2 * Copyright (c) 2009, 2010, 2011, 2012, 2013 Nicira, Inc.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at:
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
19 #include "ofproto/ofproto-provider.h"
26 #include "byte-order.h"
31 #include "dynamic-string.h"
32 #include "fail-open.h"
36 #include "mac-learning.h"
37 #include "meta-flow.h"
38 #include "multipath.h"
39 #include "netdev-vport.h"
46 #include "ofp-actions.h"
47 #include "ofp-parse.h"
48 #include "ofp-print.h"
49 #include "ofproto-dpif-governor.h"
50 #include "ofproto-dpif-ipfix.h"
51 #include "ofproto-dpif-sflow.h"
52 #include "poll-loop.h"
57 #include "unaligned.h"
59 #include "vlan-bitmap.h"
62 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
64 COVERAGE_DEFINE(ofproto_dpif_expired);
65 COVERAGE_DEFINE(ofproto_dpif_xlate);
66 COVERAGE_DEFINE(facet_changed_rule);
67 COVERAGE_DEFINE(facet_revalidate);
68 COVERAGE_DEFINE(facet_unexpected);
69 COVERAGE_DEFINE(facet_suppress);
71 /* Maximum depth of flow table recursion (due to resubmit actions) in a
72 * flow translation. */
73 #define MAX_RESUBMIT_RECURSION 64
75 /* Number of implemented OpenFlow tables. */
76 enum { N_TABLES = 255 };
77 enum { TBL_INTERNAL = N_TABLES - 1 }; /* Used for internal hidden rules. */
78 BUILD_ASSERT_DECL(N_TABLES >= 2 && N_TABLES <= 255);
90 * - Do include packets and bytes from facets that have been deleted or
91 * whose own statistics have been folded into the rule.
93 * - Do include packets and bytes sent "by hand" that were accounted to
94 * the rule without any facet being involved (this is a rare corner
95 * case in rule_execute()).
97 * - Do not include packet or bytes that can be obtained from any facet's
98 * packet_count or byte_count member or that can be obtained from the
99 * datapath by, e.g., dpif_flow_get() for any subfacet.
101 uint64_t packet_count; /* Number of packets received. */
102 uint64_t byte_count; /* Number of bytes received. */
104 tag_type tag; /* Caches rule_calculate_tag() result. */
106 struct list facets; /* List of "struct facet"s. */
109 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
111 return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
114 static struct rule_dpif *rule_dpif_lookup(struct ofproto_dpif *,
115 const struct flow *);
116 static struct rule_dpif *rule_dpif_lookup__(struct ofproto_dpif *,
119 static struct rule_dpif *rule_dpif_miss_rule(struct ofproto_dpif *ofproto,
120 const struct flow *flow);
122 static void rule_credit_stats(struct rule_dpif *,
123 const struct dpif_flow_stats *);
124 static void flow_push_stats(struct facet *, const struct dpif_flow_stats *);
125 static tag_type rule_calculate_tag(const struct flow *,
126 const struct minimask *, uint32_t basis);
127 static void rule_invalidate(const struct rule_dpif *);
129 #define MAX_MIRRORS 32
130 typedef uint32_t mirror_mask_t;
131 #define MIRROR_MASK_C(X) UINT32_C(X)
132 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
134 struct ofproto_dpif *ofproto; /* Owning ofproto. */
135 size_t idx; /* In ofproto's "mirrors" array. */
136 void *aux; /* Key supplied by ofproto's client. */
137 char *name; /* Identifier for log messages. */
139 /* Selection criteria. */
140 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
141 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
142 unsigned long *vlans; /* Bitmap of chosen VLANs, NULL selects all. */
144 /* Output (exactly one of out == NULL and out_vlan == -1 is true). */
145 struct ofbundle *out; /* Output port or NULL. */
146 int out_vlan; /* Output VLAN or -1. */
147 mirror_mask_t dup_mirrors; /* Bitmap of mirrors with the same output. */
150 int64_t packet_count; /* Number of packets sent. */
151 int64_t byte_count; /* Number of bytes sent. */
154 static void mirror_destroy(struct ofmirror *);
155 static void update_mirror_stats(struct ofproto_dpif *ofproto,
156 mirror_mask_t mirrors,
157 uint64_t packets, uint64_t bytes);
160 struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
161 struct ofproto_dpif *ofproto; /* Owning ofproto. */
162 void *aux; /* Key supplied by ofproto's client. */
163 char *name; /* Identifier for log messages. */
166 struct list ports; /* Contains "struct ofport"s. */
167 enum port_vlan_mode vlan_mode; /* VLAN mode */
168 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
