2 * Copyright (c) 2009, 2010, 2011, 2012 Nicira Networks.
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 void vsp_remove(struct ofport_dpif *);
499 static void vsp_add(struct ofport_dpif *, uint16_t realdev_ofp_port, int vid);
501 static struct ofport_dpif *
502 ofport_dpif_cast(const struct ofport *ofport)
504 assert(ofport->ofproto->ofproto_class == &ofproto_dpif_class);
505 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
508 static void port_run(struct ofport_dpif *);
509 static void port_wait(struct ofport_dpif *);
510 static int set_cfm(struct ofport *, const struct cfm_settings *);
511 static void ofport_clear_priorities(struct ofport_dpif *);
513 struct dpif_completion {
514 struct list list_node;
515 struct ofoperation *op;
518 /* Extra information about a classifier table.
519 * Currently used just for optimized flow revalidation. */
521 /* If either of these is nonnull, then this table has a form that allows
522 * flows to be tagged to avoid revalidating most flows for the most common
523 * kinds of flow table changes. */
524 struct cls_table *catchall_table; /* Table that wildcards all fields. */
525 struct cls_table *other_table; /* Table with any other wildcard set. */
526 uint32_t basis; /* Keeps each table's tags separate. */
529 struct ofproto_dpif {
530 struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
539 struct netflow *netflow;
540 struct dpif_sflow *sflow;
541 struct hmap bundles; /* Contains "struct ofbundle"s. */
542 struct mac_learning *ml;
543 struct ofmirror *mirrors[MAX_MIRRORS];
545 bool has_bonded_bundles;
548 struct timer next_expiration;
552 struct hmap subfacets;
553 struct governor *governor;
556 struct table_dpif tables[N_TABLES];
557 bool need_revalidate;
558 struct tag_set revalidate_set;
560 /* Support for debugging async flow mods. */
561 struct list completions;
563 bool has_bundle_action; /* True when the first bundle action appears. */
564 struct netdev_stats stats; /* To account packets generated and consumed in
569 long long int stp_last_tick;
571 /* VLAN splinters. */
572 struct hmap realdev_vid_map; /* (realdev,vid) -> vlandev. */
573 struct hmap vlandev_map; /* vlandev -> (realdev,vid). */
576 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
577 * for debugging the asynchronous flow_mod implementation.) */
580 /* All existing ofproto_dpif instances, indexed by ->up.name. */
581 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
583 static void ofproto_dpif_unixctl_init(void);
585 static struct ofproto_dpif *
586 ofproto_dpif_cast(const struct ofproto *ofproto)
588 assert(ofproto->ofproto_class == &ofproto_dpif_class);
589 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
592 static struct ofport_dpif *get_ofp_port(struct ofproto_dpif *,
594 static struct ofport_dpif *get_odp_port(struct ofproto_dpif *,
596 static void ofproto_trace(struct ofproto_dpif *, const struct flow *,
597 const struct ofpbuf *, ovs_be16 initial_tci,
600 /* Packet processing. */
601 static void update_learning_table(struct ofproto_dpif *,
602 const struct flow *, int vlan,
605 #define FLOW_MISS_MAX_BATCH 50
606 static int handle_upcalls(struct ofproto_dpif *, unsigned int max_batch);
608 /* Flow expiration. */
609 static int expire(struct ofproto_dpif *);
612 static void send_netflow_active_timeouts(struct ofproto_dpif *);
615 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
617 compose_sflow_action(const struct ofproto_dpif *, struct ofpbuf *odp_actions,
618 const struct flow *, uint32_t odp_port);
619 static void add_mirror_actions(struct action_xlate_ctx *ctx,
620 const struct flow *flow);
621 /* Global variables. */
622 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
624 /* Factory functions. */
627 enumerate_types(struct sset *types)
629 dp_enumerate_types(types);
633 enumerate_names(const char *type, struct sset *names)
635 return dp_enumerate_names(type, names);
639 del(const char *type, const char *name)
644 error = dpif_open(name, type, &dpif);
646 error = dpif_delete(dpif);
652 /* Basic life-cycle. */
654 static struct ofproto *
657 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
662 dealloc(struct ofproto *ofproto_)
664 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
669 construct(struct ofproto *ofproto_)
671 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
672 const char *name = ofproto->up.name;
676 error = dpif_create_and_open(name, ofproto->up.type, &ofproto->dpif);
678 VLOG_ERR("failed to open datapath %s: %s", name, strerror(error));
682 ofproto->max_ports = dpif_get_max_ports(ofproto->dpif);
683 ofproto->n_matches = 0;
685 dpif_flow_flush(ofproto->dpif);
686 dpif_recv_purge(ofproto->dpif);
688 error = dpif_recv_set(ofproto->dpif, true);
690 VLOG_ERR("failed to listen on datapath %s: %s", name, strerror(error));
691 dpif_close(ofproto->dpif);
695 ofproto->netflow = NULL;
696 ofproto->sflow = NULL;
698 hmap_init(&ofproto->bundles);
699 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
700 for (i = 0; i < MAX_MIRRORS; i++) {
701 ofproto->mirrors[i] = NULL;
703 ofproto->has_bonded_bundles = false;
705 timer_set_duration(&ofproto->next_expiration, 1000);
707 hmap_init(&ofproto->facets);
708 hmap_init(&ofproto->subfacets);
709 ofproto->governor = NULL;
711 for (i = 0; i < N_TABLES; i++) {
712 struct table_dpif *table = &ofproto->tables[i];
714 table->catchall_table = NULL;
715 table->other_table = NULL;
716 table->basis = random_uint32();
718 ofproto->need_revalidate = false;
719 tag_set_init(&ofproto->revalidate_set);
721 list_init(&ofproto->completions);
723 ofproto_dpif_unixctl_init();
725 ofproto->has_mirrors = false;
726 ofproto->has_bundle_action = false;
728 hmap_init(&ofproto->vlandev_map);
729 hmap_init(&ofproto->realdev_vid_map);
731 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
732 hash_string(ofproto->up.name, 0));
733 memset(&ofproto->stats, 0, sizeof ofproto->stats);
735 ofproto_init_tables(ofproto_, N_TABLES);
741 complete_operations(struct ofproto_dpif *ofproto)
743 struct dpif_completion *c, *next;
745 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
746 ofoperation_complete(c->op, 0);
747 list_remove(&c->list_node);
753 destruct(struct ofproto *ofproto_)
755 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
756 struct rule_dpif *rule, *next_rule;
757 struct oftable *table;
760 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
761 complete_operations(ofproto);
763 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
764 struct cls_cursor cursor;
766 cls_cursor_init(&cursor, &table->cls, NULL);
767 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
768 ofproto_rule_destroy(&rule->up);
772 for (i = 0; i < MAX_MIRRORS; i++) {
773 mirror_destroy(ofproto->mirrors[i]);
776 netflow_destroy(ofproto->netflow);
777 dpif_sflow_destroy(ofproto->sflow);
778 hmap_destroy(&ofproto->bundles);
779 mac_learning_destroy(ofproto->ml);
781 hmap_destroy(&ofproto->facets);
782 hmap_destroy(&ofproto->subfacets);
783 governor_destroy(ofproto->governor);
785 hmap_destroy(&ofproto->vlandev_map);
786 hmap_destroy(&ofproto->realdev_vid_map);
788 dpif_close(ofproto->dpif);
792 run_fast(struct ofproto *ofproto_)
794 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
797 /* Handle one or more batches of upcalls, until there's nothing left to do
798 * or until we do a fixed total amount of work.
800 * We do work in batches because it can be much cheaper to set up a number
801 * of flows and fire off their patches all at once. We do multiple batches
802 * because in some cases handling a packet can cause another packet to be
803 * queued almost immediately as part of the return flow. Both
804 * optimizations can make major improvements on some benchmarks and
805 * presumably for real traffic as well. */
807 while (work < FLOW_MISS_MAX_BATCH) {
808 int retval = handle_upcalls(ofproto, FLOW_MISS_MAX_BATCH - work);
818 run(struct ofproto *ofproto_)
820 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
821 struct ofport_dpif *ofport;
822 struct ofbundle *bundle;
826 complete_operations(ofproto);
828 dpif_run(ofproto->dpif);
830 error = run_fast(ofproto_);
835 if (timer_expired(&ofproto->next_expiration)) {
836 int delay = expire(ofproto);
837 timer_set_duration(&ofproto->next_expiration, delay);
840 if (ofproto->netflow) {
841 if (netflow_run(ofproto->netflow)) {
842 send_netflow_active_timeouts(ofproto);
845 if (ofproto->sflow) {
846 dpif_sflow_run(ofproto->sflow);
849 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
852 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
857 mac_learning_run(ofproto->ml, &ofproto->revalidate_set);
859 /* Now revalidate if there's anything to do. */
860 if (ofproto->need_revalidate
861 || !tag_set_is_empty(&ofproto->revalidate_set)) {
862 struct tag_set revalidate_set = ofproto->revalidate_set;
863 bool revalidate_all = ofproto->need_revalidate;
864 struct facet *facet, *next;
866 /* Clear the revalidation flags. */
867 tag_set_init(&ofproto->revalidate_set);
868 ofproto->need_revalidate = false;
870 HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &ofproto->facets) {
872 || tag_set_intersects(&revalidate_set, facet->tags)) {
873 facet_revalidate(facet);
878 /* Check the consistency of a random facet, to aid debugging. */
879 if (!hmap_is_empty(&ofproto->facets) && !ofproto->need_revalidate) {
882 facet = CONTAINER_OF(hmap_random_node(&ofproto->facets),
883 struct facet, hmap_node);
884 if (!tag_set_intersects(&ofproto->revalidate_set, facet->tags)) {
885 if (!facet_check_consistency(facet)) {
886 ofproto->need_revalidate = true;
891 if (ofproto->governor) {
894 governor_run(ofproto->governor);
896 /* If the governor has shrunk to its minimum size and the number of
897 * subfacets has dwindled, then drop the governor entirely.
899 * For hysteresis, the number of subfacets to drop the governor is
900 * smaller than the number needed to trigger its creation. */
901 n_subfacets = hmap_count(&ofproto->subfacets);
902 if (n_subfacets * 4 < ofproto->up.flow_eviction_threshold
903 && governor_is_idle(ofproto->governor)) {
904 governor_destroy(ofproto->governor);
905 ofproto->governor = NULL;
913 wait(struct ofproto *ofproto_)
915 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
916 struct ofport_dpif *ofport;
917 struct ofbundle *bundle;
919 if (!clogged && !list_is_empty(&ofproto->completions)) {
920 poll_immediate_wake();
923 dpif_wait(ofproto->dpif);
924 dpif_recv_wait(ofproto->dpif);
925 if (ofproto->sflow) {
926 dpif_sflow_wait(ofproto->sflow);
928 if (!tag_set_is_empty(&ofproto->revalidate_set)) {
929 poll_immediate_wake();
931 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
934 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
937 if (ofproto->netflow) {
938 netflow_wait(ofproto->netflow);
940 mac_learning_wait(ofproto->ml);
942 if (ofproto->need_revalidate) {
943 /* Shouldn't happen, but if it does just go around again. */
944 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
945 poll_immediate_wake();
947 timer_wait(&ofproto->next_expiration);
949 if (ofproto->governor) {
950 governor_wait(ofproto->governor);
955 flush(struct ofproto *ofproto_)
957 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
958 struct facet *facet, *next_facet;
960 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
961 /* Mark the facet as not installed so that facet_remove() doesn't
962 * bother trying to uninstall it. There is no point in uninstalling it
963 * individually since we are about to blow away all the facets with
964 * dpif_flow_flush(). */
965 struct subfacet *subfacet;
967 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
968 subfacet->installed = false;
969 subfacet->dp_packet_count = 0;
970 subfacet->dp_byte_count = 0;
974 dpif_flow_flush(ofproto->dpif);
978 get_features(struct ofproto *ofproto_ OVS_UNUSED,
979 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
981 *arp_match_ip = true;
982 *actions = (OFPUTIL_A_OUTPUT |
983 OFPUTIL_A_SET_VLAN_VID |
984 OFPUTIL_A_SET_VLAN_PCP |
985 OFPUTIL_A_STRIP_VLAN |
986 OFPUTIL_A_SET_DL_SRC |
987 OFPUTIL_A_SET_DL_DST |
988 OFPUTIL_A_SET_NW_SRC |
989 OFPUTIL_A_SET_NW_DST |
990 OFPUTIL_A_SET_NW_TOS |
991 OFPUTIL_A_SET_TP_SRC |
992 OFPUTIL_A_SET_TP_DST |
997 get_tables(struct ofproto *ofproto_, struct ofp_table_stats *ots)
999 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1000 struct dpif_dp_stats s;
1002 strcpy(ots->name, "classifier");
1004 dpif_get_dp_stats(ofproto->dpif, &s);
1005 put_32aligned_be64(&ots->lookup_count, htonll(s.n_hit + s.n_missed));
1006 put_32aligned_be64(&ots->matched_count,
1007 htonll(s.n_hit + ofproto->n_matches));
1010 static struct ofport *
1013 struct ofport_dpif *port = xmalloc(sizeof *port);
1018 port_dealloc(struct ofport *port_)
1020 struct ofport_dpif *port = ofport_dpif_cast(port_);
1025 port_construct(struct ofport *port_)
1027 struct ofport_dpif *port = ofport_dpif_cast(port_);
1028 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1030 ofproto->need_revalidate = true;
1031 port->odp_port = ofp_port_to_odp_port(port->up.ofp_port);
1032 port->bundle = NULL;
1034 port->tag = tag_create_random();
1035 port->may_enable = true;
1036 port->stp_port = NULL;
1037 port->stp_state = STP_DISABLED;
1038 hmap_init(&port->priorities);
1039 port->realdev_ofp_port = 0;
1040 port->vlandev_vid = 0;
1041 port->carrier_seq = netdev_get_carrier_resets(port->up.netdev);
1043 if (ofproto->sflow) {
1044 dpif_sflow_add_port(ofproto->sflow, port_);
1051 port_destruct(struct ofport *port_)
1053 struct ofport_dpif *port = ofport_dpif_cast(port_);
1054 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1056 ofproto->need_revalidate = true;
1057 bundle_remove(port_);
1058 set_cfm(port_, NULL);
1059 if (ofproto->sflow) {
1060 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1063 ofport_clear_priorities(port);
1064 hmap_destroy(&port->priorities);
1068 port_modified(struct ofport *port_)
1070 struct ofport_dpif *port = ofport_dpif_cast(port_);
1072 if (port->bundle && port->bundle->bond) {
1073 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1078 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1080 struct ofport_dpif *port = ofport_dpif_cast(port_);
1081 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1082 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1084 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1085 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD)) {
1086 ofproto->need_revalidate = true;
1088 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1089 bundle_update(port->bundle);
1095 set_sflow(struct ofproto *ofproto_,
1096 const struct ofproto_sflow_options *sflow_options)
1098 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1099 struct dpif_sflow *ds = ofproto->sflow;
1101 if (sflow_options) {
1103 struct ofport_dpif *ofport;
1105 ds = ofproto->sflow = dpif_sflow_create(ofproto->dpif);
1106 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1107 dpif_sflow_add_port(ds, &ofport->up);
1109 ofproto->need_revalidate = true;
1111 dpif_sflow_set_options(ds, sflow_options);
1114 dpif_sflow_destroy(ds);
1115 ofproto->need_revalidate = true;
1116 ofproto->sflow = NULL;
1123 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1125 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1132 struct ofproto_dpif *ofproto;
1134 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1135 ofproto->need_revalidate = true;
1136 ofport->cfm = cfm_create(netdev_get_name(ofport->up.netdev));
1139 if (cfm_configure(ofport->cfm, s)) {
1145 cfm_destroy(ofport->cfm);
1151 get_cfm_fault(const struct ofport *ofport_)
1153 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1155 return ofport->cfm ? cfm_get_fault(ofport->cfm) : -1;
1159 get_cfm_remote_mpids(const struct ofport *ofport_, const uint64_t **rmps,
1162 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1165 cfm_get_remote_mpids(ofport->cfm, rmps, n_rmps);
1173 get_cfm_health(const struct ofport *ofport_)
1175 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1177 return ofport->cfm ? cfm_get_health(ofport->cfm) : -1;
1180 /* Spanning Tree. */
1183 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
1185 struct ofproto_dpif *ofproto = ofproto_;
1186 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
1187 struct ofport_dpif *ofport;
1189 ofport = stp_port_get_aux(sp);
1191 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
1192 ofproto->up.name, port_num);
1194 struct eth_header *eth = pkt->l2;
1196 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
1197 if (eth_addr_is_zero(eth->eth_src)) {
1198 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
1199 "with unknown MAC", ofproto->up.name, port_num);
1201 send_packet(ofport, pkt);
1207 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
1209 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
1211 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1213 /* Only revalidate flows if the configuration changed. */
1214 if (!s != !ofproto->stp) {
1215 ofproto->need_revalidate = true;
1219 if (!ofproto->stp) {
1220 ofproto->stp = stp_create(ofproto_->name, s->system_id,
1221 send_bpdu_cb, ofproto);
1222 ofproto->stp_last_tick = time_msec();
1225 stp_set_bridge_id(ofproto->stp, s->system_id);
1226 stp_set_bridge_priority(ofproto->stp, s->priority);
1227 stp_set_hello_time(ofproto->stp, s->hello_time);
1228 stp_set_max_age(ofproto->stp, s->max_age);
1229 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
1231 struct ofport *ofport;
1233 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
1234 set_stp_port(ofport, NULL);
1237 stp_destroy(ofproto->stp);
1238 ofproto->stp = NULL;
1245 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
1247 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1251 s->bridge_id = stp_get_bridge_id(ofproto->stp);
1252 s->designated_root = stp_get_designated_root(ofproto->stp);
1253 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
1262 update_stp_port_state(struct ofport_dpif *ofport)
1264 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1265 enum stp_state state;
1267 /* Figure out new state. */
1268 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
1272 if (ofport->stp_state != state) {
1273 enum ofputil_port_state of_state;
1276 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
1277 netdev_get_name(ofport->up.netdev),
1278 stp_state_name(ofport->stp_state),
1279 stp_state_name(state));
1280 if (stp_learn_in_state(ofport->stp_state)
1281 != stp_learn_in_state(state)) {
1282 /* xxx Learning action flows should also be flushed. */
1283 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
1285 fwd_change = stp_forward_in_state(ofport->stp_state)
1286 != stp_forward_in_state(state);
1288 ofproto->need_revalidate = true;
1289 ofport->stp_state = state;
1290 ofport->stp_state_entered = time_msec();
1292 if (fwd_change && ofport->bundle) {
1293 bundle_update(ofport->bundle);
1296 /* Update the STP state bits in the OpenFlow port description. */
1297 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
1298 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
1299 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
1300 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
1301 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
1303 ofproto_port_set_state(&ofport->up, of_state);
1307 /* Configures STP on 'ofport_' using the settings defined in 's'. The
1308 * caller is responsible for assigning STP port numbers and ensuring
1309 * there are no duplicates. */
1311 set_stp_port(struct ofport *ofport_,
1312 const struct ofproto_port_stp_settings *s)
1314 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1315 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1316 struct stp_port *sp = ofport->stp_port;
1318 if (!s || !s->enable) {
1320 ofport->stp_port = NULL;
1321 stp_port_disable(sp);
1322 update_stp_port_state(ofport);
1325 } else if (sp && stp_port_no(sp) != s->port_num
1326 && ofport == stp_port_get_aux(sp)) {
1327 /* The port-id changed, so disable the old one if it's not
1328 * already in use by another port. */
1329 stp_port_disable(sp);
1332 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
1333 stp_port_enable(sp);
1335 stp_port_set_aux(sp, ofport);
1336 stp_port_set_priority(sp, s->priority);
1337 stp_port_set_path_cost(sp, s->path_cost);
1339 update_stp_port_state(ofport);
1345 get_stp_port_status(struct ofport *ofport_,
1346 struct ofproto_port_stp_status *s)
1348 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1349 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1350 struct stp_port *sp = ofport->stp_port;
1352 if (!ofproto->stp || !sp) {
1358 s->port_id = stp_port_get_id(sp);
1359 s->state = stp_port_get_state(sp);
1360 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
1361 s->role = stp_port_get_role(sp);
1362 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
1368 stp_run(struct ofproto_dpif *ofproto)
1371 long long int now = time_msec();
1372 long long int elapsed = now - ofproto->stp_last_tick;
1373 struct stp_port *sp;
1376 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
1377 ofproto->stp_last_tick = now;
1379 while (stp_get_changed_port(ofproto->stp, &sp)) {
1380 struct ofport_dpif *ofport = stp_port_get_aux(sp);
1383 update_stp_port_state(ofport);
1387 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
1388 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
1394 stp_wait(struct ofproto_dpif *ofproto)
1397 poll_timer_wait(1000);
1401 /* Returns true if STP should process 'flow'. */
1403 stp_should_process_flow(const struct flow *flow)
1405 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
1409 stp_process_packet(const struct ofport_dpif *ofport,
1410 const struct ofpbuf *packet)
1412 struct ofpbuf payload = *packet;
1413 struct eth_header *eth = payload.data;
1414 struct stp_port *sp = ofport->stp_port;
1416 /* Sink packets on ports that have STP disabled when the bridge has
1418 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
1422 /* Trim off padding on payload. */
1423 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
1424 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
1427 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
1428 stp_received_bpdu(sp, payload.data, payload.size);
1432 static struct priority_to_dscp *
1433 get_priority(const struct ofport_dpif *ofport, uint32_t priority)
1435 struct priority_to_dscp *pdscp;
1438 hash = hash_int(priority, 0);
1439 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) {
1440 if (pdscp->priority == priority) {
1448 ofport_clear_priorities(struct ofport_dpif *ofport)
1450 struct priority_to_dscp *pdscp, *next;
1452 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) {
1453 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
1459 set_queues(struct ofport *ofport_,
1460 const struct ofproto_port_queue *qdscp_list,
1463 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1464 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1465 struct hmap new = HMAP_INITIALIZER(&new);
1468 for (i = 0; i < n_qdscp; i++) {
1469 struct priority_to_dscp *pdscp;
1473 dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
1474 if (dpif_queue_to_priority(ofproto->dpif, qdscp_list[i].queue,
1479 pdscp = get_priority(ofport, priority);
1481 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
1483 pdscp = xmalloc(sizeof *pdscp);
1484 pdscp->priority = priority;
1486 ofproto->need_revalidate = true;
1489 if (pdscp->dscp != dscp) {
1491 ofproto->need_revalidate = true;
1494 hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0));
1497 if (!hmap_is_empty(&ofport->priorities)) {
1498 ofport_clear_priorities(ofport);
1499 ofproto->need_revalidate = true;
1502 hmap_swap(&new, &ofport->priorities);
1510 /* Expires all MAC learning entries associated with 'bundle' and forces its
1511 * ofproto to revalidate every flow.
