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-print.h"
46 #include "ofproto-dpif-governor.h"
47 #include "ofproto-dpif-sflow.h"
48 #include "poll-loop.h"
50 #include "unaligned.h"
52 #include "vlan-bitmap.h"
55 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
57 COVERAGE_DEFINE(ofproto_dpif_ctlr_action);
58 COVERAGE_DEFINE(ofproto_dpif_expired);
59 COVERAGE_DEFINE(ofproto_dpif_no_packet_in);
60 COVERAGE_DEFINE(ofproto_dpif_xlate);
61 COVERAGE_DEFINE(facet_changed_rule);
62 COVERAGE_DEFINE(facet_invalidated);
63 COVERAGE_DEFINE(facet_revalidate);
64 COVERAGE_DEFINE(facet_unexpected);
65 COVERAGE_DEFINE(facet_suppress);
67 /* Maximum depth of flow table recursion (due to resubmit actions) in a
68 * flow translation. */
69 #define MAX_RESUBMIT_RECURSION 32
71 /* Number of implemented OpenFlow tables. */
72 enum { N_TABLES = 255 };
73 BUILD_ASSERT_DECL(N_TABLES >= 1 && N_TABLES <= 255);
83 * - Do include packets and bytes from facets that have been deleted or
84 * whose own statistics have been folded into the rule.
86 * - Do include packets and bytes sent "by hand" that were accounted to
87 * the rule without any facet being involved (this is a rare corner
88 * case in rule_execute()).
90 * - Do not include packet or bytes that can be obtained from any facet's
91 * packet_count or byte_count member or that can be obtained from the
92 * datapath by, e.g., dpif_flow_get() for any subfacet.
94 uint64_t packet_count; /* Number of packets received. */
95 uint64_t byte_count; /* Number of bytes received. */
97 tag_type tag; /* Caches rule_calculate_tag() result. */
99 struct list facets; /* List of "struct facet"s. */
102 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
104 return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
107 static struct rule_dpif *rule_dpif_lookup(struct ofproto_dpif *,
108 const struct flow *, uint8_t table);
110 static void rule_credit_stats(struct rule_dpif *,
111 const struct dpif_flow_stats *);
112 static void flow_push_stats(struct rule_dpif *, const struct flow *,
113 const struct dpif_flow_stats *);
114 static tag_type rule_calculate_tag(const struct flow *,
115 const struct flow_wildcards *,
117 static void rule_invalidate(const struct rule_dpif *);
119 #define MAX_MIRRORS 32
120 typedef uint32_t mirror_mask_t;
121 #define MIRROR_MASK_C(X) UINT32_C(X)
122 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
124 struct ofproto_dpif *ofproto; /* Owning ofproto. */
125 size_t idx; /* In ofproto's "mirrors" array. */
126 void *aux; /* Key supplied by ofproto's client. */
127 char *name; /* Identifier for log messages. */
129 /* Selection criteria. */
130 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
131 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
132 unsigned long *vlans; /* Bitmap of chosen VLANs, NULL selects all. */
134 /* Output (exactly one of out == NULL and out_vlan == -1 is true). */
135 struct ofbundle *out; /* Output port or NULL. */
136 int out_vlan; /* Output VLAN or -1. */
137 mirror_mask_t dup_mirrors; /* Bitmap of mirrors with the same output. */
140 int64_t packet_count; /* Number of packets sent. */
141 int64_t byte_count; /* Number of bytes sent. */
144 static void mirror_destroy(struct ofmirror *);
145 static void update_mirror_stats(struct ofproto_dpif *ofproto,
146 mirror_mask_t mirrors,
147 uint64_t packets, uint64_t bytes);
150 struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
151 struct ofproto_dpif *ofproto; /* Owning ofproto. */
152 void *aux; /* Key supplied by ofproto's client. */
153 char *name; /* Identifier for log messages. */
156 struct list ports; /* Contains "struct ofport"s. */
157 enum port_vlan_mode vlan_mode; /* VLAN mode */
158 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
159 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
160 * NULL if all VLANs are trunked. */
161 struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */
162 struct bond *bond; /* Nonnull iff more than one port. */
163 bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */
166 bool floodable; /* True if no port has OFPUTIL_PC_NO_FLOOD set. */
168 /* Port mirroring info. */
169 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
170 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
171 mirror_mask_t mirror_out; /* Mirrors that output to this bundle. */
174 static void bundle_remove(struct ofport *);
175 static void bundle_update(struct ofbundle *);
176 static void bundle_destroy(struct ofbundle *);
177 static void bundle_del_port(struct ofport_dpif *);
178 static void bundle_run(struct ofbundle *);
179 static void bundle_wait(struct ofbundle *);
180 static struct ofbundle *lookup_input_bundle(struct ofproto_dpif *,
181 uint16_t in_port, bool warn,
182 struct ofport_dpif **in_ofportp);
184 /* A controller may use OFPP_NONE as the ingress port to indicate that
185 * it did not arrive on a "real" port. 'ofpp_none_bundle' exists for
186 * when an input bundle is needed for validation (e.g., mirroring or
187 * OFPP_NORMAL processing). It is not connected to an 'ofproto' or have
188 * any 'port' structs, so care must be taken when dealing with it. */
189 static struct ofbundle ofpp_none_bundle = {
191 .vlan_mode = PORT_VLAN_TRUNK
194 static void stp_run(struct ofproto_dpif *ofproto);
195 static void stp_wait(struct ofproto_dpif *ofproto);
196 static int set_stp_port(struct ofport *,
197 const struct ofproto_port_stp_settings *);
199 static bool ofbundle_includes_vlan(const struct ofbundle *, uint16_t vlan);
201 struct action_xlate_ctx {
202 /* action_xlate_ctx_init() initializes these members. */
205 struct ofproto_dpif *ofproto;
207 /* Flow to which the OpenFlow actions apply. xlate_actions() will modify
208 * this flow when actions change header fields. */
211 /* The packet corresponding to 'flow', or a null pointer if we are
212 * revalidating without a packet to refer to. */
213 const struct ofpbuf *packet;
215 /* Should OFPP_NORMAL update the MAC learning table? Should "learn"
216 * actions update the flow table?
218 * We want to update these tables if we are actually processing a packet,
219 * or if we are accounting for packets that the datapath has processed, but
220 * not if we are just revalidating. */
223 /* The rule that we are currently translating, or NULL. */
224 struct rule_dpif *rule;
226 /* Union of the set of TCP flags seen so far in this flow. (Used only by
227 * NXAST_FIN_TIMEOUT. Set to zero to avoid updating updating rules'
231 /* If nonnull, flow translation calls this function just before executing a
232 * resubmit or OFPP_TABLE action. In addition, disables logging of traces
233 * when the recursion depth is exceeded.
235 * 'rule' is the rule being submitted into. It will be null if the
236 * resubmit or OFPP_TABLE action didn't find a matching rule.
238 * This is normally null so the client has to set it manually after
239 * calling action_xlate_ctx_init(). */
240 void (*resubmit_hook)(struct action_xlate_ctx *, struct rule_dpif *rule);
242 /* If nonnull, flow translation credits the specified statistics to each
243 * rule reached through a resubmit or OFPP_TABLE action.
245 * This is normally null so the client has to set it manually after
246 * calling action_xlate_ctx_init(). */
247 const struct dpif_flow_stats *resubmit_stats;
249 /* xlate_actions() initializes and uses these members. The client might want
250 * to look at them after it returns. */
252 struct ofpbuf *odp_actions; /* Datapath actions. */
253 tag_type tags; /* Tags associated with actions. */
254 bool may_set_up_flow; /* True ordinarily; false if the actions must
255 * be reassessed for every packet. */
256 bool has_learn; /* Actions include NXAST_LEARN? */
257 bool has_normal; /* Actions output to OFPP_NORMAL? */
258 bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */
259 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
260 mirror_mask_t mirrors; /* Bitmap of associated mirrors. */
262 /* xlate_actions() initializes and uses these members, but the client has no
263 * reason to look at them. */
265 int recurse; /* Recursion level, via xlate_table_action. */
266 bool max_resubmit_trigger; /* Recursed too deeply during translation. */
267 struct flow base_flow; /* Flow at the last commit. */
268 uint32_t orig_skb_priority; /* Priority when packet arrived. */
269 uint8_t table_id; /* OpenFlow table ID where flow was found. */
270 uint32_t sflow_n_outputs; /* Number of output ports. */
271 uint16_t sflow_odp_port; /* Output port for composing sFlow action. */
272 uint16_t user_cookie_offset;/* Used for user_action_cookie fixup. */
273 bool exit; /* No further actions should be processed. */
274 struct flow orig_flow; /* Copy of original flow. */
277 static void action_xlate_ctx_init(struct action_xlate_ctx *,
278 struct ofproto_dpif *, const struct flow *,
279 ovs_be16 initial_tci, struct rule_dpif *,
280 uint8_t tcp_flags, const struct ofpbuf *);
281 static void xlate_actions(struct action_xlate_ctx *,
282 const union ofp_action *in, size_t n_in,
283 struct ofpbuf *odp_actions);
284 static void xlate_actions_for_side_effects(struct action_xlate_ctx *,
285 const union ofp_action *in,
288 /* A dpif flow and actions associated with a facet.
290 * See also the large comment on struct facet. */
293 struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */
294 struct list list_node; /* In struct facet's 'facets' list. */
295 struct facet *facet; /* Owning facet. */
299 * To save memory in the common case, 'key' is NULL if 'key_fitness' is
300 * ODP_FIT_PERFECT, that is, odp_flow_key_from_flow() can accurately
301 * regenerate the ODP flow key from ->facet->flow. */
302 enum odp_key_fitness key_fitness;
306 long long int used; /* Time last used; time created if not used. */
308 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
309 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
313 * These should be essentially identical for every subfacet in a facet, but
314 * may differ in trivial ways due to VLAN splinters. */
315 size_t actions_len; /* Number of bytes in actions[]. */
316 struct nlattr *actions; /* Datapath actions. */
318 bool installed; /* Installed in datapath? */
320 /* This value is normally the same as ->facet->flow.vlan_tci. Only VLAN
321 * splinters can cause it to differ. This value should be removed when
322 * the VLAN splinters feature is no longer needed. */
323 ovs_be16 initial_tci; /* Initial VLAN TCI value. */
326 static struct subfacet *subfacet_create(struct facet *, enum odp_key_fitness,
327 const struct nlattr *key,
328 size_t key_len, ovs_be16 initial_tci);
329 static struct subfacet *subfacet_find(struct ofproto_dpif *,
330 const struct nlattr *key, size_t key_len);
331 static void subfacet_destroy(struct subfacet *);
332 static void subfacet_destroy__(struct subfacet *);
333 static void subfacet_get_key(struct subfacet *, struct odputil_keybuf *,
335 static void subfacet_reset_dp_stats(struct subfacet *,
336 struct dpif_flow_stats *);
337 static void subfacet_update_time(struct subfacet *, long long int used);
338 static void subfacet_update_stats(struct subfacet *,
339 const struct dpif_flow_stats *);
340 static void subfacet_make_actions(struct subfacet *,
341 const struct ofpbuf *packet,
342 struct ofpbuf *odp_actions);
343 static int subfacet_install(struct subfacet *,
344 const struct nlattr *actions, size_t actions_len,
345 struct dpif_flow_stats *);
346 static void subfacet_uninstall(struct subfacet *);
348 /* An exact-match instantiation of an OpenFlow flow.
350 * A facet associates a "struct flow", which represents the Open vSwitch
351 * userspace idea of an exact-match flow, with one or more subfacets. Each
352 * subfacet tracks the datapath's idea of the exact-match flow equivalent to
353 * the facet. When the kernel module (or other dpif implementation) and Open
354 * vSwitch userspace agree on the definition of a flow key, there is exactly
355 * one subfacet per facet. If the dpif implementation supports more-specific
356 * flow matching than userspace, however, a facet can have more than one
357 * subfacet, each of which corresponds to some distinction in flow that
358 * userspace simply doesn't understand.
360 * Flow expiration works in terms of subfacets, so a facet must have at least
361 * one subfacet or it will never expire, leaking memory. */
364 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
365 struct list list_node; /* In owning rule's 'facets' list. */
366 struct rule_dpif *rule; /* Owning rule. */
369 struct list subfacets;
370 long long int used; /* Time last used; time created if not used. */
377 * - Do include packets and bytes sent "by hand", e.g. with
380 * - Do include packets and bytes that were obtained from the datapath
381 * when a subfacet's statistics were reset (e.g. dpif_flow_put() with
382 * DPIF_FP_ZERO_STATS).
384 * - Do not include packets or bytes that can be obtained from the
385 * datapath for any existing subfacet.
387 uint64_t packet_count; /* Number of packets received. */
388 uint64_t byte_count; /* Number of bytes received. */
390 /* Resubmit statistics. */
391 uint64_t prev_packet_count; /* Number of packets from last stats push. */
392 uint64_t prev_byte_count; /* Number of bytes from last stats push. */
393 long long int prev_used; /* Used time from last stats push. */
396 uint64_t accounted_bytes; /* Bytes processed by facet_account(). */
397 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
398 uint8_t tcp_flags; /* TCP flags seen for this 'rule'. */
400 /* Properties of datapath actions.
402 * Every subfacet has its own actions because actions can differ slightly
403 * between splintered and non-splintered subfacets due to the VLAN tag
404 * being initially different (present vs. absent). All of them have these
405 * properties in common so we just store one copy of them here. */
406 bool may_install; /* Reassess actions for every packet? */
407 bool has_learn; /* Actions include NXAST_LEARN? */
408 bool has_normal; /* Actions output to OFPP_NORMAL? */
409 bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */
410 tag_type tags; /* Tags that would require revalidation. */
411 mirror_mask_t mirrors; /* Bitmap of dependent mirrors. */
413 /* Storage for a single subfacet, to reduce malloc() time and space
414 * overhead. (A facet always has at least one subfacet and in the common
415 * case has exactly one subfacet.) */
416 struct subfacet one_subfacet;
419 static struct facet *facet_create(struct rule_dpif *,
420 const struct flow *, uint32_t hash);
421 static void facet_remove(struct facet *);
422 static void facet_free(struct facet *);
424 static struct facet *facet_find(struct ofproto_dpif *,
425 const struct flow *, uint32_t hash);
426 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
427 const struct flow *, uint32_t hash);
428 static bool facet_revalidate(struct facet *);
429 static bool facet_check_consistency(struct facet *);
431 static void facet_flush_stats(struct facet *);
433 static void facet_update_time(struct facet *, long long int used);
434 static void facet_reset_counters(struct facet *);
435 static void facet_push_stats(struct facet *);
436 static void facet_learn(struct facet *);
437 static void facet_account(struct facet *);
439 static bool facet_is_controller_flow(struct facet *);
445 struct ofbundle *bundle; /* Bundle that contains this port, if any. */
446 struct list bundle_node; /* In struct ofbundle's "ports" list. */
447 struct cfm *cfm; /* Connectivity Fault Management, if any. */
448 tag_type tag; /* Tag associated with this port. */
449 uint32_t bond_stable_id; /* stable_id to use as bond slave, or 0. */
450 bool may_enable; /* May be enabled in bonds. */
451 long long int carrier_seq; /* Carrier status changes. */
454 struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */
455 enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */
456 long long int stp_state_entered;
458 struct hmap priorities; /* Map of attached 'priority_to_dscp's. */
460 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
462 * This is deprecated. It is only for compatibility with broken device
463 * drivers in old versions of Linux that do not properly support VLANs when
464 * VLAN devices are not used. When broken device drivers are no longer in
465 * widespread use, we will delete these interfaces. */
466 uint16_t realdev_ofp_port;
470 /* Node in 'ofport_dpif''s 'priorities' map. Used to maintain a map from
471 * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
472 * traffic egressing the 'ofport' with that priority should be marked with. */
473 struct priority_to_dscp {
474 struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'priorities' map. */
475 uint32_t priority; /* Priority of this queue (see struct flow). */
477 uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */
480 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
482 * This is deprecated. It is only for compatibility with broken device drivers
483 * in old versions of Linux that do not properly support VLANs when VLAN
484 * devices are not used. When broken device drivers are no longer in
485 * widespread use, we will delete these interfaces. */
486 struct vlan_splinter {
487 struct hmap_node realdev_vid_node;
488 struct hmap_node vlandev_node;
489 uint16_t realdev_ofp_port;
490 uint16_t vlandev_ofp_port;
494 static uint32_t vsp_realdev_to_vlandev(const struct ofproto_dpif *,
495 uint32_t realdev, ovs_be16 vlan_tci);
496 static uint16_t vsp_vlandev_to_realdev(const struct ofproto_dpif *,
497 uint16_t vlandev, int *vid);
498 static bool vsp_adjust_flow(const struct ofproto_dpif *, struct flow *);
499 static void vsp_remove(struct ofport_dpif *);
500 static void vsp_add(struct ofport_dpif *, uint16_t realdev_ofp_port, int vid);
502 static struct ofport_dpif *
503 ofport_dpif_cast(const struct ofport *ofport)
505 assert(ofport->ofproto->ofproto_class == &ofproto_dpif_class);
506 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
509 static void port_run(struct ofport_dpif *);
510 static void port_wait(struct ofport_dpif *);
511 static int set_cfm(struct ofport *, const struct cfm_settings *);
512 static void ofport_clear_priorities(struct ofport_dpif *);
514 struct dpif_completion {
515 struct list list_node;
516 struct ofoperation *op;
519 /* Extra information about a classifier table.
520 * Currently used just for optimized flow revalidation. */
522 /* If either of these is nonnull, then this table has a form that allows
523 * flows to be tagged to avoid revalidating most flows for the most common
524 * kinds of flow table changes. */
525 struct cls_table *catchall_table; /* Table that wildcards all fields. */
526 struct cls_table *other_table; /* Table with any other wildcard set. */
527 uint32_t basis; /* Keeps each table's tags separate. */
530 struct ofproto_dpif {
531 struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
540 struct netflow *netflow;
541 struct dpif_sflow *sflow;
542 struct hmap bundles; /* Contains "struct ofbundle"s. */
543 struct mac_learning *ml;
544 struct ofmirror *mirrors[MAX_MIRRORS];
546 bool has_bonded_bundles;
549 struct timer next_expiration;
553 struct hmap subfacets;
554 struct governor *governor;
557 struct table_dpif tables[N_TABLES];
558 bool need_revalidate;
559 struct tag_set revalidate_set;
561 /* Support for debugging async flow mods. */
562 struct list completions;
564 bool has_bundle_action; /* True when the first bundle action appears. */
565 struct netdev_stats stats; /* To account packets generated and consumed in
570 long long int stp_last_tick;
572 /* VLAN splinters. */
573 struct hmap realdev_vid_map; /* (realdev,vid) -> vlandev. */
574 struct hmap vlandev_map; /* vlandev -> (realdev,vid). */
577 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
578 * for debugging the asynchronous flow_mod implementation.) */
581 /* All existing ofproto_dpif instances, indexed by ->up.name. */
582 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
584 static void ofproto_dpif_unixctl_init(void);
586 static struct ofproto_dpif *
587 ofproto_dpif_cast(const struct ofproto *ofproto)
589 assert(ofproto->ofproto_class == &ofproto_dpif_class);
590 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
593 static struct ofport_dpif *get_ofp_port(struct ofproto_dpif *,
595 static struct ofport_dpif *get_odp_port(struct ofproto_dpif *,
597 static void ofproto_trace(struct ofproto_dpif *, const struct flow *,
598 const struct ofpbuf *, ovs_be16 initial_tci,
601 /* Packet processing. */
602 static void update_learning_table(struct ofproto_dpif *,
603 const struct flow *, int vlan,
606 #define FLOW_MISS_MAX_BATCH 50
607 static int handle_upcalls(struct ofproto_dpif *, unsigned int max_batch);
609 /* Flow expiration. */
610 static int expire(struct ofproto_dpif *);
613 static void send_netflow_active_timeouts(struct ofproto_dpif *);
616 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
618 compose_sflow_action(const struct ofproto_dpif *, struct ofpbuf *odp_actions,
619 const struct flow *, uint32_t odp_port);
620 static void add_mirror_actions(struct action_xlate_ctx *ctx,
621 const struct flow *flow);
622 /* Global variables. */
623 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
625 /* Factory functions. */
628 enumerate_types(struct sset *types)
630 dp_enumerate_types(types);
634 enumerate_names(const char *type, struct sset *names)
636 return dp_enumerate_names(type, names);
640 del(const char *type, const char *name)
645 error = dpif_open(name, type, &dpif);
647 error = dpif_delete(dpif);
653 /* Basic life-cycle. */
655 static struct ofproto *
658 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
663 dealloc(struct ofproto *ofproto_)
665 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
670 construct(struct ofproto *ofproto_)
672 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
673 const char *name = ofproto->up.name;
677 error = dpif_create_and_open(name, ofproto->up.type, &ofproto->dpif);
679 VLOG_ERR("failed to open datapath %s: %s", name, strerror(error));
683 ofproto->max_ports = dpif_get_max_ports(ofproto->dpif);
684 ofproto->n_matches = 0;
686 dpif_flow_flush(ofproto->dpif);
687 dpif_recv_purge(ofproto->dpif);
689 error = dpif_recv_set(ofproto->dpif, true);
691 VLOG_ERR("failed to listen on datapath %s: %s", name, strerror(error));
692 dpif_close(ofproto->dpif);
696 ofproto->netflow = NULL;
697 ofproto->sflow = NULL;
699 hmap_init(&ofproto->bundles);
700 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
701 for (i = 0; i < MAX_MIRRORS; i++) {
702 ofproto->mirrors[i] = NULL;
704 ofproto->has_bonded_bundles = false;
706 timer_set_duration(&ofproto->next_expiration, 1000);
708 hmap_init(&ofproto->facets);
709 hmap_init(&ofproto->subfacets);
710 ofproto->governor = NULL;
712 for (i = 0; i < N_TABLES; i++) {
713 struct table_dpif *table = &ofproto->tables[i];
715 table->catchall_table = NULL;
716 table->other_table = NULL;
717 table->basis = random_uint32();
719 ofproto->need_revalidate = false;
720 tag_set_init(&ofproto->revalidate_set);
722 list_init(&ofproto->completions);
724 ofproto_dpif_unixctl_init();
726 ofproto->has_mirrors = false;
727 ofproto->has_bundle_action = false;
729 hmap_init(&ofproto->vlandev_map);
730 hmap_init(&ofproto->realdev_vid_map);
732 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
733 hash_string(ofproto->up.name, 0));
734 memset(&ofproto->stats, 0, sizeof ofproto->stats);
736 ofproto_init_tables(ofproto_, N_TABLES);
742 complete_operations(struct ofproto_dpif *ofproto)
744 struct dpif_completion *c, *next;
746 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
747 ofoperation_complete(c->op, 0);
748 list_remove(&c->list_node);
754 destruct(struct ofproto *ofproto_)
756 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
757 struct rule_dpif *rule, *next_rule;
758 struct oftable *table;
761 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
762 complete_operations(ofproto);
764 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
765 struct cls_cursor cursor;
767 cls_cursor_init(&cursor, &table->cls, NULL);
768 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
769 ofproto_rule_destroy(&rule->up);
773 for (i = 0; i < MAX_MIRRORS; i++) {
774 mirror_destroy(ofproto->mirrors[i]);
777 netflow_destroy(ofproto->netflow);
778 dpif_sflow_destroy(ofproto->sflow);
779 hmap_destroy(&ofproto->bundles);
780 mac_learning_destroy(ofproto->ml);
782 hmap_destroy(&ofproto->facets);
783 hmap_destroy(&ofproto->subfacets);
784 governor_destroy(ofproto->governor);
786 hmap_destroy(&ofproto->vlandev_map);
787 hmap_destroy(&ofproto->realdev_vid_map);
789 dpif_close(ofproto->dpif);
793 run_fast(struct ofproto *ofproto_)
795 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
798 /* Handle one or more batches of upcalls, until there's nothing left to do
799 * or until we do a fixed total amount of work.