169 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
170 * NULL if all VLANs are trunked. */
171 struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */
172 struct bond *bond; /* Nonnull iff more than one port. */
173 bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */
176 bool floodable; /* True if no port has OFPUTIL_PC_NO_FLOOD set. */
178 /* Port mirroring info. */
179 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
180 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
181 mirror_mask_t mirror_out; /* Mirrors that output to this bundle. */
184 static void bundle_remove(struct ofport *);
185 static void bundle_update(struct ofbundle *);
186 static void bundle_destroy(struct ofbundle *);
187 static void bundle_del_port(struct ofport_dpif *);
188 static void bundle_run(struct ofbundle *);
189 static void bundle_wait(struct ofbundle *);
190 static struct ofbundle *lookup_input_bundle(const struct ofproto_dpif *,
191 uint16_t in_port, bool warn,
192 struct ofport_dpif **in_ofportp);
194 /* A controller may use OFPP_NONE as the ingress port to indicate that
195 * it did not arrive on a "real" port. 'ofpp_none_bundle' exists for
196 * when an input bundle is needed for validation (e.g., mirroring or
197 * OFPP_NORMAL processing). It is not connected to an 'ofproto' or have
198 * any 'port' structs, so care must be taken when dealing with it. */
199 static struct ofbundle ofpp_none_bundle = {
201 .vlan_mode = PORT_VLAN_TRUNK
204 static void stp_run(struct ofproto_dpif *ofproto);
205 static void stp_wait(struct ofproto_dpif *ofproto);
206 static int set_stp_port(struct ofport *,
207 const struct ofproto_port_stp_settings *);
209 static bool ofbundle_includes_vlan(const struct ofbundle *, uint16_t vlan);
211 struct action_xlate_ctx {
212 /* action_xlate_ctx_init() initializes these members. */
215 struct ofproto_dpif *ofproto;
217 /* Flow to which the OpenFlow actions apply. xlate_actions() will modify
218 * this flow when actions change header fields. */
221 /* Flow at the last commit. */
222 struct flow base_flow;
224 /* Tunnel IP destination address as received. This is stored separately
225 * as the base_flow.tunnel is cleared on init to reflect the datapath
226 * behavior. Used to make sure not to send tunneled output to ourselves,
227 * which might lead to an infinite loop. This could happen easily
228 * if a tunnel is marked as 'ip_remote=flow', and the flow does not
229 * actually set the tun_dst field. */
230 ovs_be32 orig_tunnel_ip_dst;
232 /* stack for the push and pop actions.
233 * Each stack element is of the type "union mf_subvalue". */
235 union mf_subvalue init_stack[1024 / sizeof(union mf_subvalue)];
237 /* The packet corresponding to 'flow', or a null pointer if we are
238 * revalidating without a packet to refer to. */
239 const struct ofpbuf *packet;
241 /* Should OFPP_NORMAL update the MAC learning table? Should "learn"
242 * actions update the flow table?
244 * We want to update these tables if we are actually processing a packet,
245 * or if we are accounting for packets that the datapath has processed, but
246 * not if we are just revalidating. */
249 /* The rule that we are currently translating, or NULL. */
250 struct rule_dpif *rule;
252 /* Union of the set of TCP flags seen so far in this flow. (Used only by
253 * NXAST_FIN_TIMEOUT. Set to zero to avoid updating updating rules'
257 /* If nonnull, flow translation calls this function just before executing a
258 * resubmit or OFPP_TABLE action. In addition, disables logging of traces
259 * when the recursion depth is exceeded.
261 * 'rule' is the rule being submitted into. It will be null if the
262 * resubmit or OFPP_TABLE action didn't find a matching rule.
264 * This is normally null so the client has to set it manually after
265 * calling action_xlate_ctx_init(). */
266 void (*resubmit_hook)(struct action_xlate_ctx *, struct rule_dpif *rule);
268 /* If nonnull, flow translation calls this function to report some
269 * significant decision, e.g. to explain why OFPP_NORMAL translation
270 * dropped a packet. */
271 void (*report_hook)(struct action_xlate_ctx *, const char *s);
273 /* If nonnull, flow translation credits the specified statistics to each
274 * rule reached through a resubmit or OFPP_TABLE action.
