2 * Copyright (c) 2009, 2010, 2011, 2012, 2013 Nicira, Inc.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at:
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
19 #include "ofproto/ofproto-provider.h"
26 #include "byte-order.h"
31 #include "dynamic-string.h"
32 #include "fail-open.h"
36 #include "mac-learning.h"
37 #include "meta-flow.h"
38 #include "multipath.h"
39 #include "netdev-vport.h"
46 #include "ofp-actions.h"
47 #include "ofp-parse.h"
48 #include "ofp-print.h"
49 #include "ofproto-dpif-governor.h"
50 #include "ofproto-dpif-sflow.h"
51 #include "poll-loop.h"
56 #include "unaligned.h"
58 #include "vlan-bitmap.h"
61 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
63 COVERAGE_DEFINE(ofproto_dpif_expired);
64 COVERAGE_DEFINE(ofproto_dpif_xlate);
65 COVERAGE_DEFINE(facet_changed_rule);
66 COVERAGE_DEFINE(facet_revalidate);
67 COVERAGE_DEFINE(facet_unexpected);
68 COVERAGE_DEFINE(facet_suppress);
70 /* Maximum depth of flow table recursion (due to resubmit actions) in a
71 * flow translation. */
72 #define MAX_RESUBMIT_RECURSION 64
74 /* Number of implemented OpenFlow tables. */
75 enum { N_TABLES = 255 };
76 enum { TBL_INTERNAL = N_TABLES - 1 }; /* Used for internal hidden rules. */
77 BUILD_ASSERT_DECL(N_TABLES >= 2 && N_TABLES <= 255);
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 rule_dpif *, const struct flow *,
123 const struct dpif_flow_stats *);
124 static tag_type rule_calculate_tag(const struct flow *,
125 const struct minimask *, uint32_t basis);
126 static void rule_invalidate(const struct rule_dpif *);
128 #define MAX_MIRRORS 32
129 typedef uint32_t mirror_mask_t;
130 #define MIRROR_MASK_C(X) UINT32_C(X)
131 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
133 struct ofproto_dpif *ofproto; /* Owning ofproto. */
134 size_t idx; /* In ofproto's "mirrors" array. */
135 void *aux; /* Key supplied by ofproto's client. */
136 char *name; /* Identifier for log messages. */
138 /* Selection criteria. */
139 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
140 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
141 unsigned long *vlans; /* Bitmap of chosen VLANs, NULL selects all. */
143 /* Output (exactly one of out == NULL and out_vlan == -1 is true). */
144 struct ofbundle *out; /* Output port or NULL. */
145 int out_vlan; /* Output VLAN or -1. */
146 mirror_mask_t dup_mirrors; /* Bitmap of mirrors with the same output. */
149 int64_t packet_count; /* Number of packets sent. */
150 int64_t byte_count; /* Number of bytes sent. */
153 static void mirror_destroy(struct ofmirror *);
154 static void update_mirror_stats(struct ofproto_dpif *ofproto,
155 mirror_mask_t mirrors,
156 uint64_t packets, uint64_t bytes);
159 struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
160 struct ofproto_dpif *ofproto; /* Owning ofproto. */
161 void *aux; /* Key supplied by ofproto's client. */
162 char *name; /* Identifier for log messages. */
165 struct list ports; /* Contains "struct ofport"s. */
166 enum port_vlan_mode vlan_mode; /* VLAN mode */
167 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
168 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
169 * NULL if all VLANs are trunked. */
170 struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */
171 struct bond *bond; /* Nonnull iff more than one port. */
172 bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */
175 bool floodable; /* True if no port has OFPUTIL_PC_NO_FLOOD set. */
177 /* Port mirroring info. */
178 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
179 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
180 mirror_mask_t mirror_out; /* Mirrors that output to this bundle. */
183 static void bundle_remove(struct ofport *);
184 static void bundle_update(struct ofbundle *);
185 static void bundle_destroy(struct ofbundle *);
186 static void bundle_del_port(struct ofport_dpif *);
187 static void bundle_run(struct ofbundle *);
188 static void bundle_wait(struct ofbundle *);
189 static struct ofbundle *lookup_input_bundle(const struct ofproto_dpif *,
190 uint16_t in_port, bool warn,
191 struct ofport_dpif **in_ofportp);
193 /* A controller may use OFPP_NONE as the ingress port to indicate that
194 * it did not arrive on a "real" port. 'ofpp_none_bundle' exists for
195 * when an input bundle is needed for validation (e.g., mirroring or
196 * OFPP_NORMAL processing). It is not connected to an 'ofproto' or have
197 * any 'port' structs, so care must be taken when dealing with it. */
198 static struct ofbundle ofpp_none_bundle = {
200 .vlan_mode = PORT_VLAN_TRUNK
203 static void stp_run(struct ofproto_dpif *ofproto);
204 static void stp_wait(struct ofproto_dpif *ofproto);
205 static int set_stp_port(struct ofport *,
206 const struct ofproto_port_stp_settings *);
208 static bool ofbundle_includes_vlan(const struct ofbundle *, uint16_t vlan);
210 struct action_xlate_ctx {
211 /* action_xlate_ctx_init() initializes these members. */
214 struct ofproto_dpif *ofproto;
216 /* Flow to which the OpenFlow actions apply. xlate_actions() will modify
217 * this flow when actions change header fields. */
220 /* The packet corresponding to 'flow', or a null pointer if we are
221 * revalidating without a packet to refer to. */
222 const struct ofpbuf *packet;
224 /* Should OFPP_NORMAL update the MAC learning table? Should "learn"
225 * actions update the flow table?
227 * We want to update these tables if we are actually processing a packet,
228 * or if we are accounting for packets that the datapath has processed, but
229 * not if we are just revalidating. */
232 /* The rule that we are currently translating, or NULL. */
233 struct rule_dpif *rule;
235 /* Union of the set of TCP flags seen so far in this flow. (Used only by
236 * NXAST_FIN_TIMEOUT. Set to zero to avoid updating updating rules'
240 /* If nonnull, flow translation calls this function just before executing a
241 * resubmit or OFPP_TABLE action. In addition, disables logging of traces
242 * when the recursion depth is exceeded.
244 * 'rule' is the rule being submitted into. It will be null if the
245 * resubmit or OFPP_TABLE action didn't find a matching rule.
247 * This is normally null so the client has to set it manually after
248 * calling action_xlate_ctx_init(). */
249 void (*resubmit_hook)(struct action_xlate_ctx *, struct rule_dpif *rule);
251 /* If nonnull, flow translation calls this function to report some
252 * significant decision, e.g. to explain why OFPP_NORMAL translation
253 * dropped a packet. */
254 void (*report_hook)(struct action_xlate_ctx *, const char *s);
256 /* If nonnull, flow translation credits the specified statistics to each
257 * rule reached through a resubmit or OFPP_TABLE action.
259 * This is normally null so the client has to set it manually after
260 * calling action_xlate_ctx_init(). */
261 const struct dpif_flow_stats *resubmit_stats;
263 /* xlate_actions() initializes and uses these members. The client might want
264 * to look at them after it returns. */
266 struct ofpbuf *odp_actions; /* Datapath actions. */
267 tag_type tags; /* Tags associated with actions. */
268 enum slow_path_reason slow; /* 0 if fast path may be used. */
269 bool has_learn; /* Actions include NXAST_LEARN? */
270 bool has_normal; /* Actions output to OFPP_NORMAL? */
271 bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */
272 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
273 mirror_mask_t mirrors; /* Bitmap of associated mirrors. */
275 /* xlate_actions() initializes and uses these members, but the client has no
276 * reason to look at them. */
278 int recurse; /* Recursion level, via xlate_table_action. */
279 bool max_resubmit_trigger; /* Recursed too deeply during translation. */
280 struct flow base_flow; /* Flow at the last commit. */
281 uint32_t orig_skb_priority; /* Priority when packet arrived. */
282 uint8_t table_id; /* OpenFlow table ID where flow was found. */
283 uint32_t sflow_n_outputs; /* Number of output ports. */
284 uint32_t sflow_odp_port; /* Output port for composing sFlow action. */
285 uint16_t user_cookie_offset;/* Used for user_action_cookie fixup. */
286 bool exit; /* No further actions should be processed. */
287 struct flow orig_flow; /* Copy of original flow. */
290 static void action_xlate_ctx_init(struct action_xlate_ctx *,
291 struct ofproto_dpif *, const struct flow *,
292 ovs_be16 initial_tci, struct rule_dpif *,
293 uint8_t tcp_flags, const struct ofpbuf *);
294 static void xlate_actions(struct action_xlate_ctx *,
295 const struct ofpact *ofpacts, size_t ofpacts_len,
296 struct ofpbuf *odp_actions);
297 static void xlate_actions_for_side_effects(struct action_xlate_ctx *,
298 const struct ofpact *ofpacts,
300 static void xlate_table_action(struct action_xlate_ctx *, uint16_t in_port,
301 uint8_t table_id, bool may_packet_in);
303 static size_t put_userspace_action(const struct ofproto_dpif *,
304 struct ofpbuf *odp_actions,
306 const union user_action_cookie *);
308 static void compose_slow_path(const struct ofproto_dpif *, const struct flow *,
309 enum slow_path_reason,
310 uint64_t *stub, size_t stub_size,
311 const struct nlattr **actionsp,
312 size_t *actions_lenp);
314 static void xlate_report(struct action_xlate_ctx *ctx, const char *s);
316 /* A subfacet (see "struct subfacet" below) has three possible installation
319 * - SF_NOT_INSTALLED: Not installed in the datapath. This will only be the
320 * case just after the subfacet is created, just before the subfacet is
321 * destroyed, or if the datapath returns an error when we try to install a
324 * - SF_FAST_PATH: The subfacet's actions are installed in the datapath.
326 * - SF_SLOW_PATH: An action that sends every packet for the subfacet through
327 * ofproto_dpif is installed in the datapath.
330 SF_NOT_INSTALLED, /* No datapath flow for this subfacet. */
331 SF_FAST_PATH, /* Full actions are installed. */
332 SF_SLOW_PATH, /* Send-to-userspace action is installed. */
335 static const char *subfacet_path_to_string(enum subfacet_path);
337 /* A dpif flow and actions associated with a facet.
339 * See also the large comment on struct facet. */
342 struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */
343 struct list list_node; /* In struct facet's 'facets' list. */
344 struct facet *facet; /* Owning facet. */
348 * To save memory in the common case, 'key' is NULL if 'key_fitness' is
349 * ODP_FIT_PERFECT, that is, odp_flow_key_from_flow() can accurately
350 * regenerate the ODP flow key from ->facet->flow. */
351 enum odp_key_fitness key_fitness;
355 long long int used; /* Time last used; time created if not used. */
357 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
358 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
362 * These should be essentially identical for every subfacet in a facet, but
363 * may differ in trivial ways due to VLAN splinters. */
364 size_t actions_len; /* Number of bytes in actions[]. */
365 struct nlattr *actions; /* Datapath actions. */
367 enum slow_path_reason slow; /* 0 if fast path may be used. */
368 enum subfacet_path path; /* Installed in datapath? */
370 /* This value is normally the same as ->facet->flow.vlan_tci. Only VLAN
371 * splinters can cause it to differ. This value should be removed when
372 * the VLAN splinters feature is no longer needed. */
373 ovs_be16 initial_tci; /* Initial VLAN TCI value. */
375 /* Datapath port the packet arrived on. This is needed to remove
376 * flows for ports that are no longer part of the bridge. Since the
377 * flow definition only has the OpenFlow port number and the port is
378 * no longer part of the bridge, we can't determine the datapath port
379 * number needed to delete the flow from the datapath. */
380 uint32_t odp_in_port;
383 #define SUBFACET_DESTROY_MAX_BATCH 50
385 static struct subfacet *subfacet_create(struct facet *, struct flow_miss *miss,
387 static struct subfacet *subfacet_find(struct ofproto_dpif *,
388 const struct nlattr *key, size_t key_len,
390 const struct flow *flow);
391 static void subfacet_destroy(struct subfacet *);
392 static void subfacet_destroy__(struct subfacet *);
393 static void subfacet_destroy_batch(struct ofproto_dpif *,
394 struct subfacet **, int n);
395 static void subfacet_get_key(struct subfacet *, struct odputil_keybuf *,
397 static void subfacet_reset_dp_stats(struct subfacet *,
398 struct dpif_flow_stats *);
399 static void subfacet_update_time(struct subfacet *, long long int used);
400 static void subfacet_update_stats(struct subfacet *,
401 const struct dpif_flow_stats *);
402 static void subfacet_make_actions(struct subfacet *,
403 const struct ofpbuf *packet,
404 struct ofpbuf *odp_actions);
405 static int subfacet_install(struct subfacet *,
406 const struct nlattr *actions, size_t actions_len,
407 struct dpif_flow_stats *, enum slow_path_reason);
408 static void subfacet_uninstall(struct subfacet *);
410 static enum subfacet_path subfacet_want_path(enum slow_path_reason);
412 /* An exact-match instantiation of an OpenFlow flow.
414 * A facet associates a "struct flow", which represents the Open vSwitch
415 * userspace idea of an exact-match flow, with one or more subfacets. Each
416 * subfacet tracks the datapath's idea of the exact-match flow equivalent to
417 * the facet. When the kernel module (or other dpif implementation) and Open
418 * vSwitch userspace agree on the definition of a flow key, there is exactly
419 * one subfacet per facet. If the dpif implementation supports more-specific
420 * flow matching than userspace, however, a facet can have more than one
421 * subfacet, each of which corresponds to some distinction in flow that
422 * userspace simply doesn't understand.
424 * Flow expiration works in terms of subfacets, so a facet must have at least
425 * one subfacet or it will never expire, leaking memory. */
428 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
429 struct list list_node; /* In owning rule's 'facets' list. */
430 struct rule_dpif *rule; /* Owning rule. */
433 struct list subfacets;
434 long long int used; /* Time last used; time created if not used. */
441 * - Do include packets and bytes sent "by hand", e.g. with
444 * - Do include packets and bytes that were obtained from the datapath
445 * when a subfacet's statistics were reset (e.g. dpif_flow_put() with
446 * DPIF_FP_ZERO_STATS).
448 * - Do not include packets or bytes that can be obtained from the
449 * datapath for any existing subfacet.
451 uint64_t packet_count; /* Number of packets received. */
452 uint64_t byte_count; /* Number of bytes received. */
454 /* Resubmit statistics. */
455 uint64_t prev_packet_count; /* Number of packets from last stats push. */
456 uint64_t prev_byte_count; /* Number of bytes from last stats push. */
457 long long int prev_used; /* Used time from last stats push. */
460 uint64_t accounted_bytes; /* Bytes processed by facet_account(). */
461 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
462 uint8_t tcp_flags; /* TCP flags seen for this 'rule'. */
464 /* Properties of datapath actions.
466 * Every subfacet has its own actions because actions can differ slightly
467 * between splintered and non-splintered subfacets due to the VLAN tag
468 * being initially different (present vs. absent). All of them have these
469 * properties in common so we just store one copy of them here. */
470 bool has_learn; /* Actions include NXAST_LEARN? */
471 bool has_normal; /* Actions output to OFPP_NORMAL? */
472 bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */
473 tag_type tags; /* Tags that would require revalidation. */
474 mirror_mask_t mirrors; /* Bitmap of dependent mirrors. */
476 /* Storage for a single subfacet, to reduce malloc() time and space
477 * overhead. (A facet always has at least one subfacet and in the common
478 * case has exactly one subfacet.) */
479 struct subfacet one_subfacet;
482 static struct facet *facet_create(struct rule_dpif *,
483 const struct flow *, uint32_t hash);
484 static void facet_remove(struct facet *);
485 static void facet_free(struct facet *);
487 static struct facet *facet_find(struct ofproto_dpif *,
488 const struct flow *, uint32_t hash);
489 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
490 const struct flow *, uint32_t hash);
491 static void facet_revalidate(struct facet *);
492 static bool facet_check_consistency(struct facet *);
494 static void facet_flush_stats(struct facet *);
496 static void facet_update_time(struct facet *, long long int used);
497 static void facet_reset_counters(struct facet *);
498 static void facet_push_stats(struct facet *);
499 static void facet_learn(struct facet *);
500 static void facet_account(struct facet *);
502 static bool facet_is_controller_flow(struct facet *);
505 struct hmap_node odp_port_node; /* In dpif_backer's "odp_to_ofport_map". */
509 struct ofbundle *bundle; /* Bundle that contains this port, if any. */
510 struct list bundle_node; /* In struct ofbundle's "ports" list. */
511 struct cfm *cfm; /* Connectivity Fault Management, if any. */
512 tag_type tag; /* Tag associated with this port. */
513 uint32_t bond_stable_id; /* stable_id to use as bond slave, or 0. */
514 bool may_enable; /* May be enabled in bonds. */
515 long long int carrier_seq; /* Carrier status changes. */
516 struct tnl_port *tnl_port; /* Tunnel handle, or null. */
519 struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */
520 enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */
521 long long int stp_state_entered;
523 struct hmap priorities; /* Map of attached 'priority_to_dscp's. */
525 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
527 * This is deprecated. It is only for compatibility with broken device
528 * drivers in old versions of Linux that do not properly support VLANs when
529 * VLAN devices are not used. When broken device drivers are no longer in
530 * widespread use, we will delete these interfaces. */
531 uint16_t realdev_ofp_port;
535 /* Node in 'ofport_dpif''s 'priorities' map. Used to maintain a map from
536 * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
537 * traffic egressing the 'ofport' with that priority should be marked with. */
538 struct priority_to_dscp {
539 struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'priorities' map. */
540 uint32_t priority; /* Priority of this queue (see struct flow). */
542 uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */
545 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
547 * This is deprecated. It is only for compatibility with broken device drivers
548 * in old versions of Linux that do not properly support VLANs when VLAN
549 * devices are not used. When broken device drivers are no longer in
550 * widespread use, we will delete these interfaces. */
551 struct vlan_splinter {
552 struct hmap_node realdev_vid_node;
553 struct hmap_node vlandev_node;
554 uint16_t realdev_ofp_port;
555 uint16_t vlandev_ofp_port;
559 static uint32_t vsp_realdev_to_vlandev(const struct ofproto_dpif *,
560 uint32_t realdev, ovs_be16 vlan_tci);
561 static bool vsp_adjust_flow(const struct ofproto_dpif *, struct flow *);
562 static void vsp_remove(struct ofport_dpif *);
563 static void vsp_add(struct ofport_dpif *, uint16_t realdev_ofp_port, int vid);
565 static uint32_t ofp_port_to_odp_port(const struct ofproto_dpif *,
567 static uint16_t odp_port_to_ofp_port(const struct ofproto_dpif *,
570 static struct ofport_dpif *
571 ofport_dpif_cast(const struct ofport *ofport)
573 ovs_assert(ofport->ofproto->ofproto_class == &ofproto_dpif_class);
574 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
577 static void port_run(struct ofport_dpif *);
578 static void port_run_fast(struct ofport_dpif *);
579 static void port_wait(struct ofport_dpif *);
580 static int set_cfm(struct ofport *, const struct cfm_settings *);
581 static void ofport_clear_priorities(struct ofport_dpif *);
583 struct dpif_completion {
584 struct list list_node;
585 struct ofoperation *op;
588 /* Extra information about a classifier table.
589 * Currently used just for optimized flow revalidation. */
591 /* If either of these is nonnull, then this table has a form that allows
592 * flows to be tagged to avoid revalidating most flows for the most common
593 * kinds of flow table changes. */
594 struct cls_table *catchall_table; /* Table that wildcards all fields. */
595 struct cls_table *other_table; /* Table with any other wildcard set. */
596 uint32_t basis; /* Keeps each table's tags separate. */
599 /* Reasons that we might need to revalidate every facet, and corresponding
602 * A value of 0 means that there is no need to revalidate.
604 * It would be nice to have some cleaner way to integrate with coverage
605 * counters, but with only a few reasons I guess this is good enough for
607 enum revalidate_reason {
608 REV_RECONFIGURE = 1, /* Switch configuration changed. */
609 REV_STP, /* Spanning tree protocol port status change. */
610 REV_PORT_TOGGLED, /* Port enabled or disabled by CFM, LACP, ...*/
611 REV_FLOW_TABLE, /* Flow table changed. */
612 REV_INCONSISTENCY /* Facet self-check failed. */
614 COVERAGE_DEFINE(rev_reconfigure);
615 COVERAGE_DEFINE(rev_stp);
616 COVERAGE_DEFINE(rev_port_toggled);
617 COVERAGE_DEFINE(rev_flow_table);
618 COVERAGE_DEFINE(rev_inconsistency);
620 /* Drop keys are odp flow keys which have drop flows installed in the kernel.
621 * These are datapath flows which have no associated ofproto, if they did we
622 * would use facets. */
624 struct hmap_node hmap_node;
629 /* All datapaths of a given type share a single dpif backer instance. */
634 struct timer next_expiration;
635 struct hmap odp_to_ofport_map; /* ODP port to ofport mapping. */
637 struct sset tnl_backers; /* Set of dpif ports backing tunnels. */
639 /* Facet revalidation flags applying to facets which use this backer. */
640 enum revalidate_reason need_revalidate; /* Revalidate every facet. */
641 struct tag_set revalidate_set; /* Revalidate only matching facets. */
643 struct hmap drop_keys; /* Set of dropped odp keys. */
646 /* All existing ofproto_backer instances, indexed by ofproto->up.type. */
647 static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers);
649 static void drop_key_clear(struct dpif_backer *);
650 static struct ofport_dpif *
651 odp_port_to_ofport(const struct dpif_backer *, uint32_t odp_port);
653 struct ofproto_dpif {
654 struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
656 struct dpif_backer *backer;
658 /* Special OpenFlow rules. */
659 struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */
660 struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */
666 struct netflow *netflow;
667 struct dpif_sflow *sflow;
668 struct hmap bundles; /* Contains "struct ofbundle"s. */
669 struct mac_learning *ml;
670 struct ofmirror *mirrors[MAX_MIRRORS];
672 bool has_bonded_bundles;
676 struct hmap subfacets;
677 struct governor *governor;
680 struct table_dpif tables[N_TABLES];
682 /* Support for debugging async flow mods. */
683 struct list completions;
685 bool has_bundle_action; /* True when the first bundle action appears. */
686 struct netdev_stats stats; /* To account packets generated and consumed in
691 long long int stp_last_tick;
693 /* VLAN splinters. */
694 struct hmap realdev_vid_map; /* (realdev,vid) -> vlandev. */
695 struct hmap vlandev_map; /* vlandev -> (realdev,vid). */
698 struct sset ports; /* Set of standard port names. */
699 struct sset ghost_ports; /* Ports with no datapath port. */
700 struct sset port_poll_set; /* Queued names for port_poll() reply. */
701 int port_poll_errno; /* Last errno for port_poll() reply. */
704 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
705 * for debugging the asynchronous flow_mod implementation.) */
708 /* All existing ofproto_dpif instances, indexed by ->up.name. */
709 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
711 static void ofproto_dpif_unixctl_init(void);
713 static struct ofproto_dpif *
714 ofproto_dpif_cast(const struct ofproto *ofproto)
716 ovs_assert(ofproto->ofproto_class == &ofproto_dpif_class);
717 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
720 static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *,
722 static struct ofport_dpif *get_odp_port(const struct ofproto_dpif *,
724 static void ofproto_trace(struct ofproto_dpif *, const struct flow *,
725 const struct ofpbuf *, ovs_be16 initial_tci,
727 static bool may_dpif_port_del(struct ofport_dpif *);
729 /* Packet processing. */
730 static void update_learning_table(struct ofproto_dpif *,
731 const struct flow *, int vlan,
734 #define FLOW_MISS_MAX_BATCH 50
735 static int handle_upcalls(struct dpif_backer *, unsigned int max_batch);
737 /* Flow expiration. */
738 static int expire(struct dpif_backer *);
741 static void send_netflow_active_timeouts(struct ofproto_dpif *);
744 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
745 static size_t compose_sflow_action(const struct ofproto_dpif *,
746 struct ofpbuf *odp_actions,
747 const struct flow *, uint32_t odp_port);
748 static void add_mirror_actions(struct action_xlate_ctx *ctx,
749 const struct flow *flow);
750 /* Global variables. */
751 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
753 /* Initial mappings of port to bridge mappings. */
754 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
756 /* Factory functions. */
759 init(const struct shash *iface_hints)
761 struct shash_node *node;
763 /* Make a local copy, since we don't own 'iface_hints' elements. */
764 SHASH_FOR_EACH(node, iface_hints) {
765 const struct iface_hint *orig_hint = node->data;
766 struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
768 new_hint->br_name = xstrdup(orig_hint->br_name);
769 new_hint->br_type = xstrdup(orig_hint->br_type);
770 new_hint->ofp_port = orig_hint->ofp_port;
772 shash_add(&init_ofp_ports, node->name, new_hint);
777 enumerate_types(struct sset *types)
779 dp_enumerate_types(types);
783 enumerate_names(const char *type, struct sset *names)
785 struct ofproto_dpif *ofproto;
788 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
789 if (strcmp(type, ofproto->up.type)) {
792 sset_add(names, ofproto->up.name);
799 del(const char *type, const char *name)
804 error = dpif_open(name, type, &dpif);
806 error = dpif_delete(dpif);
813 port_open_type(const char *datapath_type, const char *port_type)
815 return dpif_port_open_type(datapath_type, port_type);
818 /* Type functions. */
820 static struct ofproto_dpif *
821 lookup_ofproto_dpif_by_port_name(const char *name)
823 struct ofproto_dpif *ofproto;
825 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
826 if (sset_contains(&ofproto->ports, name)) {
835 type_run(const char *type)
837 struct dpif_backer *backer;
841 backer = shash_find_data(&all_dpif_backers, type);
843 /* This is not necessarily a problem, since backers are only
844 * created on demand. */
848 dpif_run(backer->dpif);
850 if (backer->need_revalidate
851 || !tag_set_is_empty(&backer->revalidate_set)) {
852 struct tag_set revalidate_set = backer->revalidate_set;
853 bool need_revalidate = backer->need_revalidate;
854 struct ofproto_dpif *ofproto;
856 switch (backer->need_revalidate) {
857 case REV_RECONFIGURE: COVERAGE_INC(rev_reconfigure); break;
858 case REV_STP: COVERAGE_INC(rev_stp); break;
859 case REV_PORT_TOGGLED: COVERAGE_INC(rev_port_toggled); break;
860 case REV_FLOW_TABLE: COVERAGE_INC(rev_flow_table); break;
861 case REV_INCONSISTENCY: COVERAGE_INC(rev_inconsistency); break;
864 if (backer->need_revalidate) {
865 /* Clear the drop_keys in case we should now be accepting some
866 * formerly dropped flows. */
867 drop_key_clear(backer);
870 /* Clear the revalidation flags. */
871 tag_set_init(&backer->revalidate_set);
872 backer->need_revalidate = 0;
874 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
877 if (ofproto->backer != backer) {
881 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
883 || tag_set_intersects(&revalidate_set, facet->tags)) {
884 facet_revalidate(facet);
890 if (timer_expired(&backer->next_expiration)) {
891 int delay = expire(backer);
892 timer_set_duration(&backer->next_expiration, delay);
895 /* Check for port changes in the dpif. */
896 while ((error = dpif_port_poll(backer->dpif, &devname)) == 0) {
897 struct ofproto_dpif *ofproto;
898 struct dpif_port port;
900 /* Don't report on the datapath's device. */
901 if (!strcmp(devname, dpif_base_name(backer->dpif))) {
905 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
906 &all_ofproto_dpifs) {
907 if (sset_contains(&ofproto->backer->tnl_backers, devname)) {
912 ofproto = lookup_ofproto_dpif_by_port_name(devname);
913 if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
914 /* The port was removed. If we know the datapath,
915 * report it through poll_set(). If we don't, it may be
916 * notifying us of a removal we initiated, so ignore it.