1513 * Normally MAC learning entries are removed only from the ofproto associated
1514 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
1515 * are removed from every ofproto. When patch ports and SLB bonds are in use
1516 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
1517 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
1518 * with the host from which it migrated. */
1520 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
1522 struct ofproto_dpif *ofproto = bundle->ofproto;
1523 struct mac_learning *ml = ofproto->ml;
1524 struct mac_entry *mac, *next_mac;
1526 ofproto->need_revalidate = true;
1527 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
1528 if (mac->port.p == bundle) {
1530 struct ofproto_dpif *o;
1532 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1534 struct mac_entry *e;
1536 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan,
1539 tag_set_add(&o->revalidate_set, e->tag);
1540 mac_learning_expire(o->ml, e);
1546 mac_learning_expire(ml, mac);
1551 static struct ofbundle *
1552 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
1554 struct ofbundle *bundle;
1556 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
1557 &ofproto->bundles) {
1558 if (bundle->aux == aux) {
1565 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
1566 * ones that are found to 'bundles'. */
1568 bundle_lookup_multiple(struct ofproto_dpif *ofproto,
1569 void **auxes, size_t n_auxes,
1570 struct hmapx *bundles)
1574 hmapx_init(bundles);
1575 for (i = 0; i < n_auxes; i++) {
1576 struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
1578 hmapx_add(bundles, bundle);
1584 bundle_update(struct ofbundle *bundle)
1586 struct ofport_dpif *port;
1588 bundle->floodable = true;
1589 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1590 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
1591 || !stp_forward_in_state(port->stp_state)) {
1592 bundle->floodable = false;
1599 bundle_del_port(struct ofport_dpif *port)
1601 struct ofbundle *bundle = port->bundle;
1603 bundle->ofproto->need_revalidate = true;
1605 list_remove(&port->bundle_node);
1606 port->bundle = NULL;
1609 lacp_slave_unregister(bundle->lacp, port);
1612 bond_slave_unregister(bundle->bond, port);
1615 bundle_update(bundle);
1619 bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port,
1620 struct lacp_slave_settings *lacp,
1621 uint32_t bond_stable_id)
1623 struct ofport_dpif *port;
1625 port = get_ofp_port(bundle->ofproto, ofp_port);
1630 if (port->bundle != bundle) {
1631 bundle->ofproto->need_revalidate = true;
1633 bundle_del_port(port);
1636 port->bundle = bundle;
1637 list_push_back(&bundle->ports, &port->bundle_node);
1638 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
1639 || !stp_forward_in_state(port->stp_state)) {
1640 bundle->floodable = false;
1644 port->bundle->ofproto->need_revalidate = true;
1645 lacp_slave_register(bundle->lacp, port, lacp);
1648 port->bond_stable_id = bond_stable_id;
1654 bundle_destroy(struct ofbundle *bundle)
1656 struct ofproto_dpif *ofproto;
1657 struct ofport_dpif *port, *next_port;
1664 ofproto = bundle->ofproto;
1665 for (i = 0; i < MAX_MIRRORS; i++) {
1666 struct ofmirror *m = ofproto->mirrors[i];
1668 if (m->out == bundle) {
1670 } else if (hmapx_find_and_delete(&m->srcs, bundle)
1671 || hmapx_find_and_delete(&m->dsts, bundle)) {
1672 ofproto->need_revalidate = true;
1677 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
1678 bundle_del_port(port);
1681 bundle_flush_macs(bundle, true);
1682 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
1684 free(bundle->trunks);
1685 lacp_destroy(bundle->lacp);
1686 bond_destroy(bundle->bond);
1691 bundle_set(struct ofproto *ofproto_, void *aux,
1692 const struct ofproto_bundle_settings *s)
1694 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1695 bool need_flush = false;
1696 struct ofport_dpif *port;
1697 struct ofbundle *bundle;
1698 unsigned long *trunks;
1704 bundle_destroy(bundle_lookup(ofproto, aux));
1708 assert(s->n_slaves == 1 || s->bond != NULL);
1709 assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
1711 bundle = bundle_lookup(ofproto, aux);
1713 bundle = xmalloc(sizeof *bundle);
1715 bundle->ofproto = ofproto;
1716 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
1717 hash_pointer(aux, 0));
1719 bundle->name = NULL;
1721 list_init(&bundle->ports);
1722 bundle->vlan_mode = PORT_VLAN_TRUNK;
1724 bundle->trunks = NULL;
1725 bundle->use_priority_tags = s->use_priority_tags;
1726 bundle->lacp = NULL;
1727 bundle->bond = NULL;
1729 bundle->floodable = true;
1731 bundle->src_mirrors = 0;
1732 bundle->dst_mirrors = 0;
1733 bundle->mirror_out = 0;
1736 if (!bundle->name || strcmp(s->name, bundle->name)) {
1738 bundle->name = xstrdup(s->name);
1743 if (!bundle->lacp) {
1744 ofproto->need_revalidate = true;
1745 bundle->lacp = lacp_create();
1747 lacp_configure(bundle->lacp, s->lacp);
1749 lacp_destroy(bundle->lacp);
1750 bundle->lacp = NULL;
1753 /* Update set of ports. */
1755 for (i = 0; i < s->n_slaves; i++) {
1756 if (!bundle_add_port(bundle, s->slaves[i],
1757 s->lacp ? &s->lacp_slaves[i] : NULL,
1758 s->bond_stable_ids ? s->bond_stable_ids[i] : 0)) {
1762 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
1763 struct ofport_dpif *next_port;
1765 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
1766 for (i = 0; i < s->n_slaves; i++) {
1767 if (s->slaves[i] == port->up.ofp_port) {
1772 bundle_del_port(port);
1776 assert(list_size(&bundle->ports) <= s->n_slaves);
1778 if (list_is_empty(&bundle->ports)) {
1779 bundle_destroy(bundle);
1783 /* Set VLAN tagging mode */
1784 if (s->vlan_mode != bundle->vlan_mode
1785 || s->use_priority_tags != bundle->use_priority_tags) {
1786 bundle->vlan_mode = s->vlan_mode;
1787 bundle->use_priority_tags = s->use_priority_tags;
1792 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
1793 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
1795 if (vlan != bundle->vlan) {
1796 bundle->vlan = vlan;
1800 /* Get trunked VLANs. */
1801 switch (s->vlan_mode) {
1802 case PORT_VLAN_ACCESS:
1806 case PORT_VLAN_TRUNK:
1807 trunks = (unsigned long *) s->trunks;
1810 case PORT_VLAN_NATIVE_UNTAGGED:
1811 case PORT_VLAN_NATIVE_TAGGED:
1812 if (vlan != 0 && (!s->trunks
1813 || !bitmap_is_set(s->trunks, vlan)
1814 || bitmap_is_set(s->trunks, 0))) {
1815 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
1817 trunks = bitmap_clone(s->trunks, 4096);
1819 trunks = bitmap_allocate1(4096);
1821 bitmap_set1(trunks, vlan);
1822 bitmap_set0(trunks, 0);
1824 trunks = (unsigned long *) s->trunks;
1831 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
1832 free(bundle->trunks);
1833 if (trunks == s->trunks) {
1834 bundle->trunks = vlan_bitmap_clone(trunks);
1836 bundle->trunks = trunks;
1841 if (trunks != s->trunks) {
1846 if (!list_is_short(&bundle->ports)) {
1847 bundle->ofproto->has_bonded_bundles = true;
1849 if (bond_reconfigure(bundle->bond, s->bond)) {
1850 ofproto->need_revalidate = true;
1853 bundle->bond = bond_create(s->bond);
1854 ofproto->need_revalidate = true;
1857 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1858 bond_slave_register(bundle->bond, port, port->bond_stable_id,
1862 bond_destroy(bundle->bond);
1863 bundle->bond = NULL;
1866 /* If we changed something that would affect MAC learning, un-learn
1867 * everything on this port and force flow revalidation. */
1869 bundle_flush_macs(bundle, false);
1876 bundle_remove(struct ofport *port_)
1878 struct ofport_dpif *port = ofport_dpif_cast(port_);
1879 struct ofbundle *bundle = port->bundle;
1882 bundle_del_port(port);
1883 if (list_is_empty(&bundle->ports)) {
1884 bundle_destroy(bundle);
1885 } else if (list_is_short(&bundle->ports)) {
1886 bond_destroy(bundle->bond);
1887 bundle->bond = NULL;
1893 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
1895 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
1896 struct ofport_dpif *port = port_;
1897 uint8_t ea[ETH_ADDR_LEN];
1900 error = netdev_get_etheraddr(port->up.netdev, ea);
1902 struct ofpbuf packet;
1905 ofpbuf_init(&packet, 0);
1906 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
1908 memcpy(packet_pdu, pdu, pdu_size);
1910 send_packet(port, &packet);
1911 ofpbuf_uninit(&packet);
1913 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
1914 "%s (%s)", port->bundle->name,
1915 netdev_get_name(port->up.netdev), strerror(error));
1920 bundle_send_learning_packets(struct ofbundle *bundle)
1922 struct ofproto_dpif *ofproto = bundle->ofproto;
1923 int error, n_packets, n_errors;
1924 struct mac_entry *e;
1926 error = n_packets = n_errors = 0;
1927 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
1928 if (e->port.p != bundle) {
1929 struct ofpbuf *learning_packet;
1930 struct ofport_dpif *port;
1934 /* The assignment to "port" is unnecessary but makes "grep"ing for
1935 * struct ofport_dpif more effective. */
1936 learning_packet = bond_compose_learning_packet(bundle->bond,
1940 ret = send_packet(port, learning_packet);
1941 ofpbuf_delete(learning_packet);
1951 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1952 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
1953 "packets, last error was: %s",
1954 bundle->name, n_errors, n_packets, strerror(error));
1956 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
1957 bundle->name, n_packets);
1962 bundle_run(struct ofbundle *bundle)
1965 lacp_run(bundle->lacp, send_pdu_cb);
1968 struct ofport_dpif *port;
1970 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1971 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
1974 bond_run(bundle->bond, &bundle->ofproto->revalidate_set,
1975 lacp_status(bundle->lacp));
1976 if (bond_should_send_learning_packets(bundle->bond)) {
1977 bundle_send_learning_packets(bundle);
1983 bundle_wait(struct ofbundle *bundle)
1986 lacp_wait(bundle->lacp);
1989 bond_wait(bundle->bond);
1996 mirror_scan(struct ofproto_dpif *ofproto)
2000 for (idx = 0; idx < MAX_MIRRORS; idx++) {
2001 if (!ofproto->mirrors[idx]) {
2008 static struct ofmirror *
2009 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
2013 for (i = 0; i < MAX_MIRRORS; i++) {
2014 struct ofmirror *mirror = ofproto->mirrors[i];
2015 if (mirror && mirror->aux == aux) {
2023 /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */
2025 mirror_update_dups(struct ofproto_dpif *ofproto)
2029 for (i = 0; i < MAX_MIRRORS; i++) {
2030 struct ofmirror *m = ofproto->mirrors[i];
2033 m->dup_mirrors = MIRROR_MASK_C(1) << i;
2037 for (i = 0; i < MAX_MIRRORS; i++) {
2038 struct ofmirror *m1 = ofproto->mirrors[i];
2045 for (j = i + 1; j < MAX_MIRRORS; j++) {
2046 struct ofmirror *m2 = ofproto->mirrors[j];
2048 if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) {
2049 m1->dup_mirrors |= MIRROR_MASK_C(1) << j;
2050 m2->dup_mirrors |= m1->dup_mirrors;
2057 mirror_set(struct ofproto *ofproto_, void *aux,
2058 const struct ofproto_mirror_settings *s)
2060 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2061 mirror_mask_t mirror_bit;
2062 struct ofbundle *bundle;
2063 struct ofmirror *mirror;
2064 struct ofbundle *out;
2065 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
2066 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
2069 mirror = mirror_lookup(ofproto, aux);
2071 mirror_destroy(mirror);
2077 idx = mirror_scan(ofproto);
2079 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
2081 ofproto->up.name, MAX_MIRRORS, s->name);
2085 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
2086 mirror->ofproto = ofproto;
2089 mirror->out_vlan = -1;
2090 mirror->name = NULL;
2093 if (!mirror->name || strcmp(s->name, mirror->name)) {
2095 mirror->name = xstrdup(s->name);
2098 /* Get the new configuration. */
2099 if (s->out_bundle) {
2100 out = bundle_lookup(ofproto, s->out_bundle);
2102 mirror_destroy(mirror);
2108 out_vlan = s->out_vlan;
2110 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
2111 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
2113 /* If the configuration has not changed, do nothing. */
2114 if (hmapx_equals(&srcs, &mirror->srcs)
2115 && hmapx_equals(&dsts, &mirror->dsts)
2116 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
2117 && mirror->out == out
2118 && mirror->out_vlan == out_vlan)
2120 hmapx_destroy(&srcs);
2121 hmapx_destroy(&dsts);
2125 hmapx_swap(&srcs, &mirror->srcs);
2126 hmapx_destroy(&srcs);
2128 hmapx_swap(&dsts, &mirror->dsts);
2129 hmapx_destroy(&dsts);
2131 free(mirror->vlans);
2132 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
2135 mirror->out_vlan = out_vlan;
2137 /* Update bundles. */
2138 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2139 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
2140 if (hmapx_contains(&mirror->srcs, bundle)) {
2141 bundle->src_mirrors |= mirror_bit;
2143 bundle->src_mirrors &= ~mirror_bit;
2146 if (hmapx_contains(&mirror->dsts, bundle)) {
2147 bundle->dst_mirrors |= mirror_bit;
2149 bundle->dst_mirrors &= ~mirror_bit;
2152 if (mirror->out == bundle) {
2153 bundle->mirror_out |= mirror_bit;
2155 bundle->mirror_out &= ~mirror_bit;
2159 ofproto->need_revalidate = true;
2160 ofproto->has_mirrors = true;
2161 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
2162 mirror_update_dups(ofproto);
2168 mirror_destroy(struct ofmirror *mirror)
2170 struct ofproto_dpif *ofproto;
2171 mirror_mask_t mirror_bit;
2172 struct ofbundle *bundle;
2179 ofproto = mirror->ofproto;
2180 ofproto->need_revalidate = true;
2181 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
2183 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2184 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
2185 bundle->src_mirrors &= ~mirror_bit;
2186 bundle->dst_mirrors &= ~mirror_bit;
2187 bundle->mirror_out &= ~mirror_bit;
2190 hmapx_destroy(&mirror->srcs);
2191 hmapx_destroy(&mirror->dsts);
2192 free(mirror->vlans);
2194 ofproto->mirrors[mirror->idx] = NULL;
2198 mirror_update_dups(ofproto);
2200 ofproto->has_mirrors = false;
2201 for (i = 0; i < MAX_MIRRORS; i++) {
2202 if (ofproto->mirrors[i]) {
2203 ofproto->has_mirrors = true;
2210 mirror_get_stats(struct ofproto *ofproto_, void *aux,
2211 uint64_t *packets, uint64_t *bytes)
2213 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2214 struct ofmirror *mirror = mirror_lookup(ofproto, aux);
2217 *packets = *bytes = UINT64_MAX;
2221 *packets = mirror->packet_count;
2222 *bytes = mirror->byte_count;
2228 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2230 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2231 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2232 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
2238 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2240 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2241 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2242 return bundle && bundle->mirror_out != 0;
2246 forward_bpdu_changed(struct ofproto *ofproto_)
2248 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2249 /* Revalidate cached flows whenever forward_bpdu option changes. */
2250 ofproto->need_revalidate = true;
2254 set_mac_idle_time(struct ofproto *ofproto_, unsigned int idle_time)
2256 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2257 mac_learning_set_idle_time(ofproto->ml, idle_time);
2262 static struct ofport_dpif *
2263 get_ofp_port(struct ofproto_dpif *ofproto, uint16_t ofp_port)
2265 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2266 return ofport ? ofport_dpif_cast(ofport) : NULL;
2269 static struct ofport_dpif *
2270 get_odp_port(struct ofproto_dpif *ofproto, uint32_t odp_port)
2272 return get_ofp_port(ofproto, odp_port_to_ofp_port(odp_port));
2276 ofproto_port_from_dpif_port(struct ofproto_port *ofproto_port,
2277 struct dpif_port *dpif_port)
2279 ofproto_port->name = dpif_port->name;
2280 ofproto_port->type = dpif_port->type;
2281 ofproto_port->ofp_port = odp_port_to_ofp_port(dpif_port->port_no);
2285 port_run(struct ofport_dpif *ofport)
2287 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
2288 bool carrier_changed = carrier_seq != ofport->carrier_seq;
2289 bool enable = netdev_get_carrier(ofport->up.netdev);
2291 ofport->carrier_seq = carrier_seq;
2294 cfm_run(ofport->cfm);
2296 if (cfm_should_send_ccm(ofport->cfm)) {
2297 struct ofpbuf packet;
2299 ofpbuf_init(&packet, 0);
2300 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
2301 send_packet(ofport, &packet);
2302 ofpbuf_uninit(&packet);
2305 enable = enable && !cfm_get_fault(ofport->cfm)
2306 && cfm_get_opup(ofport->cfm);
2309 if (ofport->bundle) {
2310 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
2311 if (carrier_changed) {
2312 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
2316 if (ofport->may_enable != enable) {
2317 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2319 if (ofproto->has_bundle_action) {
2320 ofproto->need_revalidate = true;
2324 ofport->may_enable = enable;
2328 port_wait(struct ofport_dpif *ofport)
2331 cfm_wait(ofport->cfm);
2336 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
2337 struct ofproto_port *ofproto_port)
2339 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2340 struct dpif_port dpif_port;
2343 error = dpif_port_query_by_name(ofproto->dpif, devname, &dpif_port);
2345 ofproto_port_from_dpif_port(ofproto_port, &dpif_port);
2351 port_add(struct ofproto *ofproto_, struct netdev *netdev, uint16_t *ofp_portp)
2353 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2357 error = dpif_port_add(ofproto->dpif, netdev, &odp_port);
2359 *ofp_portp = odp_port_to_ofp_port(odp_port);
2365 port_del(struct ofproto *ofproto_, uint16_t ofp_port)
2367 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2370 error = dpif_port_del(ofproto->dpif, ofp_port_to_odp_port(ofp_port));
2372 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
2374 /* The caller is going to close ofport->up.netdev. If this is a
2375 * bonded port, then the bond is using that netdev, so remove it
2376 * from the bond. The client will need to reconfigure everything
2377 * after deleting ports, so then the slave will get re-added. */
2378 bundle_remove(&ofport->up);
2385 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
2387 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2390 error = netdev_get_stats(ofport->up.netdev, stats);
2392 if (!error && ofport->odp_port == OVSP_LOCAL) {
2393 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2395 /* ofproto->stats.tx_packets represents packets that we created
2396 * internally and sent to some port (e.g. packets sent with
2397 * send_packet()). Account for them as if they had come from
2398 * OFPP_LOCAL and got forwarded. */
2400 if (stats->rx_packets != UINT64_MAX) {
2401 stats->rx_packets += ofproto->stats.tx_packets;
2404 if (stats->rx_bytes != UINT64_MAX) {
2405 stats->rx_bytes += ofproto->stats.tx_bytes;
2408 /* ofproto->stats.rx_packets represents packets that were received on
2409 * some port and we processed internally and dropped (e.g. STP).
