2 * Copyright (c) 2009, 2010, 2011, 2012 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"
45 #include "ofp-actions.h"
46 #include "ofp-parse.h"
47 #include "ofp-print.h"
48 #include "ofproto-dpif-governor.h"
49 #include "ofproto-dpif-sflow.h"
50 #include "poll-loop.h"
54 #include "unaligned.h"
56 #include "vlan-bitmap.h"
59 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
61 COVERAGE_DEFINE(ofproto_dpif_expired);
62 COVERAGE_DEFINE(ofproto_dpif_xlate);
63 COVERAGE_DEFINE(facet_changed_rule);
64 COVERAGE_DEFINE(facet_revalidate);
65 COVERAGE_DEFINE(facet_unexpected);
66 COVERAGE_DEFINE(facet_suppress);
68 /* Maximum depth of flow table recursion (due to resubmit actions) in a
69 * flow translation. */
70 #define MAX_RESUBMIT_RECURSION 64
72 /* Number of implemented OpenFlow tables. */
73 enum { N_TABLES = 255 };
74 enum { TBL_INTERNAL = N_TABLES - 1 }; /* Used for internal hidden rules. */
75 BUILD_ASSERT_DECL(N_TABLES >= 2 && N_TABLES <= 255);
85 * - Do include packets and bytes from facets that have been deleted or
86 * whose own statistics have been folded into the rule.
88 * - Do include packets and bytes sent "by hand" that were accounted to
89 * the rule without any facet being involved (this is a rare corner
90 * case in rule_execute()).
92 * - Do not include packet or bytes that can be obtained from any facet's
93 * packet_count or byte_count member or that can be obtained from the
94 * datapath by, e.g., dpif_flow_get() for any subfacet.
96 uint64_t packet_count; /* Number of packets received. */
97 uint64_t byte_count; /* Number of bytes received. */
99 tag_type tag; /* Caches rule_calculate_tag() result. */
101 struct list facets; /* List of "struct facet"s. */
104 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
106 return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
109 static struct rule_dpif *rule_dpif_lookup(struct ofproto_dpif *,
110 const struct flow *);
111 static struct rule_dpif *rule_dpif_lookup__(struct ofproto_dpif *,
114 static struct rule_dpif *rule_dpif_miss_rule(struct ofproto_dpif *ofproto,
115 const struct flow *flow);
117 static void rule_credit_stats(struct rule_dpif *,
118 const struct dpif_flow_stats *);
119 static void flow_push_stats(struct rule_dpif *, const struct flow *,
120 const struct dpif_flow_stats *);
121 static tag_type rule_calculate_tag(const struct flow *,
122 const struct minimask *, uint32_t basis);
123 static void rule_invalidate(const struct rule_dpif *);
125 #define MAX_MIRRORS 32
126 typedef uint32_t mirror_mask_t;
127 #define MIRROR_MASK_C(X) UINT32_C(X)
128 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
130 struct ofproto_dpif *ofproto; /* Owning ofproto. */
131 size_t idx; /* In ofproto's "mirrors" array. */
132 void *aux; /* Key supplied by ofproto's client. */
133 char *name; /* Identifier for log messages. */
135 /* Selection criteria. */
136 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
137 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
138 unsigned long *vlans; /* Bitmap of chosen VLANs, NULL selects all. */
140 /* Output (exactly one of out == NULL and out_vlan == -1 is true). */
141 struct ofbundle *out; /* Output port or NULL. */
142 int out_vlan; /* Output VLAN or -1. */
143 mirror_mask_t dup_mirrors; /* Bitmap of mirrors with the same output. */
146 int64_t packet_count; /* Number of packets sent. */
147 int64_t byte_count; /* Number of bytes sent. */
150 static void mirror_destroy(struct ofmirror *);
151 static void update_mirror_stats(struct ofproto_dpif *ofproto,
152 mirror_mask_t mirrors,
153 uint64_t packets, uint64_t bytes);
156 struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
157 struct ofproto_dpif *ofproto; /* Owning ofproto. */
158 void *aux; /* Key supplied by ofproto's client. */
159 char *name; /* Identifier for log messages. */
162 struct list ports; /* Contains "struct ofport"s. */
163 enum port_vlan_mode vlan_mode; /* VLAN mode */
164 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
165 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
166 * NULL if all VLANs are trunked. */
167 struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */
168 struct bond *bond; /* Nonnull iff more than one port. */
169 bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */
172 bool floodable; /* True if no port has OFPUTIL_PC_NO_FLOOD set. */
174 /* Port mirroring info. */
175 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
176 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
177 mirror_mask_t mirror_out; /* Mirrors that output to this bundle. */
180 static void bundle_remove(struct ofport *);
181 static void bundle_update(struct ofbundle *);
182 static void bundle_destroy(struct ofbundle *);
183 static void bundle_del_port(struct ofport_dpif *);
184 static void bundle_run(struct ofbundle *);
185 static void bundle_wait(struct ofbundle *);
186 static struct ofbundle *lookup_input_bundle(const struct ofproto_dpif *,
187 uint16_t in_port, bool warn,
188 struct ofport_dpif **in_ofportp);
190 /* A controller may use OFPP_NONE as the ingress port to indicate that
191 * it did not arrive on a "real" port. 'ofpp_none_bundle' exists for
192 * when an input bundle is needed for validation (e.g., mirroring or
193 * OFPP_NORMAL processing). It is not connected to an 'ofproto' or have
194 * any 'port' structs, so care must be taken when dealing with it. */
195 static struct ofbundle ofpp_none_bundle = {
197 .vlan_mode = PORT_VLAN_TRUNK
200 static void stp_run(struct ofproto_dpif *ofproto);
201 static void stp_wait(struct ofproto_dpif *ofproto);
202 static int set_stp_port(struct ofport *,
203 const struct ofproto_port_stp_settings *);
205 static bool ofbundle_includes_vlan(const struct ofbundle *, uint16_t vlan);
207 struct action_xlate_ctx {
208 /* action_xlate_ctx_init() initializes these members. */
211 struct ofproto_dpif *ofproto;
213 /* Flow to which the OpenFlow actions apply. xlate_actions() will modify
214 * this flow when actions change header fields. */
217 /* The packet corresponding to 'flow', or a null pointer if we are
218 * revalidating without a packet to refer to. */
219 const struct ofpbuf *packet;
221 /* Should OFPP_NORMAL update the MAC learning table? Should "learn"
222 * actions update the flow table?
224 * We want to update these tables if we are actually processing a packet,
225 * or if we are accounting for packets that the datapath has processed, but
226 * not if we are just revalidating. */
229 /* The rule that we are currently translating, or NULL. */
230 struct rule_dpif *rule;
232 /* Union of the set of TCP flags seen so far in this flow. (Used only by
233 * NXAST_FIN_TIMEOUT. Set to zero to avoid updating updating rules'
237 /* If nonnull, flow translation calls this function just before executing a
238 * resubmit or OFPP_TABLE action. In addition, disables logging of traces
239 * when the recursion depth is exceeded.
241 * 'rule' is the rule being submitted into. It will be null if the
242 * resubmit or OFPP_TABLE action didn't find a matching rule.
244 * This is normally null so the client has to set it manually after
245 * calling action_xlate_ctx_init(). */
246 void (*resubmit_hook)(struct action_xlate_ctx *, struct rule_dpif *rule);
248 /* If nonnull, flow translation calls this function to report some
249 * significant decision, e.g. to explain why OFPP_NORMAL translation
250 * dropped a packet. */
251 void (*report_hook)(struct action_xlate_ctx *, const char *s);
253 /* If nonnull, flow translation credits the specified statistics to each
254 * rule reached through a resubmit or OFPP_TABLE action.
256 * This is normally null so the client has to set it manually after
257 * calling action_xlate_ctx_init(). */
258 const struct dpif_flow_stats *resubmit_stats;
260 /* xlate_actions() initializes and uses these members. The client might want
261 * to look at them after it returns. */
263 struct ofpbuf *odp_actions; /* Datapath actions. */
264 tag_type tags; /* Tags associated with actions. */
265 enum slow_path_reason slow; /* 0 if fast path may be used. */
266 bool has_learn; /* Actions include NXAST_LEARN? */
267 bool has_normal; /* Actions output to OFPP_NORMAL? */
268 bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */
269 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
270 mirror_mask_t mirrors; /* Bitmap of associated mirrors. */
272 /* xlate_actions() initializes and uses these members, but the client has no
273 * reason to look at them. */
275 int recurse; /* Recursion level, via xlate_table_action. */
276 bool max_resubmit_trigger; /* Recursed too deeply during translation. */
277 struct flow base_flow; /* Flow at the last commit. */
278 uint32_t orig_skb_priority; /* Priority when packet arrived. */
279 uint8_t table_id; /* OpenFlow table ID where flow was found. */
280 uint32_t sflow_n_outputs; /* Number of output ports. */
281 uint32_t sflow_odp_port; /* Output port for composing sFlow action. */
282 uint16_t user_cookie_offset;/* Used for user_action_cookie fixup. */
283 bool exit; /* No further actions should be processed. */
284 struct flow orig_flow; /* Copy of original flow. */
287 static void action_xlate_ctx_init(struct action_xlate_ctx *,
288 struct ofproto_dpif *, const struct flow *,
289 ovs_be16 initial_tci, struct rule_dpif *,
290 uint8_t tcp_flags, const struct ofpbuf *);
291 static void xlate_actions(struct action_xlate_ctx *,
292 const struct ofpact *ofpacts, size_t ofpacts_len,
293 struct ofpbuf *odp_actions);
294 static void xlate_actions_for_side_effects(struct action_xlate_ctx *,
295 const struct ofpact *ofpacts,
298 static size_t put_userspace_action(const struct ofproto_dpif *,
299 struct ofpbuf *odp_actions,
301 const union user_action_cookie *);
303 static void compose_slow_path(const struct ofproto_dpif *, const struct flow *,
304 enum slow_path_reason,
305 uint64_t *stub, size_t stub_size,
306 const struct nlattr **actionsp,
307 size_t *actions_lenp);
309 static void xlate_report(struct action_xlate_ctx *ctx, const char *s);
311 /* A subfacet (see "struct subfacet" below) has three possible installation
314 * - SF_NOT_INSTALLED: Not installed in the datapath. This will only be the
315 * case just after the subfacet is created, just before the subfacet is
316 * destroyed, or if the datapath returns an error when we try to install a
319 * - SF_FAST_PATH: The subfacet's actions are installed in the datapath.
321 * - SF_SLOW_PATH: An action that sends every packet for the subfacet through
322 * ofproto_dpif is installed in the datapath.
325 SF_NOT_INSTALLED, /* No datapath flow for this subfacet. */
326 SF_FAST_PATH, /* Full actions are installed. */
327 SF_SLOW_PATH, /* Send-to-userspace action is installed. */
330 static const char *subfacet_path_to_string(enum subfacet_path);
332 /* A dpif flow and actions associated with a facet.
334 * See also the large comment on struct facet. */
337 struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */
338 struct list list_node; /* In struct facet's 'facets' list. */
339 struct facet *facet; /* Owning facet. */
343 * To save memory in the common case, 'key' is NULL if 'key_fitness' is
344 * ODP_FIT_PERFECT, that is, odp_flow_key_from_flow() can accurately
345 * regenerate the ODP flow key from ->facet->flow. */
346 enum odp_key_fitness key_fitness;
350 long long int used; /* Time last used; time created if not used. */
352 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
353 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
357 * These should be essentially identical for every subfacet in a facet, but
358 * may differ in trivial ways due to VLAN splinters. */
359 size_t actions_len; /* Number of bytes in actions[]. */
360 struct nlattr *actions; /* Datapath actions. */
362 enum slow_path_reason slow; /* 0 if fast path may be used. */
363 enum subfacet_path path; /* Installed in datapath? */
365 /* This value is normally the same as ->facet->flow.vlan_tci. Only VLAN
366 * splinters can cause it to differ. This value should be removed when
367 * the VLAN splinters feature is no longer needed. */
368 ovs_be16 initial_tci; /* Initial VLAN TCI value. */
371 #define SUBFACET_DESTROY_MAX_BATCH 50
373 static struct subfacet *subfacet_create(struct facet *, enum odp_key_fitness,
374 const struct nlattr *key,
375 size_t key_len, ovs_be16 initial_tci,
377 static struct subfacet *subfacet_find(struct ofproto_dpif *,
378 const struct nlattr *key, size_t key_len,
380 const struct flow *flow);
381 static void subfacet_destroy(struct subfacet *);
382 static void subfacet_destroy__(struct subfacet *);
383 static void subfacet_destroy_batch(struct ofproto_dpif *,
384 struct subfacet **, int n);
385 static void subfacet_get_key(struct subfacet *, struct odputil_keybuf *,
387 static void subfacet_reset_dp_stats(struct subfacet *,
388 struct dpif_flow_stats *);
389 static void subfacet_update_time(struct subfacet *, long long int used);
390 static void subfacet_update_stats(struct subfacet *,
391 const struct dpif_flow_stats *);
392 static void subfacet_make_actions(struct subfacet *,
393 const struct ofpbuf *packet,
394 struct ofpbuf *odp_actions);
395 static int subfacet_install(struct subfacet *,
396 const struct nlattr *actions, size_t actions_len,
397 struct dpif_flow_stats *, enum slow_path_reason);
398 static void subfacet_uninstall(struct subfacet *);
400 static enum subfacet_path subfacet_want_path(enum slow_path_reason);
402 /* An exact-match instantiation of an OpenFlow flow.
404 * A facet associates a "struct flow", which represents the Open vSwitch
405 * userspace idea of an exact-match flow, with one or more subfacets. Each
406 * subfacet tracks the datapath's idea of the exact-match flow equivalent to
407 * the facet. When the kernel module (or other dpif implementation) and Open
408 * vSwitch userspace agree on the definition of a flow key, there is exactly
409 * one subfacet per facet. If the dpif implementation supports more-specific
410 * flow matching than userspace, however, a facet can have more than one
411 * subfacet, each of which corresponds to some distinction in flow that
412 * userspace simply doesn't understand.
414 * Flow expiration works in terms of subfacets, so a facet must have at least
415 * one subfacet or it will never expire, leaking memory. */
418 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
419 struct list list_node; /* In owning rule's 'facets' list. */
420 struct rule_dpif *rule; /* Owning rule. */
423 struct list subfacets;
424 long long int used; /* Time last used; time created if not used. */
431 * - Do include packets and bytes sent "by hand", e.g. with
434 * - Do include packets and bytes that were obtained from the datapath
435 * when a subfacet's statistics were reset (e.g. dpif_flow_put() with
436 * DPIF_FP_ZERO_STATS).
438 * - Do not include packets or bytes that can be obtained from the
439 * datapath for any existing subfacet.
441 uint64_t packet_count; /* Number of packets received. */
442 uint64_t byte_count; /* Number of bytes received. */
444 /* Resubmit statistics. */
445 uint64_t prev_packet_count; /* Number of packets from last stats push. */
446 uint64_t prev_byte_count; /* Number of bytes from last stats push. */
447 long long int prev_used; /* Used time from last stats push. */
450 uint64_t accounted_bytes; /* Bytes processed by facet_account(). */
451 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
452 uint8_t tcp_flags; /* TCP flags seen for this 'rule'. */
454 /* Properties of datapath actions.
456 * Every subfacet has its own actions because actions can differ slightly
457 * between splintered and non-splintered subfacets due to the VLAN tag
458 * being initially different (present vs. absent). All of them have these
459 * properties in common so we just store one copy of them here. */
460 bool has_learn; /* Actions include NXAST_LEARN? */
461 bool has_normal; /* Actions output to OFPP_NORMAL? */
462 bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */
463 tag_type tags; /* Tags that would require revalidation. */
464 mirror_mask_t mirrors; /* Bitmap of dependent mirrors. */
466 /* Storage for a single subfacet, to reduce malloc() time and space
467 * overhead. (A facet always has at least one subfacet and in the common
468 * case has exactly one subfacet.) */
469 struct subfacet one_subfacet;
472 static struct facet *facet_create(struct rule_dpif *,
473 const struct flow *, uint32_t hash);
474 static void facet_remove(struct facet *);
475 static void facet_free(struct facet *);
477 static struct facet *facet_find(struct ofproto_dpif *,
478 const struct flow *, uint32_t hash);
479 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
480 const struct flow *, uint32_t hash);
481 static void facet_revalidate(struct facet *);
482 static bool facet_check_consistency(struct facet *);
484 static void facet_flush_stats(struct facet *);
486 static void facet_update_time(struct facet *, long long int used);
487 static void facet_reset_counters(struct facet *);
488 static void facet_push_stats(struct facet *);
489 static void facet_learn(struct facet *);
490 static void facet_account(struct facet *);
492 static bool facet_is_controller_flow(struct facet *);
495 struct hmap_node odp_port_node; /* In dpif_backer's "odp_to_ofport_map". */
499 struct ofbundle *bundle; /* Bundle that contains this port, if any. */
500 struct list bundle_node; /* In struct ofbundle's "ports" list. */
501 struct cfm *cfm; /* Connectivity Fault Management, if any. */
502 tag_type tag; /* Tag associated with this port. */
503 uint32_t bond_stable_id; /* stable_id to use as bond slave, or 0. */
504 bool may_enable; /* May be enabled in bonds. */
505 long long int carrier_seq; /* Carrier status changes. */
508 struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */
509 enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */
510 long long int stp_state_entered;
512 struct hmap priorities; /* Map of attached 'priority_to_dscp's. */
514 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
516 * This is deprecated. It is only for compatibility with broken device
517 * drivers in old versions of Linux that do not properly support VLANs when
518 * VLAN devices are not used. When broken device drivers are no longer in
519 * widespread use, we will delete these interfaces. */
520 uint16_t realdev_ofp_port;
524 /* Node in 'ofport_dpif''s 'priorities' map. Used to maintain a map from
525 * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
526 * traffic egressing the 'ofport' with that priority should be marked with. */
527 struct priority_to_dscp {
528 struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'priorities' map. */
529 uint32_t priority; /* Priority of this queue (see struct flow). */
531 uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */
534 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
536 * This is deprecated. It is only for compatibility with broken device drivers
537 * in old versions of Linux that do not properly support VLANs when VLAN
538 * devices are not used. When broken device drivers are no longer in
539 * widespread use, we will delete these interfaces. */
540 struct vlan_splinter {
541 struct hmap_node realdev_vid_node;
542 struct hmap_node vlandev_node;
543 uint16_t realdev_ofp_port;
544 uint16_t vlandev_ofp_port;
548 static uint32_t vsp_realdev_to_vlandev(const struct ofproto_dpif *,
549 uint32_t realdev, ovs_be16 vlan_tci);
550 static bool vsp_adjust_flow(const struct ofproto_dpif *, struct flow *);
551 static void vsp_remove(struct ofport_dpif *);
552 static void vsp_add(struct ofport_dpif *, uint16_t realdev_ofp_port, int vid);
554 static uint32_t ofp_port_to_odp_port(const struct ofproto_dpif *,
556 static uint16_t odp_port_to_ofp_port(const struct ofproto_dpif *,
559 static struct ofport_dpif *
560 ofport_dpif_cast(const struct ofport *ofport)
562 assert(ofport->ofproto->ofproto_class == &ofproto_dpif_class);
563 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
566 static void port_run(struct ofport_dpif *);
567 static void port_run_fast(struct ofport_dpif *);
568 static void port_wait(struct ofport_dpif *);
569 static int set_cfm(struct ofport *, const struct cfm_settings *);
570 static void ofport_clear_priorities(struct ofport_dpif *);
572 struct dpif_completion {
573 struct list list_node;
574 struct ofoperation *op;
577 /* Extra information about a classifier table.
578 * Currently used just for optimized flow revalidation. */
580 /* If either of these is nonnull, then this table has a form that allows
581 * flows to be tagged to avoid revalidating most flows for the most common
582 * kinds of flow table changes. */
583 struct cls_table *catchall_table; /* Table that wildcards all fields. */
584 struct cls_table *other_table; /* Table with any other wildcard set. */
585 uint32_t basis; /* Keeps each table's tags separate. */
588 /* Reasons that we might need to revalidate every facet, and corresponding
591 * A value of 0 means that there is no need to revalidate.
593 * It would be nice to have some cleaner way to integrate with coverage
594 * counters, but with only a few reasons I guess this is good enough for
596 enum revalidate_reason {
597 REV_RECONFIGURE = 1, /* Switch configuration changed. */
598 REV_STP, /* Spanning tree protocol port status change. */
599 REV_PORT_TOGGLED, /* Port enabled or disabled by CFM, LACP, ...*/
600 REV_FLOW_TABLE, /* Flow table changed. */
601 REV_INCONSISTENCY /* Facet self-check failed. */
603 COVERAGE_DEFINE(rev_reconfigure);
604 COVERAGE_DEFINE(rev_stp);
605 COVERAGE_DEFINE(rev_port_toggled);
606 COVERAGE_DEFINE(rev_flow_table);
607 COVERAGE_DEFINE(rev_inconsistency);
609 /* All datapaths of a given type share a single dpif backer instance. */
614 struct timer next_expiration;
615 struct hmap odp_to_ofport_map; /* ODP port to ofport mapping. */
618 /* All existing ofproto_backer instances, indexed by ofproto->up.type. */
619 static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers);
621 static struct ofport_dpif *
622 odp_port_to_ofport(const struct dpif_backer *, uint32_t odp_port);
624 struct ofproto_dpif {
625 struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
627 struct dpif_backer *backer;
629 /* Special OpenFlow rules. */
630 struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */
631 struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */
637 struct netflow *netflow;
638 struct dpif_sflow *sflow;
639 struct hmap bundles; /* Contains "struct ofbundle"s. */
640 struct mac_learning *ml;
641 struct ofmirror *mirrors[MAX_MIRRORS];
643 bool has_bonded_bundles;
647 struct hmap subfacets;
648 struct governor *governor;
651 struct table_dpif tables[N_TABLES];
652 enum revalidate_reason need_revalidate;
653 struct tag_set revalidate_set;
655 /* Support for debugging async flow mods. */
656 struct list completions;
658 bool has_bundle_action; /* True when the first bundle action appears. */
659 struct netdev_stats stats; /* To account packets generated and consumed in
664 long long int stp_last_tick;
666 /* VLAN splinters. */
667 struct hmap realdev_vid_map; /* (realdev,vid) -> vlandev. */
668 struct hmap vlandev_map; /* vlandev -> (realdev,vid). */
671 struct sset ports; /* Set of port names. */
672 struct sset port_poll_set; /* Queued names for port_poll() reply. */
673 int port_poll_errno; /* Last errno for port_poll() reply. */
676 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
677 * for debugging the asynchronous flow_mod implementation.) */
680 /* All existing ofproto_dpif instances, indexed by ->up.name. */
681 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
683 static void ofproto_dpif_unixctl_init(void);
685 static struct ofproto_dpif *
686 ofproto_dpif_cast(const struct ofproto *ofproto)
688 assert(ofproto->ofproto_class == &ofproto_dpif_class);
689 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
692 static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *,
694 static struct ofport_dpif *get_odp_port(const struct ofproto_dpif *,
696 static void ofproto_trace(struct ofproto_dpif *, const struct flow *,
697 const struct ofpbuf *, ovs_be16 initial_tci,
700 /* Packet processing. */
701 static void update_learning_table(struct ofproto_dpif *,
702 const struct flow *, int vlan,
705 #define FLOW_MISS_MAX_BATCH 50
706 static int handle_upcalls(struct dpif_backer *, unsigned int max_batch);
708 /* Flow expiration. */
709 static int expire(struct dpif_backer *);
712 static void send_netflow_active_timeouts(struct ofproto_dpif *);
715 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
716 static size_t compose_sflow_action(const struct ofproto_dpif *,
717 struct ofpbuf *odp_actions,
718 const struct flow *, uint32_t odp_port);
719 static void add_mirror_actions(struct action_xlate_ctx *ctx,
720 const struct flow *flow);
721 /* Global variables. */
722 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
724 /* Initial mappings of port to bridge mappings. */
725 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
727 /* Factory functions. */
730 init(const struct shash *iface_hints)
732 struct shash_node *node;
734 /* Make a local copy, since we don't own 'iface_hints' elements. */
735 SHASH_FOR_EACH(node, iface_hints) {
736 const struct iface_hint *orig_hint = node->data;
737 struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
739 new_hint->br_name = xstrdup(orig_hint->br_name);
740 new_hint->br_type = xstrdup(orig_hint->br_type);
741 new_hint->ofp_port = orig_hint->ofp_port;
743 shash_add(&init_ofp_ports, node->name, new_hint);
748 enumerate_types(struct sset *types)
750 dp_enumerate_types(types);
754 enumerate_names(const char *type, struct sset *names)
756 struct ofproto_dpif *ofproto;
759 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
760 if (strcmp(type, ofproto->up.type)) {
763 sset_add(names, ofproto->up.name);
770 del(const char *type, const char *name)
775 error = dpif_open(name, type, &dpif);
777 error = dpif_delete(dpif);
784 port_open_type(const char *datapath_type, const char *port_type)
786 return dpif_port_open_type(datapath_type, port_type);
789 /* Type functions. */
792 type_run(const char *type)
794 struct dpif_backer *backer;
798 backer = shash_find_data(&all_dpif_backers, type);
800 /* This is not necessarily a problem, since backers are only
801 * created on demand. */
805 dpif_run(backer->dpif);
807 if (timer_expired(&backer->next_expiration)) {
808 int delay = expire(backer);
809 timer_set_duration(&backer->next_expiration, delay);
812 /* Check for port changes in the dpif. */
813 while ((error = dpif_port_poll(backer->dpif, &devname)) == 0) {
814 struct ofproto_dpif *ofproto = NULL;
815 struct dpif_port port;
817 /* Don't report on the datapath's device. */
818 if (!strcmp(devname, dpif_base_name(backer->dpif))) {
822 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
823 &all_ofproto_dpifs) {
824 if (sset_contains(&ofproto->ports, devname)) {
829 if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
830 /* The port was removed. If we know the datapath,
831 * report it through poll_set(). If we don't, it may be
832 * notifying us of a removal we initiated, so ignore it.