917 * If there's a pending ENOBUFS, let it stand, since
918 * everything will be reevaluated. */
919 if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
920 sset_add(&ofproto->port_poll_set, devname);
921 ofproto->port_poll_errno = 0;
923 } else if (!ofproto) {
924 /* The port was added, but we don't know with which
925 * ofproto we should associate it. Delete it. */
926 dpif_port_del(backer->dpif, port.port_no);
928 dpif_port_destroy(&port);
934 if (error != EAGAIN) {
935 struct ofproto_dpif *ofproto;
937 /* There was some sort of error, so propagate it to all
938 * ofprotos that use this backer. */
939 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
940 &all_ofproto_dpifs) {
941 if (ofproto->backer == backer) {
942 sset_clear(&ofproto->port_poll_set);
943 ofproto->port_poll_errno = error;
952 type_run_fast(const char *type)
954 struct dpif_backer *backer;
957 backer = shash_find_data(&all_dpif_backers, type);
959 /* This is not necessarily a problem, since backers are only
960 * created on demand. */
964 /* Handle one or more batches of upcalls, until there's nothing left to do
965 * or until we do a fixed total amount of work.
967 * We do work in batches because it can be much cheaper to set up a number
968 * of flows and fire off their patches all at once. We do multiple batches
969 * because in some cases handling a packet can cause another packet to be
970 * queued almost immediately as part of the return flow. Both
971 * optimizations can make major improvements on some benchmarks and
972 * presumably for real traffic as well. */
974 while (work < FLOW_MISS_MAX_BATCH) {
975 int retval = handle_upcalls(backer, FLOW_MISS_MAX_BATCH - work);
986 type_wait(const char *type)
988 struct dpif_backer *backer;
990 backer = shash_find_data(&all_dpif_backers, type);
992 /* This is not necessarily a problem, since backers are only
993 * created on demand. */
997 timer_wait(&backer->next_expiration);
1000 /* Basic life-cycle. */
1002 static int add_internal_flows(struct ofproto_dpif *);
1004 static struct ofproto *
1007 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
1008 return &ofproto->up;
1012 dealloc(struct ofproto *ofproto_)
1014 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1019 close_dpif_backer(struct dpif_backer *backer)
1021 struct shash_node *node;
1023 ovs_assert(backer->refcount > 0);
1025 if (--backer->refcount) {
1029 drop_key_clear(backer);
1030 hmap_destroy(&backer->drop_keys);
1032 sset_destroy(&backer->tnl_backers);
1033 hmap_destroy(&backer->odp_to_ofport_map);
1034 node = shash_find(&all_dpif_backers, backer->type);
1036 shash_delete(&all_dpif_backers, node);
1037 dpif_close(backer->dpif);
1042 /* Datapath port slated for removal from datapath. */
1043 struct odp_garbage {
1044 struct list list_node;
1049 open_dpif_backer(const char *type, struct dpif_backer **backerp)
1051 struct dpif_backer *backer;
1052 struct dpif_port_dump port_dump;
1053 struct dpif_port port;
1054 struct shash_node *node;
1055 struct list garbage_list;
1056 struct odp_garbage *garbage, *next;
1062 backer = shash_find_data(&all_dpif_backers, type);
1069 backer_name = xasprintf("ovs-%s", type);
1071 /* Remove any existing datapaths, since we assume we're the only
1072 * userspace controlling the datapath. */
1074 dp_enumerate_names(type, &names);
1075 SSET_FOR_EACH(name, &names) {
1076 struct dpif *old_dpif;
1078 /* Don't remove our backer if it exists. */
1079 if (!strcmp(name, backer_name)) {
1083 if (dpif_open(name, type, &old_dpif)) {
1084 VLOG_WARN("couldn't open old datapath %s to remove it", name);
1086 dpif_delete(old_dpif);
1087 dpif_close(old_dpif);
1090 sset_destroy(&names);
1092 backer = xmalloc(sizeof *backer);
1094 error = dpif_create_and_open(backer_name, type, &backer->dpif);
1097 VLOG_ERR("failed to open datapath of type %s: %s", type,
1103 backer->type = xstrdup(type);
1104 backer->refcount = 1;
1105 hmap_init(&backer->odp_to_ofport_map);
1106 hmap_init(&backer->drop_keys);
1107 timer_set_duration(&backer->next_expiration, 1000);
1108 backer->need_revalidate = 0;
1109 sset_init(&backer->tnl_backers);
1110 tag_set_init(&backer->revalidate_set);
1113 dpif_flow_flush(backer->dpif);
1115 /* Loop through the ports already on the datapath and remove any
1116 * that we don't need anymore. */
1117 list_init(&garbage_list);
1118 dpif_port_dump_start(&port_dump, backer->dpif);
1119 while (dpif_port_dump_next(&port_dump, &port)) {
1120 node = shash_find(&init_ofp_ports, port.name);
1121 if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
1122 garbage = xmalloc(sizeof *garbage);
1123 garbage->odp_port = port.port_no;
1124 list_push_front(&garbage_list, &garbage->list_node);
1127 dpif_port_dump_done(&port_dump);
1129 LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) {
1130 dpif_port_del(backer->dpif, garbage->odp_port);
1131 list_remove(&garbage->list_node);
1135 shash_add(&all_dpif_backers, type, backer);
1137 error = dpif_recv_set(backer->dpif, true);
1139 VLOG_ERR("failed to listen on datapath of type %s: %s",
1140 type, strerror(error));
1141 close_dpif_backer(backer);
1149 construct(struct ofproto *ofproto_)
1151 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1152 struct shash_node *node, *next;
1157 error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1162 max_ports = dpif_get_max_ports(ofproto->backer->dpif);
1163 ofproto_init_max_ports(ofproto_, MIN(max_ports, OFPP_MAX));
1165 ofproto->n_matches = 0;
1167 ofproto->netflow = NULL;
1168 ofproto->sflow = NULL;
1169 ofproto->stp = NULL;
1170 hmap_init(&ofproto->bundles);
1171 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1172 for (i = 0; i < MAX_MIRRORS; i++) {
1173 ofproto->mirrors[i] = NULL;
1175 ofproto->has_bonded_bundles = false;
1177 hmap_init(&ofproto->facets);
1178 hmap_init(&ofproto->subfacets);
1179 ofproto->governor = NULL;
1181 for (i = 0; i < N_TABLES; i++) {
1182 struct table_dpif *table = &ofproto->tables[i];
1184 table->catchall_table = NULL;
1185 table->other_table = NULL;
1186 table->basis = random_uint32();
1189 list_init(&ofproto->completions);
1191 ofproto_dpif_unixctl_init();
1193 ofproto->has_mirrors = false;
1194 ofproto->has_bundle_action = false;
1196 hmap_init(&ofproto->vlandev_map);
1197 hmap_init(&ofproto->realdev_vid_map);
1199 sset_init(&ofproto->ports);
1200 sset_init(&ofproto->ghost_ports);
1201 sset_init(&ofproto->port_poll_set);
1202 ofproto->port_poll_errno = 0;
1204 SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1205 struct iface_hint *iface_hint = node->data;
1207 if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1208 /* Check if the datapath already has this port. */
1209 if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1210 sset_add(&ofproto->ports, node->name);
1213 free(iface_hint->br_name);
1214 free(iface_hint->br_type);
1216 shash_delete(&init_ofp_ports, node);
1220 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1221 hash_string(ofproto->up.name, 0));
1222 memset(&ofproto->stats, 0, sizeof ofproto->stats);
1224 ofproto_init_tables(ofproto_, N_TABLES);
1225 error = add_internal_flows(ofproto);
1226 ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1232 add_internal_flow(struct ofproto_dpif *ofproto, int id,
1233 const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1235 struct ofputil_flow_mod fm;
1238 match_init_catchall(&fm.match);
1240 match_set_reg(&fm.match, 0, id);
1241 fm.new_cookie = htonll(0);
1242 fm.cookie = htonll(0);
1243 fm.cookie_mask = htonll(0);
1244 fm.table_id = TBL_INTERNAL;
1245 fm.command = OFPFC_ADD;
1246 fm.idle_timeout = 0;
1247 fm.hard_timeout = 0;
1251 fm.ofpacts = ofpacts->data;
1252 fm.ofpacts_len = ofpacts->size;
1254 error = ofproto_flow_mod(&ofproto->up, &fm);
1256 VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
1257 id, ofperr_to_string(error));
1261 *rulep = rule_dpif_lookup__(ofproto, &fm.match.flow, TBL_INTERNAL);
1262 ovs_assert(*rulep != NULL);
1268 add_internal_flows(struct ofproto_dpif *ofproto)
1270 struct ofpact_controller *controller;
1271 uint64_t ofpacts_stub[128 / 8];
1272 struct ofpbuf ofpacts;
1276 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1279 controller = ofpact_put_CONTROLLER(&ofpacts);
1280 controller->max_len = UINT16_MAX;
1281 controller->controller_id = 0;
1282 controller->reason = OFPR_NO_MATCH;
1283 ofpact_pad(&ofpacts);
1285 error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule);
1290 ofpbuf_clear(&ofpacts);
1291 error = add_internal_flow(ofproto, id++, &ofpacts,
1292 &ofproto->no_packet_in_rule);
1297 complete_operations(struct ofproto_dpif *ofproto)
1299 struct dpif_completion *c, *next;
1301 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
1302 ofoperation_complete(c->op, 0);
1303 list_remove(&c->list_node);
1309 destruct(struct ofproto *ofproto_)
1311 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1312 struct rule_dpif *rule, *next_rule;
1313 struct oftable *table;
1316 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1317 complete_operations(ofproto);
1319 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1320 struct cls_cursor cursor;
1322 cls_cursor_init(&cursor, &table->cls, NULL);
1323 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1324 ofproto_rule_destroy(&rule->up);
1328 for (i = 0; i < MAX_MIRRORS; i++) {
1329 mirror_destroy(ofproto->mirrors[i]);
1332 netflow_destroy(ofproto->netflow);
1333 dpif_sflow_destroy(ofproto->sflow);
1334 hmap_destroy(&ofproto->bundles);
1335 mac_learning_destroy(ofproto->ml);
1337 hmap_destroy(&ofproto->facets);
1338 hmap_destroy(&ofproto->subfacets);
1339 governor_destroy(ofproto->governor);
1341 hmap_destroy(&ofproto->vlandev_map);
1342 hmap_destroy(&ofproto->realdev_vid_map);
1344 sset_destroy(&ofproto->ports);
1345 sset_destroy(&ofproto->ghost_ports);
1346 sset_destroy(&ofproto->port_poll_set);
1348 close_dpif_backer(ofproto->backer);
1352 run_fast(struct ofproto *ofproto_)
1354 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1355 struct ofport_dpif *ofport;
1357 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1358 port_run_fast(ofport);
1365 run(struct ofproto *ofproto_)
1367 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1368 struct ofport_dpif *ofport;
1369 struct ofbundle *bundle;
1373 complete_operations(ofproto);
1376 error = run_fast(ofproto_);
1381 if (ofproto->netflow) {
1382 if (netflow_run(ofproto->netflow)) {
1383 send_netflow_active_timeouts(ofproto);
1386 if (ofproto->sflow) {
1387 dpif_sflow_run(ofproto->sflow);
1390 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1393 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1398 mac_learning_run(ofproto->ml, &ofproto->backer->revalidate_set);
1400 /* Check the consistency of a random facet, to aid debugging. */
1401 if (!hmap_is_empty(&ofproto->facets)
1402 && !ofproto->backer->need_revalidate) {
1403 struct facet *facet;
1405 facet = CONTAINER_OF(hmap_random_node(&ofproto->facets),
1406 struct facet, hmap_node);
1407 if (!tag_set_intersects(&ofproto->backer->revalidate_set,
1409 if (!facet_check_consistency(facet)) {
1410 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
1415 if (ofproto->governor) {
1418 governor_run(ofproto->governor);
1420 /* If the governor has shrunk to its minimum size and the number of
1421 * subfacets has dwindled, then drop the governor entirely.
1423 * For hysteresis, the number of subfacets to drop the governor is
1424 * smaller than the number needed to trigger its creation. */
1425 n_subfacets = hmap_count(&ofproto->subfacets);
1426 if (n_subfacets * 4 < ofproto->up.flow_eviction_threshold
1427 && governor_is_idle(ofproto->governor)) {
1428 governor_destroy(ofproto->governor);
1429 ofproto->governor = NULL;
1437 wait(struct ofproto *ofproto_)
1439 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1440 struct ofport_dpif *ofport;
1441 struct ofbundle *bundle;
1443 if (!clogged && !list_is_empty(&ofproto->completions)) {
1444 poll_immediate_wake();
1447 dpif_wait(ofproto->backer->dpif);
1448 dpif_recv_wait(ofproto->backer->dpif);
1449 if (ofproto->sflow) {
1450 dpif_sflow_wait(ofproto->sflow);
1452 if (!tag_set_is_empty(&ofproto->backer->revalidate_set)) {
1453 poll_immediate_wake();
1455 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1458 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1459 bundle_wait(bundle);
1461 if (ofproto->netflow) {
1462 netflow_wait(ofproto->netflow);
1464 mac_learning_wait(ofproto->ml);
1466 if (ofproto->backer->need_revalidate) {
1467 /* Shouldn't happen, but if it does just go around again. */
1468 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1469 poll_immediate_wake();
1471 if (ofproto->governor) {
1472 governor_wait(ofproto->governor);
1477 get_memory_usage(const struct ofproto *ofproto_, struct simap *usage)
1479 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1481 simap_increase(usage, "facets", hmap_count(&ofproto->facets));
1482 simap_increase(usage, "subfacets", hmap_count(&ofproto->subfacets));
1486 flush(struct ofproto *ofproto_)
1488 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1489 struct subfacet *subfacet, *next_subfacet;
1490 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
1494 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
1495 &ofproto->subfacets) {
1496 if (subfacet->path != SF_NOT_INSTALLED) {
1497 batch[n_batch++] = subfacet;
1498 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
1499 subfacet_destroy_batch(ofproto, batch, n_batch);
1503 subfacet_destroy(subfacet);
1508 subfacet_destroy_batch(ofproto, batch, n_batch);
1513 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1514 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1516 *arp_match_ip = true;
1517 *actions = (OFPUTIL_A_OUTPUT |
1518 OFPUTIL_A_SET_VLAN_VID |
1519 OFPUTIL_A_SET_VLAN_PCP |
1520 OFPUTIL_A_STRIP_VLAN |
1521 OFPUTIL_A_SET_DL_SRC |
1522 OFPUTIL_A_SET_DL_DST |
1523 OFPUTIL_A_SET_NW_SRC |
1524 OFPUTIL_A_SET_NW_DST |
1525 OFPUTIL_A_SET_NW_TOS |
1526 OFPUTIL_A_SET_TP_SRC |
1527 OFPUTIL_A_SET_TP_DST |
1532 get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots)
1534 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1535 struct dpif_dp_stats s;
1537 strcpy(ots->name, "classifier");
1539 dpif_get_dp_stats(ofproto->backer->dpif, &s);
1541 ots->lookup_count = htonll(s.n_hit + s.n_missed);
1542 ots->matched_count = htonll(s.n_hit + ofproto->n_matches);
1545 static struct ofport *
1548 struct ofport_dpif *port = xmalloc(sizeof *port);
1553 port_dealloc(struct ofport *port_)
1555 struct ofport_dpif *port = ofport_dpif_cast(port_);
1560 port_construct(struct ofport *port_)
1562 struct ofport_dpif *port = ofport_dpif_cast(port_);
1563 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1564 const struct netdev *netdev = port->up.netdev;
1565 struct dpif_port dpif_port;
1568 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1569 port->bundle = NULL;
1571 port->tag = tag_create_random();
1572 port->may_enable = true;
1573 port->stp_port = NULL;
1574 port->stp_state = STP_DISABLED;
1575 port->tnl_port = NULL;
1576 hmap_init(&port->priorities);
1577 port->realdev_ofp_port = 0;
1578 port->vlandev_vid = 0;
1579 port->carrier_seq = netdev_get_carrier_resets(netdev);
1581 if (netdev_vport_is_patch(netdev)) {
1582 /* XXX By bailing out here, we don't do required sFlow work. */
1583 port->odp_port = OVSP_NONE;
1587 error = dpif_port_query_by_name(ofproto->backer->dpif,
1588 netdev_vport_get_dpif_port(netdev),
1594 port->odp_port = dpif_port.port_no;
1596 if (netdev_get_tunnel_config(netdev)) {
1597 port->tnl_port = tnl_port_add(&port->up, port->odp_port);
1599 /* Sanity-check that a mapping doesn't already exist. This
1600 * shouldn't happen for non-tunnel ports. */
1601 if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1602 VLOG_ERR("port %s already has an OpenFlow port number",
1604 dpif_port_destroy(&dpif_port);
1608 hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1609 hash_int(port->odp_port, 0));
1611 dpif_port_destroy(&dpif_port);
1613 if (ofproto->sflow) {
1614 dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1621 port_destruct(struct ofport *port_)
1623 struct ofport_dpif *port = ofport_dpif_cast(port_);
1624 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1625 const char *dp_port_name = netdev_vport_get_dpif_port(port->up.netdev);
1626 const char *devname = netdev_get_name(port->up.netdev);
1628 if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)
1629 && may_dpif_port_del(port)) {
1630 /* The underlying device is still there, so delete it. This
1631 * happens when the ofproto is being destroyed, since the caller
1632 * assumes that removal of attached ports will happen as part of
1634 dpif_port_del(ofproto->backer->dpif, port->odp_port);
1635 sset_find_and_delete(&ofproto->backer->tnl_backers, dp_port_name);
1638 if (port->odp_port != OVSP_NONE && !port->tnl_port) {
1639 hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1642 tnl_port_del(port->tnl_port);
1643 sset_find_and_delete(&ofproto->ports, devname);
1644 sset_find_and_delete(&ofproto->ghost_ports, devname);
1645 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1646 bundle_remove(port_);
1647 set_cfm(port_, NULL);
1648 if (ofproto->sflow) {
1649 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1652 ofport_clear_priorities(port);
1653 hmap_destroy(&port->priorities);
1657 port_modified(struct ofport *port_)
1659 struct ofport_dpif *port = ofport_dpif_cast(port_);
1661 if (port->bundle && port->bundle->bond) {
1662 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1667 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1669 struct ofport_dpif *port = ofport_dpif_cast(port_);
1670 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1671 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1673 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1674 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1675 OFPUTIL_PC_NO_PACKET_IN)) {
1676 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1678 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1679 bundle_update(port->bundle);
1685 set_sflow(struct ofproto *ofproto_,
1686 const struct ofproto_sflow_options *sflow_options)
1688 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1689 struct dpif_sflow *ds = ofproto->sflow;
1691 if (sflow_options) {
1693 struct ofport_dpif *ofport;
1695 ds = ofproto->sflow = dpif_sflow_create();
1696 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1697 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1699 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1701 dpif_sflow_set_options(ds, sflow_options);
1704 dpif_sflow_destroy(ds);
1705 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1706 ofproto->sflow = NULL;
1713 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1715 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1722 struct ofproto_dpif *ofproto;
1724 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1725 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1726 ofport->cfm = cfm_create(netdev_get_name(ofport->up.netdev));
1729 if (cfm_configure(ofport->cfm, s)) {
1735 cfm_destroy(ofport->cfm);
1741 get_cfm_fault(const struct ofport *ofport_)
1743 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1745 return ofport->cfm ? cfm_get_fault(ofport->cfm) : -1;
1749 get_cfm_opup(const struct ofport *ofport_)
1751 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1753 return ofport->cfm ? cfm_get_opup(ofport->cfm) : -1;
1757 get_cfm_remote_mpids(const struct ofport *ofport_, const uint64_t **rmps,
1760 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1763 cfm_get_remote_mpids(ofport->cfm, rmps, n_rmps);
1771 get_cfm_health(const struct ofport *ofport_)
1773 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1775 return ofport->cfm ? cfm_get_health(ofport->cfm) : -1;
1778 /* Spanning Tree. */
1781 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
1783 struct ofproto_dpif *ofproto = ofproto_;
1784 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
1785 struct ofport_dpif *ofport;
1787 ofport = stp_port_get_aux(sp);
1789 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
1790 ofproto->up.name, port_num);
1792 struct eth_header *eth = pkt->l2;
1794 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
1795 if (eth_addr_is_zero(eth->eth_src)) {
1796 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
1797 "with unknown MAC", ofproto->up.name, port_num);
1799 send_packet(ofport, pkt);
1805 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
1807 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
1809 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1811 /* Only revalidate flows if the configuration changed. */
1812 if (!s != !ofproto->stp) {
1813 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1817 if (!ofproto->stp) {
1818 ofproto->stp = stp_create(ofproto_->name, s->system_id,
1819 send_bpdu_cb, ofproto);
1820 ofproto->stp_last_tick = time_msec();
1823 stp_set_bridge_id(ofproto->stp, s->system_id);
1824 stp_set_bridge_priority(ofproto->stp, s->priority);
1825 stp_set_hello_time(ofproto->stp, s->hello_time);
1826 stp_set_max_age(ofproto->stp, s->max_age);
1827 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
1829 struct ofport *ofport;
1831 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
1832 set_stp_port(ofport, NULL);
1835 stp_destroy(ofproto->stp);
1836 ofproto->stp = NULL;
1843 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
1845 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1849 s->bridge_id = stp_get_bridge_id(ofproto->stp);
1850 s->designated_root = stp_get_designated_root(ofproto->stp);
1851 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
1860 update_stp_port_state(struct ofport_dpif *ofport)
1862 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1863 enum stp_state state;
1865 /* Figure out new state. */
1866 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
1870 if (ofport->stp_state != state) {
1871 enum ofputil_port_state of_state;
1874 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
1875 netdev_get_name(ofport->up.netdev),
1876 stp_state_name(ofport->stp_state),
1877 stp_state_name(state));
1878 if (stp_learn_in_state(ofport->stp_state)
1879 != stp_learn_in_state(state)) {
1880 /* xxx Learning action flows should also be flushed. */
1881 mac_learning_flush(ofproto->ml,
1882 &ofproto->backer->revalidate_set);
1884 fwd_change = stp_forward_in_state(ofport->stp_state)
1885 != stp_forward_in_state(state);
1887 ofproto->backer->need_revalidate = REV_STP;
1888 ofport->stp_state = state;
1889 ofport->stp_state_entered = time_msec();
1891 if (fwd_change && ofport->bundle) {
1892 bundle_update(ofport->bundle);
1895 /* Update the STP state bits in the OpenFlow port description. */
1896 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
1897 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
1898 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
1899 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
1900 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
1902 ofproto_port_set_state(&ofport->up, of_state);
1906 /* Configures STP on 'ofport_' using the settings defined in 's'. The
1907 * caller is responsible for assigning STP port numbers and ensuring
1908 * there are no duplicates. */
1910 set_stp_port(struct ofport *ofport_,
1911 const struct ofproto_port_stp_settings *s)
1913 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1914 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1915 struct stp_port *sp = ofport->stp_port;
1917 if (!s || !s->enable) {
1919 ofport->stp_port = NULL;
1920 stp_port_disable(sp);
1921 update_stp_port_state(ofport);
1924 } else if (sp && stp_port_no(sp) != s->port_num
1925 && ofport == stp_port_get_aux(sp)) {
1926 /* The port-id changed, so disable the old one if it's not
1927 * already in use by another port. */
1928 stp_port_disable(sp);
1931 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
1932 stp_port_enable(sp);
1934 stp_port_set_aux(sp, ofport);
1935 stp_port_set_priority(sp, s->priority);
1936 stp_port_set_path_cost(sp, s->path_cost);
1938 update_stp_port_state(ofport);
1944 get_stp_port_status(struct ofport *ofport_,
1945 struct ofproto_port_stp_status *s)
1947 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1948 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1949 struct stp_port *sp = ofport->stp_port;
1951 if (!ofproto->stp || !sp) {
1957 s->port_id = stp_port_get_id(sp);
1958 s->state = stp_port_get_state(sp);
1959 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
1960 s->role = stp_port_get_role(sp);
1961 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
1967 stp_run(struct ofproto_dpif *ofproto)
1970 long long int now = time_msec();
1971 long long int elapsed = now - ofproto->stp_last_tick;
1972 struct stp_port *sp;
1975 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
1976 ofproto->stp_last_tick = now;
1978 while (stp_get_changed_port(ofproto->stp, &sp)) {
1979 struct ofport_dpif *ofport = stp_port_get_aux(sp);
1982 update_stp_port_state(ofport);
1986 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
1987 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
1993 stp_wait(struct ofproto_dpif *ofproto)
1996 poll_timer_wait(1000);
2000 /* Returns true if STP should process 'flow'. */
2002 stp_should_process_flow(const struct flow *flow)
2004 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
2008 stp_process_packet(const struct ofport_dpif *ofport,
2009 const struct ofpbuf *packet)
2011 struct ofpbuf payload = *packet;
2012 struct eth_header *eth = payload.data;
2013 struct stp_port *sp = ofport->stp_port;
2015 /* Sink packets on ports that have STP disabled when the bridge has
2017 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
2021 /* Trim off padding on payload. */
2022 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
2023 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
2026 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
2027 stp_received_bpdu(sp, payload.data, payload.size);
2031 static struct priority_to_dscp *
2032 get_priority(const struct ofport_dpif *ofport, uint32_t priority)
2034 struct priority_to_dscp *pdscp;
2037 hash = hash_int(priority, 0);
2038 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) {
2039 if (pdscp->priority == priority) {
2047 ofport_clear_priorities(struct ofport_dpif *ofport)
2049 struct priority_to_dscp *pdscp, *next;
2051 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) {
2052 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
2058 set_queues(struct ofport *ofport_,
2059 const struct ofproto_port_queue *qdscp_list,
2062 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2063 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2064 struct hmap new = HMAP_INITIALIZER(&new);
2067 for (i = 0; i < n_qdscp; i++) {
2068 struct priority_to_dscp *pdscp;
2072 dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
2073 if (dpif_queue_to_priority(ofproto->backer->dpif, qdscp_list[i].queue,
2078 pdscp = get_priority(ofport, priority);
2080 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
2082 pdscp = xmalloc(sizeof *pdscp);
2083 pdscp->priority = priority;
2085 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2088 if (pdscp->dscp != dscp) {
2090 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2093 hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0));
2096 if (!hmap_is_empty(&ofport->priorities)) {
2097 ofport_clear_priorities(ofport);
2098 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2101 hmap_swap(&new, &ofport->priorities);
2109 /* Expires all MAC learning entries associated with 'bundle' and forces its
2110 * ofproto to revalidate every flow.
