpif/dump-flows", "bridge", 1, 1,
8251 ofproto_unixctl_dpif_dump_flows, NULL);
8252 unixctl_command_register("dpif/del-flows", "bridge", 1, 1,
8253 ofproto_unixctl_dpif_del_flows, NULL);
8256 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
8258 * This is deprecated. It is only for compatibility with broken device drivers
8259 * in old versions of Linux that do not properly support VLANs when VLAN
8260 * devices are not used. When broken device drivers are no longer in
8261 * widespread use, we will delete these interfaces. */
8264 set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid)
8266 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
8267 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
8269 if (realdev_ofp_port == ofport->realdev_ofp_port
8270 && vid == ofport->vlandev_vid) {
8274 ofproto->backer->need_revalidate = REV_RECONFIGURE;
8276 if (ofport->realdev_ofp_port) {
8279 if (realdev_ofp_port && ofport->bundle) {
8280 /* vlandevs are enslaved to their realdevs, so they are not allowed to
8281 * themselves be part of a bundle. */
8282 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
8285 ofport->realdev_ofp_port = realdev_ofp_port;
8286 ofport->vlandev_vid = vid;
8288 if (realdev_ofp_port) {
8289 vsp_add(ofport, realdev_ofp_port, vid);
8296 hash_realdev_vid(uint16_t realdev_ofp_port, int vid)
8298 return hash_2words(realdev_ofp_port, vid);
8301 /* Returns the ODP port number of the Linux VLAN device that corresponds to
8302 * 'vlan_tci' on the network device with port number 'realdev_odp_port' in
8303 * 'ofproto'. For example, given 'realdev_odp_port' of eth0 and 'vlan_tci' 9,
8304 * it would return the port number of eth0.9.
8306 * Unless VLAN splinters are enabled for port 'realdev_odp_port', this
8307 * function just returns its 'realdev_odp_port' argument. */
8309 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
8310 uint32_t realdev_odp_port, ovs_be16 vlan_tci)
8312 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
8313 uint16_t realdev_ofp_port;
8314 int vid = vlan_tci_to_vid(vlan_tci);
8315 const struct vlan_splinter *vsp;
8317 realdev_ofp_port = odp_port_to_ofp_port(ofproto, realdev_odp_port);
8318 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
8319 hash_realdev_vid(realdev_ofp_port, vid),
8320 &ofproto->realdev_vid_map) {
8321 if (vsp->realdev_ofp_port == realdev_ofp_port
8322 && vsp->vid == vid) {
8323 return ofp_port_to_odp_port(ofproto, vsp->vlandev_ofp_port);
8327 return realdev_odp_port;
8330 static struct vlan_splinter *
8331 vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port)
8333 struct vlan_splinter *vsp;
8335 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, hash_int(vlandev_ofp_port, 0),
8336 &ofproto->vlandev_map) {
8337 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
8345 /* Returns the OpenFlow port number of the "real" device underlying the Linux
8346 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
8347 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
8348 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
8349 * eth0 and store 9 in '*vid'.
8351 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
8352 * VLAN device. Unless VLAN splinters are enabled, this is what this function
8355 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
8356 uint16_t vlandev_ofp_port, int *vid)
8358 if (!hmap_is_empty(&ofproto->vlandev_map)) {
8359 const struct vlan_splinter *vsp;
8361 vsp = vlandev_find(ofproto, vlandev_ofp_port);
8366 return vsp->realdev_ofp_port;
8372 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
8373 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
8374 * 'flow->in_port' to the "real" device backing the VLAN device, sets
8375 * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is
8376 * always the case unless VLAN splinters are enabled), returns false without
8377 * making any changes. */
8379 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
8384 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port, &vid);
8389 /* Cause the flow to be processed as if it came in on the real device with
8390 * the VLAN device's VLAN ID. */
8391 flow->in_port = realdev;
8392 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
8397 vsp_remove(struct ofport_dpif *port)
8399 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
8400 struct vlan_splinter *vsp;
8402 vsp = vlandev_find(ofproto, port->up.ofp_port);
8404 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
8405 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
8408 port->realdev_ofp_port = 0;
8410 VLOG_ERR("missing vlan device record");
8415 vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid)
8417 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
8419 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
8420 && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid))
8421 == realdev_ofp_port)) {
8422 struct vlan_splinter *vsp;
8424 vsp = xmalloc(sizeof *vsp);
8425 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
8426 hash_int(port->up.ofp_port, 0));
8427 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
8428 hash_realdev_vid(realdev_ofp_port, vid));
8429 vsp->realdev_ofp_port = realdev_ofp_port;
8430 vsp->vlandev_ofp_port = port->up.ofp_port;
8433 port->realdev_ofp_port = realdev_ofp_port;
8435 VLOG_ERR("duplicate vlan device record");
8440 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
8442 const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
8443 return ofport ? ofport->odp_port : OVSP_NONE;
8446 static struct ofport_dpif *
8447 odp_port_to_ofport(const struct dpif_backer *backer, uint32_t odp_port)
8449 struct ofport_dpif *port;
8451 HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node,
8452 hash_int(odp_port, 0),
8453 &backer->odp_to_ofport_map) {
8454 if (port->odp_port == odp_port) {
8463 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
8465 struct ofport_dpif *port;
8467 port = odp_port_to_ofport(ofproto->backer, odp_port);
8468 if (port && &ofproto->up == port->up.ofproto) {
8469 return port->up.ofp_port;
8475 const struct ofproto_class ofproto_dpif_class = {
8510 port_is_lacp_current,
8511 NULL, /* rule_choose_table */
8518 rule_modify_actions,
8529 get_stp_port_status,
8536 is_mirror_output_bundle,
8537 forward_bpdu_changed,
8538 set_mac_table_config,