276 * This is normally null so the client has to set it manually after
277 * calling action_xlate_ctx_init(). */
278 const struct dpif_flow_stats *resubmit_stats;
280 /* xlate_actions() initializes and uses these members. The client might want
281 * to look at them after it returns. */
283 struct ofpbuf *odp_actions; /* Datapath actions. */
284 tag_type tags; /* Tags associated with actions. */
285 enum slow_path_reason slow; /* 0 if fast path may be used. */
286 bool has_learn; /* Actions include NXAST_LEARN? */
287 bool has_normal; /* Actions output to OFPP_NORMAL? */
288 bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */
289 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
290 mirror_mask_t mirrors; /* Bitmap of associated mirrors. */
292 /* xlate_actions() initializes and uses these members, but the client has no
293 * reason to look at them. */
295 int recurse; /* Recursion level, via xlate_table_action. */
296 bool max_resubmit_trigger; /* Recursed too deeply during translation. */
297 uint32_t orig_skb_priority; /* Priority when packet arrived. */
298 uint8_t table_id; /* OpenFlow table ID where flow was found. */
299 uint32_t sflow_n_outputs; /* Number of output ports. */
300 uint32_t sflow_odp_port; /* Output port for composing sFlow action. */
301 uint16_t user_cookie_offset;/* Used for user_action_cookie fixup. */
302 bool exit; /* No further actions should be processed. */
305 /* Initial values of fields of the packet that may be changed during
306 * flow processing and needed later. */
307 struct initial_vals {
308 /* This is the value of vlan_tci in the packet as actually received from
309 * dpif. This is the same as the facet's flow.vlan_tci unless the packet
310 * was received via a VLAN splinter. In that case, this value is 0
311 * (because the packet as actually received from the dpif had no 802.1Q
312 * tag) but the facet's flow.vlan_tci is set to the VLAN that the splinter
315 * This member should be removed when the VLAN splinters feature is no
320 static void action_xlate_ctx_init(struct action_xlate_ctx *,
321 struct ofproto_dpif *, const struct flow *,
322 const struct initial_vals *initial_vals,
324 uint8_t tcp_flags, const struct ofpbuf *);
325 static void xlate_actions(struct action_xlate_ctx *,
326 const struct ofpact *ofpacts, size_t ofpacts_len,
327 struct ofpbuf *odp_actions);
328 static void xlate_actions_for_side_effects(struct action_xlate_ctx *,
329 const struct ofpact *ofpacts,
331 static void xlate_table_action(struct action_xlate_ctx *, uint16_t in_port,
332 uint8_t table_id, bool may_packet_in);
334 static size_t put_userspace_action(const struct ofproto_dpif *,
335 struct ofpbuf *odp_actions,
337 const union user_action_cookie *,
340 static void compose_slow_path(const struct ofproto_dpif *, const struct flow *,
341 enum slow_path_reason,
342 uint64_t *stub, size_t stub_size,
343 const struct nlattr **actionsp,
344 size_t *actions_lenp);
346 static void xlate_report(struct action_xlate_ctx *ctx, const char *s);
348 /* A subfacet (see "struct subfacet" below) has three possible installation
351 * - SF_NOT_INSTALLED: Not installed in the datapath. This will only be the
352 * case just after the subfacet is created, just before the subfacet is
353 * destroyed, or if the datapath returns an error when we try to install a
356 * - SF_FAST_PATH: The subfacet's actions are installed in the datapath.
358 * - SF_SLOW_PATH: An action that sends every packet for the subfacet through
359 * ofproto_dpif is installed in the datapath.
362 SF_NOT_INSTALLED, /* No datapath flow for this subfacet. */
363 SF_FAST_PATH, /* Full actions are installed. */
364 SF_SLOW_PATH, /* Send-to-userspace action is installed. */
367 /* A dpif flow and actions associated with a facet.
369 * See also the large comment on struct facet. */
372 struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */
373 struct list list_node; /* In struct facet's 'facets' list. */
374 struct facet *facet; /* Owning facet. */
376 enum odp_key_fitness key_fitness;
380 long long int used; /* Time last used; time created if not used. */
381 long long int created; /* Time created. */
383 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
384 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
386 enum subfacet_path path; /* Installed in datapath? */
388 /* Datapath port the packet arrived on. This is needed to remove
389 * flows for ports that are no longer part of the bridge. Since the
390 * flow definition only has the OpenFlow port number and the port is
391 * no longer part of the bridge, we can't determine the datapath port
392 * number needed to delete the flow from the datapath. */
393 uint32_t odp_in_port;
396 #define SUBFACET_DESTROY_MAX_BATCH 50
398 static struct subfacet *subfacet_create(struct facet *, struct flow_miss *miss,
400 static struct subfacet *subfacet_find(struct ofproto_dpif *,
401 const struct nlattr *key, size_t key_len,
403 static void subfacet_destroy(struct subfacet *);
404 static void subfacet_destroy__(struct subfacet *);
405 static void subfacet_destroy_batch(struct ofproto_dpif *,
406 struct subfacet **, int n);
407 static void subfacet_reset_dp_stats(struct subfacet *,
408 struct dpif_flow_stats *);
409 static void subfacet_update_time(struct subfacet *, long long int used);
410 static void subfacet_update_stats(struct subfacet *,
411 const struct dpif_flow_stats *);
412 static int subfacet_install(struct subfacet *,
413 const struct ofpbuf *odp_actions,
414 struct dpif_flow_stats *);
415 static void subfacet_uninstall(struct subfacet *);
417 /* An exact-match instantiation of an OpenFlow flow.