2112 * Normally MAC learning entries are removed only from the ofproto associated
2113 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2114 * are removed from every ofproto. When patch ports and SLB bonds are in use
2115 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2116 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2117 * with the host from which it migrated. */
2119 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2121 struct ofproto_dpif *ofproto = bundle->ofproto;
2122 struct mac_learning *ml = ofproto->ml;
2123 struct mac_entry *mac, *next_mac;
2125 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2126 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2127 if (mac->port.p == bundle) {
2129 struct ofproto_dpif *o;
2131 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2133 struct mac_entry *e;
2135 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan,
2138 mac_learning_expire(o->ml, e);
2144 mac_learning_expire(ml, mac);
2149 static struct ofbundle *
2150 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2152 struct ofbundle *bundle;
2154 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2155 &ofproto->bundles) {
2156 if (bundle->aux == aux) {
2163 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
2164 * ones that are found to 'bundles'. */
2166 bundle_lookup_multiple(struct ofproto_dpif *ofproto,
2167 void **auxes, size_t n_auxes,
2168 struct hmapx *bundles)
2172 hmapx_init(bundles);
2173 for (i = 0; i < n_auxes; i++) {
2174 struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
2176 hmapx_add(bundles, bundle);
2182 bundle_update(struct ofbundle *bundle)
2184 struct ofport_dpif *port;
2186 bundle->floodable = true;
2187 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2188 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2189 || !stp_forward_in_state(port->stp_state)) {
2190 bundle->floodable = false;
2197 bundle_del_port(struct ofport_dpif *port)
2199 struct ofbundle *bundle = port->bundle;
2201 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2203 list_remove(&port->bundle_node);
2204 port->bundle = NULL;
2207 lacp_slave_unregister(bundle->lacp, port);
2210 bond_slave_unregister(bundle->bond, port);
2213 bundle_update(bundle);
2217 bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port,
2218 struct lacp_slave_settings *lacp,
2219 uint32_t bond_stable_id)
2221 struct ofport_dpif *port;
2223 port = get_ofp_port(bundle->ofproto, ofp_port);
2228 if (port->bundle != bundle) {
2229 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2231 bundle_del_port(port);
2234 port->bundle = bundle;
2235 list_push_back(&bundle->ports, &port->bundle_node);
2236 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2237 || !stp_forward_in_state(port->stp_state)) {
2238 bundle->floodable = false;
2242 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2243 lacp_slave_register(bundle->lacp, port, lacp);
2246 port->bond_stable_id = bond_stable_id;
2252 bundle_destroy(struct ofbundle *bundle)
2254 struct ofproto_dpif *ofproto;
2255 struct ofport_dpif *port, *next_port;
2262 ofproto = bundle->ofproto;
2263 for (i = 0; i < MAX_MIRRORS; i++) {
2264 struct ofmirror *m = ofproto->mirrors[i];
2266 if (m->out == bundle) {
2268 } else if (hmapx_find_and_delete(&m->srcs, bundle)
2269 || hmapx_find_and_delete(&m->dsts, bundle)) {
2270 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2275 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2276 bundle_del_port(port);
2279 bundle_flush_macs(bundle, true);
2280 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2282 free(bundle->trunks);
2283 lacp_destroy(bundle->lacp);
2284 bond_destroy(bundle->bond);
2289 bundle_set(struct ofproto *ofproto_, void *aux,
2290 const struct ofproto_bundle_settings *s)
2292 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2293 bool need_flush = false;
2294 struct ofport_dpif *port;
2295 struct ofbundle *bundle;
2296 unsigned long *trunks;
2302 bundle_destroy(bundle_lookup(ofproto, aux));
2306 ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2307 ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2309 bundle = bundle_lookup(ofproto, aux);
2311 bundle = xmalloc(sizeof *bundle);
2313 bundle->ofproto = ofproto;
2314 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2315 hash_pointer(aux, 0));
2317 bundle->name = NULL;
2319 list_init(&bundle->ports);
2320 bundle->vlan_mode = PORT_VLAN_TRUNK;
2322 bundle->trunks = NULL;
2323 bundle->use_priority_tags = s->use_priority_tags;
2324 bundle->lacp = NULL;
2325 bundle->bond = NULL;
2327 bundle->floodable = true;
2329 bundle->src_mirrors = 0;
2330 bundle->dst_mirrors = 0;
2331 bundle->mirror_out = 0;
2334 if (!bundle->name || strcmp(s->name, bundle->name)) {
2336 bundle->name = xstrdup(s->name);
2341 if (!bundle->lacp) {
2342 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2343 bundle->lacp = lacp_create();
2345 lacp_configure(bundle->lacp, s->lacp);
2347 lacp_destroy(bundle->lacp);
2348 bundle->lacp = NULL;
2351 /* Update set of ports. */
2353 for (i = 0; i < s->n_slaves; i++) {
2354 if (!bundle_add_port(bundle, s->slaves[i],
2355 s->lacp ? &s->lacp_slaves[i] : NULL,
2356 s->bond_stable_ids ? s->bond_stable_ids[i] : 0)) {
2360 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2361 struct ofport_dpif *next_port;
2363 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2364 for (i = 0; i < s->n_slaves; i++) {
2365 if (s->slaves[i] == port->up.ofp_port) {
2370 bundle_del_port(port);
2374 ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2376 if (list_is_empty(&bundle->ports)) {
2377 bundle_destroy(bundle);
2381 /* Set VLAN tagging mode */
2382 if (s->vlan_mode != bundle->vlan_mode
2383 || s->use_priority_tags != bundle->use_priority_tags) {
2384 bundle->vlan_mode = s->vlan_mode;
2385 bundle->use_priority_tags = s->use_priority_tags;
2390 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2391 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2393 if (vlan != bundle->vlan) {
2394 bundle->vlan = vlan;
2398 /* Get trunked VLANs. */
2399 switch (s->vlan_mode) {
2400 case PORT_VLAN_ACCESS:
2404 case PORT_VLAN_TRUNK:
2405 trunks = CONST_CAST(unsigned long *, s->trunks);
2408 case PORT_VLAN_NATIVE_UNTAGGED:
2409 case PORT_VLAN_NATIVE_TAGGED:
2410 if (vlan != 0 && (!s->trunks
2411 || !bitmap_is_set(s->trunks, vlan)
2412 || bitmap_is_set(s->trunks, 0))) {
2413 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2415 trunks = bitmap_clone(s->trunks, 4096);
2417 trunks = bitmap_allocate1(4096);
2419 bitmap_set1(trunks, vlan);
2420 bitmap_set0(trunks, 0);
2422 trunks = CONST_CAST(unsigned long *, s->trunks);
2429 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2430 free(bundle->trunks);
2431 if (trunks == s->trunks) {
2432 bundle->trunks = vlan_bitmap_clone(trunks);
2434 bundle->trunks = trunks;
2439 if (trunks != s->trunks) {
2444 if (!list_is_short(&bundle->ports)) {
2445 bundle->ofproto->has_bonded_bundles = true;
2447 if (bond_reconfigure(bundle->bond, s->bond)) {
2448 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2451 bundle->bond = bond_create(s->bond);
2452 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2455 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2456 bond_slave_register(bundle->bond, port, port->bond_stable_id,
2460 bond_destroy(bundle->bond);
2461 bundle->bond = NULL;
2464 /* If we changed something that would affect MAC learning, un-learn
2465 * everything on this port and force flow revalidation. */
2467 bundle_flush_macs(bundle, false);
2474 bundle_remove(struct ofport *port_)
2476 struct ofport_dpif *port = ofport_dpif_cast(port_);
2477 struct ofbundle *bundle = port->bundle;
2480 bundle_del_port(port);
2481 if (list_is_empty(&bundle->ports)) {
2482 bundle_destroy(bundle);
2483 } else if (list_is_short(&bundle->ports)) {
2484 bond_destroy(bundle->bond);
2485 bundle->bond = NULL;
2491 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2493 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2494 struct ofport_dpif *port = port_;
2495 uint8_t ea[ETH_ADDR_LEN];
2498 error = netdev_get_etheraddr(port->up.netdev, ea);
2500 struct ofpbuf packet;
2503 ofpbuf_init(&packet, 0);
2504 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2506 memcpy(packet_pdu, pdu, pdu_size);
2508 send_packet(port, &packet);
2509 ofpbuf_uninit(&packet);
2511 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2512 "%s (%s)", port->bundle->name,
2513 netdev_get_name(port->up.netdev), strerror(error));
2518 bundle_send_learning_packets(struct ofbundle *bundle)
2520 struct ofproto_dpif *ofproto = bundle->ofproto;
2521 int error, n_packets, n_errors;
2522 struct mac_entry *e;
2524 error = n_packets = n_errors = 0;
2525 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2526 if (e->port.p != bundle) {
2527 struct ofpbuf *learning_packet;
2528 struct ofport_dpif *port;
2532 /* The assignment to "port" is unnecessary but makes "grep"ing for
2533 * struct ofport_dpif more effective. */
2534 learning_packet = bond_compose_learning_packet(bundle->bond,
2538 ret = send_packet(port, learning_packet);
2539 ofpbuf_delete(learning_packet);
2549 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2550 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2551 "packets, last error was: %s",
2552 bundle->name, n_errors, n_packets, strerror(error));
2554 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2555 bundle->name, n_packets);
2560 bundle_run(struct ofbundle *bundle)
2563 lacp_run(bundle->lacp, send_pdu_cb);
2566 struct ofport_dpif *port;
2568 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2569 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2572 bond_run(bundle->bond, &bundle->ofproto->backer->revalidate_set,
2573 lacp_status(bundle->lacp));
2574 if (bond_should_send_learning_packets(bundle->bond)) {
2575 bundle_send_learning_packets(bundle);
2581 bundle_wait(struct ofbundle *bundle)
2584 lacp_wait(bundle->lacp);
2587 bond_wait(bundle->bond);
2594 mirror_scan(struct ofproto_dpif *ofproto)
2598 for (idx = 0; idx < MAX_MIRRORS; idx++) {
2599 if (!ofproto->mirrors[idx]) {
2606 static struct ofmirror *
2607 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
2611 for (i = 0; i < MAX_MIRRORS; i++) {
2612 struct ofmirror *mirror = ofproto->mirrors[i];
2613 if (mirror && mirror->aux == aux) {
2621 /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */
2623 mirror_update_dups(struct ofproto_dpif *ofproto)
2627 for (i = 0; i < MAX_MIRRORS; i++) {
2628 struct ofmirror *m = ofproto->mirrors[i];
2631 m->dup_mirrors = MIRROR_MASK_C(1) << i;
2635 for (i = 0; i < MAX_MIRRORS; i++) {
2636 struct ofmirror *m1 = ofproto->mirrors[i];
2643 for (j = i + 1; j < MAX_MIRRORS; j++) {
2644 struct ofmirror *m2 = ofproto->mirrors[j];
2646 if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) {
2647 m1->dup_mirrors |= MIRROR_MASK_C(1) << j;
2648 m2->dup_mirrors |= m1->dup_mirrors;
2655 mirror_set(struct ofproto *ofproto_, void *aux,
2656 const struct ofproto_mirror_settings *s)
2658 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2659 mirror_mask_t mirror_bit;
2660 struct ofbundle *bundle;
2661 struct ofmirror *mirror;
2662 struct ofbundle *out;
2663 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
2664 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
2667 mirror = mirror_lookup(ofproto, aux);
2669 mirror_destroy(mirror);
2675 idx = mirror_scan(ofproto);
2677 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
2679 ofproto->up.name, MAX_MIRRORS, s->name);
2683 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
2684 mirror->ofproto = ofproto;
2687 mirror->out_vlan = -1;
2688 mirror->name = NULL;
2691 if (!mirror->name || strcmp(s->name, mirror->name)) {
2693 mirror->name = xstrdup(s->name);
2696 /* Get the new configuration. */
2697 if (s->out_bundle) {
2698 out = bundle_lookup(ofproto, s->out_bundle);
2700 mirror_destroy(mirror);
2706 out_vlan = s->out_vlan;
2708 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
2709 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
2711 /* If the configuration has not changed, do nothing. */
2712 if (hmapx_equals(&srcs, &mirror->srcs)
2713 && hmapx_equals(&dsts, &mirror->dsts)
2714 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
2715 && mirror->out == out
2716 && mirror->out_vlan == out_vlan)
2718 hmapx_destroy(&srcs);
2719 hmapx_destroy(&dsts);
2723 hmapx_swap(&srcs, &mirror->srcs);
2724 hmapx_destroy(&srcs);
2726 hmapx_swap(&dsts, &mirror->dsts);
2727 hmapx_destroy(&dsts);
2729 free(mirror->vlans);
2730 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
2733 mirror->out_vlan = out_vlan;
2735 /* Update bundles. */
2736 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2737 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
2738 if (hmapx_contains(&mirror->srcs, bundle)) {
2739 bundle->src_mirrors |= mirror_bit;
2741 bundle->src_mirrors &= ~mirror_bit;
2744 if (hmapx_contains(&mirror->dsts, bundle)) {
2745 bundle->dst_mirrors |= mirror_bit;
2747 bundle->dst_mirrors &= ~mirror_bit;
2750 if (mirror->out == bundle) {
2751 bundle->mirror_out |= mirror_bit;
2753 bundle->mirror_out &= ~mirror_bit;
2757 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2758 ofproto->has_mirrors = true;
2759 mac_learning_flush(ofproto->ml,
2760 &ofproto->backer->revalidate_set);
2761 mirror_update_dups(ofproto);
2767 mirror_destroy(struct ofmirror *mirror)
2769 struct ofproto_dpif *ofproto;
2770 mirror_mask_t mirror_bit;
2771 struct ofbundle *bundle;
2778 ofproto = mirror->ofproto;
2779 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2780 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
2782 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2783 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
2784 bundle->src_mirrors &= ~mirror_bit;
2785 bundle->dst_mirrors &= ~mirror_bit;
2786 bundle->mirror_out &= ~mirror_bit;
2789 hmapx_destroy(&mirror->srcs);
2790 hmapx_destroy(&mirror->dsts);
2791 free(mirror->vlans);
2793 ofproto->mirrors[mirror->idx] = NULL;
2797 mirror_update_dups(ofproto);
2799 ofproto->has_mirrors = false;
2800 for (i = 0; i < MAX_MIRRORS; i++) {
2801 if (ofproto->mirrors[i]) {
2802 ofproto->has_mirrors = true;
2809 mirror_get_stats(struct ofproto *ofproto_, void *aux,
2810 uint64_t *packets, uint64_t *bytes)
2812 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2813 struct ofmirror *mirror = mirror_lookup(ofproto, aux);
2816 *packets = *bytes = UINT64_MAX;
2820 *packets = mirror->packet_count;
2821 *bytes = mirror->byte_count;
2827 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2829 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2830 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2831 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
2837 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2839 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2840 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2841 return bundle && bundle->mirror_out != 0;
2845 forward_bpdu_changed(struct ofproto *ofproto_)
2847 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2848 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2852 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
2855 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2856 mac_learning_set_idle_time(ofproto->ml, idle_time);
2857 mac_learning_set_max_entries(ofproto->ml, max_entries);
2862 static struct ofport_dpif *
2863 get_ofp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
2865 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2866 return ofport ? ofport_dpif_cast(ofport) : NULL;
2869 static struct ofport_dpif *
2870 get_odp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
2872 struct ofport_dpif *port = odp_port_to_ofport(ofproto->backer, odp_port);
2873 return port && &ofproto->up == port->up.ofproto ? port : NULL;
2877 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
2878 struct ofproto_port *ofproto_port,
2879 struct dpif_port *dpif_port)
2881 ofproto_port->name = dpif_port->name;
2882 ofproto_port->type = dpif_port->type;
2883 ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
2886 static struct ofport_dpif *
2887 ofport_get_peer(const struct ofport_dpif *ofport_dpif)
2889 const struct ofproto_dpif *ofproto;
2892 peer = netdev_vport_patch_peer(ofport_dpif->up.netdev);
2897 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2898 struct ofport *ofport;
2900 ofport = shash_find_data(&ofproto->up.port_by_name, peer);
2901 if (ofport && ofport->ofproto->ofproto_class == &ofproto_dpif_class) {
2902 return ofport_dpif_cast(ofport);
2909 port_run_fast(struct ofport_dpif *ofport)
2911 if (ofport->cfm && cfm_should_send_ccm(ofport->cfm)) {
2912 struct ofpbuf packet;
2914 ofpbuf_init(&packet, 0);
2915 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
2916 send_packet(ofport, &packet);
2917 ofpbuf_uninit(&packet);
2922 port_run(struct ofport_dpif *ofport)
2924 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
2925 bool carrier_changed = carrier_seq != ofport->carrier_seq;
2926 bool enable = netdev_get_carrier(ofport->up.netdev);
2928 ofport->carrier_seq = carrier_seq;
2930 port_run_fast(ofport);
2932 if (ofport->tnl_port
2933 && tnl_port_reconfigure(&ofport->up, ofport->odp_port,
2934 &ofport->tnl_port)) {
2935 ofproto_dpif_cast(ofport->up.ofproto)->backer->need_revalidate = true;
2939 int cfm_opup = cfm_get_opup(ofport->cfm);
2941 cfm_run(ofport->cfm);
2942 enable = enable && !cfm_get_fault(ofport->cfm);
2944 if (cfm_opup >= 0) {
2945 enable = enable && cfm_opup;
2949 if (ofport->bundle) {
2950 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
2951 if (carrier_changed) {
2952 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
2956 if (ofport->may_enable != enable) {
2957 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2959 if (ofproto->has_bundle_action) {
2960 ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
2964 ofport->may_enable = enable;
2968 port_wait(struct ofport_dpif *ofport)
2971 cfm_wait(ofport->cfm);
2976 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
2977 struct ofproto_port *ofproto_port)
2979 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2980 struct dpif_port dpif_port;
2983 if (sset_contains(&ofproto->ghost_ports, devname)) {
2984 const char *type = netdev_get_type_from_name(devname);
2986 /* We may be called before ofproto->up.port_by_name is populated with
2987 * the appropriate ofport. For this reason, we must get the name and
2988 * type from the netdev layer directly. */
2990 const struct ofport *ofport;
2992 ofport = shash_find_data(&ofproto->up.port_by_name, devname);
2993 ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
2994 ofproto_port->name = xstrdup(devname);
2995 ofproto_port->type = xstrdup(type);
3001 if (!sset_contains(&ofproto->ports, devname)) {
3004 error = dpif_port_query_by_name(ofproto->backer->dpif,
3005 devname, &dpif_port);
3007 ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
3013 port_add(struct ofproto *ofproto_, struct netdev *netdev)
3015 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3016 const char *dp_port_name = netdev_vport_get_dpif_port(netdev);
3017 const char *devname = netdev_get_name(netdev);
3019 if (netdev_vport_is_patch(netdev)) {
3020 sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
3024 if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
3025 int error = dpif_port_add(ofproto->backer->dpif, netdev, NULL);
3031 if (netdev_get_tunnel_config(netdev)) {
3032 sset_add(&ofproto->ghost_ports, devname);
3033 sset_add(&ofproto->backer->tnl_backers, dp_port_name);
3035 sset_add(&ofproto->ports, devname);
3040 /* Returns true if the odp_port backing 'ofport' may be deleted from the
3041 * datapath. In most cases, this function simply returns true. However, for
3042 * tunnels it's possible that multiple ofports use the same odp_port, in which
3043 * case we need to keep the odp_port backer around until the last ofport is
3046 may_dpif_port_del(struct ofport_dpif *ofport)
3048 struct dpif_backer *backer = ofproto_dpif_cast(ofport->up.ofproto)->backer;
3049 struct ofproto_dpif *ofproto_iter;
3051 if (!ofport->tnl_port) {
3055 HMAP_FOR_EACH (ofproto_iter, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
3056 struct ofport_dpif *iter;
3058 if (backer != ofproto_iter->backer) {
3062 HMAP_FOR_EACH (iter, up.hmap_node, &ofproto_iter->up.ports) {
3063 if (ofport == iter) {
3067 if (!strcmp(netdev_vport_get_dpif_port(ofport->up.netdev),
3068 netdev_vport_get_dpif_port(iter->up.netdev))) {
3078 port_del(struct ofproto *ofproto_, uint16_t ofp_port)
3080 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3081 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
3088 sset_find_and_delete(&ofproto->ghost_ports,
3089 netdev_get_name(ofport->up.netdev));
3090 if (may_dpif_port_del(ofport)) {
3091 error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
3093 const char *dpif_port;
3095 /* The caller is going to close ofport->up.netdev. If this is a
3096 * bonded port, then the bond is using that netdev, so remove it
3097 * from the bond. The client will need to reconfigure everything
3098 * after deleting ports, so then the slave will get re-added. */
3099 dpif_port = netdev_vport_get_dpif_port(ofport->up.netdev);
3100 sset_find_and_delete(&ofproto->backer->tnl_backers, dpif_port);
3101 bundle_remove(&ofport->up);
3108 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
3110 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3113 error = netdev_get_stats(ofport->up.netdev, stats);
3115 if (!error && ofport_->ofp_port == OFPP_LOCAL) {
3116 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3118 /* ofproto->stats.tx_packets represents packets that we created
3119 * internally and sent to some port (e.g. packets sent with
3120 * send_packet()). Account for them as if they had come from
3121 * OFPP_LOCAL and got forwarded. */
3123 if (stats->rx_packets != UINT64_MAX) {
3124 stats->rx_packets += ofproto->stats.tx_packets;
3127 if (stats->rx_bytes != UINT64_MAX) {
3128 stats->rx_bytes += ofproto->stats.tx_bytes;
3131 /* ofproto->stats.rx_packets represents packets that were received on
3132 * some port and we processed internally and dropped (e.g. STP).
3133 * Account for them as if they had been forwarded to OFPP_LOCAL. */
3135 if (stats->tx_packets != UINT64_MAX) {
3136 stats->tx_packets += ofproto->stats.rx_packets;
3139 if (stats->tx_bytes != UINT64_MAX) {
3140 stats->tx_bytes += ofproto->stats.rx_bytes;
3147 /* Account packets for LOCAL port. */
3149 ofproto_update_local_port_stats(const struct ofproto *ofproto_,
3150 size_t tx_size, size_t rx_size)
3152 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3155 ofproto->stats.rx_packets++;
3156 ofproto->stats.rx_bytes += rx_size;
3159 ofproto->stats.tx_packets++;
3160 ofproto->stats.tx_bytes += tx_size;
3164 struct port_dump_state {
3169 struct ofproto_port port;
3174 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
3176 *statep = xzalloc(sizeof(struct port_dump_state));
3181 port_dump_next(const struct ofproto *ofproto_, void *state_,
3182 struct ofproto_port *port)
3184 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3185 struct port_dump_state *state = state_;
3186 const struct sset *sset;
3187 struct sset_node *node;
3189 if (state->has_port) {
3190 ofproto_port_destroy(&state->port);
3191 state->has_port = false;
3193 sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
3194 while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
3197 error = port_query_by_name(ofproto_, node->name, &state->port);
3199 *port = state->port;
3200 state->has_port = true;
3202 } else if (error != ENODEV) {
3207 if (!state->ghost) {
3208 state->ghost = true;
3211 return port_dump_next(ofproto_, state_, port);
3218 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3220 struct port_dump_state *state = state_;
3222 if (state->has_port) {
3223 ofproto_port_destroy(&state->port);
3230 port_poll(const struct ofproto *ofproto_, char **devnamep)
3232 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3234 if (ofproto->port_poll_errno) {
3235 int error = ofproto->port_poll_errno;
3236 ofproto->port_poll_errno = 0;
3240 if (sset_is_empty(&ofproto->port_poll_set)) {
3244 *devnamep = sset_pop(&ofproto->port_poll_set);
3249 port_poll_wait(const struct ofproto *ofproto_)
3251 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3252 dpif_port_poll_wait(ofproto->backer->dpif);
3256 port_is_lacp_current(const struct ofport *ofport_)
3258 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3259 return (ofport->bundle && ofport->bundle->lacp
3260 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3264 /* Upcall handling. */
3266 /* Flow miss batching.
3268 * Some dpifs implement operations faster when you hand them off in a batch.
3269 * To allow batching, "struct flow_miss" queues the dpif-related work needed
3270 * for a given flow. Each "struct flow_miss" corresponds to sending one or
3271 * more packets, plus possibly installing the flow in the dpif.
3273 * So far we only batch the operations that affect flow setup time the most.
3274 * It's possible to batch more than that, but the benefit might be minimal. */
3276 struct hmap_node hmap_node;
3277 struct ofproto_dpif *ofproto;
3279 enum odp_key_fitness key_fitness;
3280 const struct nlattr *key;
3282 ovs_be16 initial_tci;
3283 struct list packets;
3284 enum dpif_upcall_type upcall_type;
3285 uint32_t odp_in_port;
3288 struct flow_miss_op {
3289 struct dpif_op dpif_op;
3290 struct subfacet *subfacet; /* Subfacet */
3291 void *garbage; /* Pointer to pass to free(), NULL if none. */
3292 uint64_t stub[1024 / 8]; /* Temporary buffer. */
3295 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
3296 * OpenFlow controller as necessary according to their individual
3297 * configurations. */
3299 send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet,
3300 const struct flow *flow)
3302 struct ofputil_packet_in pin;
3304 pin.packet = packet->data;
3305 pin.packet_len = packet->size;
3306 pin.reason = OFPR_NO_MATCH;
3307 pin.controller_id = 0;
3312 pin.send_len = 0; /* not used for flow table misses */
3314 flow_get_metadata(flow, &pin.fmd);
3316 connmgr_send_packet_in(ofproto->up.connmgr, &pin);
3319 static enum slow_path_reason
3320 process_special(struct ofproto_dpif *ofproto, const struct flow *flow,
3321 const struct ofport_dpif *ofport, const struct ofpbuf *packet)
3325 } else if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) {
3327 cfm_process_heartbeat(ofport->cfm, packet);
3330 } else if (ofport->bundle && ofport->bundle->lacp
3331 && flow->dl_type == htons(ETH_TYPE_LACP)) {
3333 lacp_process_packet(ofport->bundle->lacp, ofport, packet);
3336 } else if (ofproto->stp && stp_should_process_flow(flow)) {
3338 stp_process_packet(ofport, packet);
3346 static struct flow_miss *
3347 flow_miss_find(struct hmap *todo, const struct flow *flow, uint32_t hash)
3349 struct flow_miss *miss;
3351 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
3352 if (flow_equal(&miss->flow, flow)) {
3360 /* Partially Initializes 'op' as an "execute" operation for 'miss' and
3361 * 'packet'. The caller must initialize op->actions and op->actions_len. If
3362 * 'miss' is associated with a subfacet the caller must also initialize the
3363 * returned op->subfacet, and if anything needs to be freed after processing
3364 * the op, the caller must initialize op->garbage also. */
3366 init_flow_miss_execute_op(struct flow_miss *miss, struct ofpbuf *packet,
3367 struct flow_miss_op *op)
3369 if (miss->flow.vlan_tci != miss->initial_tci) {
3370 /* This packet was received on a VLAN splinter port. We
3371 * added a VLAN to the packet to make the packet resemble
3372 * the flow, but the actions were composed assuming that
3373 * the packet contained no VLAN. So, we must remove the
3374 * VLAN header from the packet before trying to execute the
3376 eth_pop_vlan(packet);
3379 op->subfacet = NULL;
3381 op->dpif_op.type = DPIF_OP_EXECUTE;
3382 op->dpif_op.u.execute.key = miss->key;
3383 op->dpif_op.u.execute.key_len = miss->key_len;
3384 op->dpif_op.u.execute.packet = packet;
3387 /* Helper for handle_flow_miss_without_facet() and
3388 * handle_flow_miss_with_facet(). */
3390 handle_flow_miss_common(struct rule_dpif *rule,
3391 struct ofpbuf *packet, const struct flow *flow)
3393 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3395 ofproto->n_matches++;
3397 if (rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
3399 * Extra-special case for fail-open mode.
3401 * We are in fail-open mode and the packet matched the fail-open
3402 * rule, but we are connected to a controller too. We should send
3403 * the packet up to the controller in the hope that it will try to
3404 * set up a flow and thereby allow us to exit fail-open.
3406 * See the top-level comment in fail-open.c for more information.