833 * If there's a pending ENOBUFS, let it stand, since
834 * everything will be reevaluated. */
835 if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
836 sset_add(&ofproto->port_poll_set, devname);
837 ofproto->port_poll_errno = 0;
839 dpif_port_destroy(&port);
840 } else if (!ofproto) {
841 /* The port was added, but we don't know with which
842 * ofproto we should associate it. Delete it. */
843 dpif_port_del(backer->dpif, port.port_no);
849 if (error != EAGAIN) {
850 struct ofproto_dpif *ofproto;
852 /* There was some sort of error, so propagate it to all
853 * ofprotos that use this backer. */
854 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
855 &all_ofproto_dpifs) {
856 if (ofproto->backer == backer) {
857 sset_clear(&ofproto->port_poll_set);
858 ofproto->port_poll_errno = error;
867 type_run_fast(const char *type)
869 struct dpif_backer *backer;
872 backer = shash_find_data(&all_dpif_backers, type);
874 /* This is not necessarily a problem, since backers are only
875 * created on demand. */
879 /* Handle one or more batches of upcalls, until there's nothing left to do
880 * or until we do a fixed total amount of work.
882 * We do work in batches because it can be much cheaper to set up a number
883 * of flows and fire off their patches all at once. We do multiple batches
884 * because in some cases handling a packet can cause another packet to be
885 * queued almost immediately as part of the return flow. Both
886 * optimizations can make major improvements on some benchmarks and
887 * presumably for real traffic as well. */
889 while (work < FLOW_MISS_MAX_BATCH) {
890 int retval = handle_upcalls(backer, FLOW_MISS_MAX_BATCH - work);
901 type_wait(const char *type)
903 struct dpif_backer *backer;
905 backer = shash_find_data(&all_dpif_backers, type);
907 /* This is not necessarily a problem, since backers are only
908 * created on demand. */
912 timer_wait(&backer->next_expiration);
915 /* Basic life-cycle. */
917 static int add_internal_flows(struct ofproto_dpif *);
919 static struct ofproto *
922 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
927 dealloc(struct ofproto *ofproto_)
929 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
934 close_dpif_backer(struct dpif_backer *backer)
936 struct shash_node *node;
938 assert(backer->refcount > 0);
940 if (--backer->refcount) {
944 hmap_destroy(&backer->odp_to_ofport_map);
945 node = shash_find(&all_dpif_backers, backer->type);
947 shash_delete(&all_dpif_backers, node);
948 dpif_close(backer->dpif);
953 /* Datapath port slated for removal from datapath. */
955 struct list list_node;
960 open_dpif_backer(const char *type, struct dpif_backer **backerp)
962 struct dpif_backer *backer;
963 struct dpif_port_dump port_dump;
964 struct dpif_port port;
965 struct shash_node *node;
966 struct list garbage_list;
967 struct odp_garbage *garbage, *next;
973 backer = shash_find_data(&all_dpif_backers, type);
980 backer_name = xasprintf("ovs-%s", type);
982 /* Remove any existing datapaths, since we assume we're the only
983 * userspace controlling the datapath. */
985 dp_enumerate_names(type, &names);
986 SSET_FOR_EACH(name, &names) {
987 struct dpif *old_dpif;
989 /* Don't remove our backer if it exists. */
990 if (!strcmp(name, backer_name)) {
994 if (dpif_open(name, type, &old_dpif)) {
995 VLOG_WARN("couldn't open old datapath %s to remove it", name);
997 dpif_delete(old_dpif);
998 dpif_close(old_dpif);
1001 sset_destroy(&names);
1003 backer = xmalloc(sizeof *backer);
1005 error = dpif_create_and_open(backer_name, type, &backer->dpif);
1008 VLOG_ERR("failed to open datapath of type %s: %s", type,
1013 backer->type = xstrdup(type);
1014 backer->refcount = 1;
1015 hmap_init(&backer->odp_to_ofport_map);
1016 timer_set_duration(&backer->next_expiration, 1000);
1019 dpif_flow_flush(backer->dpif);
1021 /* Loop through the ports already on the datapath and remove any
1022 * that we don't need anymore. */
1023 list_init(&garbage_list);
1024 dpif_port_dump_start(&port_dump, backer->dpif);
1025 while (dpif_port_dump_next(&port_dump, &port)) {
1026 node = shash_find(&init_ofp_ports, port.name);
1027 if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
1028 garbage = xmalloc(sizeof *garbage);
1029 garbage->odp_port = port.port_no;
1030 list_push_front(&garbage_list, &garbage->list_node);
1033 dpif_port_dump_done(&port_dump);
1035 LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) {
1036 dpif_port_del(backer->dpif, garbage->odp_port);
1037 list_remove(&garbage->list_node);
1041 shash_add(&all_dpif_backers, type, backer);
1043 error = dpif_recv_set(backer->dpif, true);
1045 VLOG_ERR("failed to listen on datapath of type %s: %s",
1046 type, strerror(error));
1047 close_dpif_backer(backer);
1055 construct(struct ofproto *ofproto_)
1057 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1058 struct shash_node *node, *next;
1063 error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1068 max_ports = dpif_get_max_ports(ofproto->backer->dpif);
1069 ofproto_init_max_ports(ofproto_, MIN(max_ports, OFPP_MAX));
1071 ofproto->n_matches = 0;
1073 ofproto->netflow = NULL;
1074 ofproto->sflow = NULL;
1075 ofproto->stp = NULL;
1076 hmap_init(&ofproto->bundles);
1077 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1078 for (i = 0; i < MAX_MIRRORS; i++) {
1079 ofproto->mirrors[i] = NULL;
1081 ofproto->has_bonded_bundles = false;
1083 hmap_init(&ofproto->facets);
1084 hmap_init(&ofproto->subfacets);
1085 ofproto->governor = NULL;
1087 for (i = 0; i < N_TABLES; i++) {
1088 struct table_dpif *table = &ofproto->tables[i];
1090 table->catchall_table = NULL;
1091 table->other_table = NULL;
1092 table->basis = random_uint32();
1094 ofproto->need_revalidate = 0;
1095 tag_set_init(&ofproto->revalidate_set);
1097 list_init(&ofproto->completions);
1099 ofproto_dpif_unixctl_init();
1101 ofproto->has_mirrors = false;
1102 ofproto->has_bundle_action = false;
1104 hmap_init(&ofproto->vlandev_map);
1105 hmap_init(&ofproto->realdev_vid_map);
1107 sset_init(&ofproto->ports);
1108 sset_init(&ofproto->port_poll_set);
1109 ofproto->port_poll_errno = 0;
1111 SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1112 const struct iface_hint *iface_hint = node->data;
1114 if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1115 /* Check if the datapath already has this port. */
1116 if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1117 sset_add(&ofproto->ports, node->name);
1120 free(iface_hint->br_name);
1121 free(iface_hint->br_type);
1122 shash_delete(&init_ofp_ports, node);
1126 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1127 hash_string(ofproto->up.name, 0));
1128 memset(&ofproto->stats, 0, sizeof ofproto->stats);
1130 ofproto_init_tables(ofproto_, N_TABLES);
1131 error = add_internal_flows(ofproto);
1132 ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1138 add_internal_flow(struct ofproto_dpif *ofproto, int id,
1139 const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1141 struct ofputil_flow_mod fm;
1144 match_init_catchall(&fm.match);
1146 match_set_reg(&fm.match, 0, id);
1147 fm.new_cookie = htonll(0);
1148 fm.cookie = htonll(0);
1149 fm.cookie_mask = htonll(0);
1150 fm.table_id = TBL_INTERNAL;
1151 fm.command = OFPFC_ADD;
1152 fm.idle_timeout = 0;
1153 fm.hard_timeout = 0;
1157 fm.ofpacts = ofpacts->data;
1158 fm.ofpacts_len = ofpacts->size;
1160 error = ofproto_flow_mod(&ofproto->up, &fm);
1162 VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
1163 id, ofperr_to_string(error));
1167 *rulep = rule_dpif_lookup__(ofproto, &fm.match.flow, TBL_INTERNAL);
1168 assert(*rulep != NULL);
1174 add_internal_flows(struct ofproto_dpif *ofproto)
1176 struct ofpact_controller *controller;
1177 uint64_t ofpacts_stub[128 / 8];
1178 struct ofpbuf ofpacts;
1182 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1185 controller = ofpact_put_CONTROLLER(&ofpacts);
1186 controller->max_len = UINT16_MAX;
1187 controller->controller_id = 0;
1188 controller->reason = OFPR_NO_MATCH;
1189 ofpact_pad(&ofpacts);
1191 error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule);
1196 ofpbuf_clear(&ofpacts);
1197 error = add_internal_flow(ofproto, id++, &ofpacts,
1198 &ofproto->no_packet_in_rule);
1203 complete_operations(struct ofproto_dpif *ofproto)
1205 struct dpif_completion *c, *next;
1207 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
1208 ofoperation_complete(c->op, 0);
1209 list_remove(&c->list_node);
1215 destruct(struct ofproto *ofproto_)
1217 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1218 struct rule_dpif *rule, *next_rule;
1219 struct oftable *table;
1222 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1223 complete_operations(ofproto);
1225 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1226 struct cls_cursor cursor;
1228 cls_cursor_init(&cursor, &table->cls, NULL);
1229 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1230 ofproto_rule_destroy(&rule->up);
1234 for (i = 0; i < MAX_MIRRORS; i++) {
1235 mirror_destroy(ofproto->mirrors[i]);
1238 netflow_destroy(ofproto->netflow);
1239 dpif_sflow_destroy(ofproto->sflow);
1240 hmap_destroy(&ofproto->bundles);
1241 mac_learning_destroy(ofproto->ml);
1243 hmap_destroy(&ofproto->facets);
1244 hmap_destroy(&ofproto->subfacets);
1245 governor_destroy(ofproto->governor);
1247 hmap_destroy(&ofproto->vlandev_map);
1248 hmap_destroy(&ofproto->realdev_vid_map);
1250 sset_destroy(&ofproto->ports);
1251 sset_destroy(&ofproto->port_poll_set);
1253 close_dpif_backer(ofproto->backer);
1257 run_fast(struct ofproto *ofproto_)
1259 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1260 struct ofport_dpif *ofport;
1262 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1263 port_run_fast(ofport);
1270 run(struct ofproto *ofproto_)
1272 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1273 struct ofport_dpif *ofport;
1274 struct ofbundle *bundle;
1278 complete_operations(ofproto);
1281 error = run_fast(ofproto_);
1286 if (ofproto->netflow) {
1287 if (netflow_run(ofproto->netflow)) {
1288 send_netflow_active_timeouts(ofproto);
1291 if (ofproto->sflow) {
1292 dpif_sflow_run(ofproto->sflow);
1295 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1298 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1303 mac_learning_run(ofproto->ml, &ofproto->revalidate_set);
1305 /* Now revalidate if there's anything to do. */
1306 if (ofproto->need_revalidate
1307 || !tag_set_is_empty(&ofproto->revalidate_set)) {
1308 struct tag_set revalidate_set = ofproto->revalidate_set;
1309 bool revalidate_all = ofproto->need_revalidate;
1310 struct facet *facet;
1312 switch (ofproto->need_revalidate) {
1313 case REV_RECONFIGURE: COVERAGE_INC(rev_reconfigure); break;
1314 case REV_STP: COVERAGE_INC(rev_stp); break;
1315 case REV_PORT_TOGGLED: COVERAGE_INC(rev_port_toggled); break;
1316 case REV_FLOW_TABLE: COVERAGE_INC(rev_flow_table); break;
1317 case REV_INCONSISTENCY: COVERAGE_INC(rev_inconsistency); break;
1320 /* Clear the revalidation flags. */
1321 tag_set_init(&ofproto->revalidate_set);
1322 ofproto->need_revalidate = 0;
1324 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
1326 || tag_set_intersects(&revalidate_set, facet->tags)) {
1327 facet_revalidate(facet);
1332 /* Check the consistency of a random facet, to aid debugging. */
1333 if (!hmap_is_empty(&ofproto->facets) && !ofproto->need_revalidate) {
1334 struct facet *facet;
1336 facet = CONTAINER_OF(hmap_random_node(&ofproto->facets),
1337 struct facet, hmap_node);
1338 if (!tag_set_intersects(&ofproto->revalidate_set, facet->tags)) {
1339 if (!facet_check_consistency(facet)) {
1340 ofproto->need_revalidate = REV_INCONSISTENCY;
1345 if (ofproto->governor) {
1348 governor_run(ofproto->governor);
1350 /* If the governor has shrunk to its minimum size and the number of
1351 * subfacets has dwindled, then drop the governor entirely.
1353 * For hysteresis, the number of subfacets to drop the governor is
1354 * smaller than the number needed to trigger its creation. */
1355 n_subfacets = hmap_count(&ofproto->subfacets);
1356 if (n_subfacets * 4 < ofproto->up.flow_eviction_threshold
1357 && governor_is_idle(ofproto->governor)) {
1358 governor_destroy(ofproto->governor);
1359 ofproto->governor = NULL;
1367 wait(struct ofproto *ofproto_)
1369 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1370 struct ofport_dpif *ofport;
1371 struct ofbundle *bundle;
1373 if (!clogged && !list_is_empty(&ofproto->completions)) {
1374 poll_immediate_wake();
1377 dpif_wait(ofproto->backer->dpif);
1378 dpif_recv_wait(ofproto->backer->dpif);
1379 if (ofproto->sflow) {
1380 dpif_sflow_wait(ofproto->sflow);
1382 if (!tag_set_is_empty(&ofproto->revalidate_set)) {
1383 poll_immediate_wake();
1385 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1388 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1389 bundle_wait(bundle);
1391 if (ofproto->netflow) {
1392 netflow_wait(ofproto->netflow);
1394 mac_learning_wait(ofproto->ml);
1396 if (ofproto->need_revalidate) {
1397 /* Shouldn't happen, but if it does just go around again. */
1398 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1399 poll_immediate_wake();
1401 if (ofproto->governor) {
1402 governor_wait(ofproto->governor);
1407 get_memory_usage(const struct ofproto *ofproto_, struct simap *usage)
1409 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1411 simap_increase(usage, "facets", hmap_count(&ofproto->facets));
1412 simap_increase(usage, "subfacets", hmap_count(&ofproto->subfacets));
1416 flush(struct ofproto *ofproto_)
1418 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1419 struct subfacet *subfacet, *next_subfacet;
1420 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
1424 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
1425 &ofproto->subfacets) {
1426 if (subfacet->path != SF_NOT_INSTALLED) {
1427 batch[n_batch++] = subfacet;
1428 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
1429 subfacet_destroy_batch(ofproto, batch, n_batch);
1433 subfacet_destroy(subfacet);
1438 subfacet_destroy_batch(ofproto, batch, n_batch);
1443 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1444 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1446 *arp_match_ip = true;
1447 *actions = (OFPUTIL_A_OUTPUT |
1448 OFPUTIL_A_SET_VLAN_VID |
1449 OFPUTIL_A_SET_VLAN_PCP |
1450 OFPUTIL_A_STRIP_VLAN |
1451 OFPUTIL_A_SET_DL_SRC |
1452 OFPUTIL_A_SET_DL_DST |
1453 OFPUTIL_A_SET_NW_SRC |
1454 OFPUTIL_A_SET_NW_DST |
1455 OFPUTIL_A_SET_NW_TOS |
1456 OFPUTIL_A_SET_TP_SRC |
1457 OFPUTIL_A_SET_TP_DST |
1462 get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots)
1464 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1465 struct dpif_dp_stats s;
1467 strcpy(ots->name, "classifier");
1469 dpif_get_dp_stats(ofproto->backer->dpif, &s);
1471 ots->lookup_count = htonll(s.n_hit + s.n_missed);
1472 ots->matched_count = htonll(s.n_hit + ofproto->n_matches);
1475 static struct ofport *
1478 struct ofport_dpif *port = xmalloc(sizeof *port);
1483 port_dealloc(struct ofport *port_)
1485 struct ofport_dpif *port = ofport_dpif_cast(port_);
1490 port_construct(struct ofport *port_)
1492 struct ofport_dpif *port = ofport_dpif_cast(port_);
1493 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1494 struct dpif_port dpif_port;
1497 ofproto->need_revalidate = REV_RECONFIGURE;
1498 port->bundle = NULL;
1500 port->tag = tag_create_random();
1501 port->may_enable = true;
1502 port->stp_port = NULL;
1503 port->stp_state = STP_DISABLED;
1504 hmap_init(&port->priorities);
1505 port->realdev_ofp_port = 0;
1506 port->vlandev_vid = 0;
1507 port->carrier_seq = netdev_get_carrier_resets(port->up.netdev);
1509 error = dpif_port_query_by_name(ofproto->backer->dpif,
1510 netdev_get_name(port->up.netdev),
1516 port->odp_port = dpif_port.port_no;
1518 /* Sanity-check that a mapping doesn't already exist. This
1519 * shouldn't happen. */
1520 if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1521 VLOG_ERR("port %s already has an OpenFlow port number\n",
1526 hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1527 hash_int(port->odp_port, 0));
1529 if (ofproto->sflow) {
1530 dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1537 port_destruct(struct ofport *port_)
1539 struct ofport_dpif *port = ofport_dpif_cast(port_);
1540 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1541 struct dpif_port dpif_port;
1543 if (!dpif_port_query_by_number(ofproto->backer->dpif,
1544 port->odp_port, &dpif_port)) {
1545 /* The underlying device is still there, so delete it. This
1546 * happens when the ofproto is being destroyed, since the caller
1547 * assumes that removal of attached ports will happen as part of
1549 dpif_port_del(ofproto->backer->dpif, port->odp_port);
1550 dpif_port_destroy(&dpif_port);
1553 sset_find_and_delete(&ofproto->ports, netdev_get_name(port->up.netdev));
1554 hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1555 ofproto->need_revalidate = REV_RECONFIGURE;
1556 bundle_remove(port_);
1557 set_cfm(port_, NULL);
1558 if (ofproto->sflow) {
1559 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1562 ofport_clear_priorities(port);
1563 hmap_destroy(&port->priorities);
1567 port_modified(struct ofport *port_)
1569 struct ofport_dpif *port = ofport_dpif_cast(port_);
1571 if (port->bundle && port->bundle->bond) {
1572 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1577 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1579 struct ofport_dpif *port = ofport_dpif_cast(port_);
1580 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1581 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1583 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1584 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1585 OFPUTIL_PC_NO_PACKET_IN)) {
1586 ofproto->need_revalidate = REV_RECONFIGURE;
1588 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1589 bundle_update(port->bundle);
1595 set_sflow(struct ofproto *ofproto_,
1596 const struct ofproto_sflow_options *sflow_options)
1598 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1599 struct dpif_sflow *ds = ofproto->sflow;
1601 if (sflow_options) {
1603 struct ofport_dpif *ofport;
1605 ds = ofproto->sflow = dpif_sflow_create();
1606 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1607 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1609 ofproto->need_revalidate = REV_RECONFIGURE;
1611 dpif_sflow_set_options(ds, sflow_options);
1614 dpif_sflow_destroy(ds);
1615 ofproto->need_revalidate = REV_RECONFIGURE;
1616 ofproto->sflow = NULL;
1623 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1625 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1632 struct ofproto_dpif *ofproto;
1634 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1635 ofproto->need_revalidate = REV_RECONFIGURE;
1636 ofport->cfm = cfm_create(netdev_get_name(ofport->up.netdev));
1639 if (cfm_configure(ofport->cfm, s)) {
1645 cfm_destroy(ofport->cfm);
1651 get_cfm_fault(const struct ofport *ofport_)
1653 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1655 return ofport->cfm ? cfm_get_fault(ofport->cfm) : -1;
1659 get_cfm_opup(const struct ofport *ofport_)
1661 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1663 return ofport->cfm ? cfm_get_opup(ofport->cfm) : -1;
1667 get_cfm_remote_mpids(const struct ofport *ofport_, const uint64_t **rmps,
1670 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1673 cfm_get_remote_mpids(ofport->cfm, rmps, n_rmps);
1681 get_cfm_health(const struct ofport *ofport_)
1683 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1685 return ofport->cfm ? cfm_get_health(ofport->cfm) : -1;
1688 /* Spanning Tree. */
1691 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
1693 struct ofproto_dpif *ofproto = ofproto_;
1694 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
1695 struct ofport_dpif *ofport;
1697 ofport = stp_port_get_aux(sp);
1699 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
1700 ofproto->up.name, port_num);
1702 struct eth_header *eth = pkt->l2;
1704 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
1705 if (eth_addr_is_zero(eth->eth_src)) {
1706 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
1707 "with unknown MAC", ofproto->up.name, port_num);
1709 send_packet(ofport, pkt);
1715 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
1717 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
1719 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1721 /* Only revalidate flows if the configuration changed. */
1722 if (!s != !ofproto->stp) {
1723 ofproto->need_revalidate = REV_RECONFIGURE;
1727 if (!ofproto->stp) {
1728 ofproto->stp = stp_create(ofproto_->name, s->system_id,
1729 send_bpdu_cb, ofproto);
1730 ofproto->stp_last_tick = time_msec();
1733 stp_set_bridge_id(ofproto->stp, s->system_id);
1734 stp_set_bridge_priority(ofproto->stp, s->priority);
1735 stp_set_hello_time(ofproto->stp, s->hello_time);
1736 stp_set_max_age(ofproto->stp, s->max_age);
1737 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
1739 struct ofport *ofport;
1741 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
1742 set_stp_port(ofport, NULL);
1745 stp_destroy(ofproto->stp);
1746 ofproto->stp = NULL;
1753 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
1755 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1759 s->bridge_id = stp_get_bridge_id(ofproto->stp);
1760 s->designated_root = stp_get_designated_root(ofproto->stp);
1761 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
1770 update_stp_port_state(struct ofport_dpif *ofport)
1772 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1773 enum stp_state state;
1775 /* Figure out new state. */
1776 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
1780 if (ofport->stp_state != state) {
1781 enum ofputil_port_state of_state;
1784 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
1785 netdev_get_name(ofport->up.netdev),
1786 stp_state_name(ofport->stp_state),
1787 stp_state_name(state));
1788 if (stp_learn_in_state(ofport->stp_state)
1789 != stp_learn_in_state(state)) {
1790 /* xxx Learning action flows should also be flushed. */
1791 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
1793 fwd_change = stp_forward_in_state(ofport->stp_state)
1794 != stp_forward_in_state(state);
1796 ofproto->need_revalidate = REV_STP;
1797 ofport->stp_state = state;
1798 ofport->stp_state_entered = time_msec();
1800 if (fwd_change && ofport->bundle) {
1801 bundle_update(ofport->bundle);
1804 /* Update the STP state bits in the OpenFlow port description. */
1805 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
1806 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
1807 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
1808 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
1809 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
1811 ofproto_port_set_state(&ofport->up, of_state);
1815 /* Configures STP on 'ofport_' using the settings defined in 's'. The
1816 * caller is responsible for assigning STP port numbers and ensuring
1817 * there are no duplicates. */
1819 set_stp_port(struct ofport *ofport_,
1820 const struct ofproto_port_stp_settings *s)
1822 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1823 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1824 struct stp_port *sp = ofport->stp_port;
1826 if (!s || !s->enable) {
1828 ofport->stp_port = NULL;
1829 stp_port_disable(sp);
1830 update_stp_port_state(ofport);
1833 } else if (sp && stp_port_no(sp) != s->port_num
1834 && ofport == stp_port_get_aux(sp)) {
1835 /* The port-id changed, so disable the old one if it's not
1836 * already in use by another port. */
1837 stp_port_disable(sp);
1840 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
1841 stp_port_enable(sp);
1843 stp_port_set_aux(sp, ofport);
1844 stp_port_set_priority(sp, s->priority);
1845 stp_port_set_path_cost(sp, s->path_cost);
1847 update_stp_port_state(ofport);
1853 get_stp_port_status(struct ofport *ofport_,
1854 struct ofproto_port_stp_status *s)
1856 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1857 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1858 struct stp_port *sp = ofport->stp_port;
1860 if (!ofproto->stp || !sp) {
1866 s->port_id = stp_port_get_id(sp);
1867 s->state = stp_port_get_state(sp);
1868 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
1869 s->role = stp_port_get_role(sp);
1870 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
1876 stp_run(struct ofproto_dpif *ofproto)
1879 long long int now = time_msec();
1880 long long int elapsed = now - ofproto->stp_last_tick;
1881 struct stp_port *sp;
1884 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
1885 ofproto->stp_last_tick = now;
1887 while (stp_get_changed_port(ofproto->stp, &sp)) {
1888 struct ofport_dpif *ofport = stp_port_get_aux(sp);
1891 update_stp_port_state(ofport);
1895 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
1896 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
1902 stp_wait(struct ofproto_dpif *ofproto)
1905 poll_timer_wait(1000);
1909 /* Returns true if STP should process 'flow'. */
1911 stp_should_process_flow(const struct flow *flow)
1913 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
1917 stp_process_packet(const struct ofport_dpif *ofport,
1918 const struct ofpbuf *packet)
1920 struct ofpbuf payload = *packet;
1921 struct eth_header *eth = payload.data;
1922 struct stp_port *sp = ofport->stp_port;
1924 /* Sink packets on ports that have STP disabled when the bridge has
1926 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
1930 /* Trim off padding on payload. */
1931 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
1932 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
1935 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
1936 stp_received_bpdu(sp, payload.data, payload.size);
1940 static struct priority_to_dscp *
1941 get_priority(const struct ofport_dpif *ofport, uint32_t priority)
1943 struct priority_to_dscp *pdscp;
1946 hash = hash_int(priority, 0);
1947 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) {
1948 if (pdscp->priority == priority) {
1956 ofport_clear_priorities(struct ofport_dpif *ofport)
1958 struct priority_to_dscp *pdscp, *next;
1960 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) {
1961 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
1967 set_queues(struct ofport *ofport_,
1968 const struct ofproto_port_queue *qdscp_list,
1971 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1972 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1973 struct hmap new = HMAP_INITIALIZER(&new);
1976 for (i = 0; i < n_qdscp; i++) {
1977 struct priority_to_dscp *pdscp;
1981 dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
1982 if (dpif_queue_to_priority(ofproto->backer->dpif, qdscp_list[i].queue,
1987 pdscp = get_priority(ofport, priority);
1989 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
1991 pdscp = xmalloc(sizeof *pdscp);
1992 pdscp->priority = priority;
1994 ofproto->need_revalidate = REV_RECONFIGURE;
1997 if (pdscp->dscp != dscp) {
1999 ofproto->need_revalidate = REV_RECONFIGURE;
2002 hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0));
2005 if (!hmap_is_empty(&ofport->priorities)) {
2006 ofport_clear_priorities(ofport);
2007 ofproto->need_revalidate = REV_RECONFIGURE;
2010 hmap_swap(&new, &ofport->priorities);
2018 /* Expires all MAC learning entries associated with 'bundle' and forces its
2019 * ofproto to revalidate every flow.