2410 * Account fro them as if they had been forwarded to OFPP_LOCAL. */
2412 if (stats->tx_packets != UINT64_MAX) {
2413 stats->tx_packets += ofproto->stats.rx_packets;
2416 if (stats->tx_bytes != UINT64_MAX) {
2417 stats->tx_bytes += ofproto->stats.rx_bytes;
2424 /* Account packets for LOCAL port. */
2426 ofproto_update_local_port_stats(const struct ofproto *ofproto_,
2427 size_t tx_size, size_t rx_size)
2429 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2432 ofproto->stats.rx_packets++;
2433 ofproto->stats.rx_bytes += rx_size;
2436 ofproto->stats.tx_packets++;
2437 ofproto->stats.tx_bytes += tx_size;
2441 struct port_dump_state {
2442 struct dpif_port_dump dump;
2447 port_dump_start(const struct ofproto *ofproto_, void **statep)
2449 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2450 struct port_dump_state *state;
2452 *statep = state = xmalloc(sizeof *state);
2453 dpif_port_dump_start(&state->dump, ofproto->dpif);
2454 state->done = false;
2459 port_dump_next(const struct ofproto *ofproto_ OVS_UNUSED, void *state_,
2460 struct ofproto_port *port)
2462 struct port_dump_state *state = state_;
2463 struct dpif_port dpif_port;
2465 if (dpif_port_dump_next(&state->dump, &dpif_port)) {
2466 ofproto_port_from_dpif_port(port, &dpif_port);
2469 int error = dpif_port_dump_done(&state->dump);
2471 return error ? error : EOF;
2476 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
2478 struct port_dump_state *state = state_;
2481 dpif_port_dump_done(&state->dump);
2488 port_poll(const struct ofproto *ofproto_, char **devnamep)
2490 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2491 return dpif_port_poll(ofproto->dpif, devnamep);
2495 port_poll_wait(const struct ofproto *ofproto_)
2497 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2498 dpif_port_poll_wait(ofproto->dpif);
2502 port_is_lacp_current(const struct ofport *ofport_)
2504 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2505 return (ofport->bundle && ofport->bundle->lacp
2506 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
2510 /* Upcall handling. */
2512 /* Flow miss batching.
2514 * Some dpifs implement operations faster when you hand them off in a batch.
2515 * To allow batching, "struct flow_miss" queues the dpif-related work needed
2516 * for a given flow. Each "struct flow_miss" corresponds to sending one or
2517 * more packets, plus possibly installing the flow in the dpif.
2519 * So far we only batch the operations that affect flow setup time the most.
2520 * It's possible to batch more than that, but the benefit might be minimal. */
2522 struct hmap_node hmap_node;
2524 enum odp_key_fitness key_fitness;
2525 const struct nlattr *key;
2527 ovs_be16 initial_tci;
2528 struct list packets;
2531 struct flow_miss_op {
2532 struct dpif_op dpif_op;
2533 struct subfacet *subfacet; /* Subfacet */
2534 void *garbage; /* Pointer to pass to free(), NULL if none. */
2535 uint64_t stub[1024 / 8]; /* Temporary buffer. */
2538 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
2539 * OpenFlow controller as necessary according to their individual
2540 * configurations. */
2542 send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet,
2543 const struct flow *flow)
2545 struct ofputil_packet_in pin;
2547 pin.packet = packet->data;
2548 pin.packet_len = packet->size;
2549 pin.reason = OFPR_NO_MATCH;
2550 pin.controller_id = 0;
2555 pin.send_len = 0; /* not used for flow table misses */
2557 flow_get_metadata(flow, &pin.fmd);
2559 /* Registers aren't meaningful on a miss. */
2560 memset(pin.fmd.reg_masks, 0, sizeof pin.fmd.reg_masks);
2562 connmgr_send_packet_in(ofproto->up.connmgr, &pin);
2566 process_special(struct ofproto_dpif *ofproto, const struct flow *flow,
2567 const struct ofpbuf *packet)
2569 struct ofport_dpif *ofport = get_ofp_port(ofproto, flow->in_port);
2575 if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) {
2577 cfm_process_heartbeat(ofport->cfm, packet);
2580 } else if (ofport->bundle && ofport->bundle->lacp
2581 && flow->dl_type == htons(ETH_TYPE_LACP)) {
2583 lacp_process_packet(ofport->bundle->lacp, ofport, packet);
2586 } else if (ofproto->stp && stp_should_process_flow(flow)) {
2588 stp_process_packet(ofport, packet);
2595 static struct flow_miss *
2596 flow_miss_find(struct hmap *todo, const struct flow *flow, uint32_t hash)
2598 struct flow_miss *miss;
2600 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
2601 if (flow_equal(&miss->flow, flow)) {
2609 /* Partially Initializes 'op' as an "execute" operation for 'miss' and
2610 * 'packet'. The caller must initialize op->actions and op->actions_len. If
2611 * 'miss' is associated with a subfacet the caller must also initialize the
2612 * returned op->subfacet, and if anything needs to be freed after processing
2613 * the op, the caller must initialize op->garbage also. */
2615 init_flow_miss_execute_op(struct flow_miss *miss, struct ofpbuf *packet,
2616 struct flow_miss_op *op)
2618 if (miss->flow.vlan_tci != miss->initial_tci) {
2619 /* This packet was received on a VLAN splinter port. We
2620 * added a VLAN to the packet to make the packet resemble
2621 * the flow, but the actions were composed assuming that
2622 * the packet contained no VLAN. So, we must remove the
2623 * VLAN header from the packet before trying to execute the
2625 eth_pop_vlan(packet);
2628 op->subfacet = NULL;
2630 op->dpif_op.type = DPIF_OP_EXECUTE;
2631 op->dpif_op.u.execute.key = miss->key;
2632 op->dpif_op.u.execute.key_len = miss->key_len;
2633 op->dpif_op.u.execute.packet = packet;
2636 /* Helper for handle_flow_miss_without_facet() and
2637 * handle_flow_miss_with_facet(). */
2639 handle_flow_miss_common(struct rule_dpif *rule,
2640 struct ofpbuf *packet, const struct flow *flow)
2642 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
2644 ofproto->n_matches++;
2646 if (rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
2648 * Extra-special case for fail-open mode.
2650 * We are in fail-open mode and the packet matched the fail-open
2651 * rule, but we are connected to a controller too. We should send
2652 * the packet up to the controller in the hope that it will try to
2653 * set up a flow and thereby allow us to exit fail-open.
2655 * See the top-level comment in fail-open.c for more information.
2657 send_packet_in_miss(ofproto, packet, flow);
2661 /* Figures out whether a flow that missed in 'ofproto', whose details are in
2662 * 'miss', is likely to be worth tracking in detail in userspace and (usually)
2663 * installing a datapath flow. The answer is usually "yes" (a return value of
2664 * true). However, for short flows the cost of bookkeeping is much higher than
2665 * the benefits, so when the datapath holds a large number of flows we impose
2666 * some heuristics to decide which flows are likely to be worth tracking. */
2668 flow_miss_should_make_facet(struct ofproto_dpif *ofproto,
2669 struct flow_miss *miss, uint32_t hash)
2671 if (!ofproto->governor) {
2674 n_subfacets = hmap_count(&ofproto->subfacets);
2675 if (n_subfacets * 2 <= ofproto->up.flow_eviction_threshold) {
2679 ofproto->governor = governor_create(ofproto->up.name);
2682 return governor_should_install_flow(ofproto->governor, hash,
2683 list_size(&miss->packets));
2686 /* Handles 'miss', which matches 'rule', without creating a facet or subfacet
2687 * or creating any datapath flow. May add an "execute" operation to 'ops' and
2688 * increment '*n_ops'. */
2690 handle_flow_miss_without_facet(struct flow_miss *miss,
2691 struct rule_dpif *rule,
2692 struct flow_miss_op *ops, size_t *n_ops)
2694 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
2695 struct action_xlate_ctx ctx;
2696 struct ofpbuf *packet;
2698 LIST_FOR_EACH (packet, list_node, &miss->packets) {
2699 struct flow_miss_op *op = &ops[*n_ops];
2700 struct dpif_flow_stats stats;
2701 struct ofpbuf odp_actions;
2703 COVERAGE_INC(facet_suppress);
2705 ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub);
2707 dpif_flow_stats_extract(&miss->flow, packet, &stats);
2708 rule_credit_stats(rule, &stats);
2710 action_xlate_ctx_init(&ctx, ofproto, &miss->flow, miss->initial_tci,
2712 ctx.resubmit_stats = &stats;
2713 xlate_actions(&ctx, rule->up.actions, rule->up.n_actions,
2716 if (odp_actions.size) {
2717 struct dpif_execute *execute = &op->dpif_op.u.execute;
2719 init_flow_miss_execute_op(miss, packet, op);
2720 execute->actions = odp_actions.data;
2721 execute->actions_len = odp_actions.size;
2722 op->garbage = ofpbuf_get_uninit_pointer(&odp_actions);
2726 ofpbuf_uninit(&odp_actions);
2731 /* Handles 'miss', which matches 'facet'. May add any required datapath
2732 * operations to 'ops', incrementing '*n_ops' for each new op. */
2734 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
2735 struct flow_miss_op *ops, size_t *n_ops)
2737 struct subfacet *subfacet;
2738 struct ofpbuf *packet;
2740 subfacet = subfacet_create(facet,
2741 miss->key_fitness, miss->key, miss->key_len,
2744 LIST_FOR_EACH (packet, list_node, &miss->packets) {
2745 struct flow_miss_op *op = &ops[*n_ops];
2746 struct dpif_flow_stats stats;
2747 struct ofpbuf odp_actions;
2749 handle_flow_miss_common(facet->rule, packet, &miss->flow);
2751 ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub);
2752 if (!facet->may_install || !subfacet->actions) {
2753 subfacet_make_actions(subfacet, packet, &odp_actions);
2756 dpif_flow_stats_extract(&facet->flow, packet, &stats);
2757 subfacet_update_stats(subfacet, &stats);
2759 if (subfacet->actions_len) {
2760 struct dpif_execute *execute = &op->dpif_op.u.execute;
2762 init_flow_miss_execute_op(miss, packet, op);
2763 op->subfacet = subfacet;
2764 if (facet->may_install) {
2765 execute->actions = subfacet->actions;
2766 execute->actions_len = subfacet->actions_len;
2767 ofpbuf_uninit(&odp_actions);
2769 execute->actions = odp_actions.data;
2770 execute->actions_len = odp_actions.size;
2771 op->garbage = ofpbuf_get_uninit_pointer(&odp_actions);
2776 ofpbuf_uninit(&odp_actions);
2780 if (facet->may_install && subfacet->key_fitness != ODP_FIT_TOO_LITTLE) {
2781 struct flow_miss_op *op = &ops[(*n_ops)++];
2782 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
2784 op->subfacet = subfacet;
2786 op->dpif_op.type = DPIF_OP_FLOW_PUT;
2787 put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
2788 put->key = miss->key;
2789 put->key_len = miss->key_len;
2790 put->actions = subfacet->actions;
2791 put->actions_len = subfacet->actions_len;
2796 /* Handles flow miss 'miss' on 'ofproto'. The flow does not match any flow in
2797 * the OpenFlow flow table. */
2799 handle_flow_miss_no_rule(struct ofproto_dpif *ofproto, struct flow_miss *miss)
2801 uint16_t in_port = miss->flow.in_port;
2802 struct ofport_dpif *port = get_ofp_port(ofproto, in_port);
2805 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, in_port);
2808 if (port && port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN) {
2809 /* XXX install 'drop' flow entry */
2810 COVERAGE_INC(ofproto_dpif_no_packet_in);
2812 const struct ofpbuf *packet;
2814 LIST_FOR_EACH (packet, list_node, &miss->packets) {
2815 send_packet_in_miss(ofproto, packet, &miss->flow);
2820 /* Handles flow miss 'miss' on 'ofproto'. May add any required datapath
2821 * operations to 'ops', incrementing '*n_ops' for each new op. */
2823 handle_flow_miss(struct ofproto_dpif *ofproto, struct flow_miss *miss,
2824 struct flow_miss_op *ops, size_t *n_ops)
2826 struct facet *facet;
2829 /* The caller must ensure that miss->hmap_node.hash contains
2830 * flow_hash(miss->flow, 0). */
2831 hash = miss->hmap_node.hash;
2833 facet = facet_lookup_valid(ofproto, &miss->flow, hash);
2835 struct rule_dpif *rule = rule_dpif_lookup(ofproto, &miss->flow, 0);
2837 handle_flow_miss_no_rule(ofproto, miss);
2839 } else if (!flow_miss_should_make_facet(ofproto, miss, hash)) {
2840 handle_flow_miss_without_facet(miss, rule, ops, n_ops);
2844 facet = facet_create(rule, &miss->flow, hash);
2846 handle_flow_miss_with_facet(miss, facet, ops, n_ops);
2849 /* Like odp_flow_key_to_flow(), this function converts the 'key_len' bytes of
2850 * OVS_KEY_ATTR_* attributes in 'key' to a flow structure in 'flow' and returns
2851 * an ODP_FIT_* value that indicates how well 'key' fits our expectations for
2852 * what a flow key should contain.
2854 * This function also includes some logic to help make VLAN splinters
2855 * transparent to the rest of the upcall processing logic. In particular, if
2856 * the extracted in_port is a VLAN splinter port, it replaces flow->in_port by
2857 * the "real" port, sets flow->vlan_tci correctly for the VLAN of the VLAN
2858 * splinter port, and pushes a VLAN header onto 'packet' (if it is nonnull).
2860 * Sets '*initial_tci' to the VLAN TCI with which the packet was really
2861 * received, that is, the actual VLAN TCI extracted by odp_flow_key_to_flow().
2862 * (This differs from the value returned in flow->vlan_tci only for packets
2863 * received on VLAN splinters.)
2865 static enum odp_key_fitness
2866 ofproto_dpif_extract_flow_key(const struct ofproto_dpif *ofproto,
2867 const struct nlattr *key, size_t key_len,
2868 struct flow *flow, ovs_be16 *initial_tci,
2869 struct ofpbuf *packet)
2871 enum odp_key_fitness fitness;
2875 fitness = odp_flow_key_to_flow(key, key_len, flow);
2876 if (fitness == ODP_FIT_ERROR) {
2879 *initial_tci = flow->vlan_tci;
2881 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port, &vid);
2883 /* Cause the flow to be processed as if it came in on the real device
2884 * with the VLAN device's VLAN ID. */
2885 flow->in_port = realdev;
2886 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
2888 /* Make the packet resemble the flow, so that it gets sent to an
2889 * OpenFlow controller properly, so that it looks correct for
2890 * sFlow, and so that flow_extract() will get the correct vlan_tci
2891 * if it is called on 'packet'.
2893 * The allocated space inside 'packet' probably also contains
2894 * 'key', that is, both 'packet' and 'key' are probably part of a
2895 * struct dpif_upcall (see the large comment on that structure
2896 * definition), so pushing data on 'packet' is in general not a
2897 * good idea since it could overwrite 'key' or free it as a side
2898 * effect. However, it's OK in this special case because we know
2899 * that 'packet' is inside a Netlink attribute: pushing 4 bytes
2900 * will just overwrite the 4-byte "struct nlattr", which is fine
2901 * since we don't need that header anymore. */
2902 eth_push_vlan(packet, flow->vlan_tci);
2905 /* Let the caller know that we can't reproduce 'key' from 'flow'. */
2906 if (fitness == ODP_FIT_PERFECT) {
2907 fitness = ODP_FIT_TOO_MUCH;
2915 handle_miss_upcalls(struct ofproto_dpif *ofproto, struct dpif_upcall *upcalls,
2918 struct dpif_upcall *upcall;
2919 struct flow_miss *miss;
2920 struct flow_miss misses[FLOW_MISS_MAX_BATCH];
2921 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
2922 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
2932 /* Construct the to-do list.
2934 * This just amounts to extracting the flow from each packet and sticking
2935 * the packets that have the same flow in the same "flow_miss" structure so
2936 * that we can process them together. */
2939 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
2940 struct flow_miss *miss = &misses[n_misses];
2941 struct flow_miss *existing_miss;
2944 /* Obtain metadata and check userspace/kernel agreement on flow match,
2945 * then set 'flow''s header pointers. */
2946 miss->key_fitness = ofproto_dpif_extract_flow_key(
2947 ofproto, upcall->key, upcall->key_len,
2948 &miss->flow, &miss->initial_tci, upcall->packet);
2949 if (miss->key_fitness == ODP_FIT_ERROR) {
2952 flow_extract(upcall->packet, miss->flow.skb_priority,
2953 miss->flow.tun_id, miss->flow.in_port, &miss->flow);
2955 /* Handle 802.1ag, LACP, and STP specially. */
2956 if (process_special(ofproto, &miss->flow, upcall->packet)) {
2957 ofproto_update_local_port_stats(&ofproto->up,
2958 0, upcall->packet->size);
2959 ofproto->n_matches++;
2963 /* Add other packets to a to-do list. */
2964 hash = flow_hash(&miss->flow, 0);
2965 existing_miss = flow_miss_find(&todo, &miss->flow, hash);
2966 if (!existing_miss) {
2967 hmap_insert(&todo, &miss->hmap_node, hash);
2968 miss->key = upcall->key;
2969 miss->key_len = upcall->key_len;
2970 list_init(&miss->packets);
2974 miss = existing_miss;
2976 list_push_back(&miss->packets, &upcall->packet->list_node);
2979 /* Process each element in the to-do list, constructing the set of
2980 * operations to batch. */
2982 HMAP_FOR_EACH (miss, hmap_node, &todo) {
2983 handle_flow_miss(ofproto, miss, flow_miss_ops, &n_ops);
2985 assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
2987 /* Execute batch. */
2988 for (i = 0; i < n_ops; i++) {
2989 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
2991 dpif_operate(ofproto->dpif, dpif_ops, n_ops);
2993 /* Free memory and update facets. */
2994 for (i = 0; i < n_ops; i++) {
2995 struct flow_miss_op *op = &flow_miss_ops[i];
2997 switch (op->dpif_op.type) {
2998 case DPIF_OP_EXECUTE:
3001 case DPIF_OP_FLOW_PUT:
3002 if (!op->dpif_op.error) {
3003 op->subfacet->installed = true;
3007 case DPIF_OP_FLOW_DEL:
3013 hmap_destroy(&todo);
3017 handle_userspace_upcall(struct ofproto_dpif *ofproto,
3018 struct dpif_upcall *upcall)
3020 struct user_action_cookie cookie;
3021 enum odp_key_fitness fitness;
3022 ovs_be16 initial_tci;
3025 memcpy(&cookie, &upcall->userdata, sizeof(cookie));
3027 fitness = ofproto_dpif_extract_flow_key(ofproto, upcall->key,
3028 upcall->key_len, &flow,
3029 &initial_tci, upcall->packet);
3030 if (fitness == ODP_FIT_ERROR) {
3034 if (cookie.type == USER_ACTION_COOKIE_SFLOW) {
3035 if (ofproto->sflow) {
3036 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
3040 VLOG_WARN_RL(&rl, "invalid user cookie : 0x%"PRIx64, upcall->userdata);
3045 handle_upcalls(struct ofproto_dpif *ofproto, unsigned int max_batch)
3047 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
3048 struct ofpbuf miss_bufs[FLOW_MISS_MAX_BATCH];
3049 uint64_t miss_buf_stubs[FLOW_MISS_MAX_BATCH][4096 / 8];
3054 assert(max_batch <= FLOW_MISS_MAX_BATCH);
3058 for (n_processed = 0; n_processed < max_batch; n_processed++) {
3059 struct dpif_upcall *upcall = &misses[n_misses];
3060 struct ofpbuf *buf = &miss_bufs[n_misses];
3063 ofpbuf_use_stub(buf, miss_buf_stubs[n_misses],
3064 sizeof miss_buf_stubs[n_misses]);
3065 error = dpif_recv(ofproto->dpif, upcall, buf);
3071 switch (upcall->type) {
3072 case DPIF_UC_ACTION:
3073 handle_userspace_upcall(ofproto, upcall);
3078 /* Handle it later. */
3082 case DPIF_N_UC_TYPES:
3084 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32,
3090 handle_miss_upcalls(ofproto, misses, n_misses);
3091 for (i = 0; i < n_misses; i++) {
3092 ofpbuf_uninit(&miss_bufs[i]);
3098 /* Flow expiration. */
3100 static int subfacet_max_idle(const struct ofproto_dpif *);
3101 static void update_stats(struct ofproto_dpif *);
3102 static void rule_expire(struct rule_dpif *);
3103 static void expire_subfacets(struct ofproto_dpif *, int dp_max_idle);
3105 /* This function is called periodically by run(). Its job is to collect
3106 * updates for the flows that have been installed into the datapath, most
3107 * importantly when they last were used, and then use that information to
3108 * expire flows that have not been used recently.