801 * We do work in batches because it can be much cheaper to set up a number
802 * of flows and fire off their patches all at once. We do multiple batches
803 * because in some cases handling a packet can cause another packet to be
804 * queued almost immediately as part of the return flow. Both
805 * optimizations can make major improvements on some benchmarks and
806 * presumably for real traffic as well. */
808 while (work < FLOW_MISS_MAX_BATCH) {
809 int retval = handle_upcalls(ofproto, FLOW_MISS_MAX_BATCH - work);
819 run(struct ofproto *ofproto_)
821 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
822 struct ofport_dpif *ofport;
823 struct ofbundle *bundle;
827 complete_operations(ofproto);
829 dpif_run(ofproto->dpif);
831 error = run_fast(ofproto_);
836 if (timer_expired(&ofproto->next_expiration)) {
837 int delay = expire(ofproto);
838 timer_set_duration(&ofproto->next_expiration, delay);
841 if (ofproto->netflow) {
842 if (netflow_run(ofproto->netflow)) {
843 send_netflow_active_timeouts(ofproto);
846 if (ofproto->sflow) {
847 dpif_sflow_run(ofproto->sflow);
850 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
853 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
858 mac_learning_run(ofproto->ml, &ofproto->revalidate_set);
860 /* Now revalidate if there's anything to do. */
861 if (ofproto->need_revalidate
862 || !tag_set_is_empty(&ofproto->revalidate_set)) {
863 struct tag_set revalidate_set = ofproto->revalidate_set;
864 bool revalidate_all = ofproto->need_revalidate;
865 struct facet *facet, *next;
867 /* Clear the revalidation flags. */
868 tag_set_init(&ofproto->revalidate_set);
869 ofproto->need_revalidate = false;
871 HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &ofproto->facets) {
873 || tag_set_intersects(&revalidate_set, facet->tags)) {
874 facet_revalidate(facet);
879 /* Check the consistency of a random facet, to aid debugging. */
880 if (!hmap_is_empty(&ofproto->facets) && !ofproto->need_revalidate) {
883 facet = CONTAINER_OF(hmap_random_node(&ofproto->facets),
884 struct facet, hmap_node);
885 if (!tag_set_intersects(&ofproto->revalidate_set, facet->tags)) {
886 if (!facet_check_consistency(facet)) {
887 ofproto->need_revalidate = true;
892 if (ofproto->governor) {
895 governor_run(ofproto->governor);
897 /* If the governor has shrunk to its minimum size and the number of
898 * subfacets has dwindled, then drop the governor entirely.
900 * For hysteresis, the number of subfacets to drop the governor is
901 * smaller than the number needed to trigger its creation. */
902 n_subfacets = hmap_count(&ofproto->subfacets);
903 if (n_subfacets * 4 < ofproto->up.flow_eviction_threshold
904 && governor_is_idle(ofproto->governor)) {
905 governor_destroy(ofproto->governor);
906 ofproto->governor = NULL;
914 wait(struct ofproto *ofproto_)
916 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
917 struct ofport_dpif *ofport;
918 struct ofbundle *bundle;
920 if (!clogged && !list_is_empty(&ofproto->completions)) {
921 poll_immediate_wake();
924 dpif_wait(ofproto->dpif);
925 dpif_recv_wait(ofproto->dpif);
926 if (ofproto->sflow) {
927 dpif_sflow_wait(ofproto->sflow);
929 if (!tag_set_is_empty(&ofproto->revalidate_set)) {
930 poll_immediate_wake();
932 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
935 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
938 if (ofproto->netflow) {
939 netflow_wait(ofproto->netflow);
941 mac_learning_wait(ofproto->ml);
943 if (ofproto->need_revalidate) {
944 /* Shouldn't happen, but if it does just go around again. */
945 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
946 poll_immediate_wake();
948 timer_wait(&ofproto->next_expiration);
950 if (ofproto->governor) {
951 governor_wait(ofproto->governor);
956 flush(struct ofproto *ofproto_)
958 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
959 struct facet *facet, *next_facet;
961 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
962 /* Mark the facet as not installed so that facet_remove() doesn't
963 * bother trying to uninstall it. There is no point in uninstalling it
964 * individually since we are about to blow away all the facets with
965 * dpif_flow_flush(). */
966 struct subfacet *subfacet;
968 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
969 subfacet->installed = false;
970 subfacet->dp_packet_count = 0;
971 subfacet->dp_byte_count = 0;
975 dpif_flow_flush(ofproto->dpif);
979 get_features(struct ofproto *ofproto_ OVS_UNUSED,
980 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
982 *arp_match_ip = true;
983 *actions = (OFPUTIL_A_OUTPUT |
984 OFPUTIL_A_SET_VLAN_VID |
985 OFPUTIL_A_SET_VLAN_PCP |
986 OFPUTIL_A_STRIP_VLAN |
987 OFPUTIL_A_SET_DL_SRC |
988 OFPUTIL_A_SET_DL_DST |
989 OFPUTIL_A_SET_NW_SRC |
990 OFPUTIL_A_SET_NW_DST |
991 OFPUTIL_A_SET_NW_TOS |
992 OFPUTIL_A_SET_TP_SRC |
993 OFPUTIL_A_SET_TP_DST |
998 get_tables(struct ofproto *ofproto_, struct ofp_table_stats *ots)
1000 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1001 struct dpif_dp_stats s;
1003 strcpy(ots->name, "classifier");
1005 dpif_get_dp_stats(ofproto->dpif, &s);
1006 put_32aligned_be64(&ots->lookup_count, htonll(s.n_hit + s.n_missed));
1007 put_32aligned_be64(&ots->matched_count,
1008 htonll(s.n_hit + ofproto->n_matches));
1011 static struct ofport *
1014 struct ofport_dpif *port = xmalloc(sizeof *port);
1019 port_dealloc(struct ofport *port_)
1021 struct ofport_dpif *port = ofport_dpif_cast(port_);
1026 port_construct(struct ofport *port_)
1028 struct ofport_dpif *port = ofport_dpif_cast(port_);
1029 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1031 ofproto->need_revalidate = true;
1032 port->odp_port = ofp_port_to_odp_port(port->up.ofp_port);
1033 port->bundle = NULL;
1035 port->tag = tag_create_random();
1036 port->may_enable = true;
1037 port->stp_port = NULL;
1038 port->stp_state = STP_DISABLED;
1039 hmap_init(&port->priorities);
1040 port->realdev_ofp_port = 0;
1041 port->vlandev_vid = 0;
1042 port->carrier_seq = netdev_get_carrier_resets(port->up.netdev);
1044 if (ofproto->sflow) {
1045 dpif_sflow_add_port(ofproto->sflow, port_);
1052 port_destruct(struct ofport *port_)
1054 struct ofport_dpif *port = ofport_dpif_cast(port_);
1055 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1057 ofproto->need_revalidate = true;
1058 bundle_remove(port_);
1059 set_cfm(port_, NULL);
1060 if (ofproto->sflow) {
1061 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1064 ofport_clear_priorities(port);
1065 hmap_destroy(&port->priorities);
1069 port_modified(struct ofport *port_)
1071 struct ofport_dpif *port = ofport_dpif_cast(port_);
1073 if (port->bundle && port->bundle->bond) {
1074 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1079 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1081 struct ofport_dpif *port = ofport_dpif_cast(port_);
1082 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1083 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1085 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1086 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD)) {
1087 ofproto->need_revalidate = true;
1089 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1090 bundle_update(port->bundle);
1096 set_sflow(struct ofproto *ofproto_,
1097 const struct ofproto_sflow_options *sflow_options)
1099 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1100 struct dpif_sflow *ds = ofproto->sflow;
1102 if (sflow_options) {
1104 struct ofport_dpif *ofport;
1106 ds = ofproto->sflow = dpif_sflow_create(ofproto->dpif);
1107 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1108 dpif_sflow_add_port(ds, &ofport->up);
1110 ofproto->need_revalidate = true;
1112 dpif_sflow_set_options(ds, sflow_options);
1115 dpif_sflow_destroy(ds);
1116 ofproto->need_revalidate = true;
1117 ofproto->sflow = NULL;
1124 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1126 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1133 struct ofproto_dpif *ofproto;
1135 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1136 ofproto->need_revalidate = true;
1137 ofport->cfm = cfm_create(netdev_get_name(ofport->up.netdev));
1140 if (cfm_configure(ofport->cfm, s)) {
1146 cfm_destroy(ofport->cfm);
1152 get_cfm_fault(const struct ofport *ofport_)
1154 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1156 return ofport->cfm ? cfm_get_fault(ofport->cfm) : -1;
1160 get_cfm_remote_mpids(const struct ofport *ofport_, const uint64_t **rmps,
1163 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1166 cfm_get_remote_mpids(ofport->cfm, rmps, n_rmps);
1174 get_cfm_health(const struct ofport *ofport_)
1176 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1178 return ofport->cfm ? cfm_get_health(ofport->cfm) : -1;
1181 /* Spanning Tree. */
1184 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
1186 struct ofproto_dpif *ofproto = ofproto_;
1187 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
1188 struct ofport_dpif *ofport;
1190 ofport = stp_port_get_aux(sp);
1192 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
1193 ofproto->up.name, port_num);
1195 struct eth_header *eth = pkt->l2;
1197 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
1198 if (eth_addr_is_zero(eth->eth_src)) {
1199 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
1200 "with unknown MAC", ofproto->up.name, port_num);
1202 send_packet(ofport, pkt);
1208 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
1210 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
1212 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1214 /* Only revalidate flows if the configuration changed. */
1215 if (!s != !ofproto->stp) {
1216 ofproto->need_revalidate = true;
1220 if (!ofproto->stp) {
1221 ofproto->stp = stp_create(ofproto_->name, s->system_id,
1222 send_bpdu_cb, ofproto);
1223 ofproto->stp_last_tick = time_msec();
1226 stp_set_bridge_id(ofproto->stp, s->system_id);
1227 stp_set_bridge_priority(ofproto->stp, s->priority);
1228 stp_set_hello_time(ofproto->stp, s->hello_time);
1229 stp_set_max_age(ofproto->stp, s->max_age);
1230 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
1232 struct ofport *ofport;
1234 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
1235 set_stp_port(ofport, NULL);
1238 stp_destroy(ofproto->stp);
1239 ofproto->stp = NULL;
1246 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
1248 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1252 s->bridge_id = stp_get_bridge_id(ofproto->stp);
1253 s->designated_root = stp_get_designated_root(ofproto->stp);
1254 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
1263 update_stp_port_state(struct ofport_dpif *ofport)
1265 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1266 enum stp_state state;
1268 /* Figure out new state. */
1269 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
1273 if (ofport->stp_state != state) {
1274 enum ofputil_port_state of_state;
1277 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
1278 netdev_get_name(ofport->up.netdev),
1279 stp_state_name(ofport->stp_state),
1280 stp_state_name(state));
1281 if (stp_learn_in_state(ofport->stp_state)
1282 != stp_learn_in_state(state)) {
1283 /* xxx Learning action flows should also be flushed. */
1284 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
1286 fwd_change = stp_forward_in_state(ofport->stp_state)
1287 != stp_forward_in_state(state);
1289 ofproto->need_revalidate = true;
1290 ofport->stp_state = state;
1291 ofport->stp_state_entered = time_msec();
1293 if (fwd_change && ofport->bundle) {
1294 bundle_update(ofport->bundle);
1297 /* Update the STP state bits in the OpenFlow port description. */
1298 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
1299 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
1300 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
1301 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
1302 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
1304 ofproto_port_set_state(&ofport->up, of_state);
1308 /* Configures STP on 'ofport_' using the settings defined in 's'. The
1309 * caller is responsible for assigning STP port numbers and ensuring
1310 * there are no duplicates. */
1312 set_stp_port(struct ofport *ofport_,
1313 const struct ofproto_port_stp_settings *s)
1315 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1316 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1317 struct stp_port *sp = ofport->stp_port;
1319 if (!s || !s->enable) {
1321 ofport->stp_port = NULL;
1322 stp_port_disable(sp);
1323 update_stp_port_state(ofport);
1326 } else if (sp && stp_port_no(sp) != s->port_num
1327 && ofport == stp_port_get_aux(sp)) {
1328 /* The port-id changed, so disable the old one if it's not
1329 * already in use by another port. */
1330 stp_port_disable(sp);
1333 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
1334 stp_port_enable(sp);
1336 stp_port_set_aux(sp, ofport);
1337 stp_port_set_priority(sp, s->priority);
1338 stp_port_set_path_cost(sp, s->path_cost);
1340 update_stp_port_state(ofport);
1346 get_stp_port_status(struct ofport *ofport_,
1347 struct ofproto_port_stp_status *s)
1349 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1350 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1351 struct stp_port *sp = ofport->stp_port;
1353 if (!ofproto->stp || !sp) {
1359 s->port_id = stp_port_get_id(sp);
1360 s->state = stp_port_get_state(sp);
1361 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
1362 s->role = stp_port_get_role(sp);
1363 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
1369 stp_run(struct ofproto_dpif *ofproto)
1372 long long int now = time_msec();
1373 long long int elapsed = now - ofproto->stp_last_tick;
1374 struct stp_port *sp;
1377 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
1378 ofproto->stp_last_tick = now;
1380 while (stp_get_changed_port(ofproto->stp, &sp)) {
1381 struct ofport_dpif *ofport = stp_port_get_aux(sp);
1384 update_stp_port_state(ofport);
1388 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
1389 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
1395 stp_wait(struct ofproto_dpif *ofproto)
1398 poll_timer_wait(1000);
1402 /* Returns true if STP should process 'flow'. */
1404 stp_should_process_flow(const struct flow *flow)
1406 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
1410 stp_process_packet(const struct ofport_dpif *ofport,
1411 const struct ofpbuf *packet)
1413 struct ofpbuf payload = *packet;
1414 struct eth_header *eth = payload.data;
1415 struct stp_port *sp = ofport->stp_port;
1417 /* Sink packets on ports that have STP disabled when the bridge has
1419 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
1423 /* Trim off padding on payload. */
1424 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
1425 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
1428 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
1429 stp_received_bpdu(sp, payload.data, payload.size);
1433 static struct priority_to_dscp *
1434 get_priority(const struct ofport_dpif *ofport, uint32_t priority)
1436 struct priority_to_dscp *pdscp;
1439 hash = hash_int(priority, 0);
1440 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) {
1441 if (pdscp->priority == priority) {
1449 ofport_clear_priorities(struct ofport_dpif *ofport)
1451 struct priority_to_dscp *pdscp, *next;
1453 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) {
1454 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
1460 set_queues(struct ofport *ofport_,
1461 const struct ofproto_port_queue *qdscp_list,
1464 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1465 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1466 struct hmap new = HMAP_INITIALIZER(&new);
1469 for (i = 0; i < n_qdscp; i++) {
1470 struct priority_to_dscp *pdscp;
1474 dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
1475 if (dpif_queue_to_priority(ofproto->dpif, qdscp_list[i].queue,
1480 pdscp = get_priority(ofport, priority);
1482 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
1484 pdscp = xmalloc(sizeof *pdscp);
1485 pdscp->priority = priority;
1487 ofproto->need_revalidate = true;
1490 if (pdscp->dscp != dscp) {
1492 ofproto->need_revalidate = true;
1495 hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0));
1498 if (!hmap_is_empty(&ofport->priorities)) {
1499 ofport_clear_priorities(ofport);
1500 ofproto->need_revalidate = true;
1503 hmap_swap(&new, &ofport->priorities);
1511 /* Expires all MAC learning entries associated with 'bundle' and forces its
1512 * ofproto to revalidate every flow.