3408 send_packet_in_miss(ofproto, packet, flow);
3412 /* Figures out whether a flow that missed in 'ofproto', whose details are in
3413 * 'miss', is likely to be worth tracking in detail in userspace and (usually)
3414 * installing a datapath flow. The answer is usually "yes" (a return value of
3415 * true). However, for short flows the cost of bookkeeping is much higher than
3416 * the benefits, so when the datapath holds a large number of flows we impose
3417 * some heuristics to decide which flows are likely to be worth tracking. */
3419 flow_miss_should_make_facet(struct ofproto_dpif *ofproto,
3420 struct flow_miss *miss, uint32_t hash)
3422 if (!ofproto->governor) {
3425 n_subfacets = hmap_count(&ofproto->subfacets);
3426 if (n_subfacets * 2 <= ofproto->up.flow_eviction_threshold) {
3430 ofproto->governor = governor_create(ofproto->up.name);
3433 return governor_should_install_flow(ofproto->governor, hash,
3434 list_size(&miss->packets));
3437 /* Handles 'miss', which matches 'rule', without creating a facet or subfacet
3438 * or creating any datapath flow. May add an "execute" operation to 'ops' and
3439 * increment '*n_ops'. */
3441 handle_flow_miss_without_facet(struct flow_miss *miss,
3442 struct rule_dpif *rule,
3443 struct flow_miss_op *ops, size_t *n_ops)
3445 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3446 long long int now = time_msec();
3447 struct action_xlate_ctx ctx;
3448 struct ofpbuf *packet;
3450 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3451 struct flow_miss_op *op = &ops[*n_ops];
3452 struct dpif_flow_stats stats;
3453 struct ofpbuf odp_actions;
3455 COVERAGE_INC(facet_suppress);
3457 ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub);
3459 dpif_flow_stats_extract(&miss->flow, packet, now, &stats);
3460 rule_credit_stats(rule, &stats);
3462 action_xlate_ctx_init(&ctx, ofproto, &miss->flow, miss->initial_tci,
3464 ctx.resubmit_stats = &stats;
3465 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len,
3468 if (odp_actions.size) {
3469 struct dpif_execute *execute = &op->dpif_op.u.execute;
3471 init_flow_miss_execute_op(miss, packet, op);
3472 execute->actions = odp_actions.data;
3473 execute->actions_len = odp_actions.size;
3474 op->garbage = ofpbuf_get_uninit_pointer(&odp_actions);
3478 ofpbuf_uninit(&odp_actions);
3483 /* Handles 'miss', which matches 'facet'. May add any required datapath
3484 * operations to 'ops', incrementing '*n_ops' for each new op.
3486 * All of the packets in 'miss' are considered to have arrived at time 'now'.
3487 * This is really important only for new facets: if we just called time_msec()
3488 * here, then the new subfacet or its packets could look (occasionally) as
3489 * though it was used some time after the facet was used. That can make a
3490 * one-packet flow look like it has a nonzero duration, which looks odd in
3491 * e.g. NetFlow statistics. */
3493 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
3495 struct flow_miss_op *ops, size_t *n_ops)
3497 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3498 enum subfacet_path want_path;
3499 struct subfacet *subfacet;
3500 struct ofpbuf *packet;
3502 subfacet = subfacet_create(facet, miss, now);
3504 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3505 struct flow_miss_op *op = &ops[*n_ops];
3506 struct dpif_flow_stats stats;
3507 struct ofpbuf odp_actions;
3509 handle_flow_miss_common(facet->rule, packet, &miss->flow);
3511 ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub);
3512 if (!subfacet->actions || subfacet->slow) {
3513 subfacet_make_actions(subfacet, packet, &odp_actions);
3516 dpif_flow_stats_extract(&facet->flow, packet, now, &stats);
3517 subfacet_update_stats(subfacet, &stats);
3519 if (subfacet->actions_len) {
3520 struct dpif_execute *execute = &op->dpif_op.u.execute;
3522 init_flow_miss_execute_op(miss, packet, op);
3523 op->subfacet = subfacet;
3524 if (!subfacet->slow) {
3525 execute->actions = subfacet->actions;
3526 execute->actions_len = subfacet->actions_len;
3527 ofpbuf_uninit(&odp_actions);
3529 execute->actions = odp_actions.data;
3530 execute->actions_len = odp_actions.size;
3531 op->garbage = ofpbuf_get_uninit_pointer(&odp_actions);
3536 ofpbuf_uninit(&odp_actions);
3540 want_path = subfacet_want_path(subfacet->slow);
3541 if (miss->upcall_type == DPIF_UC_MISS || subfacet->path != want_path) {
3542 struct flow_miss_op *op = &ops[(*n_ops)++];
3543 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
3545 op->subfacet = subfacet;
3547 op->dpif_op.type = DPIF_OP_FLOW_PUT;
3548 put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
3549 put->key = miss->key;
3550 put->key_len = miss->key_len;
3551 if (want_path == SF_FAST_PATH) {
3552 put->actions = subfacet->actions;
3553 put->actions_len = subfacet->actions_len;
3555 compose_slow_path(ofproto, &facet->flow, subfacet->slow,
3556 op->stub, sizeof op->stub,
3557 &put->actions, &put->actions_len);
3563 /* Handles flow miss 'miss'. May add any required datapath operations
3564 * to 'ops', incrementing '*n_ops' for each new op. */
3566 handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops,
3569 struct ofproto_dpif *ofproto = miss->ofproto;
3570 struct facet *facet;
3574 /* The caller must ensure that miss->hmap_node.hash contains
3575 * flow_hash(miss->flow, 0). */
3576 hash = miss->hmap_node.hash;
3578 facet = facet_lookup_valid(ofproto, &miss->flow, hash);
3580 struct rule_dpif *rule = rule_dpif_lookup(ofproto, &miss->flow);
3582 if (!flow_miss_should_make_facet(ofproto, miss, hash)) {
3583 handle_flow_miss_without_facet(miss, rule, ops, n_ops);
3587 facet = facet_create(rule, &miss->flow, hash);
3592 handle_flow_miss_with_facet(miss, facet, now, ops, n_ops);
3595 static struct drop_key *
3596 drop_key_lookup(const struct dpif_backer *backer, const struct nlattr *key,
3599 struct drop_key *drop_key;
3601 HMAP_FOR_EACH_WITH_HASH (drop_key, hmap_node, hash_bytes(key, key_len, 0),
3602 &backer->drop_keys) {
3603 if (drop_key->key_len == key_len
3604 && !memcmp(drop_key->key, key, key_len)) {
3612 drop_key_clear(struct dpif_backer *backer)
3614 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3615 struct drop_key *drop_key, *next;
3617 HMAP_FOR_EACH_SAFE (drop_key, next, hmap_node, &backer->drop_keys) {
3620 error = dpif_flow_del(backer->dpif, drop_key->key, drop_key->key_len,
3622 if (error && !VLOG_DROP_WARN(&rl)) {
3623 struct ds ds = DS_EMPTY_INITIALIZER;
3624 odp_flow_key_format(drop_key->key, drop_key->key_len, &ds);
3625 VLOG_WARN("Failed to delete drop key (%s) (%s)", strerror(error),
3630 hmap_remove(&backer->drop_keys, &drop_key->hmap_node);
3631 free(drop_key->key);
3636 /* Given a datpath, packet, and flow metadata ('backer', 'packet', and 'key'
3637 * respectively), populates 'flow' with the result of odp_flow_key_to_flow().
3638 * Optionally, if nonnull, populates 'fitnessp' with the fitness of 'flow' as
3639 * returned by odp_flow_key_to_flow(). Also, optionally populates 'ofproto'
3640 * with the ofproto_dpif, and 'odp_in_port' with the datapath in_port, that
3641 * 'packet' ingressed.
3643 * If 'ofproto' is nonnull, requires 'flow''s in_port to exist. Otherwise sets
3644 * 'flow''s in_port to OFPP_NONE.
3646 * This function does post-processing on data returned from
3647 * odp_flow_key_to_flow() to help make VLAN splinters transparent to the rest
3648 * of the upcall processing logic. In particular, if the extracted in_port is
3649 * a VLAN splinter port, it replaces flow->in_port by the "real" port, sets
3650 * flow->vlan_tci correctly for the VLAN of the VLAN splinter port, and pushes
3651 * a VLAN header onto 'packet' (if it is nonnull).
3653 * Optionally, if nonnull, sets '*initial_tci' to the VLAN TCI with which the
3654 * packet was really received, that is, the actual VLAN TCI extracted by
3655 * odp_flow_key_to_flow(). (This differs from the value returned in
3656 * flow->vlan_tci only for packets received on VLAN splinters.)
3658 * Similarly, this function also includes some logic to help with tunnels. It
3659 * may modify 'flow' as necessary to make the tunneling implementation
3660 * transparent to the upcall processing logic.
3662 * Returns 0 if successful, ENODEV if the parsed flow has no associated ofport,
3663 * or some other positive errno if there are other problems. */
3665 ofproto_receive(const struct dpif_backer *backer, struct ofpbuf *packet,
3666 const struct nlattr *key, size_t key_len,
3667 struct flow *flow, enum odp_key_fitness *fitnessp,
3668 struct ofproto_dpif **ofproto, uint32_t *odp_in_port,
3669 ovs_be16 *initial_tci)
3671 const struct ofport_dpif *port;
3672 enum odp_key_fitness fitness;
3675 fitness = odp_flow_key_to_flow(key, key_len, flow);
3676 if (fitness == ODP_FIT_ERROR) {
3682 *initial_tci = flow->vlan_tci;
3686 *odp_in_port = flow->in_port;
3689 if (tnl_port_should_receive(flow)) {
3690 const struct ofport *ofport = tnl_port_receive(flow);
3692 flow->in_port = OFPP_NONE;
3695 port = ofport_dpif_cast(ofport);
3697 /* We can't reproduce 'key' from 'flow'. */
3698 fitness = fitness == ODP_FIT_PERFECT ? ODP_FIT_TOO_MUCH : fitness;
3700 /* XXX: Since the tunnel module is not scoped per backer, it's
3701 * theoretically possible that we'll receive an ofport belonging to an
3702 * entirely different datapath. In practice, this can't happen because
3703 * no platforms has two separate datapaths which each support
3705 ovs_assert(ofproto_dpif_cast(port->up.ofproto)->backer == backer);
3707 port = odp_port_to_ofport(backer, flow->in_port);
3709 flow->in_port = OFPP_NONE;
3713 flow->in_port = port->up.ofp_port;
3714 if (vsp_adjust_flow(ofproto_dpif_cast(port->up.ofproto), flow)) {
3716 /* Make the packet resemble the flow, so that it gets sent to
3717 * an OpenFlow controller properly, so that it looks correct
3718 * for sFlow, and so that flow_extract() will get the correct
3719 * vlan_tci if it is called on 'packet'.
3721 * The allocated space inside 'packet' probably also contains
3722 * 'key', that is, both 'packet' and 'key' are probably part of
3723 * a struct dpif_upcall (see the large comment on that
3724 * structure definition), so pushing data on 'packet' is in
3725 * general not a good idea since it could overwrite 'key' or
3726 * free it as a side effect. However, it's OK in this special
3727 * case because we know that 'packet' is inside a Netlink
3728 * attribute: pushing 4 bytes will just overwrite the 4-byte
3729 * "struct nlattr", which is fine since we don't need that
3730 * header anymore. */
3731 eth_push_vlan(packet, flow->vlan_tci);
3733 /* We can't reproduce 'key' from 'flow'. */
3734 fitness = fitness == ODP_FIT_PERFECT ? ODP_FIT_TOO_MUCH : fitness;
3740 *ofproto = ofproto_dpif_cast(port->up.ofproto);
3745 *fitnessp = fitness;
3751 handle_miss_upcalls(struct dpif_backer *backer, struct dpif_upcall *upcalls,
3754 struct dpif_upcall *upcall;
3755 struct flow_miss *miss;
3756 struct flow_miss misses[FLOW_MISS_MAX_BATCH];
3757 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
3758 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
3768 /* Construct the to-do list.
3770 * This just amounts to extracting the flow from each packet and sticking
3771 * the packets that have the same flow in the same "flow_miss" structure so
3772 * that we can process them together. */
3775 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
3776 struct flow_miss *miss = &misses[n_misses];
3777 struct flow_miss *existing_miss;
3778 struct ofproto_dpif *ofproto;
3779 uint32_t odp_in_port;
3784 error = ofproto_receive(backer, upcall->packet, upcall->key,
3785 upcall->key_len, &flow, &miss->key_fitness,
3786 &ofproto, &odp_in_port, &miss->initial_tci);
3787 if (error == ENODEV) {
3788 struct drop_key *drop_key;
3790 /* Received packet on port for which we couldn't associate
3791 * an ofproto. This can happen if a port is removed while
3792 * traffic is being received. Print a rate-limited message
3793 * in case it happens frequently. Install a drop flow so
3794 * that future packets of the flow are inexpensively dropped
3796 VLOG_INFO_RL(&rl, "received packet on unassociated port %"PRIu32,
3799 drop_key = drop_key_lookup(backer, upcall->key, upcall->key_len);
3801 drop_key = xmalloc(sizeof *drop_key);
3802 drop_key->key = xmemdup(upcall->key, upcall->key_len);
3803 drop_key->key_len = upcall->key_len;
3805 hmap_insert(&backer->drop_keys, &drop_key->hmap_node,
3806 hash_bytes(drop_key->key, drop_key->key_len, 0));
3807 dpif_flow_put(backer->dpif, DPIF_FP_CREATE | DPIF_FP_MODIFY,
3808 drop_key->key, drop_key->key_len, NULL, 0, NULL);
3815 flow_extract(upcall->packet, flow.skb_priority, flow.skb_mark,
3816 &flow.tunnel, flow.in_port, &miss->flow);
3818 /* Add other packets to a to-do list. */
3819 hash = flow_hash(&miss->flow, 0);
3820 existing_miss = flow_miss_find(&todo, &miss->flow, hash);
3821 if (!existing_miss) {
3822 hmap_insert(&todo, &miss->hmap_node, hash);
3823 miss->ofproto = ofproto;
3824 miss->key = upcall->key;
3825 miss->key_len = upcall->key_len;
3826 miss->upcall_type = upcall->type;
3827 miss->odp_in_port = odp_in_port;
3828 list_init(&miss->packets);
3832 miss = existing_miss;
3834 list_push_back(&miss->packets, &upcall->packet->list_node);
3837 /* Process each element in the to-do list, constructing the set of
3838 * operations to batch. */
3840 HMAP_FOR_EACH (miss, hmap_node, &todo) {
3841 handle_flow_miss(miss, flow_miss_ops, &n_ops);
3843 ovs_assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
3845 /* Execute batch. */
3846 for (i = 0; i < n_ops; i++) {
3847 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
3849 dpif_operate(backer->dpif, dpif_ops, n_ops);
3851 /* Free memory and update facets. */
3852 for (i = 0; i < n_ops; i++) {
3853 struct flow_miss_op *op = &flow_miss_ops[i];
3855 switch (op->dpif_op.type) {
3856 case DPIF_OP_EXECUTE:
3859 case DPIF_OP_FLOW_PUT:
3860 if (!op->dpif_op.error) {
3861 op->subfacet->path = subfacet_want_path(op->subfacet->slow);
3865 case DPIF_OP_FLOW_DEL:
3871 hmap_destroy(&todo);
3874 static enum { SFLOW_UPCALL, MISS_UPCALL, BAD_UPCALL }
3875 classify_upcall(const struct dpif_upcall *upcall)
3877 union user_action_cookie cookie;
3879 /* First look at the upcall type. */
3880 switch (upcall->type) {
3881 case DPIF_UC_ACTION:
3887 case DPIF_N_UC_TYPES:
3889 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
3893 /* "action" upcalls need a closer look. */
3894 memcpy(&cookie, &upcall->userdata, sizeof(cookie));
3895 switch (cookie.type) {
3896 case USER_ACTION_COOKIE_SFLOW:
3897 return SFLOW_UPCALL;
3899 case USER_ACTION_COOKIE_SLOW_PATH:
3902 case USER_ACTION_COOKIE_UNSPEC:
3904 VLOG_WARN_RL(&rl, "invalid user cookie : 0x%"PRIx64, upcall->userdata);
3910 handle_sflow_upcall(struct dpif_backer *backer,
3911 const struct dpif_upcall *upcall)
3913 struct ofproto_dpif *ofproto;
3914 union user_action_cookie cookie;
3916 uint32_t odp_in_port;
3918 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3919 &flow, NULL, &ofproto, &odp_in_port, NULL)
3920 || !ofproto->sflow) {
3924 memcpy(&cookie, &upcall->userdata, sizeof(cookie));
3925 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
3926 odp_in_port, &cookie);
3930 handle_upcalls(struct dpif_backer *backer, unsigned int max_batch)
3932 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
3933 struct ofpbuf miss_bufs[FLOW_MISS_MAX_BATCH];
3934 uint64_t miss_buf_stubs[FLOW_MISS_MAX_BATCH][4096 / 8];
3939 ovs_assert(max_batch <= FLOW_MISS_MAX_BATCH);
3942 for (n_processed = 0; n_processed < max_batch; n_processed++) {
3943 struct dpif_upcall *upcall = &misses[n_misses];
3944 struct ofpbuf *buf = &miss_bufs[n_misses];
3947 ofpbuf_use_stub(buf, miss_buf_stubs[n_misses],
3948 sizeof miss_buf_stubs[n_misses]);
3949 error = dpif_recv(backer->dpif, upcall, buf);
3955 switch (classify_upcall(upcall)) {
3957 /* Handle it later. */
3962 handle_sflow_upcall(backer, upcall);
3972 /* Handle deferred MISS_UPCALL processing. */
3973 handle_miss_upcalls(backer, misses, n_misses);
3974 for (i = 0; i < n_misses; i++) {
3975 ofpbuf_uninit(&miss_bufs[i]);
3981 /* Flow expiration. */
3983 static int subfacet_max_idle(const struct ofproto_dpif *);
3984 static void update_stats(struct dpif_backer *);
3985 static void rule_expire(struct rule_dpif *);
3986 static void expire_subfacets(struct ofproto_dpif *, int dp_max_idle);
3988 /* This function is called periodically by run(). Its job is to collect
3989 * updates for the flows that have been installed into the datapath, most
3990 * importantly when they last were used, and then use that information to
3991 * expire flows that have not been used recently.
3993 * Returns the number of milliseconds after which it should be called again. */
3995 expire(struct dpif_backer *backer)
3997 struct ofproto_dpif *ofproto;
3998 int max_idle = INT32_MAX;
4000 /* Periodically clear out the drop keys in an effort to keep them
4001 * relatively few. */
4002 drop_key_clear(backer);
4004 /* Update stats for each flow in the backer. */
4005 update_stats(backer);
4007 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4008 struct rule *rule, *next_rule;
4011 if (ofproto->backer != backer) {
4015 /* Expire subfacets that have been idle too long. */
4016 dp_max_idle = subfacet_max_idle(ofproto);
4017 expire_subfacets(ofproto, dp_max_idle);
4019 max_idle = MIN(max_idle, dp_max_idle);
4021 /* Expire OpenFlow flows whose idle_timeout or hard_timeout
4023 LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
4024 &ofproto->up.expirable) {
4025 rule_expire(rule_dpif_cast(rule));
4028 /* All outstanding data in existing flows has been accounted, so it's a
4029 * good time to do bond rebalancing. */
4030 if (ofproto->has_bonded_bundles) {
4031 struct ofbundle *bundle;
4033 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
4035 bond_rebalance(bundle->bond, &backer->revalidate_set);
4041 return MIN(max_idle, 1000);
4044 /* Updates flow table statistics given that the datapath just reported 'stats'
4045 * as 'subfacet''s statistics. */
4047 update_subfacet_stats(struct subfacet *subfacet,
4048 const struct dpif_flow_stats *stats)
4050 struct facet *facet = subfacet->facet;
4052 if (stats->n_packets >= subfacet->dp_packet_count) {
4053 uint64_t extra = stats->n_packets - subfacet->dp_packet_count;
4054 facet->packet_count += extra;
4056 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
4059 if (stats->n_bytes >= subfacet->dp_byte_count) {
4060 facet->byte_count += stats->n_bytes - subfacet->dp_byte_count;
4062 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
4065 subfacet->dp_packet_count = stats->n_packets;
4066 subfacet->dp_byte_count = stats->n_bytes;
4068 facet->tcp_flags |= stats->tcp_flags;
4070 subfacet_update_time(subfacet, stats->used);
4071 if (facet->accounted_bytes < facet->byte_count) {
4073 facet_account(facet);
4074 facet->accounted_bytes = facet->byte_count;
4076 facet_push_stats(facet);
4079 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
4080 * about, or a flow that shouldn't be installed but was anyway. Delete it. */
4082 delete_unexpected_flow(struct ofproto_dpif *ofproto,
4083 const struct nlattr *key, size_t key_len)
4085 if (!VLOG_DROP_WARN(&rl)) {
4089 odp_flow_key_format(key, key_len, &s);
4090 VLOG_WARN("unexpected flow on %s: %s", ofproto->up.name, ds_cstr(&s));
4094 COVERAGE_INC(facet_unexpected);
4095 dpif_flow_del(ofproto->backer->dpif, key, key_len, NULL);
4098 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
4100 * This function also pushes statistics updates to rules which each facet
4101 * resubmits into. Generally these statistics will be accurate. However, if a
4102 * facet changes the rule it resubmits into at some time in between
4103 * update_stats() runs, it is possible that statistics accrued to the
4104 * old rule will be incorrectly attributed to the new rule. This could be
4105 * avoided by calling update_stats() whenever rules are created or
4106 * deleted. However, the performance impact of making so many calls to the
4107 * datapath do not justify the benefit of having perfectly accurate statistics.
4110 update_stats(struct dpif_backer *backer)
4112 const struct dpif_flow_stats *stats;
4113 struct dpif_flow_dump dump;
4114 const struct nlattr *key;
4117 dpif_flow_dump_start(&dump, backer->dpif);
4118 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
4120 struct subfacet *subfacet;
4121 struct ofproto_dpif *ofproto;
4122 struct ofport_dpif *ofport;
4125 if (ofproto_receive(backer, NULL, key, key_len, &flow, NULL, &ofproto,
4130 ofport = get_ofp_port(ofproto, flow.in_port);
4131 if (ofport && ofport->tnl_port) {
4132 netdev_vport_inc_rx(ofport->up.netdev, stats);
4135 key_hash = odp_flow_key_hash(key, key_len);
4136 subfacet = subfacet_find(ofproto, key, key_len, key_hash, &flow);
4137 switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
4139 update_subfacet_stats(subfacet, stats);
4143 /* Stats are updated per-packet. */
4146 case SF_NOT_INSTALLED:
4148 delete_unexpected_flow(ofproto, key, key_len);
4152 dpif_flow_dump_done(&dump);
4155 /* Calculates and returns the number of milliseconds of idle time after which
4156 * subfacets should expire from the datapath. When a subfacet expires, we fold
4157 * its statistics into its facet, and when a facet's last subfacet expires, we
4158 * fold its statistic into its rule. */
4160 subfacet_max_idle(const struct ofproto_dpif *ofproto)
4163 * Idle time histogram.
4165 * Most of the time a switch has a relatively small number of subfacets.
4166 * When this is the case we might as well keep statistics for all of them
4167 * in userspace and to cache them in the kernel datapath for performance as
4170 * As the number of subfacets increases, the memory required to maintain
4171 * statistics about them in userspace and in the kernel becomes
4172 * significant. However, with a large number of subfacets it is likely
4173 * that only a few of them are "heavy hitters" that consume a large amount
4174 * of bandwidth. At this point, only heavy hitters are worth caching in
4175 * the kernel and maintaining in userspaces; other subfacets we can
4178 * The technique used to compute the idle time is to build a histogram with
4179 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
4180 * that is installed in the kernel gets dropped in the appropriate bucket.
4181 * After the histogram has been built, we compute the cutoff so that only
4182 * the most-recently-used 1% of subfacets (but at least
4183 * ofproto->up.flow_eviction_threshold flows) are kept cached. At least
4184 * the most-recently-used bucket of subfacets is kept, so actually an
4185 * arbitrary number of subfacets can be kept in any given expiration run
4186 * (though the next run will delete most of those unless they receive
4189 * This requires a second pass through the subfacets, in addition to the
4190 * pass made by update_stats(), because the former function never looks at
4191 * uninstallable subfacets.
4193 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
4194 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
4195 int buckets[N_BUCKETS] = { 0 };
4196 int total, subtotal, bucket;
4197 struct subfacet *subfacet;
4201 total = hmap_count(&ofproto->subfacets);
4202 if (total <= ofproto->up.flow_eviction_threshold) {
4203 return N_BUCKETS * BUCKET_WIDTH;
4206 /* Build histogram. */
4208 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
4209 long long int idle = now - subfacet->used;
4210 int bucket = (idle <= 0 ? 0
4211 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
4212 : (unsigned int) idle / BUCKET_WIDTH);
4216 /* Find the first bucket whose flows should be expired. */
4217 subtotal = bucket = 0;
4219 subtotal += buckets[bucket++];
4220 } while (bucket < N_BUCKETS &&
4221 subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100));
4223 if (VLOG_IS_DBG_ENABLED()) {
4227 ds_put_cstr(&s, "keep");
4228 for (i = 0; i < N_BUCKETS; i++) {
4230 ds_put_cstr(&s, ", drop");
4233 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
4236 VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s));
4240 return bucket * BUCKET_WIDTH;
4244 expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle)
4246 /* Cutoff time for most flows. */
4247 long long int normal_cutoff = time_msec() - dp_max_idle;
4249 /* We really want to keep flows for special protocols around, so use a more
4250 * conservative cutoff. */
4251 long long int special_cutoff = time_msec() - 10000;
4253 struct subfacet *subfacet, *next_subfacet;
4254 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
4258 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
4259 &ofproto->subfacets) {
4260 long long int cutoff;
4262 cutoff = (subfacet->slow & (SLOW_CFM | SLOW_LACP | SLOW_STP)
4265 if (subfacet->used < cutoff) {
4266 if (subfacet->path != SF_NOT_INSTALLED) {
4267 batch[n_batch++] = subfacet;
4268 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
4269 subfacet_destroy_batch(ofproto, batch, n_batch);
4273 subfacet_destroy(subfacet);
4279 subfacet_destroy_batch(ofproto, batch, n_batch);
4283 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
4284 * then delete it entirely. */
4286 rule_expire(struct rule_dpif *rule)
4288 struct facet *facet, *next_facet;
4292 if (rule->up.pending) {
4293 /* We'll have to expire it later. */
4297 /* Has 'rule' expired? */
4299 if (rule->up.hard_timeout
4300 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
4301 reason = OFPRR_HARD_TIMEOUT;
4302 } else if (rule->up.idle_timeout
4303 && now > rule->up.used + rule->up.idle_timeout * 1000) {
4304 reason = OFPRR_IDLE_TIMEOUT;
4309 COVERAGE_INC(ofproto_dpif_expired);
4311 /* Update stats. (This is a no-op if the rule expired due to an idle
4312 * timeout, because that only happens when the rule has no facets left.) */
4313 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4314 facet_remove(facet);
4317 /* Get rid of the rule. */
4318 ofproto_rule_expire(&rule->up, reason);
4323 /* Creates and returns a new facet owned by 'rule', given a 'flow'.
4325 * The caller must already have determined that no facet with an identical
4326 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
4327 * the ofproto's classifier table.
4329 * 'hash' must be the return value of flow_hash(flow, 0).
4331 * The facet will initially have no subfacets. The caller should create (at
4332 * least) one subfacet with subfacet_create(). */
4333 static struct facet *
4334 facet_create(struct rule_dpif *rule, const struct flow *flow, uint32_t hash)
4336 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4337 struct facet *facet;
4339 facet = xzalloc(sizeof *facet);
4340 facet->used = time_msec();
4341 hmap_insert(&ofproto->facets, &facet->hmap_node, hash);
4342 list_push_back(&rule->facets, &facet->list_node);
4344 facet->flow = *flow;
4345 list_init(&facet->subfacets);
4346 netflow_flow_init(&facet->nf_flow);
4347 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
4353 facet_free(struct facet *facet)
4358 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
4359 * 'packet', which arrived on 'in_port'. */
4361 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
4362 const struct nlattr *odp_actions, size_t actions_len,
4363 struct ofpbuf *packet)
4365 struct odputil_keybuf keybuf;
4369 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4370 odp_flow_key_from_flow(&key, flow,
4371 ofp_port_to_odp_port(ofproto, flow->in_port));
4373 error = dpif_execute(ofproto->backer->dpif, key.data, key.size,
4374 odp_actions, actions_len, packet);
4378 /* Remove 'facet' from 'ofproto' and free up the associated memory:
4380 * - If 'facet' was installed in the datapath, uninstalls it and updates its
4381 * rule's statistics, via subfacet_uninstall().