419 * A facet associates a "struct flow", which represents the Open vSwitch
420 * userspace idea of an exact-match flow, with one or more subfacets. Each
421 * subfacet tracks the datapath's idea of the exact-match flow equivalent to
422 * the facet. When the kernel module (or other dpif implementation) and Open
423 * vSwitch userspace agree on the definition of a flow key, there is exactly
424 * one subfacet per facet. If the dpif implementation supports more-specific
425 * flow matching than userspace, however, a facet can have more than one
426 * subfacet, each of which corresponds to some distinction in flow that
427 * userspace simply doesn't understand.
429 * Flow expiration works in terms of subfacets, so a facet must have at least
430 * one subfacet or it will never expire, leaking memory. */
433 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
434 struct list list_node; /* In owning rule's 'facets' list. */
435 struct rule_dpif *rule; /* Owning rule. */
438 struct list subfacets;
439 long long int used; /* Time last used; time created if not used. */
446 * - Do include packets and bytes sent "by hand", e.g. with
449 * - Do include packets and bytes that were obtained from the datapath
450 * when a subfacet's statistics were reset (e.g. dpif_flow_put() with
451 * DPIF_FP_ZERO_STATS).
453 * - Do not include packets or bytes that can be obtained from the
454 * datapath for any existing subfacet.
456 uint64_t packet_count; /* Number of packets received. */
457 uint64_t byte_count; /* Number of bytes received. */
459 /* Resubmit statistics. */
460 uint64_t prev_packet_count; /* Number of packets from last stats push. */
461 uint64_t prev_byte_count; /* Number of bytes from last stats push. */
462 long long int prev_used; /* Used time from last stats push. */
465 uint64_t accounted_bytes; /* Bytes processed by facet_account(). */
466 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
467 uint8_t tcp_flags; /* TCP flags seen for this 'rule'. */
469 bool has_learn; /* Actions include NXAST_LEARN? */
470 bool has_normal; /* Actions output to OFPP_NORMAL? */
471 bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */
472 tag_type tags; /* Tags that would require revalidation. */
473 mirror_mask_t mirrors; /* Bitmap of dependent mirrors. */
475 /* Datapath actions. */
476 struct ofpbuf odp_actions;
477 uint64_t odp_actions_stub[256 / 8];
479 /* Initial values of the packet that may be needed later. */
480 struct initial_vals initial_vals;
482 enum slow_path_reason slow; /* 0 if fast path may be used. */
484 /* Storage for a single subfacet, to reduce malloc() time and space
485 * overhead. (A facet always has at least one subfacet and in the common
486 * case has exactly one subfacet. However, 'one_subfacet' may not
487 * always be valid, since it could have been removed after newer
488 * subfacets were pushed onto the 'subfacets' list.) */
489 struct subfacet one_subfacet;
491 long long int learn_rl; /* Rate limiter for facet_learn(). */
494 static struct facet *facet_create(const struct flow_miss *, uint32_t hash);
495 static void facet_remove(struct facet *);
496 static void facet_free(struct facet *);
498 static struct facet *facet_find(struct ofproto_dpif *,
499 const struct flow *, uint32_t hash);
500 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
501 const struct flow *, uint32_t hash);
502 static bool facet_revalidate(struct facet *);
503 static bool facet_check_consistency(struct facet *);
505 static void facet_flush_stats(struct facet *);
507 static void facet_update_time(struct facet *, long long int used);
508 static void facet_reset_counters(struct facet *);
509 static void facet_push_stats(struct facet *);
510 static void facet_learn(struct facet *);
511 static void facet_account(struct facet *);
512 static void push_all_stats(void);
514 static bool facet_is_controller_flow(struct facet *);
517 struct hmap_node odp_port_node; /* In dpif_backer's "odp_to_ofport_map". */
521 struct ofbundle *bundle; /* Bundle that contains this port, if any. */
522 struct list bundle_node; /* In struct ofbundle's "ports" list. */
523 struct cfm *cfm; /* Connectivity Fault Management, if any. */
524 struct bfd *bfd; /* BFD, if any. */
525 tag_type tag; /* Tag associated with this port. */