2021 * Normally MAC learning entries are removed only from the ofproto associated
2022 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2023 * are removed from every ofproto. When patch ports and SLB bonds are in use
2024 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2025 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2026 * with the host from which it migrated. */
2028 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2030 struct ofproto_dpif *ofproto = bundle->ofproto;
2031 struct mac_learning *ml = ofproto->ml;
2032 struct mac_entry *mac, *next_mac;
2034 ofproto->need_revalidate = REV_RECONFIGURE;
2035 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2036 if (mac->port.p == bundle) {
2038 struct ofproto_dpif *o;
2040 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2042 struct mac_entry *e;
2044 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan,
2047 tag_set_add(&o->revalidate_set, e->tag);
2048 mac_learning_expire(o->ml, e);
2054 mac_learning_expire(ml, mac);
2059 static struct ofbundle *
2060 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2062 struct ofbundle *bundle;
2064 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2065 &ofproto->bundles) {
2066 if (bundle->aux == aux) {
2073 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
2074 * ones that are found to 'bundles'. */
2076 bundle_lookup_multiple(struct ofproto_dpif *ofproto,
2077 void **auxes, size_t n_auxes,
2078 struct hmapx *bundles)
2082 hmapx_init(bundles);
2083 for (i = 0; i < n_auxes; i++) {
2084 struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
2086 hmapx_add(bundles, bundle);
2092 bundle_update(struct ofbundle *bundle)
2094 struct ofport_dpif *port;
2096 bundle->floodable = true;
2097 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2098 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2099 || !stp_forward_in_state(port->stp_state)) {
2100 bundle->floodable = false;
2107 bundle_del_port(struct ofport_dpif *port)
2109 struct ofbundle *bundle = port->bundle;
2111 bundle->ofproto->need_revalidate = REV_RECONFIGURE;
2113 list_remove(&port->bundle_node);
2114 port->bundle = NULL;
2117 lacp_slave_unregister(bundle->lacp, port);
2120 bond_slave_unregister(bundle->bond, port);
2123 bundle_update(bundle);
2127 bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port,
2128 struct lacp_slave_settings *lacp,
2129 uint32_t bond_stable_id)
2131 struct ofport_dpif *port;
2133 port = get_ofp_port(bundle->ofproto, ofp_port);
2138 if (port->bundle != bundle) {
2139 bundle->ofproto->need_revalidate = REV_RECONFIGURE;
2141 bundle_del_port(port);
2144 port->bundle = bundle;
2145 list_push_back(&bundle->ports, &port->bundle_node);
2146 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2147 || !stp_forward_in_state(port->stp_state)) {
2148 bundle->floodable = false;
2152 port->bundle->ofproto->need_revalidate = REV_RECONFIGURE;
2153 lacp_slave_register(bundle->lacp, port, lacp);
2156 port->bond_stable_id = bond_stable_id;
2162 bundle_destroy(struct ofbundle *bundle)
2164 struct ofproto_dpif *ofproto;
2165 struct ofport_dpif *port, *next_port;
2172 ofproto = bundle->ofproto;
2173 for (i = 0; i < MAX_MIRRORS; i++) {
2174 struct ofmirror *m = ofproto->mirrors[i];
2176 if (m->out == bundle) {
2178 } else if (hmapx_find_and_delete(&m->srcs, bundle)
2179 || hmapx_find_and_delete(&m->dsts, bundle)) {
2180 ofproto->need_revalidate = REV_RECONFIGURE;
2185 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2186 bundle_del_port(port);
2189 bundle_flush_macs(bundle, true);
2190 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2192 free(bundle->trunks);
2193 lacp_destroy(bundle->lacp);
2194 bond_destroy(bundle->bond);
2199 bundle_set(struct ofproto *ofproto_, void *aux,
2200 const struct ofproto_bundle_settings *s)
2202 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2203 bool need_flush = false;
2204 struct ofport_dpif *port;
2205 struct ofbundle *bundle;
2206 unsigned long *trunks;
2212 bundle_destroy(bundle_lookup(ofproto, aux));
2216 assert(s->n_slaves == 1 || s->bond != NULL);
2217 assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2219 bundle = bundle_lookup(ofproto, aux);
2221 bundle = xmalloc(sizeof *bundle);
2223 bundle->ofproto = ofproto;
2224 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2225 hash_pointer(aux, 0));
2227 bundle->name = NULL;
2229 list_init(&bundle->ports);
2230 bundle->vlan_mode = PORT_VLAN_TRUNK;
2232 bundle->trunks = NULL;
2233 bundle->use_priority_tags = s->use_priority_tags;
2234 bundle->lacp = NULL;
2235 bundle->bond = NULL;
2237 bundle->floodable = true;
2239 bundle->src_mirrors = 0;
2240 bundle->dst_mirrors = 0;
2241 bundle->mirror_out = 0;
2244 if (!bundle->name || strcmp(s->name, bundle->name)) {
2246 bundle->name = xstrdup(s->name);
2251 if (!bundle->lacp) {
2252 ofproto->need_revalidate = REV_RECONFIGURE;
2253 bundle->lacp = lacp_create();
2255 lacp_configure(bundle->lacp, s->lacp);
2257 lacp_destroy(bundle->lacp);
2258 bundle->lacp = NULL;
2261 /* Update set of ports. */
2263 for (i = 0; i < s->n_slaves; i++) {
2264 if (!bundle_add_port(bundle, s->slaves[i],
2265 s->lacp ? &s->lacp_slaves[i] : NULL,
2266 s->bond_stable_ids ? s->bond_stable_ids[i] : 0)) {
2270 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2271 struct ofport_dpif *next_port;
2273 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2274 for (i = 0; i < s->n_slaves; i++) {
2275 if (s->slaves[i] == port->up.ofp_port) {
2280 bundle_del_port(port);
2284 assert(list_size(&bundle->ports) <= s->n_slaves);
2286 if (list_is_empty(&bundle->ports)) {
2287 bundle_destroy(bundle);
2291 /* Set VLAN tagging mode */
2292 if (s->vlan_mode != bundle->vlan_mode
2293 || s->use_priority_tags != bundle->use_priority_tags) {
2294 bundle->vlan_mode = s->vlan_mode;
2295 bundle->use_priority_tags = s->use_priority_tags;
2300 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2301 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2303 if (vlan != bundle->vlan) {
2304 bundle->vlan = vlan;
2308 /* Get trunked VLANs. */
2309 switch (s->vlan_mode) {
2310 case PORT_VLAN_ACCESS:
2314 case PORT_VLAN_TRUNK:
2315 trunks = CONST_CAST(unsigned long *, s->trunks);
2318 case PORT_VLAN_NATIVE_UNTAGGED:
2319 case PORT_VLAN_NATIVE_TAGGED:
2320 if (vlan != 0 && (!s->trunks
2321 || !bitmap_is_set(s->trunks, vlan)
2322 || bitmap_is_set(s->trunks, 0))) {
2323 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2325 trunks = bitmap_clone(s->trunks, 4096);
2327 trunks = bitmap_allocate1(4096);
2329 bitmap_set1(trunks, vlan);
2330 bitmap_set0(trunks, 0);
2332 trunks = CONST_CAST(unsigned long *, s->trunks);
2339 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2340 free(bundle->trunks);
2341 if (trunks == s->trunks) {
2342 bundle->trunks = vlan_bitmap_clone(trunks);
2344 bundle->trunks = trunks;
2349 if (trunks != s->trunks) {
2354 if (!list_is_short(&bundle->ports)) {
2355 bundle->ofproto->has_bonded_bundles = true;
2357 if (bond_reconfigure(bundle->bond, s->bond)) {
2358 ofproto->need_revalidate = REV_RECONFIGURE;
2361 bundle->bond = bond_create(s->bond);
2362 ofproto->need_revalidate = REV_RECONFIGURE;
2365 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2366 bond_slave_register(bundle->bond, port, port->bond_stable_id,
2370 bond_destroy(bundle->bond);
2371 bundle->bond = NULL;
2374 /* If we changed something that would affect MAC learning, un-learn
2375 * everything on this port and force flow revalidation. */
2377 bundle_flush_macs(bundle, false);
2384 bundle_remove(struct ofport *port_)
2386 struct ofport_dpif *port = ofport_dpif_cast(port_);
2387 struct ofbundle *bundle = port->bundle;
2390 bundle_del_port(port);
2391 if (list_is_empty(&bundle->ports)) {
2392 bundle_destroy(bundle);
2393 } else if (list_is_short(&bundle->ports)) {
2394 bond_destroy(bundle->bond);
2395 bundle->bond = NULL;
2401 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2403 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2404 struct ofport_dpif *port = port_;
2405 uint8_t ea[ETH_ADDR_LEN];
2408 error = netdev_get_etheraddr(port->up.netdev, ea);
2410 struct ofpbuf packet;
2413 ofpbuf_init(&packet, 0);
2414 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2416 memcpy(packet_pdu, pdu, pdu_size);
2418 send_packet(port, &packet);
2419 ofpbuf_uninit(&packet);
2421 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2422 "%s (%s)", port->bundle->name,
2423 netdev_get_name(port->up.netdev), strerror(error));
2428 bundle_send_learning_packets(struct ofbundle *bundle)
2430 struct ofproto_dpif *ofproto = bundle->ofproto;
2431 int error, n_packets, n_errors;
2432 struct mac_entry *e;
2434 error = n_packets = n_errors = 0;
2435 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2436 if (e->port.p != bundle) {
2437 struct ofpbuf *learning_packet;
2438 struct ofport_dpif *port;
2442 /* The assignment to "port" is unnecessary but makes "grep"ing for
2443 * struct ofport_dpif more effective. */
2444 learning_packet = bond_compose_learning_packet(bundle->bond,
2448 ret = send_packet(port, learning_packet);
2449 ofpbuf_delete(learning_packet);
2459 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2460 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2461 "packets, last error was: %s",
2462 bundle->name, n_errors, n_packets, strerror(error));
2464 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2465 bundle->name, n_packets);
2470 bundle_run(struct ofbundle *bundle)
2473 lacp_run(bundle->lacp, send_pdu_cb);
2476 struct ofport_dpif *port;
2478 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2479 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2482 bond_run(bundle->bond, &bundle->ofproto->revalidate_set,
2483 lacp_status(bundle->lacp));
2484 if (bond_should_send_learning_packets(bundle->bond)) {
2485 bundle_send_learning_packets(bundle);
2491 bundle_wait(struct ofbundle *bundle)
2494 lacp_wait(bundle->lacp);
2497 bond_wait(bundle->bond);
2504 mirror_scan(struct ofproto_dpif *ofproto)
2508 for (idx = 0; idx < MAX_MIRRORS; idx++) {
2509 if (!ofproto->mirrors[idx]) {
2516 static struct ofmirror *
2517 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
2521 for (i = 0; i < MAX_MIRRORS; i++) {
2522 struct ofmirror *mirror = ofproto->mirrors[i];
2523 if (mirror && mirror->aux == aux) {
2531 /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */
2533 mirror_update_dups(struct ofproto_dpif *ofproto)
2537 for (i = 0; i < MAX_MIRRORS; i++) {
2538 struct ofmirror *m = ofproto->mirrors[i];
2541 m->dup_mirrors = MIRROR_MASK_C(1) << i;
2545 for (i = 0; i < MAX_MIRRORS; i++) {
2546 struct ofmirror *m1 = ofproto->mirrors[i];
2553 for (j = i + 1; j < MAX_MIRRORS; j++) {
2554 struct ofmirror *m2 = ofproto->mirrors[j];
2556 if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) {
2557 m1->dup_mirrors |= MIRROR_MASK_C(1) << j;
2558 m2->dup_mirrors |= m1->dup_mirrors;
2565 mirror_set(struct ofproto *ofproto_, void *aux,
2566 const struct ofproto_mirror_settings *s)
2568 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2569 mirror_mask_t mirror_bit;
2570 struct ofbundle *bundle;
2571 struct ofmirror *mirror;
2572 struct ofbundle *out;
2573 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
2574 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
2577 mirror = mirror_lookup(ofproto, aux);
2579 mirror_destroy(mirror);
2585 idx = mirror_scan(ofproto);
2587 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
2589 ofproto->up.name, MAX_MIRRORS, s->name);
2593 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
2594 mirror->ofproto = ofproto;
2597 mirror->out_vlan = -1;
2598 mirror->name = NULL;
2601 if (!mirror->name || strcmp(s->name, mirror->name)) {
2603 mirror->name = xstrdup(s->name);
2606 /* Get the new configuration. */
2607 if (s->out_bundle) {
2608 out = bundle_lookup(ofproto, s->out_bundle);
2610 mirror_destroy(mirror);
2616 out_vlan = s->out_vlan;
2618 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
2619 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
2621 /* If the configuration has not changed, do nothing. */
2622 if (hmapx_equals(&srcs, &mirror->srcs)
2623 && hmapx_equals(&dsts, &mirror->dsts)
2624 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
2625 && mirror->out == out
2626 && mirror->out_vlan == out_vlan)
2628 hmapx_destroy(&srcs);
2629 hmapx_destroy(&dsts);
2633 hmapx_swap(&srcs, &mirror->srcs);
2634 hmapx_destroy(&srcs);
2636 hmapx_swap(&dsts, &mirror->dsts);
2637 hmapx_destroy(&dsts);
2639 free(mirror->vlans);
2640 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
2643 mirror->out_vlan = out_vlan;
2645 /* Update bundles. */
2646 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2647 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
2648 if (hmapx_contains(&mirror->srcs, bundle)) {
2649 bundle->src_mirrors |= mirror_bit;
2651 bundle->src_mirrors &= ~mirror_bit;
2654 if (hmapx_contains(&mirror->dsts, bundle)) {
2655 bundle->dst_mirrors |= mirror_bit;
2657 bundle->dst_mirrors &= ~mirror_bit;
2660 if (mirror->out == bundle) {
2661 bundle->mirror_out |= mirror_bit;
2663 bundle->mirror_out &= ~mirror_bit;
2667 ofproto->need_revalidate = REV_RECONFIGURE;
2668 ofproto->has_mirrors = true;
2669 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
2670 mirror_update_dups(ofproto);
2676 mirror_destroy(struct ofmirror *mirror)
2678 struct ofproto_dpif *ofproto;
2679 mirror_mask_t mirror_bit;
2680 struct ofbundle *bundle;
2687 ofproto = mirror->ofproto;
2688 ofproto->need_revalidate = REV_RECONFIGURE;
2689 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
2691 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2692 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
2693 bundle->src_mirrors &= ~mirror_bit;
2694 bundle->dst_mirrors &= ~mirror_bit;
2695 bundle->mirror_out &= ~mirror_bit;
2698 hmapx_destroy(&mirror->srcs);
2699 hmapx_destroy(&mirror->dsts);
2700 free(mirror->vlans);
2702 ofproto->mirrors[mirror->idx] = NULL;
2706 mirror_update_dups(ofproto);
2708 ofproto->has_mirrors = false;
2709 for (i = 0; i < MAX_MIRRORS; i++) {
2710 if (ofproto->mirrors[i]) {
2711 ofproto->has_mirrors = true;
2718 mirror_get_stats(struct ofproto *ofproto_, void *aux,
2719 uint64_t *packets, uint64_t *bytes)
2721 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2722 struct ofmirror *mirror = mirror_lookup(ofproto, aux);
2725 *packets = *bytes = UINT64_MAX;
2729 *packets = mirror->packet_count;
2730 *bytes = mirror->byte_count;
2736 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2738 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2739 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2740 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
2746 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2748 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2749 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2750 return bundle && bundle->mirror_out != 0;
2754 forward_bpdu_changed(struct ofproto *ofproto_)
2756 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2757 ofproto->need_revalidate = REV_RECONFIGURE;
2761 set_mac_idle_time(struct ofproto *ofproto_, unsigned int idle_time)
2763 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2764 mac_learning_set_idle_time(ofproto->ml, idle_time);
2769 static struct ofport_dpif *
2770 get_ofp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
2772 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2773 return ofport ? ofport_dpif_cast(ofport) : NULL;
2776 static struct ofport_dpif *
2777 get_odp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
2779 return get_ofp_port(ofproto, odp_port_to_ofp_port(ofproto, odp_port));
2783 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
2784 struct ofproto_port *ofproto_port,
2785 struct dpif_port *dpif_port)
2787 ofproto_port->name = dpif_port->name;
2788 ofproto_port->type = dpif_port->type;
2789 ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
2793 port_run_fast(struct ofport_dpif *ofport)
2795 if (ofport->cfm && cfm_should_send_ccm(ofport->cfm)) {
2796 struct ofpbuf packet;
2798 ofpbuf_init(&packet, 0);
2799 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
2800 send_packet(ofport, &packet);
2801 ofpbuf_uninit(&packet);
2806 port_run(struct ofport_dpif *ofport)
2808 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
2809 bool carrier_changed = carrier_seq != ofport->carrier_seq;
2810 bool enable = netdev_get_carrier(ofport->up.netdev);
2812 ofport->carrier_seq = carrier_seq;
2814 port_run_fast(ofport);
2816 int cfm_opup = cfm_get_opup(ofport->cfm);
2818 cfm_run(ofport->cfm);
2819 enable = enable && !cfm_get_fault(ofport->cfm);
2821 if (cfm_opup >= 0) {
2822 enable = enable && cfm_opup;
2826 if (ofport->bundle) {
2827 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
2828 if (carrier_changed) {
2829 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
2833 if (ofport->may_enable != enable) {
2834 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2836 if (ofproto->has_bundle_action) {
2837 ofproto->need_revalidate = REV_PORT_TOGGLED;
2841 ofport->may_enable = enable;
2845 port_wait(struct ofport_dpif *ofport)
2848 cfm_wait(ofport->cfm);
2853 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
2854 struct ofproto_port *ofproto_port)
2856 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2857 struct dpif_port dpif_port;
2860 if (!sset_contains(&ofproto->ports, devname)) {
2863 error = dpif_port_query_by_name(ofproto->backer->dpif,
2864 devname, &dpif_port);
2866 ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
2872 port_add(struct ofproto *ofproto_, struct netdev *netdev)
2874 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2875 uint32_t odp_port = UINT32_MAX;
2878 error = dpif_port_add(ofproto->backer->dpif, netdev, &odp_port);
2880 sset_add(&ofproto->ports, netdev_get_name(netdev));
2886 port_del(struct ofproto *ofproto_, uint16_t ofp_port)
2888 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2889 uint32_t odp_port = ofp_port_to_odp_port(ofproto, ofp_port);
2892 if (odp_port != OFPP_NONE) {
2893 error = dpif_port_del(ofproto->backer->dpif, odp_port);
2896 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
2898 /* The caller is going to close ofport->up.netdev. If this is a
2899 * bonded port, then the bond is using that netdev, so remove it
2900 * from the bond. The client will need to reconfigure everything
2901 * after deleting ports, so then the slave will get re-added. */
2902 bundle_remove(&ofport->up);
2909 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
2911 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2914 error = netdev_get_stats(ofport->up.netdev, stats);
2916 if (!error && ofport->odp_port == OVSP_LOCAL) {
2917 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2919 /* ofproto->stats.tx_packets represents packets that we created
2920 * internally and sent to some port (e.g. packets sent with
2921 * send_packet()). Account for them as if they had come from
2922 * OFPP_LOCAL and got forwarded. */
2924 if (stats->rx_packets != UINT64_MAX) {
2925 stats->rx_packets += ofproto->stats.tx_packets;
2928 if (stats->rx_bytes != UINT64_MAX) {
2929 stats->rx_bytes += ofproto->stats.tx_bytes;
2932 /* ofproto->stats.rx_packets represents packets that were received on
2933 * some port and we processed internally and dropped (e.g. STP).
2934 * Account for them as if they had been forwarded to OFPP_LOCAL. */
2936 if (stats->tx_packets != UINT64_MAX) {
2937 stats->tx_packets += ofproto->stats.rx_packets;
2940 if (stats->tx_bytes != UINT64_MAX) {
2941 stats->tx_bytes += ofproto->stats.rx_bytes;
2948 /* Account packets for LOCAL port. */
2950 ofproto_update_local_port_stats(const struct ofproto *ofproto_,
2951 size_t tx_size, size_t rx_size)
2953 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2956 ofproto->stats.rx_packets++;
2957 ofproto->stats.rx_bytes += rx_size;
2960 ofproto->stats.tx_packets++;
2961 ofproto->stats.tx_bytes += tx_size;
2965 struct port_dump_state {
2971 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
2973 struct port_dump_state *state;
2975 *statep = state = xmalloc(sizeof *state);
2982 port_dump_next(const struct ofproto *ofproto_ OVS_UNUSED, void *state_,
2983 struct ofproto_port *port)
2985 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2986 struct port_dump_state *state = state_;
2987 struct sset_node *node;
2989 while ((node = sset_at_position(&ofproto->ports, &state->bucket,
2993 error = port_query_by_name(ofproto_, node->name, port);
2994 if (error != ENODEV) {
3003 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3005 struct port_dump_state *state = state_;
3012 port_poll(const struct ofproto *ofproto_, char **devnamep)
3014 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3016 if (ofproto->port_poll_errno) {
3017 int error = ofproto->port_poll_errno;
3018 ofproto->port_poll_errno = 0;
3022 if (sset_is_empty(&ofproto->port_poll_set)) {
3026 *devnamep = sset_pop(&ofproto->port_poll_set);
3031 port_poll_wait(const struct ofproto *ofproto_)
3033 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3034 dpif_port_poll_wait(ofproto->backer->dpif);
3038 port_is_lacp_current(const struct ofport *ofport_)
3040 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3041 return (ofport->bundle && ofport->bundle->lacp
3042 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3046 /* Upcall handling. */
3048 /* Flow miss batching.
3050 * Some dpifs implement operations faster when you hand them off in a batch.
3051 * To allow batching, "struct flow_miss" queues the dpif-related work needed
3052 * for a given flow. Each "struct flow_miss" corresponds to sending one or
3053 * more packets, plus possibly installing the flow in the dpif.
3055 * So far we only batch the operations that affect flow setup time the most.
3056 * It's possible to batch more than that, but the benefit might be minimal. */
3058 struct hmap_node hmap_node;
3059 struct ofproto_dpif *ofproto;
3061 enum odp_key_fitness key_fitness;
3062 const struct nlattr *key;
3064 ovs_be16 initial_tci;
3065 struct list packets;
3066 enum dpif_upcall_type upcall_type;
3069 struct flow_miss_op {
3070 struct dpif_op dpif_op;
3071 struct subfacet *subfacet; /* Subfacet */
3072 void *garbage; /* Pointer to pass to free(), NULL if none. */
3073 uint64_t stub[1024 / 8]; /* Temporary buffer. */
3076 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
3077 * OpenFlow controller as necessary according to their individual
3078 * configurations. */
3080 send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet,
3081 const struct flow *flow)
3083 struct ofputil_packet_in pin;
3085 pin.packet = packet->data;
3086 pin.packet_len = packet->size;
3087 pin.reason = OFPR_NO_MATCH;
3088 pin.controller_id = 0;
3093 pin.send_len = 0; /* not used for flow table misses */
3095 flow_get_metadata(flow, &pin.fmd);
3097 connmgr_send_packet_in(ofproto->up.connmgr, &pin);
3100 static enum slow_path_reason
3101 process_special(struct ofproto_dpif *ofproto, const struct flow *flow,
3102 const struct ofpbuf *packet)
3104 struct ofport_dpif *ofport = get_ofp_port(ofproto, flow->in_port);
3110 if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) {
3112 cfm_process_heartbeat(ofport->cfm, packet);
3115 } else if (ofport->bundle && ofport->bundle->lacp
3116 && flow->dl_type == htons(ETH_TYPE_LACP)) {
3118 lacp_process_packet(ofport->bundle->lacp, ofport, packet);
3121 } else if (ofproto->stp && stp_should_process_flow(flow)) {
3123 stp_process_packet(ofport, packet);
3130 static struct flow_miss *
3131 flow_miss_find(struct hmap *todo, const struct flow *flow, uint32_t hash)
3133 struct flow_miss *miss;
3135 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
3136 if (flow_equal(&miss->flow, flow)) {
3144 /* Partially Initializes 'op' as an "execute" operation for 'miss' and
3145 * 'packet'. The caller must initialize op->actions and op->actions_len. If
3146 * 'miss' is associated with a subfacet the caller must also initialize the
3147 * returned op->subfacet, and if anything needs to be freed after processing
3148 * the op, the caller must initialize op->garbage also. */
3150 init_flow_miss_execute_op(struct flow_miss *miss, struct ofpbuf *packet,
3151 struct flow_miss_op *op)
3153 if (miss->flow.vlan_tci != miss->initial_tci) {
3154 /* This packet was received on a VLAN splinter port. We
3155 * added a VLAN to the packet to make the packet resemble
3156 * the flow, but the actions were composed assuming that
3157 * the packet contained no VLAN. So, we must remove the
3158 * VLAN header from the packet before trying to execute the
3160 eth_pop_vlan(packet);
3163 op->subfacet = NULL;
3165 op->dpif_op.type = DPIF_OP_EXECUTE;
3166 op->dpif_op.u.execute.key = miss->key;
3167 op->dpif_op.u.execute.key_len = miss->key_len;
3168 op->dpif_op.u.execute.packet = packet;
3171 /* Helper for handle_flow_miss_without_facet() and
3172 * handle_flow_miss_with_facet(). */
3174 handle_flow_miss_common(struct rule_dpif *rule,
3175 struct ofpbuf *packet, const struct flow *flow)
3177 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3179 ofproto->n_matches++;
3181 if (rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
3183 * Extra-special case for fail-open mode.
3185 * We are in fail-open mode and the packet matched the fail-open
3186 * rule, but we are connected to a controller too. We should send
3187 * the packet up to the controller in the hope that it will try to
3188 * set up a flow and thereby allow us to exit fail-open.
3190 * See the top-level comment in fail-open.c for more information.