4383 * - Removes 'facet' from its rule and from ofproto->facets.
4386 facet_remove(struct facet *facet)
4388 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4389 struct subfacet *subfacet, *next_subfacet;
4391 ovs_assert(!list_is_empty(&facet->subfacets));
4393 /* First uninstall all of the subfacets to get final statistics. */
4394 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4395 subfacet_uninstall(subfacet);
4398 /* Flush the final stats to the rule.
4400 * This might require us to have at least one subfacet around so that we
4401 * can use its actions for accounting in facet_account(), which is why we
4402 * have uninstalled but not yet destroyed the subfacets. */
4403 facet_flush_stats(facet);
4405 /* Now we're really all done so destroy everything. */
4406 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
4407 &facet->subfacets) {
4408 subfacet_destroy__(subfacet);
4410 hmap_remove(&ofproto->facets, &facet->hmap_node);
4411 list_remove(&facet->list_node);
4415 /* Feed information from 'facet' back into the learning table to keep it in
4416 * sync with what is actually flowing through the datapath. */
4418 facet_learn(struct facet *facet)
4420 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4421 struct action_xlate_ctx ctx;
4423 if (!facet->has_learn
4424 && !facet->has_normal
4425 && (!facet->has_fin_timeout
4426 || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
4430 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
4431 facet->flow.vlan_tci,
4432 facet->rule, facet->tcp_flags, NULL);
4433 ctx.may_learn = true;
4434 xlate_actions_for_side_effects(&ctx, facet->rule->up.ofpacts,
4435 facet->rule->up.ofpacts_len);
4439 facet_account(struct facet *facet)
4441 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4442 struct subfacet *subfacet;
4443 const struct nlattr *a;
4448 if (!facet->has_normal || !ofproto->has_bonded_bundles) {
4451 n_bytes = facet->byte_count - facet->accounted_bytes;
4453 /* This loop feeds byte counters to bond_account() for rebalancing to use
4454 * as a basis. We also need to track the actual VLAN on which the packet
4455 * is going to be sent to ensure that it matches the one passed to
4456 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
4459 * We use the actions from an arbitrary subfacet because they should all
4460 * be equally valid for our purpose. */
4461 subfacet = CONTAINER_OF(list_front(&facet->subfacets),
4462 struct subfacet, list_node);
4463 vlan_tci = facet->flow.vlan_tci;
4464 NL_ATTR_FOR_EACH_UNSAFE (a, left,
4465 subfacet->actions, subfacet->actions_len) {
4466 const struct ovs_action_push_vlan *vlan;
4467 struct ofport_dpif *port;
4469 switch (nl_attr_type(a)) {
4470 case OVS_ACTION_ATTR_OUTPUT:
4471 port = get_odp_port(ofproto, nl_attr_get_u32(a));
4472 if (port && port->bundle && port->bundle->bond) {
4473 bond_account(port->bundle->bond, &facet->flow,
4474 vlan_tci_to_vid(vlan_tci), n_bytes);
4478 case OVS_ACTION_ATTR_POP_VLAN:
4479 vlan_tci = htons(0);
4482 case OVS_ACTION_ATTR_PUSH_VLAN:
4483 vlan = nl_attr_get(a);
4484 vlan_tci = vlan->vlan_tci;
4490 /* Returns true if the only action for 'facet' is to send to the controller.
4491 * (We don't report NetFlow expiration messages for such facets because they
4492 * are just part of the control logic for the network, not real traffic). */
4494 facet_is_controller_flow(struct facet *facet)
4497 const struct rule *rule = &facet->rule->up;
4498 const struct ofpact *ofpacts = rule->ofpacts;
4499 size_t ofpacts_len = rule->ofpacts_len;
4501 if (ofpacts_len > 0 &&
4502 ofpacts->type == OFPACT_CONTROLLER &&
4503 ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len)) {
4510 /* Folds all of 'facet''s statistics into its rule. Also updates the
4511 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
4512 * 'facet''s statistics in the datapath should have been zeroed and folded into
4513 * its packet and byte counts before this function is called. */
4515 facet_flush_stats(struct facet *facet)
4517 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4518 struct subfacet *subfacet;
4520 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4521 ovs_assert(!subfacet->dp_byte_count);
4522 ovs_assert(!subfacet->dp_packet_count);
4525 facet_push_stats(facet);
4526 if (facet->accounted_bytes < facet->byte_count) {
4527 facet_account(facet);
4528 facet->accounted_bytes = facet->byte_count;
4531 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
4532 struct ofexpired expired;
4533 expired.flow = facet->flow;
4534 expired.packet_count = facet->packet_count;
4535 expired.byte_count = facet->byte_count;
4536 expired.used = facet->used;
4537 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4540 facet->rule->packet_count += facet->packet_count;
4541 facet->rule->byte_count += facet->byte_count;
4543 /* Reset counters to prevent double counting if 'facet' ever gets
4545 facet_reset_counters(facet);
4547 netflow_flow_clear(&facet->nf_flow);
4548 facet->tcp_flags = 0;
4551 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
4552 * Returns it if found, otherwise a null pointer.
4554 * 'hash' must be the return value of flow_hash(flow, 0).
4556 * The returned facet might need revalidation; use facet_lookup_valid()
4557 * instead if that is important. */
4558 static struct facet *
4559 facet_find(struct ofproto_dpif *ofproto,
4560 const struct flow *flow, uint32_t hash)
4562 struct facet *facet;
4564 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, hash, &ofproto->facets) {
4565 if (flow_equal(flow, &facet->flow)) {
4573 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
4574 * Returns it if found, otherwise a null pointer.
4576 * 'hash' must be the return value of flow_hash(flow, 0).
4578 * The returned facet is guaranteed to be valid. */
4579 static struct facet *
4580 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow,
4583 struct facet *facet;
4585 facet = facet_find(ofproto, flow, hash);
4587 && (ofproto->backer->need_revalidate
4588 || tag_set_intersects(&ofproto->backer->revalidate_set,
4590 facet_revalidate(facet);
4597 subfacet_path_to_string(enum subfacet_path path)
4600 case SF_NOT_INSTALLED:
4601 return "not installed";
4603 return "in fast path";
4605 return "in slow path";
4611 /* Returns the path in which a subfacet should be installed if its 'slow'
4612 * member has the specified value. */
4613 static enum subfacet_path
4614 subfacet_want_path(enum slow_path_reason slow)
4616 return slow ? SF_SLOW_PATH : SF_FAST_PATH;
4619 /* Returns true if 'subfacet' needs to have its datapath flow updated,
4620 * supposing that its actions have been recalculated as 'want_actions' and that
4621 * 'slow' is nonzero iff 'subfacet' should be in the slow path. */
4623 subfacet_should_install(struct subfacet *subfacet, enum slow_path_reason slow,
4624 const struct ofpbuf *want_actions)
4626 enum subfacet_path want_path = subfacet_want_path(slow);
4627 return (want_path != subfacet->path
4628 || (want_path == SF_FAST_PATH
4629 && (subfacet->actions_len != want_actions->size
4630 || memcmp(subfacet->actions, want_actions->data,
4631 subfacet->actions_len))));
4635 facet_check_consistency(struct facet *facet)
4637 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
4639 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4641 uint64_t odp_actions_stub[1024 / 8];
4642 struct ofpbuf odp_actions;
4644 struct rule_dpif *rule;
4645 struct subfacet *subfacet;
4646 bool may_log = false;
4649 /* Check the rule for consistency. */
4650 rule = rule_dpif_lookup(ofproto, &facet->flow);
4651 ok = rule == facet->rule;
4653 may_log = !VLOG_DROP_WARN(&rl);
4658 flow_format(&s, &facet->flow);
4659 ds_put_format(&s, ": facet associated with wrong rule (was "
4660 "table=%"PRIu8",", facet->rule->up.table_id);
4661 cls_rule_format(&facet->rule->up.cr, &s);
4662 ds_put_format(&s, ") (should have been table=%"PRIu8",",
4664 cls_rule_format(&rule->up.cr, &s);
4665 ds_put_char(&s, ')');
4667 VLOG_WARN("%s", ds_cstr(&s));
4672 /* Check the datapath actions for consistency. */
4673 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
4674 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4675 enum subfacet_path want_path;
4676 struct odputil_keybuf keybuf;
4677 struct action_xlate_ctx ctx;
4681 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
4682 subfacet->initial_tci, rule, 0, NULL);
4683 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len,
4686 if (subfacet->path == SF_NOT_INSTALLED) {
4687 /* This only happens if the datapath reported an error when we
4688 * tried to install the flow. Don't flag another error here. */
4692 want_path = subfacet_want_path(subfacet->slow);
4693 if (want_path == SF_SLOW_PATH && subfacet->path == SF_SLOW_PATH) {
4694 /* The actions for slow-path flows may legitimately vary from one
4695 * packet to the next. We're done. */
4699 if (!subfacet_should_install(subfacet, subfacet->slow, &odp_actions)) {
4703 /* Inconsistency! */
4705 may_log = !VLOG_DROP_WARN(&rl);
4709 /* Rate-limited, skip reporting. */
4714 subfacet_get_key(subfacet, &keybuf, &key);
4715 odp_flow_key_format(key.data, key.size, &s);
4717 ds_put_cstr(&s, ": inconsistency in subfacet");
4718 if (want_path != subfacet->path) {
4719 enum odp_key_fitness fitness = subfacet->key_fitness;
4721 ds_put_format(&s, " (%s, fitness=%s)",
4722 subfacet_path_to_string(subfacet->path),
4723 odp_key_fitness_to_string(fitness));
4724 ds_put_format(&s, " (should have been %s)",
4725 subfacet_path_to_string(want_path));
4726 } else if (want_path == SF_FAST_PATH) {
4727 ds_put_cstr(&s, " (actions were: ");
4728 format_odp_actions(&s, subfacet->actions,
4729 subfacet->actions_len);
4730 ds_put_cstr(&s, ") (correct actions: ");
4731 format_odp_actions(&s, odp_actions.data, odp_actions.size);
4732 ds_put_char(&s, ')');
4734 ds_put_cstr(&s, " (actions: ");
4735 format_odp_actions(&s, subfacet->actions,
4736 subfacet->actions_len);
4737 ds_put_char(&s, ')');
4739 VLOG_WARN("%s", ds_cstr(&s));
4742 ofpbuf_uninit(&odp_actions);
4747 /* Re-searches the classifier for 'facet':
4749 * - If the rule found is different from 'facet''s current rule, moves
4750 * 'facet' to the new rule and recompiles its actions.
4752 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
4753 * where it is and recompiles its actions anyway. */
4755 facet_revalidate(struct facet *facet)
4757 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4759 struct nlattr *odp_actions;
4762 struct actions *new_actions;
4764 struct action_xlate_ctx ctx;
4765 uint64_t odp_actions_stub[1024 / 8];
4766 struct ofpbuf odp_actions;
4768 struct rule_dpif *new_rule;
4769 struct subfacet *subfacet;
4772 COVERAGE_INC(facet_revalidate);
4774 new_rule = rule_dpif_lookup(ofproto, &facet->flow);
4776 /* Calculate new datapath actions.
4778 * We do not modify any 'facet' state yet, because we might need to, e.g.,
4779 * emit a NetFlow expiration and, if so, we need to have the old state
4780 * around to properly compose it. */
4782 /* If the datapath actions changed or the installability changed,
4783 * then we need to talk to the datapath. */
4786 memset(&ctx, 0, sizeof ctx);
4787 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
4788 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4789 enum slow_path_reason slow;
4791 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
4792 subfacet->initial_tci, new_rule, 0, NULL);
4793 xlate_actions(&ctx, new_rule->up.ofpacts, new_rule->up.ofpacts_len,
4796 slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
4797 if (subfacet_should_install(subfacet, slow, &odp_actions)) {
4798 struct dpif_flow_stats stats;
4800 subfacet_install(subfacet,
4801 odp_actions.data, odp_actions.size, &stats, slow);
4802 subfacet_update_stats(subfacet, &stats);
4805 new_actions = xcalloc(list_size(&facet->subfacets),
4806 sizeof *new_actions);
4808 new_actions[i].odp_actions = xmemdup(odp_actions.data,
4810 new_actions[i].actions_len = odp_actions.size;
4815 ofpbuf_uninit(&odp_actions);
4818 facet_flush_stats(facet);
4821 /* Update 'facet' now that we've taken care of all the old state. */
4822 facet->tags = ctx.tags;
4823 facet->nf_flow.output_iface = ctx.nf_output_iface;
4824 facet->has_learn = ctx.has_learn;
4825 facet->has_normal = ctx.has_normal;
4826 facet->has_fin_timeout = ctx.has_fin_timeout;
4827 facet->mirrors = ctx.mirrors;
4830 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4831 subfacet->slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
4833 if (new_actions && new_actions[i].odp_actions) {
4834 free(subfacet->actions);
4835 subfacet->actions = new_actions[i].odp_actions;
4836 subfacet->actions_len = new_actions[i].actions_len;
4842 if (facet->rule != new_rule) {
4843 COVERAGE_INC(facet_changed_rule);
4844 list_remove(&facet->list_node);
4845 list_push_back(&new_rule->facets, &facet->list_node);
4846 facet->rule = new_rule;
4847 facet->used = new_rule->up.created;
4848 facet->prev_used = facet->used;
4852 /* Updates 'facet''s used time. Caller is responsible for calling
4853 * facet_push_stats() to update the flows which 'facet' resubmits into. */
4855 facet_update_time(struct facet *facet, long long int used)
4857 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4858 if (used > facet->used) {
4860 ofproto_rule_update_used(&facet->rule->up, used);
4861 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
4866 facet_reset_counters(struct facet *facet)
4868 facet->packet_count = 0;
4869 facet->byte_count = 0;
4870 facet->prev_packet_count = 0;
4871 facet->prev_byte_count = 0;
4872 facet->accounted_bytes = 0;
4876 facet_push_stats(struct facet *facet)
4878 struct dpif_flow_stats stats;
4880 ovs_assert(facet->packet_count >= facet->prev_packet_count);
4881 ovs_assert(facet->byte_count >= facet->prev_byte_count);
4882 ovs_assert(facet->used >= facet->prev_used);
4884 stats.n_packets = facet->packet_count - facet->prev_packet_count;
4885 stats.n_bytes = facet->byte_count - facet->prev_byte_count;
4886 stats.used = facet->used;
4887 stats.tcp_flags = 0;
4889 if (stats.n_packets || stats.n_bytes || facet->used > facet->prev_used) {
4890 facet->prev_packet_count = facet->packet_count;
4891 facet->prev_byte_count = facet->byte_count;
4892 facet->prev_used = facet->used;
4894 flow_push_stats(facet->rule, &facet->flow, &stats);
4896 update_mirror_stats(ofproto_dpif_cast(facet->rule->up.ofproto),
4897 facet->mirrors, stats.n_packets, stats.n_bytes);
4902 rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats)
4904 rule->packet_count += stats->n_packets;
4905 rule->byte_count += stats->n_bytes;
4906 ofproto_rule_update_used(&rule->up, stats->used);
4909 /* Pushes flow statistics to the rules which 'flow' resubmits into given
4910 * 'rule''s actions and mirrors. */
4912 flow_push_stats(struct rule_dpif *rule,
4913 const struct flow *flow, const struct dpif_flow_stats *stats)
4915 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4916 struct action_xlate_ctx ctx;
4918 ofproto_rule_update_used(&rule->up, stats->used);
4920 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci, rule,
4922 ctx.resubmit_stats = stats;
4923 xlate_actions_for_side_effects(&ctx, rule->up.ofpacts,
4924 rule->up.ofpacts_len);
4929 static struct subfacet *
4930 subfacet_find(struct ofproto_dpif *ofproto,
4931 const struct nlattr *key, size_t key_len, uint32_t key_hash,
4932 const struct flow *flow)
4934 struct subfacet *subfacet;
4936 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
4937 &ofproto->subfacets) {
4939 ? (subfacet->key_len == key_len
4940 && !memcmp(key, subfacet->key, key_len))
4941 : flow_equal(flow, &subfacet->facet->flow)) {
4949 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
4950 * 'key_fitness', 'key', and 'key_len' members in 'miss'. Returns the
4951 * existing subfacet if there is one, otherwise creates and returns a
4954 * If the returned subfacet is new, then subfacet->actions will be NULL, in
4955 * which case the caller must populate the actions with
4956 * subfacet_make_actions(). */
4957 static struct subfacet *
4958 subfacet_create(struct facet *facet, struct flow_miss *miss,
4961 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4962 enum odp_key_fitness key_fitness = miss->key_fitness;
4963 const struct nlattr *key = miss->key;
4964 size_t key_len = miss->key_len;
4966 struct subfacet *subfacet;
4968 key_hash = odp_flow_key_hash(key, key_len);
4970 if (list_is_empty(&facet->subfacets)) {
4971 subfacet = &facet->one_subfacet;
4973 subfacet = subfacet_find(ofproto, key, key_len, key_hash,
4976 if (subfacet->facet == facet) {
4980 /* This shouldn't happen. */
4981 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
4982 subfacet_destroy(subfacet);
4985 subfacet = xmalloc(sizeof *subfacet);
4988 hmap_insert(&ofproto->subfacets, &subfacet->hmap_node, key_hash);
4989 list_push_back(&facet->subfacets, &subfacet->list_node);
4990 subfacet->facet = facet;
4991 subfacet->key_fitness = key_fitness;
4992 if (key_fitness != ODP_FIT_PERFECT) {
4993 subfacet->key = xmemdup(key, key_len);
4994 subfacet->key_len = key_len;
4996 subfacet->key = NULL;
4997 subfacet->key_len = 0;
4999 subfacet->used = now;
5000 subfacet->dp_packet_count = 0;
5001 subfacet->dp_byte_count = 0;
5002 subfacet->actions_len = 0;
5003 subfacet->actions = NULL;
5004 subfacet->slow = (subfacet->key_fitness == ODP_FIT_TOO_LITTLE
5007 subfacet->path = SF_NOT_INSTALLED;
5008 subfacet->initial_tci = miss->initial_tci;
5009 subfacet->odp_in_port = miss->odp_in_port;
5014 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
5015 * its facet within 'ofproto', and frees it. */
5017 subfacet_destroy__(struct subfacet *subfacet)
5019 struct facet *facet = subfacet->facet;
5020 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5022 subfacet_uninstall(subfacet);
5023 hmap_remove(&ofproto->subfacets, &subfacet->hmap_node);
5024 list_remove(&subfacet->list_node);
5025 free(subfacet->key);
5026 free(subfacet->actions);
5027 if (subfacet != &facet->one_subfacet) {
5032 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
5033 * last remaining subfacet in its facet destroys the facet too. */
5035 subfacet_destroy(struct subfacet *subfacet)
5037 struct facet *facet = subfacet->facet;
5039 if (list_is_singleton(&facet->subfacets)) {
5040 /* facet_remove() needs at least one subfacet (it will remove it). */
5041 facet_remove(facet);
5043 subfacet_destroy__(subfacet);
5048 subfacet_destroy_batch(struct ofproto_dpif *ofproto,
5049 struct subfacet **subfacets, int n)
5051 struct odputil_keybuf keybufs[SUBFACET_DESTROY_MAX_BATCH];
5052 struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH];
5053 struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH];
5054 struct ofpbuf keys[SUBFACET_DESTROY_MAX_BATCH];
5055 struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH];
5058 for (i = 0; i < n; i++) {
5059 ops[i].type = DPIF_OP_FLOW_DEL;
5060 subfacet_get_key(subfacets[i], &keybufs[i], &keys[i]);
5061 ops[i].u.flow_del.key = keys[i].data;
5062 ops[i].u.flow_del.key_len = keys[i].size;
5063 ops[i].u.flow_del.stats = &stats[i];
5067 dpif_operate(ofproto->backer->dpif, opsp, n);
5068 for (i = 0; i < n; i++) {
5069 subfacet_reset_dp_stats(subfacets[i], &stats[i]);
5070 subfacets[i]->path = SF_NOT_INSTALLED;
5071 subfacet_destroy(subfacets[i]);
5075 /* Initializes 'key' with the sequence of OVS_KEY_ATTR_* Netlink attributes
5076 * that can be used to refer to 'subfacet'. The caller must provide 'keybuf'
5077 * for use as temporary storage. */
5079 subfacet_get_key(struct subfacet *subfacet, struct odputil_keybuf *keybuf,
5083 if (!subfacet->key) {
5084 struct flow *flow = &subfacet->facet->flow;
5086 ofpbuf_use_stack(key, keybuf, sizeof *keybuf);
5087 odp_flow_key_from_flow(key, flow, subfacet->odp_in_port);
5089 ofpbuf_use_const(key, subfacet->key, subfacet->key_len);
5093 /* Composes the datapath actions for 'subfacet' based on its rule's actions.
5094 * Translates the actions into 'odp_actions', which the caller must have
5095 * initialized and is responsible for uninitializing. */
5097 subfacet_make_actions(struct subfacet *subfacet, const struct ofpbuf *packet,
5098 struct ofpbuf *odp_actions)
5100 struct facet *facet = subfacet->facet;
5101 struct rule_dpif *rule = facet->rule;
5102 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5104 struct action_xlate_ctx ctx;
5106 action_xlate_ctx_init(&ctx, ofproto, &facet->flow, subfacet->initial_tci,
5108 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, odp_actions);
5109 facet->tags = ctx.tags;
5110 facet->has_learn = ctx.has_learn;
5111 facet->has_normal = ctx.has_normal;
5112 facet->has_fin_timeout = ctx.has_fin_timeout;
5113 facet->nf_flow.output_iface = ctx.nf_output_iface;
5114 facet->mirrors = ctx.mirrors;
5116 subfacet->slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
5117 if (subfacet->actions_len != odp_actions->size
5118 || memcmp(subfacet->actions, odp_actions->data, odp_actions->size)) {
5119 free(subfacet->actions);
5120 subfacet->actions_len = odp_actions->size;
5121 subfacet->actions = xmemdup(odp_actions->data, odp_actions->size);
5125 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
5126 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
5127 * in the datapath will be zeroed and 'stats' will be updated with traffic new
5128 * since 'subfacet' was last updated.
5130 * Returns 0 if successful, otherwise a positive errno value. */
5132 subfacet_install(struct subfacet *subfacet,
5133 const struct nlattr *actions, size_t actions_len,
5134 struct dpif_flow_stats *stats,
5135 enum slow_path_reason slow)
5137 struct facet *facet = subfacet->facet;
5138 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5139 enum subfacet_path path = subfacet_want_path(slow);
5140 uint64_t slow_path_stub[128 / 8];
5141 struct odputil_keybuf keybuf;
5142 enum dpif_flow_put_flags flags;
5146 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
5148 flags |= DPIF_FP_ZERO_STATS;
5151 if (path == SF_SLOW_PATH) {
5152 compose_slow_path(ofproto, &facet->flow, slow,
5153 slow_path_stub, sizeof slow_path_stub,
5154 &actions, &actions_len);
5157 subfacet_get_key(subfacet, &keybuf, &key);
5158 ret = dpif_flow_put(ofproto->backer->dpif, flags, key.data, key.size,
5159 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 odputil_keybuf keybuf;
5186 struct dpif_flow_stats stats;
5190 subfacet_get_key(subfacet, &keybuf, &key);
5191 error = dpif_flow_del(ofproto->backer->dpif,
5192 key.data, key.size, &stats);
5193 subfacet_reset_dp_stats(subfacet, &stats);
5195 subfacet_update_stats(subfacet, &stats);
5197 subfacet->path = SF_NOT_INSTALLED;
5199 ovs_assert(subfacet->dp_packet_count == 0);
5200 ovs_assert(subfacet->dp_byte_count == 0);
5204 /* Resets 'subfacet''s datapath statistics counters. This should be called
5205 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
5206 * non-null, it should contain the statistics returned by dpif when 'subfacet'
5207 * was reset in the datapath. 'stats' will be modified to include only
5208 * statistics new since 'subfacet' was last updated. */
5210 subfacet_reset_dp_stats(struct subfacet *subfacet,
5211 struct dpif_flow_stats *stats)
5214 && subfacet->dp_packet_count <= stats->n_packets
5215 && subfacet->dp_byte_count <= stats->n_bytes) {
5216 stats->n_packets -= subfacet->dp_packet_count;
5217 stats->n_bytes -= subfacet->dp_byte_count;
5220 subfacet->dp_packet_count = 0;
5221 subfacet->dp_byte_count = 0;
5224 /* Updates 'subfacet''s used time. The caller is responsible for calling
5225 * facet_push_stats() to update the flows which 'subfacet' resubmits into. */
5227 subfacet_update_time(struct subfacet *subfacet, long long int used)
5229 if (used > subfacet->used) {
5230 subfacet->used = used;
5231 facet_update_time(subfacet->facet, used);
5235 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
5237 * Because of the meaning of a subfacet's counters, it only makes sense to do
5238 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
5239 * represents a packet that was sent by hand or if it represents statistics
5240 * that have been cleared out of the datapath. */
5242 subfacet_update_stats(struct subfacet *subfacet,
5243 const struct dpif_flow_stats *stats)
5245 if (stats->n_packets || stats->used > subfacet->used) {
5246 struct facet *facet = subfacet->facet;
5248 subfacet_update_time(subfacet, stats->used);
5249 facet->packet_count += stats->n_packets;
5250 facet->byte_count += stats->n_bytes;
5251 facet->tcp_flags |= stats->tcp_flags;
5252 facet_push_stats(facet);
5253 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
5259 static struct rule_dpif *
5260 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow)
5262 struct rule_dpif *rule;
5264 rule = rule_dpif_lookup__(ofproto, flow, 0);
5269 return rule_dpif_miss_rule(ofproto, flow);
5272 static struct rule_dpif *
5273 rule_dpif_lookup__(struct ofproto_dpif *ofproto, const struct flow *flow,
5276 struct cls_rule *cls_rule;
5277 struct classifier *cls;
5279 if (table_id >= N_TABLES) {
5283 cls = &ofproto->up.tables[table_id].cls;
5284 if (flow->nw_frag & FLOW_NW_FRAG_ANY
5285 && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
5286 /* For OFPC_NORMAL frag_handling, we must pretend that transport ports
5287 * are unavailable. */
5288 struct flow ofpc_normal_flow = *flow;
5289 ofpc_normal_flow.tp_src = htons(0);
5290 ofpc_normal_flow.tp_dst = htons(0);
5291 cls_rule = classifier_lookup(cls, &ofpc_normal_flow);
5293 cls_rule = classifier_lookup(cls, flow);
5295 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
5298 static struct rule_dpif *
5299 rule_dpif_miss_rule(struct ofproto_dpif *ofproto, const struct flow *flow)
5301 struct ofport_dpif *port;
5303 port = get_ofp_port(ofproto, flow->in_port);
5305 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, flow->in_port);
5306 return ofproto->miss_rule;
5309 if (port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN) {
5310 return ofproto->no_packet_in_rule;
5312 return ofproto->miss_rule;
5316 complete_operation(struct rule_dpif *rule)
5318 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5320 rule_invalidate(rule);
5322 struct dpif_completion *c = xmalloc(sizeof *c);
5323 c->op = rule->up.pending;
5324 list_push_back(&ofproto->completions, &c->list_node);
5326 ofoperation_complete(rule->up.pending, 0);
5330 static struct rule *
5333 struct rule_dpif *rule = xmalloc(sizeof *rule);
5338 rule_dealloc(struct rule *rule_)
5340 struct rule_dpif *rule = rule_dpif_cast(rule_);
5345 rule_construct(struct rule *rule_)
5347 struct rule_dpif *rule = rule_dpif_cast(rule_);
5348 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5349 struct rule_dpif *victim;
5352 rule->packet_count = 0;
5353 rule->byte_count = 0;
5355 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
5356 if (victim && !list_is_empty(&victim->facets)) {
5357 struct facet *facet;
5359 rule->facets = victim->facets;
5360 list_moved(&rule->facets);
5361 LIST_FOR_EACH (facet, list_node, &rule->facets) {
5362 /* XXX: We're only clearing our local counters here. It's possible
5363 * that quite a few packets are unaccounted for in the datapath
5364 * statistics. These will be accounted to the new rule instead of
5365 * cleared as required. This could be fixed by clearing out the
5366 * datapath statistics for this facet, but currently it doesn't
5368 facet_reset_counters(facet);
5372 /* Must avoid list_moved() in this case. */
5373 list_init(&rule->facets);
5376 table_id = rule->up.table_id;
5378 rule->tag = victim->tag;
5379 } else if (table_id == 0) {
5384 miniflow_expand(&rule->up.cr.match.flow, &flow);
5385 rule->tag = rule_calculate_tag(&flow, &rule->up.cr.match.mask,
5386 ofproto->tables[table_id].basis);
5389 complete_operation(rule);
5394 rule_destruct(struct rule *rule_)
5396 struct rule_dpif *rule = rule_dpif_cast(rule_);
5397 struct facet *facet, *next_facet;
5399 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
5400 facet_revalidate(facet);
5403 complete_operation(rule);
5407 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
5409 struct rule_dpif *rule = rule_dpif_cast(rule_);
5410 struct facet *facet;
5412 /* Start from historical data for 'rule' itself that are no longer tracked
5413 * in facets. This counts, for example, facets that have expired. */
5414 *packets = rule->packet_count;
5415 *bytes = rule->byte_count;
5417 /* Add any statistics that are tracked by facets. This includes
5418 * statistical data recently updated by ofproto_update_stats() as well as
5419 * stats for packets that were executed "by hand" via dpif_execute(). */
5420 LIST_FOR_EACH (facet, list_node, &rule->facets) {
5421 *packets += facet->packet_count;
5422 *bytes += facet->byte_count;
5427 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
5428 struct ofpbuf *packet)
5430 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5432 struct dpif_flow_stats stats;
5434 struct action_xlate_ctx ctx;
5435 uint64_t odp_actions_stub[1024 / 8];
5436 struct ofpbuf odp_actions;
5438 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
5439 rule_credit_stats(rule, &stats);
5441 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5442 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci,
5443 rule, stats.tcp_flags, packet);
5444 ctx.resubmit_stats = &stats;
5445 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, &odp_actions);
5447 execute_odp_actions(ofproto, flow, odp_actions.data,
5448 odp_actions.size, packet);
5450 ofpbuf_uninit(&odp_actions);
5454 rule_execute(struct rule *rule, const struct flow *flow,
5455 struct ofpbuf *packet)
5457 rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
5458 ofpbuf_delete(packet);
5463 rule_modify_actions(struct rule *rule_)
5465 struct rule_dpif *rule = rule_dpif_cast(rule_);
5467 complete_operation(rule);
5470 /* Sends 'packet' out 'ofport'.