526 bool may_enable; /* May be enabled in bonds. */
527 long long int carrier_seq; /* Carrier status changes. */
528 struct tnl_port *tnl_port; /* Tunnel handle, or null. */
531 struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */
532 enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */
533 long long int stp_state_entered;
535 struct hmap priorities; /* Map of attached 'priority_to_dscp's. */
537 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
539 * This is deprecated. It is only for compatibility with broken device
540 * drivers in old versions of Linux that do not properly support VLANs when
541 * VLAN devices are not used. When broken device drivers are no longer in
542 * widespread use, we will delete these interfaces. */
543 uint16_t realdev_ofp_port;
547 /* Node in 'ofport_dpif''s 'priorities' map. Used to maintain a map from
548 * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
549 * traffic egressing the 'ofport' with that priority should be marked with. */
550 struct priority_to_dscp {
551 struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'priorities' map. */
552 uint32_t priority; /* Priority of this queue (see struct flow). */
554 uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */
557 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
559 * This is deprecated. It is only for compatibility with broken device drivers
560 * in old versions of Linux that do not properly support VLANs when VLAN
561 * devices are not used. When broken device drivers are no longer in
562 * widespread use, we will delete these interfaces. */
563 struct vlan_splinter {
564 struct hmap_node realdev_vid_node;
565 struct hmap_node vlandev_node;
566 uint16_t realdev_ofp_port;
567 uint16_t vlandev_ofp_port;
571 static uint32_t vsp_realdev_to_vlandev(const struct ofproto_dpif *,
572 uint32_t realdev, ovs_be16 vlan_tci);
573 static bool vsp_adjust_flow(const struct ofproto_dpif *, struct flow *);
574 static void vsp_remove(struct ofport_dpif *);
575 static void vsp_add(struct ofport_dpif *, uint16_t realdev_ofp_port, int vid);
577 static uint32_t ofp_port_to_odp_port(const struct ofproto_dpif *,
579 static uint16_t odp_port_to_ofp_port(const struct ofproto_dpif *,
582 static struct ofport_dpif *
583 ofport_dpif_cast(const struct ofport *ofport)
585 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
588 static void port_run(struct ofport_dpif *);
589 static void port_run_fast(struct ofport_dpif *);
590 static void port_wait(struct ofport_dpif *);
591 static int set_bfd(struct ofport *, const struct smap *);
592 static int set_cfm(struct ofport *, const struct cfm_settings *);
593 static void ofport_clear_priorities(struct ofport_dpif *);
594 static void run_fast_rl(void);
596 struct dpif_completion {
597 struct list list_node;
598 struct ofoperation *op;
601 /* Extra information about a classifier table.
602 * Currently used just for optimized flow revalidation. */
604 /* If either of these is nonnull, then this table has a form that allows
605 * flows to be tagged to avoid revalidating most flows for the most common
606 * kinds of flow table changes. */
607 struct cls_table *catchall_table; /* Table that wildcards all fields. */
608 struct cls_table *other_table; /* Table with any other wildcard set. */
609 uint32_t basis; /* Keeps each table's tags separate. */
612 /* Reasons that we might need to revalidate every facet, and corresponding
615 * A value of 0 means that there is no need to revalidate.
617 * It would be nice to have some cleaner way to integrate with coverage
618 * counters, but with only a few reasons I guess this is good enough for
620 enum revalidate_reason {
621 REV_RECONFIGURE = 1, /* Switch configuration changed. */
622 REV_STP, /* Spanning tree protocol port status change. */
623 REV_PORT_TOGGLED, /* Port enabled or disabled by CFM, LACP, ...*/
624 REV_FLOW_TABLE, /* Flow table changed. */
625 REV_INCONSISTENCY /* Facet self-check failed. */
627 COVERAGE_DEFINE(rev_reconfigure);
628 COVERAGE_DEFINE(rev_stp);
629 COVERAGE_DEFINE(rev_port_toggled);
630 COVERAGE_DEFINE(rev_flow_table);
631 COVERAGE_DEFINE(rev_inconsistency);
633 /* Drop keys are odp flow keys which have drop flows installed in the kernel.