3192 send_packet_in_miss(ofproto, packet, flow);
3196 /* Figures out whether a flow that missed in 'ofproto', whose details are in
3197 * 'miss', is likely to be worth tracking in detail in userspace and (usually)
3198 * installing a datapath flow. The answer is usually "yes" (a return value of
3199 * true). However, for short flows the cost of bookkeeping is much higher than
3200 * the benefits, so when the datapath holds a large number of flows we impose
3201 * some heuristics to decide which flows are likely to be worth tracking. */
3203 flow_miss_should_make_facet(struct ofproto_dpif *ofproto,
3204 struct flow_miss *miss, uint32_t hash)
3206 if (!ofproto->governor) {
3209 n_subfacets = hmap_count(&ofproto->subfacets);
3210 if (n_subfacets * 2 <= ofproto->up.flow_eviction_threshold) {
3214 ofproto->governor = governor_create(ofproto->up.name);
3217 return governor_should_install_flow(ofproto->governor, hash,
3218 list_size(&miss->packets));
3221 /* Handles 'miss', which matches 'rule', without creating a facet or subfacet
3222 * or creating any datapath flow. May add an "execute" operation to 'ops' and
3223 * increment '*n_ops'. */
3225 handle_flow_miss_without_facet(struct flow_miss *miss,
3226 struct rule_dpif *rule,
3227 struct flow_miss_op *ops, size_t *n_ops)
3229 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3230 long long int now = time_msec();
3231 struct action_xlate_ctx ctx;
3232 struct ofpbuf *packet;
3234 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3235 struct flow_miss_op *op = &ops[*n_ops];
3236 struct dpif_flow_stats stats;
3237 struct ofpbuf odp_actions;
3239 COVERAGE_INC(facet_suppress);
3241 ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub);
3243 dpif_flow_stats_extract(&miss->flow, packet, now, &stats);
3244 rule_credit_stats(rule, &stats);
3246 action_xlate_ctx_init(&ctx, ofproto, &miss->flow, miss->initial_tci,
3248 ctx.resubmit_stats = &stats;
3249 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len,
3252 if (odp_actions.size) {
3253 struct dpif_execute *execute = &op->dpif_op.u.execute;
3255 init_flow_miss_execute_op(miss, packet, op);
3256 execute->actions = odp_actions.data;
3257 execute->actions_len = odp_actions.size;
3258 op->garbage = ofpbuf_get_uninit_pointer(&odp_actions);
3262 ofpbuf_uninit(&odp_actions);
3267 /* Handles 'miss', which matches 'facet'. May add any required datapath
3268 * operations to 'ops', incrementing '*n_ops' for each new op.
3270 * All of the packets in 'miss' are considered to have arrived at time 'now'.
3271 * This is really important only for new facets: if we just called time_msec()
3272 * here, then the new subfacet or its packets could look (occasionally) as
3273 * though it was used some time after the facet was used. That can make a
3274 * one-packet flow look like it has a nonzero duration, which looks odd in
3275 * e.g. NetFlow statistics. */
3277 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
3279 struct flow_miss_op *ops, size_t *n_ops)
3281 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3282 enum subfacet_path want_path;
3283 struct subfacet *subfacet;
3284 struct ofpbuf *packet;
3286 subfacet = subfacet_create(facet,
3287 miss->key_fitness, miss->key, miss->key_len,
3288 miss->initial_tci, now);
3290 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3291 struct flow_miss_op *op = &ops[*n_ops];
3292 struct dpif_flow_stats stats;
3293 struct ofpbuf odp_actions;
3295 handle_flow_miss_common(facet->rule, packet, &miss->flow);
3297 ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub);
3298 if (!subfacet->actions || subfacet->slow) {
3299 subfacet_make_actions(subfacet, packet, &odp_actions);
3302 dpif_flow_stats_extract(&facet->flow, packet, now, &stats);
3303 subfacet_update_stats(subfacet, &stats);
3305 if (subfacet->actions_len) {
3306 struct dpif_execute *execute = &op->dpif_op.u.execute;
3308 init_flow_miss_execute_op(miss, packet, op);
3309 op->subfacet = subfacet;
3310 if (!subfacet->slow) {
3311 execute->actions = subfacet->actions;
3312 execute->actions_len = subfacet->actions_len;
3313 ofpbuf_uninit(&odp_actions);
3315 execute->actions = odp_actions.data;
3316 execute->actions_len = odp_actions.size;
3317 op->garbage = ofpbuf_get_uninit_pointer(&odp_actions);
3322 ofpbuf_uninit(&odp_actions);
3326 want_path = subfacet_want_path(subfacet->slow);
3327 if (miss->upcall_type == DPIF_UC_MISS || subfacet->path != want_path) {
3328 struct flow_miss_op *op = &ops[(*n_ops)++];
3329 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
3331 op->subfacet = subfacet;
3333 op->dpif_op.type = DPIF_OP_FLOW_PUT;
3334 put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
3335 put->key = miss->key;
3336 put->key_len = miss->key_len;
3337 if (want_path == SF_FAST_PATH) {
3338 put->actions = subfacet->actions;
3339 put->actions_len = subfacet->actions_len;
3341 compose_slow_path(ofproto, &facet->flow, subfacet->slow,
3342 op->stub, sizeof op->stub,
3343 &put->actions, &put->actions_len);
3349 /* Handles flow miss 'miss'. May add any required datapath operations
3350 * to 'ops', incrementing '*n_ops' for each new op. */
3352 handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops,
3355 struct ofproto_dpif *ofproto = miss->ofproto;
3356 struct facet *facet;
3360 /* The caller must ensure that miss->hmap_node.hash contains
3361 * flow_hash(miss->flow, 0). */
3362 hash = miss->hmap_node.hash;
3364 facet = facet_lookup_valid(ofproto, &miss->flow, hash);
3366 struct rule_dpif *rule = rule_dpif_lookup(ofproto, &miss->flow);
3368 if (!flow_miss_should_make_facet(ofproto, miss, hash)) {
3369 handle_flow_miss_without_facet(miss, rule, ops, n_ops);
3373 facet = facet_create(rule, &miss->flow, hash);
3378 handle_flow_miss_with_facet(miss, facet, now, ops, n_ops);
3381 /* This function does post-processing on data returned from
3382 * odp_flow_key_to_flow() to help make VLAN splinters transparent to the
3383 * rest of the upcall processing logic. In particular, if the extracted
3384 * in_port is a VLAN splinter port, it replaces flow->in_port by the "real"
3385 * port, sets flow->vlan_tci correctly for the VLAN of the VLAN splinter
3386 * port, and pushes a VLAN header onto 'packet' (if it is nonnull). The
3387 * caller must have called odp_flow_key_to_flow() and supply 'fitness' and
3388 * 'flow' from its output. The 'flow' argument must have had the "in_port"
3389 * member converted to the OpenFlow number.
3391 * Sets '*initial_tci' to the VLAN TCI with which the packet was really
3392 * received, that is, the actual VLAN TCI extracted by odp_flow_key_to_flow().
3393 * (This differs from the value returned in flow->vlan_tci only for packets
3394 * received on VLAN splinters.) */
3395 static enum odp_key_fitness
3396 ofproto_dpif_vsp_adjust(const struct ofproto_dpif *ofproto,
3397 enum odp_key_fitness fitness,
3398 struct flow *flow, ovs_be16 *initial_tci,
3399 struct ofpbuf *packet)
3401 if (fitness == ODP_FIT_ERROR) {
3404 *initial_tci = flow->vlan_tci;
3406 if (vsp_adjust_flow(ofproto, flow)) {
3408 /* Make the packet resemble the flow, so that it gets sent to an
3409 * OpenFlow controller properly, so that it looks correct for
3410 * sFlow, and so that flow_extract() will get the correct vlan_tci
3411 * if it is called on 'packet'.
3413 * The allocated space inside 'packet' probably also contains
3414 * 'key', that is, both 'packet' and 'key' are probably part of a
3415 * struct dpif_upcall (see the large comment on that structure
3416 * definition), so pushing data on 'packet' is in general not a
3417 * good idea since it could overwrite 'key' or free it as a side
3418 * effect. However, it's OK in this special case because we know
3419 * that 'packet' is inside a Netlink attribute: pushing 4 bytes
3420 * will just overwrite the 4-byte "struct nlattr", which is fine
3421 * since we don't need that header anymore. */
3422 eth_push_vlan(packet, flow->vlan_tci);
3425 /* Let the caller know that we can't reproduce 'key' from 'flow'. */
3426 if (fitness == ODP_FIT_PERFECT) {
3427 fitness = ODP_FIT_TOO_MUCH;
3435 handle_miss_upcalls(struct dpif_backer *backer, struct dpif_upcall *upcalls,
3438 struct dpif_upcall *upcall;
3439 struct flow_miss *miss;
3440 struct flow_miss misses[FLOW_MISS_MAX_BATCH];
3441 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
3442 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
3452 /* Construct the to-do list.
3454 * This just amounts to extracting the flow from each packet and sticking
3455 * the packets that have the same flow in the same "flow_miss" structure so
3456 * that we can process them together. */
3459 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
3460 struct flow_miss *miss = &misses[n_misses];
3461 struct flow_miss *existing_miss;
3462 enum odp_key_fitness fitness;
3463 struct ofproto_dpif *ofproto;
3464 struct ofport_dpif *port;
3468 fitness = odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
3469 port = odp_port_to_ofport(backer, flow.in_port);
3471 /* Received packet on port for which we couldn't associate
3472 * an ofproto. This can happen if a port is removed while
3473 * traffic is being received. Print a rate-limited message
3474 * in case it happens frequently. */
3475 VLOG_INFO_RL(&rl, "received packet on unassociated port %"PRIu32,
3479 ofproto = ofproto_dpif_cast(port->up.ofproto);
3480 flow.in_port = port->up.ofp_port;
3482 /* Obtain metadata and check userspace/kernel agreement on flow match,
3483 * then set 'flow''s header pointers. */
3484 miss->key_fitness = ofproto_dpif_vsp_adjust(ofproto, fitness,
3485 &flow, &miss->initial_tci, upcall->packet);
3486 if (miss->key_fitness == ODP_FIT_ERROR) {
3489 flow_extract(upcall->packet, flow.skb_priority,
3490 &flow.tunnel, flow.in_port, &miss->flow);
3492 /* Add other packets to a to-do list. */
3493 hash = flow_hash(&miss->flow, 0);
3494 existing_miss = flow_miss_find(&todo, &miss->flow, hash);
3495 if (!existing_miss) {
3496 hmap_insert(&todo, &miss->hmap_node, hash);
3497 miss->ofproto = ofproto;
3498 miss->key = upcall->key;
3499 miss->key_len = upcall->key_len;
3500 miss->upcall_type = upcall->type;
3501 list_init(&miss->packets);
3505 miss = existing_miss;
3507 list_push_back(&miss->packets, &upcall->packet->list_node);
3510 /* Process each element in the to-do list, constructing the set of
3511 * operations to batch. */
3513 HMAP_FOR_EACH (miss, hmap_node, &todo) {
3514 handle_flow_miss(miss, flow_miss_ops, &n_ops);
3516 assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
3518 /* Execute batch. */
3519 for (i = 0; i < n_ops; i++) {
3520 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
3522 dpif_operate(backer->dpif, dpif_ops, n_ops);
3524 /* Free memory and update facets. */
3525 for (i = 0; i < n_ops; i++) {
3526 struct flow_miss_op *op = &flow_miss_ops[i];
3528 switch (op->dpif_op.type) {
3529 case DPIF_OP_EXECUTE:
3532 case DPIF_OP_FLOW_PUT:
3533 if (!op->dpif_op.error) {
3534 op->subfacet->path = subfacet_want_path(op->subfacet->slow);
3538 case DPIF_OP_FLOW_DEL:
3544 hmap_destroy(&todo);
3547 static enum { SFLOW_UPCALL, MISS_UPCALL, BAD_UPCALL }
3548 classify_upcall(const struct dpif_upcall *upcall)
3550 union user_action_cookie cookie;
3552 /* First look at the upcall type. */
3553 switch (upcall->type) {
3554 case DPIF_UC_ACTION:
3560 case DPIF_N_UC_TYPES:
3562 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
3566 /* "action" upcalls need a closer look. */
3567 memcpy(&cookie, &upcall->userdata, sizeof(cookie));
3568 switch (cookie.type) {
3569 case USER_ACTION_COOKIE_SFLOW:
3570 return SFLOW_UPCALL;
3572 case USER_ACTION_COOKIE_SLOW_PATH:
3575 case USER_ACTION_COOKIE_UNSPEC:
3577 VLOG_WARN_RL(&rl, "invalid user cookie : 0x%"PRIx64, upcall->userdata);
3583 handle_sflow_upcall(struct dpif_backer *backer,
3584 const struct dpif_upcall *upcall)
3586 struct ofproto_dpif *ofproto;
3587 union user_action_cookie cookie;
3588 enum odp_key_fitness fitness;
3589 struct ofport_dpif *port;
3590 ovs_be16 initial_tci;
3592 uint32_t odp_in_port;
3594 fitness = odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
3596 port = odp_port_to_ofport(backer, flow.in_port);
3601 ofproto = ofproto_dpif_cast(port->up.ofproto);
3602 if (!ofproto->sflow) {
3606 odp_in_port = flow.in_port;
3607 flow.in_port = port->up.ofp_port;
3608 fitness = ofproto_dpif_vsp_adjust(ofproto, fitness, &flow,
3609 &initial_tci, upcall->packet);
3610 if (fitness == ODP_FIT_ERROR) {
3614 memcpy(&cookie, &upcall->userdata, sizeof(cookie));
3615 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
3616 odp_in_port, &cookie);
3620 handle_upcalls(struct dpif_backer *backer, unsigned int max_batch)
3622 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
3623 struct ofpbuf miss_bufs[FLOW_MISS_MAX_BATCH];
3624 uint64_t miss_buf_stubs[FLOW_MISS_MAX_BATCH][4096 / 8];
3629 assert(max_batch <= FLOW_MISS_MAX_BATCH);
3632 for (n_processed = 0; n_processed < max_batch; n_processed++) {
3633 struct dpif_upcall *upcall = &misses[n_misses];
3634 struct ofpbuf *buf = &miss_bufs[n_misses];
3637 ofpbuf_use_stub(buf, miss_buf_stubs[n_misses],
3638 sizeof miss_buf_stubs[n_misses]);
3639 error = dpif_recv(backer->dpif, upcall, buf);
3645 switch (classify_upcall(upcall)) {
3647 /* Handle it later. */
3652 handle_sflow_upcall(backer, upcall);
3662 /* Handle deferred MISS_UPCALL processing. */
3663 handle_miss_upcalls(backer, misses, n_misses);
3664 for (i = 0; i < n_misses; i++) {
3665 ofpbuf_uninit(&miss_bufs[i]);
3671 /* Flow expiration. */
3673 static int subfacet_max_idle(const struct ofproto_dpif *);
3674 static void update_stats(struct dpif_backer *);
3675 static void rule_expire(struct rule_dpif *);
3676 static void expire_subfacets(struct ofproto_dpif *, int dp_max_idle);
3678 /* This function is called periodically by run(). Its job is to collect
3679 * updates for the flows that have been installed into the datapath, most
3680 * importantly when they last were used, and then use that information to
3681 * expire flows that have not been used recently.
3683 * Returns the number of milliseconds after which it should be called again. */
3685 expire(struct dpif_backer *backer)
3687 struct ofproto_dpif *ofproto;
3688 int max_idle = INT32_MAX;
3690 /* Update stats for each flow in the backer. */
3691 update_stats(backer);
3693 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
3694 struct rule_dpif *rule, *next_rule;
3695 struct oftable *table;
3698 if (ofproto->backer != backer) {
3702 /* Expire subfacets that have been idle too long. */
3703 dp_max_idle = subfacet_max_idle(ofproto);
3704 expire_subfacets(ofproto, dp_max_idle);
3706 max_idle = MIN(max_idle, dp_max_idle);
3708 /* Expire OpenFlow flows whose idle_timeout or hard_timeout
3710 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
3711 struct cls_cursor cursor;
3713 cls_cursor_init(&cursor, &table->cls, NULL);
3714 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
3719 /* All outstanding data in existing flows has been accounted, so it's a
3720 * good time to do bond rebalancing. */
3721 if (ofproto->has_bonded_bundles) {
3722 struct ofbundle *bundle;
3724 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
3726 bond_rebalance(bundle->bond, &ofproto->revalidate_set);
3732 return MIN(max_idle, 1000);
3735 /* Updates flow table statistics given that the datapath just reported 'stats'
3736 * as 'subfacet''s statistics. */
3738 update_subfacet_stats(struct subfacet *subfacet,
3739 const struct dpif_flow_stats *stats)
3741 struct facet *facet = subfacet->facet;
3743 if (stats->n_packets >= subfacet->dp_packet_count) {
3744 uint64_t extra = stats->n_packets - subfacet->dp_packet_count;
3745 facet->packet_count += extra;
3747 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
3750 if (stats->n_bytes >= subfacet->dp_byte_count) {
3751 facet->byte_count += stats->n_bytes - subfacet->dp_byte_count;
3753 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
3756 subfacet->dp_packet_count = stats->n_packets;
3757 subfacet->dp_byte_count = stats->n_bytes;
3759 facet->tcp_flags |= stats->tcp_flags;
3761 subfacet_update_time(subfacet, stats->used);
3762 if (facet->accounted_bytes < facet->byte_count) {
3764 facet_account(facet);
3765 facet->accounted_bytes = facet->byte_count;
3767 facet_push_stats(facet);
3770 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
3771 * about, or a flow that shouldn't be installed but was anyway. Delete it. */
3773 delete_unexpected_flow(struct ofproto_dpif *ofproto,
3774 const struct nlattr *key, size_t key_len)
3776 if (!VLOG_DROP_WARN(&rl)) {
3780 odp_flow_key_format(key, key_len, &s);
3781 VLOG_WARN("unexpected flow on %s: %s", ofproto->up.name, ds_cstr(&s));
3785 COVERAGE_INC(facet_unexpected);
3786 dpif_flow_del(ofproto->backer->dpif, key, key_len, NULL);
3789 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
3791 * This function also pushes statistics updates to rules which each facet
3792 * resubmits into. Generally these statistics will be accurate. However, if a
3793 * facet changes the rule it resubmits into at some time in between
3794 * update_stats() runs, it is possible that statistics accrued to the
3795 * old rule will be incorrectly attributed to the new rule. This could be
3796 * avoided by calling update_stats() whenever rules are created or
3797 * deleted. However, the performance impact of making so many calls to the
3798 * datapath do not justify the benefit of having perfectly accurate statistics.
3801 update_stats(struct dpif_backer *backer)
3803 const struct dpif_flow_stats *stats;
3804 struct dpif_flow_dump dump;
3805 const struct nlattr *key;
3808 dpif_flow_dump_start(&dump, backer->dpif);
3809 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
3811 struct subfacet *subfacet;
3812 enum odp_key_fitness fitness;
3813 struct ofproto_dpif *ofproto;
3814 struct ofport_dpif *port;
3817 fitness = odp_flow_key_to_flow(key, key_len, &flow);
3818 if (fitness == ODP_FIT_ERROR) {
3822 port = odp_port_to_ofport(backer, flow.in_port);
3824 /* This flow is for a port for which we couldn't associate an
3825 * ofproto. This can happen if a port is removed while
3826 * traffic is being received. Print a rate-limited message
3827 * in case it happens frequently. */
3829 "stats update for flow with unassociated port %"PRIu32,
3834 ofproto = ofproto_dpif_cast(port->up.ofproto);
3835 flow.in_port = port->up.ofp_port;
3836 key_hash = odp_flow_key_hash(key, key_len);
3838 subfacet = subfacet_find(ofproto, key, key_len, key_hash, &flow);
3839 switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
3841 update_subfacet_stats(subfacet, stats);
3845 /* Stats are updated per-packet. */
3848 case SF_NOT_INSTALLED:
3850 delete_unexpected_flow(ofproto, key, key_len);
3854 dpif_flow_dump_done(&dump);
3857 /* Calculates and returns the number of milliseconds of idle time after which
3858 * subfacets should expire from the datapath. When a subfacet expires, we fold
3859 * its statistics into its facet, and when a facet's last subfacet expires, we
3860 * fold its statistic into its rule. */
3862 subfacet_max_idle(const struct ofproto_dpif *ofproto)
3865 * Idle time histogram.
3867 * Most of the time a switch has a relatively small number of subfacets.
3868 * When this is the case we might as well keep statistics for all of them
3869 * in userspace and to cache them in the kernel datapath for performance as
3872 * As the number of subfacets increases, the memory required to maintain
3873 * statistics about them in userspace and in the kernel becomes
3874 * significant. However, with a large number of subfacets it is likely
3875 * that only a few of them are "heavy hitters" that consume a large amount
3876 * of bandwidth. At this point, only heavy hitters are worth caching in
3877 * the kernel and maintaining in userspaces; other subfacets we can
3880 * The technique used to compute the idle time is to build a histogram with
3881 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
3882 * that is installed in the kernel gets dropped in the appropriate bucket.
3883 * After the histogram has been built, we compute the cutoff so that only
3884 * the most-recently-used 1% of subfacets (but at least
3885 * ofproto->up.flow_eviction_threshold flows) are kept cached. At least
3886 * the most-recently-used bucket of subfacets is kept, so actually an
3887 * arbitrary number of subfacets can be kept in any given expiration run
3888 * (though the next run will delete most of those unless they receive
3891 * This requires a second pass through the subfacets, in addition to the
3892 * pass made by update_stats(), because the former function never looks at
3893 * uninstallable subfacets.
3895 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
3896 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
3897 int buckets[N_BUCKETS] = { 0 };
3898 int total, subtotal, bucket;
3899 struct subfacet *subfacet;
3903 total = hmap_count(&ofproto->subfacets);
3904 if (total <= ofproto->up.flow_eviction_threshold) {
3905 return N_BUCKETS * BUCKET_WIDTH;
3908 /* Build histogram. */
3910 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
3911 long long int idle = now - subfacet->used;
3912 int bucket = (idle <= 0 ? 0
3913 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
3914 : (unsigned int) idle / BUCKET_WIDTH);
3918 /* Find the first bucket whose flows should be expired. */
3919 subtotal = bucket = 0;
3921 subtotal += buckets[bucket++];
3922 } while (bucket < N_BUCKETS &&
3923 subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100));
3925 if (VLOG_IS_DBG_ENABLED()) {
3929 ds_put_cstr(&s, "keep");
3930 for (i = 0; i < N_BUCKETS; i++) {
3932 ds_put_cstr(&s, ", drop");
3935 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
3938 VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s));
3942 return bucket * BUCKET_WIDTH;
3946 expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle)
3948 /* Cutoff time for most flows. */
3949 long long int normal_cutoff = time_msec() - dp_max_idle;
3951 /* We really want to keep flows for special protocols around, so use a more
3952 * conservative cutoff. */
3953 long long int special_cutoff = time_msec() - 10000;
3955 struct subfacet *subfacet, *next_subfacet;
3956 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
3960 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
3961 &ofproto->subfacets) {
3962 long long int cutoff;
3964 cutoff = (subfacet->slow & (SLOW_CFM | SLOW_LACP | SLOW_STP)
3967 if (subfacet->used < cutoff) {
3968 if (subfacet->path != SF_NOT_INSTALLED) {
3969 batch[n_batch++] = subfacet;
3970 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
3971 subfacet_destroy_batch(ofproto, batch, n_batch);
3975 subfacet_destroy(subfacet);
3981 subfacet_destroy_batch(ofproto, batch, n_batch);
3985 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
3986 * then delete it entirely. */
3988 rule_expire(struct rule_dpif *rule)
3990 struct facet *facet, *next_facet;
3994 if (rule->up.pending) {
3995 /* We'll have to expire it later. */
3999 /* Has 'rule' expired? */
4001 if (rule->up.hard_timeout
4002 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
4003 reason = OFPRR_HARD_TIMEOUT;
4004 } else if (rule->up.idle_timeout
4005 && now > rule->up.used + rule->up.idle_timeout * 1000) {
4006 reason = OFPRR_IDLE_TIMEOUT;
4011 COVERAGE_INC(ofproto_dpif_expired);
4013 /* Update stats. (This is a no-op if the rule expired due to an idle
4014 * timeout, because that only happens when the rule has no facets left.) */
4015 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4016 facet_remove(facet);
4019 /* Get rid of the rule. */
4020 ofproto_rule_expire(&rule->up, reason);
4025 /* Creates and returns a new facet owned by 'rule', given a 'flow'.
4027 * The caller must already have determined that no facet with an identical
4028 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
4029 * the ofproto's classifier table.
4031 * 'hash' must be the return value of flow_hash(flow, 0).
4033 * The facet will initially have no subfacets. The caller should create (at
4034 * least) one subfacet with subfacet_create(). */
4035 static struct facet *
4036 facet_create(struct rule_dpif *rule, const struct flow *flow, uint32_t hash)
4038 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4039 struct facet *facet;
4041 facet = xzalloc(sizeof *facet);
4042 facet->used = time_msec();
4043 hmap_insert(&ofproto->facets, &facet->hmap_node, hash);
4044 list_push_back(&rule->facets, &facet->list_node);
4046 facet->flow = *flow;
4047 list_init(&facet->subfacets);
4048 netflow_flow_init(&facet->nf_flow);
4049 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
4055 facet_free(struct facet *facet)
4060 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
4061 * 'packet', which arrived on 'in_port'.
4063 * Takes ownership of 'packet'. */
4065 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
4066 const struct nlattr *odp_actions, size_t actions_len,
4067 struct ofpbuf *packet)
4069 struct odputil_keybuf keybuf;
4073 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4074 odp_flow_key_from_flow(&key, flow,
4075 ofp_port_to_odp_port(ofproto, flow->in_port));
4077 error = dpif_execute(ofproto->backer->dpif, key.data, key.size,
4078 odp_actions, actions_len, packet);
4080 ofpbuf_delete(packet);
4084 /* Remove 'facet' from 'ofproto' and free up the associated memory:
4086 * - If 'facet' was installed in the datapath, uninstalls it and updates its
4087 * rule's statistics, via subfacet_uninstall().
4089 * - Removes 'facet' from its rule and from ofproto->facets.
4092 facet_remove(struct facet *facet)
4094 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4095 struct subfacet *subfacet, *next_subfacet;
4097 assert(!list_is_empty(&facet->subfacets));
4099 /* First uninstall all of the subfacets to get final statistics. */
4100 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4101 subfacet_uninstall(subfacet);
4104 /* Flush the final stats to the rule.