5471 * May modify 'packet'.
5472 * Returns 0 if successful, otherwise a positive errno value. */
5474 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
5476 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
5477 uint64_t odp_actions_stub[1024 / 8];
5478 struct ofpbuf key, odp_actions;
5479 struct odputil_keybuf keybuf;
5484 flow_extract(packet, 0, 0, NULL, OFPP_LOCAL, &flow);
5485 if (netdev_vport_is_patch(ofport->up.netdev)) {
5486 struct ofproto_dpif *peer_ofproto;
5487 struct dpif_flow_stats stats;
5488 struct ofport_dpif *peer;
5489 struct rule_dpif *rule;
5491 peer = ofport_get_peer(ofport);
5496 dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
5497 netdev_vport_inc_tx(ofport->up.netdev, &stats);
5498 netdev_vport_inc_rx(peer->up.netdev, &stats);
5500 flow.in_port = peer->up.ofp_port;
5501 peer_ofproto = ofproto_dpif_cast(peer->up.ofproto);
5502 rule = rule_dpif_lookup(peer_ofproto, &flow);
5503 rule_dpif_execute(rule, &flow, packet);
5508 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5510 if (ofport->tnl_port) {
5511 struct dpif_flow_stats stats;
5513 odp_port = tnl_port_send(ofport->tnl_port, &flow);
5514 if (odp_port == OVSP_NONE) {
5518 dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
5519 netdev_vport_inc_tx(ofport->up.netdev, &stats);
5520 odp_put_tunnel_action(&flow.tunnel, &odp_actions);
5521 odp_put_skb_mark_action(flow.skb_mark, &odp_actions);
5523 odp_port = vsp_realdev_to_vlandev(ofproto, ofport->odp_port,
5525 if (odp_port != ofport->odp_port) {
5526 eth_pop_vlan(packet);
5527 flow.vlan_tci = htons(0);
5531 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5532 odp_flow_key_from_flow(&key, &flow,
5533 ofp_port_to_odp_port(ofproto, flow.in_port));
5535 compose_sflow_action(ofproto, &odp_actions, &flow, odp_port);
5537 nl_msg_put_u32(&odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port);
5538 error = dpif_execute(ofproto->backer->dpif,
5540 odp_actions.data, odp_actions.size,
5542 ofpbuf_uninit(&odp_actions);
5545 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)",
5546 ofproto->up.name, odp_port, strerror(error));
5548 ofproto_update_local_port_stats(ofport->up.ofproto, packet->size, 0);
5552 /* OpenFlow to datapath action translation. */
5554 static bool may_receive(const struct ofport_dpif *, struct action_xlate_ctx *);
5555 static void do_xlate_actions(const struct ofpact *, size_t ofpacts_len,
5556 struct action_xlate_ctx *);
5557 static void xlate_normal(struct action_xlate_ctx *);
5559 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
5560 * The action will state 'slow' as the reason that the action is in the slow
5561 * path. (This is purely informational: it allows a human viewing "ovs-dpctl
5562 * dump-flows" output to see why a flow is in the slow path.)
5564 * The 'stub_size' bytes in 'stub' will be used to store the action.
5565 * 'stub_size' must be large enough for the action.
5567 * The action and its size will be stored in '*actionsp' and '*actions_lenp',
5570 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
5571 enum slow_path_reason slow,
5572 uint64_t *stub, size_t stub_size,
5573 const struct nlattr **actionsp, size_t *actions_lenp)
5575 union user_action_cookie cookie;
5578 cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
5579 cookie.slow_path.unused = 0;
5580 cookie.slow_path.reason = slow;
5582 ofpbuf_use_stack(&buf, stub, stub_size);
5583 if (slow & (SLOW_CFM | SLOW_LACP | SLOW_STP)) {
5584 uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif, UINT32_MAX);
5585 odp_put_userspace_action(pid, &cookie, &buf);
5587 put_userspace_action(ofproto, &buf, flow, &cookie);
5589 *actionsp = buf.data;
5590 *actions_lenp = buf.size;
5594 put_userspace_action(const struct ofproto_dpif *ofproto,
5595 struct ofpbuf *odp_actions,
5596 const struct flow *flow,
5597 const union user_action_cookie *cookie)
5601 pid = dpif_port_get_pid(ofproto->backer->dpif,
5602 ofp_port_to_odp_port(ofproto, flow->in_port));
5604 return odp_put_userspace_action(pid, cookie, odp_actions);
5608 compose_sflow_cookie(const struct ofproto_dpif *ofproto,
5609 ovs_be16 vlan_tci, uint32_t odp_port,
5610 unsigned int n_outputs, union user_action_cookie *cookie)
5614 cookie->type = USER_ACTION_COOKIE_SFLOW;
5615 cookie->sflow.vlan_tci = vlan_tci;
5617 /* See http://www.sflow.org/sflow_version_5.txt (search for "Input/output
5618 * port information") for the interpretation of cookie->output. */
5619 switch (n_outputs) {
5621 /* 0x40000000 | 256 means "packet dropped for unknown reason". */
5622 cookie->sflow.output = 0x40000000 | 256;
5626 ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port);
5628 cookie->sflow.output = ifindex;
5633 /* 0x80000000 means "multiple output ports. */
5634 cookie->sflow.output = 0x80000000 | n_outputs;
5639 /* Compose SAMPLE action for sFlow. */
5641 compose_sflow_action(const struct ofproto_dpif *ofproto,
5642 struct ofpbuf *odp_actions,
5643 const struct flow *flow,
5646 uint32_t probability;
5647 union user_action_cookie cookie;
5648 size_t sample_offset, actions_offset;
5651 if (!ofproto->sflow || flow->in_port == OFPP_NONE) {
5655 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
5657 /* Number of packets out of UINT_MAX to sample. */
5658 probability = dpif_sflow_get_probability(ofproto->sflow);
5659 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
5661 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
5662 compose_sflow_cookie(ofproto, htons(0), odp_port,
5663 odp_port == OVSP_NONE ? 0 : 1, &cookie);
5664 cookie_offset = put_userspace_action(ofproto, odp_actions, flow, &cookie);
5666 nl_msg_end_nested(odp_actions, actions_offset);
5667 nl_msg_end_nested(odp_actions, sample_offset);
5668 return cookie_offset;
5671 /* SAMPLE action must be first action in any given list of actions.
5672 * At this point we do not have all information required to build it. So try to
5673 * build sample action as complete as possible. */
5675 add_sflow_action(struct action_xlate_ctx *ctx)
5677 ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto,
5679 &ctx->flow, OVSP_NONE);
5680 ctx->sflow_odp_port = 0;
5681 ctx->sflow_n_outputs = 0;
5684 /* Fix SAMPLE action according to data collected while composing ODP actions.
5685 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
5686 * USERSPACE action's user-cookie which is required for sflow. */
5688 fix_sflow_action(struct action_xlate_ctx *ctx)
5690 const struct flow *base = &ctx->base_flow;
5691 union user_action_cookie *cookie;
5693 if (!ctx->user_cookie_offset) {
5697 cookie = ofpbuf_at(ctx->odp_actions, ctx->user_cookie_offset,
5699 ovs_assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
5701 compose_sflow_cookie(ctx->ofproto, base->vlan_tci,
5702 ctx->sflow_odp_port, ctx->sflow_n_outputs, cookie);
5706 compose_output_action__(struct action_xlate_ctx *ctx, uint16_t ofp_port,
5709 const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
5710 ovs_be16 flow_vlan_tci = ctx->flow.vlan_tci;
5711 ovs_be64 flow_tun_id = ctx->flow.tunnel.tun_id;
5712 uint8_t flow_nw_tos = ctx->flow.nw_tos;
5713 struct priority_to_dscp *pdscp;
5714 uint32_t out_port, odp_port;
5716 /* If 'struct flow' gets additional metadata, we'll need to zero it out
5717 * before traversing a patch port. */
5718 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 18);
5721 xlate_report(ctx, "Nonexistent output port");
5723 } else if (ofport->up.pp.config & OFPUTIL_PC_NO_FWD) {
5724 xlate_report(ctx, "OFPPC_NO_FWD set, skipping output");
5726 } else if (check_stp && !stp_forward_in_state(ofport->stp_state)) {
5727 xlate_report(ctx, "STP not in forwarding state, skipping output");
5731 if (netdev_vport_is_patch(ofport->up.netdev)) {
5732 struct ofport_dpif *peer = ofport_get_peer(ofport);
5733 struct flow old_flow = ctx->flow;
5734 const struct ofproto_dpif *peer_ofproto;
5735 enum slow_path_reason special;
5736 struct ofport_dpif *in_port;
5739 xlate_report(ctx, "Nonexistent patch port peer");
5743 peer_ofproto = ofproto_dpif_cast(peer->up.ofproto);
5744 if (peer_ofproto->backer != ctx->ofproto->backer) {
5745 xlate_report(ctx, "Patch port peer on a different datapath");
5749 ctx->ofproto = ofproto_dpif_cast(peer->up.ofproto);
5750 ctx->flow.in_port = peer->up.ofp_port;
5751 ctx->flow.metadata = htonll(0);
5752 memset(&ctx->flow.tunnel, 0, sizeof ctx->flow.tunnel);
5753 memset(ctx->flow.regs, 0, sizeof ctx->flow.regs);
5755 in_port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
5756 special = process_special(ctx->ofproto, &ctx->flow, in_port,
5759 ctx->slow |= special;
5760 } else if (!in_port || may_receive(in_port, ctx)) {
5761 if (!in_port || stp_forward_in_state(in_port->stp_state)) {
5762 xlate_table_action(ctx, ctx->flow.in_port, 0, true);
5764 /* Forwarding is disabled by STP. Let OFPP_NORMAL and the
5765 * learning action look at the packet, then drop it. */
5766 struct flow old_base_flow = ctx->base_flow;
5767 size_t old_size = ctx->odp_actions->size;
5768 xlate_table_action(ctx, ctx->flow.in_port, 0, true);
5769 ctx->base_flow = old_base_flow;
5770 ctx->odp_actions->size = old_size;
5774 ctx->flow = old_flow;
5775 ctx->ofproto = ofproto_dpif_cast(ofport->up.ofproto);
5777 if (ctx->resubmit_stats) {
5778 netdev_vport_inc_tx(ofport->up.netdev, ctx->resubmit_stats);
5779 netdev_vport_inc_rx(peer->up.netdev, ctx->resubmit_stats);
5785 pdscp = get_priority(ofport, ctx->flow.skb_priority);
5787 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
5788 ctx->flow.nw_tos |= pdscp->dscp;
5791 odp_port = ofp_port_to_odp_port(ctx->ofproto, ofp_port);
5792 if (ofport->tnl_port) {
5793 odp_port = tnl_port_send(ofport->tnl_port, &ctx->flow);
5794 if (odp_port == OVSP_NONE) {
5795 xlate_report(ctx, "Tunneling decided against output");
5799 if (ctx->resubmit_stats) {
5800 netdev_vport_inc_tx(ofport->up.netdev, ctx->resubmit_stats);
5802 out_port = odp_port;
5803 commit_odp_tunnel_action(&ctx->flow, &ctx->base_flow,
5806 out_port = vsp_realdev_to_vlandev(ctx->ofproto, odp_port,
5807 ctx->flow.vlan_tci);
5808 if (out_port != odp_port) {
5809 ctx->flow.vlan_tci = htons(0);
5812 commit_odp_actions(&ctx->flow, &ctx->base_flow, ctx->odp_actions);
5813 nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_OUTPUT, out_port);
5815 ctx->sflow_odp_port = odp_port;
5816 ctx->sflow_n_outputs++;
5817 ctx->nf_output_iface = ofp_port;
5818 ctx->flow.tunnel.tun_id = flow_tun_id;
5819 ctx->flow.vlan_tci = flow_vlan_tci;
5820 ctx->flow.nw_tos = flow_nw_tos;
5824 compose_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port)
5826 compose_output_action__(ctx, ofp_port, true);
5830 xlate_table_action(struct action_xlate_ctx *ctx,
5831 uint16_t in_port, uint8_t table_id, bool may_packet_in)
5833 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
5834 struct ofproto_dpif *ofproto = ctx->ofproto;
5835 struct rule_dpif *rule;
5836 uint16_t old_in_port;
5837 uint8_t old_table_id;
5839 old_table_id = ctx->table_id;
5840 ctx->table_id = table_id;
5842 /* Look up a flow with 'in_port' as the input port. */
5843 old_in_port = ctx->flow.in_port;
5844 ctx->flow.in_port = in_port;
5845 rule = rule_dpif_lookup__(ofproto, &ctx->flow, table_id);
5848 if (table_id > 0 && table_id < N_TABLES) {
5849 struct table_dpif *table = &ofproto->tables[table_id];
5850 if (table->other_table) {
5851 ctx->tags |= (rule && rule->tag
5853 : rule_calculate_tag(&ctx->flow,
5854 &table->other_table->mask,
5859 /* Restore the original input port. Otherwise OFPP_NORMAL and
5860 * OFPP_IN_PORT will have surprising behavior. */
5861 ctx->flow.in_port = old_in_port;
5863 if (ctx->resubmit_hook) {
5864 ctx->resubmit_hook(ctx, rule);
5867 if (rule == NULL && may_packet_in) {
5869 * check if table configuration flags
5870 * OFPTC_TABLE_MISS_CONTROLLER, default.
5871 * OFPTC_TABLE_MISS_CONTINUE,
5872 * OFPTC_TABLE_MISS_DROP
5873 * When OF1.0, OFPTC_TABLE_MISS_CONTINUE is used. What to do?
5875 rule = rule_dpif_miss_rule(ofproto, &ctx->flow);
5879 struct rule_dpif *old_rule = ctx->rule;
5881 if (ctx->resubmit_stats) {
5882 rule_credit_stats(rule, ctx->resubmit_stats);
5887 do_xlate_actions(rule->up.ofpacts, rule->up.ofpacts_len, ctx);
5888 ctx->rule = old_rule;
5892 ctx->table_id = old_table_id;
5894 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
5896 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
5897 MAX_RESUBMIT_RECURSION);
5898 ctx->max_resubmit_trigger = true;
5903 xlate_ofpact_resubmit(struct action_xlate_ctx *ctx,
5904 const struct ofpact_resubmit *resubmit)
5909 in_port = resubmit->in_port;
5910 if (in_port == OFPP_IN_PORT) {
5911 in_port = ctx->flow.in_port;
5914 table_id = resubmit->table_id;
5915 if (table_id == 255) {
5916 table_id = ctx->table_id;
5919 xlate_table_action(ctx, in_port, table_id, false);
5923 flood_packets(struct action_xlate_ctx *ctx, bool all)
5925 struct ofport_dpif *ofport;
5927 HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) {
5928 uint16_t ofp_port = ofport->up.ofp_port;
5930 if (ofp_port == ctx->flow.in_port) {
5935 compose_output_action__(ctx, ofp_port, false);
5936 } else if (!(ofport->up.pp.config & OFPUTIL_PC_NO_FLOOD)) {
5937 compose_output_action(ctx, ofp_port);
5941 ctx->nf_output_iface = NF_OUT_FLOOD;
5945 execute_controller_action(struct action_xlate_ctx *ctx, int len,
5946 enum ofp_packet_in_reason reason,
5947 uint16_t controller_id)
5949 struct ofputil_packet_in pin;
5950 struct ofpbuf *packet;
5952 ctx->slow |= SLOW_CONTROLLER;
5957 packet = ofpbuf_clone(ctx->packet);
5959 if (packet->l2 && packet->l3) {
5960 struct eth_header *eh;
5962 eth_pop_vlan(packet);
5965 /* If the Ethernet type is less than ETH_TYPE_MIN, it's likely an 802.2
5966 * LLC frame. Calculating the Ethernet type of these frames is more
5967 * trouble than seems appropriate for a simple assertion. */
5968 ovs_assert(ntohs(eh->eth_type) < ETH_TYPE_MIN
5969 || eh->eth_type == ctx->flow.dl_type);
5971 memcpy(eh->eth_src, ctx->flow.dl_src, sizeof eh->eth_src);
5972 memcpy(eh->eth_dst, ctx->flow.dl_dst, sizeof eh->eth_dst);
5974 if (ctx->flow.vlan_tci & htons(VLAN_CFI)) {
5975 eth_push_vlan(packet, ctx->flow.vlan_tci);
5979 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
5980 packet_set_ipv4(packet, ctx->flow.nw_src, ctx->flow.nw_dst,
5981 ctx->flow.nw_tos, ctx->flow.nw_ttl);
5985 if (ctx->flow.nw_proto == IPPROTO_TCP) {
5986 packet_set_tcp_port(packet, ctx->flow.tp_src,
5988 } else if (ctx->flow.nw_proto == IPPROTO_UDP) {
5989 packet_set_udp_port(packet, ctx->flow.tp_src,
5996 pin.packet = packet->data;
5997 pin.packet_len = packet->size;
5998 pin.reason = reason;
5999 pin.controller_id = controller_id;
6000 pin.table_id = ctx->table_id;
6001 pin.cookie = ctx->rule ? ctx->rule->up.flow_cookie : 0;
6004 flow_get_metadata(&ctx->flow, &pin.fmd);
6006 connmgr_send_packet_in(ctx->ofproto->up.connmgr, &pin);
6007 ofpbuf_delete(packet);
6011 compose_dec_ttl(struct action_xlate_ctx *ctx, struct ofpact_cnt_ids *ids)
6013 if (ctx->flow.dl_type != htons(ETH_TYPE_IP) &&
6014 ctx->flow.dl_type != htons(ETH_TYPE_IPV6)) {
6018 if (ctx->flow.nw_ttl > 1) {
6024 for (i = 0; i < ids->n_controllers; i++) {
6025 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL,
6029 /* Stop processing for current table. */
6035 xlate_output_action(struct action_xlate_ctx *ctx,
6036 uint16_t port, uint16_t max_len, bool may_packet_in)
6038 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
6040 ctx->nf_output_iface = NF_OUT_DROP;
6044 compose_output_action(ctx, ctx->flow.in_port);
6047 xlate_table_action(ctx, ctx->flow.in_port, 0, may_packet_in);
6053 flood_packets(ctx, false);
6056 flood_packets(ctx, true);
6058 case OFPP_CONTROLLER:
6059 execute_controller_action(ctx, max_len, OFPR_ACTION, 0);
6065 if (port != ctx->flow.in_port) {
6066 compose_output_action(ctx, port);
6068 xlate_report(ctx, "skipping output to input port");
6073 if (prev_nf_output_iface == NF_OUT_FLOOD) {
6074 ctx->nf_output_iface = NF_OUT_FLOOD;
6075 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
6076 ctx->nf_output_iface = prev_nf_output_iface;
6077 } else if (prev_nf_output_iface != NF_OUT_DROP &&
6078 ctx->nf_output_iface != NF_OUT_FLOOD) {
6079 ctx->nf_output_iface = NF_OUT_MULTI;
6084 xlate_output_reg_action(struct action_xlate_ctx *ctx,
6085 const struct ofpact_output_reg *or)
6087 uint64_t port = mf_get_subfield(&or->src, &ctx->flow);
6088 if (port <= UINT16_MAX) {
6089 xlate_output_action(ctx, port, or->max_len, false);
6094 xlate_enqueue_action(struct action_xlate_ctx *ctx,
6095 const struct ofpact_enqueue *enqueue)
6097 uint16_t ofp_port = enqueue->port;
6098 uint32_t queue_id = enqueue->queue;
6099 uint32_t flow_priority, priority;
6102 /* Translate queue to priority. */
6103 error = dpif_queue_to_priority(ctx->ofproto->backer->dpif,
6104 queue_id, &priority);
6106 /* Fall back to ordinary output action. */
6107 xlate_output_action(ctx, enqueue->port, 0, false);
6111 /* Check output port. */
6112 if (ofp_port == OFPP_IN_PORT) {
6113 ofp_port = ctx->flow.in_port;
6114 } else if (ofp_port == ctx->flow.in_port) {
6118 /* Add datapath actions. */
6119 flow_priority = ctx->flow.skb_priority;
6120 ctx->flow.skb_priority = priority;
6121 compose_output_action(ctx, ofp_port);
6122 ctx->flow.skb_priority = flow_priority;
6124 /* Update NetFlow output port. */
6125 if (ctx->nf_output_iface == NF_OUT_DROP) {
6126 ctx->nf_output_iface = ofp_port;
6127 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
6128 ctx->nf_output_iface = NF_OUT_MULTI;
6133 xlate_set_queue_action(struct action_xlate_ctx *ctx, uint32_t queue_id)
6135 uint32_t skb_priority;
6137 if (!dpif_queue_to_priority(ctx->ofproto->backer->dpif,
6138 queue_id, &skb_priority)) {
6139 ctx->flow.skb_priority = skb_priority;
6141 /* Couldn't translate queue to a priority. Nothing to do. A warning
6142 * has already been logged. */
6146 struct xlate_reg_state {
6152 xlate_autopath(struct action_xlate_ctx *ctx,
6153 const struct ofpact_autopath *ap)
6155 uint16_t ofp_port = ap->port;
6156 struct ofport_dpif *port = get_ofp_port(ctx->ofproto, ofp_port);
6158 if (!port || !port->bundle) {
6159 ofp_port = OFPP_NONE;
6160 } else if (port->bundle->bond) {
6161 /* Autopath does not support VLAN hashing. */
6162 struct ofport_dpif *slave = bond_choose_output_slave(
6163 port->bundle->bond, &ctx->flow, 0, &ctx->tags);
6165 ofp_port = slave->up.ofp_port;
6168 nxm_reg_load(&ap->dst, ofp_port, &ctx->flow);
6172 slave_enabled_cb(uint16_t ofp_port, void *ofproto_)
6174 struct ofproto_dpif *ofproto = ofproto_;
6175 struct ofport_dpif *port;
6185 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
6188 port = get_ofp_port(ofproto, ofp_port);
6189 return port ? port->may_enable : false;
6194 xlate_bundle_action(struct action_xlate_ctx *ctx,
6195 const struct ofpact_bundle *bundle)
6199 port = bundle_execute(bundle, &ctx->flow, slave_enabled_cb, ctx->ofproto);
6200 if (bundle->dst.field) {
6201 nxm_reg_load(&bundle->dst, port, &ctx->flow);
6203 xlate_output_action(ctx, port, 0, false);
6208 xlate_learn_action(struct action_xlate_ctx *ctx,
6209 const struct ofpact_learn *learn)
6211 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
6212 struct ofputil_flow_mod fm;
6213 uint64_t ofpacts_stub[1024 / 8];
6214 struct ofpbuf ofpacts;
6217 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
6218 learn_execute(learn, &ctx->flow, &fm, &ofpacts);
6220 error = ofproto_flow_mod(&ctx->ofproto->up, &fm);
6221 if (error && !VLOG_DROP_WARN(&rl)) {
6222 VLOG_WARN("learning action failed to modify flow table (%s)",
6223 ofperr_get_name(error));
6226 ofpbuf_uninit(&ofpacts);
6229 /* Reduces '*timeout' to no more than 'max'. A value of zero in either case
6230 * means "infinite". */
6232 reduce_timeout(uint16_t max, uint16_t *timeout)
6234 if (max && (!*timeout || *timeout > max)) {
6240 xlate_fin_timeout(struct action_xlate_ctx *ctx,
6241 const struct ofpact_fin_timeout *oft)
6243 if (ctx->tcp_flags & (TCP_FIN | TCP_RST) && ctx->rule) {
6244 struct rule_dpif *rule = ctx->rule;
6246 reduce_timeout(oft->fin_idle_timeout, &rule->up.idle_timeout);
6247 reduce_timeout(oft->fin_hard_timeout, &rule->up.hard_timeout);
6252 may_receive(const struct ofport_dpif *port, struct action_xlate_ctx *ctx)
6254 if (port->up.pp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
6255 ? OFPUTIL_PC_NO_RECV_STP
6256 : OFPUTIL_PC_NO_RECV)) {
6260 /* Only drop packets here if both forwarding and learning are
6261 * disabled. If just learning is enabled, we need to have
6262 * OFPP_NORMAL and the learning action have a look at the packet
6263 * before we can drop it. */
6264 if (!stp_forward_in_state(port->stp_state)
6265 && !stp_learn_in_state(port->stp_state)) {
6273 do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len,
6274 struct action_xlate_ctx *ctx)
6276 bool was_evictable = true;
6277 const struct ofpact *a;
6280 /* Don't let the rule we're working on get evicted underneath us. */
6281 was_evictable = ctx->rule->up.evictable;
6282 ctx->rule->up.evictable = false;
6284 OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) {
6285 struct ofpact_controller *controller;
6286 const struct ofpact_metadata *metadata;
6294 xlate_output_action(ctx, ofpact_get_OUTPUT(a)->port,
6295 ofpact_get_OUTPUT(a)->max_len, true);
6298 case OFPACT_CONTROLLER:
6299 controller = ofpact_get_CONTROLLER(a);
6300 execute_controller_action(ctx, controller->max_len,
6302 controller->controller_id);
6305 case OFPACT_ENQUEUE:
6306 xlate_enqueue_action(ctx, ofpact_get_ENQUEUE(a));
6309 case OFPACT_SET_VLAN_VID:
6310 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
6311 ctx->flow.vlan_tci |= (htons(ofpact_get_SET_VLAN_VID(a)->vlan_vid)
6315 case OFPACT_SET_VLAN_PCP:
6316 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
6317 ctx->flow.vlan_tci |= htons((ofpact_get_SET_VLAN_PCP(a)->vlan_pcp
6322 case OFPACT_STRIP_VLAN:
6323 ctx->flow.vlan_tci = htons(0);
6326 case OFPACT_PUSH_VLAN:
6327 /* XXX 802.1AD(QinQ) */
6328 ctx->flow.vlan_tci = htons(VLAN_CFI);
6331 case OFPACT_SET_ETH_SRC:
6332 memcpy(ctx->flow.dl_src, ofpact_get_SET_ETH_SRC(a)->mac,
6336 case OFPACT_SET_ETH_DST:
6337 memcpy(ctx->flow.dl_dst, ofpact_get_SET_ETH_DST(a)->mac,
6341 case OFPACT_SET_IPV4_SRC:
6342 ctx->flow.nw_src = ofpact_get_SET_IPV4_SRC(a)->ipv4;
6345 case OFPACT_SET_IPV4_DST:
6346 ctx->flow.nw_dst = ofpact_get_SET_IPV4_DST(a)->ipv4;
6349 case OFPACT_SET_IPV4_DSCP:
6350 /* OpenFlow 1.0 only supports IPv4. */
6351 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
6352 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
6353 ctx->flow.nw_tos |= ofpact_get_SET_IPV4_DSCP(a)->dscp;
6357 case OFPACT_SET_L4_SRC_PORT:
6358 ctx->flow.tp_src = htons(ofpact_get_SET_L4_SRC_PORT(a)->port);
6361 case OFPACT_SET_L4_DST_PORT:
6362 ctx->flow.tp_dst = htons(ofpact_get_SET_L4_DST_PORT(a)->port);
6365 case OFPACT_RESUBMIT:
6366 xlate_ofpact_resubmit(ctx, ofpact_get_RESUBMIT(a));
6369 case OFPACT_SET_TUNNEL:
6370 ctx->flow.tunnel.tun_id = htonll(ofpact_get_SET_TUNNEL(a)->tun_id);
6373 case OFPACT_SET_QUEUE:
6374 xlate_set_queue_action(ctx, ofpact_get_SET_QUEUE(a)->queue_id);
6377 case OFPACT_POP_QUEUE:
6378 ctx->flow.skb_priority = ctx->orig_skb_priority;
6381 case OFPACT_REG_MOVE:
6382 nxm_execute_reg_move(ofpact_get_REG_MOVE(a), &ctx->flow);
6385 case OFPACT_REG_LOAD:
6386 nxm_execute_reg_load(ofpact_get_REG_LOAD(a), &ctx->flow);
6389 case OFPACT_DEC_TTL:
6390 if (compose_dec_ttl(ctx, ofpact_get_DEC_TTL(a))) {
6396 /* Nothing to do. */
6399 case OFPACT_MULTIPATH:
6400 multipath_execute(ofpact_get_MULTIPATH(a), &ctx->flow);
6403 case OFPACT_AUTOPATH:
6404 xlate_autopath(ctx, ofpact_get_AUTOPATH(a));
6408 ctx->ofproto->has_bundle_action = true;
6409 xlate_bundle_action(ctx, ofpact_get_BUNDLE(a));
6412 case OFPACT_OUTPUT_REG:
6413 xlate_output_reg_action(ctx, ofpact_get_OUTPUT_REG(a));
6417 ctx->has_learn = true;
6418 if (ctx->may_learn) {
6419 xlate_learn_action(ctx, ofpact_get_LEARN(a));
6427 case OFPACT_FIN_TIMEOUT:
6428 ctx->has_fin_timeout = true;
6429 xlate_fin_timeout(ctx, ofpact_get_FIN_TIMEOUT(a));
6432 case OFPACT_CLEAR_ACTIONS:
6434 * Nothing to do because writa-actions is not supported for now.