634 * These are datapath flows which have no associated ofproto, if they did we
635 * would use facets. */
637 struct hmap_node hmap_node;
642 /* All datapaths of a given type share a single dpif backer instance. */
647 struct timer next_expiration;
648 struct hmap odp_to_ofport_map; /* ODP port to ofport mapping. */
650 struct simap tnl_backers; /* Set of dpif ports backing tunnels. */
652 /* Facet revalidation flags applying to facets which use this backer. */
653 enum revalidate_reason need_revalidate; /* Revalidate every facet. */
654 struct tag_set revalidate_set; /* Revalidate only matching facets. */
656 struct hmap drop_keys; /* Set of dropped odp keys. */
659 /* All existing ofproto_backer instances, indexed by ofproto->up.type. */
660 static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers);
662 static void drop_key_clear(struct dpif_backer *);
663 static struct ofport_dpif *
664 odp_port_to_ofport(const struct dpif_backer *, uint32_t odp_port);
666 static void dpif_stats_update_hit_count(struct ofproto_dpif *ofproto,
668 struct avg_subfacet_rates {
669 double add_rate; /* Moving average of new flows created per minute. */
670 double del_rate; /* Moving average of flows deleted per minute. */
672 static void show_dp_rates(struct ds *ds, const char *heading,
673 const struct avg_subfacet_rates *rates);
674 static void exp_mavg(double *avg, int base, double new);
676 struct ofproto_dpif {
677 struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
679 struct dpif_backer *backer;
681 /* Special OpenFlow rules. */
682 struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */
683 struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */
689 struct netflow *netflow;
690 struct dpif_sflow *sflow;
691 struct dpif_ipfix *ipfix;
692 struct hmap bundles; /* Contains "struct ofbundle"s. */
693 struct mac_learning *ml;
694 struct ofmirror *mirrors[MAX_MIRRORS];
696 bool has_bonded_bundles;
700 struct hmap subfacets;
701 struct governor *governor;
702 long long int consistency_rl;
705 struct table_dpif tables[N_TABLES];
707 /* Support for debugging async flow mods. */
708 struct list completions;
710 bool has_bundle_action; /* True when the first bundle action appears. */
711 struct netdev_stats stats; /* To account packets generated and consumed in
716 long long int stp_last_tick;
718 /* VLAN splinters. */
719 struct hmap realdev_vid_map; /* (realdev,vid) -> vlandev. */
720 struct hmap vlandev_map; /* vlandev -> (realdev,vid). */
723 struct sset ports; /* Set of standard port names. */
724 struct sset ghost_ports; /* Ports with no datapath port. */
725 struct sset port_poll_set; /* Queued names for port_poll() reply. */
726 int port_poll_errno; /* Last errno for port_poll() reply. */
728 /* Per ofproto's dpif stats. */
732 /* Subfacet statistics.
734 * These keep track of the total number of subfacets added and deleted and
735 * flow life span. They are useful for computing the flow rates stats
736 * exposed via "ovs-appctl dpif/show". The goal is to learn about
737 * traffic patterns in ways that we can use later to improve Open vSwitch
738 * performance in new situations. */
739 long long int created; /* Time when it is created. */
740 unsigned int max_n_subfacet; /* Maximum number of flows */
742 /* The average number of subfacets... */
743 struct avg_subfacet_rates hourly; /* ...over the last hour. */
744 struct avg_subfacet_rates daily; /* ...over the last day. */
745 long long int last_minute; /* Last time 'hourly' was updated. */
747 /* Number of subfacets added or deleted since 'last_minute'. */
748 unsigned int subfacet_add_count;
749 unsigned int subfacet_del_count;
751 /* Number of subfacets added or deleted from 'created' to 'last_minute.' */
752 unsigned long long int total_subfacet_add_count;
753 unsigned long long int total_subfacet_del_count;
755 /* Sum of the number of milliseconds that each subfacet existed,
756 * over the subfacets that have been added and then later deleted. */
757 unsigned long long int total_subfacet_life_span;
759 /* Incremented by the number of currently existing subfacets, each
760 * time we pull statistics from the kernel. */
761 unsigned long long int total_subfacet_count;
763 /* Number of times we pull statistics from the kernel. */
764 unsigned long long int n_update_stats;
766 static unsigned long long int avg_subfacet_life_span(
767 const struct ofproto_dpif *);
768 static double avg_subfacet_count(const struct ofproto_dpif *ofproto);
769 static void update_moving_averages(struct ofproto_dpif *ofproto);
770 static void dpif_stats_update_hit_count(struct ofproto_dpif *ofproto,
772 static void update_max_subfacet_count(struct ofproto_dpif *ofproto);
774 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