4106 * This might require us to have at least one subfacet around so that we
4107 * can use its actions for accounting in facet_account(), which is why we
4108 * have uninstalled but not yet destroyed the subfacets. */
4109 facet_flush_stats(facet);
4111 /* Now we're really all done so destroy everything. */
4112 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
4113 &facet->subfacets) {
4114 subfacet_destroy__(subfacet);
4116 hmap_remove(&ofproto->facets, &facet->hmap_node);
4117 list_remove(&facet->list_node);
4121 /* Feed information from 'facet' back into the learning table to keep it in
4122 * sync with what is actually flowing through the datapath. */
4124 facet_learn(struct facet *facet)
4126 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4127 struct action_xlate_ctx ctx;
4129 if (!facet->has_learn
4130 && !facet->has_normal
4131 && (!facet->has_fin_timeout
4132 || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
4136 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
4137 facet->flow.vlan_tci,
4138 facet->rule, facet->tcp_flags, NULL);
4139 ctx.may_learn = true;
4140 xlate_actions_for_side_effects(&ctx, facet->rule->up.ofpacts,
4141 facet->rule->up.ofpacts_len);
4145 facet_account(struct facet *facet)
4147 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4148 struct subfacet *subfacet;
4149 const struct nlattr *a;
4154 if (!facet->has_normal || !ofproto->has_bonded_bundles) {
4157 n_bytes = facet->byte_count - facet->accounted_bytes;
4159 /* This loop feeds byte counters to bond_account() for rebalancing to use
4160 * as a basis. We also need to track the actual VLAN on which the packet
4161 * is going to be sent to ensure that it matches the one passed to
4162 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
4165 * We use the actions from an arbitrary subfacet because they should all
4166 * be equally valid for our purpose. */
4167 subfacet = CONTAINER_OF(list_front(&facet->subfacets),
4168 struct subfacet, list_node);
4169 vlan_tci = facet->flow.vlan_tci;
4170 NL_ATTR_FOR_EACH_UNSAFE (a, left,
4171 subfacet->actions, subfacet->actions_len) {
4172 const struct ovs_action_push_vlan *vlan;
4173 struct ofport_dpif *port;
4175 switch (nl_attr_type(a)) {
4176 case OVS_ACTION_ATTR_OUTPUT:
4177 port = get_odp_port(ofproto, nl_attr_get_u32(a));
4178 if (port && port->bundle && port->bundle->bond) {
4179 bond_account(port->bundle->bond, &facet->flow,
4180 vlan_tci_to_vid(vlan_tci), n_bytes);
4184 case OVS_ACTION_ATTR_POP_VLAN:
4185 vlan_tci = htons(0);
4188 case OVS_ACTION_ATTR_PUSH_VLAN:
4189 vlan = nl_attr_get(a);
4190 vlan_tci = vlan->vlan_tci;
4196 /* Returns true if the only action for 'facet' is to send to the controller.
4197 * (We don't report NetFlow expiration messages for such facets because they
4198 * are just part of the control logic for the network, not real traffic). */
4200 facet_is_controller_flow(struct facet *facet)
4203 const struct rule *rule = &facet->rule->up;
4204 const struct ofpact *ofpacts = rule->ofpacts;
4205 size_t ofpacts_len = rule->ofpacts_len;
4207 if (ofpacts_len > 0 &&
4208 ofpacts->type == OFPACT_CONTROLLER &&
4209 ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len)) {
4216 /* Folds all of 'facet''s statistics into its rule. Also updates the
4217 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
4218 * 'facet''s statistics in the datapath should have been zeroed and folded into
4219 * its packet and byte counts before this function is called. */
4221 facet_flush_stats(struct facet *facet)
4223 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4224 struct subfacet *subfacet;
4226 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4227 assert(!subfacet->dp_byte_count);
4228 assert(!subfacet->dp_packet_count);
4231 facet_push_stats(facet);
4232 if (facet->accounted_bytes < facet->byte_count) {
4233 facet_account(facet);
4234 facet->accounted_bytes = facet->byte_count;
4237 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
4238 struct ofexpired expired;
4239 expired.flow = facet->flow;
4240 expired.packet_count = facet->packet_count;
4241 expired.byte_count = facet->byte_count;
4242 expired.used = facet->used;
4243 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4246 facet->rule->packet_count += facet->packet_count;
4247 facet->rule->byte_count += facet->byte_count;
4249 /* Reset counters to prevent double counting if 'facet' ever gets
4251 facet_reset_counters(facet);
4253 netflow_flow_clear(&facet->nf_flow);
4254 facet->tcp_flags = 0;
4257 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
4258 * Returns it if found, otherwise a null pointer.
4260 * 'hash' must be the return value of flow_hash(flow, 0).
4262 * The returned facet might need revalidation; use facet_lookup_valid()
4263 * instead if that is important. */
4264 static struct facet *
4265 facet_find(struct ofproto_dpif *ofproto,
4266 const struct flow *flow, uint32_t hash)
4268 struct facet *facet;
4270 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, hash, &ofproto->facets) {
4271 if (flow_equal(flow, &facet->flow)) {
4279 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
4280 * Returns it if found, otherwise a null pointer.
4282 * 'hash' must be the return value of flow_hash(flow, 0).
4284 * The returned facet is guaranteed to be valid. */
4285 static struct facet *
4286 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow,
4289 struct facet *facet;
4291 facet = facet_find(ofproto, flow, hash);
4293 && (ofproto->need_revalidate
4294 || tag_set_intersects(&ofproto->revalidate_set, facet->tags))) {
4295 facet_revalidate(facet);
4302 subfacet_path_to_string(enum subfacet_path path)
4305 case SF_NOT_INSTALLED:
4306 return "not installed";
4308 return "in fast path";
4310 return "in slow path";
4316 /* Returns the path in which a subfacet should be installed if its 'slow'
4317 * member has the specified value. */
4318 static enum subfacet_path
4319 subfacet_want_path(enum slow_path_reason slow)
4321 return slow ? SF_SLOW_PATH : SF_FAST_PATH;
4324 /* Returns true if 'subfacet' needs to have its datapath flow updated,
4325 * supposing that its actions have been recalculated as 'want_actions' and that
4326 * 'slow' is nonzero iff 'subfacet' should be in the slow path. */
4328 subfacet_should_install(struct subfacet *subfacet, enum slow_path_reason slow,
4329 const struct ofpbuf *want_actions)
4331 enum subfacet_path want_path = subfacet_want_path(slow);
4332 return (want_path != subfacet->path
4333 || (want_path == SF_FAST_PATH
4334 && (subfacet->actions_len != want_actions->size
4335 || memcmp(subfacet->actions, want_actions->data,
4336 subfacet->actions_len))));
4340 facet_check_consistency(struct facet *facet)
4342 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
4344 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4346 uint64_t odp_actions_stub[1024 / 8];
4347 struct ofpbuf odp_actions;
4349 struct rule_dpif *rule;
4350 struct subfacet *subfacet;
4351 bool may_log = false;
4354 /* Check the rule for consistency. */
4355 rule = rule_dpif_lookup(ofproto, &facet->flow);
4356 ok = rule == facet->rule;
4358 may_log = !VLOG_DROP_WARN(&rl);
4363 flow_format(&s, &facet->flow);
4364 ds_put_format(&s, ": facet associated with wrong rule (was "
4365 "table=%"PRIu8",", facet->rule->up.table_id);
4366 cls_rule_format(&facet->rule->up.cr, &s);
4367 ds_put_format(&s, ") (should have been table=%"PRIu8",",
4369 cls_rule_format(&rule->up.cr, &s);
4370 ds_put_char(&s, ')');
4372 VLOG_WARN("%s", ds_cstr(&s));
4377 /* Check the datapath actions for consistency. */
4378 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
4379 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4380 enum subfacet_path want_path;
4381 struct odputil_keybuf keybuf;
4382 struct action_xlate_ctx ctx;
4386 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
4387 subfacet->initial_tci, rule, 0, NULL);
4388 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len,
4391 if (subfacet->path == SF_NOT_INSTALLED) {
4392 /* This only happens if the datapath reported an error when we
4393 * tried to install the flow. Don't flag another error here. */
4397 want_path = subfacet_want_path(subfacet->slow);
4398 if (want_path == SF_SLOW_PATH && subfacet->path == SF_SLOW_PATH) {
4399 /* The actions for slow-path flows may legitimately vary from one
4400 * packet to the next. We're done. */
4404 if (!subfacet_should_install(subfacet, subfacet->slow, &odp_actions)) {
4408 /* Inconsistency! */
4410 may_log = !VLOG_DROP_WARN(&rl);
4414 /* Rate-limited, skip reporting. */
4419 subfacet_get_key(subfacet, &keybuf, &key);
4420 odp_flow_key_format(key.data, key.size, &s);
4422 ds_put_cstr(&s, ": inconsistency in subfacet");
4423 if (want_path != subfacet->path) {
4424 enum odp_key_fitness fitness = subfacet->key_fitness;
4426 ds_put_format(&s, " (%s, fitness=%s)",
4427 subfacet_path_to_string(subfacet->path),
4428 odp_key_fitness_to_string(fitness));
4429 ds_put_format(&s, " (should have been %s)",
4430 subfacet_path_to_string(want_path));
4431 } else if (want_path == SF_FAST_PATH) {
4432 ds_put_cstr(&s, " (actions were: ");
4433 format_odp_actions(&s, subfacet->actions,
4434 subfacet->actions_len);
4435 ds_put_cstr(&s, ") (correct actions: ");
4436 format_odp_actions(&s, odp_actions.data, odp_actions.size);
4437 ds_put_char(&s, ')');
4439 ds_put_cstr(&s, " (actions: ");
4440 format_odp_actions(&s, subfacet->actions,
4441 subfacet->actions_len);
4442 ds_put_char(&s, ')');
4444 VLOG_WARN("%s", ds_cstr(&s));
4447 ofpbuf_uninit(&odp_actions);
4452 /* Re-searches the classifier for 'facet':
4454 * - If the rule found is different from 'facet''s current rule, moves
4455 * 'facet' to the new rule and recompiles its actions.
4457 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
4458 * where it is and recompiles its actions anyway. */
4460 facet_revalidate(struct facet *facet)
4462 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4464 struct nlattr *odp_actions;
4467 struct actions *new_actions;
4469 struct action_xlate_ctx ctx;
4470 uint64_t odp_actions_stub[1024 / 8];
4471 struct ofpbuf odp_actions;
4473 struct rule_dpif *new_rule;
4474 struct subfacet *subfacet;
4477 COVERAGE_INC(facet_revalidate);
4479 new_rule = rule_dpif_lookup(ofproto, &facet->flow);
4481 /* Calculate new datapath actions.
4483 * We do not modify any 'facet' state yet, because we might need to, e.g.,
4484 * emit a NetFlow expiration and, if so, we need to have the old state
4485 * around to properly compose it. */
4487 /* If the datapath actions changed or the installability changed,
4488 * then we need to talk to the datapath. */
4491 memset(&ctx, 0, sizeof ctx);
4492 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
4493 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4494 enum slow_path_reason slow;
4496 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
4497 subfacet->initial_tci, new_rule, 0, NULL);
4498 xlate_actions(&ctx, new_rule->up.ofpacts, new_rule->up.ofpacts_len,
4501 slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
4502 if (subfacet_should_install(subfacet, slow, &odp_actions)) {
4503 struct dpif_flow_stats stats;
4505 subfacet_install(subfacet,
4506 odp_actions.data, odp_actions.size, &stats, slow);
4507 subfacet_update_stats(subfacet, &stats);
4510 new_actions = xcalloc(list_size(&facet->subfacets),
4511 sizeof *new_actions);
4513 new_actions[i].odp_actions = xmemdup(odp_actions.data,
4515 new_actions[i].actions_len = odp_actions.size;
4520 ofpbuf_uninit(&odp_actions);
4523 facet_flush_stats(facet);
4526 /* Update 'facet' now that we've taken care of all the old state. */
4527 facet->tags = ctx.tags;
4528 facet->nf_flow.output_iface = ctx.nf_output_iface;
4529 facet->has_learn = ctx.has_learn;
4530 facet->has_normal = ctx.has_normal;
4531 facet->has_fin_timeout = ctx.has_fin_timeout;
4532 facet->mirrors = ctx.mirrors;
4535 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4536 subfacet->slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
4538 if (new_actions && new_actions[i].odp_actions) {
4539 free(subfacet->actions);
4540 subfacet->actions = new_actions[i].odp_actions;
4541 subfacet->actions_len = new_actions[i].actions_len;
4547 if (facet->rule != new_rule) {
4548 COVERAGE_INC(facet_changed_rule);
4549 list_remove(&facet->list_node);
4550 list_push_back(&new_rule->facets, &facet->list_node);
4551 facet->rule = new_rule;
4552 facet->used = new_rule->up.created;
4553 facet->prev_used = facet->used;
4557 /* Updates 'facet''s used time. Caller is responsible for calling
4558 * facet_push_stats() to update the flows which 'facet' resubmits into. */
4560 facet_update_time(struct facet *facet, long long int used)
4562 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4563 if (used > facet->used) {
4565 ofproto_rule_update_used(&facet->rule->up, used);
4566 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
4571 facet_reset_counters(struct facet *facet)
4573 facet->packet_count = 0;
4574 facet->byte_count = 0;
4575 facet->prev_packet_count = 0;
4576 facet->prev_byte_count = 0;
4577 facet->accounted_bytes = 0;
4581 facet_push_stats(struct facet *facet)
4583 struct dpif_flow_stats stats;
4585 assert(facet->packet_count >= facet->prev_packet_count);
4586 assert(facet->byte_count >= facet->prev_byte_count);
4587 assert(facet->used >= facet->prev_used);
4589 stats.n_packets = facet->packet_count - facet->prev_packet_count;
4590 stats.n_bytes = facet->byte_count - facet->prev_byte_count;
4591 stats.used = facet->used;
4592 stats.tcp_flags = 0;
4594 if (stats.n_packets || stats.n_bytes || facet->used > facet->prev_used) {
4595 facet->prev_packet_count = facet->packet_count;
4596 facet->prev_byte_count = facet->byte_count;
4597 facet->prev_used = facet->used;
4599 flow_push_stats(facet->rule, &facet->flow, &stats);
4601 update_mirror_stats(ofproto_dpif_cast(facet->rule->up.ofproto),
4602 facet->mirrors, stats.n_packets, stats.n_bytes);
4607 rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats)
4609 rule->packet_count += stats->n_packets;
4610 rule->byte_count += stats->n_bytes;
4611 ofproto_rule_update_used(&rule->up, stats->used);
4614 /* Pushes flow statistics to the rules which 'flow' resubmits into given
4615 * 'rule''s actions and mirrors. */
4617 flow_push_stats(struct rule_dpif *rule,
4618 const struct flow *flow, const struct dpif_flow_stats *stats)
4620 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4621 struct action_xlate_ctx ctx;
4623 ofproto_rule_update_used(&rule->up, stats->used);
4625 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci, rule,
4627 ctx.resubmit_stats = stats;
4628 xlate_actions_for_side_effects(&ctx, rule->up.ofpacts,
4629 rule->up.ofpacts_len);
4634 static struct subfacet *
4635 subfacet_find(struct ofproto_dpif *ofproto,
4636 const struct nlattr *key, size_t key_len, uint32_t key_hash,
4637 const struct flow *flow)
4639 struct subfacet *subfacet;
4641 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
4642 &ofproto->subfacets) {
4644 ? (subfacet->key_len == key_len
4645 && !memcmp(key, subfacet->key, key_len))
4646 : flow_equal(flow, &subfacet->facet->flow)) {
4654 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
4655 * 'key_fitness', 'key', and 'key_len'. Returns the existing subfacet if
4656 * there is one, otherwise creates and returns a new subfacet.
4658 * If the returned subfacet is new, then subfacet->actions will be NULL, in
4659 * which case the caller must populate the actions with
4660 * subfacet_make_actions(). */
4661 static struct subfacet *
4662 subfacet_create(struct facet *facet, enum odp_key_fitness key_fitness,
4663 const struct nlattr *key, size_t key_len,
4664 ovs_be16 initial_tci, long long int now)
4666 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4667 uint32_t key_hash = odp_flow_key_hash(key, key_len);
4668 struct subfacet *subfacet;
4670 if (list_is_empty(&facet->subfacets)) {
4671 subfacet = &facet->one_subfacet;
4673 subfacet = subfacet_find(ofproto, key, key_len, key_hash,
4676 if (subfacet->facet == facet) {
4680 /* This shouldn't happen. */
4681 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
4682 subfacet_destroy(subfacet);
4685 subfacet = xmalloc(sizeof *subfacet);
4688 hmap_insert(&ofproto->subfacets, &subfacet->hmap_node, key_hash);
4689 list_push_back(&facet->subfacets, &subfacet->list_node);
4690 subfacet->facet = facet;
4691 subfacet->key_fitness = key_fitness;
4692 if (key_fitness != ODP_FIT_PERFECT) {
4693 subfacet->key = xmemdup(key, key_len);
4694 subfacet->key_len = key_len;
4696 subfacet->key = NULL;
4697 subfacet->key_len = 0;
4699 subfacet->used = now;
4700 subfacet->dp_packet_count = 0;
4701 subfacet->dp_byte_count = 0;
4702 subfacet->actions_len = 0;
4703 subfacet->actions = NULL;
4704 subfacet->slow = (subfacet->key_fitness == ODP_FIT_TOO_LITTLE
4707 subfacet->path = SF_NOT_INSTALLED;
4708 subfacet->initial_tci = initial_tci;
4713 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
4714 * its facet within 'ofproto', and frees it. */
4716 subfacet_destroy__(struct subfacet *subfacet)
4718 struct facet *facet = subfacet->facet;
4719 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4721 subfacet_uninstall(subfacet);
4722 hmap_remove(&ofproto->subfacets, &subfacet->hmap_node);
4723 list_remove(&subfacet->list_node);
4724 free(subfacet->key);
4725 free(subfacet->actions);
4726 if (subfacet != &facet->one_subfacet) {
4731 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
4732 * last remaining subfacet in its facet destroys the facet too. */
4734 subfacet_destroy(struct subfacet *subfacet)
4736 struct facet *facet = subfacet->facet;
4738 if (list_is_singleton(&facet->subfacets)) {
4739 /* facet_remove() needs at least one subfacet (it will remove it). */
4740 facet_remove(facet);
4742 subfacet_destroy__(subfacet);
4747 subfacet_destroy_batch(struct ofproto_dpif *ofproto,
4748 struct subfacet **subfacets, int n)
4750 struct odputil_keybuf keybufs[SUBFACET_DESTROY_MAX_BATCH];
4751 struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH];
4752 struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH];
4753 struct ofpbuf keys[SUBFACET_DESTROY_MAX_BATCH];
4754 struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH];
4757 for (i = 0; i < n; i++) {
4758 ops[i].type = DPIF_OP_FLOW_DEL;
4759 subfacet_get_key(subfacets[i], &keybufs[i], &keys[i]);
4760 ops[i].u.flow_del.key = keys[i].data;
4761 ops[i].u.flow_del.key_len = keys[i].size;
4762 ops[i].u.flow_del.stats = &stats[i];
4766 dpif_operate(ofproto->backer->dpif, opsp, n);
4767 for (i = 0; i < n; i++) {
4768 subfacet_reset_dp_stats(subfacets[i], &stats[i]);
4769 subfacets[i]->path = SF_NOT_INSTALLED;
4770 subfacet_destroy(subfacets[i]);
4774 /* Initializes 'key' with the sequence of OVS_KEY_ATTR_* Netlink attributes
4775 * that can be used to refer to 'subfacet'. The caller must provide 'keybuf'
4776 * for use as temporary storage. */
4778 subfacet_get_key(struct subfacet *subfacet, struct odputil_keybuf *keybuf,
4782 if (!subfacet->key) {
4783 struct ofproto_dpif *ofproto;
4784 struct flow *flow = &subfacet->facet->flow;
4786 ofpbuf_use_stack(key, keybuf, sizeof *keybuf);
4787 ofproto = ofproto_dpif_cast(subfacet->facet->rule->up.ofproto);
4788 odp_flow_key_from_flow(key, flow,
4789 ofp_port_to_odp_port(ofproto, flow->in_port));
4791 ofpbuf_use_const(key, subfacet->key, subfacet->key_len);
4795 /* Composes the datapath actions for 'subfacet' based on its rule's actions.
4796 * Translates the actions into 'odp_actions', which the caller must have
4797 * initialized and is responsible for uninitializing. */
4799 subfacet_make_actions(struct subfacet *subfacet, const struct ofpbuf *packet,
4800 struct ofpbuf *odp_actions)
4802 struct facet *facet = subfacet->facet;
4803 struct rule_dpif *rule = facet->rule;
4804 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4806 struct action_xlate_ctx ctx;
4808 action_xlate_ctx_init(&ctx, ofproto, &facet->flow, subfacet->initial_tci,
4810 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, odp_actions);
4811 facet->tags = ctx.tags;
4812 facet->has_learn = ctx.has_learn;
4813 facet->has_normal = ctx.has_normal;
4814 facet->has_fin_timeout = ctx.has_fin_timeout;
4815 facet->nf_flow.output_iface = ctx.nf_output_iface;
4816 facet->mirrors = ctx.mirrors;
4818 subfacet->slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
4819 if (subfacet->actions_len != odp_actions->size
4820 || memcmp(subfacet->actions, odp_actions->data, odp_actions->size)) {
4821 free(subfacet->actions);
4822 subfacet->actions_len = odp_actions->size;
4823 subfacet->actions = xmemdup(odp_actions->data, odp_actions->size);
4827 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
4828 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
4829 * in the datapath will be zeroed and 'stats' will be updated with traffic new
4830 * since 'subfacet' was last updated.
4832 * Returns 0 if successful, otherwise a positive errno value. */
4834 subfacet_install(struct subfacet *subfacet,
4835 const struct nlattr *actions, size_t actions_len,
4836 struct dpif_flow_stats *stats,
4837 enum slow_path_reason slow)
4839 struct facet *facet = subfacet->facet;
4840 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4841 enum subfacet_path path = subfacet_want_path(slow);
4842 uint64_t slow_path_stub[128 / 8];
4843 struct odputil_keybuf keybuf;
4844 enum dpif_flow_put_flags flags;
4848 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
4850 flags |= DPIF_FP_ZERO_STATS;
4853 if (path == SF_SLOW_PATH) {
4854 compose_slow_path(ofproto, &facet->flow, slow,
4855 slow_path_stub, sizeof slow_path_stub,
4856 &actions, &actions_len);
4859 subfacet_get_key(subfacet, &keybuf, &key);
4860 ret = dpif_flow_put(ofproto->backer->dpif, flags, key.data, key.size,
4861 actions, actions_len, stats);
4864 subfacet_reset_dp_stats(subfacet, stats);
4868 subfacet->path = path;
4874 subfacet_reinstall(struct subfacet *subfacet, struct dpif_flow_stats *stats)
4876 return subfacet_install(subfacet, subfacet->actions, subfacet->actions_len,
4877 stats, subfacet->slow);
4880 /* If 'subfacet' is installed in the datapath, uninstalls it. */
4882 subfacet_uninstall(struct subfacet *subfacet)
4884 if (subfacet->path != SF_NOT_INSTALLED) {
4885 struct rule_dpif *rule = subfacet->facet->rule;
4886 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4887 struct odputil_keybuf keybuf;
4888 struct dpif_flow_stats stats;
4892 subfacet_get_key(subfacet, &keybuf, &key);
4893 error = dpif_flow_del(ofproto->backer->dpif,
4894 key.data, key.size, &stats);
4895 subfacet_reset_dp_stats(subfacet, &stats);
4897 subfacet_update_stats(subfacet, &stats);
4899 subfacet->path = SF_NOT_INSTALLED;
4901 assert(subfacet->dp_packet_count == 0);
4902 assert(subfacet->dp_byte_count == 0);
4906 /* Resets 'subfacet''s datapath statistics counters. This should be called
4907 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
4908 * non-null, it should contain the statistics returned by dpif when 'subfacet'
4909 * was reset in the datapath. 'stats' will be modified to include only
4910 * statistics new since 'subfacet' was last updated. */
4912 subfacet_reset_dp_stats(struct subfacet *subfacet,
4913 struct dpif_flow_stats *stats)
4916 && subfacet->dp_packet_count <= stats->n_packets
4917 && subfacet->dp_byte_count <= stats->n_bytes) {
4918 stats->n_packets -= subfacet->dp_packet_count;
4919 stats->n_bytes -= subfacet->dp_byte_count;
4922 subfacet->dp_packet_count = 0;
4923 subfacet->dp_byte_count = 0;
4926 /* Updates 'subfacet''s used time. The caller is responsible for calling
4927 * facet_push_stats() to update the flows which 'subfacet' resubmits into. */
4929 subfacet_update_time(struct subfacet *subfacet, long long int used)
4931 if (used > subfacet->used) {
4932 subfacet->used = used;
4933 facet_update_time(subfacet->facet, used);
4937 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
4939 * Because of the meaning of a subfacet's counters, it only makes sense to do
4940 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
4941 * represents a packet that was sent by hand or if it represents statistics
4942 * that have been cleared out of the datapath. */
4944 subfacet_update_stats(struct subfacet *subfacet,
4945 const struct dpif_flow_stats *stats)
4947 if (stats->n_packets || stats->used > subfacet->used) {
4948 struct facet *facet = subfacet->facet;
4950 subfacet_update_time(subfacet, stats->used);
4951 facet->packet_count += stats->n_packets;
4952 facet->byte_count += stats->n_bytes;
4953 facet->tcp_flags |= stats->tcp_flags;
4954 facet_push_stats(facet);
4955 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
4961 static struct rule_dpif *
4962 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow)
4964 struct rule_dpif *rule;
4966 rule = rule_dpif_lookup__(ofproto, flow, 0);
4971 return rule_dpif_miss_rule(ofproto, flow);
4974 static struct rule_dpif *
4975 rule_dpif_lookup__(struct ofproto_dpif *ofproto, const struct flow *flow,
4978 struct cls_rule *cls_rule;
4979 struct classifier *cls;
4981 if (table_id >= N_TABLES) {
4985 cls = &ofproto->up.tables[table_id].cls;
4986 if (flow->nw_frag & FLOW_NW_FRAG_ANY
4987 && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
4988 /* For OFPC_NORMAL frag_handling, we must pretend that transport ports
4989 * are unavailable. */
4990 struct flow ofpc_normal_flow = *flow;
4991 ofpc_normal_flow.tp_src = htons(0);
4992 ofpc_normal_flow.tp_dst = htons(0);
4993 cls_rule = classifier_lookup(cls, &ofpc_normal_flow);
4995 cls_rule = classifier_lookup(cls, flow);
4997 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
5000 static struct rule_dpif *
5001 rule_dpif_miss_rule(struct ofproto_dpif *ofproto, const struct flow *flow)
5003 struct ofport_dpif *port;
5005 port = get_ofp_port(ofproto, flow->in_port);
5007 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, flow->in_port);
5008 return ofproto->miss_rule;
5011 if (port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN) {
5012 return ofproto->no_packet_in_rule;
5014 return ofproto->miss_rule;
5018 complete_operation(struct rule_dpif *rule)
5020 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5022 rule_invalidate(rule);
5024 struct dpif_completion *c = xmalloc(sizeof *c);
5025 c->op = rule->up.pending;
5026 list_push_back(&ofproto->completions, &c->list_node);
5028 ofoperation_complete(rule->up.pending, 0);
5032 static struct rule *
5035 struct rule_dpif *rule = xmalloc(sizeof *rule);
5040 rule_dealloc(struct rule *rule_)
5042 struct rule_dpif *rule = rule_dpif_cast(rule_);
5047 rule_construct(struct rule *rule_)
5049 struct rule_dpif *rule = rule_dpif_cast(rule_);
5050 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5051 struct rule_dpif *victim;
5054 rule->packet_count = 0;
5055 rule->byte_count = 0;
5057 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
5058 if (victim && !list_is_empty(&victim->facets)) {
5059 struct facet *facet;
5061 rule->facets = victim->facets;
5062 list_moved(&rule->facets);
5063 LIST_FOR_EACH (facet, list_node, &rule->facets) {
5064 /* XXX: We're only clearing our local counters here. It's possible
5065 * that quite a few packets are unaccounted for in the datapath
5066 * statistics. These will be accounted to the new rule instead of
5067 * cleared as required. This could be fixed by clearing out the
5068 * datapath statistics for this facet, but currently it doesn't
5070 facet_reset_counters(facet);
5074 /* Must avoid list_moved() in this case. */
5075 list_init(&rule->facets);
5078 table_id = rule->up.table_id;
5080 rule->tag = victim->tag;
5081 } else if (table_id == 0) {
5086 miniflow_expand(&rule->up.cr.match.flow, &flow);
5087 rule->tag = rule_calculate_tag(&flow, &rule->up.cr.match.mask,
5088 ofproto->tables[table_id].basis);
5091 complete_operation(rule);
5096 rule_destruct(struct rule *rule_)
5098 struct rule_dpif *rule = rule_dpif_cast(rule_);
5099 struct facet *facet, *next_facet;
5101 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
5102 facet_revalidate(facet);
5105 complete_operation(rule);
5109 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
5111 struct rule_dpif *rule = rule_dpif_cast(rule_);
5112 struct facet *facet;
5114 /* Start from historical data for 'rule' itself that are no longer tracked
5115 * in facets. This counts, for example, facets that have expired. */
5116 *packets = rule->packet_count;
5117 *bytes = rule->byte_count;
5119 /* Add any statistics that are tracked by facets. This includes
5120 * statistical data recently updated by ofproto_update_stats() as well as
5121 * stats for packets that were executed "by hand" via dpif_execute(). */
5122 LIST_FOR_EACH (facet, list_node, &rule->facets) {
5123 *packets += facet->packet_count;
5124 *bytes += facet->byte_count;
5129 rule_execute(struct rule *rule_, const struct flow *flow,
5130 struct ofpbuf *packet)
5132 struct rule_dpif *rule = rule_dpif_cast(rule_);
5133 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5135 struct dpif_flow_stats stats;
5137 struct action_xlate_ctx ctx;
5138 uint64_t odp_actions_stub[1024 / 8];
5139 struct ofpbuf odp_actions;
5141 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
5142 rule_credit_stats(rule, &stats);
5144 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5145 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci,
5146 rule, stats.tcp_flags, packet);
5147 ctx.resubmit_stats = &stats;
5148 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, &odp_actions);
5150 execute_odp_actions(ofproto, flow, odp_actions.data,
5151 odp_actions.size, packet);
5153 ofpbuf_uninit(&odp_actions);
5159 rule_modify_actions(struct rule *rule_)
5161 struct rule_dpif *rule = rule_dpif_cast(rule_);
5163 complete_operation(rule);
5166 /* Sends 'packet' out 'ofport'.