3110 * Returns the number of milliseconds after which it should be called again. */
3112 expire(struct ofproto_dpif *ofproto)
3114 struct rule_dpif *rule, *next_rule;
3115 struct oftable *table;
3118 /* Update stats for each flow in the datapath. */
3119 update_stats(ofproto);
3121 /* Expire subfacets that have been idle too long. */
3122 dp_max_idle = subfacet_max_idle(ofproto);
3123 expire_subfacets(ofproto, dp_max_idle);
3125 /* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */
3126 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
3127 struct cls_cursor cursor;
3129 cls_cursor_init(&cursor, &table->cls, NULL);
3130 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
3135 /* All outstanding data in existing flows has been accounted, so it's a
3136 * good time to do bond rebalancing. */
3137 if (ofproto->has_bonded_bundles) {
3138 struct ofbundle *bundle;
3140 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
3142 bond_rebalance(bundle->bond, &ofproto->revalidate_set);
3147 return MIN(dp_max_idle, 1000);
3150 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
3152 * This function also pushes statistics updates to rules which each facet
3153 * resubmits into. Generally these statistics will be accurate. However, if a
3154 * facet changes the rule it resubmits into at some time in between
3155 * update_stats() runs, it is possible that statistics accrued to the
3156 * old rule will be incorrectly attributed to the new rule. This could be
3157 * avoided by calling update_stats() whenever rules are created or
3158 * deleted. However, the performance impact of making so many calls to the
3159 * datapath do not justify the benefit of having perfectly accurate statistics.
3162 update_stats(struct ofproto_dpif *p)
3164 const struct dpif_flow_stats *stats;
3165 struct dpif_flow_dump dump;
3166 const struct nlattr *key;
3169 dpif_flow_dump_start(&dump, p->dpif);
3170 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
3171 struct subfacet *subfacet;
3173 subfacet = subfacet_find(p, key, key_len);
3174 if (subfacet && subfacet->installed) {
3175 struct facet *facet = subfacet->facet;
3177 if (stats->n_packets >= subfacet->dp_packet_count) {
3178 uint64_t extra = stats->n_packets - subfacet->dp_packet_count;
3179 facet->packet_count += extra;
3181 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
3184 if (stats->n_bytes >= subfacet->dp_byte_count) {
3185 facet->byte_count += stats->n_bytes - subfacet->dp_byte_count;
3187 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
3190 subfacet->dp_packet_count = stats->n_packets;
3191 subfacet->dp_byte_count = stats->n_bytes;
3193 facet->tcp_flags |= stats->tcp_flags;
3195 subfacet_update_time(subfacet, stats->used);
3196 if (facet->accounted_bytes < facet->byte_count) {
3198 facet_account(facet);
3199 facet->accounted_bytes = facet->byte_count;
3201 facet_push_stats(facet);
3203 if (!VLOG_DROP_WARN(&rl)) {
3207 odp_flow_key_format(key, key_len, &s);
3208 VLOG_WARN("unexpected flow from datapath %s", ds_cstr(&s));
3212 COVERAGE_INC(facet_unexpected);
3213 /* There's a flow in the datapath that we know nothing about, or a
3214 * flow that shouldn't be installed but was anyway. Delete it. */
3215 dpif_flow_del(p->dpif, key, key_len, NULL);
3218 dpif_flow_dump_done(&dump);
3221 /* Calculates and returns the number of milliseconds of idle time after which
3222 * subfacets should expire from the datapath. When a subfacet expires, we fold
3223 * its statistics into its facet, and when a facet's last subfacet expires, we
3224 * fold its statistic into its rule. */
3226 subfacet_max_idle(const struct ofproto_dpif *ofproto)
3229 * Idle time histogram.
3231 * Most of the time a switch has a relatively small number of subfacets.
3232 * When this is the case we might as well keep statistics for all of them
3233 * in userspace and to cache them in the kernel datapath for performance as
3236 * As the number of subfacets increases, the memory required to maintain
3237 * statistics about them in userspace and in the kernel becomes
3238 * significant. However, with a large number of subfacets it is likely
3239 * that only a few of them are "heavy hitters" that consume a large amount
3240 * of bandwidth. At this point, only heavy hitters are worth caching in
3241 * the kernel and maintaining in userspaces; other subfacets we can
3244 * The technique used to compute the idle time is to build a histogram with
3245 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
3246 * that is installed in the kernel gets dropped in the appropriate bucket.
3247 * After the histogram has been built, we compute the cutoff so that only
3248 * the most-recently-used 1% of subfacets (but at least
3249 * ofproto->up.flow_eviction_threshold flows) are kept cached. At least
3250 * the most-recently-used bucket of subfacets is kept, so actually an
3251 * arbitrary number of subfacets can be kept in any given expiration run
3252 * (though the next run will delete most of those unless they receive
3255 * This requires a second pass through the subfacets, in addition to the
3256 * pass made by update_stats(), because the former function never looks at
3257 * uninstallable subfacets.
3259 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
3260 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
3261 int buckets[N_BUCKETS] = { 0 };
3262 int total, subtotal, bucket;
3263 struct subfacet *subfacet;
3267 total = hmap_count(&ofproto->subfacets);
3268 if (total <= ofproto->up.flow_eviction_threshold) {
3269 return N_BUCKETS * BUCKET_WIDTH;
3272 /* Build histogram. */
3274 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
3275 long long int idle = now - subfacet->used;
3276 int bucket = (idle <= 0 ? 0
3277 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
3278 : (unsigned int) idle / BUCKET_WIDTH);
3282 /* Find the first bucket whose flows should be expired. */
3283 subtotal = bucket = 0;
3285 subtotal += buckets[bucket++];
3286 } while (bucket < N_BUCKETS &&
3287 subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100));
3289 if (VLOG_IS_DBG_ENABLED()) {
3293 ds_put_cstr(&s, "keep");
3294 for (i = 0; i < N_BUCKETS; i++) {
3296 ds_put_cstr(&s, ", drop");
3299 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
3302 VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s));
3306 return bucket * BUCKET_WIDTH;
3309 enum { EXPIRE_MAX_BATCH = 50 };
3312 expire_batch(struct ofproto_dpif *ofproto, struct subfacet **subfacets, int n)
3314 struct odputil_keybuf keybufs[EXPIRE_MAX_BATCH];
3315 struct dpif_op ops[EXPIRE_MAX_BATCH];
3316 struct dpif_op *opsp[EXPIRE_MAX_BATCH];
3317 struct ofpbuf keys[EXPIRE_MAX_BATCH];
3318 struct dpif_flow_stats stats[EXPIRE_MAX_BATCH];
3321 for (i = 0; i < n; i++) {
3322 ops[i].type = DPIF_OP_FLOW_DEL;
3323 subfacet_get_key(subfacets[i], &keybufs[i], &keys[i]);
3324 ops[i].u.flow_del.key = keys[i].data;
3325 ops[i].u.flow_del.key_len = keys[i].size;
3326 ops[i].u.flow_del.stats = &stats[i];
3330 dpif_operate(ofproto->dpif, opsp, n);
3331 for (i = 0; i < n; i++) {
3332 subfacet_reset_dp_stats(subfacets[i], &stats[i]);
3333 subfacets[i]->installed = false;
3334 subfacet_destroy(subfacets[i]);
3339 expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle)
3341 long long int cutoff = time_msec() - dp_max_idle;
3343 struct subfacet *subfacet, *next_subfacet;
3344 struct subfacet *batch[EXPIRE_MAX_BATCH];
3348 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
3349 &ofproto->subfacets) {
3350 if (subfacet->used < cutoff) {
3351 if (subfacet->installed) {
3352 batch[n_batch++] = subfacet;
3353 if (n_batch >= EXPIRE_MAX_BATCH) {
3354 expire_batch(ofproto, batch, n_batch);
3358 subfacet_destroy(subfacet);
3364 expire_batch(ofproto, batch, n_batch);
3368 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
3369 * then delete it entirely. */
3371 rule_expire(struct rule_dpif *rule)
3373 struct facet *facet, *next_facet;
3377 /* Has 'rule' expired? */
3379 if (rule->up.hard_timeout
3380 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
3381 reason = OFPRR_HARD_TIMEOUT;
3382 } else if (rule->up.idle_timeout
3383 && now > rule->up.used + rule->up.idle_timeout * 1000) {
3384 reason = OFPRR_IDLE_TIMEOUT;
3389 COVERAGE_INC(ofproto_dpif_expired);
3391 /* Update stats. (This is a no-op if the rule expired due to an idle
3392 * timeout, because that only happens when the rule has no facets left.) */
3393 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
3394 facet_remove(facet);
3397 /* Get rid of the rule. */
3398 ofproto_rule_expire(&rule->up, reason);
3403 /* Creates and returns a new facet owned by 'rule', given a 'flow'.
3405 * The caller must already have determined that no facet with an identical
3406 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
3407 * the ofproto's classifier table.
3409 * 'hash' must be the return value of flow_hash(flow, 0).
3411 * The facet will initially have no subfacets. The caller should create (at
3412 * least) one subfacet with subfacet_create(). */
3413 static struct facet *
3414 facet_create(struct rule_dpif *rule, const struct flow *flow, uint32_t hash)
3416 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3417 struct facet *facet;
3419 facet = xzalloc(sizeof *facet);
3420 facet->used = time_msec();
3421 hmap_insert(&ofproto->facets, &facet->hmap_node, hash);
3422 list_push_back(&rule->facets, &facet->list_node);
3424 facet->flow = *flow;
3425 list_init(&facet->subfacets);
3426 netflow_flow_init(&facet->nf_flow);
3427 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
3433 facet_free(struct facet *facet)
3438 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
3439 * 'packet', which arrived on 'in_port'.
3441 * Takes ownership of 'packet'. */
3443 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
3444 const struct nlattr *odp_actions, size_t actions_len,
3445 struct ofpbuf *packet)
3447 struct odputil_keybuf keybuf;
3451 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
3452 odp_flow_key_from_flow(&key, flow);
3454 error = dpif_execute(ofproto->dpif, key.data, key.size,
3455 odp_actions, actions_len, packet);
3457 ofpbuf_delete(packet);
3461 /* Remove 'facet' from 'ofproto' and free up the associated memory:
3463 * - If 'facet' was installed in the datapath, uninstalls it and updates its
3464 * rule's statistics, via subfacet_uninstall().
3466 * - Removes 'facet' from its rule and from ofproto->facets.
3469 facet_remove(struct facet *facet)
3471 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3472 struct subfacet *subfacet, *next_subfacet;
3474 assert(!list_is_empty(&facet->subfacets));
3476 /* First uninstall all of the subfacets to get final statistics. */
3477 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3478 subfacet_uninstall(subfacet);
3481 /* Flush the final stats to the rule.
3483 * This might require us to have at least one subfacet around so that we
3484 * can use its actions for accounting in facet_account(), which is why we
3485 * have uninstalled but not yet destroyed the subfacets. */
3486 facet_flush_stats(facet);
3488 /* Now we're really all done so destroy everything. */
3489 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
3490 &facet->subfacets) {
3491 subfacet_destroy__(subfacet);
3493 hmap_remove(&ofproto->facets, &facet->hmap_node);
3494 list_remove(&facet->list_node);
3498 /* Feed information from 'facet' back into the learning table to keep it in
3499 * sync with what is actually flowing through the datapath. */
3501 facet_learn(struct facet *facet)
3503 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3504 struct action_xlate_ctx ctx;
3506 if (!facet->has_learn
3507 && !facet->has_normal
3508 && (!facet->has_fin_timeout
3509 || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
3513 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
3514 facet->flow.vlan_tci,
3515 facet->rule, facet->tcp_flags, NULL);
3516 ctx.may_learn = true;
3517 xlate_actions_for_side_effects(&ctx, facet->rule->up.actions,
3518 facet->rule->up.n_actions);
3522 facet_account(struct facet *facet)
3524 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3525 struct subfacet *subfacet;
3526 const struct nlattr *a;
3531 if (!facet->has_normal || !ofproto->has_bonded_bundles) {
3534 n_bytes = facet->byte_count - facet->accounted_bytes;
3536 /* This loop feeds byte counters to bond_account() for rebalancing to use
3537 * as a basis. We also need to track the actual VLAN on which the packet
3538 * is going to be sent to ensure that it matches the one passed to
3539 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
3542 * We use the actions from an arbitrary subfacet because they should all
3543 * be equally valid for our purpose. */
3544 subfacet = CONTAINER_OF(list_front(&facet->subfacets),
3545 struct subfacet, list_node);
3546 vlan_tci = facet->flow.vlan_tci;
3547 NL_ATTR_FOR_EACH_UNSAFE (a, left,
3548 subfacet->actions, subfacet->actions_len) {
3549 const struct ovs_action_push_vlan *vlan;
3550 struct ofport_dpif *port;
3552 switch (nl_attr_type(a)) {
3553 case OVS_ACTION_ATTR_OUTPUT:
3554 port = get_odp_port(ofproto, nl_attr_get_u32(a));
3555 if (port && port->bundle && port->bundle->bond) {
3556 bond_account(port->bundle->bond, &facet->flow,
3557 vlan_tci_to_vid(vlan_tci), n_bytes);
3561 case OVS_ACTION_ATTR_POP_VLAN:
3562 vlan_tci = htons(0);
3565 case OVS_ACTION_ATTR_PUSH_VLAN:
3566 vlan = nl_attr_get(a);
3567 vlan_tci = vlan->vlan_tci;
3573 /* Returns true if the only action for 'facet' is to send to the controller.
3574 * (We don't report NetFlow expiration messages for such facets because they
3575 * are just part of the control logic for the network, not real traffic). */
3577 facet_is_controller_flow(struct facet *facet)
3580 && facet->rule->up.n_actions == 1
3581 && action_outputs_to_port(&facet->rule->up.actions[0],
3582 htons(OFPP_CONTROLLER)));
3585 /* Folds all of 'facet''s statistics into its rule. Also updates the
3586 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
3587 * 'facet''s statistics in the datapath should have been zeroed and folded into
3588 * its packet and byte counts before this function is called. */
3590 facet_flush_stats(struct facet *facet)
3592 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3593 struct subfacet *subfacet;
3595 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3596 assert(!subfacet->dp_byte_count);
3597 assert(!subfacet->dp_packet_count);
3600 facet_push_stats(facet);
3601 if (facet->accounted_bytes < facet->byte_count) {
3602 facet_account(facet);
3603 facet->accounted_bytes = facet->byte_count;
3606 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
3607 struct ofexpired expired;
3608 expired.flow = facet->flow;
3609 expired.packet_count = facet->packet_count;
3610 expired.byte_count = facet->byte_count;
3611 expired.used = facet->used;
3612 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
3615 facet->rule->packet_count += facet->packet_count;
3616 facet->rule->byte_count += facet->byte_count;
3618 /* Reset counters to prevent double counting if 'facet' ever gets
3620 facet_reset_counters(facet);
3622 netflow_flow_clear(&facet->nf_flow);
3623 facet->tcp_flags = 0;
3626 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
3627 * Returns it if found, otherwise a null pointer.
3629 * 'hash' must be the return value of flow_hash(flow, 0).
3631 * The returned facet might need revalidation; use facet_lookup_valid()
3632 * instead if that is important. */
3633 static struct facet *
3634 facet_find(struct ofproto_dpif *ofproto,
3635 const struct flow *flow, uint32_t hash)
3637 struct facet *facet;
3639 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, hash, &ofproto->facets) {
3640 if (flow_equal(flow, &facet->flow)) {
3648 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
3649 * Returns it if found, otherwise a null pointer.
3651 * 'hash' must be the return value of flow_hash(flow, 0).
3653 * The returned facet is guaranteed to be valid. */
3654 static struct facet *
3655 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow,
3658 struct facet *facet = facet_find(ofproto, flow, hash);
3660 /* The facet we found might not be valid, since we could be in need of
3661 * revalidation. If it is not valid, don't return it. */
3663 && (ofproto->need_revalidate
3664 || tag_set_intersects(&ofproto->revalidate_set, facet->tags))
3665 && !facet_revalidate(facet)) {
3666 COVERAGE_INC(facet_invalidated);
3674 facet_check_consistency(struct facet *facet)
3676 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3678 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3680 uint64_t odp_actions_stub[1024 / 8];
3681 struct ofpbuf odp_actions;
3683 struct rule_dpif *rule;
3684 struct subfacet *subfacet;
3685 bool may_log = false;
3688 /* Check the rule for consistency. */
3689 rule = rule_dpif_lookup(ofproto, &facet->flow, 0);
3691 if (!VLOG_DROP_WARN(&rl)) {
3692 char *s = flow_to_string(&facet->flow);
3693 VLOG_WARN("%s: facet should not exist", s);
3697 } else if (rule != facet->rule) {
3698 may_log = !VLOG_DROP_WARN(&rl);
3704 flow_format(&s, &facet->flow);
3705 ds_put_format(&s, ": facet associated with wrong rule (was "
3706 "table=%"PRIu8",", facet->rule->up.table_id);
3707 cls_rule_format(&facet->rule->up.cr, &s);
3708 ds_put_format(&s, ") (should have been table=%"PRIu8",",
3710 cls_rule_format(&rule->up.cr, &s);
3711 ds_put_char(&s, ')');
3713 VLOG_WARN("%s", ds_cstr(&s));
3720 /* Check the datapath actions for consistency. */
3721 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
3722 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3723 struct action_xlate_ctx ctx;
3724 bool actions_changed;
3725 bool should_install;
3727 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
3728 subfacet->initial_tci, rule, 0, NULL);
3729 xlate_actions(&ctx, rule->up.actions, rule->up.n_actions,
3732 should_install = (ctx.may_set_up_flow
3733 && subfacet->key_fitness != ODP_FIT_TOO_LITTLE);
3734 if (!should_install && !subfacet->installed) {
3735 /* The actions for uninstallable flows may vary from one packet to
3736 * the next, so don't compare the actions. */
3740 actions_changed = (subfacet->actions_len != odp_actions.size
3741 || memcmp(subfacet->actions, odp_actions.data,
3742 subfacet->actions_len));
3743 if (should_install != subfacet->installed || actions_changed) {
3745 may_log = !VLOG_DROP_WARN(&rl);
3750 struct odputil_keybuf keybuf;
3755 subfacet_get_key(subfacet, &keybuf, &key);
3756 odp_flow_key_format(key.data, key.size, &s);
3758 ds_put_cstr(&s, ": inconsistency in subfacet");
3759 if (should_install != subfacet->installed) {
3760 enum odp_key_fitness fitness = subfacet->key_fitness;
3762 ds_put_format(&s, " (should%s have been installed)",
3763 should_install ? "" : " not");
3764 ds_put_format(&s, " (may_set_up_flow=%s, fitness=%s)",
3765 ctx.may_set_up_flow ? "true" : "false",
3766 odp_key_fitness_to_string(fitness));
3768 if (actions_changed) {
3769 ds_put_cstr(&s, " (actions were: ");
3770 format_odp_actions(&s, subfacet->actions,
3771 subfacet->actions_len);
3772 ds_put_cstr(&s, ") (correct actions: ");
3773 format_odp_actions(&s, odp_actions.data, odp_actions.size);
3774 ds_put_char(&s, ')');
3776 ds_put_cstr(&s, " (actions: ");
3777 format_odp_actions(&s, subfacet->actions,
3778 subfacet->actions_len);
3779 ds_put_char(&s, ')');
3781 VLOG_WARN("%s", ds_cstr(&s));
3786 ofpbuf_uninit(&odp_actions);
3791 /* Re-searches the classifier for 'facet':
3793 * - If the rule found is different from 'facet''s current rule, moves
3794 * 'facet' to the new rule and recompiles its actions.
3796 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
3797 * where it is and recompiles its actions anyway.
3799 * - If there is none, destroys 'facet'.