775 * for debugging the asynchronous flow_mod implementation.) */
778 /* All existing ofproto_dpif instances, indexed by ->up.name. */
779 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
781 static void ofproto_dpif_unixctl_init(void);
783 static struct ofproto_dpif *
784 ofproto_dpif_cast(const struct ofproto *ofproto)
786 ovs_assert(ofproto->ofproto_class == &ofproto_dpif_class);
787 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
790 static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *,
792 static struct ofport_dpif *get_odp_port(const struct ofproto_dpif *,
794 static void ofproto_trace(struct ofproto_dpif *, const struct flow *,
795 const struct ofpbuf *,
796 const struct initial_vals *, struct ds *);
798 /* Packet processing. */
799 static void update_learning_table(struct ofproto_dpif *,
800 const struct flow *, int vlan,
803 #define FLOW_MISS_MAX_BATCH 50
804 static int handle_upcalls(struct dpif_backer *, unsigned int max_batch);
806 /* Flow expiration. */
807 static int expire(struct dpif_backer *);
810 static void send_netflow_active_timeouts(struct ofproto_dpif *);
813 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
814 static size_t compose_sflow_action(const struct ofproto_dpif *,
815 struct ofpbuf *odp_actions,
816 const struct flow *, uint32_t odp_port);
817 static void compose_ipfix_action(const struct ofproto_dpif *,
818 struct ofpbuf *odp_actions,
819 const struct flow *);
820 static void add_mirror_actions(struct action_xlate_ctx *ctx,
821 const struct flow *flow);
822 /* Global variables. */
823 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
825 /* Initial mappings of port to bridge mappings. */
826 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
828 /* Factory functions. */
831 init(const struct shash *iface_hints)
833 struct shash_node *node;
835 /* Make a local copy, since we don't own 'iface_hints' elements. */
836 SHASH_FOR_EACH(node, iface_hints) {
837 const struct iface_hint *orig_hint = node->data;
838 struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
840 new_hint->br_name = xstrdup(orig_hint->br_name);
841 new_hint->br_type = xstrdup(orig_hint->br_type);
842 new_hint->ofp_port = orig_hint->ofp_port;
844 shash_add(&init_ofp_ports, node->name, new_hint);
849 enumerate_types(struct sset *types)
851 dp_enumerate_types(types);
855 enumerate_names(const char *type, struct sset *names)
857 struct ofproto_dpif *ofproto;
860 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
861 if (strcmp(type, ofproto->up.type)) {
864 sset_add(names, ofproto->up.name);
871 del(const char *type, const char *name)
876 error = dpif_open(name, type, &dpif);
878 error = dpif_delete(dpif);
885 port_open_type(const char *datapath_type, const char *port_type)
887 return dpif_port_open_type(datapath_type, port_type);
890 /* Type functions. */
892 static struct ofproto_dpif *
893 lookup_ofproto_dpif_by_port_name(const char *name)
895 struct ofproto_dpif *ofproto;
897 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
898 if (sset_contains(&ofproto->ports, name)) {
907 type_run(const char *type)
909 static long long int push_timer = LLONG_MIN;
910 struct dpif_backer *backer;
914 backer = shash_find_data(&all_dpif_backers, type);
916 /* This is not necessarily a problem, since backers are only
917 * created on demand. */
921 dpif_run(backer->dpif);
923 /* The most natural place to push facet statistics is when they're pulled
924 * from the datapath. However, when there are many flows in the datapath,
925 * this expensive operation can occur so frequently, that it reduces our
926 * ability to quickly set up flows. To reduce the cost, we push statistics
928 if (time_msec() > push_timer) {
929 push_timer = time_msec() + 2000;
933 if (backer->need_revalidate
934 || !tag_set_is_empty(&backer->revalidate_set)) {
935 struct tag_set revalidate_set = backer->revalidate_set;
936 bool need_revalidate = backer->need_revalidate;
937 struct ofproto_dpif *ofproto;
938 struct simap_node *node;
939 struct simap tmp_backers;
941 /* Handle tunnel garbage collection. */
942 simap_init(&tmp_backers);
943 simap_swap(&backer->tnl_backers, &tmp_backers);
945 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
946 struct ofport_dpif *iter;
948 if (backer != ofproto->backer) {
952 HMAP_FOR_EACH (iter, up.hmap_node, &ofproto->up.ports) {
955 if (!iter->tnl_port) {
959 dp_port = netdev_vport_get_dpif_port(iter->up.netdev);
960 node = simap_find(&tmp_backers, dp_port);
962 simap_put(&backer->tnl_backers, dp_port, node->data);
963 simap_delete(&tmp_backers, node);
964 node = simap_find(&backer->tnl_backers, dp_port);
966 node = simap_find(&backer->tnl_backers, dp_port);
968 uint32_t odp_port = UINT32_MAX;
970 if (!dpif_port_add(backer->dpif, iter->up.netdev,
972 simap_put(&backer->tnl_backers, dp_port, odp_port);