1514 * Normally MAC learning entries are removed only from the ofproto associated
1515 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
1516 * are removed from every ofproto. When patch ports and SLB bonds are in use
1517 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
1518 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
1519 * with the host from which it migrated. */
1521 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
1523 struct ofproto_dpif *ofproto = bundle->ofproto;
1524 struct mac_learning *ml = ofproto->ml;
1525 struct mac_entry *mac, *next_mac;
1527 ofproto->need_revalidate = true;
1528 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
1529 if (mac->port.p == bundle) {
1531 struct ofproto_dpif *o;
1533 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1535 struct mac_entry *e;
1537 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan,
1540 tag_set_add(&o->revalidate_set, e->tag);
1541 mac_learning_expire(o->ml, e);
1547 mac_learning_expire(ml, mac);
1552 static struct ofbundle *
1553 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
1555 struct ofbundle *bundle;
1557 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
1558 &ofproto->bundles) {
1559 if (bundle->aux == aux) {
1566 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
1567 * ones that are found to 'bundles'. */
1569 bundle_lookup_multiple(struct ofproto_dpif *ofproto,
1570 void **auxes, size_t n_auxes,
1571 struct hmapx *bundles)
1575 hmapx_init(bundles);
1576 for (i = 0; i < n_auxes; i++) {
1577 struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
1579 hmapx_add(bundles, bundle);
1585 bundle_update(struct ofbundle *bundle)
1587 struct ofport_dpif *port;
1589 bundle->floodable = true;
1590 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1591 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
1592 || !stp_forward_in_state(port->stp_state)) {
1593 bundle->floodable = false;
1600 bundle_del_port(struct ofport_dpif *port)
1602 struct ofbundle *bundle = port->bundle;
1604 bundle->ofproto->need_revalidate = true;
1606 list_remove(&port->bundle_node);
1607 port->bundle = NULL;
1610 lacp_slave_unregister(bundle->lacp, port);
1613 bond_slave_unregister(bundle->bond, port);
1616 bundle_update(bundle);
1620 bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port,
1621 struct lacp_slave_settings *lacp,
1622 uint32_t bond_stable_id)
1624 struct ofport_dpif *port;
1626 port = get_ofp_port(bundle->ofproto, ofp_port);
1631 if (port->bundle != bundle) {
1632 bundle->ofproto->need_revalidate = true;
1634 bundle_del_port(port);
1637 port->bundle = bundle;
1638 list_push_back(&bundle->ports, &port->bundle_node);
1639 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
1640 || !stp_forward_in_state(port->stp_state)) {
1641 bundle->floodable = false;
1645 port->bundle->ofproto->need_revalidate = true;
1646 lacp_slave_register(bundle->lacp, port, lacp);
1649 port->bond_stable_id = bond_stable_id;
1655 bundle_destroy(struct ofbundle *bundle)
1657 struct ofproto_dpif *ofproto;
1658 struct ofport_dpif *port, *next_port;
1665 ofproto = bundle->ofproto;
1666 for (i = 0; i < MAX_MIRRORS; i++) {
1667 struct ofmirror *m = ofproto->mirrors[i];
1669 if (m->out == bundle) {
1671 } else if (hmapx_find_and_delete(&m->srcs, bundle)
1672 || hmapx_find_and_delete(&m->dsts, bundle)) {
1673 ofproto->need_revalidate = true;
1678 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
1679 bundle_del_port(port);
1682 bundle_flush_macs(bundle, true);
1683 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
1685 free(bundle->trunks);
1686 lacp_destroy(bundle->lacp);
1687 bond_destroy(bundle->bond);
1692 bundle_set(struct ofproto *ofproto_, void *aux,
1693 const struct ofproto_bundle_settings *s)
1695 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1696 bool need_flush = false;
1697 struct ofport_dpif *port;
1698 struct ofbundle *bundle;
1699 unsigned long *trunks;
1705 bundle_destroy(bundle_lookup(ofproto, aux));
1709 assert(s->n_slaves == 1 || s->bond != NULL);
1710 assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
1712 bundle = bundle_lookup(ofproto, aux);
1714 bundle = xmalloc(sizeof *bundle);
1716 bundle->ofproto = ofproto;
1717 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
1718 hash_pointer(aux, 0));
1720 bundle->name = NULL;
1722 list_init(&bundle->ports);
1723 bundle->vlan_mode = PORT_VLAN_TRUNK;
1725 bundle->trunks = NULL;
1726 bundle->use_priority_tags = s->use_priority_tags;
1727 bundle->lacp = NULL;
1728 bundle->bond = NULL;
1730 bundle->floodable = true;
1732 bundle->src_mirrors = 0;
1733 bundle->dst_mirrors = 0;
1734 bundle->mirror_out = 0;
1737 if (!bundle->name || strcmp(s->name, bundle->name)) {
1739 bundle->name = xstrdup(s->name);
1744 if (!bundle->lacp) {
1745 ofproto->need_revalidate = true;
1746 bundle->lacp = lacp_create();
1748 lacp_configure(bundle->lacp, s->lacp);
1750 lacp_destroy(bundle->lacp);
1751 bundle->lacp = NULL;
1754 /* Update set of ports. */
1756 for (i = 0; i < s->n_slaves; i++) {
1757 if (!bundle_add_port(bundle, s->slaves[i],
1758 s->lacp ? &s->lacp_slaves[i] : NULL,
1759 s->bond_stable_ids ? s->bond_stable_ids[i] : 0)) {
1763 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
1764 struct ofport_dpif *next_port;
1766 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
1767 for (i = 0; i < s->n_slaves; i++) {
1768 if (s->slaves[i] == port->up.ofp_port) {
1773 bundle_del_port(port);
1777 assert(list_size(&bundle->ports) <= s->n_slaves);
1779 if (list_is_empty(&bundle->ports)) {
1780 bundle_destroy(bundle);
1784 /* Set VLAN tagging mode */
1785 if (s->vlan_mode != bundle->vlan_mode
1786 || s->use_priority_tags != bundle->use_priority_tags) {
1787 bundle->vlan_mode = s->vlan_mode;
1788 bundle->use_priority_tags = s->use_priority_tags;
1793 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
1794 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
1796 if (vlan != bundle->vlan) {
1797 bundle->vlan = vlan;
1801 /* Get trunked VLANs. */
1802 switch (s->vlan_mode) {
1803 case PORT_VLAN_ACCESS:
1807 case PORT_VLAN_TRUNK:
1808 trunks = (unsigned long *) s->trunks;
1811 case PORT_VLAN_NATIVE_UNTAGGED:
1812 case PORT_VLAN_NATIVE_TAGGED:
1813 if (vlan != 0 && (!s->trunks
1814 || !bitmap_is_set(s->trunks, vlan)
1815 || bitmap_is_set(s->trunks, 0))) {
1816 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
1818 trunks = bitmap_clone(s->trunks, 4096);
1820 trunks = bitmap_allocate1(4096);
1822 bitmap_set1(trunks, vlan);
1823 bitmap_set0(trunks, 0);
1825 trunks = (unsigned long *) s->trunks;
1832 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
1833 free(bundle->trunks);
1834 if (trunks == s->trunks) {
1835 bundle->trunks = vlan_bitmap_clone(trunks);
1837 bundle->trunks = trunks;
1842 if (trunks != s->trunks) {
1847 if (!list_is_short(&bundle->ports)) {
1848 bundle->ofproto->has_bonded_bundles = true;
1850 if (bond_reconfigure(bundle->bond, s->bond)) {
1851 ofproto->need_revalidate = true;
1854 bundle->bond = bond_create(s->bond);
1855 ofproto->need_revalidate = true;
1858 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1859 bond_slave_register(bundle->bond, port, port->bond_stable_id,
1863 bond_destroy(bundle->bond);
1864 bundle->bond = NULL;
1867 /* If we changed something that would affect MAC learning, un-learn
1868 * everything on this port and force flow revalidation. */
1870 bundle_flush_macs(bundle, false);
1877 bundle_remove(struct ofport *port_)
1879 struct ofport_dpif *port = ofport_dpif_cast(port_);
1880 struct ofbundle *bundle = port->bundle;
1883 bundle_del_port(port);
1884 if (list_is_empty(&bundle->ports)) {
1885 bundle_destroy(bundle);
1886 } else if (list_is_short(&bundle->ports)) {
1887 bond_destroy(bundle->bond);
1888 bundle->bond = NULL;
1894 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
1896 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
1897 struct ofport_dpif *port = port_;
1898 uint8_t ea[ETH_ADDR_LEN];
1901 error = netdev_get_etheraddr(port->up.netdev, ea);
1903 struct ofpbuf packet;
1906 ofpbuf_init(&packet, 0);
1907 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
1909 memcpy(packet_pdu, pdu, pdu_size);
1911 send_packet(port, &packet);
1912 ofpbuf_uninit(&packet);
1914 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
1915 "%s (%s)", port->bundle->name,
1916 netdev_get_name(port->up.netdev), strerror(error));
1921 bundle_send_learning_packets(struct ofbundle *bundle)
1923 struct ofproto_dpif *ofproto = bundle->ofproto;
1924 int error, n_packets, n_errors;
1925 struct mac_entry *e;
1927 error = n_packets = n_errors = 0;
1928 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
1929 if (e->port.p != bundle) {
1930 struct ofpbuf *learning_packet;
1931 struct ofport_dpif *port;
1935 /* The assignment to "port" is unnecessary but makes "grep"ing for
1936 * struct ofport_dpif more effective. */
1937 learning_packet = bond_compose_learning_packet(bundle->bond,
1941 ret = send_packet(port, learning_packet);
1942 ofpbuf_delete(learning_packet);
1952 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1953 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
1954 "packets, last error was: %s",
1955 bundle->name, n_errors, n_packets, strerror(error));
1957 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
1958 bundle->name, n_packets);
1963 bundle_run(struct ofbundle *bundle)
1966 lacp_run(bundle->lacp, send_pdu_cb);
1969 struct ofport_dpif *port;
1971 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1972 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
1975 bond_run(bundle->bond, &bundle->ofproto->revalidate_set,
1976 lacp_status(bundle->lacp));
1977 if (bond_should_send_learning_packets(bundle->bond)) {
1978 bundle_send_learning_packets(bundle);
1984 bundle_wait(struct ofbundle *bundle)
1987 lacp_wait(bundle->lacp);
1990 bond_wait(bundle->bond);
1997 mirror_scan(struct ofproto_dpif *ofproto)
2001 for (idx = 0; idx < MAX_MIRRORS; idx++) {
2002 if (!ofproto->mirrors[idx]) {
2009 static struct ofmirror *
2010 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
2014 for (i = 0; i < MAX_MIRRORS; i++) {
2015 struct ofmirror *mirror = ofproto->mirrors[i];
2016 if (mirror && mirror->aux == aux) {
2024 /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */
2026 mirror_update_dups(struct ofproto_dpif *ofproto)
2030 for (i = 0; i < MAX_MIRRORS; i++) {
2031 struct ofmirror *m = ofproto->mirrors[i];
2034 m->dup_mirrors = MIRROR_MASK_C(1) << i;
2038 for (i = 0; i < MAX_MIRRORS; i++) {
2039 struct ofmirror *m1 = ofproto->mirrors[i];
2046 for (j = i + 1; j < MAX_MIRRORS; j++) {
2047 struct ofmirror *m2 = ofproto->mirrors[j];
2049 if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) {
2050 m1->dup_mirrors |= MIRROR_MASK_C(1) << j;
2051 m2->dup_mirrors |= m1->dup_mirrors;
2058 mirror_set(struct ofproto *ofproto_, void *aux,
2059 const struct ofproto_mirror_settings *s)
2061 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2062 mirror_mask_t mirror_bit;
2063 struct ofbundle *bundle;
2064 struct ofmirror *mirror;
2065 struct ofbundle *out;
2066 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
2067 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
2070 mirror = mirror_lookup(ofproto, aux);
2072 mirror_destroy(mirror);
2078 idx = mirror_scan(ofproto);
2080 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
2082 ofproto->up.name, MAX_MIRRORS, s->name);
2086 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
2087 mirror->ofproto = ofproto;
2090 mirror->out_vlan = -1;
2091 mirror->name = NULL;
2094 if (!mirror->name || strcmp(s->name, mirror->name)) {
2096 mirror->name = xstrdup(s->name);
2099 /* Get the new configuration. */
2100 if (s->out_bundle) {
2101 out = bundle_lookup(ofproto, s->out_bundle);
2103 mirror_destroy(mirror);
2109 out_vlan = s->out_vlan;
2111 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
2112 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
2114 /* If the configuration has not changed, do nothing. */
2115 if (hmapx_equals(&srcs, &mirror->srcs)
2116 && hmapx_equals(&dsts, &mirror->dsts)
2117 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
2118 && mirror->out == out
2119 && mirror->out_vlan == out_vlan)
2121 hmapx_destroy(&srcs);
2122 hmapx_destroy(&dsts);
2126 hmapx_swap(&srcs, &mirror->srcs);
2127 hmapx_destroy(&srcs);
2129 hmapx_swap(&dsts, &mirror->dsts);
2130 hmapx_destroy(&dsts);
2132 free(mirror->vlans);
2133 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
2136 mirror->out_vlan = out_vlan;
2138 /* Update bundles. */
2139 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2140 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
2141 if (hmapx_contains(&mirror->srcs, bundle)) {
2142 bundle->src_mirrors |= mirror_bit;
2144 bundle->src_mirrors &= ~mirror_bit;
2147 if (hmapx_contains(&mirror->dsts, bundle)) {
2148 bundle->dst_mirrors |= mirror_bit;
2150 bundle->dst_mirrors &= ~mirror_bit;
2153 if (mirror->out == bundle) {
2154 bundle->mirror_out |= mirror_bit;
2156 bundle->mirror_out &= ~mirror_bit;
2160 ofproto->need_revalidate = true;
2161 ofproto->has_mirrors = true;
2162 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
2163 mirror_update_dups(ofproto);
2169 mirror_destroy(struct ofmirror *mirror)
2171 struct ofproto_dpif *ofproto;
2172 mirror_mask_t mirror_bit;
2173 struct ofbundle *bundle;
2180 ofproto = mirror->ofproto;
2181 ofproto->need_revalidate = true;
2182 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
2184 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2185 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
2186 bundle->src_mirrors &= ~mirror_bit;
2187 bundle->dst_mirrors &= ~mirror_bit;
2188 bundle->mirror_out &= ~mirror_bit;
2191 hmapx_destroy(&mirror->srcs);
2192 hmapx_destroy(&mirror->dsts);
2193 free(mirror->vlans);
2195 ofproto->mirrors[mirror->idx] = NULL;
2199 mirror_update_dups(ofproto);
2201 ofproto->has_mirrors = false;
2202 for (i = 0; i < MAX_MIRRORS; i++) {
2203 if (ofproto->mirrors[i]) {
2204 ofproto->has_mirrors = true;
2211 mirror_get_stats(struct ofproto *ofproto_, void *aux,
2212 uint64_t *packets, uint64_t *bytes)
2214 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2215 struct ofmirror *mirror = mirror_lookup(ofproto, aux);
2218 *packets = *bytes = UINT64_MAX;
2222 *packets = mirror->packet_count;
2223 *bytes = mirror->byte_count;
2229 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2231 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2232 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2233 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
2239 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2241 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2242 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2243 return bundle && bundle->mirror_out != 0;
2247 forward_bpdu_changed(struct ofproto *ofproto_)
2249 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2250 /* Revalidate cached flows whenever forward_bpdu option changes. */
2251 ofproto->need_revalidate = true;
2255 set_mac_idle_time(struct ofproto *ofproto_, unsigned int idle_time)
2257 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2258 mac_learning_set_idle_time(ofproto->ml, idle_time);
2263 static struct ofport_dpif *
2264 get_ofp_port(struct ofproto_dpif *ofproto, uint16_t ofp_port)
2266 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2267 return ofport ? ofport_dpif_cast(ofport) : NULL;
2270 static struct ofport_dpif *
2271 get_odp_port(struct ofproto_dpif *ofproto, uint32_t odp_port)
2273 return get_ofp_port(ofproto, odp_port_to_ofp_port(odp_port));
2277 ofproto_port_from_dpif_port(struct ofproto_port *ofproto_port,
2278 struct dpif_port *dpif_port)
2280 ofproto_port->name = dpif_port->name;
2281 ofproto_port->type = dpif_port->type;
2282 ofproto_port->ofp_port = odp_port_to_ofp_port(dpif_port->port_no);
2286 port_run(struct ofport_dpif *ofport)
2288 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
2289 bool carrier_changed = carrier_seq != ofport->carrier_seq;
2290 bool enable = netdev_get_carrier(ofport->up.netdev);
2292 ofport->carrier_seq = carrier_seq;
2295 cfm_run(ofport->cfm);
2297 if (cfm_should_send_ccm(ofport->cfm)) {
2298 struct ofpbuf packet;
2300 ofpbuf_init(&packet, 0);
2301 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
2302 send_packet(ofport, &packet);
2303 ofpbuf_uninit(&packet);
2306 enable = enable && !cfm_get_fault(ofport->cfm)
2307 && cfm_get_opup(ofport->cfm);
2310 if (ofport->bundle) {
2311 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
2312 if (carrier_changed) {
2313 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
2317 if (ofport->may_enable != enable) {
2318 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2320 if (ofproto->has_bundle_action) {
2321 ofproto->need_revalidate = true;
2325 ofport->may_enable = enable;
2329 port_wait(struct ofport_dpif *ofport)
2332 cfm_wait(ofport->cfm);
2337 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
2338 struct ofproto_port *ofproto_port)
2340 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2341 struct dpif_port dpif_port;
2344 error = dpif_port_query_by_name(ofproto->dpif, devname, &dpif_port);
2346 ofproto_port_from_dpif_port(ofproto_port, &dpif_port);
2352 port_add(struct ofproto *ofproto_, struct netdev *netdev, uint16_t *ofp_portp)
2354 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2358 error = dpif_port_add(ofproto->dpif, netdev, &odp_port);
2360 *ofp_portp = odp_port_to_ofp_port(odp_port);
2366 port_del(struct ofproto *ofproto_, uint16_t ofp_port)
2368 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2371 error = dpif_port_del(ofproto->dpif, ofp_port_to_odp_port(ofp_port));
2373 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
2375 /* The caller is going to close ofport->up.netdev. If this is a
2376 * bonded port, then the bond is using that netdev, so remove it
2377 * from the bond. The client will need to reconfigure everything
2378 * after deleting ports, so then the slave will get re-added. */
2379 bundle_remove(&ofport->up);
2386 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
2388 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2391 error = netdev_get_stats(ofport->up.netdev, stats);
2393 if (!error && ofport->odp_port == OVSP_LOCAL) {
2394 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2396 /* ofproto->stats.tx_packets represents packets that we created
2397 * internally and sent to some port (e.g. packets sent with
2398 * send_packet()). Account for them as if they had come from
2399 * OFPP_LOCAL and got forwarded. */
2401 if (stats->rx_packets != UINT64_MAX) {
2402 stats->rx_packets += ofproto->stats.tx_packets;
2405 if (stats->rx_bytes != UINT64_MAX) {
2406 stats->rx_bytes += ofproto->stats.tx_bytes;
2409 /* ofproto->stats.rx_packets represents packets that were received on
2410 * some port and we processed internally and dropped (e.g. STP).
2411 * Account fro them as if they had been forwarded to OFPP_LOCAL. */
2413 if (stats->tx_packets != UINT64_MAX) {
2414 stats->tx_packets += ofproto->stats.rx_packets;
2417 if (stats->tx_bytes != UINT64_MAX) {
2418 stats->tx_bytes += ofproto->stats.rx_bytes;
2425 /* Account packets for LOCAL port. */
2427 ofproto_update_local_port_stats(const struct ofproto *ofproto_,
2428 size_t tx_size, size_t rx_size)
2430 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2433 ofproto->stats.rx_packets++;
2434 ofproto->stats.rx_bytes += rx_size;
2437 ofproto->stats.tx_packets++;
2438 ofproto->stats.tx_bytes += tx_size;
2442 struct port_dump_state {
2443 struct dpif_port_dump dump;
2448 port_dump_start(const struct ofproto *ofproto_, void **statep)
2450 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2451 struct port_dump_state *state;
2453 *statep = state = xmalloc(sizeof *state);
2454 dpif_port_dump_start(&state->dump, ofproto->dpif);
2455 state->done = false;
2460 port_dump_next(const struct ofproto *ofproto_ OVS_UNUSED, void *state_,
2461 struct ofproto_port *port)
2463 struct port_dump_state *state = state_;
2464 struct dpif_port dpif_port;
2466 if (dpif_port_dump_next(&state->dump, &dpif_port)) {
2467 ofproto_port_from_dpif_port(port, &dpif_port);
2470 int error = dpif_port_dump_done(&state->dump);
2472 return error ? error : EOF;
2477 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
2479 struct port_dump_state *state = state_;
2482 dpif_port_dump_done(&state->dump);
2489 port_poll(const struct ofproto *ofproto_, char **devnamep)
2491 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2492 return dpif_port_poll(ofproto->dpif, devnamep);
2496 port_poll_wait(const struct ofproto *ofproto_)
2498 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2499 dpif_port_poll_wait(ofproto->dpif);
2503 port_is_lacp_current(const struct ofport *ofport_)
2505 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2506 return (ofport->bundle && ofport->bundle->lacp
2507 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
2511 /* Upcall handling. */
2513 /* Flow miss batching.
2515 * Some dpifs implement operations faster when you hand them off in a batch.
2516 * To allow batching, "struct flow_miss" queues the dpif-related work needed
2517 * for a given flow. Each "struct flow_miss" corresponds to sending one or
2518 * more packets, plus possibly installing the flow in the dpif.
2520 * So far we only batch the operations that affect flow setup time the most.
2521 * It's possible to batch more than that, but the benefit might be minimal. */
2523 struct hmap_node hmap_node;
2525 enum odp_key_fitness key_fitness;
2526 const struct nlattr *key;
2528 ovs_be16 initial_tci;
2529 struct list packets;
2532 struct flow_miss_op {
2533 struct dpif_op dpif_op;
2534 struct subfacet *subfacet; /* Subfacet */
2535 void *garbage; /* Pointer to pass to free(), NULL if none. */
2536 uint64_t stub[1024 / 8]; /* Temporary buffer. */
2539 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
2540 * OpenFlow controller as necessary according to their individual
2541 * configurations. */
2543 send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet,
2544 const struct flow *flow)
2546 struct ofputil_packet_in pin;
2548 pin.packet = packet->data;
2549 pin.packet_len = packet->size;
2550 pin.reason = OFPR_NO_MATCH;
2551 pin.controller_id = 0;
2556 pin.send_len = 0; /* not used for flow table misses */
2558 flow_get_metadata(flow, &pin.fmd);
2560 /* Registers aren't meaningful on a miss. */
2561 memset(pin.fmd.reg_masks, 0, sizeof pin.fmd.reg_masks);
2563 connmgr_send_packet_in(ofproto->up.connmgr, &pin);
2567 process_special(struct ofproto_dpif *ofproto, const struct flow *flow,
2568 const struct ofpbuf *packet)
2570 struct ofport_dpif *ofport = get_ofp_port(ofproto, flow->in_port);
2576 if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) {
2578 cfm_process_heartbeat(ofport->cfm, packet);
2581 } else if (ofport->bundle && ofport->bundle->lacp
2582 && flow->dl_type == htons(ETH_TYPE_LACP)) {
2584 lacp_process_packet(ofport->bundle->lacp, ofport, packet);
2587 } else if (ofproto->stp && stp_should_process_flow(flow)) {
2589 stp_process_packet(ofport, packet);
2596 static struct flow_miss *
2597 flow_miss_find(struct hmap *todo, const struct flow *flow, uint32_t hash)
2599 struct flow_miss *miss;
2601 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
2602 if (flow_equal(&miss->flow, flow)) {
2610 /* Partially Initializes 'op' as an "execute" operation for 'miss' and
2611 * 'packet'. The caller must initialize op->actions and op->actions_len. If
2612 * 'miss' is associated with a subfacet the caller must also initialize the
2613 * returned op->subfacet, and if anything needs to be freed after processing
2614 * the op, the caller must initialize op->garbage also. */
2616 init_flow_miss_execute_op(struct flow_miss *miss, struct ofpbuf *packet,
2617 struct flow_miss_op *op)
2619 if (miss->flow.vlan_tci != miss->initial_tci) {
2620 /* This packet was received on a VLAN splinter port. We
2621 * added a VLAN to the packet to make the packet resemble
2622 * the flow, but the actions were composed assuming that
2623 * the packet contained no VLAN. So, we must remove the
2624 * VLAN header from the packet before trying to execute the
2626 eth_pop_vlan(packet);
2629 op->subfacet = NULL;
2631 op->dpif_op.type = DPIF_OP_EXECUTE;
2632 op->dpif_op.u.execute.key = miss->key;
2633 op->dpif_op.u.execute.key_len = miss->key_len;
2634 op->dpif_op.u.execute.packet = packet;
2637 /* Helper for handle_flow_miss_without_facet() and
2638 * handle_flow_miss_with_facet(). */
2640 handle_flow_miss_common(struct rule_dpif *rule,
2641 struct ofpbuf *packet, const struct flow *flow)
2643 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
2645 ofproto->n_matches++;
2647 if (rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
2649 * Extra-special case for fail-open mode.
2651 * We are in fail-open mode and the packet matched the fail-open
2652 * rule, but we are connected to a controller too. We should send
2653 * the packet up to the controller in the hope that it will try to
2654 * set up a flow and thereby allow us to exit fail-open.
2656 * See the top-level comment in fail-open.c for more information.