3801 * Returns true if 'facet' still exists, false if it has been destroyed. */
3803 facet_revalidate(struct facet *facet)
3805 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3807 struct nlattr *odp_actions;
3810 struct actions *new_actions;
3812 struct action_xlate_ctx ctx;
3813 uint64_t odp_actions_stub[1024 / 8];
3814 struct ofpbuf odp_actions;
3816 struct rule_dpif *new_rule;
3817 struct subfacet *subfacet;
3818 bool actions_changed;
3821 COVERAGE_INC(facet_revalidate);
3823 /* Determine the new rule. */
3824 new_rule = rule_dpif_lookup(ofproto, &facet->flow, 0);
3826 /* No new rule, so delete the facet. */
3827 facet_remove(facet);
3831 /* Calculate new datapath actions.
3833 * We do not modify any 'facet' state yet, because we might need to, e.g.,
3834 * emit a NetFlow expiration and, if so, we need to have the old state
3835 * around to properly compose it. */
3837 /* If the datapath actions changed or the installability changed,
3838 * then we need to talk to the datapath. */
3841 memset(&ctx, 0, sizeof ctx);
3842 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
3843 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3844 bool should_install;
3846 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
3847 subfacet->initial_tci, new_rule, 0, NULL);
3848 xlate_actions(&ctx, new_rule->up.actions, new_rule->up.n_actions,
3850 actions_changed = (subfacet->actions_len != odp_actions.size
3851 || memcmp(subfacet->actions, odp_actions.data,
3852 subfacet->actions_len));
3854 should_install = (ctx.may_set_up_flow
3855 && subfacet->key_fitness != ODP_FIT_TOO_LITTLE);
3856 if (actions_changed || should_install != subfacet->installed) {
3857 if (should_install) {
3858 struct dpif_flow_stats stats;
3860 subfacet_install(subfacet,
3861 odp_actions.data, odp_actions.size, &stats);
3862 subfacet_update_stats(subfacet, &stats);
3864 subfacet_uninstall(subfacet);
3868 new_actions = xcalloc(list_size(&facet->subfacets),
3869 sizeof *new_actions);
3871 new_actions[i].odp_actions = xmemdup(odp_actions.data,
3873 new_actions[i].actions_len = odp_actions.size;
3878 ofpbuf_uninit(&odp_actions);
3881 facet_flush_stats(facet);
3884 /* Update 'facet' now that we've taken care of all the old state. */
3885 facet->tags = ctx.tags;
3886 facet->nf_flow.output_iface = ctx.nf_output_iface;
3887 facet->may_install = ctx.may_set_up_flow;
3888 facet->has_learn = ctx.has_learn;
3889 facet->has_normal = ctx.has_normal;
3890 facet->has_fin_timeout = ctx.has_fin_timeout;
3891 facet->mirrors = ctx.mirrors;
3894 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3895 if (new_actions[i].odp_actions) {
3896 free(subfacet->actions);
3897 subfacet->actions = new_actions[i].odp_actions;
3898 subfacet->actions_len = new_actions[i].actions_len;
3904 if (facet->rule != new_rule) {
3905 COVERAGE_INC(facet_changed_rule);
3906 list_remove(&facet->list_node);
3907 list_push_back(&new_rule->facets, &facet->list_node);
3908 facet->rule = new_rule;
3909 facet->used = new_rule->up.created;
3910 facet->prev_used = facet->used;
3916 /* Updates 'facet''s used time. Caller is responsible for calling
3917 * facet_push_stats() to update the flows which 'facet' resubmits into. */
3919 facet_update_time(struct facet *facet, long long int used)
3921 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3922 if (used > facet->used) {
3924 ofproto_rule_update_used(&facet->rule->up, used);
3925 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
3930 facet_reset_counters(struct facet *facet)
3932 facet->packet_count = 0;
3933 facet->byte_count = 0;
3934 facet->prev_packet_count = 0;
3935 facet->prev_byte_count = 0;
3936 facet->accounted_bytes = 0;
3940 facet_push_stats(struct facet *facet)
3942 struct dpif_flow_stats stats;
3944 assert(facet->packet_count >= facet->prev_packet_count);
3945 assert(facet->byte_count >= facet->prev_byte_count);
3946 assert(facet->used >= facet->prev_used);
3948 stats.n_packets = facet->packet_count - facet->prev_packet_count;
3949 stats.n_bytes = facet->byte_count - facet->prev_byte_count;
3950 stats.used = facet->used;
3951 stats.tcp_flags = 0;
3953 if (stats.n_packets || stats.n_bytes || facet->used > facet->prev_used) {
3954 facet->prev_packet_count = facet->packet_count;
3955 facet->prev_byte_count = facet->byte_count;
3956 facet->prev_used = facet->used;
3958 flow_push_stats(facet->rule, &facet->flow, &stats);
3960 update_mirror_stats(ofproto_dpif_cast(facet->rule->up.ofproto),
3961 facet->mirrors, stats.n_packets, stats.n_bytes);
3966 rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats)
3968 rule->packet_count += stats->n_packets;
3969 rule->byte_count += stats->n_bytes;
3970 ofproto_rule_update_used(&rule->up, stats->used);
3973 /* Pushes flow statistics to the rules which 'flow' resubmits into given
3974 * 'rule''s actions and mirrors. */
3976 flow_push_stats(struct rule_dpif *rule,
3977 const struct flow *flow, const struct dpif_flow_stats *stats)
3979 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3980 struct action_xlate_ctx ctx;
3982 ofproto_rule_update_used(&rule->up, stats->used);
3984 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci, rule,
3986 ctx.resubmit_stats = stats;
3987 xlate_actions_for_side_effects(&ctx, rule->up.actions, rule->up.n_actions);
3992 static struct subfacet *
3993 subfacet_find__(struct ofproto_dpif *ofproto,
3994 const struct nlattr *key, size_t key_len, uint32_t key_hash,
3995 const struct flow *flow)
3997 struct subfacet *subfacet;
3999 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
4000 &ofproto->subfacets) {
4002 ? (subfacet->key_len == key_len
4003 && !memcmp(key, subfacet->key, key_len))
4004 : flow_equal(flow, &subfacet->facet->flow)) {
4012 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
4013 * 'key_fitness', 'key', and 'key_len'. Returns the existing subfacet if
4014 * there is one, otherwise creates and returns a new subfacet.
4016 * If the returned subfacet is new, then subfacet->actions will be NULL, in
4017 * which case the caller must populate the actions with
4018 * subfacet_make_actions(). */
4019 static struct subfacet *
4020 subfacet_create(struct facet *facet, enum odp_key_fitness key_fitness,
4021 const struct nlattr *key, size_t key_len, ovs_be16 initial_tci)
4023 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4024 uint32_t key_hash = odp_flow_key_hash(key, key_len);
4025 struct subfacet *subfacet;
4027 subfacet = subfacet_find__(ofproto, key, key_len, key_hash, &facet->flow);
4029 if (subfacet->facet == facet) {
4033 /* This shouldn't happen. */
4034 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
4035 subfacet_destroy(subfacet);
4038 subfacet = (list_is_empty(&facet->subfacets)
4039 ? &facet->one_subfacet
4040 : xmalloc(sizeof *subfacet));
4041 hmap_insert(&ofproto->subfacets, &subfacet->hmap_node, key_hash);
4042 list_push_back(&facet->subfacets, &subfacet->list_node);
4043 subfacet->facet = facet;
4044 subfacet->key_fitness = key_fitness;
4045 if (key_fitness != ODP_FIT_PERFECT) {
4046 subfacet->key = xmemdup(key, key_len);
4047 subfacet->key_len = key_len;
4049 subfacet->key = NULL;
4050 subfacet->key_len = 0;
4052 subfacet->used = time_msec();
4053 subfacet->dp_packet_count = 0;
4054 subfacet->dp_byte_count = 0;
4055 subfacet->actions_len = 0;
4056 subfacet->actions = NULL;
4057 subfacet->installed = false;
4058 subfacet->initial_tci = initial_tci;
4063 /* Searches 'ofproto' for a subfacet with the given 'key', 'key_len', and
4064 * 'flow'. Returns the subfacet if one exists, otherwise NULL. */
4065 static struct subfacet *
4066 subfacet_find(struct ofproto_dpif *ofproto,
4067 const struct nlattr *key, size_t key_len)
4069 uint32_t key_hash = odp_flow_key_hash(key, key_len);
4070 enum odp_key_fitness fitness;
4073 fitness = odp_flow_key_to_flow(key, key_len, &flow);
4074 if (fitness == ODP_FIT_ERROR) {
4078 return subfacet_find__(ofproto, key, key_len, key_hash, &flow);
4081 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
4082 * its facet within 'ofproto', and frees it. */
4084 subfacet_destroy__(struct subfacet *subfacet)
4086 struct facet *facet = subfacet->facet;
4087 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4089 subfacet_uninstall(subfacet);
4090 hmap_remove(&ofproto->subfacets, &subfacet->hmap_node);
4091 list_remove(&subfacet->list_node);
4092 free(subfacet->key);
4093 free(subfacet->actions);
4094 if (subfacet != &facet->one_subfacet) {
4099 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
4100 * last remaining subfacet in its facet destroys the facet too. */
4102 subfacet_destroy(struct subfacet *subfacet)
4104 struct facet *facet = subfacet->facet;
4106 if (list_is_singleton(&facet->subfacets)) {
4107 /* facet_remove() needs at least one subfacet (it will remove it). */
4108 facet_remove(facet);
4110 subfacet_destroy__(subfacet);
4114 /* Initializes 'key' with the sequence of OVS_KEY_ATTR_* Netlink attributes
4115 * that can be used to refer to 'subfacet'. The caller must provide 'keybuf'
4116 * for use as temporary storage. */
4118 subfacet_get_key(struct subfacet *subfacet, struct odputil_keybuf *keybuf,
4121 if (!subfacet->key) {
4122 ofpbuf_use_stack(key, keybuf, sizeof *keybuf);
4123 odp_flow_key_from_flow(key, &subfacet->facet->flow);
4125 ofpbuf_use_const(key, subfacet->key, subfacet->key_len);
4129 /* Composes the datapath actions for 'subfacet' based on its rule's actions.
4130 * Translates the actions into 'odp_actions', which the caller must have
4131 * initialized and is responsible for uninitializing. */
4133 subfacet_make_actions(struct subfacet *subfacet, const struct ofpbuf *packet,
4134 struct ofpbuf *odp_actions)
4136 struct facet *facet = subfacet->facet;
4137 struct rule_dpif *rule = facet->rule;
4138 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4140 struct action_xlate_ctx ctx;
4142 action_xlate_ctx_init(&ctx, ofproto, &facet->flow, subfacet->initial_tci,
4144 xlate_actions(&ctx, rule->up.actions, rule->up.n_actions, odp_actions);
4145 facet->tags = ctx.tags;
4146 facet->may_install = ctx.may_set_up_flow;
4147 facet->has_learn = ctx.has_learn;
4148 facet->has_normal = ctx.has_normal;
4149 facet->has_fin_timeout = ctx.has_fin_timeout;
4150 facet->nf_flow.output_iface = ctx.nf_output_iface;
4151 facet->mirrors = ctx.mirrors;
4153 if (subfacet->actions_len != odp_actions->size
4154 || memcmp(subfacet->actions, odp_actions->data, odp_actions->size)) {
4155 free(subfacet->actions);
4156 subfacet->actions_len = odp_actions->size;
4157 subfacet->actions = xmemdup(odp_actions->data, odp_actions->size);
4161 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
4162 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
4163 * in the datapath will be zeroed and 'stats' will be updated with traffic new
4164 * since 'subfacet' was last updated.
4166 * Returns 0 if successful, otherwise a positive errno value. */
4168 subfacet_install(struct subfacet *subfacet,
4169 const struct nlattr *actions, size_t actions_len,
4170 struct dpif_flow_stats *stats)
4172 struct facet *facet = subfacet->facet;
4173 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4174 struct odputil_keybuf keybuf;
4175 enum dpif_flow_put_flags flags;
4179 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
4181 flags |= DPIF_FP_ZERO_STATS;
4184 subfacet_get_key(subfacet, &keybuf, &key);
4185 ret = dpif_flow_put(ofproto->dpif, flags, key.data, key.size,
4186 actions, actions_len, stats);
4189 subfacet_reset_dp_stats(subfacet, stats);
4195 /* If 'subfacet' is installed in the datapath, uninstalls it. */
4197 subfacet_uninstall(struct subfacet *subfacet)
4199 if (subfacet->installed) {
4200 struct rule_dpif *rule = subfacet->facet->rule;
4201 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4202 struct odputil_keybuf keybuf;
4203 struct dpif_flow_stats stats;
4207 subfacet_get_key(subfacet, &keybuf, &key);
4208 error = dpif_flow_del(ofproto->dpif, key.data, key.size, &stats);
4209 subfacet_reset_dp_stats(subfacet, &stats);
4211 subfacet_update_stats(subfacet, &stats);
4213 subfacet->installed = false;
4215 assert(subfacet->dp_packet_count == 0);
4216 assert(subfacet->dp_byte_count == 0);
4220 /* Resets 'subfacet''s datapath statistics counters. This should be called
4221 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
4222 * non-null, it should contain the statistics returned by dpif when 'subfacet'
4223 * was reset in the datapath. 'stats' will be modified to include only
4224 * statistics new since 'subfacet' was last updated. */
4226 subfacet_reset_dp_stats(struct subfacet *subfacet,
4227 struct dpif_flow_stats *stats)
4230 && subfacet->dp_packet_count <= stats->n_packets
4231 && subfacet->dp_byte_count <= stats->n_bytes) {
4232 stats->n_packets -= subfacet->dp_packet_count;
4233 stats->n_bytes -= subfacet->dp_byte_count;
4236 subfacet->dp_packet_count = 0;
4237 subfacet->dp_byte_count = 0;
4240 /* Updates 'subfacet''s used time. The caller is responsible for calling
4241 * facet_push_stats() to update the flows which 'subfacet' resubmits into. */
4243 subfacet_update_time(struct subfacet *subfacet, long long int used)
4245 if (used > subfacet->used) {
4246 subfacet->used = used;
4247 facet_update_time(subfacet->facet, used);
4251 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
4253 * Because of the meaning of a subfacet's counters, it only makes sense to do
4254 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
4255 * represents a packet that was sent by hand or if it represents statistics
4256 * that have been cleared out of the datapath. */
4258 subfacet_update_stats(struct subfacet *subfacet,
4259 const struct dpif_flow_stats *stats)
4261 if (stats->n_packets || stats->used > subfacet->used) {
4262 struct facet *facet = subfacet->facet;
4264 subfacet_update_time(subfacet, stats->used);
4265 facet->packet_count += stats->n_packets;
4266 facet->byte_count += stats->n_bytes;
4267 facet->tcp_flags |= stats->tcp_flags;
4268 facet_push_stats(facet);
4269 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
4275 static struct rule_dpif *
4276 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow,
4279 struct cls_rule *cls_rule;
4280 struct classifier *cls;
4282 if (table_id >= N_TABLES) {
4286 cls = &ofproto->up.tables[table_id].cls;
4287 if (flow->nw_frag & FLOW_NW_FRAG_ANY
4288 && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
4289 /* For OFPC_NORMAL frag_handling, we must pretend that transport ports
4290 * are unavailable. */
4291 struct flow ofpc_normal_flow = *flow;
4292 ofpc_normal_flow.tp_src = htons(0);
4293 ofpc_normal_flow.tp_dst = htons(0);
4294 cls_rule = classifier_lookup(cls, &ofpc_normal_flow);
4296 cls_rule = classifier_lookup(cls, flow);
4298 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
4302 complete_operation(struct rule_dpif *rule)
4304 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4306 rule_invalidate(rule);
4308 struct dpif_completion *c = xmalloc(sizeof *c);
4309 c->op = rule->up.pending;
4310 list_push_back(&ofproto->completions, &c->list_node);
4312 ofoperation_complete(rule->up.pending, 0);
4316 static struct rule *
4319 struct rule_dpif *rule = xmalloc(sizeof *rule);
4324 rule_dealloc(struct rule *rule_)
4326 struct rule_dpif *rule = rule_dpif_cast(rule_);
4331 rule_construct(struct rule *rule_)
4333 struct rule_dpif *rule = rule_dpif_cast(rule_);
4334 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4335 struct rule_dpif *victim;
4339 error = validate_actions(rule->up.actions, rule->up.n_actions,
4340 &rule->up.cr.flow, ofproto->max_ports);
4345 rule->packet_count = 0;
4346 rule->byte_count = 0;
4348 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
4349 if (victim && !list_is_empty(&victim->facets)) {
4350 struct facet *facet;
4352 rule->facets = victim->facets;
4353 list_moved(&rule->facets);
4354 LIST_FOR_EACH (facet, list_node, &rule->facets) {
4355 /* XXX: We're only clearing our local counters here. It's possible
4356 * that quite a few packets are unaccounted for in the datapath
4357 * statistics. These will be accounted to the new rule instead of
4358 * cleared as required. This could be fixed by clearing out the
4359 * datapath statistics for this facet, but currently it doesn't
4361 facet_reset_counters(facet);
4365 /* Must avoid list_moved() in this case. */
4366 list_init(&rule->facets);
4369 table_id = rule->up.table_id;
4370 rule->tag = (victim ? victim->tag
4372 : rule_calculate_tag(&rule->up.cr.flow, &rule->up.cr.wc,
4373 ofproto->tables[table_id].basis));
4375 complete_operation(rule);
4380 rule_destruct(struct rule *rule_)
4382 struct rule_dpif *rule = rule_dpif_cast(rule_);
4383 struct facet *facet, *next_facet;
4385 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4386 facet_revalidate(facet);
4389 complete_operation(rule);
4393 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
4395 struct rule_dpif *rule = rule_dpif_cast(rule_);
4396 struct facet *facet;
4398 /* Start from historical data for 'rule' itself that are no longer tracked
4399 * in facets. This counts, for example, facets that have expired. */
4400 *packets = rule->packet_count;
4401 *bytes = rule->byte_count;
4403 /* Add any statistics that are tracked by facets. This includes
4404 * statistical data recently updated by ofproto_update_stats() as well as
4405 * stats for packets that were executed "by hand" via dpif_execute(). */
4406 LIST_FOR_EACH (facet, list_node, &rule->facets) {
4407 *packets += facet->packet_count;
4408 *bytes += facet->byte_count;
4413 rule_execute(struct rule *rule_, const struct flow *flow,
4414 struct ofpbuf *packet)
4416 struct rule_dpif *rule = rule_dpif_cast(rule_);
4417 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4419 struct dpif_flow_stats stats;
4421 struct action_xlate_ctx ctx;
4422 uint64_t odp_actions_stub[1024 / 8];
4423 struct ofpbuf odp_actions;
4425 dpif_flow_stats_extract(flow, packet, &stats);
4426 rule_credit_stats(rule, &stats);
4428 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
4429 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci,
4430 rule, stats.tcp_flags, packet);
4431 ctx.resubmit_stats = &stats;
4432 xlate_actions(&ctx, rule->up.actions, rule->up.n_actions, &odp_actions);
4434 execute_odp_actions(ofproto, flow, odp_actions.data,
4435 odp_actions.size, packet);
4437 ofpbuf_uninit(&odp_actions);
4443 rule_modify_actions(struct rule *rule_)
4445 struct rule_dpif *rule = rule_dpif_cast(rule_);
4446 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4449 error = validate_actions(rule->up.actions, rule->up.n_actions,
4450 &rule->up.cr.flow, ofproto->max_ports);
4452 ofoperation_complete(rule->up.pending, error);
4456 complete_operation(rule);
4459 /* Sends 'packet' out 'ofport'.
4460 * May modify 'packet'.