5167 * May modify 'packet'.
5168 * Returns 0 if successful, otherwise a positive errno value. */
5170 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
5172 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
5173 struct ofpbuf key, odp_actions;
5174 struct odputil_keybuf keybuf;
5179 flow_extract(packet, 0, NULL, OFPP_LOCAL, &flow);
5180 odp_port = vsp_realdev_to_vlandev(ofproto, ofport->odp_port,
5182 if (odp_port != ofport->odp_port) {
5183 eth_pop_vlan(packet);
5184 flow.vlan_tci = htons(0);
5187 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5188 odp_flow_key_from_flow(&key, &flow,
5189 ofp_port_to_odp_port(ofproto, flow.in_port));
5191 ofpbuf_init(&odp_actions, 32);
5192 compose_sflow_action(ofproto, &odp_actions, &flow, odp_port);
5194 nl_msg_put_u32(&odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port);
5195 error = dpif_execute(ofproto->backer->dpif,
5197 odp_actions.data, odp_actions.size,
5199 ofpbuf_uninit(&odp_actions);
5202 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)",
5203 ofproto->up.name, odp_port, strerror(error));
5205 ofproto_update_local_port_stats(ofport->up.ofproto, packet->size, 0);
5209 /* OpenFlow to datapath action translation. */
5211 static void do_xlate_actions(const struct ofpact *, size_t ofpacts_len,
5212 struct action_xlate_ctx *);
5213 static void xlate_normal(struct action_xlate_ctx *);
5215 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
5216 * The action will state 'slow' as the reason that the action is in the slow
5217 * path. (This is purely informational: it allows a human viewing "ovs-dpctl
5218 * dump-flows" output to see why a flow is in the slow path.)
5220 * The 'stub_size' bytes in 'stub' will be used to store the action.
5221 * 'stub_size' must be large enough for the action.
5223 * The action and its size will be stored in '*actionsp' and '*actions_lenp',
5226 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
5227 enum slow_path_reason slow,
5228 uint64_t *stub, size_t stub_size,
5229 const struct nlattr **actionsp, size_t *actions_lenp)
5231 union user_action_cookie cookie;
5234 cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
5235 cookie.slow_path.unused = 0;
5236 cookie.slow_path.reason = slow;
5238 ofpbuf_use_stack(&buf, stub, stub_size);
5239 if (slow & (SLOW_CFM | SLOW_LACP | SLOW_STP)) {
5240 uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif, UINT16_MAX);
5241 odp_put_userspace_action(pid, &cookie, &buf);
5243 put_userspace_action(ofproto, &buf, flow, &cookie);
5245 *actionsp = buf.data;
5246 *actions_lenp = buf.size;
5250 put_userspace_action(const struct ofproto_dpif *ofproto,
5251 struct ofpbuf *odp_actions,
5252 const struct flow *flow,
5253 const union user_action_cookie *cookie)
5257 pid = dpif_port_get_pid(ofproto->backer->dpif,
5258 ofp_port_to_odp_port(ofproto, flow->in_port));
5260 return odp_put_userspace_action(pid, cookie, odp_actions);
5264 compose_sflow_cookie(const struct ofproto_dpif *ofproto,
5265 ovs_be16 vlan_tci, uint32_t odp_port,
5266 unsigned int n_outputs, union user_action_cookie *cookie)
5270 cookie->type = USER_ACTION_COOKIE_SFLOW;
5271 cookie->sflow.vlan_tci = vlan_tci;
5273 /* See http://www.sflow.org/sflow_version_5.txt (search for "Input/output
5274 * port information") for the interpretation of cookie->output. */
5275 switch (n_outputs) {
5277 /* 0x40000000 | 256 means "packet dropped for unknown reason". */
5278 cookie->sflow.output = 0x40000000 | 256;
5282 ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port);
5284 cookie->sflow.output = ifindex;
5289 /* 0x80000000 means "multiple output ports. */
5290 cookie->sflow.output = 0x80000000 | n_outputs;
5295 /* Compose SAMPLE action for sFlow. */
5297 compose_sflow_action(const struct ofproto_dpif *ofproto,
5298 struct ofpbuf *odp_actions,
5299 const struct flow *flow,
5302 uint32_t probability;
5303 union user_action_cookie cookie;
5304 size_t sample_offset, actions_offset;
5307 if (!ofproto->sflow || flow->in_port == OFPP_NONE) {
5311 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
5313 /* Number of packets out of UINT_MAX to sample. */
5314 probability = dpif_sflow_get_probability(ofproto->sflow);
5315 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
5317 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
5318 compose_sflow_cookie(ofproto, htons(0), odp_port,
5319 odp_port == OVSP_NONE ? 0 : 1, &cookie);
5320 cookie_offset = put_userspace_action(ofproto, odp_actions, flow, &cookie);
5322 nl_msg_end_nested(odp_actions, actions_offset);
5323 nl_msg_end_nested(odp_actions, sample_offset);
5324 return cookie_offset;
5327 /* SAMPLE action must be first action in any given list of actions.
5328 * At this point we do not have all information required to build it. So try to
5329 * build sample action as complete as possible. */
5331 add_sflow_action(struct action_xlate_ctx *ctx)
5333 ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto,
5335 &ctx->flow, OVSP_NONE);
5336 ctx->sflow_odp_port = 0;
5337 ctx->sflow_n_outputs = 0;
5340 /* Fix SAMPLE action according to data collected while composing ODP actions.
5341 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
5342 * USERSPACE action's user-cookie which is required for sflow. */
5344 fix_sflow_action(struct action_xlate_ctx *ctx)
5346 const struct flow *base = &ctx->base_flow;
5347 union user_action_cookie *cookie;
5349 if (!ctx->user_cookie_offset) {
5353 cookie = ofpbuf_at(ctx->odp_actions, ctx->user_cookie_offset,
5355 assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
5357 compose_sflow_cookie(ctx->ofproto, base->vlan_tci,
5358 ctx->sflow_odp_port, ctx->sflow_n_outputs, cookie);
5362 compose_output_action__(struct action_xlate_ctx *ctx, uint16_t ofp_port,
5365 const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
5366 uint32_t odp_port = ofp_port_to_odp_port(ctx->ofproto, ofp_port);
5367 ovs_be16 flow_vlan_tci = ctx->flow.vlan_tci;
5368 uint8_t flow_nw_tos = ctx->flow.nw_tos;
5372 struct priority_to_dscp *pdscp;
5374 if (ofport->up.pp.config & OFPUTIL_PC_NO_FWD) {
5375 xlate_report(ctx, "OFPPC_NO_FWD set, skipping output");
5377 } else if (check_stp && !stp_forward_in_state(ofport->stp_state)) {
5378 xlate_report(ctx, "STP not in forwarding state, skipping output");
5382 pdscp = get_priority(ofport, ctx->flow.skb_priority);
5384 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
5385 ctx->flow.nw_tos |= pdscp->dscp;
5388 /* We may not have an ofport record for this port, but it doesn't hurt
5389 * to allow forwarding to it anyhow. Maybe such a port will appear
5390 * later and we're pre-populating the flow table. */
5393 out_port = vsp_realdev_to_vlandev(ctx->ofproto, odp_port,
5394 ctx->flow.vlan_tci);
5395 if (out_port != odp_port) {
5396 ctx->flow.vlan_tci = htons(0);
5398 commit_odp_actions(&ctx->flow, &ctx->base_flow, ctx->odp_actions);
5399 nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_OUTPUT, out_port);
5401 ctx->sflow_odp_port = odp_port;
5402 ctx->sflow_n_outputs++;
5403 ctx->nf_output_iface = ofp_port;
5404 ctx->flow.vlan_tci = flow_vlan_tci;
5405 ctx->flow.nw_tos = flow_nw_tos;
5409 compose_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port)
5411 compose_output_action__(ctx, ofp_port, true);
5415 xlate_table_action(struct action_xlate_ctx *ctx,
5416 uint16_t in_port, uint8_t table_id, bool may_packet_in)
5418 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
5419 struct ofproto_dpif *ofproto = ctx->ofproto;
5420 struct rule_dpif *rule;
5421 uint16_t old_in_port;
5422 uint8_t old_table_id;
5424 old_table_id = ctx->table_id;
5425 ctx->table_id = table_id;
5427 /* Look up a flow with 'in_port' as the input port. */
5428 old_in_port = ctx->flow.in_port;
5429 ctx->flow.in_port = in_port;
5430 rule = rule_dpif_lookup__(ofproto, &ctx->flow, table_id);
5433 if (table_id > 0 && table_id < N_TABLES) {
5434 struct table_dpif *table = &ofproto->tables[table_id];
5435 if (table->other_table) {
5436 ctx->tags |= (rule && rule->tag
5438 : rule_calculate_tag(&ctx->flow,
5439 &table->other_table->mask,
5444 /* Restore the original input port. Otherwise OFPP_NORMAL and
5445 * OFPP_IN_PORT will have surprising behavior. */
5446 ctx->flow.in_port = old_in_port;
5448 if (ctx->resubmit_hook) {
5449 ctx->resubmit_hook(ctx, rule);
5452 if (rule == NULL && may_packet_in) {
5454 * check if table configuration flags
5455 * OFPTC_TABLE_MISS_CONTROLLER, default.
5456 * OFPTC_TABLE_MISS_CONTINUE,
5457 * OFPTC_TABLE_MISS_DROP
5458 * When OF1.0, OFPTC_TABLE_MISS_CONTINUE is used. What to do?
5460 rule = rule_dpif_miss_rule(ofproto, &ctx->flow);
5464 struct rule_dpif *old_rule = ctx->rule;
5466 if (ctx->resubmit_stats) {
5467 rule_credit_stats(rule, ctx->resubmit_stats);
5472 do_xlate_actions(rule->up.ofpacts, rule->up.ofpacts_len, ctx);
5473 ctx->rule = old_rule;
5477 ctx->table_id = old_table_id;
5479 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
5481 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
5482 MAX_RESUBMIT_RECURSION);
5483 ctx->max_resubmit_trigger = true;
5488 xlate_ofpact_resubmit(struct action_xlate_ctx *ctx,
5489 const struct ofpact_resubmit *resubmit)
5494 in_port = resubmit->in_port;
5495 if (in_port == OFPP_IN_PORT) {
5496 in_port = ctx->flow.in_port;
5499 table_id = resubmit->table_id;
5500 if (table_id == 255) {
5501 table_id = ctx->table_id;
5504 xlate_table_action(ctx, in_port, table_id, false);
5508 flood_packets(struct action_xlate_ctx *ctx, bool all)
5510 struct ofport_dpif *ofport;
5512 HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) {
5513 uint16_t ofp_port = ofport->up.ofp_port;
5515 if (ofp_port == ctx->flow.in_port) {
5520 compose_output_action__(ctx, ofp_port, false);
5521 } else if (!(ofport->up.pp.config & OFPUTIL_PC_NO_FLOOD)) {
5522 compose_output_action(ctx, ofp_port);
5526 ctx->nf_output_iface = NF_OUT_FLOOD;
5530 execute_controller_action(struct action_xlate_ctx *ctx, int len,
5531 enum ofp_packet_in_reason reason,
5532 uint16_t controller_id)
5534 struct ofputil_packet_in pin;
5535 struct ofpbuf *packet;
5537 ctx->slow |= SLOW_CONTROLLER;
5542 packet = ofpbuf_clone(ctx->packet);
5544 if (packet->l2 && packet->l3) {
5545 struct eth_header *eh;
5547 eth_pop_vlan(packet);
5550 /* If the Ethernet type is less than ETH_TYPE_MIN, it's likely an 802.2
5551 * LLC frame. Calculating the Ethernet type of these frames is more
5552 * trouble than seems appropriate for a simple assertion. */
5553 assert(ntohs(eh->eth_type) < ETH_TYPE_MIN
5554 || eh->eth_type == ctx->flow.dl_type);
5556 memcpy(eh->eth_src, ctx->flow.dl_src, sizeof eh->eth_src);
5557 memcpy(eh->eth_dst, ctx->flow.dl_dst, sizeof eh->eth_dst);
5559 if (ctx->flow.vlan_tci & htons(VLAN_CFI)) {
5560 eth_push_vlan(packet, ctx->flow.vlan_tci);
5564 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
5565 packet_set_ipv4(packet, ctx->flow.nw_src, ctx->flow.nw_dst,
5566 ctx->flow.nw_tos, ctx->flow.nw_ttl);
5570 if (ctx->flow.nw_proto == IPPROTO_TCP) {
5571 packet_set_tcp_port(packet, ctx->flow.tp_src,
5573 } else if (ctx->flow.nw_proto == IPPROTO_UDP) {
5574 packet_set_udp_port(packet, ctx->flow.tp_src,
5581 pin.packet = packet->data;
5582 pin.packet_len = packet->size;
5583 pin.reason = reason;
5584 pin.controller_id = controller_id;
5585 pin.table_id = ctx->table_id;
5586 pin.cookie = ctx->rule ? ctx->rule->up.flow_cookie : 0;
5589 flow_get_metadata(&ctx->flow, &pin.fmd);
5591 connmgr_send_packet_in(ctx->ofproto->up.connmgr, &pin);
5592 ofpbuf_delete(packet);
5596 compose_dec_ttl(struct action_xlate_ctx *ctx, struct ofpact_cnt_ids *ids)
5598 if (ctx->flow.dl_type != htons(ETH_TYPE_IP) &&
5599 ctx->flow.dl_type != htons(ETH_TYPE_IPV6)) {
5603 if (ctx->flow.nw_ttl > 1) {
5609 for (i = 0; i < ids->n_controllers; i++) {
5610 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL,
5614 /* Stop processing for current table. */
5620 xlate_output_action(struct action_xlate_ctx *ctx,
5621 uint16_t port, uint16_t max_len, bool may_packet_in)
5623 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
5625 ctx->nf_output_iface = NF_OUT_DROP;
5629 compose_output_action(ctx, ctx->flow.in_port);
5632 xlate_table_action(ctx, ctx->flow.in_port, 0, may_packet_in);
5638 flood_packets(ctx, false);
5641 flood_packets(ctx, true);
5643 case OFPP_CONTROLLER:
5644 execute_controller_action(ctx, max_len, OFPR_ACTION, 0);
5650 if (port != ctx->flow.in_port) {
5651 compose_output_action(ctx, port);
5653 xlate_report(ctx, "skipping output to input port");
5658 if (prev_nf_output_iface == NF_OUT_FLOOD) {
5659 ctx->nf_output_iface = NF_OUT_FLOOD;
5660 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
5661 ctx->nf_output_iface = prev_nf_output_iface;
5662 } else if (prev_nf_output_iface != NF_OUT_DROP &&
5663 ctx->nf_output_iface != NF_OUT_FLOOD) {
5664 ctx->nf_output_iface = NF_OUT_MULTI;
5669 xlate_output_reg_action(struct action_xlate_ctx *ctx,
5670 const struct ofpact_output_reg *or)
5672 uint64_t port = mf_get_subfield(&or->src, &ctx->flow);
5673 if (port <= UINT16_MAX) {
5674 xlate_output_action(ctx, port, or->max_len, false);
5679 xlate_enqueue_action(struct action_xlate_ctx *ctx,
5680 const struct ofpact_enqueue *enqueue)
5682 uint16_t ofp_port = enqueue->port;
5683 uint32_t queue_id = enqueue->queue;
5684 uint32_t flow_priority, priority;
5687 /* Translate queue to priority. */
5688 error = dpif_queue_to_priority(ctx->ofproto->backer->dpif,
5689 queue_id, &priority);
5691 /* Fall back to ordinary output action. */
5692 xlate_output_action(ctx, enqueue->port, 0, false);
5696 /* Check output port. */
5697 if (ofp_port == OFPP_IN_PORT) {
5698 ofp_port = ctx->flow.in_port;
5699 } else if (ofp_port == ctx->flow.in_port) {
5703 /* Add datapath actions. */
5704 flow_priority = ctx->flow.skb_priority;
5705 ctx->flow.skb_priority = priority;
5706 compose_output_action(ctx, ofp_port);
5707 ctx->flow.skb_priority = flow_priority;
5709 /* Update NetFlow output port. */
5710 if (ctx->nf_output_iface == NF_OUT_DROP) {
5711 ctx->nf_output_iface = ofp_port;
5712 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
5713 ctx->nf_output_iface = NF_OUT_MULTI;
5718 xlate_set_queue_action(struct action_xlate_ctx *ctx, uint32_t queue_id)
5720 uint32_t skb_priority;
5722 if (!dpif_queue_to_priority(ctx->ofproto->backer->dpif,
5723 queue_id, &skb_priority)) {
5724 ctx->flow.skb_priority = skb_priority;
5726 /* Couldn't translate queue to a priority. Nothing to do. A warning
5727 * has already been logged. */
5731 struct xlate_reg_state {
5737 xlate_autopath(struct action_xlate_ctx *ctx,
5738 const struct ofpact_autopath *ap)
5740 uint16_t ofp_port = ap->port;
5741 struct ofport_dpif *port = get_ofp_port(ctx->ofproto, ofp_port);
5743 if (!port || !port->bundle) {
5744 ofp_port = OFPP_NONE;
5745 } else if (port->bundle->bond) {
5746 /* Autopath does not support VLAN hashing. */
5747 struct ofport_dpif *slave = bond_choose_output_slave(
5748 port->bundle->bond, &ctx->flow, 0, &ctx->tags);
5750 ofp_port = slave->up.ofp_port;
5753 nxm_reg_load(&ap->dst, ofp_port, &ctx->flow);
5757 slave_enabled_cb(uint16_t ofp_port, void *ofproto_)
5759 struct ofproto_dpif *ofproto = ofproto_;
5760 struct ofport_dpif *port;
5770 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
5773 port = get_ofp_port(ofproto, ofp_port);
5774 return port ? port->may_enable : false;
5779 xlate_bundle_action(struct action_xlate_ctx *ctx,
5780 const struct ofpact_bundle *bundle)
5784 port = bundle_execute(bundle, &ctx->flow, slave_enabled_cb, ctx->ofproto);
5785 if (bundle->dst.field) {
5786 nxm_reg_load(&bundle->dst, port, &ctx->flow);
5788 xlate_output_action(ctx, port, 0, false);
5793 xlate_learn_action(struct action_xlate_ctx *ctx,
5794 const struct ofpact_learn *learn)
5796 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
5797 struct ofputil_flow_mod fm;
5798 uint64_t ofpacts_stub[1024 / 8];
5799 struct ofpbuf ofpacts;
5802 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
5803 learn_execute(learn, &ctx->flow, &fm, &ofpacts);
5805 error = ofproto_flow_mod(&ctx->ofproto->up, &fm);
5806 if (error && !VLOG_DROP_WARN(&rl)) {
5807 VLOG_WARN("learning action failed to modify flow table (%s)",
5808 ofperr_get_name(error));
5811 ofpbuf_uninit(&ofpacts);
5814 /* Reduces '*timeout' to no more than 'max'. A value of zero in either case
5815 * means "infinite". */
5817 reduce_timeout(uint16_t max, uint16_t *timeout)
5819 if (max && (!*timeout || *timeout > max)) {
5825 xlate_fin_timeout(struct action_xlate_ctx *ctx,
5826 const struct ofpact_fin_timeout *oft)
5828 if (ctx->tcp_flags & (TCP_FIN | TCP_RST) && ctx->rule) {
5829 struct rule_dpif *rule = ctx->rule;
5831 reduce_timeout(oft->fin_idle_timeout, &rule->up.idle_timeout);
5832 reduce_timeout(oft->fin_hard_timeout, &rule->up.hard_timeout);
5837 may_receive(const struct ofport_dpif *port, struct action_xlate_ctx *ctx)
5839 if (port->up.pp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
5840 ? OFPUTIL_PC_NO_RECV_STP
5841 : OFPUTIL_PC_NO_RECV)) {
5845 /* Only drop packets here if both forwarding and learning are
5846 * disabled. If just learning is enabled, we need to have
5847 * OFPP_NORMAL and the learning action have a look at the packet
5848 * before we can drop it. */
5849 if (!stp_forward_in_state(port->stp_state)
5850 && !stp_learn_in_state(port->stp_state)) {
5858 do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len,
5859 struct action_xlate_ctx *ctx)
5861 const struct ofport_dpif *port;
5862 bool was_evictable = true;
5863 const struct ofpact *a;
5865 port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
5866 if (port && !may_receive(port, ctx)) {
5867 /* Drop this flow. */
5872 /* Don't let the rule we're working on get evicted underneath us. */
5873 was_evictable = ctx->rule->up.evictable;
5874 ctx->rule->up.evictable = false;
5876 OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) {
5877 struct ofpact_controller *controller;
5878 const struct ofpact_metadata *metadata;
5886 xlate_output_action(ctx, ofpact_get_OUTPUT(a)->port,
5887 ofpact_get_OUTPUT(a)->max_len, true);
5890 case OFPACT_CONTROLLER:
5891 controller = ofpact_get_CONTROLLER(a);
5892 execute_controller_action(ctx, controller->max_len,
5894 controller->controller_id);
5897 case OFPACT_ENQUEUE:
5898 xlate_enqueue_action(ctx, ofpact_get_ENQUEUE(a));
5901 case OFPACT_SET_VLAN_VID:
5902 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
5903 ctx->flow.vlan_tci |= (htons(ofpact_get_SET_VLAN_VID(a)->vlan_vid)
5907 case OFPACT_SET_VLAN_PCP:
5908 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
5909 ctx->flow.vlan_tci |= htons((ofpact_get_SET_VLAN_PCP(a)->vlan_pcp
5914 case OFPACT_STRIP_VLAN:
5915 ctx->flow.vlan_tci = htons(0);
5918 case OFPACT_PUSH_VLAN:
5919 /* TODO:XXX 802.1AD(QinQ) */
5920 ctx->flow.vlan_tci = htons(VLAN_CFI);
5923 case OFPACT_SET_ETH_SRC:
5924 memcpy(ctx->flow.dl_src, ofpact_get_SET_ETH_SRC(a)->mac,
5928 case OFPACT_SET_ETH_DST:
5929 memcpy(ctx->flow.dl_dst, ofpact_get_SET_ETH_DST(a)->mac,
5933 case OFPACT_SET_IPV4_SRC:
5934 ctx->flow.nw_src = ofpact_get_SET_IPV4_SRC(a)->ipv4;
5937 case OFPACT_SET_IPV4_DST:
5938 ctx->flow.nw_dst = ofpact_get_SET_IPV4_DST(a)->ipv4;
5941 case OFPACT_SET_IPV4_DSCP:
5942 /* OpenFlow 1.0 only supports IPv4. */
5943 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
5944 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
5945 ctx->flow.nw_tos |= ofpact_get_SET_IPV4_DSCP(a)->dscp;
5949 case OFPACT_SET_L4_SRC_PORT:
5950 ctx->flow.tp_src = htons(ofpact_get_SET_L4_SRC_PORT(a)->port);
5953 case OFPACT_SET_L4_DST_PORT:
5954 ctx->flow.tp_dst = htons(ofpact_get_SET_L4_DST_PORT(a)->port);
5957 case OFPACT_RESUBMIT:
5958 xlate_ofpact_resubmit(ctx, ofpact_get_RESUBMIT(a));
5961 case OFPACT_SET_TUNNEL:
5962 ctx->flow.tunnel.tun_id = htonll(ofpact_get_SET_TUNNEL(a)->tun_id);
5965 case OFPACT_SET_QUEUE:
5966 xlate_set_queue_action(ctx, ofpact_get_SET_QUEUE(a)->queue_id);
5969 case OFPACT_POP_QUEUE:
5970 ctx->flow.skb_priority = ctx->orig_skb_priority;
5973 case OFPACT_REG_MOVE:
5974 nxm_execute_reg_move(ofpact_get_REG_MOVE(a), &ctx->flow);
5977 case OFPACT_REG_LOAD:
5978 nxm_execute_reg_load(ofpact_get_REG_LOAD(a), &ctx->flow);
5981 case OFPACT_DEC_TTL:
5982 if (compose_dec_ttl(ctx, ofpact_get_DEC_TTL(a))) {
5988 /* Nothing to do. */
5991 case OFPACT_MULTIPATH:
5992 multipath_execute(ofpact_get_MULTIPATH(a), &ctx->flow);
5995 case OFPACT_AUTOPATH:
5996 xlate_autopath(ctx, ofpact_get_AUTOPATH(a));
6000 ctx->ofproto->has_bundle_action = true;
6001 xlate_bundle_action(ctx, ofpact_get_BUNDLE(a));
6004 case OFPACT_OUTPUT_REG:
6005 xlate_output_reg_action(ctx, ofpact_get_OUTPUT_REG(a));
6009 ctx->has_learn = true;
6010 if (ctx->may_learn) {
6011 xlate_learn_action(ctx, ofpact_get_LEARN(a));
6019 case OFPACT_FIN_TIMEOUT:
6020 ctx->has_fin_timeout = true;
6021 xlate_fin_timeout(ctx, ofpact_get_FIN_TIMEOUT(a));
6024 case OFPACT_CLEAR_ACTIONS:
6026 * Nothing to do because writa-actions is not supported for now.