6435 * When writa-actions is supported, clear-actions also must
6436 * be supported at the same time.
6440 case OFPACT_WRITE_METADATA:
6441 metadata = ofpact_get_WRITE_METADATA(a);
6442 ctx->flow.metadata &= ~metadata->mask;
6443 ctx->flow.metadata |= metadata->metadata & metadata->mask;
6446 case OFPACT_GOTO_TABLE: {
6447 /* XXX remove recursion */
6448 /* It is assumed that goto-table is last action */
6449 struct ofpact_goto_table *ogt = ofpact_get_GOTO_TABLE(a);
6450 ovs_assert(ctx->table_id < ogt->table_id);
6451 xlate_table_action(ctx, ctx->flow.in_port, ogt->table_id, true);
6459 ctx->rule->up.evictable = was_evictable;
6464 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
6465 struct ofproto_dpif *ofproto, const struct flow *flow,
6466 ovs_be16 initial_tci, struct rule_dpif *rule,
6467 uint8_t tcp_flags, const struct ofpbuf *packet)
6469 ovs_be64 initial_tun_id = flow->tunnel.tun_id;
6471 /* Flow initialization rules:
6472 * - 'base_flow' must match the kernel's view of the packet at the
6473 * time that action processing starts. 'flow' represents any
6474 * transformations we wish to make through actions.
6475 * - By default 'base_flow' and 'flow' are the same since the input
6476 * packet matches the output before any actions are applied.
6477 * - When using VLAN splinters, 'base_flow''s VLAN is set to the value
6478 * of the received packet as seen by the kernel. If we later output
6479 * to another device without any modifications this will cause us to
6480 * insert a new tag since the original one was stripped off by the
6482 * - Tunnel 'flow' is largely cleared when transitioning between
6483 * the input and output stages since it does not make sense to output
6484 * a packet with the exact headers that it was received with (i.e.
6485 * the destination IP is us). The one exception is the tun_id, which
6486 * is preserved to allow use in later resubmit lookups and loads into
6488 * - Tunnel 'base_flow' is completely cleared since that is what the
6489 * kernel does. If we wish to maintain the original values an action
6490 * needs to be generated. */
6492 ctx->ofproto = ofproto;
6494 memset(&ctx->flow.tunnel, 0, sizeof ctx->flow.tunnel);
6495 ctx->base_flow = ctx->flow;
6496 ctx->base_flow.vlan_tci = initial_tci;
6497 ctx->flow.tunnel.tun_id = initial_tun_id;
6499 ctx->packet = packet;
6500 ctx->may_learn = packet != NULL;
6501 ctx->tcp_flags = tcp_flags;
6502 ctx->resubmit_hook = NULL;
6503 ctx->report_hook = NULL;
6504 ctx->resubmit_stats = NULL;
6507 /* Translates the 'ofpacts_len' bytes of "struct ofpacts" starting at 'ofpacts'
6508 * into datapath actions in 'odp_actions', using 'ctx'. */
6510 xlate_actions(struct action_xlate_ctx *ctx,
6511 const struct ofpact *ofpacts, size_t ofpacts_len,
6512 struct ofpbuf *odp_actions)
6514 /* Normally false. Set to true if we ever hit MAX_RESUBMIT_RECURSION, so
6515 * that in the future we always keep a copy of the original flow for
6516 * tracing purposes. */
6517 static bool hit_resubmit_limit;
6519 enum slow_path_reason special;
6520 struct ofport_dpif *in_port;
6522 COVERAGE_INC(ofproto_dpif_xlate);
6524 ofpbuf_clear(odp_actions);
6525 ofpbuf_reserve(odp_actions, NL_A_U32_SIZE);
6527 ctx->odp_actions = odp_actions;
6530 ctx->has_learn = false;
6531 ctx->has_normal = false;
6532 ctx->has_fin_timeout = false;
6533 ctx->nf_output_iface = NF_OUT_DROP;
6536 ctx->max_resubmit_trigger = false;
6537 ctx->orig_skb_priority = ctx->flow.skb_priority;
6541 if (ctx->ofproto->has_mirrors || hit_resubmit_limit) {
6542 /* Do this conditionally because the copy is expensive enough that it
6543 * shows up in profiles.
6545 * We keep orig_flow in 'ctx' only because I couldn't make GCC 4.4
6546 * believe that I wasn't using it without initializing it if I kept it
6547 * in a local variable. */
6548 ctx->orig_flow = ctx->flow;
6551 if (ctx->flow.nw_frag & FLOW_NW_FRAG_ANY) {
6552 switch (ctx->ofproto->up.frag_handling) {
6553 case OFPC_FRAG_NORMAL:
6554 /* We must pretend that transport ports are unavailable. */
6555 ctx->flow.tp_src = ctx->base_flow.tp_src = htons(0);
6556 ctx->flow.tp_dst = ctx->base_flow.tp_dst = htons(0);
6559 case OFPC_FRAG_DROP:
6562 case OFPC_FRAG_REASM:
6565 case OFPC_FRAG_NX_MATCH:
6566 /* Nothing to do. */
6569 case OFPC_INVALID_TTL_TO_CONTROLLER:
6574 in_port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
6575 special = process_special(ctx->ofproto, &ctx->flow, in_port, ctx->packet);
6577 ctx->slow |= special;
6579 static struct vlog_rate_limit trace_rl = VLOG_RATE_LIMIT_INIT(1, 1);
6580 ovs_be16 initial_tci = ctx->base_flow.vlan_tci;
6581 uint32_t local_odp_port;
6583 add_sflow_action(ctx);
6585 if (!in_port || may_receive(in_port, ctx)) {
6586 do_xlate_actions(ofpacts, ofpacts_len, ctx);
6588 /* We've let OFPP_NORMAL and the learning action look at the
6589 * packet, so drop it now if forwarding is disabled. */
6590 if (in_port && !stp_forward_in_state(in_port->stp_state)) {
6591 ofpbuf_clear(ctx->odp_actions);
6592 add_sflow_action(ctx);
6596 if (ctx->max_resubmit_trigger && !ctx->resubmit_hook) {
6597 if (!hit_resubmit_limit) {
6598 /* We didn't record the original flow. Make sure we do from
6600 hit_resubmit_limit = true;
6601 } else if (!VLOG_DROP_ERR(&trace_rl)) {
6602 struct ds ds = DS_EMPTY_INITIALIZER;
6604 ofproto_trace(ctx->ofproto, &ctx->orig_flow, ctx->packet,
6606 VLOG_ERR("Trace triggered by excessive resubmit "
6607 "recursion:\n%s", ds_cstr(&ds));
6612 local_odp_port = ofp_port_to_odp_port(ctx->ofproto, OFPP_LOCAL);
6613 if (!connmgr_may_set_up_flow(ctx->ofproto->up.connmgr, &ctx->flow,
6615 ctx->odp_actions->data,
6616 ctx->odp_actions->size)) {
6617 ctx->slow |= SLOW_IN_BAND;
6619 && connmgr_msg_in_hook(ctx->ofproto->up.connmgr, &ctx->flow,
6621 compose_output_action(ctx, OFPP_LOCAL);
6624 if (ctx->ofproto->has_mirrors) {
6625 add_mirror_actions(ctx, &ctx->orig_flow);
6627 fix_sflow_action(ctx);
6631 /* Translates the 'ofpacts_len' bytes of "struct ofpact"s starting at 'ofpacts'
6632 * into datapath actions, using 'ctx', and discards the datapath actions. */
6634 xlate_actions_for_side_effects(struct action_xlate_ctx *ctx,
6635 const struct ofpact *ofpacts,
6638 uint64_t odp_actions_stub[1024 / 8];
6639 struct ofpbuf odp_actions;
6641 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
6642 xlate_actions(ctx, ofpacts, ofpacts_len, &odp_actions);
6643 ofpbuf_uninit(&odp_actions);
6647 xlate_report(struct action_xlate_ctx *ctx, const char *s)
6649 if (ctx->report_hook) {
6650 ctx->report_hook(ctx, s);
6654 /* OFPP_NORMAL implementation. */
6656 static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *);
6658 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
6659 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle',
6660 * the bundle on which the packet was received, returns the VLAN to which the
6663 * Both 'vid' and the return value are in the range 0...4095. */
6665 input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid)
6667 switch (in_bundle->vlan_mode) {
6668 case PORT_VLAN_ACCESS:
6669 return in_bundle->vlan;
6672 case PORT_VLAN_TRUNK:
6675 case PORT_VLAN_NATIVE_UNTAGGED:
6676 case PORT_VLAN_NATIVE_TAGGED:
6677 return vid ? vid : in_bundle->vlan;
6684 /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'.
6685 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
6688 * 'vid' should be the VID obtained from the 802.1Q header that was received as
6689 * part of a packet (specify 0 if there was no 802.1Q header), in the range
6692 input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn)
6694 /* Allow any VID on the OFPP_NONE port. */
6695 if (in_bundle == &ofpp_none_bundle) {
6699 switch (in_bundle->vlan_mode) {
6700 case PORT_VLAN_ACCESS:
6703 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6704 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
6705 "packet received on port %s configured as VLAN "
6706 "%"PRIu16" access port",
6707 in_bundle->ofproto->up.name, vid,
6708 in_bundle->name, in_bundle->vlan);
6714 case PORT_VLAN_NATIVE_UNTAGGED:
6715 case PORT_VLAN_NATIVE_TAGGED:
6717 /* Port must always carry its native VLAN. */
6721 case PORT_VLAN_TRUNK:
6722 if (!ofbundle_includes_vlan(in_bundle, vid)) {
6724 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6725 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet "
6726 "received on port %s not configured for trunking "
6728 in_bundle->ofproto->up.name, vid,
6729 in_bundle->name, vid);
6741 /* Given 'vlan', the VLAN that a packet belongs to, and
6742 * 'out_bundle', a bundle on which the packet is to be output, returns the VID
6743 * that should be included in the 802.1Q header. (If the return value is 0,
6744 * then the 802.1Q header should only be included in the packet if there is a
6747 * Both 'vlan' and the return value are in the range 0...4095. */
6749 output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan)
6751 switch (out_bundle->vlan_mode) {
6752 case PORT_VLAN_ACCESS:
6755 case PORT_VLAN_TRUNK:
6756 case PORT_VLAN_NATIVE_TAGGED:
6759 case PORT_VLAN_NATIVE_UNTAGGED:
6760 return vlan == out_bundle->vlan ? 0 : vlan;
6768 output_normal(struct action_xlate_ctx *ctx, const struct ofbundle *out_bundle,
6771 struct ofport_dpif *port;
6773 ovs_be16 tci, old_tci;
6775 vid = output_vlan_to_vid(out_bundle, vlan);
6776 if (!out_bundle->bond) {
6777 port = ofbundle_get_a_port(out_bundle);
6779 port = bond_choose_output_slave(out_bundle->bond, &ctx->flow,
6782 /* No slaves enabled, so drop packet. */
6787 old_tci = ctx->flow.vlan_tci;
6789 if (tci || out_bundle->use_priority_tags) {
6790 tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK);
6792 tci |= htons(VLAN_CFI);
6795 ctx->flow.vlan_tci = tci;
6797 compose_output_action(ctx, port->up.ofp_port);
6798 ctx->flow.vlan_tci = old_tci;
6802 mirror_mask_ffs(mirror_mask_t mask)
6804 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
6809 ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
6811 return (bundle->vlan_mode != PORT_VLAN_ACCESS
6812 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
6816 ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan)
6818 return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan);
6821 /* Returns an arbitrary interface within 'bundle'. */
6822 static struct ofport_dpif *
6823 ofbundle_get_a_port(const struct ofbundle *bundle)
6825 return CONTAINER_OF(list_front(&bundle->ports),
6826 struct ofport_dpif, bundle_node);
6830 vlan_is_mirrored(const struct ofmirror *m, int vlan)
6832 return !m->vlans || bitmap_is_set(m->vlans, vlan);
6836 add_mirror_actions(struct action_xlate_ctx *ctx, const struct flow *orig_flow)
6838 struct ofproto_dpif *ofproto = ctx->ofproto;
6839 mirror_mask_t mirrors;
6840 struct ofbundle *in_bundle;
6843 const struct nlattr *a;
6846 in_bundle = lookup_input_bundle(ctx->ofproto, orig_flow->in_port,
6847 ctx->packet != NULL, NULL);
6851 mirrors = in_bundle->src_mirrors;
6853 /* Drop frames on bundles reserved for mirroring. */
6854 if (in_bundle->mirror_out) {
6855 if (ctx->packet != NULL) {
6856 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6857 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
6858 "%s, which is reserved exclusively for mirroring",
6859 ctx->ofproto->up.name, in_bundle->name);
6865 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
6866 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
6869 vlan = input_vid_to_vlan(in_bundle, vid);
6871 /* Look at the output ports to check for destination selections. */
6873 NL_ATTR_FOR_EACH (a, left, ctx->odp_actions->data,
6874 ctx->odp_actions->size) {
6875 enum ovs_action_attr type = nl_attr_type(a);
6876 struct ofport_dpif *ofport;
6878 if (type != OVS_ACTION_ATTR_OUTPUT) {
6882 ofport = get_odp_port(ofproto, nl_attr_get_u32(a));
6883 if (ofport && ofport->bundle) {
6884 mirrors |= ofport->bundle->dst_mirrors;
6892 /* Restore the original packet before adding the mirror actions. */
6893 ctx->flow = *orig_flow;
6898 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
6900 if (!vlan_is_mirrored(m, vlan)) {
6901 mirrors = zero_rightmost_1bit(mirrors);
6905 mirrors &= ~m->dup_mirrors;
6906 ctx->mirrors |= m->dup_mirrors;
6908 output_normal(ctx, m->out, vlan);
6909 } else if (vlan != m->out_vlan
6910 && !eth_addr_is_reserved(orig_flow->dl_dst)) {
6911 struct ofbundle *bundle;
6913 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
6914 if (ofbundle_includes_vlan(bundle, m->out_vlan)
6915 && !bundle->mirror_out) {
6916 output_normal(ctx, bundle, m->out_vlan);
6924 update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors,
6925 uint64_t packets, uint64_t bytes)
6931 for (; mirrors; mirrors = zero_rightmost_1bit(mirrors)) {
6934 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
6937 /* In normal circumstances 'm' will not be NULL. However,
6938 * if mirrors are reconfigured, we can temporarily get out
6939 * of sync in facet_revalidate(). We could "correct" the
6940 * mirror list before reaching here, but doing that would
6941 * not properly account the traffic stats we've currently
6942 * accumulated for previous mirror configuration. */
6946 m->packet_count += packets;
6947 m->byte_count += bytes;
6951 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
6952 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
6953 * indicate this; newer upstream kernels use gratuitous ARP requests. */
6955 is_gratuitous_arp(const struct flow *flow)
6957 return (flow->dl_type == htons(ETH_TYPE_ARP)
6958 && eth_addr_is_broadcast(flow->dl_dst)
6959 && (flow->nw_proto == ARP_OP_REPLY
6960 || (flow->nw_proto == ARP_OP_REQUEST
6961 && flow->nw_src == flow->nw_dst)));
6965 update_learning_table(struct ofproto_dpif *ofproto,
6966 const struct flow *flow, int vlan,
6967 struct ofbundle *in_bundle)
6969 struct mac_entry *mac;
6971 /* Don't learn the OFPP_NONE port. */
6972 if (in_bundle == &ofpp_none_bundle) {
6976 if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
6980 mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan);
6981 if (is_gratuitous_arp(flow)) {
6982 /* We don't want to learn from gratuitous ARP packets that are
6983 * reflected back over bond slaves so we lock the learning table. */
6984 if (!in_bundle->bond) {
6985 mac_entry_set_grat_arp_lock(mac);
6986 } else if (mac_entry_is_grat_arp_locked(mac)) {
6991 if (mac_entry_is_new(mac) || mac->port.p != in_bundle) {
6992 /* The log messages here could actually be useful in debugging,
6993 * so keep the rate limit relatively high. */
6994 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
6995 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
6996 "on port %s in VLAN %d",
6997 ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src),
6998 in_bundle->name, vlan);
7000 mac->port.p = in_bundle;
7001 tag_set_add(&ofproto->backer->revalidate_set,
7002 mac_learning_changed(ofproto->ml, mac));
7006 static struct ofbundle *
7007 lookup_input_bundle(const struct ofproto_dpif *ofproto, uint16_t in_port,
7008 bool warn, struct ofport_dpif **in_ofportp)
7010 struct ofport_dpif *ofport;
7012 /* Find the port and bundle for the received packet. */
7013 ofport = get_ofp_port(ofproto, in_port);
7015 *in_ofportp = ofport;
7017 if (ofport && ofport->bundle) {
7018 return ofport->bundle;
7021 /* Special-case OFPP_NONE, which a controller may use as the ingress
7022 * port for traffic that it is sourcing. */
7023 if (in_port == OFPP_NONE) {
7024 return &ofpp_none_bundle;
7027 /* Odd. A few possible reasons here:
7029 * - We deleted a port but there are still a few packets queued up
7032 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
7033 * we don't know about.
7035 * - The ofproto client didn't configure the port as part of a bundle.
7036 * This is particularly likely to happen if a packet was received on the
7037 * port after it was created, but before the client had a chance to
7038 * configure its bundle.
7041 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7043 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
7044 "port %"PRIu16, ofproto->up.name, in_port);
7049 /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or
7050 * dropped. Returns true if they may be forwarded, false if they should be
7053 * 'in_port' must be the ofport_dpif that corresponds to flow->in_port.
7054 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
7056 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
7057 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
7058 * checked by input_vid_is_valid().
7060 * May also add tags to '*tags', although the current implementation only does
7061 * so in one special case.
7064 is_admissible(struct action_xlate_ctx *ctx, struct ofport_dpif *in_port,
7067 struct ofproto_dpif *ofproto = ctx->ofproto;
7068 struct flow *flow = &ctx->flow;
7069 struct ofbundle *in_bundle = in_port->bundle;
7071 /* Drop frames for reserved multicast addresses
7072 * only if forward_bpdu option is absent. */
7073 if (!ofproto->up.forward_bpdu && eth_addr_is_reserved(flow->dl_dst)) {
7074 xlate_report(ctx, "packet has reserved destination MAC, dropping");
7078 if (in_bundle->bond) {
7079 struct mac_entry *mac;
7081 switch (bond_check_admissibility(in_bundle->bond, in_port,
7082 flow->dl_dst, &ctx->tags)) {
7087 xlate_report(ctx, "bonding refused admissibility, dropping");
7090 case BV_DROP_IF_MOVED:
7091 mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL);
7092 if (mac && mac->port.p != in_bundle &&
7093 (!is_gratuitous_arp(flow)
7094 || mac_entry_is_grat_arp_locked(mac))) {
7095 xlate_report(ctx, "SLB bond thinks this packet looped back, "
7107 xlate_normal(struct action_xlate_ctx *ctx)
7109 struct ofport_dpif *in_port;
7110 struct ofbundle *in_bundle;
7111 struct mac_entry *mac;
7115 ctx->has_normal = true;
7117 in_bundle = lookup_input_bundle(ctx->ofproto, ctx->flow.in_port,
7118 ctx->packet != NULL, &in_port);
7120 xlate_report(ctx, "no input bundle, dropping");
7124 /* Drop malformed frames. */
7125 if (ctx->flow.dl_type == htons(ETH_TYPE_VLAN) &&
7126 !(ctx->flow.vlan_tci & htons(VLAN_CFI))) {
7127 if (ctx->packet != NULL) {
7128 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7129 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
7130 "VLAN tag received on port %s",
7131 ctx->ofproto->up.name, in_bundle->name);
7133 xlate_report(ctx, "partial VLAN tag, dropping");
7137 /* Drop frames on bundles reserved for mirroring. */
7138 if (in_bundle->mirror_out) {
7139 if (ctx->packet != NULL) {
7140 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7141 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
7142 "%s, which is reserved exclusively for mirroring",
7143 ctx->ofproto->up.name, in_bundle->name);
7145 xlate_report(ctx, "input port is mirror output port, dropping");
7150 vid = vlan_tci_to_vid(ctx->flow.vlan_tci);
7151 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
7152 xlate_report(ctx, "disallowed VLAN VID for this input port, dropping");
7155 vlan = input_vid_to_vlan(in_bundle, vid);
7157 /* Check other admissibility requirements. */
7158 if (in_port && !is_admissible(ctx, in_port, vlan)) {
7162 /* Learn source MAC. */
7163 if (ctx->may_learn) {
7164 update_learning_table(ctx->ofproto, &ctx->flow, vlan, in_bundle);
7167 /* Determine output bundle. */
7168 mac = mac_learning_lookup(ctx->ofproto->ml, ctx->flow.dl_dst, vlan,
7171 if (mac->port.p != in_bundle) {
7172 xlate_report(ctx, "forwarding to learned port");
7173 output_normal(ctx, mac->port.p, vlan);
7175 xlate_report(ctx, "learned port is input port, dropping");
7178 struct ofbundle *bundle;
7180 xlate_report(ctx, "no learned MAC for destination, flooding");
7181 HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
7182 if (bundle != in_bundle
7183 && ofbundle_includes_vlan(bundle, vlan)
7184 && bundle->floodable
7185 && !bundle->mirror_out) {
7186 output_normal(ctx, bundle, vlan);
7189 ctx->nf_output_iface = NF_OUT_FLOOD;
7193 /* Optimized flow revalidation.
7195 * It's a difficult problem, in general, to tell which facets need to have
7196 * their actions recalculated whenever the OpenFlow flow table changes. We
7197 * don't try to solve that general problem: for most kinds of OpenFlow flow
7198 * table changes, we recalculate the actions for every facet. This is
7199 * relatively expensive, but it's good enough if the OpenFlow flow table
7200 * doesn't change very often.
7202 * However, we can expect one particular kind of OpenFlow flow table change to
7203 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
7204 * of CPU on revalidating every facet whenever MAC learning modifies the flow
7205 * table, we add a special case that applies to flow tables in which every rule
7206 * has the same form (that is, the same wildcards), except that the table is
7207 * also allowed to have a single "catch-all" flow that matches all packets. We
7208 * optimize this case by tagging all of the facets that resubmit into the table
7209 * and invalidating the same tag whenever a flow changes in that table. The
7210 * end result is that we revalidate just the facets that need it (and sometimes
7211 * a few more, but not all of the facets or even all of the facets that
7212 * resubmit to the table modified by MAC learning). */
7214 /* Calculates the tag to use for 'flow' and mask 'mask' when it is inserted
7215 * into an OpenFlow table with the given 'basis'. */
7217 rule_calculate_tag(const struct flow *flow, const struct minimask *mask,
7220 if (minimask_is_catchall(mask)) {
7223 uint32_t hash = flow_hash_in_minimask(flow, mask, secret);
7224 return tag_create_deterministic(hash);
7228 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
7229 * taggability of that table.
7231 * This function must be called after *each* change to a flow table. If you
7232 * skip calling it on some changes then the pointer comparisons at the end can
7233 * be invalid if you get unlucky. For example, if a flow removal causes a
7234 * cls_table to be destroyed and then a flow insertion causes a cls_table with
7235 * different wildcards to be created with the same address, then this function
7236 * will incorrectly skip revalidation. */
7238 table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
7240 struct table_dpif *table = &ofproto->tables[table_id];
7241 const struct oftable *oftable = &ofproto->up.tables[table_id];
7242 struct cls_table *catchall, *other;
7243 struct cls_table *t;
7245 catchall = other = NULL;
7247 switch (hmap_count(&oftable->cls.tables)) {
7249 /* We could tag this OpenFlow table but it would make the logic a
7250 * little harder and it's a corner case that doesn't seem worth it
7256 HMAP_FOR_EACH (t, hmap_node, &oftable->cls.tables) {
7257 if (cls_table_is_catchall(t)) {
7259 } else if (!other) {
7262 /* Indicate that we can't tag this by setting both tables to
7263 * NULL. (We know that 'catchall' is already NULL.) */
7270 /* Can't tag this table. */
7274 if (table->catchall_table != catchall || table->other_table != other) {
7275 table->catchall_table = catchall;
7276 table->other_table = other;
7277 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
7281 /* Given 'rule' that has changed in some way (either it is a rule being
7282 * inserted, a rule being deleted, or a rule whose actions are being
7283 * modified), marks facets for revalidation to ensure that packets will be
7284 * forwarded correctly according to the new state of the flow table.