2658 send_packet_in_miss(ofproto, packet, flow);
2662 /* Figures out whether a flow that missed in 'ofproto', whose details are in
2663 * 'miss', is likely to be worth tracking in detail in userspace and (usually)
2664 * installing a datapath flow. The answer is usually "yes" (a return value of
2665 * true). However, for short flows the cost of bookkeeping is much higher than
2666 * the benefits, so when the datapath holds a large number of flows we impose
2667 * some heuristics to decide which flows are likely to be worth tracking. */
2669 flow_miss_should_make_facet(struct ofproto_dpif *ofproto,
2670 struct flow_miss *miss, uint32_t hash)
2672 if (!ofproto->governor) {
2675 n_subfacets = hmap_count(&ofproto->subfacets);
2676 if (n_subfacets * 2 <= ofproto->up.flow_eviction_threshold) {
2680 ofproto->governor = governor_create(ofproto->up.name);
2683 return governor_should_install_flow(ofproto->governor, hash,
2684 list_size(&miss->packets));
2687 /* Handles 'miss', which matches 'rule', without creating a facet or subfacet
2688 * or creating any datapath flow. May add an "execute" operation to 'ops' and
2689 * increment '*n_ops'. */
2691 handle_flow_miss_without_facet(struct flow_miss *miss,
2692 struct rule_dpif *rule,
2693 struct flow_miss_op *ops, size_t *n_ops)
2695 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
2696 struct action_xlate_ctx ctx;
2697 struct ofpbuf *packet;
2699 LIST_FOR_EACH (packet, list_node, &miss->packets) {
2700 struct flow_miss_op *op = &ops[*n_ops];
2701 struct dpif_flow_stats stats;
2702 struct ofpbuf odp_actions;
2704 COVERAGE_INC(facet_suppress);
2706 ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub);
2708 dpif_flow_stats_extract(&miss->flow, packet, &stats);
2709 rule_credit_stats(rule, &stats);
2711 action_xlate_ctx_init(&ctx, ofproto, &miss->flow, miss->initial_tci,
2713 ctx.resubmit_stats = &stats;
2714 xlate_actions(&ctx, rule->up.actions, rule->up.n_actions,
2717 if (odp_actions.size) {
2718 struct dpif_execute *execute = &op->dpif_op.u.execute;
2720 init_flow_miss_execute_op(miss, packet, op);
2721 execute->actions = odp_actions.data;
2722 execute->actions_len = odp_actions.size;
2723 op->garbage = ofpbuf_get_uninit_pointer(&odp_actions);
2727 ofpbuf_uninit(&odp_actions);
2732 /* Handles 'miss', which matches 'facet'. May add any required datapath
2733 * operations to 'ops', incrementing '*n_ops' for each new op. */
2735 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
2736 struct flow_miss_op *ops, size_t *n_ops)
2738 struct subfacet *subfacet;
2739 struct ofpbuf *packet;
2741 subfacet = subfacet_create(facet,
2742 miss->key_fitness, miss->key, miss->key_len,
2745 LIST_FOR_EACH (packet, list_node, &miss->packets) {
2746 struct flow_miss_op *op = &ops[*n_ops];
2747 struct dpif_flow_stats stats;
2748 struct ofpbuf odp_actions;
2750 handle_flow_miss_common(facet->rule, packet, &miss->flow);
2752 ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub);
2753 if (!facet->may_install || !subfacet->actions) {
2754 subfacet_make_actions(subfacet, packet, &odp_actions);
2757 dpif_flow_stats_extract(&facet->flow, packet, &stats);
2758 subfacet_update_stats(subfacet, &stats);
2760 if (subfacet->actions_len) {
2761 struct dpif_execute *execute = &op->dpif_op.u.execute;
2763 init_flow_miss_execute_op(miss, packet, op);
2764 op->subfacet = subfacet;
2765 if (facet->may_install) {
2766 execute->actions = subfacet->actions;
2767 execute->actions_len = subfacet->actions_len;
2768 ofpbuf_uninit(&odp_actions);
2770 execute->actions = odp_actions.data;
2771 execute->actions_len = odp_actions.size;
2772 op->garbage = ofpbuf_get_uninit_pointer(&odp_actions);
2777 ofpbuf_uninit(&odp_actions);
2781 if (facet->may_install && subfacet->key_fitness != ODP_FIT_TOO_LITTLE) {
2782 struct flow_miss_op *op = &ops[(*n_ops)++];
2783 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
2785 op->subfacet = subfacet;
2787 op->dpif_op.type = DPIF_OP_FLOW_PUT;
2788 put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
2789 put->key = miss->key;
2790 put->key_len = miss->key_len;
2791 put->actions = subfacet->actions;
2792 put->actions_len = subfacet->actions_len;
2797 /* Handles flow miss 'miss' on 'ofproto'. The flow does not match any flow in
2798 * the OpenFlow flow table. */
2800 handle_flow_miss_no_rule(struct ofproto_dpif *ofproto, struct flow_miss *miss)
2802 uint16_t in_port = miss->flow.in_port;
2803 struct ofport_dpif *port = get_ofp_port(ofproto, in_port);
2806 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, in_port);
2809 if (port && port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN) {
2810 /* XXX install 'drop' flow entry */
2811 COVERAGE_INC(ofproto_dpif_no_packet_in);
2813 const struct ofpbuf *packet;
2815 LIST_FOR_EACH (packet, list_node, &miss->packets) {
2816 send_packet_in_miss(ofproto, packet, &miss->flow);
2821 /* Handles flow miss 'miss' on 'ofproto'. May add any required datapath
2822 * operations to 'ops', incrementing '*n_ops' for each new op. */
2824 handle_flow_miss(struct ofproto_dpif *ofproto, struct flow_miss *miss,
2825 struct flow_miss_op *ops, size_t *n_ops)
2827 struct facet *facet;
2830 /* The caller must ensure that miss->hmap_node.hash contains
2831 * flow_hash(miss->flow, 0). */
2832 hash = miss->hmap_node.hash;
2834 facet = facet_lookup_valid(ofproto, &miss->flow, hash);
2836 struct rule_dpif *rule = rule_dpif_lookup(ofproto, &miss->flow, 0);
2838 handle_flow_miss_no_rule(ofproto, miss);
2840 } else if (!flow_miss_should_make_facet(ofproto, miss, hash)) {
2841 handle_flow_miss_without_facet(miss, rule, ops, n_ops);
2845 facet = facet_create(rule, &miss->flow, hash);
2847 handle_flow_miss_with_facet(miss, facet, ops, n_ops);
2850 /* Like odp_flow_key_to_flow(), this function converts the 'key_len' bytes of
2851 * OVS_KEY_ATTR_* attributes in 'key' to a flow structure in 'flow' and returns
2852 * an ODP_FIT_* value that indicates how well 'key' fits our expectations for
2853 * what a flow key should contain.
2855 * This function also includes some logic to help make VLAN splinters
2856 * transparent to the rest of the upcall processing logic. In particular, if
2857 * the extracted in_port is a VLAN splinter port, it replaces flow->in_port by
2858 * the "real" port, sets flow->vlan_tci correctly for the VLAN of the VLAN
2859 * splinter port, and pushes a VLAN header onto 'packet' (if it is nonnull).
2861 * Sets '*initial_tci' to the VLAN TCI with which the packet was really
2862 * received, that is, the actual VLAN TCI extracted by odp_flow_key_to_flow().
2863 * (This differs from the value returned in flow->vlan_tci only for packets
2864 * received on VLAN splinters.)
2866 static enum odp_key_fitness
2867 ofproto_dpif_extract_flow_key(const struct ofproto_dpif *ofproto,
2868 const struct nlattr *key, size_t key_len,
2869 struct flow *flow, ovs_be16 *initial_tci,
2870 struct ofpbuf *packet)
2872 enum odp_key_fitness fitness;
2874 fitness = odp_flow_key_to_flow(key, key_len, flow);
2875 if (fitness == ODP_FIT_ERROR) {
2878 *initial_tci = flow->vlan_tci;
2880 if (vsp_adjust_flow(ofproto, flow)) {
2882 /* Make the packet resemble the flow, so that it gets sent to an
2883 * OpenFlow controller properly, so that it looks correct for
2884 * sFlow, and so that flow_extract() will get the correct vlan_tci
2885 * if it is called on 'packet'.
2887 * The allocated space inside 'packet' probably also contains
2888 * 'key', that is, both 'packet' and 'key' are probably part of a
2889 * struct dpif_upcall (see the large comment on that structure
2890 * definition), so pushing data on 'packet' is in general not a
2891 * good idea since it could overwrite 'key' or free it as a side
2892 * effect. However, it's OK in this special case because we know
2893 * that 'packet' is inside a Netlink attribute: pushing 4 bytes
2894 * will just overwrite the 4-byte "struct nlattr", which is fine
2895 * since we don't need that header anymore. */
2896 eth_push_vlan(packet, flow->vlan_tci);
2899 /* Let the caller know that we can't reproduce 'key' from 'flow'. */
2900 if (fitness == ODP_FIT_PERFECT) {
2901 fitness = ODP_FIT_TOO_MUCH;
2909 handle_miss_upcalls(struct ofproto_dpif *ofproto, struct dpif_upcall *upcalls,
2912 struct dpif_upcall *upcall;
2913 struct flow_miss *miss;
2914 struct flow_miss misses[FLOW_MISS_MAX_BATCH];
2915 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
2916 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
2926 /* Construct the to-do list.
2928 * This just amounts to extracting the flow from each packet and sticking
2929 * the packets that have the same flow in the same "flow_miss" structure so
2930 * that we can process them together. */
2933 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
2934 struct flow_miss *miss = &misses[n_misses];
2935 struct flow_miss *existing_miss;
2938 /* Obtain metadata and check userspace/kernel agreement on flow match,
2939 * then set 'flow''s header pointers. */
2940 miss->key_fitness = ofproto_dpif_extract_flow_key(
2941 ofproto, upcall->key, upcall->key_len,
2942 &miss->flow, &miss->initial_tci, upcall->packet);
2943 if (miss->key_fitness == ODP_FIT_ERROR) {
2946 flow_extract(upcall->packet, miss->flow.skb_priority,
2947 miss->flow.tun_id, miss->flow.in_port, &miss->flow);
2949 /* Handle 802.1ag, LACP, and STP specially. */
2950 if (process_special(ofproto, &miss->flow, upcall->packet)) {
2951 ofproto_update_local_port_stats(&ofproto->up,
2952 0, upcall->packet->size);
2953 ofproto->n_matches++;
2957 /* Add other packets to a to-do list. */
2958 hash = flow_hash(&miss->flow, 0);
2959 existing_miss = flow_miss_find(&todo, &miss->flow, hash);
2960 if (!existing_miss) {
2961 hmap_insert(&todo, &miss->hmap_node, hash);
2962 miss->key = upcall->key;
2963 miss->key_len = upcall->key_len;
2964 list_init(&miss->packets);
2968 miss = existing_miss;
2970 list_push_back(&miss->packets, &upcall->packet->list_node);
2973 /* Process each element in the to-do list, constructing the set of
2974 * operations to batch. */
2976 HMAP_FOR_EACH (miss, hmap_node, &todo) {
2977 handle_flow_miss(ofproto, miss, flow_miss_ops, &n_ops);
2979 assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
2981 /* Execute batch. */
2982 for (i = 0; i < n_ops; i++) {
2983 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
2985 dpif_operate(ofproto->dpif, dpif_ops, n_ops);
2987 /* Free memory and update facets. */
2988 for (i = 0; i < n_ops; i++) {
2989 struct flow_miss_op *op = &flow_miss_ops[i];
2991 switch (op->dpif_op.type) {
2992 case DPIF_OP_EXECUTE:
2995 case DPIF_OP_FLOW_PUT:
2996 if (!op->dpif_op.error) {
2997 op->subfacet->installed = true;
3001 case DPIF_OP_FLOW_DEL:
3007 hmap_destroy(&todo);
3011 handle_userspace_upcall(struct ofproto_dpif *ofproto,
3012 struct dpif_upcall *upcall)
3014 struct user_action_cookie cookie;
3015 enum odp_key_fitness fitness;
3016 ovs_be16 initial_tci;
3019 memcpy(&cookie, &upcall->userdata, sizeof(cookie));
3021 fitness = ofproto_dpif_extract_flow_key(ofproto, upcall->key,
3022 upcall->key_len, &flow,
3023 &initial_tci, upcall->packet);
3024 if (fitness == ODP_FIT_ERROR) {
3028 if (cookie.type == USER_ACTION_COOKIE_SFLOW) {
3029 if (ofproto->sflow) {
3030 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
3034 VLOG_WARN_RL(&rl, "invalid user cookie : 0x%"PRIx64, upcall->userdata);
3039 handle_upcalls(struct ofproto_dpif *ofproto, unsigned int max_batch)
3041 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
3042 struct ofpbuf miss_bufs[FLOW_MISS_MAX_BATCH];
3043 uint64_t miss_buf_stubs[FLOW_MISS_MAX_BATCH][4096 / 8];
3048 assert(max_batch <= FLOW_MISS_MAX_BATCH);
3052 for (n_processed = 0; n_processed < max_batch; n_processed++) {
3053 struct dpif_upcall *upcall = &misses[n_misses];
3054 struct ofpbuf *buf = &miss_bufs[n_misses];
3057 ofpbuf_use_stub(buf, miss_buf_stubs[n_misses],
3058 sizeof miss_buf_stubs[n_misses]);
3059 error = dpif_recv(ofproto->dpif, upcall, buf);
3065 switch (upcall->type) {
3066 case DPIF_UC_ACTION:
3067 handle_userspace_upcall(ofproto, upcall);
3072 /* Handle it later. */
3076 case DPIF_N_UC_TYPES:
3078 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32,
3084 handle_miss_upcalls(ofproto, misses, n_misses);
3085 for (i = 0; i < n_misses; i++) {
3086 ofpbuf_uninit(&miss_bufs[i]);
3092 /* Flow expiration. */
3094 static int subfacet_max_idle(const struct ofproto_dpif *);
3095 static void update_stats(struct ofproto_dpif *);
3096 static void rule_expire(struct rule_dpif *);
3097 static void expire_subfacets(struct ofproto_dpif *, int dp_max_idle);
3099 /* This function is called periodically by run(). Its job is to collect
3100 * updates for the flows that have been installed into the datapath, most
3101 * importantly when they last were used, and then use that information to
3102 * expire flows that have not been used recently.
3104 * Returns the number of milliseconds after which it should be called again. */
3106 expire(struct ofproto_dpif *ofproto)
3108 struct rule_dpif *rule, *next_rule;
3109 struct oftable *table;
3112 /* Update stats for each flow in the datapath. */
3113 update_stats(ofproto);
3115 /* Expire subfacets that have been idle too long. */
3116 dp_max_idle = subfacet_max_idle(ofproto);
3117 expire_subfacets(ofproto, dp_max_idle);
3119 /* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */
3120 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
3121 struct cls_cursor cursor;
3123 cls_cursor_init(&cursor, &table->cls, NULL);
3124 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
3129 /* All outstanding data in existing flows has been accounted, so it's a
3130 * good time to do bond rebalancing. */
3131 if (ofproto->has_bonded_bundles) {
3132 struct ofbundle *bundle;
3134 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
3136 bond_rebalance(bundle->bond, &ofproto->revalidate_set);
3141 return MIN(dp_max_idle, 1000);
3144 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
3146 * This function also pushes statistics updates to rules which each facet
3147 * resubmits into. Generally these statistics will be accurate. However, if a
3148 * facet changes the rule it resubmits into at some time in between
3149 * update_stats() runs, it is possible that statistics accrued to the
3150 * old rule will be incorrectly attributed to the new rule. This could be
3151 * avoided by calling update_stats() whenever rules are created or
3152 * deleted. However, the performance impact of making so many calls to the
3153 * datapath do not justify the benefit of having perfectly accurate statistics.
3156 update_stats(struct ofproto_dpif *p)
3158 const struct dpif_flow_stats *stats;
3159 struct dpif_flow_dump dump;
3160 const struct nlattr *key;
3163 dpif_flow_dump_start(&dump, p->dpif);
3164 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
3165 struct subfacet *subfacet;
3167 subfacet = subfacet_find(p, key, key_len);
3168 if (subfacet && subfacet->installed) {
3169 struct facet *facet = subfacet->facet;
3171 if (stats->n_packets >= subfacet->dp_packet_count) {
3172 uint64_t extra = stats->n_packets - subfacet->dp_packet_count;
3173 facet->packet_count += extra;
3175 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
3178 if (stats->n_bytes >= subfacet->dp_byte_count) {
3179 facet->byte_count += stats->n_bytes - subfacet->dp_byte_count;
3181 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
3184 subfacet->dp_packet_count = stats->n_packets;
3185 subfacet->dp_byte_count = stats->n_bytes;
3187 facet->tcp_flags |= stats->tcp_flags;
3189 subfacet_update_time(subfacet, stats->used);
3190 if (facet->accounted_bytes < facet->byte_count) {
3192 facet_account(facet);
3193 facet->accounted_bytes = facet->byte_count;
3195 facet_push_stats(facet);
3197 if (!VLOG_DROP_WARN(&rl)) {
3201 odp_flow_key_format(key, key_len, &s);
3202 VLOG_WARN("unexpected flow from datapath %s", ds_cstr(&s));
3206 COVERAGE_INC(facet_unexpected);
3207 /* There's a flow in the datapath that we know nothing about, or a
3208 * flow that shouldn't be installed but was anyway. Delete it. */
3209 dpif_flow_del(p->dpif, key, key_len, NULL);
3212 dpif_flow_dump_done(&dump);
3215 /* Calculates and returns the number of milliseconds of idle time after which
3216 * subfacets should expire from the datapath. When a subfacet expires, we fold
3217 * its statistics into its facet, and when a facet's last subfacet expires, we
3218 * fold its statistic into its rule. */
3220 subfacet_max_idle(const struct ofproto_dpif *ofproto)
3223 * Idle time histogram.
3225 * Most of the time a switch has a relatively small number of subfacets.
3226 * When this is the case we might as well keep statistics for all of them
3227 * in userspace and to cache them in the kernel datapath for performance as
3230 * As the number of subfacets increases, the memory required to maintain
3231 * statistics about them in userspace and in the kernel becomes
3232 * significant. However, with a large number of subfacets it is likely
3233 * that only a few of them are "heavy hitters" that consume a large amount
3234 * of bandwidth. At this point, only heavy hitters are worth caching in
3235 * the kernel and maintaining in userspaces; other subfacets we can
3238 * The technique used to compute the idle time is to build a histogram with
3239 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
3240 * that is installed in the kernel gets dropped in the appropriate bucket.
3241 * After the histogram has been built, we compute the cutoff so that only
3242 * the most-recently-used 1% of subfacets (but at least
3243 * ofproto->up.flow_eviction_threshold flows) are kept cached. At least
3244 * the most-recently-used bucket of subfacets is kept, so actually an
3245 * arbitrary number of subfacets can be kept in any given expiration run
3246 * (though the next run will delete most of those unless they receive
3249 * This requires a second pass through the subfacets, in addition to the
3250 * pass made by update_stats(), because the former function never looks at
3251 * uninstallable subfacets.
3253 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
3254 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
3255 int buckets[N_BUCKETS] = { 0 };
3256 int total, subtotal, bucket;
3257 struct subfacet *subfacet;
3261 total = hmap_count(&ofproto->subfacets);
3262 if (total <= ofproto->up.flow_eviction_threshold) {
3263 return N_BUCKETS * BUCKET_WIDTH;
3266 /* Build histogram. */
3268 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
3269 long long int idle = now - subfacet->used;
3270 int bucket = (idle <= 0 ? 0
3271 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
3272 : (unsigned int) idle / BUCKET_WIDTH);
3276 /* Find the first bucket whose flows should be expired. */
3277 subtotal = bucket = 0;
3279 subtotal += buckets[bucket++];
3280 } while (bucket < N_BUCKETS &&
3281 subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100));
3283 if (VLOG_IS_DBG_ENABLED()) {
3287 ds_put_cstr(&s, "keep");
3288 for (i = 0; i < N_BUCKETS; i++) {
3290 ds_put_cstr(&s, ", drop");
3293 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
3296 VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s));
3300 return bucket * BUCKET_WIDTH;
3303 enum { EXPIRE_MAX_BATCH = 50 };
3306 expire_batch(struct ofproto_dpif *ofproto, struct subfacet **subfacets, int n)
3308 struct odputil_keybuf keybufs[EXPIRE_MAX_BATCH];
3309 struct dpif_op ops[EXPIRE_MAX_BATCH];
3310 struct dpif_op *opsp[EXPIRE_MAX_BATCH];
3311 struct ofpbuf keys[EXPIRE_MAX_BATCH];
3312 struct dpif_flow_stats stats[EXPIRE_MAX_BATCH];
3315 for (i = 0; i < n; i++) {
3316 ops[i].type = DPIF_OP_FLOW_DEL;
3317 subfacet_get_key(subfacets[i], &keybufs[i], &keys[i]);
3318 ops[i].u.flow_del.key = keys[i].data;
3319 ops[i].u.flow_del.key_len = keys[i].size;
3320 ops[i].u.flow_del.stats = &stats[i];
3324 dpif_operate(ofproto->dpif, opsp, n);
3325 for (i = 0; i < n; i++) {
3326 subfacet_reset_dp_stats(subfacets[i], &stats[i]);
3327 subfacets[i]->installed = false;
3328 subfacet_destroy(subfacets[i]);
3333 expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle)
3335 long long int cutoff = time_msec() - dp_max_idle;
3337 struct subfacet *subfacet, *next_subfacet;
3338 struct subfacet *batch[EXPIRE_MAX_BATCH];
3342 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
3343 &ofproto->subfacets) {
3344 if (subfacet->used < cutoff) {
3345 if (subfacet->installed) {
3346 batch[n_batch++] = subfacet;
3347 if (n_batch >= EXPIRE_MAX_BATCH) {
3348 expire_batch(ofproto, batch, n_batch);
3352 subfacet_destroy(subfacet);
3358 expire_batch(ofproto, batch, n_batch);
3362 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
3363 * then delete it entirely. */
3365 rule_expire(struct rule_dpif *rule)
3367 struct facet *facet, *next_facet;
3371 /* Has 'rule' expired? */
3373 if (rule->up.hard_timeout
3374 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
3375 reason = OFPRR_HARD_TIMEOUT;
3376 } else if (rule->up.idle_timeout
3377 && now > rule->up.used + rule->up.idle_timeout * 1000) {
3378 reason = OFPRR_IDLE_TIMEOUT;
3383 COVERAGE_INC(ofproto_dpif_expired);
3385 /* Update stats. (This is a no-op if the rule expired due to an idle
3386 * timeout, because that only happens when the rule has no facets left.) */
3387 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
3388 facet_remove(facet);
3391 /* Get rid of the rule. */
3392 ofproto_rule_expire(&rule->up, reason);
3397 /* Creates and returns a new facet owned by 'rule', given a 'flow'.
3399 * The caller must already have determined that no facet with an identical
3400 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
3401 * the ofproto's classifier table.
3403 * 'hash' must be the return value of flow_hash(flow, 0).
3405 * The facet will initially have no subfacets. The caller should create (at
3406 * least) one subfacet with subfacet_create(). */
3407 static struct facet *
3408 facet_create(struct rule_dpif *rule, const struct flow *flow, uint32_t hash)
3410 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3411 struct facet *facet;
3413 facet = xzalloc(sizeof *facet);
3414 facet->used = time_msec();
3415 hmap_insert(&ofproto->facets, &facet->hmap_node, hash);
3416 list_push_back(&rule->facets, &facet->list_node);
3418 facet->flow = *flow;
3419 list_init(&facet->subfacets);
3420 netflow_flow_init(&facet->nf_flow);
3421 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
3427 facet_free(struct facet *facet)
3432 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
3433 * 'packet', which arrived on 'in_port'.
3435 * Takes ownership of 'packet'. */
3437 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
3438 const struct nlattr *odp_actions, size_t actions_len,
3439 struct ofpbuf *packet)
3441 struct odputil_keybuf keybuf;
3445 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
3446 odp_flow_key_from_flow(&key, flow);
3448 error = dpif_execute(ofproto->dpif, key.data, key.size,
3449 odp_actions, actions_len, packet);
3451 ofpbuf_delete(packet);
3455 /* Remove 'facet' from 'ofproto' and free up the associated memory:
3457 * - If 'facet' was installed in the datapath, uninstalls it and updates its
3458 * rule's statistics, via subfacet_uninstall().
3460 * - Removes 'facet' from its rule and from ofproto->facets.
3463 facet_remove(struct facet *facet)
3465 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3466 struct subfacet *subfacet, *next_subfacet;
3468 assert(!list_is_empty(&facet->subfacets));
3470 /* First uninstall all of the subfacets to get final statistics. */
3471 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3472 subfacet_uninstall(subfacet);
3475 /* Flush the final stats to the rule.