4461 * Returns 0 if successful, otherwise a positive errno value. */
4463 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
4465 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
4466 struct ofpbuf key, odp_actions;
4467 struct odputil_keybuf keybuf;
4472 flow_extract((struct ofpbuf *) packet, 0, 0, 0, &flow);
4473 odp_port = vsp_realdev_to_vlandev(ofproto, ofport->odp_port,
4475 if (odp_port != ofport->odp_port) {
4476 eth_pop_vlan(packet);
4477 flow.vlan_tci = htons(0);
4480 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4481 odp_flow_key_from_flow(&key, &flow);
4483 ofpbuf_init(&odp_actions, 32);
4484 compose_sflow_action(ofproto, &odp_actions, &flow, odp_port);
4486 nl_msg_put_u32(&odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port);
4487 error = dpif_execute(ofproto->dpif,
4489 odp_actions.data, odp_actions.size,
4491 ofpbuf_uninit(&odp_actions);
4494 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)",
4495 ofproto->up.name, odp_port, strerror(error));
4497 ofproto_update_local_port_stats(ofport->up.ofproto, packet->size, 0);
4501 /* OpenFlow to datapath action translation. */
4503 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
4504 struct action_xlate_ctx *ctx);
4505 static void xlate_normal(struct action_xlate_ctx *);
4508 put_userspace_action(const struct ofproto_dpif *ofproto,
4509 struct ofpbuf *odp_actions,
4510 const struct flow *flow,
4511 const struct user_action_cookie *cookie)
4515 pid = dpif_port_get_pid(ofproto->dpif,
4516 ofp_port_to_odp_port(flow->in_port));
4518 return odp_put_userspace_action(pid, cookie, odp_actions);
4521 /* Compose SAMPLE action for sFlow. */
4523 compose_sflow_action(const struct ofproto_dpif *ofproto,
4524 struct ofpbuf *odp_actions,
4525 const struct flow *flow,
4528 uint32_t port_ifindex;
4529 uint32_t probability;
4530 struct user_action_cookie cookie;
4531 size_t sample_offset, actions_offset;
4532 int cookie_offset, n_output;
4534 if (!ofproto->sflow || flow->in_port == OFPP_NONE) {
4538 if (odp_port == OVSP_NONE) {
4542 port_ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port);
4546 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
4548 /* Number of packets out of UINT_MAX to sample. */
4549 probability = dpif_sflow_get_probability(ofproto->sflow);
4550 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
4552 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
4554 cookie.type = USER_ACTION_COOKIE_SFLOW;
4555 cookie.data = port_ifindex;
4556 cookie.n_output = n_output;
4557 cookie.vlan_tci = 0;
4558 cookie_offset = put_userspace_action(ofproto, odp_actions, flow, &cookie);
4560 nl_msg_end_nested(odp_actions, actions_offset);
4561 nl_msg_end_nested(odp_actions, sample_offset);
4562 return cookie_offset;
4565 /* SAMPLE action must be first action in any given list of actions.
4566 * At this point we do not have all information required to build it. So try to
4567 * build sample action as complete as possible. */
4569 add_sflow_action(struct action_xlate_ctx *ctx)
4571 ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto,
4573 &ctx->flow, OVSP_NONE);
4574 ctx->sflow_odp_port = 0;
4575 ctx->sflow_n_outputs = 0;
4578 /* Fix SAMPLE action according to data collected while composing ODP actions.
4579 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
4580 * USERSPACE action's user-cookie which is required for sflow. */
4582 fix_sflow_action(struct action_xlate_ctx *ctx)
4584 const struct flow *base = &ctx->base_flow;
4585 struct user_action_cookie *cookie;
4587 if (!ctx->user_cookie_offset) {
4591 cookie = ofpbuf_at(ctx->odp_actions, ctx->user_cookie_offset,
4593 assert(cookie != NULL);
4594 assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
4596 if (ctx->sflow_n_outputs) {
4597 cookie->data = dpif_sflow_odp_port_to_ifindex(ctx->ofproto->sflow,
4598 ctx->sflow_odp_port);
4600 if (ctx->sflow_n_outputs >= 255) {
4601 cookie->n_output = 255;
4603 cookie->n_output = ctx->sflow_n_outputs;
4605 cookie->vlan_tci = base->vlan_tci;
4609 compose_output_action__(struct action_xlate_ctx *ctx, uint16_t ofp_port,
4612 const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
4613 uint16_t odp_port = ofp_port_to_odp_port(ofp_port);
4614 ovs_be16 flow_vlan_tci = ctx->flow.vlan_tci;
4615 uint8_t flow_nw_tos = ctx->flow.nw_tos;
4619 struct priority_to_dscp *pdscp;
4621 if (ofport->up.pp.config & OFPUTIL_PC_NO_FWD
4622 || (check_stp && !stp_forward_in_state(ofport->stp_state))) {
4626 pdscp = get_priority(ofport, ctx->flow.skb_priority);
4628 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
4629 ctx->flow.nw_tos |= pdscp->dscp;
4632 /* We may not have an ofport record for this port, but it doesn't hurt
4633 * to allow forwarding to it anyhow. Maybe such a port will appear
4634 * later and we're pre-populating the flow table. */
4637 out_port = vsp_realdev_to_vlandev(ctx->ofproto, odp_port,
4638 ctx->flow.vlan_tci);
4639 if (out_port != odp_port) {
4640 ctx->flow.vlan_tci = htons(0);
4642 commit_odp_actions(&ctx->flow, &ctx->base_flow, ctx->odp_actions);
4643 nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_OUTPUT, out_port);
4645 ctx->sflow_odp_port = odp_port;
4646 ctx->sflow_n_outputs++;
4647 ctx->nf_output_iface = ofp_port;
4648 ctx->flow.vlan_tci = flow_vlan_tci;
4649 ctx->flow.nw_tos = flow_nw_tos;
4653 compose_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port)
4655 compose_output_action__(ctx, ofp_port, true);
4659 xlate_table_action(struct action_xlate_ctx *ctx,
4660 uint16_t in_port, uint8_t table_id)
4662 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
4663 struct ofproto_dpif *ofproto = ctx->ofproto;
4664 struct rule_dpif *rule;
4665 uint16_t old_in_port;
4666 uint8_t old_table_id;
4668 old_table_id = ctx->table_id;
4669 ctx->table_id = table_id;
4671 /* Look up a flow with 'in_port' as the input port. */
4672 old_in_port = ctx->flow.in_port;
4673 ctx->flow.in_port = in_port;
4674 rule = rule_dpif_lookup(ofproto, &ctx->flow, table_id);
4677 if (table_id > 0 && table_id < N_TABLES) {
4678 struct table_dpif *table = &ofproto->tables[table_id];
4679 if (table->other_table) {
4680 ctx->tags |= (rule && rule->tag
4682 : rule_calculate_tag(&ctx->flow,
4683 &table->other_table->wc,
4688 /* Restore the original input port. Otherwise OFPP_NORMAL and
4689 * OFPP_IN_PORT will have surprising behavior. */
4690 ctx->flow.in_port = old_in_port;
4692 if (ctx->resubmit_hook) {
4693 ctx->resubmit_hook(ctx, rule);
4697 struct rule_dpif *old_rule = ctx->rule;
4699 if (ctx->resubmit_stats) {
4700 rule_credit_stats(rule, ctx->resubmit_stats);
4705 do_xlate_actions(rule->up.actions, rule->up.n_actions, ctx);
4706 ctx->rule = old_rule;
4710 ctx->table_id = old_table_id;
4712 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
4714 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
4715 MAX_RESUBMIT_RECURSION);
4716 ctx->max_resubmit_trigger = true;
4721 xlate_resubmit_table(struct action_xlate_ctx *ctx,
4722 const struct nx_action_resubmit *nar)
4727 in_port = (nar->in_port == htons(OFPP_IN_PORT)
4729 : ntohs(nar->in_port));
4730 table_id = nar->table == 255 ? ctx->table_id : nar->table;
4732 xlate_table_action(ctx, in_port, table_id);
4736 flood_packets(struct action_xlate_ctx *ctx, bool all)
4738 struct ofport_dpif *ofport;
4740 HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) {
4741 uint16_t ofp_port = ofport->up.ofp_port;
4743 if (ofp_port == ctx->flow.in_port) {
4748 compose_output_action__(ctx, ofp_port, false);
4749 } else if (!(ofport->up.pp.config & OFPUTIL_PC_NO_FLOOD)) {
4750 compose_output_action(ctx, ofp_port);
4754 ctx->nf_output_iface = NF_OUT_FLOOD;
4758 execute_controller_action(struct action_xlate_ctx *ctx, int len,
4759 enum ofp_packet_in_reason reason,
4760 uint16_t controller_id)
4762 struct ofputil_packet_in pin;
4763 struct ofpbuf *packet;
4765 ctx->may_set_up_flow = false;
4770 packet = ofpbuf_clone(ctx->packet);
4772 if (packet->l2 && packet->l3) {
4773 struct eth_header *eh;
4775 eth_pop_vlan(packet);
4778 /* If the Ethernet type is less than ETH_TYPE_MIN, it's likely an 802.2
4779 * LLC frame. Calculating the Ethernet type of these frames is more
4780 * trouble than seems appropriate for a simple assertion. */
4781 assert(ntohs(eh->eth_type) < ETH_TYPE_MIN
4782 || eh->eth_type == ctx->flow.dl_type);
4784 memcpy(eh->eth_src, ctx->flow.dl_src, sizeof eh->eth_src);
4785 memcpy(eh->eth_dst, ctx->flow.dl_dst, sizeof eh->eth_dst);
4787 if (ctx->flow.vlan_tci & htons(VLAN_CFI)) {
4788 eth_push_vlan(packet, ctx->flow.vlan_tci);
4792 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
4793 packet_set_ipv4(packet, ctx->flow.nw_src, ctx->flow.nw_dst,
4794 ctx->flow.nw_tos, ctx->flow.nw_ttl);
4798 if (ctx->flow.nw_proto == IPPROTO_TCP) {
4799 packet_set_tcp_port(packet, ctx->flow.tp_src,
4801 } else if (ctx->flow.nw_proto == IPPROTO_UDP) {
4802 packet_set_udp_port(packet, ctx->flow.tp_src,
4809 pin.packet = packet->data;
4810 pin.packet_len = packet->size;
4811 pin.reason = reason;
4812 pin.controller_id = controller_id;
4813 pin.table_id = ctx->table_id;
4814 pin.cookie = ctx->rule ? ctx->rule->up.flow_cookie : 0;
4817 flow_get_metadata(&ctx->flow, &pin.fmd);
4819 connmgr_send_packet_in(ctx->ofproto->up.connmgr, &pin);
4820 ofpbuf_delete(packet);
4824 compose_dec_ttl(struct action_xlate_ctx *ctx)
4826 if (ctx->flow.dl_type != htons(ETH_TYPE_IP) &&
4827 ctx->flow.dl_type != htons(ETH_TYPE_IPV6)) {
4831 if (ctx->flow.nw_ttl > 1) {
4835 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL, 0);
4837 /* Stop processing for current table. */
4843 xlate_output_action__(struct action_xlate_ctx *ctx,
4844 uint16_t port, uint16_t max_len)
4846 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
4848 ctx->nf_output_iface = NF_OUT_DROP;
4852 compose_output_action(ctx, ctx->flow.in_port);
4855 xlate_table_action(ctx, ctx->flow.in_port, ctx->table_id);
4861 flood_packets(ctx, false);
4864 flood_packets(ctx, true);
4866 case OFPP_CONTROLLER:
4867 execute_controller_action(ctx, max_len, OFPR_ACTION, 0);
4873 if (port != ctx->flow.in_port) {
4874 compose_output_action(ctx, port);
4879 if (prev_nf_output_iface == NF_OUT_FLOOD) {
4880 ctx->nf_output_iface = NF_OUT_FLOOD;
4881 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
4882 ctx->nf_output_iface = prev_nf_output_iface;
4883 } else if (prev_nf_output_iface != NF_OUT_DROP &&
4884 ctx->nf_output_iface != NF_OUT_FLOOD) {
4885 ctx->nf_output_iface = NF_OUT_MULTI;
4890 xlate_output_reg_action(struct action_xlate_ctx *ctx,
4891 const struct nx_action_output_reg *naor)
4893 struct mf_subfield src;
4896 nxm_decode(&src, naor->src, naor->ofs_nbits);
4897 ofp_port = mf_get_subfield(&src, &ctx->flow);
4899 if (ofp_port <= UINT16_MAX) {
4900 xlate_output_action__(ctx, ofp_port, ntohs(naor->max_len));
4905 xlate_output_action(struct action_xlate_ctx *ctx,
4906 const struct ofp_action_output *oao)
4908 xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len));
4912 xlate_enqueue_action(struct action_xlate_ctx *ctx,
4913 const struct ofp_action_enqueue *oae)
4916 uint32_t flow_priority, priority;
4919 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id),
4922 /* Fall back to ordinary output action. */
4923 xlate_output_action__(ctx, ntohs(oae->port), 0);
4927 /* Figure out datapath output port. */
4928 ofp_port = ntohs(oae->port);
4929 if (ofp_port == OFPP_IN_PORT) {
4930 ofp_port = ctx->flow.in_port;
4931 } else if (ofp_port == ctx->flow.in_port) {
4935 /* Add datapath actions. */
4936 flow_priority = ctx->flow.skb_priority;
4937 ctx->flow.skb_priority = priority;
4938 compose_output_action(ctx, ofp_port);
4939 ctx->flow.skb_priority = flow_priority;
4941 /* Update NetFlow output port. */
4942 if (ctx->nf_output_iface == NF_OUT_DROP) {
4943 ctx->nf_output_iface = ofp_port;
4944 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
4945 ctx->nf_output_iface = NF_OUT_MULTI;
4950 xlate_set_queue_action(struct action_xlate_ctx *ctx,
4951 const struct nx_action_set_queue *nasq)
4956 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(nasq->queue_id),
4959 /* Couldn't translate queue to a priority, so ignore. A warning
4960 * has already been logged. */
4964 ctx->flow.skb_priority = priority;
4967 struct xlate_reg_state {
4973 xlate_autopath(struct action_xlate_ctx *ctx,
4974 const struct nx_action_autopath *naa)
4976 uint16_t ofp_port = ntohl(naa->id);
4977 struct ofport_dpif *port = get_ofp_port(ctx->ofproto, ofp_port);
4979 if (!port || !port->bundle) {
4980 ofp_port = OFPP_NONE;
4981 } else if (port->bundle->bond) {
4982 /* Autopath does not support VLAN hashing. */
4983 struct ofport_dpif *slave = bond_choose_output_slave(
4984 port->bundle->bond, &ctx->flow, 0, &ctx->tags);
4986 ofp_port = slave->up.ofp_port;
4989 autopath_execute(naa, &ctx->flow, ofp_port);
4993 slave_enabled_cb(uint16_t ofp_port, void *ofproto_)
4995 struct ofproto_dpif *ofproto = ofproto_;
4996 struct ofport_dpif *port;
5006 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
5009 port = get_ofp_port(ofproto, ofp_port);
5010 return port ? port->may_enable : false;
5015 xlate_learn_action(struct action_xlate_ctx *ctx,
5016 const struct nx_action_learn *learn)
5018 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
5019 struct ofputil_flow_mod fm;
5022 learn_execute(learn, &ctx->flow, &fm);
5024 error = ofproto_flow_mod(&ctx->ofproto->up, &fm);
5025 if (error && !VLOG_DROP_WARN(&rl)) {
5026 VLOG_WARN("learning action failed to modify flow table (%s)",
5027 ofperr_get_name(error));
5033 /* Reduces '*timeout' to no more than 'max'. A value of zero in either case
5034 * means "infinite". */
5036 reduce_timeout(uint16_t max, uint16_t *timeout)
5038 if (max && (!*timeout || *timeout > max)) {
5044 xlate_fin_timeout(struct action_xlate_ctx *ctx,
5045 const struct nx_action_fin_timeout *naft)
5047 if (ctx->tcp_flags & (TCP_FIN | TCP_RST) && ctx->rule) {
5048 struct rule_dpif *rule = ctx->rule;
5050 reduce_timeout(ntohs(naft->fin_idle_timeout), &rule->up.idle_timeout);
5051 reduce_timeout(ntohs(naft->fin_hard_timeout), &rule->up.hard_timeout);
5056 may_receive(const struct ofport_dpif *port, struct action_xlate_ctx *ctx)
5058 if (port->up.pp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
5059 ? OFPUTIL_PC_NO_RECV_STP
5060 : OFPUTIL_PC_NO_RECV)) {
5064 /* Only drop packets here if both forwarding and learning are
5065 * disabled. If just learning is enabled, we need to have
5066 * OFPP_NORMAL and the learning action have a look at the packet
5067 * before we can drop it. */
5068 if (!stp_forward_in_state(port->stp_state)
5069 && !stp_learn_in_state(port->stp_state)) {
5077 do_xlate_actions(const union ofp_action *in, size_t n_in,
5078 struct action_xlate_ctx *ctx)
5080 const struct ofport_dpif *port;
5081 const union ofp_action *ia;
5082 bool was_evictable = true;
5085 port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
5086 if (port && !may_receive(port, ctx)) {
5087 /* Drop this flow. */
5092 /* Don't let the rule we're working on get evicted underneath us. */
5093 was_evictable = ctx->rule->up.evictable;
5094 ctx->rule->up.evictable = false;
5096 OFPUTIL_ACTION_FOR_EACH_UNSAFE (ia, left, in, n_in) {
5097 const struct ofp_action_dl_addr *oada;
5098 const struct nx_action_resubmit *nar;
5099 const struct nx_action_set_tunnel *nast;
5100 const struct nx_action_set_queue *nasq;
5101 const struct nx_action_multipath *nam;
5102 const struct nx_action_autopath *naa;
5103 const struct nx_action_bundle *nab;
5104 const struct nx_action_output_reg *naor;
5105 const struct nx_action_controller *nac;
5106 enum ofputil_action_code code;
5113 code = ofputil_decode_action_unsafe(ia);
5115 case OFPUTIL_OFPAT10_OUTPUT:
5116 xlate_output_action(ctx, &ia->output);
5119 case OFPUTIL_OFPAT10_SET_VLAN_VID:
5120 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
5121 ctx->flow.vlan_tci |= ia->vlan_vid.vlan_vid | htons(VLAN_CFI);
5124 case OFPUTIL_OFPAT10_SET_VLAN_PCP:
5125 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
5126 ctx->flow.vlan_tci |= htons(
5127 (ia->vlan_pcp.vlan_pcp << VLAN_PCP_SHIFT) | VLAN_CFI);
5130 case OFPUTIL_OFPAT10_STRIP_VLAN:
5131 ctx->flow.vlan_tci = htons(0);
5134 case OFPUTIL_OFPAT10_SET_DL_SRC:
5135 oada = ((struct ofp_action_dl_addr *) ia);
5136 memcpy(ctx->flow.dl_src, oada->dl_addr, ETH_ADDR_LEN);
5139 case OFPUTIL_OFPAT10_SET_DL_DST:
5140 oada = ((struct ofp_action_dl_addr *) ia);
5141 memcpy(ctx->flow.dl_dst, oada->dl_addr, ETH_ADDR_LEN);
5144 case OFPUTIL_OFPAT10_SET_NW_SRC:
5145 ctx->flow.nw_src = ia->nw_addr.nw_addr;
5148 case OFPUTIL_OFPAT10_SET_NW_DST:
5149 ctx->flow.nw_dst = ia->nw_addr.nw_addr;
5152 case OFPUTIL_OFPAT10_SET_NW_TOS:
5153 /* OpenFlow 1.0 only supports IPv4. */
5154 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
5155 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
5156 ctx->flow.nw_tos |= ia->nw_tos.nw_tos & IP_DSCP_MASK;
5160 case OFPUTIL_OFPAT10_SET_TP_SRC:
5161 ctx->flow.tp_src = ia->tp_port.tp_port;
5164 case OFPUTIL_OFPAT10_SET_TP_DST:
5165 ctx->flow.tp_dst = ia->tp_port.tp_port;
5168 case OFPUTIL_OFPAT10_ENQUEUE:
5169 xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia);
5172 case OFPUTIL_NXAST_RESUBMIT:
5173 nar = (const struct nx_action_resubmit *) ia;
5174 xlate_table_action(ctx, ntohs(nar->in_port), ctx->table_id);
5177 case OFPUTIL_NXAST_RESUBMIT_TABLE:
5178 xlate_resubmit_table(ctx, (const struct nx_action_resubmit *) ia);
5181 case OFPUTIL_NXAST_SET_TUNNEL:
5182 nast = (const struct nx_action_set_tunnel *) ia;
5183 tun_id = htonll(ntohl(nast->tun_id));
5184 ctx->flow.tun_id = tun_id;
5187 case OFPUTIL_NXAST_SET_QUEUE:
5188 nasq = (const struct nx_action_set_queue *) ia;
5189 xlate_set_queue_action(ctx, nasq);
5192 case OFPUTIL_NXAST_POP_QUEUE:
5193 ctx->flow.skb_priority = ctx->orig_skb_priority;
5196 case OFPUTIL_NXAST_REG_MOVE:
5197 nxm_execute_reg_move((const struct nx_action_reg_move *) ia,
5201 case OFPUTIL_NXAST_REG_LOAD:
5202 nxm_execute_reg_load((const struct nx_action_reg_load *) ia,
5206 case OFPUTIL_NXAST_NOTE:
5207 /* Nothing to do. */
5210 case OFPUTIL_NXAST_SET_TUNNEL64:
5211 tun_id = ((const struct nx_action_set_tunnel64 *) ia)->tun_id;
5212 ctx->flow.tun_id = tun_id;
5215 case OFPUTIL_NXAST_MULTIPATH:
5216 nam = (const struct nx_action_multipath *) ia;
5217 multipath_execute(nam, &ctx->flow);
5220 case OFPUTIL_NXAST_AUTOPATH:
5221 naa = (const struct nx_action_autopath *) ia;
5222 xlate_autopath(ctx, naa);
5225 case OFPUTIL_NXAST_BUNDLE:
5226 ctx->ofproto->has_bundle_action = true;
5227 nab = (const struct nx_action_bundle *) ia;
5228 xlate_output_action__(ctx, bundle_execute(nab, &ctx->flow,
5233 case OFPUTIL_NXAST_BUNDLE_LOAD:
5234 ctx->ofproto->has_bundle_action = true;
5235 nab = (const struct nx_action_bundle *) ia;
5236 bundle_execute_load(nab, &ctx->flow, slave_enabled_cb,
5240 case OFPUTIL_NXAST_OUTPUT_REG:
5241 naor = (const struct nx_action_output_reg *) ia;
5242 xlate_output_reg_action(ctx, naor);
5245 case OFPUTIL_NXAST_LEARN:
5246 ctx->has_learn = true;
5247 if (ctx->may_learn) {
5248 xlate_learn_action(ctx, (const struct nx_action_learn *) ia);
5252 case OFPUTIL_NXAST_DEC_TTL:
5253 if (compose_dec_ttl(ctx)) {
5258 case OFPUTIL_NXAST_EXIT:
5262 case OFPUTIL_NXAST_FIN_TIMEOUT:
5263 ctx->has_fin_timeout = true;
5264 xlate_fin_timeout(ctx, (const struct nx_action_fin_timeout *) ia);
5267 case OFPUTIL_NXAST_CONTROLLER:
5268 nac = (const struct nx_action_controller *) ia;
5269 execute_controller_action(ctx, ntohs(nac->max_len), nac->reason,
5270 ntohs(nac->controller_id));
5276 /* We've let OFPP_NORMAL and the learning action look at the packet,
5277 * so drop it now if forwarding is disabled. */
5278 if (port && !stp_forward_in_state(port->stp_state)) {
5279 ofpbuf_clear(ctx->odp_actions);
5280 add_sflow_action(ctx);
5283 ctx->rule->up.evictable = was_evictable;
5288 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
5289 struct ofproto_dpif *ofproto, const struct flow *flow,
5290 ovs_be16 initial_tci, struct rule_dpif *rule,
5291 uint8_t tcp_flags, const struct ofpbuf *packet)
5293 ctx->ofproto = ofproto;
5295 ctx->base_flow = ctx->flow;
5296 ctx->base_flow.tun_id = 0;
5297 ctx->base_flow.vlan_tci = initial_tci;
5299 ctx->packet = packet;
5300 ctx->may_learn = packet != NULL;
5301 ctx->tcp_flags = tcp_flags;
5302 ctx->resubmit_hook = NULL;
5303 ctx->resubmit_stats = NULL;
5306 /* Translates the 'n_in' "union ofp_action"s in 'in' into datapath actions in
5307 * 'odp_actions', using 'ctx'. */
5309 xlate_actions(struct action_xlate_ctx *ctx,
5310 const union ofp_action *in, size_t n_in,
5311 struct ofpbuf *odp_actions)
5313 /* Normally false. Set to true if we ever hit MAX_RESUBMIT_RECURSION, so
5314 * that in the future we always keep a copy of the original flow for
5315 * tracing purposes. */
5316 static bool hit_resubmit_limit;
5318 COVERAGE_INC(ofproto_dpif_xlate);
5320 ofpbuf_clear(odp_actions);
5321 ofpbuf_reserve(odp_actions, NL_A_U32_SIZE);
5323 ctx->odp_actions = odp_actions;
5325 ctx->may_set_up_flow = true;
5326 ctx->has_learn = false;
5327 ctx->has_normal = false;
5328 ctx->has_fin_timeout = false;
5329 ctx->nf_output_iface = NF_OUT_DROP;
5332 ctx->max_resubmit_trigger = false;
5333 ctx->orig_skb_priority = ctx->flow.skb_priority;
5337 if (ctx->ofproto->has_mirrors || hit_resubmit_limit) {
5338 /* Do this conditionally because the copy is expensive enough that it
5339 * shows up in profiles.