6027 * When writa-actions is supported, clear-actions also must
6028 * be supported at the same time.
6032 case OFPACT_WRITE_METADATA:
6033 metadata = ofpact_get_WRITE_METADATA(a);
6034 ctx->flow.metadata &= ~metadata->mask;
6035 ctx->flow.metadata |= metadata->metadata & metadata->mask;
6038 case OFPACT_GOTO_TABLE: {
6039 /* TODO:XXX remove recursion */
6040 /* It is assumed that goto-table is last action */
6041 struct ofpact_goto_table *ogt = ofpact_get_GOTO_TABLE(a);
6042 assert(ctx->table_id < ogt->table_id);
6043 xlate_table_action(ctx, ctx->flow.in_port, ogt->table_id, true);
6050 /* We've let OFPP_NORMAL and the learning action look at the packet,
6051 * so drop it now if forwarding is disabled. */
6052 if (port && !stp_forward_in_state(port->stp_state)) {
6053 ofpbuf_clear(ctx->odp_actions);
6054 add_sflow_action(ctx);
6057 ctx->rule->up.evictable = was_evictable;
6062 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
6063 struct ofproto_dpif *ofproto, const struct flow *flow,
6064 ovs_be16 initial_tci, struct rule_dpif *rule,
6065 uint8_t tcp_flags, const struct ofpbuf *packet)
6067 ctx->ofproto = ofproto;
6069 ctx->base_flow = ctx->flow;
6070 memset(&ctx->base_flow.tunnel, 0, sizeof ctx->base_flow.tunnel);
6071 ctx->base_flow.vlan_tci = initial_tci;
6073 ctx->packet = packet;
6074 ctx->may_learn = packet != NULL;
6075 ctx->tcp_flags = tcp_flags;
6076 ctx->resubmit_hook = NULL;
6077 ctx->report_hook = NULL;
6078 ctx->resubmit_stats = NULL;
6081 /* Translates the 'ofpacts_len' bytes of "struct ofpacts" starting at 'ofpacts'
6082 * into datapath actions in 'odp_actions', using 'ctx'. */
6084 xlate_actions(struct action_xlate_ctx *ctx,
6085 const struct ofpact *ofpacts, size_t ofpacts_len,
6086 struct ofpbuf *odp_actions)
6088 /* Normally false. Set to true if we ever hit MAX_RESUBMIT_RECURSION, so
6089 * that in the future we always keep a copy of the original flow for
6090 * tracing purposes. */
6091 static bool hit_resubmit_limit;
6093 enum slow_path_reason special;
6095 COVERAGE_INC(ofproto_dpif_xlate);
6097 ofpbuf_clear(odp_actions);
6098 ofpbuf_reserve(odp_actions, NL_A_U32_SIZE);
6100 ctx->odp_actions = odp_actions;
6103 ctx->has_learn = false;
6104 ctx->has_normal = false;
6105 ctx->has_fin_timeout = false;
6106 ctx->nf_output_iface = NF_OUT_DROP;
6109 ctx->max_resubmit_trigger = false;
6110 ctx->orig_skb_priority = ctx->flow.skb_priority;
6114 if (ctx->ofproto->has_mirrors || hit_resubmit_limit) {
6115 /* Do this conditionally because the copy is expensive enough that it
6116 * shows up in profiles.
6118 * We keep orig_flow in 'ctx' only because I couldn't make GCC 4.4
6119 * believe that I wasn't using it without initializing it if I kept it
6120 * in a local variable. */
6121 ctx->orig_flow = ctx->flow;
6124 if (ctx->flow.nw_frag & FLOW_NW_FRAG_ANY) {
6125 switch (ctx->ofproto->up.frag_handling) {
6126 case OFPC_FRAG_NORMAL:
6127 /* We must pretend that transport ports are unavailable. */
6128 ctx->flow.tp_src = ctx->base_flow.tp_src = htons(0);
6129 ctx->flow.tp_dst = ctx->base_flow.tp_dst = htons(0);
6132 case OFPC_FRAG_DROP:
6135 case OFPC_FRAG_REASM:
6138 case OFPC_FRAG_NX_MATCH:
6139 /* Nothing to do. */
6142 case OFPC_INVALID_TTL_TO_CONTROLLER:
6147 special = process_special(ctx->ofproto, &ctx->flow, ctx->packet);
6149 ctx->slow |= special;
6151 static struct vlog_rate_limit trace_rl = VLOG_RATE_LIMIT_INIT(1, 1);
6152 ovs_be16 initial_tci = ctx->base_flow.vlan_tci;
6154 add_sflow_action(ctx);
6155 do_xlate_actions(ofpacts, ofpacts_len, ctx);
6157 if (ctx->max_resubmit_trigger && !ctx->resubmit_hook) {
6158 if (!hit_resubmit_limit) {
6159 /* We didn't record the original flow. Make sure we do from
6161 hit_resubmit_limit = true;
6162 } else if (!VLOG_DROP_ERR(&trace_rl)) {
6163 struct ds ds = DS_EMPTY_INITIALIZER;
6165 ofproto_trace(ctx->ofproto, &ctx->orig_flow, ctx->packet,
6167 VLOG_ERR("Trace triggered by excessive resubmit "
6168 "recursion:\n%s", ds_cstr(&ds));
6173 if (!connmgr_may_set_up_flow(ctx->ofproto->up.connmgr, &ctx->flow,
6174 ctx->odp_actions->data,
6175 ctx->odp_actions->size)) {
6176 ctx->slow |= SLOW_IN_BAND;
6178 && connmgr_msg_in_hook(ctx->ofproto->up.connmgr, &ctx->flow,
6180 compose_output_action(ctx, OFPP_LOCAL);
6183 if (ctx->ofproto->has_mirrors) {
6184 add_mirror_actions(ctx, &ctx->orig_flow);
6186 fix_sflow_action(ctx);
6190 /* Translates the 'ofpacts_len' bytes of "struct ofpact"s starting at 'ofpacts'
6191 * into datapath actions, using 'ctx', and discards the datapath actions. */
6193 xlate_actions_for_side_effects(struct action_xlate_ctx *ctx,
6194 const struct ofpact *ofpacts,
6197 uint64_t odp_actions_stub[1024 / 8];
6198 struct ofpbuf odp_actions;
6200 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
6201 xlate_actions(ctx, ofpacts, ofpacts_len, &odp_actions);
6202 ofpbuf_uninit(&odp_actions);
6206 xlate_report(struct action_xlate_ctx *ctx, const char *s)
6208 if (ctx->report_hook) {
6209 ctx->report_hook(ctx, s);
6213 /* OFPP_NORMAL implementation. */
6215 static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *);
6217 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
6218 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle',
6219 * the bundle on which the packet was received, returns the VLAN to which the
6222 * Both 'vid' and the return value are in the range 0...4095. */
6224 input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid)
6226 switch (in_bundle->vlan_mode) {
6227 case PORT_VLAN_ACCESS:
6228 return in_bundle->vlan;
6231 case PORT_VLAN_TRUNK:
6234 case PORT_VLAN_NATIVE_UNTAGGED:
6235 case PORT_VLAN_NATIVE_TAGGED:
6236 return vid ? vid : in_bundle->vlan;
6243 /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'.
6244 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
6247 * 'vid' should be the VID obtained from the 802.1Q header that was received as
6248 * part of a packet (specify 0 if there was no 802.1Q header), in the range
6251 input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn)
6253 /* Allow any VID on the OFPP_NONE port. */
6254 if (in_bundle == &ofpp_none_bundle) {
6258 switch (in_bundle->vlan_mode) {
6259 case PORT_VLAN_ACCESS:
6262 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6263 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
6264 "packet received on port %s configured as VLAN "
6265 "%"PRIu16" access port",
6266 in_bundle->ofproto->up.name, vid,
6267 in_bundle->name, in_bundle->vlan);
6273 case PORT_VLAN_NATIVE_UNTAGGED:
6274 case PORT_VLAN_NATIVE_TAGGED:
6276 /* Port must always carry its native VLAN. */
6280 case PORT_VLAN_TRUNK:
6281 if (!ofbundle_includes_vlan(in_bundle, vid)) {
6283 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6284 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet "
6285 "received on port %s not configured for trunking "
6287 in_bundle->ofproto->up.name, vid,
6288 in_bundle->name, vid);
6300 /* Given 'vlan', the VLAN that a packet belongs to, and
6301 * 'out_bundle', a bundle on which the packet is to be output, returns the VID
6302 * that should be included in the 802.1Q header. (If the return value is 0,
6303 * then the 802.1Q header should only be included in the packet if there is a
6306 * Both 'vlan' and the return value are in the range 0...4095. */
6308 output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan)
6310 switch (out_bundle->vlan_mode) {
6311 case PORT_VLAN_ACCESS:
6314 case PORT_VLAN_TRUNK:
6315 case PORT_VLAN_NATIVE_TAGGED:
6318 case PORT_VLAN_NATIVE_UNTAGGED:
6319 return vlan == out_bundle->vlan ? 0 : vlan;
6327 output_normal(struct action_xlate_ctx *ctx, const struct ofbundle *out_bundle,
6330 struct ofport_dpif *port;
6332 ovs_be16 tci, old_tci;
6334 vid = output_vlan_to_vid(out_bundle, vlan);
6335 if (!out_bundle->bond) {
6336 port = ofbundle_get_a_port(out_bundle);
6338 port = bond_choose_output_slave(out_bundle->bond, &ctx->flow,
6341 /* No slaves enabled, so drop packet. */
6346 old_tci = ctx->flow.vlan_tci;
6348 if (tci || out_bundle->use_priority_tags) {
6349 tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK);
6351 tci |= htons(VLAN_CFI);
6354 ctx->flow.vlan_tci = tci;
6356 compose_output_action(ctx, port->up.ofp_port);
6357 ctx->flow.vlan_tci = old_tci;
6361 mirror_mask_ffs(mirror_mask_t mask)
6363 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
6368 ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
6370 return (bundle->vlan_mode != PORT_VLAN_ACCESS
6371 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
6375 ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan)
6377 return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan);
6380 /* Returns an arbitrary interface within 'bundle'. */
6381 static struct ofport_dpif *
6382 ofbundle_get_a_port(const struct ofbundle *bundle)
6384 return CONTAINER_OF(list_front(&bundle->ports),
6385 struct ofport_dpif, bundle_node);
6389 vlan_is_mirrored(const struct ofmirror *m, int vlan)
6391 return !m->vlans || bitmap_is_set(m->vlans, vlan);
6395 add_mirror_actions(struct action_xlate_ctx *ctx, const struct flow *orig_flow)
6397 struct ofproto_dpif *ofproto = ctx->ofproto;
6398 mirror_mask_t mirrors;
6399 struct ofbundle *in_bundle;
6402 const struct nlattr *a;
6405 in_bundle = lookup_input_bundle(ctx->ofproto, orig_flow->in_port,
6406 ctx->packet != NULL, NULL);
6410 mirrors = in_bundle->src_mirrors;
6412 /* Drop frames on bundles reserved for mirroring. */
6413 if (in_bundle->mirror_out) {
6414 if (ctx->packet != NULL) {
6415 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6416 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
6417 "%s, which is reserved exclusively for mirroring",
6418 ctx->ofproto->up.name, in_bundle->name);
6424 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
6425 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
6428 vlan = input_vid_to_vlan(in_bundle, vid);
6430 /* Look at the output ports to check for destination selections. */
6432 NL_ATTR_FOR_EACH (a, left, ctx->odp_actions->data,
6433 ctx->odp_actions->size) {
6434 enum ovs_action_attr type = nl_attr_type(a);
6435 struct ofport_dpif *ofport;
6437 if (type != OVS_ACTION_ATTR_OUTPUT) {
6441 ofport = get_odp_port(ofproto, nl_attr_get_u32(a));
6442 if (ofport && ofport->bundle) {
6443 mirrors |= ofport->bundle->dst_mirrors;
6451 /* Restore the original packet before adding the mirror actions. */
6452 ctx->flow = *orig_flow;
6457 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
6459 if (!vlan_is_mirrored(m, vlan)) {
6460 mirrors = zero_rightmost_1bit(mirrors);
6464 mirrors &= ~m->dup_mirrors;
6465 ctx->mirrors |= m->dup_mirrors;
6467 output_normal(ctx, m->out, vlan);
6468 } else if (vlan != m->out_vlan
6469 && !eth_addr_is_reserved(orig_flow->dl_dst)) {
6470 struct ofbundle *bundle;
6472 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
6473 if (ofbundle_includes_vlan(bundle, m->out_vlan)
6474 && !bundle->mirror_out) {
6475 output_normal(ctx, bundle, m->out_vlan);
6483 update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors,
6484 uint64_t packets, uint64_t bytes)
6490 for (; mirrors; mirrors = zero_rightmost_1bit(mirrors)) {
6493 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
6496 /* In normal circumstances 'm' will not be NULL. However,
6497 * if mirrors are reconfigured, we can temporarily get out
6498 * of sync in facet_revalidate(). We could "correct" the
6499 * mirror list before reaching here, but doing that would
6500 * not properly account the traffic stats we've currently
6501 * accumulated for previous mirror configuration. */
6505 m->packet_count += packets;
6506 m->byte_count += bytes;
6510 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
6511 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
6512 * indicate this; newer upstream kernels use gratuitous ARP requests. */
6514 is_gratuitous_arp(const struct flow *flow)
6516 return (flow->dl_type == htons(ETH_TYPE_ARP)
6517 && eth_addr_is_broadcast(flow->dl_dst)
6518 && (flow->nw_proto == ARP_OP_REPLY
6519 || (flow->nw_proto == ARP_OP_REQUEST
6520 && flow->nw_src == flow->nw_dst)));
6524 update_learning_table(struct ofproto_dpif *ofproto,
6525 const struct flow *flow, int vlan,
6526 struct ofbundle *in_bundle)
6528 struct mac_entry *mac;
6530 /* Don't learn the OFPP_NONE port. */
6531 if (in_bundle == &ofpp_none_bundle) {
6535 if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
6539 mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan);
6540 if (is_gratuitous_arp(flow)) {
6541 /* We don't want to learn from gratuitous ARP packets that are
6542 * reflected back over bond slaves so we lock the learning table. */
6543 if (!in_bundle->bond) {
6544 mac_entry_set_grat_arp_lock(mac);
6545 } else if (mac_entry_is_grat_arp_locked(mac)) {
6550 if (mac_entry_is_new(mac) || mac->port.p != in_bundle) {
6551 /* The log messages here could actually be useful in debugging,
6552 * so keep the rate limit relatively high. */
6553 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
6554 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
6555 "on port %s in VLAN %d",
6556 ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src),
6557 in_bundle->name, vlan);
6559 mac->port.p = in_bundle;
6560 tag_set_add(&ofproto->revalidate_set,
6561 mac_learning_changed(ofproto->ml, mac));
6565 static struct ofbundle *
6566 lookup_input_bundle(const struct ofproto_dpif *ofproto, uint16_t in_port,
6567 bool warn, struct ofport_dpif **in_ofportp)
6569 struct ofport_dpif *ofport;
6571 /* Find the port and bundle for the received packet. */
6572 ofport = get_ofp_port(ofproto, in_port);
6574 *in_ofportp = ofport;
6576 if (ofport && ofport->bundle) {
6577 return ofport->bundle;
6580 /* Special-case OFPP_NONE, which a controller may use as the ingress
6581 * port for traffic that it is sourcing. */
6582 if (in_port == OFPP_NONE) {
6583 return &ofpp_none_bundle;
6586 /* Odd. A few possible reasons here:
6588 * - We deleted a port but there are still a few packets queued up
6591 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
6592 * we don't know about.
6594 * - The ofproto client didn't configure the port as part of a bundle.
6595 * This is particularly likely to happen if a packet was received on the
6596 * port after it was created, but before the client had a chance to
6597 * configure its bundle.
6600 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6602 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
6603 "port %"PRIu16, ofproto->up.name, in_port);
6608 /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or
6609 * dropped. Returns true if they may be forwarded, false if they should be
6612 * 'in_port' must be the ofport_dpif that corresponds to flow->in_port.
6613 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
6615 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
6616 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
6617 * checked by input_vid_is_valid().
6619 * May also add tags to '*tags', although the current implementation only does
6620 * so in one special case.
6623 is_admissible(struct action_xlate_ctx *ctx, struct ofport_dpif *in_port,
6626 struct ofproto_dpif *ofproto = ctx->ofproto;
6627 struct flow *flow = &ctx->flow;
6628 struct ofbundle *in_bundle = in_port->bundle;
6630 /* Drop frames for reserved multicast addresses
6631 * only if forward_bpdu option is absent. */
6632 if (!ofproto->up.forward_bpdu && eth_addr_is_reserved(flow->dl_dst)) {
6633 xlate_report(ctx, "packet has reserved destination MAC, dropping");
6637 if (in_bundle->bond) {
6638 struct mac_entry *mac;
6640 switch (bond_check_admissibility(in_bundle->bond, in_port,
6641 flow->dl_dst, &ctx->tags)) {
6646 xlate_report(ctx, "bonding refused admissibility, dropping");
6649 case BV_DROP_IF_MOVED:
6650 mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL);
6651 if (mac && mac->port.p != in_bundle &&
6652 (!is_gratuitous_arp(flow)
6653 || mac_entry_is_grat_arp_locked(mac))) {
6654 xlate_report(ctx, "SLB bond thinks this packet looped back, "
6666 xlate_normal(struct action_xlate_ctx *ctx)
6668 struct ofport_dpif *in_port;
6669 struct ofbundle *in_bundle;
6670 struct mac_entry *mac;
6674 ctx->has_normal = true;
6676 in_bundle = lookup_input_bundle(ctx->ofproto, ctx->flow.in_port,
6677 ctx->packet != NULL, &in_port);
6679 xlate_report(ctx, "no input bundle, dropping");
6683 /* Drop malformed frames. */
6684 if (ctx->flow.dl_type == htons(ETH_TYPE_VLAN) &&
6685 !(ctx->flow.vlan_tci & htons(VLAN_CFI))) {
6686 if (ctx->packet != NULL) {
6687 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6688 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
6689 "VLAN tag received on port %s",
6690 ctx->ofproto->up.name, in_bundle->name);
6692 xlate_report(ctx, "partial VLAN tag, dropping");
6696 /* Drop frames on bundles reserved for mirroring. */
6697 if (in_bundle->mirror_out) {
6698 if (ctx->packet != NULL) {
6699 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6700 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
6701 "%s, which is reserved exclusively for mirroring",
6702 ctx->ofproto->up.name, in_bundle->name);
6704 xlate_report(ctx, "input port is mirror output port, dropping");
6709 vid = vlan_tci_to_vid(ctx->flow.vlan_tci);
6710 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
6711 xlate_report(ctx, "disallowed VLAN VID for this input port, dropping");
6714 vlan = input_vid_to_vlan(in_bundle, vid);
6716 /* Check other admissibility requirements. */
6717 if (in_port && !is_admissible(ctx, in_port, vlan)) {
6721 /* Learn source MAC. */
6722 if (ctx->may_learn) {
6723 update_learning_table(ctx->ofproto, &ctx->flow, vlan, in_bundle);
6726 /* Determine output bundle. */
6727 mac = mac_learning_lookup(ctx->ofproto->ml, ctx->flow.dl_dst, vlan,
6730 if (mac->port.p != in_bundle) {
6731 xlate_report(ctx, "forwarding to learned port");
6732 output_normal(ctx, mac->port.p, vlan);
6734 xlate_report(ctx, "learned port is input port, dropping");
6737 struct ofbundle *bundle;
6739 xlate_report(ctx, "no learned MAC for destination, flooding");
6740 HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
6741 if (bundle != in_bundle
6742 && ofbundle_includes_vlan(bundle, vlan)
6743 && bundle->floodable
6744 && !bundle->mirror_out) {
6745 output_normal(ctx, bundle, vlan);
6748 ctx->nf_output_iface = NF_OUT_FLOOD;
6752 /* Optimized flow revalidation.
6754 * It's a difficult problem, in general, to tell which facets need to have
6755 * their actions recalculated whenever the OpenFlow flow table changes. We
6756 * don't try to solve that general problem: for most kinds of OpenFlow flow
6757 * table changes, we recalculate the actions for every facet. This is
6758 * relatively expensive, but it's good enough if the OpenFlow flow table
6759 * doesn't change very often.
6761 * However, we can expect one particular kind of OpenFlow flow table change to
6762 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
6763 * of CPU on revalidating every facet whenever MAC learning modifies the flow
6764 * table, we add a special case that applies to flow tables in which every rule
6765 * has the same form (that is, the same wildcards), except that the table is
6766 * also allowed to have a single "catch-all" flow that matches all packets. We
6767 * optimize this case by tagging all of the facets that resubmit into the table
6768 * and invalidating the same tag whenever a flow changes in that table. The
6769 * end result is that we revalidate just the facets that need it (and sometimes
6770 * a few more, but not all of the facets or even all of the facets that
6771 * resubmit to the table modified by MAC learning). */
6773 /* Calculates the tag to use for 'flow' and mask 'mask' when it is inserted
6774 * into an OpenFlow table with the given 'basis'. */
6776 rule_calculate_tag(const struct flow *flow, const struct minimask *mask,
6779 if (minimask_is_catchall(mask)) {
6782 uint32_t hash = flow_hash_in_minimask(flow, mask, secret);
6783 return tag_create_deterministic(hash);
6787 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
6788 * taggability of that table.
6790 * This function must be called after *each* change to a flow table. If you
6791 * skip calling it on some changes then the pointer comparisons at the end can
6792 * be invalid if you get unlucky. For example, if a flow removal causes a
6793 * cls_table to be destroyed and then a flow insertion causes a cls_table with
6794 * different wildcards to be created with the same address, then this function
6795 * will incorrectly skip revalidation. */
6797 table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
6799 struct table_dpif *table = &ofproto->tables[table_id];
6800 const struct oftable *oftable = &ofproto->up.tables[table_id];
6801 struct cls_table *catchall, *other;
6802 struct cls_table *t;
6804 catchall = other = NULL;
6806 switch (hmap_count(&oftable->cls.tables)) {
6808 /* We could tag this OpenFlow table but it would make the logic a
6809 * little harder and it's a corner case that doesn't seem worth it
6815 HMAP_FOR_EACH (t, hmap_node, &oftable->cls.tables) {
6816 if (cls_table_is_catchall(t)) {
6818 } else if (!other) {
6821 /* Indicate that we can't tag this by setting both tables to
6822 * NULL. (We know that 'catchall' is already NULL.) */
6829 /* Can't tag this table. */
6833 if (table->catchall_table != catchall || table->other_table != other) {
6834 table->catchall_table = catchall;
6835 table->other_table = other;
6836 ofproto->need_revalidate = REV_FLOW_TABLE;
6840 /* Given 'rule' that has changed in some way (either it is a rule being
6841 * inserted, a rule being deleted, or a rule whose actions are being
6842 * modified), marks facets for revalidation to ensure that packets will be
6843 * forwarded correctly according to the new state of the flow table.