3477 * This might require us to have at least one subfacet around so that we
3478 * can use its actions for accounting in facet_account(), which is why we
3479 * have uninstalled but not yet destroyed the subfacets. */
3480 facet_flush_stats(facet);
3482 /* Now we're really all done so destroy everything. */
3483 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
3484 &facet->subfacets) {
3485 subfacet_destroy__(subfacet);
3487 hmap_remove(&ofproto->facets, &facet->hmap_node);
3488 list_remove(&facet->list_node);
3492 /* Feed information from 'facet' back into the learning table to keep it in
3493 * sync with what is actually flowing through the datapath. */
3495 facet_learn(struct facet *facet)
3497 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3498 struct action_xlate_ctx ctx;
3500 if (!facet->has_learn
3501 && !facet->has_normal
3502 && (!facet->has_fin_timeout
3503 || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
3507 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
3508 facet->flow.vlan_tci,
3509 facet->rule, facet->tcp_flags, NULL);
3510 ctx.may_learn = true;
3511 xlate_actions_for_side_effects(&ctx, facet->rule->up.actions,
3512 facet->rule->up.n_actions);
3516 facet_account(struct facet *facet)
3518 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3519 struct subfacet *subfacet;
3520 const struct nlattr *a;
3525 if (!facet->has_normal || !ofproto->has_bonded_bundles) {
3528 n_bytes = facet->byte_count - facet->accounted_bytes;
3530 /* This loop feeds byte counters to bond_account() for rebalancing to use
3531 * as a basis. We also need to track the actual VLAN on which the packet
3532 * is going to be sent to ensure that it matches the one passed to
3533 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
3536 * We use the actions from an arbitrary subfacet because they should all
3537 * be equally valid for our purpose. */
3538 subfacet = CONTAINER_OF(list_front(&facet->subfacets),
3539 struct subfacet, list_node);
3540 vlan_tci = facet->flow.vlan_tci;
3541 NL_ATTR_FOR_EACH_UNSAFE (a, left,
3542 subfacet->actions, subfacet->actions_len) {
3543 const struct ovs_action_push_vlan *vlan;
3544 struct ofport_dpif *port;
3546 switch (nl_attr_type(a)) {
3547 case OVS_ACTION_ATTR_OUTPUT:
3548 port = get_odp_port(ofproto, nl_attr_get_u32(a));
3549 if (port && port->bundle && port->bundle->bond) {
3550 bond_account(port->bundle->bond, &facet->flow,
3551 vlan_tci_to_vid(vlan_tci), n_bytes);
3555 case OVS_ACTION_ATTR_POP_VLAN:
3556 vlan_tci = htons(0);
3559 case OVS_ACTION_ATTR_PUSH_VLAN:
3560 vlan = nl_attr_get(a);
3561 vlan_tci = vlan->vlan_tci;
3567 /* Returns true if the only action for 'facet' is to send to the controller.
3568 * (We don't report NetFlow expiration messages for such facets because they
3569 * are just part of the control logic for the network, not real traffic). */
3571 facet_is_controller_flow(struct facet *facet)
3574 && facet->rule->up.n_actions == 1
3575 && action_outputs_to_port(&facet->rule->up.actions[0],
3576 htons(OFPP_CONTROLLER)));
3579 /* Folds all of 'facet''s statistics into its rule. Also updates the
3580 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
3581 * 'facet''s statistics in the datapath should have been zeroed and folded into
3582 * its packet and byte counts before this function is called. */
3584 facet_flush_stats(struct facet *facet)
3586 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3587 struct subfacet *subfacet;
3589 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3590 assert(!subfacet->dp_byte_count);
3591 assert(!subfacet->dp_packet_count);
3594 facet_push_stats(facet);
3595 if (facet->accounted_bytes < facet->byte_count) {
3596 facet_account(facet);
3597 facet->accounted_bytes = facet->byte_count;
3600 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
3601 struct ofexpired expired;
3602 expired.flow = facet->flow;
3603 expired.packet_count = facet->packet_count;
3604 expired.byte_count = facet->byte_count;
3605 expired.used = facet->used;
3606 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
3609 facet->rule->packet_count += facet->packet_count;
3610 facet->rule->byte_count += facet->byte_count;
3612 /* Reset counters to prevent double counting if 'facet' ever gets
3614 facet_reset_counters(facet);
3616 netflow_flow_clear(&facet->nf_flow);
3617 facet->tcp_flags = 0;
3620 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
3621 * Returns it if found, otherwise a null pointer.
3623 * 'hash' must be the return value of flow_hash(flow, 0).
3625 * The returned facet might need revalidation; use facet_lookup_valid()
3626 * instead if that is important. */
3627 static struct facet *
3628 facet_find(struct ofproto_dpif *ofproto,
3629 const struct flow *flow, uint32_t hash)
3631 struct facet *facet;
3633 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, hash, &ofproto->facets) {
3634 if (flow_equal(flow, &facet->flow)) {
3642 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
3643 * Returns it if found, otherwise a null pointer.
3645 * 'hash' must be the return value of flow_hash(flow, 0).
3647 * The returned facet is guaranteed to be valid. */
3648 static struct facet *
3649 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow,
3652 struct facet *facet = facet_find(ofproto, flow, hash);
3654 /* The facet we found might not be valid, since we could be in need of
3655 * revalidation. If it is not valid, don't return it. */
3657 && (ofproto->need_revalidate
3658 || tag_set_intersects(&ofproto->revalidate_set, facet->tags))
3659 && !facet_revalidate(facet)) {
3660 COVERAGE_INC(facet_invalidated);
3668 facet_check_consistency(struct facet *facet)
3670 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3672 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3674 uint64_t odp_actions_stub[1024 / 8];
3675 struct ofpbuf odp_actions;
3677 struct rule_dpif *rule;
3678 struct subfacet *subfacet;
3679 bool may_log = false;
3682 /* Check the rule for consistency. */
3683 rule = rule_dpif_lookup(ofproto, &facet->flow, 0);
3685 if (!VLOG_DROP_WARN(&rl)) {
3686 char *s = flow_to_string(&facet->flow);
3687 VLOG_WARN("%s: facet should not exist", s);
3691 } else if (rule != facet->rule) {
3692 may_log = !VLOG_DROP_WARN(&rl);
3698 flow_format(&s, &facet->flow);
3699 ds_put_format(&s, ": facet associated with wrong rule (was "
3700 "table=%"PRIu8",", facet->rule->up.table_id);
3701 cls_rule_format(&facet->rule->up.cr, &s);
3702 ds_put_format(&s, ") (should have been table=%"PRIu8",",
3704 cls_rule_format(&rule->up.cr, &s);
3705 ds_put_char(&s, ')');
3707 VLOG_WARN("%s", ds_cstr(&s));
3714 /* Check the datapath actions for consistency. */
3715 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
3716 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3717 struct action_xlate_ctx ctx;
3718 bool actions_changed;
3719 bool should_install;
3721 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
3722 subfacet->initial_tci, rule, 0, NULL);
3723 xlate_actions(&ctx, rule->up.actions, rule->up.n_actions,
3726 should_install = (ctx.may_set_up_flow
3727 && subfacet->key_fitness != ODP_FIT_TOO_LITTLE);
3728 if (!should_install && !subfacet->installed) {
3729 /* The actions for uninstallable flows may vary from one packet to
3730 * the next, so don't compare the actions. */
3734 actions_changed = (subfacet->actions_len != odp_actions.size
3735 || memcmp(subfacet->actions, odp_actions.data,
3736 subfacet->actions_len));
3737 if (should_install != subfacet->installed || actions_changed) {
3739 may_log = !VLOG_DROP_WARN(&rl);
3744 struct odputil_keybuf keybuf;
3749 subfacet_get_key(subfacet, &keybuf, &key);
3750 odp_flow_key_format(key.data, key.size, &s);
3752 ds_put_cstr(&s, ": inconsistency in subfacet");
3753 if (should_install != subfacet->installed) {
3754 enum odp_key_fitness fitness = subfacet->key_fitness;
3756 ds_put_format(&s, " (should%s have been installed)",
3757 should_install ? "" : " not");
3758 ds_put_format(&s, " (may_set_up_flow=%s, fitness=%s)",
3759 ctx.may_set_up_flow ? "true" : "false",
3760 odp_key_fitness_to_string(fitness));
3762 if (actions_changed) {
3763 ds_put_cstr(&s, " (actions were: ");
3764 format_odp_actions(&s, subfacet->actions,
3765 subfacet->actions_len);
3766 ds_put_cstr(&s, ") (correct actions: ");
3767 format_odp_actions(&s, odp_actions.data, odp_actions.size);
3768 ds_put_char(&s, ')');
3770 ds_put_cstr(&s, " (actions: ");
3771 format_odp_actions(&s, subfacet->actions,
3772 subfacet->actions_len);
3773 ds_put_char(&s, ')');
3775 VLOG_WARN("%s", ds_cstr(&s));
3780 ofpbuf_uninit(&odp_actions);
3785 /* Re-searches the classifier for 'facet':
3787 * - If the rule found is different from 'facet''s current rule, moves
3788 * 'facet' to the new rule and recompiles its actions.
3790 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
3791 * where it is and recompiles its actions anyway.
3793 * - If there is none, destroys 'facet'.
3795 * Returns true if 'facet' still exists, false if it has been destroyed. */
3797 facet_revalidate(struct facet *facet)
3799 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3801 struct nlattr *odp_actions;
3804 struct actions *new_actions;
3806 struct action_xlate_ctx ctx;
3807 uint64_t odp_actions_stub[1024 / 8];
3808 struct ofpbuf odp_actions;
3810 struct rule_dpif *new_rule;
3811 struct subfacet *subfacet;
3812 bool actions_changed;
3815 COVERAGE_INC(facet_revalidate);
3817 /* Determine the new rule. */
3818 new_rule = rule_dpif_lookup(ofproto, &facet->flow, 0);
3820 /* No new rule, so delete the facet. */
3821 facet_remove(facet);
3825 /* Calculate new datapath actions.
3827 * We do not modify any 'facet' state yet, because we might need to, e.g.,
3828 * emit a NetFlow expiration and, if so, we need to have the old state
3829 * around to properly compose it. */
3831 /* If the datapath actions changed or the installability changed,
3832 * then we need to talk to the datapath. */
3835 memset(&ctx, 0, sizeof ctx);
3836 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
3837 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3838 bool should_install;
3840 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
3841 subfacet->initial_tci, new_rule, 0, NULL);
3842 xlate_actions(&ctx, new_rule->up.actions, new_rule->up.n_actions,
3844 actions_changed = (subfacet->actions_len != odp_actions.size
3845 || memcmp(subfacet->actions, odp_actions.data,
3846 subfacet->actions_len));
3848 should_install = (ctx.may_set_up_flow
3849 && subfacet->key_fitness != ODP_FIT_TOO_LITTLE);
3850 if (actions_changed || should_install != subfacet->installed) {
3851 if (should_install) {
3852 struct dpif_flow_stats stats;
3854 subfacet_install(subfacet,
3855 odp_actions.data, odp_actions.size, &stats);
3856 subfacet_update_stats(subfacet, &stats);
3858 subfacet_uninstall(subfacet);
3862 new_actions = xcalloc(list_size(&facet->subfacets),
3863 sizeof *new_actions);
3865 new_actions[i].odp_actions = xmemdup(odp_actions.data,
3867 new_actions[i].actions_len = odp_actions.size;
3872 ofpbuf_uninit(&odp_actions);
3875 facet_flush_stats(facet);
3878 /* Update 'facet' now that we've taken care of all the old state. */
3879 facet->tags = ctx.tags;
3880 facet->nf_flow.output_iface = ctx.nf_output_iface;
3881 facet->may_install = ctx.may_set_up_flow;
3882 facet->has_learn = ctx.has_learn;
3883 facet->has_normal = ctx.has_normal;
3884 facet->has_fin_timeout = ctx.has_fin_timeout;
3885 facet->mirrors = ctx.mirrors;
3888 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3889 if (new_actions[i].odp_actions) {
3890 free(subfacet->actions);
3891 subfacet->actions = new_actions[i].odp_actions;
3892 subfacet->actions_len = new_actions[i].actions_len;
3898 if (facet->rule != new_rule) {
3899 COVERAGE_INC(facet_changed_rule);
3900 list_remove(&facet->list_node);
3901 list_push_back(&new_rule->facets, &facet->list_node);
3902 facet->rule = new_rule;
3903 facet->used = new_rule->up.created;
3904 facet->prev_used = facet->used;
3910 /* Updates 'facet''s used time. Caller is responsible for calling
3911 * facet_push_stats() to update the flows which 'facet' resubmits into. */
3913 facet_update_time(struct facet *facet, long long int used)
3915 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3916 if (used > facet->used) {
3918 ofproto_rule_update_used(&facet->rule->up, used);
3919 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
3924 facet_reset_counters(struct facet *facet)
3926 facet->packet_count = 0;
3927 facet->byte_count = 0;
3928 facet->prev_packet_count = 0;
3929 facet->prev_byte_count = 0;
3930 facet->accounted_bytes = 0;
3934 facet_push_stats(struct facet *facet)
3936 struct dpif_flow_stats stats;
3938 assert(facet->packet_count >= facet->prev_packet_count);
3939 assert(facet->byte_count >= facet->prev_byte_count);
3940 assert(facet->used >= facet->prev_used);
3942 stats.n_packets = facet->packet_count - facet->prev_packet_count;
3943 stats.n_bytes = facet->byte_count - facet->prev_byte_count;
3944 stats.used = facet->used;
3945 stats.tcp_flags = 0;
3947 if (stats.n_packets || stats.n_bytes || facet->used > facet->prev_used) {
3948 facet->prev_packet_count = facet->packet_count;
3949 facet->prev_byte_count = facet->byte_count;
3950 facet->prev_used = facet->used;
3952 flow_push_stats(facet->rule, &facet->flow, &stats);
3954 update_mirror_stats(ofproto_dpif_cast(facet->rule->up.ofproto),
3955 facet->mirrors, stats.n_packets, stats.n_bytes);
3960 rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats)
3962 rule->packet_count += stats->n_packets;
3963 rule->byte_count += stats->n_bytes;
3964 ofproto_rule_update_used(&rule->up, stats->used);
3967 /* Pushes flow statistics to the rules which 'flow' resubmits into given
3968 * 'rule''s actions and mirrors. */
3970 flow_push_stats(struct rule_dpif *rule,
3971 const struct flow *flow, const struct dpif_flow_stats *stats)
3973 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3974 struct action_xlate_ctx ctx;
3976 ofproto_rule_update_used(&rule->up, stats->used);
3978 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci, rule,
3980 ctx.resubmit_stats = stats;
3981 xlate_actions_for_side_effects(&ctx, rule->up.actions, rule->up.n_actions);
3986 static struct subfacet *
3987 subfacet_find__(struct ofproto_dpif *ofproto,
3988 const struct nlattr *key, size_t key_len, uint32_t key_hash,
3989 const struct flow *flow)
3991 struct subfacet *subfacet;
3993 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
3994 &ofproto->subfacets) {
3996 ? (subfacet->key_len == key_len
3997 && !memcmp(key, subfacet->key, key_len))
3998 : flow_equal(flow, &subfacet->facet->flow)) {
4006 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
4007 * 'key_fitness', 'key', and 'key_len'. Returns the existing subfacet if
4008 * there is one, otherwise creates and returns a new subfacet.
4010 * If the returned subfacet is new, then subfacet->actions will be NULL, in
4011 * which case the caller must populate the actions with
4012 * subfacet_make_actions(). */
4013 static struct subfacet *
4014 subfacet_create(struct facet *facet, enum odp_key_fitness key_fitness,
4015 const struct nlattr *key, size_t key_len, ovs_be16 initial_tci)
4017 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4018 uint32_t key_hash = odp_flow_key_hash(key, key_len);
4019 struct subfacet *subfacet;
4021 subfacet = subfacet_find__(ofproto, key, key_len, key_hash, &facet->flow);
4023 if (subfacet->facet == facet) {
4027 /* This shouldn't happen. */
4028 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
4029 subfacet_destroy(subfacet);
4032 subfacet = (list_is_empty(&facet->subfacets)
4033 ? &facet->one_subfacet
4034 : xmalloc(sizeof *subfacet));
4035 hmap_insert(&ofproto->subfacets, &subfacet->hmap_node, key_hash);
4036 list_push_back(&facet->subfacets, &subfacet->list_node);
4037 subfacet->facet = facet;
4038 subfacet->key_fitness = key_fitness;
4039 if (key_fitness != ODP_FIT_PERFECT) {
4040 subfacet->key = xmemdup(key, key_len);
4041 subfacet->key_len = key_len;
4043 subfacet->key = NULL;
4044 subfacet->key_len = 0;
4046 subfacet->used = time_msec();
4047 subfacet->dp_packet_count = 0;
4048 subfacet->dp_byte_count = 0;
4049 subfacet->actions_len = 0;
4050 subfacet->actions = NULL;
4051 subfacet->installed = false;
4052 subfacet->initial_tci = initial_tci;
4057 /* Searches 'ofproto' for a subfacet with the given 'key', 'key_len', and
4058 * 'flow'. Returns the subfacet if one exists, otherwise NULL. */
4059 static struct subfacet *
4060 subfacet_find(struct ofproto_dpif *ofproto,
4061 const struct nlattr *key, size_t key_len)
4063 uint32_t key_hash = odp_flow_key_hash(key, key_len);
4064 enum odp_key_fitness fitness;
4067 fitness = odp_flow_key_to_flow(key, key_len, &flow);
4068 if (fitness == ODP_FIT_ERROR) {
4072 return subfacet_find__(ofproto, key, key_len, key_hash, &flow);
4075 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
4076 * its facet within 'ofproto', and frees it. */
4078 subfacet_destroy__(struct subfacet *subfacet)
4080 struct facet *facet = subfacet->facet;
4081 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4083 subfacet_uninstall(subfacet);
4084 hmap_remove(&ofproto->subfacets, &subfacet->hmap_node);
4085 list_remove(&subfacet->list_node);
4086 free(subfacet->key);
4087 free(subfacet->actions);
4088 if (subfacet != &facet->one_subfacet) {
4093 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
4094 * last remaining subfacet in its facet destroys the facet too. */
4096 subfacet_destroy(struct subfacet *subfacet)
4098 struct facet *facet = subfacet->facet;
4100 if (list_is_singleton(&facet->subfacets)) {
4101 /* facet_remove() needs at least one subfacet (it will remove it). */
4102 facet_remove(facet);
4104 subfacet_destroy__(subfacet);
4108 /* Initializes 'key' with the sequence of OVS_KEY_ATTR_* Netlink attributes
4109 * that can be used to refer to 'subfacet'. The caller must provide 'keybuf'
4110 * for use as temporary storage. */
4112 subfacet_get_key(struct subfacet *subfacet, struct odputil_keybuf *keybuf,
4115 if (!subfacet->key) {
4116 ofpbuf_use_stack(key, keybuf, sizeof *keybuf);
4117 odp_flow_key_from_flow(key, &subfacet->facet->flow);
4119 ofpbuf_use_const(key, subfacet->key, subfacet->key_len);
4123 /* Composes the datapath actions for 'subfacet' based on its rule's actions.
4124 * Translates the actions into 'odp_actions', which the caller must have
4125 * initialized and is responsible for uninitializing. */
4127 subfacet_make_actions(struct subfacet *subfacet, const struct ofpbuf *packet,
4128 struct ofpbuf *odp_actions)
4130 struct facet *facet = subfacet->facet;
4131 struct rule_dpif *rule = facet->rule;
4132 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4134 struct action_xlate_ctx ctx;
4136 action_xlate_ctx_init(&ctx, ofproto, &facet->flow, subfacet->initial_tci,
4138 xlate_actions(&ctx, rule->up.actions, rule->up.n_actions, odp_actions);
4139 facet->tags = ctx.tags;
4140 facet->may_install = ctx.may_set_up_flow;
4141 facet->has_learn = ctx.has_learn;
4142 facet->has_normal = ctx.has_normal;
4143 facet->has_fin_timeout = ctx.has_fin_timeout;
4144 facet->nf_flow.output_iface = ctx.nf_output_iface;
4145 facet->mirrors = ctx.mirrors;
4147 if (subfacet->actions_len != odp_actions->size
4148 || memcmp(subfacet->actions, odp_actions->data, odp_actions->size)) {
4149 free(subfacet->actions);
4150 subfacet->actions_len = odp_actions->size;
4151 subfacet->actions = xmemdup(odp_actions->data, odp_actions->size);
4155 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
4156 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
4157 * in the datapath will be zeroed and 'stats' will be updated with traffic new
4158 * since 'subfacet' was last updated.