5341 * We keep orig_flow in 'ctx' only because I couldn't make GCC 4.4
5342 * believe that I wasn't using it without initializing it if I kept it
5343 * in a local variable. */
5344 ctx->orig_flow = ctx->flow;
5347 if (ctx->flow.nw_frag & FLOW_NW_FRAG_ANY) {
5348 switch (ctx->ofproto->up.frag_handling) {
5349 case OFPC_FRAG_NORMAL:
5350 /* We must pretend that transport ports are unavailable. */
5351 ctx->flow.tp_src = ctx->base_flow.tp_src = htons(0);
5352 ctx->flow.tp_dst = ctx->base_flow.tp_dst = htons(0);
5355 case OFPC_FRAG_DROP:
5358 case OFPC_FRAG_REASM:
5361 case OFPC_FRAG_NX_MATCH:
5362 /* Nothing to do. */
5365 case OFPC_INVALID_TTL_TO_CONTROLLER:
5370 if (process_special(ctx->ofproto, &ctx->flow, ctx->packet)) {
5371 ctx->may_set_up_flow = false;
5373 static struct vlog_rate_limit trace_rl = VLOG_RATE_LIMIT_INIT(1, 1);
5374 ovs_be16 initial_tci = ctx->base_flow.vlan_tci;
5376 add_sflow_action(ctx);
5377 do_xlate_actions(in, n_in, ctx);
5379 if (ctx->max_resubmit_trigger && !ctx->resubmit_hook) {
5380 if (!hit_resubmit_limit) {
5381 /* We didn't record the original flow. Make sure we do from
5383 hit_resubmit_limit = true;
5384 } else if (!VLOG_DROP_ERR(&trace_rl)) {
5385 struct ds ds = DS_EMPTY_INITIALIZER;
5387 ofproto_trace(ctx->ofproto, &ctx->orig_flow, ctx->packet,
5389 VLOG_ERR("Trace triggered by excessive resubmit "
5390 "recursion:\n%s", ds_cstr(&ds));
5395 if (!connmgr_may_set_up_flow(ctx->ofproto->up.connmgr, &ctx->flow,
5396 ctx->odp_actions->data,
5397 ctx->odp_actions->size)) {
5398 ctx->may_set_up_flow = false;
5400 && connmgr_msg_in_hook(ctx->ofproto->up.connmgr, &ctx->flow,
5402 compose_output_action(ctx, OFPP_LOCAL);
5405 if (ctx->ofproto->has_mirrors) {
5406 add_mirror_actions(ctx, &ctx->orig_flow);
5408 fix_sflow_action(ctx);
5412 /* Translates the 'n_in' "union ofp_action"s in 'in' into datapath actions,
5413 * using 'ctx', and discards the datapath actions. */
5415 xlate_actions_for_side_effects(struct action_xlate_ctx *ctx,
5416 const union ofp_action *in, size_t n_in)
5418 uint64_t odp_actions_stub[1024 / 8];
5419 struct ofpbuf odp_actions;
5421 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5422 xlate_actions(ctx, in, n_in, &odp_actions);
5423 ofpbuf_uninit(&odp_actions);
5426 /* OFPP_NORMAL implementation. */
5428 static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *);
5430 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
5431 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle',
5432 * the bundle on which the packet was received, returns the VLAN to which the
5435 * Both 'vid' and the return value are in the range 0...4095. */
5437 input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid)
5439 switch (in_bundle->vlan_mode) {
5440 case PORT_VLAN_ACCESS:
5441 return in_bundle->vlan;
5444 case PORT_VLAN_TRUNK:
5447 case PORT_VLAN_NATIVE_UNTAGGED:
5448 case PORT_VLAN_NATIVE_TAGGED:
5449 return vid ? vid : in_bundle->vlan;
5456 /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'.
5457 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
5460 * 'vid' should be the VID obtained from the 802.1Q header that was received as
5461 * part of a packet (specify 0 if there was no 802.1Q header), in the range
5464 input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn)
5466 /* Allow any VID on the OFPP_NONE port. */
5467 if (in_bundle == &ofpp_none_bundle) {
5471 switch (in_bundle->vlan_mode) {
5472 case PORT_VLAN_ACCESS:
5475 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5476 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
5477 "packet received on port %s configured as VLAN "
5478 "%"PRIu16" access port",
5479 in_bundle->ofproto->up.name, vid,
5480 in_bundle->name, in_bundle->vlan);
5486 case PORT_VLAN_NATIVE_UNTAGGED:
5487 case PORT_VLAN_NATIVE_TAGGED:
5489 /* Port must always carry its native VLAN. */
5493 case PORT_VLAN_TRUNK:
5494 if (!ofbundle_includes_vlan(in_bundle, vid)) {
5496 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5497 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet "
5498 "received on port %s not configured for trunking "
5500 in_bundle->ofproto->up.name, vid,
5501 in_bundle->name, vid);
5513 /* Given 'vlan', the VLAN that a packet belongs to, and
5514 * 'out_bundle', a bundle on which the packet is to be output, returns the VID
5515 * that should be included in the 802.1Q header. (If the return value is 0,
5516 * then the 802.1Q header should only be included in the packet if there is a
5519 * Both 'vlan' and the return value are in the range 0...4095. */
5521 output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan)
5523 switch (out_bundle->vlan_mode) {
5524 case PORT_VLAN_ACCESS:
5527 case PORT_VLAN_TRUNK:
5528 case PORT_VLAN_NATIVE_TAGGED:
5531 case PORT_VLAN_NATIVE_UNTAGGED:
5532 return vlan == out_bundle->vlan ? 0 : vlan;
5540 output_normal(struct action_xlate_ctx *ctx, const struct ofbundle *out_bundle,
5543 struct ofport_dpif *port;
5545 ovs_be16 tci, old_tci;
5547 vid = output_vlan_to_vid(out_bundle, vlan);
5548 if (!out_bundle->bond) {
5549 port = ofbundle_get_a_port(out_bundle);
5551 port = bond_choose_output_slave(out_bundle->bond, &ctx->flow,
5554 /* No slaves enabled, so drop packet. */
5559 old_tci = ctx->flow.vlan_tci;
5561 if (tci || out_bundle->use_priority_tags) {
5562 tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK);
5564 tci |= htons(VLAN_CFI);
5567 ctx->flow.vlan_tci = tci;
5569 compose_output_action(ctx, port->up.ofp_port);
5570 ctx->flow.vlan_tci = old_tci;
5574 mirror_mask_ffs(mirror_mask_t mask)
5576 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
5581 ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
5583 return (bundle->vlan_mode != PORT_VLAN_ACCESS
5584 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
5588 ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan)
5590 return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan);
5593 /* Returns an arbitrary interface within 'bundle'. */
5594 static struct ofport_dpif *
5595 ofbundle_get_a_port(const struct ofbundle *bundle)
5597 return CONTAINER_OF(list_front(&bundle->ports),
5598 struct ofport_dpif, bundle_node);
5602 vlan_is_mirrored(const struct ofmirror *m, int vlan)
5604 return !m->vlans || bitmap_is_set(m->vlans, vlan);
5607 /* Returns true if a packet with Ethernet destination MAC 'dst' may be mirrored
5608 * to a VLAN. In general most packets may be mirrored but we want to drop
5609 * protocols that may confuse switches. */
5611 eth_dst_may_rspan(const uint8_t dst[ETH_ADDR_LEN])
5613 /* If you change this function's behavior, please update corresponding
5614 * documentation in vswitch.xml at the same time. */
5615 if (dst[0] != 0x01) {
5616 /* All the currently banned MACs happen to start with 01 currently, so
5617 * this is a quick way to eliminate most of the good ones. */
5619 if (eth_addr_is_reserved(dst)) {
5620 /* Drop STP, IEEE pause frames, and other reserved protocols
5621 * (01-80-c2-00-00-0x). */
5625 if (dst[0] == 0x01 && dst[1] == 0x00 && dst[2] == 0x0c) {
5627 if ((dst[3] & 0xfe) == 0xcc &&
5628 (dst[4] & 0xfe) == 0xcc &&
5629 (dst[5] & 0xfe) == 0xcc) {
5630 /* Drop the following protocols plus others following the same
5633 CDP, VTP, DTP, PAgP (01-00-0c-cc-cc-cc)
5634 Spanning Tree PVSTP+ (01-00-0c-cc-cc-cd)
5635 STP Uplink Fast (01-00-0c-cd-cd-cd) */
5639 if (!(dst[3] | dst[4] | dst[5])) {
5640 /* Drop Inter Switch Link packets (01-00-0c-00-00-00). */
5649 add_mirror_actions(struct action_xlate_ctx *ctx, const struct flow *orig_flow)
5651 struct ofproto_dpif *ofproto = ctx->ofproto;
5652 mirror_mask_t mirrors;
5653 struct ofbundle *in_bundle;
5656 const struct nlattr *a;
5659 in_bundle = lookup_input_bundle(ctx->ofproto, orig_flow->in_port,
5660 ctx->packet != NULL, NULL);
5664 mirrors = in_bundle->src_mirrors;
5666 /* Drop frames on bundles reserved for mirroring. */
5667 if (in_bundle->mirror_out) {
5668 if (ctx->packet != NULL) {
5669 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5670 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
5671 "%s, which is reserved exclusively for mirroring",
5672 ctx->ofproto->up.name, in_bundle->name);
5678 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
5679 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
5682 vlan = input_vid_to_vlan(in_bundle, vid);
5684 /* Look at the output ports to check for destination selections. */
5686 NL_ATTR_FOR_EACH (a, left, ctx->odp_actions->data,
5687 ctx->odp_actions->size) {
5688 enum ovs_action_attr type = nl_attr_type(a);
5689 struct ofport_dpif *ofport;
5691 if (type != OVS_ACTION_ATTR_OUTPUT) {
5695 ofport = get_odp_port(ofproto, nl_attr_get_u32(a));
5696 if (ofport && ofport->bundle) {
5697 mirrors |= ofport->bundle->dst_mirrors;
5705 /* Restore the original packet before adding the mirror actions. */
5706 ctx->flow = *orig_flow;
5711 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
5713 if (!vlan_is_mirrored(m, vlan)) {
5714 mirrors &= mirrors - 1;
5718 mirrors &= ~m->dup_mirrors;
5719 ctx->mirrors |= m->dup_mirrors;
5721 output_normal(ctx, m->out, vlan);
5722 } else if (eth_dst_may_rspan(orig_flow->dl_dst)
5723 && vlan != m->out_vlan) {
5724 struct ofbundle *bundle;
5726 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
5727 if (ofbundle_includes_vlan(bundle, m->out_vlan)
5728 && !bundle->mirror_out) {
5729 output_normal(ctx, bundle, m->out_vlan);
5737 update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors,
5738 uint64_t packets, uint64_t bytes)
5744 for (; mirrors; mirrors &= mirrors - 1) {
5747 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
5750 /* In normal circumstances 'm' will not be NULL. However,
5751 * if mirrors are reconfigured, we can temporarily get out
5752 * of sync in facet_revalidate(). We could "correct" the
5753 * mirror list before reaching here, but doing that would
5754 * not properly account the traffic stats we've currently
5755 * accumulated for previous mirror configuration. */
5759 m->packet_count += packets;
5760 m->byte_count += bytes;
5764 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
5765 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
5766 * indicate this; newer upstream kernels use gratuitous ARP requests. */
5768 is_gratuitous_arp(const struct flow *flow)
5770 return (flow->dl_type == htons(ETH_TYPE_ARP)
5771 && eth_addr_is_broadcast(flow->dl_dst)
5772 && (flow->nw_proto == ARP_OP_REPLY
5773 || (flow->nw_proto == ARP_OP_REQUEST
5774 && flow->nw_src == flow->nw_dst)));
5778 update_learning_table(struct ofproto_dpif *ofproto,
5779 const struct flow *flow, int vlan,
5780 struct ofbundle *in_bundle)
5782 struct mac_entry *mac;
5784 /* Don't learn the OFPP_NONE port. */
5785 if (in_bundle == &ofpp_none_bundle) {
5789 if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
5793 mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan);
5794 if (is_gratuitous_arp(flow)) {
5795 /* We don't want to learn from gratuitous ARP packets that are
5796 * reflected back over bond slaves so we lock the learning table. */
5797 if (!in_bundle->bond) {
5798 mac_entry_set_grat_arp_lock(mac);
5799 } else if (mac_entry_is_grat_arp_locked(mac)) {
5804 if (mac_entry_is_new(mac) || mac->port.p != in_bundle) {
5805 /* The log messages here could actually be useful in debugging,
5806 * so keep the rate limit relatively high. */
5807 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
5808 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
5809 "on port %s in VLAN %d",
5810 ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src),
5811 in_bundle->name, vlan);
5813 mac->port.p = in_bundle;
5814 tag_set_add(&ofproto->revalidate_set,
5815 mac_learning_changed(ofproto->ml, mac));
5819 static struct ofbundle *
5820 lookup_input_bundle(struct ofproto_dpif *ofproto, uint16_t in_port, bool warn,
5821 struct ofport_dpif **in_ofportp)
5823 struct ofport_dpif *ofport;
5825 /* Find the port and bundle for the received packet. */
5826 ofport = get_ofp_port(ofproto, in_port);
5828 *in_ofportp = ofport;
5830 if (ofport && ofport->bundle) {
5831 return ofport->bundle;
5834 /* Special-case OFPP_NONE, which a controller may use as the ingress
5835 * port for traffic that it is sourcing. */
5836 if (in_port == OFPP_NONE) {
5837 return &ofpp_none_bundle;
5840 /* Odd. A few possible reasons here:
5842 * - We deleted a port but there are still a few packets queued up
5845 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
5846 * we don't know about.
5848 * - The ofproto client didn't configure the port as part of a bundle.
5851 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5853 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
5854 "port %"PRIu16, ofproto->up.name, in_port);
5859 /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or
5860 * dropped. Returns true if they may be forwarded, false if they should be
5863 * 'in_port' must be the ofport_dpif that corresponds to flow->in_port.
5864 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
5866 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
5867 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
5868 * checked by input_vid_is_valid().
5870 * May also add tags to '*tags', although the current implementation only does
5871 * so in one special case.
5874 is_admissible(struct ofproto_dpif *ofproto, const struct flow *flow,
5875 struct ofport_dpif *in_port, uint16_t vlan, tag_type *tags)
5877 struct ofbundle *in_bundle = in_port->bundle;
5879 /* Drop frames for reserved multicast addresses
5880 * only if forward_bpdu option is absent. */
5881 if (eth_addr_is_reserved(flow->dl_dst) && !ofproto->up.forward_bpdu) {
5885 if (in_bundle->bond) {
5886 struct mac_entry *mac;
5888 switch (bond_check_admissibility(in_bundle->bond, in_port,
5889 flow->dl_dst, tags)) {
5896 case BV_DROP_IF_MOVED:
5897 mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL);
5898 if (mac && mac->port.p != in_bundle &&
5899 (!is_gratuitous_arp(flow)
5900 || mac_entry_is_grat_arp_locked(mac))) {
5911 xlate_normal(struct action_xlate_ctx *ctx)
5913 struct ofport_dpif *in_port;
5914 struct ofbundle *in_bundle;
5915 struct mac_entry *mac;
5919 ctx->has_normal = true;
5921 in_bundle = lookup_input_bundle(ctx->ofproto, ctx->flow.in_port,
5922 ctx->packet != NULL, &in_port);
5927 /* Drop malformed frames. */
5928 if (ctx->flow.dl_type == htons(ETH_TYPE_VLAN) &&
5929 !(ctx->flow.vlan_tci & htons(VLAN_CFI))) {
5930 if (ctx->packet != NULL) {
5931 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5932 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
5933 "VLAN tag received on port %s",
5934 ctx->ofproto->up.name, in_bundle->name);
5939 /* Drop frames on bundles reserved for mirroring. */
5940 if (in_bundle->mirror_out) {
5941 if (ctx->packet != NULL) {
5942 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5943 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
5944 "%s, which is reserved exclusively for mirroring",
5945 ctx->ofproto->up.name, in_bundle->name);
5951 vid = vlan_tci_to_vid(ctx->flow.vlan_tci);
5952 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
5955 vlan = input_vid_to_vlan(in_bundle, vid);
5957 /* Check other admissibility requirements. */
5959 !is_admissible(ctx->ofproto, &ctx->flow, in_port, vlan, &ctx->tags)) {
5963 /* Learn source MAC. */
5964 if (ctx->may_learn) {
5965 update_learning_table(ctx->ofproto, &ctx->flow, vlan, in_bundle);
5968 /* Determine output bundle. */
5969 mac = mac_learning_lookup(ctx->ofproto->ml, ctx->flow.dl_dst, vlan,
5972 if (mac->port.p != in_bundle) {
5973 output_normal(ctx, mac->port.p, vlan);
5976 struct ofbundle *bundle;
5978 HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
5979 if (bundle != in_bundle
5980 && ofbundle_includes_vlan(bundle, vlan)
5981 && bundle->floodable
5982 && !bundle->mirror_out) {
5983 output_normal(ctx, bundle, vlan);
5986 ctx->nf_output_iface = NF_OUT_FLOOD;
5990 /* Optimized flow revalidation.