7286 * This function must be called after *each* change to a flow table. See
7287 * the comment on table_update_taggable() for more information. */
7289 rule_invalidate(const struct rule_dpif *rule)
7291 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
7293 table_update_taggable(ofproto, rule->up.table_id);
7295 if (!ofproto->backer->need_revalidate) {
7296 struct table_dpif *table = &ofproto->tables[rule->up.table_id];
7298 if (table->other_table && rule->tag) {
7299 tag_set_add(&ofproto->backer->revalidate_set, rule->tag);
7301 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
7307 set_frag_handling(struct ofproto *ofproto_,
7308 enum ofp_config_flags frag_handling)
7310 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7311 if (frag_handling != OFPC_FRAG_REASM) {
7312 ofproto->backer->need_revalidate = REV_RECONFIGURE;
7320 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
7321 const struct flow *flow,
7322 const struct ofpact *ofpacts, size_t ofpacts_len)
7324 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7325 struct odputil_keybuf keybuf;
7326 struct dpif_flow_stats stats;
7330 struct action_xlate_ctx ctx;
7331 uint64_t odp_actions_stub[1024 / 8];
7332 struct ofpbuf odp_actions;
7334 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
7335 odp_flow_key_from_flow(&key, flow,
7336 ofp_port_to_odp_port(ofproto, flow->in_port));
7338 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
7340 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci, NULL,
7341 packet_get_tcp_flags(packet, flow), packet);
7342 ctx.resubmit_stats = &stats;
7344 ofpbuf_use_stub(&odp_actions,
7345 odp_actions_stub, sizeof odp_actions_stub);
7346 xlate_actions(&ctx, ofpacts, ofpacts_len, &odp_actions);
7347 dpif_execute(ofproto->backer->dpif, key.data, key.size,
7348 odp_actions.data, odp_actions.size, packet);
7349 ofpbuf_uninit(&odp_actions);
7357 set_netflow(struct ofproto *ofproto_,
7358 const struct netflow_options *netflow_options)
7360 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7362 if (netflow_options) {
7363 if (!ofproto->netflow) {
7364 ofproto->netflow = netflow_create();
7366 return netflow_set_options(ofproto->netflow, netflow_options);
7368 netflow_destroy(ofproto->netflow);
7369 ofproto->netflow = NULL;
7375 get_netflow_ids(const struct ofproto *ofproto_,
7376 uint8_t *engine_type, uint8_t *engine_id)
7378 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7380 dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
7384 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
7386 if (!facet_is_controller_flow(facet) &&
7387 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
7388 struct subfacet *subfacet;
7389 struct ofexpired expired;
7391 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
7392 if (subfacet->path == SF_FAST_PATH) {
7393 struct dpif_flow_stats stats;
7395 subfacet_reinstall(subfacet, &stats);
7396 subfacet_update_stats(subfacet, &stats);
7400 expired.flow = facet->flow;
7401 expired.packet_count = facet->packet_count;
7402 expired.byte_count = facet->byte_count;
7403 expired.used = facet->used;
7404 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
7409 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
7411 struct facet *facet;
7413 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
7414 send_active_timeout(ofproto, facet);
7418 static struct ofproto_dpif *
7419 ofproto_dpif_lookup(const char *name)
7421 struct ofproto_dpif *ofproto;
7423 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
7424 hash_string(name, 0), &all_ofproto_dpifs) {
7425 if (!strcmp(ofproto->up.name, name)) {
7433 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
7434 const char *argv[], void *aux OVS_UNUSED)
7436 struct ofproto_dpif *ofproto;
7439 ofproto = ofproto_dpif_lookup(argv[1]);
7441 unixctl_command_reply_error(conn, "no such bridge");
7444 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
7446 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
7447 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
7451 unixctl_command_reply(conn, "table successfully flushed");
7455 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
7456 const char *argv[], void *aux OVS_UNUSED)
7458 struct ds ds = DS_EMPTY_INITIALIZER;
7459 const struct ofproto_dpif *ofproto;
7460 const struct mac_entry *e;
7462 ofproto = ofproto_dpif_lookup(argv[1]);
7464 unixctl_command_reply_error(conn, "no such bridge");
7468 ds_put_cstr(&ds, " port VLAN MAC Age\n");
7469 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
7470 struct ofbundle *bundle = e->port.p;
7471 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
7472 ofbundle_get_a_port(bundle)->odp_port,
7473 e->vlan, ETH_ADDR_ARGS(e->mac),
7474 mac_entry_age(ofproto->ml, e));
7476 unixctl_command_reply(conn, ds_cstr(&ds));
7481 struct action_xlate_ctx ctx;
7487 trace_format_rule(struct ds *result, uint8_t table_id, int level,
7488 const struct rule_dpif *rule)
7490 ds_put_char_multiple(result, '\t', level);
7492 ds_put_cstr(result, "No match\n");
7496 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
7497 table_id, ntohll(rule->up.flow_cookie));
7498 cls_rule_format(&rule->up.cr, result);
7499 ds_put_char(result, '\n');
7501 ds_put_char_multiple(result, '\t', level);
7502 ds_put_cstr(result, "OpenFlow ");
7503 ofpacts_format(rule->up.ofpacts, rule->up.ofpacts_len, result);
7504 ds_put_char(result, '\n');
7508 trace_format_flow(struct ds *result, int level, const char *title,
7509 struct trace_ctx *trace)
7511 ds_put_char_multiple(result, '\t', level);
7512 ds_put_format(result, "%s: ", title);
7513 if (flow_equal(&trace->ctx.flow, &trace->flow)) {
7514 ds_put_cstr(result, "unchanged");
7516 flow_format(result, &trace->ctx.flow);
7517 trace->flow = trace->ctx.flow;
7519 ds_put_char(result, '\n');
7523 trace_format_regs(struct ds *result, int level, const char *title,
7524 struct trace_ctx *trace)
7528 ds_put_char_multiple(result, '\t', level);
7529 ds_put_format(result, "%s:", title);
7530 for (i = 0; i < FLOW_N_REGS; i++) {
7531 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
7533 ds_put_char(result, '\n');
7537 trace_format_odp(struct ds *result, int level, const char *title,
7538 struct trace_ctx *trace)
7540 struct ofpbuf *odp_actions = trace->ctx.odp_actions;
7542 ds_put_char_multiple(result, '\t', level);
7543 ds_put_format(result, "%s: ", title);
7544 format_odp_actions(result, odp_actions->data, odp_actions->size);
7545 ds_put_char(result, '\n');
7549 trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
7551 struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx);
7552 struct ds *result = trace->result;
7554 ds_put_char(result, '\n');
7555 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
7556 trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace);
7557 trace_format_odp(result, ctx->recurse + 1, "Resubmitted odp", trace);
7558 trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule);
7562 trace_report(struct action_xlate_ctx *ctx, const char *s)
7564 struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx);
7565 struct ds *result = trace->result;
7567 ds_put_char_multiple(result, '\t', ctx->recurse);
7568 ds_put_cstr(result, s);
7569 ds_put_char(result, '\n');
7573 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
7574 void *aux OVS_UNUSED)
7576 const char *dpname = argv[1];
7577 struct ofproto_dpif *ofproto;
7578 struct ofpbuf odp_key;
7579 struct ofpbuf *packet;
7580 ovs_be16 initial_tci;
7586 ofpbuf_init(&odp_key, 0);
7589 ofproto = ofproto_dpif_lookup(dpname);
7591 unixctl_command_reply_error(conn, "Unknown ofproto (use ofproto/list "
7595 if (argc == 3 || (argc == 4 && !strcmp(argv[3], "-generate"))) {
7596 /* ofproto/trace dpname flow [-generate] */
7597 const char *flow_s = argv[2];
7598 const char *generate_s = argv[3];
7600 /* Allow 'flow_s' to be either a datapath flow or an OpenFlow-like
7601 * flow. We guess which type it is based on whether 'flow_s' contains
7602 * an '(', since a datapath flow always contains '(') but an
7603 * OpenFlow-like flow should not (in fact it's allowed but I believe
7604 * that's not documented anywhere).
7606 * An alternative would be to try to parse 'flow_s' both ways, but then
7607 * it would be tricky giving a sensible error message. After all, do
7608 * you just say "syntax error" or do you present both error messages?
7609 * Both choices seem lousy. */
7610 if (strchr(flow_s, '(')) {
7613 /* Convert string to datapath key. */
7614 ofpbuf_init(&odp_key, 0);
7615 error = odp_flow_key_from_string(flow_s, NULL, &odp_key);
7617 unixctl_command_reply_error(conn, "Bad flow syntax");
7621 /* XXX: Since we allow the user to specify an ofproto, it's
7622 * possible they will specify a different ofproto than the one the
7623 * port actually belongs too. Ideally we should simply remove the
7624 * ability to specify the ofproto. */
7625 if (ofproto_receive(ofproto->backer, NULL, odp_key.data,
7626 odp_key.size, &flow, NULL, NULL, NULL,
7628 unixctl_command_reply_error(conn, "Invalid flow");
7634 error_s = parse_ofp_exact_flow(&flow, argv[2]);
7636 unixctl_command_reply_error(conn, error_s);
7641 initial_tci = flow.vlan_tci;
7644 /* Generate a packet, if requested. */
7646 packet = ofpbuf_new(0);
7647 flow_compose(packet, &flow);
7649 } else if (argc == 7) {
7650 /* ofproto/trace dpname priority tun_id in_port mark packet */
7651 const char *priority_s = argv[2];
7652 const char *tun_id_s = argv[3];
7653 const char *in_port_s = argv[4];
7654 const char *mark_s = argv[5];
7655 const char *packet_s = argv[6];
7656 uint32_t in_port = atoi(in_port_s);
7657 ovs_be64 tun_id = htonll(strtoull(tun_id_s, NULL, 0));
7658 uint32_t priority = atoi(priority_s);
7659 uint32_t mark = atoi(mark_s);
7662 msg = eth_from_hex(packet_s, &packet);
7664 unixctl_command_reply_error(conn, msg);
7668 ds_put_cstr(&result, "Packet: ");
7669 s = ofp_packet_to_string(packet->data, packet->size);
7670 ds_put_cstr(&result, s);
7673 flow_extract(packet, priority, mark, NULL, in_port, &flow);
7674 flow.tunnel.tun_id = tun_id;
7675 initial_tci = flow.vlan_tci;
7677 unixctl_command_reply_error(conn, "Bad command syntax");
7681 ofproto_trace(ofproto, &flow, packet, initial_tci, &result);
7682 unixctl_command_reply(conn, ds_cstr(&result));
7685 ds_destroy(&result);
7686 ofpbuf_delete(packet);
7687 ofpbuf_uninit(&odp_key);
7691 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
7692 const struct ofpbuf *packet, ovs_be16 initial_tci,
7695 struct rule_dpif *rule;
7697 ds_put_cstr(ds, "Flow: ");
7698 flow_format(ds, flow);
7699 ds_put_char(ds, '\n');
7701 rule = rule_dpif_lookup(ofproto, flow);
7703 trace_format_rule(ds, 0, 0, rule);
7704 if (rule == ofproto->miss_rule) {
7705 ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
7706 } else if (rule == ofproto->no_packet_in_rule) {
7707 ds_put_cstr(ds, "\nNo match, packets dropped because "
7708 "OFPPC_NO_PACKET_IN is set on in_port.\n");
7712 uint64_t odp_actions_stub[1024 / 8];
7713 struct ofpbuf odp_actions;
7715 struct trace_ctx trace;
7718 tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
7721 ofpbuf_use_stub(&odp_actions,
7722 odp_actions_stub, sizeof odp_actions_stub);
7723 action_xlate_ctx_init(&trace.ctx, ofproto, flow, initial_tci,
7724 rule, tcp_flags, packet);
7725 trace.ctx.resubmit_hook = trace_resubmit;
7726 trace.ctx.report_hook = trace_report;
7727 xlate_actions(&trace.ctx, rule->up.ofpacts, rule->up.ofpacts_len,
7730 ds_put_char(ds, '\n');
7731 trace_format_flow(ds, 0, "Final flow", &trace);
7732 ds_put_cstr(ds, "Datapath actions: ");
7733 format_odp_actions(ds, odp_actions.data, odp_actions.size);
7734 ofpbuf_uninit(&odp_actions);
7736 if (trace.ctx.slow) {
7737 enum slow_path_reason slow;
7739 ds_put_cstr(ds, "\nThis flow is handled by the userspace "
7740 "slow path because it:");
7741 for (slow = trace.ctx.slow; slow; ) {
7742 enum slow_path_reason bit = rightmost_1bit(slow);
7746 ds_put_cstr(ds, "\n\t- Consists of CFM packets.");
7749 ds_put_cstr(ds, "\n\t- Consists of LACP packets.");
7752 ds_put_cstr(ds, "\n\t- Consists of STP packets.");
7755 ds_put_cstr(ds, "\n\t- Needs in-band special case "
7758 ds_put_cstr(ds, "\n\t (The datapath actions are "
7759 "incomplete--for complete actions, "
7760 "please supply a packet.)");
7763 case SLOW_CONTROLLER:
7764 ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages "
7765 "to the OpenFlow controller.");
7768 ds_put_cstr(ds, "\n\t- Needs more specific matching "
7769 "than the datapath supports.");
7776 if (slow & ~SLOW_MATCH) {
7777 ds_put_cstr(ds, "\nThe datapath actions above do not reflect "
7778 "the special slow-path processing.");
7785 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
7786 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
7789 unixctl_command_reply(conn, NULL);
7793 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
7794 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
7797 unixctl_command_reply(conn, NULL);
7800 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
7801 * 'reply' describing the results. */
7803 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
7805 struct facet *facet;
7809 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
7810 if (!facet_check_consistency(facet)) {
7815 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
7819 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
7820 ofproto->up.name, errors);
7822 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
7827 ofproto_dpif_self_check(struct unixctl_conn *conn,
7828 int argc, const char *argv[], void *aux OVS_UNUSED)
7830 struct ds reply = DS_EMPTY_INITIALIZER;
7831 struct ofproto_dpif *ofproto;
7834 ofproto = ofproto_dpif_lookup(argv[1]);
7836 unixctl_command_reply_error(conn, "Unknown ofproto (use "
7837 "ofproto/list for help)");
7840 ofproto_dpif_self_check__(ofproto, &reply);
7842 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
7843 ofproto_dpif_self_check__(ofproto, &reply);
7847 unixctl_command_reply(conn, ds_cstr(&reply));
7851 /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list
7852 * of the 'ofproto_shash' nodes. It is the responsibility of the caller
7853 * to destroy 'ofproto_shash' and free the returned value. */
7854 static const struct shash_node **
7855 get_ofprotos(struct shash *ofproto_shash)
7857 const struct ofproto_dpif *ofproto;
7859 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
7860 char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
7861 shash_add_nocopy(ofproto_shash, name, ofproto);
7864 return shash_sort(ofproto_shash);
7868 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
7869 const char *argv[] OVS_UNUSED,
7870 void *aux OVS_UNUSED)
7872 struct ds ds = DS_EMPTY_INITIALIZER;
7873 struct shash ofproto_shash;
7874 const struct shash_node **sorted_ofprotos;
7877 shash_init(&ofproto_shash);
7878 sorted_ofprotos = get_ofprotos(&ofproto_shash);
7879 for (i = 0; i < shash_count(&ofproto_shash); i++) {
7880 const struct shash_node *node = sorted_ofprotos[i];
7881 ds_put_format(&ds, "%s\n", node->name);
7884 shash_destroy(&ofproto_shash);
7885 free(sorted_ofprotos);
7887 unixctl_command_reply(conn, ds_cstr(&ds));
7892 show_dp_format(const struct ofproto_dpif *ofproto, struct ds *ds)
7894 struct dpif_dp_stats s;
7895 const struct shash_node **ports;
7898 dpif_get_dp_stats(ofproto->backer->dpif, &s);
7900 ds_put_format(ds, "%s (%s):\n", ofproto->up.name,
7901 dpif_name(ofproto->backer->dpif));
7902 /* xxx It would be better to show bridge-specific stats instead
7903 * xxx of dp ones. */
7905 "\tlookups: hit:%"PRIu64" missed:%"PRIu64" lost:%"PRIu64"\n",
7906 s.n_hit, s.n_missed, s.n_lost);
7907 ds_put_format(ds, "\tflows: %zu\n",
7908 hmap_count(&ofproto->subfacets));
7910 ports = shash_sort(&ofproto->up.port_by_name);
7911 for (i = 0; i < shash_count(&ofproto->up.port_by_name); i++) {
7912 const struct shash_node *node = ports[i];
7913 struct ofport *ofport = node->data;
7914 const char *name = netdev_get_name(ofport->netdev);
7915 const char *type = netdev_get_type(ofport->netdev);
7918 ds_put_format(ds, "\t%s %u/", name, ofport->ofp_port);
7920 odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
7921 if (odp_port != OVSP_NONE) {
7922 ds_put_format(ds, "%"PRIu32":", odp_port);
7924 ds_put_cstr(ds, "none:");
7927 if (strcmp(type, "system")) {
7928 struct netdev *netdev;
7931 ds_put_format(ds, " (%s", type);
7933 error = netdev_open(name, type, &netdev);
7938 error = netdev_get_config(netdev, &config);
7940 const struct smap_node **nodes;
7943 nodes = smap_sort(&config);
7944 for (i = 0; i < smap_count(&config); i++) {
7945 const struct smap_node *node = nodes[i];
7946 ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
7947 node->key, node->value);
7951 smap_destroy(&config);
7953 netdev_close(netdev);
7955 ds_put_char(ds, ')');
7957 ds_put_char(ds, '\n');
7963 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc,
7964 const char *argv[], void *aux OVS_UNUSED)
7966 struct ds ds = DS_EMPTY_INITIALIZER;
7967 const struct ofproto_dpif *ofproto;
7971 for (i = 1; i < argc; i++) {
7972 ofproto = ofproto_dpif_lookup(argv[i]);
7974 ds_put_format(&ds, "Unknown bridge %s (use dpif/dump-dps "
7975 "for help)", argv[i]);
7976 unixctl_command_reply_error(conn, ds_cstr(&ds));
7979 show_dp_format(ofproto, &ds);
7982 struct shash ofproto_shash;
7983 const struct shash_node **sorted_ofprotos;
7986 shash_init(&ofproto_shash);
7987 sorted_ofprotos = get_ofprotos(&ofproto_shash);
7988 for (i = 0; i < shash_count(&ofproto_shash); i++) {
7989 const struct shash_node *node = sorted_ofprotos[i];
7990 show_dp_format(node->data, &ds);
7993 shash_destroy(&ofproto_shash);
7994 free(sorted_ofprotos);
7997 unixctl_command_reply(conn, ds_cstr(&ds));
8002 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
8003 int argc OVS_UNUSED, const char *argv[],
8004 void *aux OVS_UNUSED)
8006 struct ds ds = DS_EMPTY_INITIALIZER;
8007 const struct ofproto_dpif *ofproto;
8008 struct subfacet *subfacet;
8010 ofproto = ofproto_dpif_lookup(argv[1]);
8012 unixctl_command_reply_error(conn, "no such bridge");
8016 update_stats(ofproto->backer);
8018 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
8019 struct odputil_keybuf keybuf;
8022 subfacet_get_key(subfacet, &keybuf, &key);
8023 odp_flow_key_format(key.data, key.size, &ds);
8025 ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:",
8026 subfacet->dp_packet_count, subfacet->dp_byte_count);
8027 if (subfacet->used) {
8028 ds_put_format(&ds, "%.3fs",
8029 (time_msec() - subfacet->used) / 1000.0);
8031 ds_put_format(&ds, "never");
8033 if (subfacet->facet->tcp_flags) {
8034 ds_put_cstr(&ds, ", flags:");
8035 packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags);
8038 ds_put_cstr(&ds, ", actions:");
8039 format_odp_actions(&ds, subfacet->actions, subfacet->actions_len);
8040 ds_put_char(&ds, '\n');
8043 unixctl_command_reply(conn, ds_cstr(&ds));
8048 ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn,
8049 int argc OVS_UNUSED, const char *argv[],
8050 void *aux OVS_UNUSED)
8052 struct ds ds = DS_EMPTY_INITIALIZER;
8053 struct ofproto_dpif *ofproto;
8055 ofproto = ofproto_dpif_lookup(argv[1]);
8057 unixctl_command_reply_error(conn, "no such bridge");
8061 flush(&ofproto->up);
8063 unixctl_command_reply(conn, ds_cstr(&ds));
8068 ofproto_dpif_unixctl_init(void)
8070 static bool registered;
8076 unixctl_command_register(
8078 "bridge {priority tun_id in_port mark packet | odp_flow [-generate]}",
8079 2, 6, ofproto_unixctl_trace, NULL);
8080 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
8081 ofproto_unixctl_fdb_flush, NULL);
8082 unixctl_command_register("fdb/show", "bridge", 1, 1,
8083 ofproto_unixctl_fdb_show, NULL);
8084 unixctl_command_register("ofproto/clog", "", 0, 0,
8085 ofproto_dpif_clog, NULL);
8086 unixctl_command_register("ofproto/unclog", "", 0, 0,
8087 ofproto_dpif_unclog, NULL);
8088 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
8089 ofproto_dpif_self_check, NULL);
8090 unixctl_command_register("dpif/dump-dps", "", 0, 0,
8091 ofproto_unixctl_dpif_dump_dps, NULL);
8092 unixctl_command_register("dpif/show", "[bridge]", 0, INT_MAX,
8093 ofproto_unixctl_dpif_show, NULL);
8094 unixctl_command_register("dpif/dump-flows", "bridge", 1, 1,
8095 ofproto_unixctl_dpif_dump_flows, NULL);
8096 unixctl_command_register("dpif/del-flows", "bridge", 1, 1,
8097 ofproto_unixctl_dpif_del_flows, NULL);
8100 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
8102 * This is deprecated. It is only for compatibility with broken device drivers
8103 * in old versions of Linux that do not properly support VLANs when VLAN
8104 * devices are not used. When broken device drivers are no longer in
8105 * widespread use, we will delete these interfaces. */
8108 set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid)
8110 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
8111 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
8113 if (realdev_ofp_port == ofport->realdev_ofp_port
8114 && vid == ofport->vlandev_vid) {
8118 ofproto->backer->need_revalidate = REV_RECONFIGURE;
8120 if (ofport->realdev_ofp_port) {
8123 if (realdev_ofp_port && ofport->bundle) {
8124 /* vlandevs are enslaved to their realdevs, so they are not allowed to
8125 * themselves be part of a bundle. */
8126 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
8129 ofport->realdev_ofp_port = realdev_ofp_port;
8130 ofport->vlandev_vid = vid;
8132 if (realdev_ofp_port) {
8133 vsp_add(ofport, realdev_ofp_port, vid);
8140 hash_realdev_vid(uint16_t realdev_ofp_port, int vid)
8142 return hash_2words(realdev_ofp_port, vid);
8145 /* Returns the ODP port number of the Linux VLAN device that corresponds to
8146 * 'vlan_tci' on the network device with port number 'realdev_odp_port' in
8147 * 'ofproto'. For example, given 'realdev_odp_port' of eth0 and 'vlan_tci' 9,
8148 * it would return the port number of eth0.9.
8150 * Unless VLAN splinters are enabled for port 'realdev_odp_port', this
8151 * function just returns its 'realdev_odp_port' argument. */
8153 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
8154 uint32_t realdev_odp_port, ovs_be16 vlan_tci)
8156 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
8157 uint16_t realdev_ofp_port;
8158 int vid = vlan_tci_to_vid(vlan_tci);
8159 const struct vlan_splinter *vsp;
8161 realdev_ofp_port = odp_port_to_ofp_port(ofproto, realdev_odp_port);
8162 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
8163 hash_realdev_vid(realdev_ofp_port, vid),
8164 &ofproto->realdev_vid_map) {
8165 if (vsp->realdev_ofp_port == realdev_ofp_port
8166 && vsp->vid == vid) {
8167 return ofp_port_to_odp_port(ofproto, vsp->vlandev_ofp_port);
8171 return realdev_odp_port;
8174 static struct vlan_splinter *
8175 vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port)
8177 struct vlan_splinter *vsp;
8179 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, hash_int(vlandev_ofp_port, 0),
8180 &ofproto->vlandev_map) {
8181 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
8189 /* Returns the OpenFlow port number of the "real" device underlying the Linux
8190 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
8191 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
8192 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
8193 * eth0 and store 9 in '*vid'.
8195 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
8196 * VLAN device. Unless VLAN splinters are enabled, this is what this function
8199 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
8200 uint16_t vlandev_ofp_port, int *vid)
8202 if (!hmap_is_empty(&ofproto->vlandev_map)) {
8203 const struct vlan_splinter *vsp;
8205 vsp = vlandev_find(ofproto, vlandev_ofp_port);
8210 return vsp->realdev_ofp_port;
8216 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
8217 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
8218 * 'flow->in_port' to the "real" device backing the VLAN device, sets
8219 * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is
8220 * always the case unless VLAN splinters are enabled), returns false without
8221 * making any changes. */
8223 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
8228 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port, &vid);
8233 /* Cause the flow to be processed as if it came in on the real device with
8234 * the VLAN device's VLAN ID. */
8235 flow->in_port = realdev;
8236 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
8241 vsp_remove(struct ofport_dpif *port)
8243 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
8244 struct vlan_splinter *vsp;
8246 vsp = vlandev_find(ofproto, port->up.ofp_port);
8248 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
8249 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
8252 port->realdev_ofp_port = 0;
8254 VLOG_ERR("missing vlan device record");
8259 vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid)
8261 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
8263 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
8264 && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid))
8265 == realdev_ofp_port)) {
8266 struct vlan_splinter *vsp;
8268 vsp = xmalloc(sizeof *vsp);
8269 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
8270 hash_int(port->up.ofp_port, 0));
8271 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
8272 hash_realdev_vid(realdev_ofp_port, vid));
8273 vsp->realdev_ofp_port = realdev_ofp_port;
8274 vsp->vlandev_ofp_port = port->up.ofp_port;
8277 port->realdev_ofp_port = realdev_ofp_port;
8279 VLOG_ERR("duplicate vlan device record");
8284 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
8286 const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
8287 return ofport ? ofport->odp_port : OVSP_NONE;
8290 static struct ofport_dpif *
8291 odp_port_to_ofport(const struct dpif_backer *backer, uint32_t odp_port)
8293 struct ofport_dpif *port;
8295 HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node,
8296 hash_int(odp_port, 0),
8297 &backer->odp_to_ofport_map) {
8298 if (port->odp_port == odp_port) {
8307 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
8309 struct ofport_dpif *port;
8311 port = odp_port_to_ofport(ofproto->backer, odp_port);
8312 if (port && &ofproto->up == port->up.ofproto) {
8313 return port->up.ofp_port;
8319 const struct ofproto_class ofproto_dpif_class = {
8354 port_is_lacp_current,
8355 NULL, /* rule_choose_table */
8362 rule_modify_actions,
8371 get_cfm_remote_mpids,
8376 get_stp_port_status,
8383 is_mirror_output_bundle,
8384 forward_bpdu_changed,
8385 set_mac_table_config,