4160 * Returns 0 if successful, otherwise a positive errno value. */
4162 subfacet_install(struct subfacet *subfacet,
4163 const struct nlattr *actions, size_t actions_len,
4164 struct dpif_flow_stats *stats)
4166 struct facet *facet = subfacet->facet;
4167 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4168 struct odputil_keybuf keybuf;
4169 enum dpif_flow_put_flags flags;
4173 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
4175 flags |= DPIF_FP_ZERO_STATS;
4178 subfacet_get_key(subfacet, &keybuf, &key);
4179 ret = dpif_flow_put(ofproto->dpif, flags, key.data, key.size,
4180 actions, actions_len, stats);
4183 subfacet_reset_dp_stats(subfacet, stats);
4189 /* If 'subfacet' is installed in the datapath, uninstalls it. */
4191 subfacet_uninstall(struct subfacet *subfacet)
4193 if (subfacet->installed) {
4194 struct rule_dpif *rule = subfacet->facet->rule;
4195 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4196 struct odputil_keybuf keybuf;
4197 struct dpif_flow_stats stats;
4201 subfacet_get_key(subfacet, &keybuf, &key);
4202 error = dpif_flow_del(ofproto->dpif, key.data, key.size, &stats);
4203 subfacet_reset_dp_stats(subfacet, &stats);
4205 subfacet_update_stats(subfacet, &stats);
4207 subfacet->installed = false;
4209 assert(subfacet->dp_packet_count == 0);
4210 assert(subfacet->dp_byte_count == 0);
4214 /* Resets 'subfacet''s datapath statistics counters. This should be called
4215 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
4216 * non-null, it should contain the statistics returned by dpif when 'subfacet'
4217 * was reset in the datapath. 'stats' will be modified to include only
4218 * statistics new since 'subfacet' was last updated. */
4220 subfacet_reset_dp_stats(struct subfacet *subfacet,
4221 struct dpif_flow_stats *stats)
4224 && subfacet->dp_packet_count <= stats->n_packets
4225 && subfacet->dp_byte_count <= stats->n_bytes) {
4226 stats->n_packets -= subfacet->dp_packet_count;
4227 stats->n_bytes -= subfacet->dp_byte_count;
4230 subfacet->dp_packet_count = 0;
4231 subfacet->dp_byte_count = 0;
4234 /* Updates 'subfacet''s used time. The caller is responsible for calling
4235 * facet_push_stats() to update the flows which 'subfacet' resubmits into. */
4237 subfacet_update_time(struct subfacet *subfacet, long long int used)
4239 if (used > subfacet->used) {
4240 subfacet->used = used;
4241 facet_update_time(subfacet->facet, used);
4245 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
4247 * Because of the meaning of a subfacet's counters, it only makes sense to do
4248 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
4249 * represents a packet that was sent by hand or if it represents statistics
4250 * that have been cleared out of the datapath. */
4252 subfacet_update_stats(struct subfacet *subfacet,
4253 const struct dpif_flow_stats *stats)
4255 if (stats->n_packets || stats->used > subfacet->used) {
4256 struct facet *facet = subfacet->facet;
4258 subfacet_update_time(subfacet, stats->used);
4259 facet->packet_count += stats->n_packets;
4260 facet->byte_count += stats->n_bytes;
4261 facet->tcp_flags |= stats->tcp_flags;
4262 facet_push_stats(facet);
4263 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
4269 static struct rule_dpif *
4270 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow,
4273 struct cls_rule *cls_rule;
4274 struct classifier *cls;
4276 if (table_id >= N_TABLES) {
4280 cls = &ofproto->up.tables[table_id].cls;
4281 if (flow->nw_frag & FLOW_NW_FRAG_ANY
4282 && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
4283 /* For OFPC_NORMAL frag_handling, we must pretend that transport ports
4284 * are unavailable. */
4285 struct flow ofpc_normal_flow = *flow;
4286 ofpc_normal_flow.tp_src = htons(0);
4287 ofpc_normal_flow.tp_dst = htons(0);
4288 cls_rule = classifier_lookup(cls, &ofpc_normal_flow);
4290 cls_rule = classifier_lookup(cls, flow);
4292 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
4296 complete_operation(struct rule_dpif *rule)
4298 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4300 rule_invalidate(rule);
4302 struct dpif_completion *c = xmalloc(sizeof *c);
4303 c->op = rule->up.pending;
4304 list_push_back(&ofproto->completions, &c->list_node);
4306 ofoperation_complete(rule->up.pending, 0);
4310 static struct rule *
4313 struct rule_dpif *rule = xmalloc(sizeof *rule);
4318 rule_dealloc(struct rule *rule_)
4320 struct rule_dpif *rule = rule_dpif_cast(rule_);
4325 rule_construct(struct rule *rule_)
4327 struct rule_dpif *rule = rule_dpif_cast(rule_);
4328 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4329 struct rule_dpif *victim;
4333 error = validate_actions(rule->up.actions, rule->up.n_actions,
4334 &rule->up.cr.flow, ofproto->max_ports);
4339 rule->packet_count = 0;
4340 rule->byte_count = 0;
4342 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
4343 if (victim && !list_is_empty(&victim->facets)) {
4344 struct facet *facet;
4346 rule->facets = victim->facets;
4347 list_moved(&rule->facets);
4348 LIST_FOR_EACH (facet, list_node, &rule->facets) {
4349 /* XXX: We're only clearing our local counters here. It's possible
4350 * that quite a few packets are unaccounted for in the datapath
4351 * statistics. These will be accounted to the new rule instead of
4352 * cleared as required. This could be fixed by clearing out the
4353 * datapath statistics for this facet, but currently it doesn't
4355 facet_reset_counters(facet);
4359 /* Must avoid list_moved() in this case. */
4360 list_init(&rule->facets);
4363 table_id = rule->up.table_id;
4364 rule->tag = (victim ? victim->tag
4366 : rule_calculate_tag(&rule->up.cr.flow, &rule->up.cr.wc,
4367 ofproto->tables[table_id].basis));
4369 complete_operation(rule);
4374 rule_destruct(struct rule *rule_)
4376 struct rule_dpif *rule = rule_dpif_cast(rule_);
4377 struct facet *facet, *next_facet;
4379 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4380 facet_revalidate(facet);
4383 complete_operation(rule);
4387 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
4389 struct rule_dpif *rule = rule_dpif_cast(rule_);
4390 struct facet *facet;
4392 /* Start from historical data for 'rule' itself that are no longer tracked
4393 * in facets. This counts, for example, facets that have expired. */
4394 *packets = rule->packet_count;
4395 *bytes = rule->byte_count;
4397 /* Add any statistics that are tracked by facets. This includes
4398 * statistical data recently updated by ofproto_update_stats() as well as
4399 * stats for packets that were executed "by hand" via dpif_execute(). */
4400 LIST_FOR_EACH (facet, list_node, &rule->facets) {
4401 *packets += facet->packet_count;
4402 *bytes += facet->byte_count;
4407 rule_execute(struct rule *rule_, const struct flow *flow,
4408 struct ofpbuf *packet)
4410 struct rule_dpif *rule = rule_dpif_cast(rule_);
4411 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4413 struct dpif_flow_stats stats;
4415 struct action_xlate_ctx ctx;
4416 uint64_t odp_actions_stub[1024 / 8];
4417 struct ofpbuf odp_actions;
4419 dpif_flow_stats_extract(flow, packet, &stats);
4420 rule_credit_stats(rule, &stats);
4422 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
4423 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci,
4424 rule, stats.tcp_flags, packet);
4425 ctx.resubmit_stats = &stats;
4426 xlate_actions(&ctx, rule->up.actions, rule->up.n_actions, &odp_actions);
4428 execute_odp_actions(ofproto, flow, odp_actions.data,
4429 odp_actions.size, packet);
4431 ofpbuf_uninit(&odp_actions);
4437 rule_modify_actions(struct rule *rule_)
4439 struct rule_dpif *rule = rule_dpif_cast(rule_);
4440 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4443 error = validate_actions(rule->up.actions, rule->up.n_actions,
4444 &rule->up.cr.flow, ofproto->max_ports);
4446 ofoperation_complete(rule->up.pending, error);
4450 complete_operation(rule);
4453 /* Sends 'packet' out 'ofport'.
4454 * May modify 'packet'.
4455 * Returns 0 if successful, otherwise a positive errno value. */
4457 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
4459 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
4460 struct ofpbuf key, odp_actions;
4461 struct odputil_keybuf keybuf;
4466 flow_extract((struct ofpbuf *) packet, 0, 0, 0, &flow);
4467 odp_port = vsp_realdev_to_vlandev(ofproto, ofport->odp_port,
4469 if (odp_port != ofport->odp_port) {
4470 eth_pop_vlan(packet);
4471 flow.vlan_tci = htons(0);
4474 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4475 odp_flow_key_from_flow(&key, &flow);
4477 ofpbuf_init(&odp_actions, 32);
4478 compose_sflow_action(ofproto, &odp_actions, &flow, odp_port);
4480 nl_msg_put_u32(&odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port);
4481 error = dpif_execute(ofproto->dpif,
4483 odp_actions.data, odp_actions.size,
4485 ofpbuf_uninit(&odp_actions);
4488 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)",
4489 ofproto->up.name, odp_port, strerror(error));
4491 ofproto_update_local_port_stats(ofport->up.ofproto, packet->size, 0);
4495 /* OpenFlow to datapath action translation. */
4497 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
4498 struct action_xlate_ctx *ctx);
4499 static void xlate_normal(struct action_xlate_ctx *);
4502 put_userspace_action(const struct ofproto_dpif *ofproto,
4503 struct ofpbuf *odp_actions,
4504 const struct flow *flow,
4505 const struct user_action_cookie *cookie)
4509 pid = dpif_port_get_pid(ofproto->dpif,
4510 ofp_port_to_odp_port(flow->in_port));
4512 return odp_put_userspace_action(pid, cookie, odp_actions);
4515 /* Compose SAMPLE action for sFlow. */
4517 compose_sflow_action(const struct ofproto_dpif *ofproto,
4518 struct ofpbuf *odp_actions,
4519 const struct flow *flow,
4522 uint32_t port_ifindex;
4523 uint32_t probability;
4524 struct user_action_cookie cookie;
4525 size_t sample_offset, actions_offset;
4526 int cookie_offset, n_output;
4528 if (!ofproto->sflow || flow->in_port == OFPP_NONE) {
4532 if (odp_port == OVSP_NONE) {
4536 port_ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port);
4540 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
4542 /* Number of packets out of UINT_MAX to sample. */
4543 probability = dpif_sflow_get_probability(ofproto->sflow);
4544 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
4546 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
4548 cookie.type = USER_ACTION_COOKIE_SFLOW;
4549 cookie.data = port_ifindex;
4550 cookie.n_output = n_output;
4551 cookie.vlan_tci = 0;
4552 cookie_offset = put_userspace_action(ofproto, odp_actions, flow, &cookie);
4554 nl_msg_end_nested(odp_actions, actions_offset);
4555 nl_msg_end_nested(odp_actions, sample_offset);
4556 return cookie_offset;
4559 /* SAMPLE action must be first action in any given list of actions.
4560 * At this point we do not have all information required to build it. So try to
4561 * build sample action as complete as possible. */
4563 add_sflow_action(struct action_xlate_ctx *ctx)
4565 ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto,
4567 &ctx->flow, OVSP_NONE);
4568 ctx->sflow_odp_port = 0;
4569 ctx->sflow_n_outputs = 0;
4572 /* Fix SAMPLE action according to data collected while composing ODP actions.
4573 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
4574 * USERSPACE action's user-cookie which is required for sflow. */
4576 fix_sflow_action(struct action_xlate_ctx *ctx)
4578 const struct flow *base = &ctx->base_flow;
4579 struct user_action_cookie *cookie;
4581 if (!ctx->user_cookie_offset) {
4585 cookie = ofpbuf_at(ctx->odp_actions, ctx->user_cookie_offset,
4587 assert(cookie != NULL);
4588 assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
4590 if (ctx->sflow_n_outputs) {
4591 cookie->data = dpif_sflow_odp_port_to_ifindex(ctx->ofproto->sflow,
4592 ctx->sflow_odp_port);
4594 if (ctx->sflow_n_outputs >= 255) {
4595 cookie->n_output = 255;
4597 cookie->n_output = ctx->sflow_n_outputs;
4599 cookie->vlan_tci = base->vlan_tci;
4603 compose_output_action__(struct action_xlate_ctx *ctx, uint16_t ofp_port,
4606 const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
4607 uint16_t odp_port = ofp_port_to_odp_port(ofp_port);
4608 ovs_be16 flow_vlan_tci = ctx->flow.vlan_tci;
4609 uint8_t flow_nw_tos = ctx->flow.nw_tos;
4613 struct priority_to_dscp *pdscp;
4615 if (ofport->up.pp.config & OFPUTIL_PC_NO_FWD
4616 || (check_stp && !stp_forward_in_state(ofport->stp_state))) {
4620 pdscp = get_priority(ofport, ctx->flow.skb_priority);
4622 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
4623 ctx->flow.nw_tos |= pdscp->dscp;
4626 /* We may not have an ofport record for this port, but it doesn't hurt
4627 * to allow forwarding to it anyhow. Maybe such a port will appear
4628 * later and we're pre-populating the flow table. */
4631 out_port = vsp_realdev_to_vlandev(ctx->ofproto, odp_port,
4632 ctx->flow.vlan_tci);
4633 if (out_port != odp_port) {
4634 ctx->flow.vlan_tci = htons(0);
4636 commit_odp_actions(&ctx->flow, &ctx->base_flow, ctx->odp_actions);
4637 nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_OUTPUT, out_port);
4639 ctx->sflow_odp_port = odp_port;
4640 ctx->sflow_n_outputs++;
4641 ctx->nf_output_iface = ofp_port;
4642 ctx->flow.vlan_tci = flow_vlan_tci;
4643 ctx->flow.nw_tos = flow_nw_tos;
4647 compose_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port)
4649 compose_output_action__(ctx, ofp_port, true);
4653 xlate_table_action(struct action_xlate_ctx *ctx,
4654 uint16_t in_port, uint8_t table_id)
4656 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
4657 struct ofproto_dpif *ofproto = ctx->ofproto;
4658 struct rule_dpif *rule;
4659 uint16_t old_in_port;
4660 uint8_t old_table_id;
4662 old_table_id = ctx->table_id;
4663 ctx->table_id = table_id;
4665 /* Look up a flow with 'in_port' as the input port. */
4666 old_in_port = ctx->flow.in_port;
4667 ctx->flow.in_port = in_port;
4668 rule = rule_dpif_lookup(ofproto, &ctx->flow, table_id);
4671 if (table_id > 0 && table_id < N_TABLES) {
4672 struct table_dpif *table = &ofproto->tables[table_id];
4673 if (table->other_table) {
4674 ctx->tags |= (rule && rule->tag
4676 : rule_calculate_tag(&ctx->flow,
4677 &table->other_table->wc,
4682 /* Restore the original input port. Otherwise OFPP_NORMAL and
4683 * OFPP_IN_PORT will have surprising behavior. */
4684 ctx->flow.in_port = old_in_port;
4686 if (ctx->resubmit_hook) {
4687 ctx->resubmit_hook(ctx, rule);
4691 struct rule_dpif *old_rule = ctx->rule;
4693 if (ctx->resubmit_stats) {
4694 rule_credit_stats(rule, ctx->resubmit_stats);
4699 do_xlate_actions(rule->up.actions, rule->up.n_actions, ctx);
4700 ctx->rule = old_rule;
4704 ctx->table_id = old_table_id;
4706 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
4708 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
4709 MAX_RESUBMIT_RECURSION);
4710 ctx->max_resubmit_trigger = true;
4715 xlate_resubmit_table(struct action_xlate_ctx *ctx,
4716 const struct nx_action_resubmit *nar)
4721 in_port = (nar->in_port == htons(OFPP_IN_PORT)
4723 : ntohs(nar->in_port));
4724 table_id = nar->table == 255 ? ctx->table_id : nar->table;
4726 xlate_table_action(ctx, in_port, table_id);
4730 flood_packets(struct action_xlate_ctx *ctx, bool all)
4732 struct ofport_dpif *ofport;
4734 HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) {
4735 uint16_t ofp_port = ofport->up.ofp_port;
4737 if (ofp_port == ctx->flow.in_port) {
4742 compose_output_action__(ctx, ofp_port, false);
4743 } else if (!(ofport->up.pp.config & OFPUTIL_PC_NO_FLOOD)) {
4744 compose_output_action(ctx, ofp_port);
4748 ctx->nf_output_iface = NF_OUT_FLOOD;
4752 execute_controller_action(struct action_xlate_ctx *ctx, int len,
4753 enum ofp_packet_in_reason reason,
4754 uint16_t controller_id)
4756 struct ofputil_packet_in pin;
4757 struct ofpbuf *packet;
4759 ctx->may_set_up_flow = false;
4764 packet = ofpbuf_clone(ctx->packet);
4766 if (packet->l2 && packet->l3) {
4767 struct eth_header *eh;
4769 eth_pop_vlan(packet);
4772 /* If the Ethernet type is less than ETH_TYPE_MIN, it's likely an 802.2
4773 * LLC frame. Calculating the Ethernet type of these frames is more
4774 * trouble than seems appropriate for a simple assertion. */
4775 assert(ntohs(eh->eth_type) < ETH_TYPE_MIN
4776 || eh->eth_type == ctx->flow.dl_type);
4778 memcpy(eh->eth_src, ctx->flow.dl_src, sizeof eh->eth_src);
4779 memcpy(eh->eth_dst, ctx->flow.dl_dst, sizeof eh->eth_dst);
4781 if (ctx->flow.vlan_tci & htons(VLAN_CFI)) {
4782 eth_push_vlan(packet, ctx->flow.vlan_tci);
4786 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
4787 packet_set_ipv4(packet, ctx->flow.nw_src, ctx->flow.nw_dst,
4788 ctx->flow.nw_tos, ctx->flow.nw_ttl);
4792 if (ctx->flow.nw_proto == IPPROTO_TCP) {
4793 packet_set_tcp_port(packet, ctx->flow.tp_src,
4795 } else if (ctx->flow.nw_proto == IPPROTO_UDP) {
4796 packet_set_udp_port(packet, ctx->flow.tp_src,
4803 pin.packet = packet->data;
4804 pin.packet_len = packet->size;
4805 pin.reason = reason;
4806 pin.controller_id = controller_id;
4807 pin.table_id = ctx->table_id;
4808 pin.cookie = ctx->rule ? ctx->rule->up.flow_cookie : 0;
4811 flow_get_metadata(&ctx->flow, &pin.fmd);
4813 connmgr_send_packet_in(ctx->ofproto->up.connmgr, &pin);
4814 ofpbuf_delete(packet);
4818 compose_dec_ttl(struct action_xlate_ctx *ctx)
4820 if (ctx->flow.dl_type != htons(ETH_TYPE_IP) &&
4821 ctx->flow.dl_type != htons(ETH_TYPE_IPV6)) {
4825 if (ctx->flow.nw_ttl > 1) {
4829 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL, 0);
4831 /* Stop processing for current table. */
4837 xlate_output_action__(struct action_xlate_ctx *ctx,
4838 uint16_t port, uint16_t max_len)
4840 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
4842 ctx->nf_output_iface = NF_OUT_DROP;
4846 compose_output_action(ctx, ctx->flow.in_port);
4849 xlate_table_action(ctx, ctx->flow.in_port, ctx->table_id);
4855 flood_packets(ctx, false);
4858 flood_packets(ctx, true);
4860 case OFPP_CONTROLLER:
4861 execute_controller_action(ctx, max_len, OFPR_ACTION, 0);
4867 if (port != ctx->flow.in_port) {
4868 compose_output_action(ctx, port);
4873 if (prev_nf_output_iface == NF_OUT_FLOOD) {
4874 ctx->nf_output_iface = NF_OUT_FLOOD;
4875 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
4876 ctx->nf_output_iface = prev_nf_output_iface;
4877 } else if (prev_nf_output_iface != NF_OUT_DROP &&
4878 ctx->nf_output_iface != NF_OUT_FLOOD) {
4879 ctx->nf_output_iface = NF_OUT_MULTI;
4884 xlate_output_reg_action(struct action_xlate_ctx *ctx,
4885 const struct nx_action_output_reg *naor)
4887 struct mf_subfield src;
4890 nxm_decode(&src, naor->src, naor->ofs_nbits);
4891 ofp_port = mf_get_subfield(&src, &ctx->flow);
4893 if (ofp_port <= UINT16_MAX) {
4894 xlate_output_action__(ctx, ofp_port, ntohs(naor->max_len));
4899 xlate_output_action(struct action_xlate_ctx *ctx,
4900 const struct ofp_action_output *oao)
4902 xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len));
4906 xlate_enqueue_action(struct action_xlate_ctx *ctx,
4907 const struct ofp_action_enqueue *oae)
4910 uint32_t flow_priority, priority;
4913 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id),
4916 /* Fall back to ordinary output action. */
4917 xlate_output_action__(ctx, ntohs(oae->port), 0);
4921 /* Figure out datapath output port. */
4922 ofp_port = ntohs(oae->port);
4923 if (ofp_port == OFPP_IN_PORT) {
4924 ofp_port = ctx->flow.in_port;
4925 } else if (ofp_port == ctx->flow.in_port) {
4929 /* Add datapath actions. */
4930 flow_priority = ctx->flow.skb_priority;
4931 ctx->flow.skb_priority = priority;
4932 compose_output_action(ctx, ofp_port);
4933 ctx->flow.skb_priority = flow_priority;
4935 /* Update NetFlow output port. */
4936 if (ctx->nf_output_iface == NF_OUT_DROP) {
4937 ctx->nf_output_iface = ofp_port;
4938 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
4939 ctx->nf_output_iface = NF_OUT_MULTI;
4944 xlate_set_queue_action(struct action_xlate_ctx *ctx,
4945 const struct nx_action_set_queue *nasq)
4950 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(nasq->queue_id),
4953 /* Couldn't translate queue to a priority, so ignore. A warning
4954 * has already been logged. */
4958 ctx->flow.skb_priority = priority;
4961 struct xlate_reg_state {
4967 xlate_autopath(struct action_xlate_ctx *ctx,
4968 const struct nx_action_autopath *naa)
4970 uint16_t ofp_port = ntohl(naa->id);
4971 struct ofport_dpif *port = get_ofp_port(ctx->ofproto, ofp_port);
4973 if (!port || !port->bundle) {
4974 ofp_port = OFPP_NONE;
4975 } else if (port->bundle->bond) {
4976 /* Autopath does not support VLAN hashing. */
4977 struct ofport_dpif *slave = bond_choose_output_slave(
4978 port->bundle->bond, &ctx->flow, 0, &ctx->tags);
4980 ofp_port = slave->up.ofp_port;
4983 autopath_execute(naa, &ctx->flow, ofp_port);
4987 slave_enabled_cb(uint16_t ofp_port, void *ofproto_)
4989 struct ofproto_dpif *ofproto = ofproto_;
4990 struct ofport_dpif *port;
5000 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
5003 port = get_ofp_port(ofproto, ofp_port);
5004 return port ? port->may_enable : false;
5009 xlate_learn_action(struct action_xlate_ctx *ctx,
5010 const struct nx_action_learn *learn)
5012 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
5013 struct ofputil_flow_mod fm;
5016 learn_execute(learn, &ctx->flow, &fm);
5018 error = ofproto_flow_mod(&ctx->ofproto->up, &fm);