5992 * It's a difficult problem, in general, to tell which facets need to have
5993 * their actions recalculated whenever the OpenFlow flow table changes. We
5994 * don't try to solve that general problem: for most kinds of OpenFlow flow
5995 * table changes, we recalculate the actions for every facet. This is
5996 * relatively expensive, but it's good enough if the OpenFlow flow table
5997 * doesn't change very often.
5999 * However, we can expect one particular kind of OpenFlow flow table change to
6000 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
6001 * of CPU on revalidating every facet whenever MAC learning modifies the flow
6002 * table, we add a special case that applies to flow tables in which every rule
6003 * has the same form (that is, the same wildcards), except that the table is
6004 * also allowed to have a single "catch-all" flow that matches all packets. We
6005 * optimize this case by tagging all of the facets that resubmit into the table
6006 * and invalidating the same tag whenever a flow changes in that table. The
6007 * end result is that we revalidate just the facets that need it (and sometimes
6008 * a few more, but not all of the facets or even all of the facets that
6009 * resubmit to the table modified by MAC learning). */
6011 /* Calculates the tag to use for 'flow' and wildcards 'wc' when it is inserted
6012 * into an OpenFlow table with the given 'basis'. */
6014 rule_calculate_tag(const struct flow *flow, const struct flow_wildcards *wc,
6017 if (flow_wildcards_is_catchall(wc)) {
6020 struct flow tag_flow = *flow;
6021 flow_zero_wildcards(&tag_flow, wc);
6022 return tag_create_deterministic(flow_hash(&tag_flow, secret));
6026 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
6027 * taggability of that table.
6029 * This function must be called after *each* change to a flow table. If you
6030 * skip calling it on some changes then the pointer comparisons at the end can
6031 * be invalid if you get unlucky. For example, if a flow removal causes a
6032 * cls_table to be destroyed and then a flow insertion causes a cls_table with
6033 * different wildcards to be created with the same address, then this function
6034 * will incorrectly skip revalidation. */
6036 table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
6038 struct table_dpif *table = &ofproto->tables[table_id];
6039 const struct oftable *oftable = &ofproto->up.tables[table_id];
6040 struct cls_table *catchall, *other;
6041 struct cls_table *t;
6043 catchall = other = NULL;
6045 switch (hmap_count(&oftable->cls.tables)) {
6047 /* We could tag this OpenFlow table but it would make the logic a
6048 * little harder and it's a corner case that doesn't seem worth it
6054 HMAP_FOR_EACH (t, hmap_node, &oftable->cls.tables) {
6055 if (cls_table_is_catchall(t)) {
6057 } else if (!other) {
6060 /* Indicate that we can't tag this by setting both tables to
6061 * NULL. (We know that 'catchall' is already NULL.) */
6068 /* Can't tag this table. */
6072 if (table->catchall_table != catchall || table->other_table != other) {
6073 table->catchall_table = catchall;
6074 table->other_table = other;
6075 ofproto->need_revalidate = true;
6079 /* Given 'rule' that has changed in some way (either it is a rule being
6080 * inserted, a rule being deleted, or a rule whose actions are being
6081 * modified), marks facets for revalidation to ensure that packets will be
6082 * forwarded correctly according to the new state of the flow table.
6084 * This function must be called after *each* change to a flow table. See
6085 * the comment on table_update_taggable() for more information. */
6087 rule_invalidate(const struct rule_dpif *rule)
6089 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
6091 table_update_taggable(ofproto, rule->up.table_id);
6093 if (!ofproto->need_revalidate) {
6094 struct table_dpif *table = &ofproto->tables[rule->up.table_id];
6096 if (table->other_table && rule->tag) {
6097 tag_set_add(&ofproto->revalidate_set, rule->tag);
6099 ofproto->need_revalidate = true;
6105 set_frag_handling(struct ofproto *ofproto_,
6106 enum ofp_config_flags frag_handling)
6108 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
6110 if (frag_handling != OFPC_FRAG_REASM) {
6111 ofproto->need_revalidate = true;
6119 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
6120 const struct flow *flow,
6121 const union ofp_action *ofp_actions, size_t n_ofp_actions)
6123 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
6126 if (flow->in_port >= ofproto->max_ports && flow->in_port < OFPP_MAX) {
6127 return OFPERR_NXBRC_BAD_IN_PORT;
6130 error = validate_actions(ofp_actions, n_ofp_actions, flow,
6131 ofproto->max_ports);
6133 struct odputil_keybuf keybuf;
6134 struct dpif_flow_stats stats;
6138 struct action_xlate_ctx ctx;
6139 uint64_t odp_actions_stub[1024 / 8];
6140 struct ofpbuf odp_actions;
6142 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
6143 odp_flow_key_from_flow(&key, flow);
6145 dpif_flow_stats_extract(flow, packet, &stats);
6147 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci, NULL,
6148 packet_get_tcp_flags(packet, flow), packet);
6149 ctx.resubmit_stats = &stats;
6151 ofpbuf_use_stub(&odp_actions,
6152 odp_actions_stub, sizeof odp_actions_stub);
6153 xlate_actions(&ctx, ofp_actions, n_ofp_actions, &odp_actions);
6154 dpif_execute(ofproto->dpif, key.data, key.size,
6155 odp_actions.data, odp_actions.size, packet);
6156 ofpbuf_uninit(&odp_actions);
6164 set_netflow(struct ofproto *ofproto_,
6165 const struct netflow_options *netflow_options)
6167 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
6169 if (netflow_options) {
6170 if (!ofproto->netflow) {
6171 ofproto->netflow = netflow_create();
6173 return netflow_set_options(ofproto->netflow, netflow_options);
6175 netflow_destroy(ofproto->netflow);
6176 ofproto->netflow = NULL;
6182 get_netflow_ids(const struct ofproto *ofproto_,
6183 uint8_t *engine_type, uint8_t *engine_id)
6185 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
6187 dpif_get_netflow_ids(ofproto->dpif, engine_type, engine_id);
6191 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
6193 if (!facet_is_controller_flow(facet) &&
6194 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
6195 struct subfacet *subfacet;
6196 struct ofexpired expired;
6198 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
6199 if (subfacet->installed) {
6200 struct dpif_flow_stats stats;
6202 subfacet_install(subfacet, subfacet->actions,
6203 subfacet->actions_len, &stats);
6204 subfacet_update_stats(subfacet, &stats);
6208 expired.flow = facet->flow;
6209 expired.packet_count = facet->packet_count;
6210 expired.byte_count = facet->byte_count;
6211 expired.used = facet->used;
6212 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
6217 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
6219 struct facet *facet;
6221 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
6222 send_active_timeout(ofproto, facet);
6226 static struct ofproto_dpif *
6227 ofproto_dpif_lookup(const char *name)
6229 struct ofproto_dpif *ofproto;
6231 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
6232 hash_string(name, 0), &all_ofproto_dpifs) {
6233 if (!strcmp(ofproto->up.name, name)) {
6241 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
6242 const char *argv[], void *aux OVS_UNUSED)
6244 struct ofproto_dpif *ofproto;
6247 ofproto = ofproto_dpif_lookup(argv[1]);
6249 unixctl_command_reply_error(conn, "no such bridge");
6252 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
6254 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
6255 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
6259 unixctl_command_reply(conn, "table successfully flushed");
6263 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
6264 const char *argv[], void *aux OVS_UNUSED)
6266 struct ds ds = DS_EMPTY_INITIALIZER;
6267 const struct ofproto_dpif *ofproto;
6268 const struct mac_entry *e;
6270 ofproto = ofproto_dpif_lookup(argv[1]);
6272 unixctl_command_reply_error(conn, "no such bridge");
6276 ds_put_cstr(&ds, " port VLAN MAC Age\n");
6277 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
6278 struct ofbundle *bundle = e->port.p;
6279 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
6280 ofbundle_get_a_port(bundle)->odp_port,
6281 e->vlan, ETH_ADDR_ARGS(e->mac),
6282 mac_entry_age(ofproto->ml, e));
6284 unixctl_command_reply(conn, ds_cstr(&ds));
6289 struct action_xlate_ctx ctx;
6295 trace_format_rule(struct ds *result, uint8_t table_id, int level,
6296 const struct rule_dpif *rule)
6298 ds_put_char_multiple(result, '\t', level);
6300 ds_put_cstr(result, "No match\n");
6304 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
6305 table_id, ntohll(rule->up.flow_cookie));
6306 cls_rule_format(&rule->up.cr, result);
6307 ds_put_char(result, '\n');
6309 ds_put_char_multiple(result, '\t', level);
6310 ds_put_cstr(result, "OpenFlow ");
6311 ofp_print_actions(result, rule->up.actions, rule->up.n_actions);
6312 ds_put_char(result, '\n');
6316 trace_format_flow(struct ds *result, int level, const char *title,
6317 struct trace_ctx *trace)
6319 ds_put_char_multiple(result, '\t', level);
6320 ds_put_format(result, "%s: ", title);
6321 if (flow_equal(&trace->ctx.flow, &trace->flow)) {
6322 ds_put_cstr(result, "unchanged");
6324 flow_format(result, &trace->ctx.flow);
6325 trace->flow = trace->ctx.flow;
6327 ds_put_char(result, '\n');
6331 trace_format_regs(struct ds *result, int level, const char *title,
6332 struct trace_ctx *trace)
6336 ds_put_char_multiple(result, '\t', level);
6337 ds_put_format(result, "%s:", title);
6338 for (i = 0; i < FLOW_N_REGS; i++) {
6339 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
6341 ds_put_char(result, '\n');
6345 trace_format_odp(struct ds *result, int level, const char *title,
6346 struct trace_ctx *trace)
6348 struct ofpbuf *odp_actions = trace->ctx.odp_actions;
6350 ds_put_char_multiple(result, '\t', level);
6351 ds_put_format(result, "%s: ", title);
6352 format_odp_actions(result, odp_actions->data, odp_actions->size);
6353 ds_put_char(result, '\n');
6357 trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
6359 struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx);
6360 struct ds *result = trace->result;
6362 ds_put_char(result, '\n');
6363 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
6364 trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace);
6365 trace_format_odp(result, ctx->recurse + 1, "Resubmitted odp", trace);
6366 trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule);
6370 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
6371 void *aux OVS_UNUSED)
6373 const char *dpname = argv[1];
6374 struct ofproto_dpif *ofproto;
6375 struct ofpbuf odp_key;
6376 struct ofpbuf *packet;
6377 ovs_be16 initial_tci;
6383 ofpbuf_init(&odp_key, 0);
6386 ofproto = ofproto_dpif_lookup(dpname);
6388 unixctl_command_reply_error(conn, "Unknown ofproto (use ofproto/list "
6392 if (argc == 3 || (argc == 4 && !strcmp(argv[3], "-generate"))) {
6393 /* ofproto/trace dpname flow [-generate] */
6394 const char *flow_s = argv[2];
6395 const char *generate_s = argv[3];
6398 /* Convert string to datapath key. */
6399 ofpbuf_init(&odp_key, 0);
6400 error = odp_flow_key_from_string(flow_s, NULL, &odp_key);
6402 unixctl_command_reply_error(conn, "Bad flow syntax");
6406 /* Convert odp_key to flow. */
6407 error = ofproto_dpif_extract_flow_key(ofproto, odp_key.data,
6408 odp_key.size, &flow,
6409 &initial_tci, NULL);
6410 if (error == ODP_FIT_ERROR) {
6411 unixctl_command_reply_error(conn, "Invalid flow");
6415 /* Generate a packet, if requested. */
6417 packet = ofpbuf_new(0);
6418 flow_compose(packet, &flow);
6420 } else if (argc == 6) {
6421 /* ofproto/trace dpname priority tun_id in_port packet */
6422 const char *priority_s = argv[2];
6423 const char *tun_id_s = argv[3];
6424 const char *in_port_s = argv[4];
6425 const char *packet_s = argv[5];
6426 uint16_t in_port = ofp_port_to_odp_port(atoi(in_port_s));
6427 ovs_be64 tun_id = htonll(strtoull(tun_id_s, NULL, 0));
6428 uint32_t priority = atoi(priority_s);
6431 msg = eth_from_hex(packet_s, &packet);
6433 unixctl_command_reply_error(conn, msg);
6437 ds_put_cstr(&result, "Packet: ");
6438 s = ofp_packet_to_string(packet->data, packet->size);
6439 ds_put_cstr(&result, s);
6442 flow_extract(packet, priority, tun_id, in_port, &flow);
6443 initial_tci = flow.vlan_tci;
6445 unixctl_command_reply_error(conn, "Bad command syntax");
6449 ofproto_trace(ofproto, &flow, packet, initial_tci, &result);
6450 unixctl_command_reply(conn, ds_cstr(&result));
6453 ds_destroy(&result);
6454 ofpbuf_delete(packet);
6455 ofpbuf_uninit(&odp_key);
6459 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
6460 const struct ofpbuf *packet, ovs_be16 initial_tci,
6463 struct rule_dpif *rule;
6465 ds_put_cstr(ds, "Flow: ");
6466 flow_format(ds, flow);
6467 ds_put_char(ds, '\n');
6469 rule = rule_dpif_lookup(ofproto, flow, 0);
6470 trace_format_rule(ds, 0, 0, rule);
6472 uint64_t odp_actions_stub[1024 / 8];
6473 struct ofpbuf odp_actions;
6475 struct trace_ctx trace;
6478 tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
6481 ofpbuf_use_stub(&odp_actions,
6482 odp_actions_stub, sizeof odp_actions_stub);
6483 action_xlate_ctx_init(&trace.ctx, ofproto, flow, initial_tci,
6484 rule, tcp_flags, packet);
6485 trace.ctx.resubmit_hook = trace_resubmit;
6486 xlate_actions(&trace.ctx, rule->up.actions, rule->up.n_actions,
6489 ds_put_char(ds, '\n');
6490 trace_format_flow(ds, 0, "Final flow", &trace);
6491 ds_put_cstr(ds, "Datapath actions: ");
6492 format_odp_actions(ds, odp_actions.data, odp_actions.size);
6493 ofpbuf_uninit(&odp_actions);
6495 if (!trace.ctx.may_set_up_flow) {
6497 ds_put_cstr(ds, "\nThis flow is not cachable.");
6499 ds_put_cstr(ds, "\nThe datapath actions are incomplete--"
6500 "for complete actions, please supply a packet.");
6507 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
6508 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
6511 unixctl_command_reply(conn, NULL);
6515 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
6516 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
6519 unixctl_command_reply(conn, NULL);
6522 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
6523 * 'reply' describing the results. */
6525 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
6527 struct facet *facet;
6531 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
6532 if (!facet_check_consistency(facet)) {
6537 ofproto->need_revalidate = true;
6541 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
6542 ofproto->up.name, errors);
6544 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
6549 ofproto_dpif_self_check(struct unixctl_conn *conn,
6550 int argc, const char *argv[], void *aux OVS_UNUSED)
6552 struct ds reply = DS_EMPTY_INITIALIZER;
6553 struct ofproto_dpif *ofproto;
6556 ofproto = ofproto_dpif_lookup(argv[1]);
6558 unixctl_command_reply_error(conn, "Unknown ofproto (use "
6559 "ofproto/list for help)");
6562 ofproto_dpif_self_check__(ofproto, &reply);
6564 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
6565 ofproto_dpif_self_check__(ofproto, &reply);
6569 unixctl_command_reply(conn, ds_cstr(&reply));
6574 ofproto_dpif_unixctl_init(void)
6576 static bool registered;
6582 unixctl_command_register(
6584 "bridge {tun_id in_port packet | odp_flow [-generate]}",
6585 2, 5, ofproto_unixctl_trace, NULL);
6586 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
6587 ofproto_unixctl_fdb_flush, NULL);
6588 unixctl_command_register("fdb/show", "bridge", 1, 1,
6589 ofproto_unixctl_fdb_show, NULL);
6590 unixctl_command_register("ofproto/clog", "", 0, 0,
6591 ofproto_dpif_clog, NULL);
6592 unixctl_command_register("ofproto/unclog", "", 0, 0,
6593 ofproto_dpif_unclog, NULL);
6594 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
6595 ofproto_dpif_self_check, NULL);
6598 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
6600 * This is deprecated. It is only for compatibility with broken device drivers
6601 * in old versions of Linux that do not properly support VLANs when VLAN
6602 * devices are not used. When broken device drivers are no longer in
6603 * widespread use, we will delete these interfaces. */
6606 set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid)
6608 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
6609 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
6611 if (realdev_ofp_port == ofport->realdev_ofp_port
6612 && vid == ofport->vlandev_vid) {
6616 ofproto->need_revalidate = true;
6618 if (ofport->realdev_ofp_port) {
6621 if (realdev_ofp_port && ofport->bundle) {
6622 /* vlandevs are enslaved to their realdevs, so they are not allowed to
6623 * themselves be part of a bundle. */
6624 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
6627 ofport->realdev_ofp_port = realdev_ofp_port;
6628 ofport->vlandev_vid = vid;
6630 if (realdev_ofp_port) {
6631 vsp_add(ofport, realdev_ofp_port, vid);
6638 hash_realdev_vid(uint16_t realdev_ofp_port, int vid)
6640 return hash_2words(realdev_ofp_port, vid);
6643 /* Returns the ODP port number of the Linux VLAN device that corresponds to
6644 * 'vlan_tci' on the network device with port number 'realdev_odp_port' in
6645 * 'ofproto'. For example, given 'realdev_odp_port' of eth0 and 'vlan_tci' 9,
6646 * it would return the port number of eth0.9.
6648 * Unless VLAN splinters are enabled for port 'realdev_odp_port', this
6649 * function just returns its 'realdev_odp_port' argument. */
6651 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
6652 uint32_t realdev_odp_port, ovs_be16 vlan_tci)
6654 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
6655 uint16_t realdev_ofp_port = odp_port_to_ofp_port(realdev_odp_port);
6656 int vid = vlan_tci_to_vid(vlan_tci);
6657 const struct vlan_splinter *vsp;
6659 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
6660 hash_realdev_vid(realdev_ofp_port, vid),
6661 &ofproto->realdev_vid_map) {
6662 if (vsp->realdev_ofp_port == realdev_ofp_port
6663 && vsp->vid == vid) {
6664 return ofp_port_to_odp_port(vsp->vlandev_ofp_port);
6668 return realdev_odp_port;
6671 static struct vlan_splinter *
6672 vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port)
6674 struct vlan_splinter *vsp;
6676 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, hash_int(vlandev_ofp_port, 0),
6677 &ofproto->vlandev_map) {
6678 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
6686 /* Returns the OpenFlow port number of the "real" device underlying the Linux
6687 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
6688 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
6689 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
6690 * eth0 and store 9 in '*vid'.
6692 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
6693 * VLAN device. Unless VLAN splinters are enabled, this is what this function
6696 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
6697 uint16_t vlandev_ofp_port, int *vid)
6699 if (!hmap_is_empty(&ofproto->vlandev_map)) {
6700 const struct vlan_splinter *vsp;
6702 vsp = vlandev_find(ofproto, vlandev_ofp_port);
6707 return vsp->realdev_ofp_port;
6714 vsp_remove(struct ofport_dpif *port)
6716 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6717 struct vlan_splinter *vsp;
6719 vsp = vlandev_find(ofproto, port->up.ofp_port);
6721 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
6722 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
6725 port->realdev_ofp_port = 0;
6727 VLOG_ERR("missing vlan device record");
6732 vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid)
6734 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6736 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
6737 && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid))
6738 == realdev_ofp_port)) {
6739 struct vlan_splinter *vsp;
6741 vsp = xmalloc(sizeof *vsp);
6742 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
6743 hash_int(port->up.ofp_port, 0));
6744 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
6745 hash_realdev_vid(realdev_ofp_port, vid));
6746 vsp->realdev_ofp_port = realdev_ofp_port;
6747 vsp->vlandev_ofp_port = port->up.ofp_port;
6750 port->realdev_ofp_port = realdev_ofp_port;
6752 VLOG_ERR("duplicate vlan device record");
6756 const struct ofproto_class ofproto_dpif_class = {
6785 port_is_lacp_current,
6786 NULL, /* rule_choose_table */
6793 rule_modify_actions,
6801 get_cfm_remote_mpids,
6806 get_stp_port_status,
6813 is_mirror_output_bundle,
6814 forward_bpdu_changed,