6845 * This function must be called after *each* change to a flow table. See
6846 * the comment on table_update_taggable() for more information. */
6848 rule_invalidate(const struct rule_dpif *rule)
6850 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
6852 table_update_taggable(ofproto, rule->up.table_id);
6854 if (!ofproto->need_revalidate) {
6855 struct table_dpif *table = &ofproto->tables[rule->up.table_id];
6857 if (table->other_table && rule->tag) {
6858 tag_set_add(&ofproto->revalidate_set, rule->tag);
6860 ofproto->need_revalidate = REV_FLOW_TABLE;
6866 set_frag_handling(struct ofproto *ofproto_,
6867 enum ofp_config_flags frag_handling)
6869 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
6871 if (frag_handling != OFPC_FRAG_REASM) {
6872 ofproto->need_revalidate = REV_RECONFIGURE;
6880 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
6881 const struct flow *flow,
6882 const struct ofpact *ofpacts, size_t ofpacts_len)
6884 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
6885 struct odputil_keybuf keybuf;
6886 struct dpif_flow_stats stats;
6890 struct action_xlate_ctx ctx;
6891 uint64_t odp_actions_stub[1024 / 8];
6892 struct ofpbuf odp_actions;
6894 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
6895 odp_flow_key_from_flow(&key, flow,
6896 ofp_port_to_odp_port(ofproto, flow->in_port));
6898 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
6900 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci, NULL,
6901 packet_get_tcp_flags(packet, flow), packet);
6902 ctx.resubmit_stats = &stats;
6904 ofpbuf_use_stub(&odp_actions,
6905 odp_actions_stub, sizeof odp_actions_stub);
6906 xlate_actions(&ctx, ofpacts, ofpacts_len, &odp_actions);
6907 dpif_execute(ofproto->backer->dpif, key.data, key.size,
6908 odp_actions.data, odp_actions.size, packet);
6909 ofpbuf_uninit(&odp_actions);
6917 set_netflow(struct ofproto *ofproto_,
6918 const struct netflow_options *netflow_options)
6920 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
6922 if (netflow_options) {
6923 if (!ofproto->netflow) {
6924 ofproto->netflow = netflow_create();
6926 return netflow_set_options(ofproto->netflow, netflow_options);
6928 netflow_destroy(ofproto->netflow);
6929 ofproto->netflow = NULL;
6935 get_netflow_ids(const struct ofproto *ofproto_,
6936 uint8_t *engine_type, uint8_t *engine_id)
6938 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
6940 dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
6944 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
6946 if (!facet_is_controller_flow(facet) &&
6947 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
6948 struct subfacet *subfacet;
6949 struct ofexpired expired;
6951 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
6952 if (subfacet->path == SF_FAST_PATH) {
6953 struct dpif_flow_stats stats;
6955 subfacet_reinstall(subfacet, &stats);
6956 subfacet_update_stats(subfacet, &stats);
6960 expired.flow = facet->flow;
6961 expired.packet_count = facet->packet_count;
6962 expired.byte_count = facet->byte_count;
6963 expired.used = facet->used;
6964 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
6969 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
6971 struct facet *facet;
6973 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
6974 send_active_timeout(ofproto, facet);
6978 static struct ofproto_dpif *
6979 ofproto_dpif_lookup(const char *name)
6981 struct ofproto_dpif *ofproto;
6983 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
6984 hash_string(name, 0), &all_ofproto_dpifs) {
6985 if (!strcmp(ofproto->up.name, name)) {
6993 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
6994 const char *argv[], void *aux OVS_UNUSED)
6996 struct ofproto_dpif *ofproto;
6999 ofproto = ofproto_dpif_lookup(argv[1]);
7001 unixctl_command_reply_error(conn, "no such bridge");
7004 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
7006 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
7007 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
7011 unixctl_command_reply(conn, "table successfully flushed");
7015 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
7016 const char *argv[], void *aux OVS_UNUSED)
7018 struct ds ds = DS_EMPTY_INITIALIZER;
7019 const struct ofproto_dpif *ofproto;
7020 const struct mac_entry *e;
7022 ofproto = ofproto_dpif_lookup(argv[1]);
7024 unixctl_command_reply_error(conn, "no such bridge");
7028 ds_put_cstr(&ds, " port VLAN MAC Age\n");
7029 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
7030 struct ofbundle *bundle = e->port.p;
7031 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
7032 ofbundle_get_a_port(bundle)->odp_port,
7033 e->vlan, ETH_ADDR_ARGS(e->mac),
7034 mac_entry_age(ofproto->ml, e));
7036 unixctl_command_reply(conn, ds_cstr(&ds));
7041 struct action_xlate_ctx ctx;
7047 trace_format_rule(struct ds *result, uint8_t table_id, int level,
7048 const struct rule_dpif *rule)
7050 ds_put_char_multiple(result, '\t', level);
7052 ds_put_cstr(result, "No match\n");
7056 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
7057 table_id, ntohll(rule->up.flow_cookie));
7058 cls_rule_format(&rule->up.cr, result);
7059 ds_put_char(result, '\n');
7061 ds_put_char_multiple(result, '\t', level);
7062 ds_put_cstr(result, "OpenFlow ");
7063 ofpacts_format(rule->up.ofpacts, rule->up.ofpacts_len, result);
7064 ds_put_char(result, '\n');
7068 trace_format_flow(struct ds *result, int level, const char *title,
7069 struct trace_ctx *trace)
7071 ds_put_char_multiple(result, '\t', level);
7072 ds_put_format(result, "%s: ", title);
7073 if (flow_equal(&trace->ctx.flow, &trace->flow)) {
7074 ds_put_cstr(result, "unchanged");
7076 flow_format(result, &trace->ctx.flow);
7077 trace->flow = trace->ctx.flow;
7079 ds_put_char(result, '\n');
7083 trace_format_regs(struct ds *result, int level, const char *title,
7084 struct trace_ctx *trace)
7088 ds_put_char_multiple(result, '\t', level);
7089 ds_put_format(result, "%s:", title);
7090 for (i = 0; i < FLOW_N_REGS; i++) {
7091 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
7093 ds_put_char(result, '\n');
7097 trace_format_odp(struct ds *result, int level, const char *title,
7098 struct trace_ctx *trace)
7100 struct ofpbuf *odp_actions = trace->ctx.odp_actions;
7102 ds_put_char_multiple(result, '\t', level);
7103 ds_put_format(result, "%s: ", title);
7104 format_odp_actions(result, odp_actions->data, odp_actions->size);
7105 ds_put_char(result, '\n');
7109 trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
7111 struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx);
7112 struct ds *result = trace->result;
7114 ds_put_char(result, '\n');
7115 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
7116 trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace);
7117 trace_format_odp(result, ctx->recurse + 1, "Resubmitted odp", trace);
7118 trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule);
7122 trace_report(struct action_xlate_ctx *ctx, const char *s)
7124 struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx);
7125 struct ds *result = trace->result;
7127 ds_put_char_multiple(result, '\t', ctx->recurse);
7128 ds_put_cstr(result, s);
7129 ds_put_char(result, '\n');
7133 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
7134 void *aux OVS_UNUSED)
7136 const char *dpname = argv[1];
7137 struct ofproto_dpif *ofproto;
7138 struct ofpbuf odp_key;
7139 struct ofpbuf *packet;
7140 ovs_be16 initial_tci;
7146 ofpbuf_init(&odp_key, 0);
7149 ofproto = ofproto_dpif_lookup(dpname);
7151 unixctl_command_reply_error(conn, "Unknown ofproto (use ofproto/list "
7155 if (argc == 3 || (argc == 4 && !strcmp(argv[3], "-generate"))) {
7156 /* ofproto/trace dpname flow [-generate] */
7157 const char *flow_s = argv[2];
7158 const char *generate_s = argv[3];
7160 /* Allow 'flow_s' to be either a datapath flow or an OpenFlow-like
7161 * flow. We guess which type it is based on whether 'flow_s' contains
7162 * an '(', since a datapath flow always contains '(') but an
7163 * OpenFlow-like flow should not (in fact it's allowed but I believe
7164 * that's not documented anywhere).
7166 * An alternative would be to try to parse 'flow_s' both ways, but then
7167 * it would be tricky giving a sensible error message. After all, do
7168 * you just say "syntax error" or do you present both error messages?
7169 * Both choices seem lousy. */
7170 if (strchr(flow_s, '(')) {
7171 enum odp_key_fitness fitness;
7174 /* Convert string to datapath key. */
7175 ofpbuf_init(&odp_key, 0);
7176 error = odp_flow_key_from_string(flow_s, NULL, &odp_key);
7178 unixctl_command_reply_error(conn, "Bad flow syntax");
7182 fitness = odp_flow_key_to_flow(odp_key.data, odp_key.size, &flow);
7183 flow.in_port = odp_port_to_ofp_port(ofproto, flow.in_port);
7185 /* Convert odp_key to flow. */
7186 error = ofproto_dpif_vsp_adjust(ofproto, fitness, &flow,
7187 &initial_tci, NULL);
7188 if (error == ODP_FIT_ERROR) {
7189 unixctl_command_reply_error(conn, "Invalid flow");
7195 error_s = parse_ofp_exact_flow(&flow, argv[2]);
7197 unixctl_command_reply_error(conn, error_s);
7202 initial_tci = flow.vlan_tci;
7203 vsp_adjust_flow(ofproto, &flow);
7206 /* Generate a packet, if requested. */
7208 packet = ofpbuf_new(0);
7209 flow_compose(packet, &flow);
7211 } else if (argc == 6) {
7212 /* ofproto/trace dpname priority tun_id in_port packet */
7213 const char *priority_s = argv[2];
7214 const char *tun_id_s = argv[3];
7215 const char *in_port_s = argv[4];
7216 const char *packet_s = argv[5];
7217 uint32_t in_port = atoi(in_port_s);
7218 ovs_be64 tun_id = htonll(strtoull(tun_id_s, NULL, 0));
7219 uint32_t priority = atoi(priority_s);
7222 msg = eth_from_hex(packet_s, &packet);
7224 unixctl_command_reply_error(conn, msg);
7228 ds_put_cstr(&result, "Packet: ");
7229 s = ofp_packet_to_string(packet->data, packet->size);
7230 ds_put_cstr(&result, s);
7233 flow_extract(packet, priority, NULL, in_port, &flow);
7234 flow.tunnel.tun_id = tun_id;
7235 initial_tci = flow.vlan_tci;
7237 unixctl_command_reply_error(conn, "Bad command syntax");
7241 ofproto_trace(ofproto, &flow, packet, initial_tci, &result);
7242 unixctl_command_reply(conn, ds_cstr(&result));
7245 ds_destroy(&result);
7246 ofpbuf_delete(packet);
7247 ofpbuf_uninit(&odp_key);
7251 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
7252 const struct ofpbuf *packet, ovs_be16 initial_tci,
7255 struct rule_dpif *rule;
7257 ds_put_cstr(ds, "Flow: ");
7258 flow_format(ds, flow);
7259 ds_put_char(ds, '\n');
7261 rule = rule_dpif_lookup(ofproto, flow);
7263 trace_format_rule(ds, 0, 0, rule);
7264 if (rule == ofproto->miss_rule) {
7265 ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
7266 } else if (rule == ofproto->no_packet_in_rule) {
7267 ds_put_cstr(ds, "\nNo match, packets dropped because "
7268 "OFPPC_NO_PACKET_IN is set on in_port.\n");
7272 uint64_t odp_actions_stub[1024 / 8];
7273 struct ofpbuf odp_actions;
7275 struct trace_ctx trace;
7278 tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
7281 ofpbuf_use_stub(&odp_actions,
7282 odp_actions_stub, sizeof odp_actions_stub);
7283 action_xlate_ctx_init(&trace.ctx, ofproto, flow, initial_tci,
7284 rule, tcp_flags, packet);
7285 trace.ctx.resubmit_hook = trace_resubmit;
7286 trace.ctx.report_hook = trace_report;
7287 xlate_actions(&trace.ctx, rule->up.ofpacts, rule->up.ofpacts_len,
7290 ds_put_char(ds, '\n');
7291 trace_format_flow(ds, 0, "Final flow", &trace);
7292 ds_put_cstr(ds, "Datapath actions: ");
7293 format_odp_actions(ds, odp_actions.data, odp_actions.size);
7294 ofpbuf_uninit(&odp_actions);
7296 if (trace.ctx.slow) {
7297 enum slow_path_reason slow;
7299 ds_put_cstr(ds, "\nThis flow is handled by the userspace "
7300 "slow path because it:");
7301 for (slow = trace.ctx.slow; slow; ) {
7302 enum slow_path_reason bit = rightmost_1bit(slow);
7306 ds_put_cstr(ds, "\n\t- Consists of CFM packets.");
7309 ds_put_cstr(ds, "\n\t- Consists of LACP packets.");
7312 ds_put_cstr(ds, "\n\t- Consists of STP packets.");
7315 ds_put_cstr(ds, "\n\t- Needs in-band special case "
7318 ds_put_cstr(ds, "\n\t (The datapath actions are "
7319 "incomplete--for complete actions, "
7320 "please supply a packet.)");
7323 case SLOW_CONTROLLER:
7324 ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages "
7325 "to the OpenFlow controller.");
7328 ds_put_cstr(ds, "\n\t- Needs more specific matching "
7329 "than the datapath supports.");
7336 if (slow & ~SLOW_MATCH) {
7337 ds_put_cstr(ds, "\nThe datapath actions above do not reflect "
7338 "the special slow-path processing.");
7345 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
7346 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
7349 unixctl_command_reply(conn, NULL);
7353 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
7354 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
7357 unixctl_command_reply(conn, NULL);
7360 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
7361 * 'reply' describing the results. */
7363 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
7365 struct facet *facet;
7369 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
7370 if (!facet_check_consistency(facet)) {
7375 ofproto->need_revalidate = REV_INCONSISTENCY;
7379 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
7380 ofproto->up.name, errors);
7382 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
7387 ofproto_dpif_self_check(struct unixctl_conn *conn,
7388 int argc, const char *argv[], void *aux OVS_UNUSED)
7390 struct ds reply = DS_EMPTY_INITIALIZER;
7391 struct ofproto_dpif *ofproto;
7394 ofproto = ofproto_dpif_lookup(argv[1]);
7396 unixctl_command_reply_error(conn, "Unknown ofproto (use "
7397 "ofproto/list for help)");
7400 ofproto_dpif_self_check__(ofproto, &reply);
7402 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
7403 ofproto_dpif_self_check__(ofproto, &reply);
7407 unixctl_command_reply(conn, ds_cstr(&reply));
7411 /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list
7412 * of the 'ofproto_shash' nodes. It is the responsibility of the caller
7413 * to destroy 'ofproto_shash' and free the returned value. */
7414 static const struct shash_node **
7415 get_ofprotos(struct shash *ofproto_shash)
7417 const struct ofproto_dpif *ofproto;
7419 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
7420 char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
7421 shash_add_nocopy(ofproto_shash, name, ofproto);
7424 return shash_sort(ofproto_shash);
7428 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
7429 const char *argv[] OVS_UNUSED,
7430 void *aux OVS_UNUSED)
7432 struct ds ds = DS_EMPTY_INITIALIZER;
7433 struct shash ofproto_shash;
7434 const struct shash_node **sorted_ofprotos;
7437 shash_init(&ofproto_shash);
7438 sorted_ofprotos = get_ofprotos(&ofproto_shash);
7439 for (i = 0; i < shash_count(&ofproto_shash); i++) {
7440 const struct shash_node *node = sorted_ofprotos[i];
7441 ds_put_format(&ds, "%s\n", node->name);
7444 shash_destroy(&ofproto_shash);
7445 free(sorted_ofprotos);
7447 unixctl_command_reply(conn, ds_cstr(&ds));
7452 show_dp_format(const struct ofproto_dpif *ofproto, struct ds *ds)
7454 struct dpif_dp_stats s;
7455 const struct shash_node **ports;
7458 dpif_get_dp_stats(ofproto->backer->dpif, &s);
7460 ds_put_format(ds, "%s (%s):\n", ofproto->up.name,
7461 dpif_name(ofproto->backer->dpif));
7462 /* xxx It would be better to show bridge-specific stats instead
7463 * xxx of dp ones. */
7465 "\tlookups: hit:%"PRIu64" missed:%"PRIu64" lost:%"PRIu64"\n",
7466 s.n_hit, s.n_missed, s.n_lost);
7467 ds_put_format(ds, "\tflows: %zu\n",
7468 hmap_count(&ofproto->subfacets));
7470 ports = shash_sort(&ofproto->up.port_by_name);
7471 for (i = 0; i < shash_count(&ofproto->up.port_by_name); i++) {
7472 const struct shash_node *node = ports[i];
7473 struct ofport *ofport = node->data;
7474 const char *name = netdev_get_name(ofport->netdev);
7475 const char *type = netdev_get_type(ofport->netdev);
7477 ds_put_format(ds, "\t%s %u/%u:", name, ofport->ofp_port,
7478 ofp_port_to_odp_port(ofproto, ofport->ofp_port));
7479 if (strcmp(type, "system")) {
7480 struct netdev *netdev;
7483 ds_put_format(ds, " (%s", type);
7485 error = netdev_open(name, type, &netdev);
7490 error = netdev_get_config(netdev, &config);
7492 const struct smap_node **nodes;
7495 nodes = smap_sort(&config);
7496 for (i = 0; i < smap_count(&config); i++) {
7497 const struct smap_node *node = nodes[i];
7498 ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
7499 node->key, node->value);
7503 smap_destroy(&config);
7505 netdev_close(netdev);
7507 ds_put_char(ds, ')');
7509 ds_put_char(ds, '\n');
7515 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc,
7516 const char *argv[], void *aux OVS_UNUSED)
7518 struct ds ds = DS_EMPTY_INITIALIZER;
7519 const struct ofproto_dpif *ofproto;
7523 for (i = 1; i < argc; i++) {
7524 ofproto = ofproto_dpif_lookup(argv[i]);
7526 ds_put_format(&ds, "Unknown bridge %s (use dpif/dump-dps "
7527 "for help)", argv[i]);
7528 unixctl_command_reply_error(conn, ds_cstr(&ds));
7531 show_dp_format(ofproto, &ds);
7534 struct shash ofproto_shash;
7535 const struct shash_node **sorted_ofprotos;
7538 shash_init(&ofproto_shash);
7539 sorted_ofprotos = get_ofprotos(&ofproto_shash);
7540 for (i = 0; i < shash_count(&ofproto_shash); i++) {
7541 const struct shash_node *node = sorted_ofprotos[i];
7542 show_dp_format(node->data, &ds);
7545 shash_destroy(&ofproto_shash);
7546 free(sorted_ofprotos);
7549 unixctl_command_reply(conn, ds_cstr(&ds));
7554 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
7555 int argc OVS_UNUSED, const char *argv[],
7556 void *aux OVS_UNUSED)
7558 struct ds ds = DS_EMPTY_INITIALIZER;
7559 const struct ofproto_dpif *ofproto;
7560 struct subfacet *subfacet;
7562 ofproto = ofproto_dpif_lookup(argv[1]);
7564 unixctl_command_reply_error(conn, "no such bridge");
7568 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
7569 struct odputil_keybuf keybuf;
7572 subfacet_get_key(subfacet, &keybuf, &key);
7573 odp_flow_key_format(key.data, key.size, &ds);
7575 ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:",
7576 subfacet->dp_packet_count, subfacet->dp_byte_count);
7577 if (subfacet->used) {
7578 ds_put_format(&ds, "%.3fs",
7579 (time_msec() - subfacet->used) / 1000.0);
7581 ds_put_format(&ds, "never");
7583 if (subfacet->facet->tcp_flags) {
7584 ds_put_cstr(&ds, ", flags:");
7585 packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags);
7588 ds_put_cstr(&ds, ", actions:");
7589 format_odp_actions(&ds, subfacet->actions, subfacet->actions_len);
7590 ds_put_char(&ds, '\n');
7593 unixctl_command_reply(conn, ds_cstr(&ds));
7598 ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn,
7599 int argc OVS_UNUSED, const char *argv[],
7600 void *aux OVS_UNUSED)
7602 struct ds ds = DS_EMPTY_INITIALIZER;
7603 struct ofproto_dpif *ofproto;
7605 ofproto = ofproto_dpif_lookup(argv[1]);
7607 unixctl_command_reply_error(conn, "no such bridge");
7611 flush(&ofproto->up);
7613 unixctl_command_reply(conn, ds_cstr(&ds));
7618 ofproto_dpif_unixctl_init(void)
7620 static bool registered;
7626 unixctl_command_register(
7628 "bridge {tun_id in_port packet | odp_flow [-generate]}",
7629 2, 5, ofproto_unixctl_trace, NULL);
7630 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
7631 ofproto_unixctl_fdb_flush, NULL);
7632 unixctl_command_register("fdb/show", "bridge", 1, 1,
7633 ofproto_unixctl_fdb_show, NULL);
7634 unixctl_command_register("ofproto/clog", "", 0, 0,
7635 ofproto_dpif_clog, NULL);
7636 unixctl_command_register("ofproto/unclog", "", 0, 0,
7637 ofproto_dpif_unclog, NULL);
7638 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
7639 ofproto_dpif_self_check, NULL);
7640 unixctl_command_register("dpif/dump-dps", "", 0, 0,
7641 ofproto_unixctl_dpif_dump_dps, NULL);
7642 unixctl_command_register("dpif/show", "[bridge]", 0, INT_MAX,
7643 ofproto_unixctl_dpif_show, NULL);
7644 unixctl_command_register("dpif/dump-flows", "bridge", 1, 1,
7645 ofproto_unixctl_dpif_dump_flows, NULL);
7646 unixctl_command_register("dpif/del-flows", "bridge", 1, 1,
7647 ofproto_unixctl_dpif_del_flows, NULL);
7650 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
7652 * This is deprecated. It is only for compatibility with broken device drivers
7653 * in old versions of Linux that do not properly support VLANs when VLAN
7654 * devices are not used. When broken device drivers are no longer in
7655 * widespread use, we will delete these interfaces. */
7658 set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid)
7660 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
7661 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
7663 if (realdev_ofp_port == ofport->realdev_ofp_port
7664 && vid == ofport->vlandev_vid) {
7668 ofproto->need_revalidate = REV_RECONFIGURE;
7670 if (ofport->realdev_ofp_port) {
7673 if (realdev_ofp_port && ofport->bundle) {
7674 /* vlandevs are enslaved to their realdevs, so they are not allowed to
7675 * themselves be part of a bundle. */
7676 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
7679 ofport->realdev_ofp_port = realdev_ofp_port;
7680 ofport->vlandev_vid = vid;
7682 if (realdev_ofp_port) {
7683 vsp_add(ofport, realdev_ofp_port, vid);
7690 hash_realdev_vid(uint16_t realdev_ofp_port, int vid)
7692 return hash_2words(realdev_ofp_port, vid);
7695 /* Returns the ODP port number of the Linux VLAN device that corresponds to
7696 * 'vlan_tci' on the network device with port number 'realdev_odp_port' in
7697 * 'ofproto'. For example, given 'realdev_odp_port' of eth0 and 'vlan_tci' 9,
7698 * it would return the port number of eth0.9.
7700 * Unless VLAN splinters are enabled for port 'realdev_odp_port', this
7701 * function just returns its 'realdev_odp_port' argument. */
7703 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
7704 uint32_t realdev_odp_port, ovs_be16 vlan_tci)
7706 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
7707 uint16_t realdev_ofp_port;
7708 int vid = vlan_tci_to_vid(vlan_tci);
7709 const struct vlan_splinter *vsp;
7711 realdev_ofp_port = odp_port_to_ofp_port(ofproto, realdev_odp_port);
7712 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
7713 hash_realdev_vid(realdev_ofp_port, vid),
7714 &ofproto->realdev_vid_map) {
7715 if (vsp->realdev_ofp_port == realdev_ofp_port
7716 && vsp->vid == vid) {
7717 return ofp_port_to_odp_port(ofproto, vsp->vlandev_ofp_port);
7721 return realdev_odp_port;
7724 static struct vlan_splinter *
7725 vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port)
7727 struct vlan_splinter *vsp;
7729 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, hash_int(vlandev_ofp_port, 0),
7730 &ofproto->vlandev_map) {
7731 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
7739 /* Returns the OpenFlow port number of the "real" device underlying the Linux
7740 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
7741 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
7742 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
7743 * eth0 and store 9 in '*vid'.
7745 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
7746 * VLAN device. Unless VLAN splinters are enabled, this is what this function
7749 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
7750 uint16_t vlandev_ofp_port, int *vid)
7752 if (!hmap_is_empty(&ofproto->vlandev_map)) {
7753 const struct vlan_splinter *vsp;
7755 vsp = vlandev_find(ofproto, vlandev_ofp_port);
7760 return vsp->realdev_ofp_port;
7766 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
7767 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
7768 * 'flow->in_port' to the "real" device backing the VLAN device, sets
7769 * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is
7770 * always the case unless VLAN splinters are enabled), returns false without
7771 * making any changes. */
7773 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
7778 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port, &vid);
7783 /* Cause the flow to be processed as if it came in on the real device with
7784 * the VLAN device's VLAN ID. */
7785 flow->in_port = realdev;
7786 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
7791 vsp_remove(struct ofport_dpif *port)
7793 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
7794 struct vlan_splinter *vsp;
7796 vsp = vlandev_find(ofproto, port->up.ofp_port);
7798 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
7799 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
7802 port->realdev_ofp_port = 0;
7804 VLOG_ERR("missing vlan device record");
7809 vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid)
7811 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
7813 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
7814 && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid))
7815 == realdev_ofp_port)) {
7816 struct vlan_splinter *vsp;
7818 vsp = xmalloc(sizeof *vsp);
7819 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
7820 hash_int(port->up.ofp_port, 0));
7821 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
7822 hash_realdev_vid(realdev_ofp_port, vid));
7823 vsp->realdev_ofp_port = realdev_ofp_port;
7824 vsp->vlandev_ofp_port = port->up.ofp_port;
7827 port->realdev_ofp_port = realdev_ofp_port;
7829 VLOG_ERR("duplicate vlan device record");
7834 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
7836 const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
7837 return ofport ? ofport->odp_port : OVSP_NONE;
7840 static struct ofport_dpif *
7841 odp_port_to_ofport(const struct dpif_backer *backer, uint32_t odp_port)
7843 struct ofport_dpif *port;
7845 HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node,
7846 hash_int(odp_port, 0),
7847 &backer->odp_to_ofport_map) {
7848 if (port->odp_port == odp_port) {
7857 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
7859 struct ofport_dpif *port;
7861 port = odp_port_to_ofport(ofproto->backer, odp_port);
7862 if (port && ofproto == ofproto_dpif_cast(port->up.ofproto)) {
7863 return port->up.ofp_port;
7869 const struct ofproto_class ofproto_dpif_class = {
7904 port_is_lacp_current,
7905 NULL, /* rule_choose_table */
7912 rule_modify_actions,
7921 get_cfm_remote_mpids,
7926 get_stp_port_status,
7933 is_mirror_output_bundle,
7934 forward_bpdu_changed,