5019 if (error && !VLOG_DROP_WARN(&rl)) {
5020 VLOG_WARN("learning action failed to modify flow table (%s)",
5021 ofperr_get_name(error));
5027 /* Reduces '*timeout' to no more than 'max'. A value of zero in either case
5028 * means "infinite". */
5030 reduce_timeout(uint16_t max, uint16_t *timeout)
5032 if (max && (!*timeout || *timeout > max)) {
5038 xlate_fin_timeout(struct action_xlate_ctx *ctx,
5039 const struct nx_action_fin_timeout *naft)
5041 if (ctx->tcp_flags & (TCP_FIN | TCP_RST) && ctx->rule) {
5042 struct rule_dpif *rule = ctx->rule;
5044 reduce_timeout(ntohs(naft->fin_idle_timeout), &rule->up.idle_timeout);
5045 reduce_timeout(ntohs(naft->fin_hard_timeout), &rule->up.hard_timeout);
5050 may_receive(const struct ofport_dpif *port, struct action_xlate_ctx *ctx)
5052 if (port->up.pp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
5053 ? OFPUTIL_PC_NO_RECV_STP
5054 : OFPUTIL_PC_NO_RECV)) {
5058 /* Only drop packets here if both forwarding and learning are
5059 * disabled. If just learning is enabled, we need to have
5060 * OFPP_NORMAL and the learning action have a look at the packet
5061 * before we can drop it. */
5062 if (!stp_forward_in_state(port->stp_state)
5063 && !stp_learn_in_state(port->stp_state)) {
5071 do_xlate_actions(const union ofp_action *in, size_t n_in,
5072 struct action_xlate_ctx *ctx)
5074 const struct ofport_dpif *port;
5075 const union ofp_action *ia;
5076 bool was_evictable = true;
5079 port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
5080 if (port && !may_receive(port, ctx)) {
5081 /* Drop this flow. */
5086 /* Don't let the rule we're working on get evicted underneath us. */
5087 was_evictable = ctx->rule->up.evictable;
5088 ctx->rule->up.evictable = false;
5090 OFPUTIL_ACTION_FOR_EACH_UNSAFE (ia, left, in, n_in) {
5091 const struct ofp_action_dl_addr *oada;
5092 const struct nx_action_resubmit *nar;
5093 const struct nx_action_set_tunnel *nast;
5094 const struct nx_action_set_queue *nasq;
5095 const struct nx_action_multipath *nam;
5096 const struct nx_action_autopath *naa;
5097 const struct nx_action_bundle *nab;
5098 const struct nx_action_output_reg *naor;
5099 const struct nx_action_controller *nac;
5100 enum ofputil_action_code code;
5107 code = ofputil_decode_action_unsafe(ia);
5109 case OFPUTIL_OFPAT10_OUTPUT:
5110 xlate_output_action(ctx, &ia->output);
5113 case OFPUTIL_OFPAT10_SET_VLAN_VID:
5114 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
5115 ctx->flow.vlan_tci |= ia->vlan_vid.vlan_vid | htons(VLAN_CFI);
5118 case OFPUTIL_OFPAT10_SET_VLAN_PCP:
5119 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
5120 ctx->flow.vlan_tci |= htons(
5121 (ia->vlan_pcp.vlan_pcp << VLAN_PCP_SHIFT) | VLAN_CFI);
5124 case OFPUTIL_OFPAT10_STRIP_VLAN:
5125 ctx->flow.vlan_tci = htons(0);
5128 case OFPUTIL_OFPAT10_SET_DL_SRC:
5129 oada = ((struct ofp_action_dl_addr *) ia);
5130 memcpy(ctx->flow.dl_src, oada->dl_addr, ETH_ADDR_LEN);
5133 case OFPUTIL_OFPAT10_SET_DL_DST:
5134 oada = ((struct ofp_action_dl_addr *) ia);
5135 memcpy(ctx->flow.dl_dst, oada->dl_addr, ETH_ADDR_LEN);
5138 case OFPUTIL_OFPAT10_SET_NW_SRC:
5139 ctx->flow.nw_src = ia->nw_addr.nw_addr;
5142 case OFPUTIL_OFPAT10_SET_NW_DST:
5143 ctx->flow.nw_dst = ia->nw_addr.nw_addr;
5146 case OFPUTIL_OFPAT10_SET_NW_TOS:
5147 /* OpenFlow 1.0 only supports IPv4. */
5148 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
5149 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
5150 ctx->flow.nw_tos |= ia->nw_tos.nw_tos & IP_DSCP_MASK;
5154 case OFPUTIL_OFPAT10_SET_TP_SRC:
5155 ctx->flow.tp_src = ia->tp_port.tp_port;
5158 case OFPUTIL_OFPAT10_SET_TP_DST:
5159 ctx->flow.tp_dst = ia->tp_port.tp_port;
5162 case OFPUTIL_OFPAT10_ENQUEUE:
5163 xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia);
5166 case OFPUTIL_NXAST_RESUBMIT:
5167 nar = (const struct nx_action_resubmit *) ia;
5168 xlate_table_action(ctx, ntohs(nar->in_port), ctx->table_id);
5171 case OFPUTIL_NXAST_RESUBMIT_TABLE:
5172 xlate_resubmit_table(ctx, (const struct nx_action_resubmit *) ia);
5175 case OFPUTIL_NXAST_SET_TUNNEL:
5176 nast = (const struct nx_action_set_tunnel *) ia;
5177 tun_id = htonll(ntohl(nast->tun_id));
5178 ctx->flow.tun_id = tun_id;
5181 case OFPUTIL_NXAST_SET_QUEUE:
5182 nasq = (const struct nx_action_set_queue *) ia;
5183 xlate_set_queue_action(ctx, nasq);
5186 case OFPUTIL_NXAST_POP_QUEUE:
5187 ctx->flow.skb_priority = ctx->orig_skb_priority;
5190 case OFPUTIL_NXAST_REG_MOVE:
5191 nxm_execute_reg_move((const struct nx_action_reg_move *) ia,
5195 case OFPUTIL_NXAST_REG_LOAD:
5196 nxm_execute_reg_load((const struct nx_action_reg_load *) ia,
5200 case OFPUTIL_NXAST_NOTE:
5201 /* Nothing to do. */
5204 case OFPUTIL_NXAST_SET_TUNNEL64:
5205 tun_id = ((const struct nx_action_set_tunnel64 *) ia)->tun_id;
5206 ctx->flow.tun_id = tun_id;
5209 case OFPUTIL_NXAST_MULTIPATH:
5210 nam = (const struct nx_action_multipath *) ia;
5211 multipath_execute(nam, &ctx->flow);
5214 case OFPUTIL_NXAST_AUTOPATH:
5215 naa = (const struct nx_action_autopath *) ia;
5216 xlate_autopath(ctx, naa);
5219 case OFPUTIL_NXAST_BUNDLE:
5220 ctx->ofproto->has_bundle_action = true;
5221 nab = (const struct nx_action_bundle *) ia;
5222 xlate_output_action__(ctx, bundle_execute(nab, &ctx->flow,
5227 case OFPUTIL_NXAST_BUNDLE_LOAD:
5228 ctx->ofproto->has_bundle_action = true;
5229 nab = (const struct nx_action_bundle *) ia;
5230 bundle_execute_load(nab, &ctx->flow, slave_enabled_cb,
5234 case OFPUTIL_NXAST_OUTPUT_REG:
5235 naor = (const struct nx_action_output_reg *) ia;
5236 xlate_output_reg_action(ctx, naor);
5239 case OFPUTIL_NXAST_LEARN:
5240 ctx->has_learn = true;
5241 if (ctx->may_learn) {
5242 xlate_learn_action(ctx, (const struct nx_action_learn *) ia);
5246 case OFPUTIL_NXAST_DEC_TTL:
5247 if (compose_dec_ttl(ctx)) {
5252 case OFPUTIL_NXAST_EXIT:
5256 case OFPUTIL_NXAST_FIN_TIMEOUT:
5257 ctx->has_fin_timeout = true;
5258 xlate_fin_timeout(ctx, (const struct nx_action_fin_timeout *) ia);
5261 case OFPUTIL_NXAST_CONTROLLER:
5262 nac = (const struct nx_action_controller *) ia;
5263 execute_controller_action(ctx, ntohs(nac->max_len), nac->reason,
5264 ntohs(nac->controller_id));
5270 /* We've let OFPP_NORMAL and the learning action look at the packet,
5271 * so drop it now if forwarding is disabled. */
5272 if (port && !stp_forward_in_state(port->stp_state)) {
5273 ofpbuf_clear(ctx->odp_actions);
5274 add_sflow_action(ctx);
5277 ctx->rule->up.evictable = was_evictable;
5282 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
5283 struct ofproto_dpif *ofproto, const struct flow *flow,
5284 ovs_be16 initial_tci, struct rule_dpif *rule,
5285 uint8_t tcp_flags, const struct ofpbuf *packet)
5287 ctx->ofproto = ofproto;
5289 ctx->base_flow = ctx->flow;
5290 ctx->base_flow.tun_id = 0;
5291 ctx->base_flow.vlan_tci = initial_tci;
5293 ctx->packet = packet;
5294 ctx->may_learn = packet != NULL;
5295 ctx->tcp_flags = tcp_flags;
5296 ctx->resubmit_hook = NULL;
5297 ctx->resubmit_stats = NULL;
5300 /* Translates the 'n_in' "union ofp_action"s in 'in' into datapath actions in
5301 * 'odp_actions', using 'ctx'. */
5303 xlate_actions(struct action_xlate_ctx *ctx,
5304 const union ofp_action *in, size_t n_in,
5305 struct ofpbuf *odp_actions)
5307 /* Normally false. Set to true if we ever hit MAX_RESUBMIT_RECURSION, so
5308 * that in the future we always keep a copy of the original flow for
5309 * tracing purposes. */
5310 static bool hit_resubmit_limit;
5312 COVERAGE_INC(ofproto_dpif_xlate);
5314 ofpbuf_clear(odp_actions);
5315 ofpbuf_reserve(odp_actions, NL_A_U32_SIZE);
5317 ctx->odp_actions = odp_actions;
5319 ctx->may_set_up_flow = true;
5320 ctx->has_learn = false;
5321 ctx->has_normal = false;
5322 ctx->has_fin_timeout = false;
5323 ctx->nf_output_iface = NF_OUT_DROP;
5326 ctx->max_resubmit_trigger = false;
5327 ctx->orig_skb_priority = ctx->flow.skb_priority;
5331 if (ctx->ofproto->has_mirrors || hit_resubmit_limit) {
5332 /* Do this conditionally because the copy is expensive enough that it
5333 * shows up in profiles.
5335 * We keep orig_flow in 'ctx' only because I couldn't make GCC 4.4
5336 * believe that I wasn't using it without initializing it if I kept it
5337 * in a local variable. */
5338 ctx->orig_flow = ctx->flow;
5341 if (ctx->flow.nw_frag & FLOW_NW_FRAG_ANY) {
5342 switch (ctx->ofproto->up.frag_handling) {
5343 case OFPC_FRAG_NORMAL:
5344 /* We must pretend that transport ports are unavailable. */
5345 ctx->flow.tp_src = ctx->base_flow.tp_src = htons(0);
5346 ctx->flow.tp_dst = ctx->base_flow.tp_dst = htons(0);
5349 case OFPC_FRAG_DROP:
5352 case OFPC_FRAG_REASM:
5355 case OFPC_FRAG_NX_MATCH:
5356 /* Nothing to do. */
5359 case OFPC_INVALID_TTL_TO_CONTROLLER:
5364 if (process_special(ctx->ofproto, &ctx->flow, ctx->packet)) {
5365 ctx->may_set_up_flow = false;
5367 static struct vlog_rate_limit trace_rl = VLOG_RATE_LIMIT_INIT(1, 1);
5368 ovs_be16 initial_tci = ctx->base_flow.vlan_tci;
5370 add_sflow_action(ctx);
5371 do_xlate_actions(in, n_in, ctx);
5373 if (ctx->max_resubmit_trigger && !ctx->resubmit_hook) {
5374 if (!hit_resubmit_limit) {
5375 /* We didn't record the original flow. Make sure we do from
5377 hit_resubmit_limit = true;
5378 } else if (!VLOG_DROP_ERR(&trace_rl)) {
5379 struct ds ds = DS_EMPTY_INITIALIZER;
5381 ofproto_trace(ctx->ofproto, &ctx->orig_flow, ctx->packet,
5383 VLOG_ERR("Trace triggered by excessive resubmit "
5384 "recursion:\n%s", ds_cstr(&ds));
5389 if (!connmgr_may_set_up_flow(ctx->ofproto->up.connmgr, &ctx->flow,
5390 ctx->odp_actions->data,
5391 ctx->odp_actions->size)) {
5392 ctx->may_set_up_flow = false;
5394 && connmgr_msg_in_hook(ctx->ofproto->up.connmgr, &ctx->flow,
5396 compose_output_action(ctx, OFPP_LOCAL);
5399 if (ctx->ofproto->has_mirrors) {
5400 add_mirror_actions(ctx, &ctx->orig_flow);
5402 fix_sflow_action(ctx);
5406 /* Translates the 'n_in' "union ofp_action"s in 'in' into datapath actions,
5407 * using 'ctx', and discards the datapath actions. */
5409 xlate_actions_for_side_effects(struct action_xlate_ctx *ctx,
5410 const union ofp_action *in, size_t n_in)
5412 uint64_t odp_actions_stub[1024 / 8];
5413 struct ofpbuf odp_actions;
5415 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5416 xlate_actions(ctx, in, n_in, &odp_actions);
5417 ofpbuf_uninit(&odp_actions);
5420 /* OFPP_NORMAL implementation. */
5422 static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *);
5424 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
5425 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle',
5426 * the bundle on which the packet was received, returns the VLAN to which the
5429 * Both 'vid' and the return value are in the range 0...4095. */
5431 input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid)
5433 switch (in_bundle->vlan_mode) {
5434 case PORT_VLAN_ACCESS:
5435 return in_bundle->vlan;
5438 case PORT_VLAN_TRUNK:
5441 case PORT_VLAN_NATIVE_UNTAGGED:
5442 case PORT_VLAN_NATIVE_TAGGED:
5443 return vid ? vid : in_bundle->vlan;
5450 /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'.
5451 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
5454 * 'vid' should be the VID obtained from the 802.1Q header that was received as
5455 * part of a packet (specify 0 if there was no 802.1Q header), in the range
5458 input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn)
5460 /* Allow any VID on the OFPP_NONE port. */
5461 if (in_bundle == &ofpp_none_bundle) {
5465 switch (in_bundle->vlan_mode) {
5466 case PORT_VLAN_ACCESS:
5469 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5470 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
5471 "packet received on port %s configured as VLAN "
5472 "%"PRIu16" access port",
5473 in_bundle->ofproto->up.name, vid,
5474 in_bundle->name, in_bundle->vlan);
5480 case PORT_VLAN_NATIVE_UNTAGGED:
5481 case PORT_VLAN_NATIVE_TAGGED:
5483 /* Port must always carry its native VLAN. */
5487 case PORT_VLAN_TRUNK:
5488 if (!ofbundle_includes_vlan(in_bundle, vid)) {
5490 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5491 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet "
5492 "received on port %s not configured for trunking "
5494 in_bundle->ofproto->up.name, vid,
5495 in_bundle->name, vid);
5507 /* Given 'vlan', the VLAN that a packet belongs to, and
5508 * 'out_bundle', a bundle on which the packet is to be output, returns the VID
5509 * that should be included in the 802.1Q header. (If the return value is 0,
5510 * then the 802.1Q header should only be included in the packet if there is a
5513 * Both 'vlan' and the return value are in the range 0...4095. */
5515 output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan)
5517 switch (out_bundle->vlan_mode) {
5518 case PORT_VLAN_ACCESS:
5521 case PORT_VLAN_TRUNK:
5522 case PORT_VLAN_NATIVE_TAGGED:
5525 case PORT_VLAN_NATIVE_UNTAGGED:
5526 return vlan == out_bundle->vlan ? 0 : vlan;
5534 output_normal(struct action_xlate_ctx *ctx, const struct ofbundle *out_bundle,
5537 struct ofport_dpif *port;
5539 ovs_be16 tci, old_tci;
5541 vid = output_vlan_to_vid(out_bundle, vlan);
5542 if (!out_bundle->bond) {
5543 port = ofbundle_get_a_port(out_bundle);
5545 port = bond_choose_output_slave(out_bundle->bond, &ctx->flow,
5548 /* No slaves enabled, so drop packet. */
5553 old_tci = ctx->flow.vlan_tci;
5555 if (tci || out_bundle->use_priority_tags) {
5556 tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK);
5558 tci |= htons(VLAN_CFI);
5561 ctx->flow.vlan_tci = tci;
5563 compose_output_action(ctx, port->up.ofp_port);
5564 ctx->flow.vlan_tci = old_tci;
5568 mirror_mask_ffs(mirror_mask_t mask)
5570 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
5575 ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
5577 return (bundle->vlan_mode != PORT_VLAN_ACCESS
5578 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
5582 ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan)
5584 return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan);
5587 /* Returns an arbitrary interface within 'bundle'. */
5588 static struct ofport_dpif *
5589 ofbundle_get_a_port(const struct ofbundle *bundle)
5591 return CONTAINER_OF(list_front(&bundle->ports),
5592 struct ofport_dpif, bundle_node);
5596 vlan_is_mirrored(const struct ofmirror *m, int vlan)
5598 return !m->vlans || bitmap_is_set(m->vlans, vlan);
5601 /* Returns true if a packet with Ethernet destination MAC 'dst' may be mirrored
5602 * to a VLAN. In general most packets may be mirrored but we want to drop
5603 * protocols that may confuse switches. */
5605 eth_dst_may_rspan(const uint8_t dst[ETH_ADDR_LEN])
5607 /* If you change this function's behavior, please update corresponding
5608 * documentation in vswitch.xml at the same time. */
5609 if (dst[0] != 0x01) {
5610 /* All the currently banned MACs happen to start with 01 currently, so
5611 * this is a quick way to eliminate most of the good ones. */
5613 if (eth_addr_is_reserved(dst)) {
5614 /* Drop STP, IEEE pause frames, and other reserved protocols
5615 * (01-80-c2-00-00-0x). */
5619 if (dst[0] == 0x01 && dst[1] == 0x00 && dst[2] == 0x0c) {
5621 if ((dst[3] & 0xfe) == 0xcc &&
5622 (dst[4] & 0xfe) == 0xcc &&
5623 (dst[5] & 0xfe) == 0xcc) {
5624 /* Drop the following protocols plus others following the same
5627 CDP, VTP, DTP, PAgP (01-00-0c-cc-cc-cc)
5628 Spanning Tree PVSTP+ (01-00-0c-cc-cc-cd)
5629 STP Uplink Fast (01-00-0c-cd-cd-cd) */
5633 if (!(dst[3] | dst[4] | dst[5])) {
5634 /* Drop Inter Switch Link packets (01-00-0c-00-00-00). */
5643 add_mirror_actions(struct action_xlate_ctx *ctx, const struct flow *orig_flow)
5645 struct ofproto_dpif *ofproto = ctx->ofproto;
5646 mirror_mask_t mirrors;
5647 struct ofbundle *in_bundle;
5650 const struct nlattr *a;
5653 in_bundle = lookup_input_bundle(ctx->ofproto, orig_flow->in_port,
5654 ctx->packet != NULL, NULL);
5658 mirrors = in_bundle->src_mirrors;
5660 /* Drop frames on bundles reserved for mirroring. */
5661 if (in_bundle->mirror_out) {
5662 if (ctx->packet != NULL) {
5663 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5664 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
5665 "%s, which is reserved exclusively for mirroring",
5666 ctx->ofproto->up.name, in_bundle->name);
5672 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
5673 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
5676 vlan = input_vid_to_vlan(in_bundle, vid);
5678 /* Look at the output ports to check for destination selections. */
5680 NL_ATTR_FOR_EACH (a, left, ctx->odp_actions->data,
5681 ctx->odp_actions->size) {
5682 enum ovs_action_attr type = nl_attr_type(a);
5683 struct ofport_dpif *ofport;
5685 if (type != OVS_ACTION_ATTR_OUTPUT) {
5689 ofport = get_odp_port(ofproto, nl_attr_get_u32(a));
5690 if (ofport && ofport->bundle) {
5691 mirrors |= ofport->bundle->dst_mirrors;
5699 /* Restore the original packet before adding the mirror actions. */
5700 ctx->flow = *orig_flow;
5705 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
5707 if (!vlan_is_mirrored(m, vlan)) {
5708 mirrors &= mirrors - 1;
5712 mirrors &= ~m->dup_mirrors;
5713 ctx->mirrors |= m->dup_mirrors;
5715 output_normal(ctx, m->out, vlan);
5716 } else if (eth_dst_may_rspan(orig_flow->dl_dst)
5717 && vlan != m->out_vlan) {
5718 struct ofbundle *bundle;
5720 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
5721 if (ofbundle_includes_vlan(bundle, m->out_vlan)
5722 && !bundle->mirror_out) {
5723 output_normal(ctx, bundle, m->out_vlan);
5731 update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors,
5732 uint64_t packets, uint64_t bytes)
5738 for (; mirrors; mirrors &= mirrors - 1) {
5741 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
5744 /* In normal circumstances 'm' will not be NULL. However,
5745 * if mirrors are reconfigured, we can temporarily get out
5746 * of sync in facet_revalidate(). We could "correct" the
5747 * mirror list before reaching here, but doing that would
5748 * not properly account the traffic stats we've currently
5749 * accumulated for previous mirror configuration. */
5753 m->packet_count += packets;
5754 m->byte_count += bytes;
5758 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
5759 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
5760 * indicate this; newer upstream kernels use gratuitous ARP requests. */
5762 is_gratuitous_arp(const struct flow *flow)
5764 return (flow->dl_type == htons(ETH_TYPE_ARP)
5765 && eth_addr_is_broadcast(flow->dl_dst)
5766 && (flow->nw_proto == ARP_OP_REPLY
5767 || (flow->nw_proto == ARP_OP_REQUEST
5768 && flow->nw_src == flow->nw_dst)));
5772 update_learning_table(struct ofproto_dpif *ofproto,
5773 const struct flow *flow, int vlan,
5774 struct ofbundle *in_bundle)
5776 struct mac_entry *mac;
5778 /* Don't learn the OFPP_NONE port. */
5779 if (in_bundle == &ofpp_none_bundle) {
5783 if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
5787 mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan);
5788 if (is_gratuitous_arp(flow)) {
5789 /* We don't want to learn from gratuitous ARP packets that are
5790 * reflected back over bond slaves so we lock the learning table. */
5791 if (!in_bundle->bond) {
5792 mac_entry_set_grat_arp_lock(mac);
5793 } else if (mac_entry_is_grat_arp_locked(mac)) {
5798 if (mac_entry_is_new(mac) || mac->port.p != in_bundle) {
5799 /* The log messages here could actually be useful in debugging,
5800 * so keep the rate limit relatively high. */
5801 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
5802 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
5803 "on port %s in VLAN %d",
5804 ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src),
5805 in_bundle->name, vlan);
5807 mac->port.p = in_bundle;
5808 tag_set_add(&ofproto->revalidate_set,
5809 mac_learning_changed(ofproto->ml, mac));
5813 static struct ofbundle *
5814 lookup_input_bundle(struct ofproto_dpif *ofproto, uint16_t in_port, bool warn,
5815 struct ofport_dpif **in_ofportp)
5817 struct ofport_dpif *ofport;
5819 /* Find the port and bundle for the received packet. */
5820 ofport = get_ofp_port(ofproto, in_port);
5822 *in_ofportp = ofport;
5824 if (ofport && ofport->bundle) {
5825 return ofport->bundle;
5828 /* Special-case OFPP_NONE, which a controller may use as the ingress
5829 * port for traffic that it is sourcing. */
5830 if (in_port == OFPP_NONE) {
5831 return &ofpp_none_bundle;
5834 /* Odd. A few possible reasons here:
5836 * - We deleted a port but there are still a few packets queued up
5839 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
5840 * we don't know about.
5842 * - The ofproto client didn't configure the port as part of a bundle.
5845 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5847 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
5848 "port %"PRIu16, ofproto->up.name, in_port);
5853 /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or
5854 * dropped. Returns true if they may be forwarded, false if they should be
5857 * 'in_port' must be the ofport_dpif that corresponds to flow->in_port.
5858 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
5860 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
5861 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
5862 * checked by input_vid_is_valid().
5864 * May also add tags to '*tags', although the current implementation only does
5865 * so in one special case.
5868 is_admissible(struct ofproto_dpif *ofproto, const struct flow *flow,
5869 struct ofport_dpif *in_port, uint16_t vlan, tag_type *tags)
5871 struct ofbundle *in_bundle = in_port->bundle;
5873 /* Drop frames for reserved multicast addresses
5874 * only if forward_bpdu option is absent. */
5875 if (eth_addr_is_reserved(flow->dl_dst) && !ofproto->up.forward_bpdu) {
5879 if (in_bundle->bond) {
5880 struct mac_entry *mac;
5882 switch (bond_check_admissibility(in_bundle->bond, in_port,
5883 flow->dl_dst, tags)) {
5890 case BV_DROP_IF_MOVED:
5891 mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL);
5892 if (mac && mac->port.p != in_bundle &&
5893 (!is_gratuitous_arp(flow)
5894 || mac_entry_is_grat_arp_locked(mac))) {
5905 xlate_normal(struct action_xlate_ctx *ctx)
5907 struct ofport_dpif *in_port;
5908 struct ofbundle *in_bundle;
5909 struct mac_entry *mac;
5913 ctx->has_normal = true;
5915 in_bundle = lookup_input_bundle(ctx->ofproto, ctx->flow.in_port,
5916 ctx->packet != NULL, &in_port);
5921 /* Drop malformed frames. */
5922 if (ctx->flow.dl_type == htons(ETH_TYPE_VLAN) &&
5923 !(ctx->flow.vlan_tci & htons(VLAN_CFI))) {
5924 if (ctx->packet != NULL) {
5925 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5926 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
5927 "VLAN tag received on port %s",
5928 ctx->ofproto->up.name, in_bundle->name);
5933 /* Drop frames on bundles reserved for mirroring. */
5934 if (in_bundle->mirror_out) {
5935 if (ctx->packet != NULL) {
5936 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5937 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
5938 "%s, which is reserved exclusively for mirroring",
5939 ctx->ofproto->up.name, in_bundle->name);
5945 vid = vlan_tci_to_vid(ctx->flow.vlan_tci);
5946 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
5949 vlan = input_vid_to_vlan(in_bundle, vid);
5951 /* Check other admissibility requirements. */
5953 !is_admissible(ctx->ofproto, &ctx->flow, in_port, vlan, &ctx->tags)) {
5957 /* Learn source MAC. */
5958 if (ctx->may_learn) {
5959 update_learning_table(ctx->ofproto, &ctx->flow, vlan, in_bundle);
5962 /* Determine output bundle. */
5963 mac = mac_learning_lookup(ctx->ofproto->ml, ctx->flow.dl_dst, vlan,
5966 if (mac->port.p != in_bundle) {
5967 output_normal(ctx, mac->port.p, vlan);
5970 struct ofbundle *bundle;
5972 HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
5973 if (bundle != in_bundle
5974 && ofbundle_includes_vlan(bundle, vlan)
5975 && bundle->floodable
5976 && !bundle->mirror_out) {
5977 output_normal(ctx, bundle, vlan);
5980 ctx->nf_output_iface = NF_OUT_FLOOD;
5984 /* Optimized flow revalidation.
5986 * It's a difficult problem, in general, to tell which facets need to have
5987 * their actions recalculated whenever the OpenFlow flow table changes. We
5988 * don't try to solve that general problem: for most kinds of OpenFlow flow
5989 * table changes, we recalculate the actions for every facet. This is
5990 * relatively expensive, but it's good enough if the OpenFlow flow table
5991 * doesn't change very often.
5993 * However, we can expect one particular kind of OpenFlow flow table change to
5994 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
5995 * of CPU on revalidating every facet whenever MAC learning modifies the flow
5996 * table, we add a special case that applies to flow tables in which every rule
5997 * has the same form (that is, the same wildcards), except that the table is
5998 * also allowed to have a single "catch-all" flow that matches all packets. We
5999 * optimize this case by tagging all of the facets that resubmit into the table
6000 * and invalidating the same tag whenever a flow changes in that table. The
6001 * end result is that we revalidate just the facets that need it (and sometimes
6002 * a few more, but not all of the facets or even all of the facets that
6003 * resubmit to the table modified by MAC learning). */
6005 /* Calculates the tag to use for 'flow' and wildcards 'wc' when it is inserted
6006 * into an OpenFlow table with the given 'basis'. */
6008 rule_calculate_tag(const struct flow *flow, const struct flow_wildcards *wc,
6011 if (flow_wildcards_is_catchall(wc)) {
6014 struct flow tag_flow = *flow;
6015 flow_zero_wildcards(&tag_flow, wc);
6016 return tag_create_deterministic(flow_hash(&tag_flow, secret));
6020 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
6021 * taggability of that table.
6023 * This function must be called after *each* change to a flow table. If you
6024 * skip calling it on some changes then the pointer comparisons at the end can
6025 * be invalid if you get unlucky. For example, if a flow removal causes a
6026 * cls_table to be destroyed and then a flow insertion causes a cls_table with
6027 * different wildcards to be created with the same address, then this function
6028 * will incorrectly skip revalidation. */
6030 table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
6032 struct table_dpif *table = &ofproto->tables[table_id];
6033 const struct oftable *oftable = &ofproto->up.tables[table_id];
6034 struct cls_table *catchall, *other;
6035 struct cls_table *t;
6037 catchall = other = NULL;
6039 switch (hmap_count(&oftable->cls.tables)) {
6041 /* We could tag this OpenFlow table but it would make the logic a
6042 * little harder and it's a corner case that doesn't seem worth it
6048 HMAP_FOR_EACH (t, hmap_node, &oftable->cls.tables) {
6049 if (cls_table_is_catchall(t)) {
6051 } else if (!other) {
6054 /* Indicate that we can't tag this by setting both tables to
6055 * NULL. (We know that 'catchall' is already NULL.) */
6062 /* Can't tag this table. */
6066 if (table->catchall_table != catchall || table->other_table != other) {
6067 table->catchall_table = catchall;
6068 table->other_table = other;
6069 ofproto->need_revalidate = true;
6073 /* Given 'rule' that has changed in some way (either it is a rule being
6074 * inserted, a rule being deleted, or a rule whose actions are being
6075 * modified), marks facets for revalidation to ensure that packets will be
6076 * forwarded correctly according to the new state of the flow table.
6078 * This function must be called after *each* change to a flow table. See
6079 * the comment on table_update_taggable() for more information. */
6081 rule_invalidate(const struct rule_dpif *rule)
6083 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
6085 table_update_taggable(ofproto, rule->up.table_id);
6087 if (!ofproto->need_revalidate) {
6088 struct table_dpif *table = &ofproto->tables[rule->up.table_id];
6090 if (table->other_table && rule->tag) {
6091 tag_set_add(&ofproto->revalidate_set, rule->tag);
6093 ofproto->need_revalidate = true;
6099 set_frag_handling(struct ofproto *ofproto_,
6100 enum ofp_config_flags frag_handling)
6102 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
6104 if (frag_handling != OFPC_FRAG_REASM) {
6105 ofproto->need_revalidate = true;
6113 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
6114 const struct flow *flow,
6115 const union ofp_action *ofp_actions, size_t n_ofp_actions)
6117 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
6120 if (flow->in_port >= ofproto->max_ports && flow->in_port < OFPP_MAX) {
6121 return OFPERR_NXBRC_BAD_IN_PORT;
6124 error = validate_actions(ofp_actions, n_ofp_actions, flow,
6125 ofproto->max_ports);
6127 struct odputil_keybuf keybuf;
6128 struct dpif_flow_stats stats;
6132 struct action_xlate_ctx ctx;
6133 uint64_t odp_actions_stub[1024 / 8];
6134 struct ofpbuf odp_actions;
6136 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
6137 odp_flow_key_from_flow(&key, flow);
6139 dpif_flow_stats_extract(flow, packet, &stats);
6141 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci, NULL,
6142 packet_get_tcp_flags(packet, flow), packet);
6143 ctx.resubmit_stats = &stats;
6145 ofpbuf_use_stub(&odp_actions,
6146 odp_actions_stub, sizeof odp_actions_stub);
6147 xlate_actions(&ctx, ofp_actions, n_ofp_actions, &odp_actions);
6148 dpif_execute(ofproto->dpif, key.data, key.size,
6149 odp_actions.data, odp_actions.size, packet);
6150 ofpbuf_uninit(&odp_actions);
6158 set_netflow(struct ofproto *ofproto_,
6159 const struct netflow_options *netflow_options)
6161 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
6163 if (netflow_options) {
6164 if (!ofproto->netflow) {
6165 ofproto->netflow = netflow_create();
6167 return netflow_set_options(ofproto->netflow, netflow_options);
6169 netflow_destroy(ofproto->netflow);
6170 ofproto->netflow = NULL;
6176 get_netflow_ids(const struct ofproto *ofproto_,
6177 uint8_t *engine_type, uint8_t *engine_id)
6179 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
6181 dpif_get_netflow_ids(ofproto->dpif, engine_type, engine_id);
6185 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
6187 if (!facet_is_controller_flow(facet) &&
6188 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
6189 struct subfacet *subfacet;
6190 struct ofexpired expired;
6192 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
6193 if (subfacet->installed) {
6194 struct dpif_flow_stats stats;
6196 subfacet_install(subfacet, subfacet->actions,
6197 subfacet->actions_len, &stats);
6198 subfacet_update_stats(subfacet, &stats);
6202 expired.flow = facet->flow;
6203 expired.packet_count = facet->packet_count;
6204 expired.byte_count = facet->byte_count;
6205 expired.used = facet->used;
6206 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
6211 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
6213 struct facet *facet;
6215 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
6216 send_active_timeout(ofproto, facet);
6220 static struct ofproto_dpif *
6221 ofproto_dpif_lookup(const char *name)
6223 struct ofproto_dpif *ofproto;
6225 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
6226 hash_string(name, 0), &all_ofproto_dpifs) {
6227 if (!strcmp(ofproto->up.name, name)) {
6235 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
6236 const char *argv[], void *aux OVS_UNUSED)
6238 struct ofproto_dpif *ofproto;
6241 ofproto = ofproto_dpif_lookup(argv[1]);
6243 unixctl_command_reply_error(conn, "no such bridge");
6246 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
6248 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
6249 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
6253 unixctl_command_reply(conn, "table successfully flushed");
6257 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
6258 const char *argv[], void *aux OVS_UNUSED)
6260 struct ds ds = DS_EMPTY_INITIALIZER;
6261 const struct ofproto_dpif *ofproto;
6262 const struct mac_entry *e;
6264 ofproto = ofproto_dpif_lookup(argv[1]);
6266 unixctl_command_reply_error(conn, "no such bridge");
6270 ds_put_cstr(&ds, " port VLAN MAC Age\n");
6271 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
6272 struct ofbundle *bundle = e->port.p;
6273 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
6274 ofbundle_get_a_port(bundle)->odp_port,
6275 e->vlan, ETH_ADDR_ARGS(e->mac),
6276 mac_entry_age(ofproto->ml, e));
6278 unixctl_command_reply(conn, ds_cstr(&ds));
6283 struct action_xlate_ctx ctx;
6289 trace_format_rule(struct ds *result, uint8_t table_id, int level,
6290 const struct rule_dpif *rule)
6292 ds_put_char_multiple(result, '\t', level);
6294 ds_put_cstr(result, "No match\n");
6298 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
6299 table_id, ntohll(rule->up.flow_cookie));
6300 cls_rule_format(&rule->up.cr, result);
6301 ds_put_char(result, '\n');
6303 ds_put_char_multiple(result, '\t', level);
6304 ds_put_cstr(result, "OpenFlow ");
6305 ofp_print_actions(result, rule->up.actions, rule->up.n_actions);
6306 ds_put_char(result, '\n');
6310 trace_format_flow(struct ds *result, int level, const char *title,
6311 struct trace_ctx *trace)
6313 ds_put_char_multiple(result, '\t', level);
6314 ds_put_format(result, "%s: ", title);
6315 if (flow_equal(&trace->ctx.flow, &trace->flow)) {
6316 ds_put_cstr(result, "unchanged");
6318 flow_format(result, &trace->ctx.flow);
6319 trace->flow = trace->ctx.flow;
6321 ds_put_char(result, '\n');
6325 trace_format_regs(struct ds *result, int level, const char *title,
6326 struct trace_ctx *trace)
6330 ds_put_char_multiple(result, '\t', level);
6331 ds_put_format(result, "%s:", title);
6332 for (i = 0; i < FLOW_N_REGS; i++) {
6333 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
6335 ds_put_char(result, '\n');
6339 trace_format_odp(struct ds *result, int level, const char *title,
6340 struct trace_ctx *trace)
6342 struct ofpbuf *odp_actions = trace->ctx.odp_actions;
6344 ds_put_char_multiple(result, '\t', level);
6345 ds_put_format(result, "%s: ", title);
6346 format_odp_actions(result, odp_actions->data, odp_actions->size);
6347 ds_put_char(result, '\n');
6351 trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
6353 struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx);
6354 struct ds *result = trace->result;
6356 ds_put_char(result, '\n');
6357 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
6358 trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace);
6359 trace_format_odp(result, ctx->recurse + 1, "Resubmitted odp", trace);
6360 trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule);
6364 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
6365 void *aux OVS_UNUSED)
6367 const char *dpname = argv[1];
6368 struct ofproto_dpif *ofproto;
6369 struct ofpbuf odp_key;
6370 struct ofpbuf *packet;
6371 ovs_be16 initial_tci;
6377 ofpbuf_init(&odp_key, 0);
6380 ofproto = ofproto_dpif_lookup(dpname);
6382 unixctl_command_reply_error(conn, "Unknown ofproto (use ofproto/list "
6386 if (argc == 3 || (argc == 4 && !strcmp(argv[3], "-generate"))) {
6387 /* ofproto/trace dpname flow [-generate] */
6388 const char *flow_s = argv[2];
6389 const char *generate_s = argv[3];
6392 /* Convert string to datapath key. */
6393 ofpbuf_init(&odp_key, 0);
6394 error = odp_flow_key_from_string(flow_s, NULL, &odp_key);
6396 unixctl_command_reply_error(conn, "Bad flow syntax");
6400 /* Convert odp_key to flow. */
6401 error = ofproto_dpif_extract_flow_key(ofproto, odp_key.data,
6402 odp_key.size, &flow,
6403 &initial_tci, NULL);
6404 if (error == ODP_FIT_ERROR) {
6405 unixctl_command_reply_error(conn, "Invalid flow");
6409 /* Generate a packet, if requested. */
6411 packet = ofpbuf_new(0);
6412 flow_compose(packet, &flow);
6414 } else if (argc == 6) {
6415 /* ofproto/trace dpname priority tun_id in_port packet */
6416 const char *priority_s = argv[2];
6417 const char *tun_id_s = argv[3];
6418 const char *in_port_s = argv[4];
6419 const char *packet_s = argv[5];
6420 uint16_t in_port = ofp_port_to_odp_port(atoi(in_port_s));
6421 ovs_be64 tun_id = htonll(strtoull(tun_id_s, NULL, 0));
6422 uint32_t priority = atoi(priority_s);
6425 msg = eth_from_hex(packet_s, &packet);
6427 unixctl_command_reply_error(conn, msg);
6431 ds_put_cstr(&result, "Packet: ");
6432 s = ofp_packet_to_string(packet->data, packet->size);
6433 ds_put_cstr(&result, s);
6436 flow_extract(packet, priority, tun_id, in_port, &flow);
6437 initial_tci = flow.vlan_tci;
6439 unixctl_command_reply_error(conn, "Bad command syntax");
6443 ofproto_trace(ofproto, &flow, packet, initial_tci, &result);
6444 unixctl_command_reply(conn, ds_cstr(&result));
6447 ds_destroy(&result);
6448 ofpbuf_delete(packet);
6449 ofpbuf_uninit(&odp_key);
6453 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
6454 const struct ofpbuf *packet, ovs_be16 initial_tci,
6457 struct rule_dpif *rule;
6459 ds_put_cstr(ds, "Flow: ");
6460 flow_format(ds, flow);
6461 ds_put_char(ds, '\n');
6463 rule = rule_dpif_lookup(ofproto, flow, 0);
6464 trace_format_rule(ds, 0, 0, rule);
6466 uint64_t odp_actions_stub[1024 / 8];
6467 struct ofpbuf odp_actions;
6469 struct trace_ctx trace;
6472 tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
6475 ofpbuf_use_stub(&odp_actions,
6476 odp_actions_stub, sizeof odp_actions_stub);
6477 action_xlate_ctx_init(&trace.ctx, ofproto, flow, initial_tci,
6478 rule, tcp_flags, packet);
6479 trace.ctx.resubmit_hook = trace_resubmit;
6480 xlate_actions(&trace.ctx, rule->up.actions, rule->up.n_actions,
6483 ds_put_char(ds, '\n');
6484 trace_format_flow(ds, 0, "Final flow", &trace);
6485 ds_put_cstr(ds, "Datapath actions: ");
6486 format_odp_actions(ds, odp_actions.data, odp_actions.size);
6487 ofpbuf_uninit(&odp_actions);
6489 if (!trace.ctx.may_set_up_flow) {
6491 ds_put_cstr(ds, "\nThis flow is not cachable.");
6493 ds_put_cstr(ds, "\nThe datapath actions are incomplete--"
6494 "for complete actions, please supply a packet.");
6501 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
6502 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
6505 unixctl_command_reply(conn, NULL);
6509 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
6510 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
6513 unixctl_command_reply(conn, NULL);
6516 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
6517 * 'reply' describing the results. */
6519 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
6521 struct facet *facet;
6525 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
6526 if (!facet_check_consistency(facet)) {
6531 ofproto->need_revalidate = true;
6535 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
6536 ofproto->up.name, errors);
6538 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
6543 ofproto_dpif_self_check(struct unixctl_conn *conn,
6544 int argc, const char *argv[], void *aux OVS_UNUSED)
6546 struct ds reply = DS_EMPTY_INITIALIZER;
6547 struct ofproto_dpif *ofproto;
6550 ofproto = ofproto_dpif_lookup(argv[1]);
6552 unixctl_command_reply_error(conn, "Unknown ofproto (use "
6553 "ofproto/list for help)");
6556 ofproto_dpif_self_check__(ofproto, &reply);
6558 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
6559 ofproto_dpif_self_check__(ofproto, &reply);
6563 unixctl_command_reply(conn, ds_cstr(&reply));
6568 ofproto_dpif_unixctl_init(void)
6570 static bool registered;
6576 unixctl_command_register(
6578 "bridge {tun_id in_port packet | odp_flow [-generate]}",
6579 2, 5, ofproto_unixctl_trace, NULL);
6580 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
6581 ofproto_unixctl_fdb_flush, NULL);
6582 unixctl_command_register("fdb/show", "bridge", 1, 1,
6583 ofproto_unixctl_fdb_show, NULL);
6584 unixctl_command_register("ofproto/clog", "", 0, 0,
6585 ofproto_dpif_clog, NULL);
6586 unixctl_command_register("ofproto/unclog", "", 0, 0,
6587 ofproto_dpif_unclog, NULL);
6588 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
6589 ofproto_dpif_self_check, NULL);
6592 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
6594 * This is deprecated. It is only for compatibility with broken device drivers
6595 * in old versions of Linux that do not properly support VLANs when VLAN
6596 * devices are not used. When broken device drivers are no longer in
6597 * widespread use, we will delete these interfaces. */
6600 set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid)
6602 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
6603 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
6605 if (realdev_ofp_port == ofport->realdev_ofp_port
6606 && vid == ofport->vlandev_vid) {
6610 ofproto->need_revalidate = true;
6612 if (ofport->realdev_ofp_port) {
6615 if (realdev_ofp_port && ofport->bundle) {
6616 /* vlandevs are enslaved to their realdevs, so they are not allowed to
6617 * themselves be part of a bundle. */
6618 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
6621 ofport->realdev_ofp_port = realdev_ofp_port;
6622 ofport->vlandev_vid = vid;
6624 if (realdev_ofp_port) {
6625 vsp_add(ofport, realdev_ofp_port, vid);
6632 hash_realdev_vid(uint16_t realdev_ofp_port, int vid)
6634 return hash_2words(realdev_ofp_port, vid);
6637 /* Returns the ODP port number of the Linux VLAN device that corresponds to
6638 * 'vlan_tci' on the network device with port number 'realdev_odp_port' in
6639 * 'ofproto'. For example, given 'realdev_odp_port' of eth0 and 'vlan_tci' 9,
6640 * it would return the port number of eth0.9.
6642 * Unless VLAN splinters are enabled for port 'realdev_odp_port', this
6643 * function just returns its 'realdev_odp_port' argument. */
6645 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
6646 uint32_t realdev_odp_port, ovs_be16 vlan_tci)
6648 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
6649 uint16_t realdev_ofp_port = odp_port_to_ofp_port(realdev_odp_port);
6650 int vid = vlan_tci_to_vid(vlan_tci);
6651 const struct vlan_splinter *vsp;
6653 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
6654 hash_realdev_vid(realdev_ofp_port, vid),
6655 &ofproto->realdev_vid_map) {
6656 if (vsp->realdev_ofp_port == realdev_ofp_port
6657 && vsp->vid == vid) {
6658 return ofp_port_to_odp_port(vsp->vlandev_ofp_port);
6662 return realdev_odp_port;
6665 static struct vlan_splinter *
6666 vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port)
6668 struct vlan_splinter *vsp;
6670 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, hash_int(vlandev_ofp_port, 0),
6671 &ofproto->vlandev_map) {
6672 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
6680 /* Returns the OpenFlow port number of the "real" device underlying the Linux
6681 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
6682 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
6683 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
6684 * eth0 and store 9 in '*vid'.
6686 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
6687 * VLAN device. Unless VLAN splinters are enabled, this is what this function
6690 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
6691 uint16_t vlandev_ofp_port, int *vid)
6693 if (!hmap_is_empty(&ofproto->vlandev_map)) {
6694 const struct vlan_splinter *vsp;
6696 vsp = vlandev_find(ofproto, vlandev_ofp_port);
6701 return vsp->realdev_ofp_port;
6707 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
6708 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
6709 * 'flow->in_port' to the "real" device backing the VLAN device, sets
6710 * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is
6711 * always the case unless VLAN splinters are enabled), returns false without
6712 * making any changes. */
6714 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
6719 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port, &vid);
6724 /* Cause the flow to be processed as if it came in on the real device with
6725 * the VLAN device's VLAN ID. */
6726 flow->in_port = realdev;
6727 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
6732 vsp_remove(struct ofport_dpif *port)
6734 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6735 struct vlan_splinter *vsp;
6737 vsp = vlandev_find(ofproto, port->up.ofp_port);
6739 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
6740 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
6743 port->realdev_ofp_port = 0;
6745 VLOG_ERR("missing vlan device record");
6750 vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid)
6752 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6754 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
6755 && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid))
6756 == realdev_ofp_port)) {
6757 struct vlan_splinter *vsp;
6759 vsp = xmalloc(sizeof *vsp);
6760 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
6761 hash_int(port->up.ofp_port, 0));
6762 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
6763 hash_realdev_vid(realdev_ofp_port, vid));
6764 vsp->realdev_ofp_port = realdev_ofp_port;
6765 vsp->vlandev_ofp_port = port->up.ofp_port;
6768 port->realdev_ofp_port = realdev_ofp_port;
6770 VLOG_ERR("duplicate vlan device record");
6774 const struct ofproto_class ofproto_dpif_class = {
6803 port_is_lacp_current,
6804 NULL, /* rule_choose_table */
6811 rule_modify_actions,
6819 get_cfm_remote_mpids,
6824 get_stp_port_status,
6831 is_mirror_output_bundle,
6832 forward_bpdu_changed,