973 node = simap_find(&backer->tnl_backers, dp_port);
978 iter->odp_port = node ? node->data : OVSP_NONE;
979 if (tnl_port_reconfigure(&iter->up, iter->odp_port,
981 backer->need_revalidate = REV_RECONFIGURE;
986 SIMAP_FOR_EACH (node, &tmp_backers) {
987 dpif_port_del(backer->dpif, node->data);
989 simap_destroy(&tmp_backers);
991 switch (backer->need_revalidate) {
992 case REV_RECONFIGURE: COVERAGE_INC(rev_reconfigure); break;
993 case REV_STP: COVERAGE_INC(rev_stp); break;
994 case REV_PORT_TOGGLED: COVERAGE_INC(rev_port_toggled); break;
995 case REV_FLOW_TABLE: COVERAGE_INC(rev_flow_table); break;
996 case REV_INCONSISTENCY: COVERAGE_INC(rev_inconsistency); break;
999 if (backer->need_revalidate) {
1000 /* Clear the drop_keys in case we should now be accepting some
1001 * formerly dropped flows. */
1002 drop_key_clear(backer);
1005 /* Clear the revalidation flags. */
1006 tag_set_init(&backer->revalidate_set);
1007 backer->need_revalidate = 0;
1009 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1010 struct facet *facet, *next;
1012 if (ofproto->backer != backer) {
1016 HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &ofproto->facets) {
1018 || tag_set_intersects(&revalidate_set, facet->tags)) {
1019 facet_revalidate(facet);
1026 if (timer_expired(&backer->next_expiration)) {
1027 int delay = expire(backer);
1028 timer_set_duration(&backer->next_expiration, delay);
1031 /* Check for port changes in the dpif. */
1032 while ((error = dpif_port_poll(backer->dpif, &devname)) == 0) {
1033 struct ofproto_dpif *ofproto;
1034 struct dpif_port port;
1036 /* Don't report on the datapath's device. */
1037 if (!strcmp(devname, dpif_base_name(backer->dpif))) {
1041 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
1042 &all_ofproto_dpifs) {
1043 if (simap_contains(&ofproto->backer->tnl_backers, devname)) {
1048 ofproto = lookup_ofproto_dpif_by_port_name(devname);
1049 if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
1050 /* The port was removed. If we know the datapath,
1051 * report it through poll_set(). If we don't, it may be
1052 * notifying us of a removal we initiated, so ignore it.
1053 * If there's a pending ENOBUFS, let it stand, since
1054 * everything will be reevaluated. */
1055 if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
1056 sset_add(&ofproto->port_poll_set, devname);
1057 ofproto->port_poll_errno = 0;
1059 } else if (!ofproto) {
1060 /* The port was added, but we don't know with which
1061 * ofproto we should associate it. Delete it. */
1062 dpif_port_del(backer->dpif, port.port_no);
1064 dpif_port_destroy(&port);
1070 if (error != EAGAIN) {
1071 struct ofproto_dpif *ofproto;
1073 /* There was some sort of error, so propagate it to all
1074 * ofprotos that use this backer. */
1075 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
1076 &all_ofproto_dpifs) {
1077 if (ofproto->backer == backer) {
1078 sset_clear(&ofproto->port_poll_set);
1079 ofproto->port_poll_errno = error;
1088 dpif_backer_run_fast(struct dpif_backer *backer, int max_batch)
1092 /* Handle one or more batches of upcalls, until there's nothing left to do
1093 * or until we do a fixed total amount of work.
1095 * We do work in batches because it can be much cheaper to set up a number
1096 * of flows and fire off their patches all at once. We do multiple batches
1097 * because in some cases handling a packet can cause another packet to be
1098 * queued almost immediately as part of the return flow. Both
1099 * optimizations can make major improvements on some benchmarks and
1100 * presumably for real traffic as well. */
1102 while (work < max_batch) {
1103 int retval = handle_upcalls(backer, max_batch - work);
1114 type_run_fast(const char *type)
1116 struct dpif_backer *backer;
1118 backer = shash_find_data(&all_dpif_backers, type);
1120 /* This is not necessarily a problem, since backers are only
1121 * created on demand. */
1125 return dpif_backer_run_fast(backer, FLOW_MISS_MAX_BATCH);
1131 static long long int port_rl = LLONG_MIN;
1132 static unsigned int backer_rl = 0;
1134 if (time_msec() >= port_rl) {
1135 struct ofproto_dpif *ofproto;
1136 struct ofport_dpif *ofport;
1138 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1140 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1141 port_run_fast(ofport);
1144 port_rl = time_msec() + 200;
1147 /* XXX: We have to be careful not to do too much work in this function. If
1148 * we call dpif_backer_run_fast() too often, or with too large a batch,
1149 * performance improves signifcantly, but at a cost. It's possible for the
1150 * number of flows in the datapath to increase without bound, and for poll
1151 * loops to take 10s of seconds. The correct solution to this problem,
1152 * long term, is to separate flow miss handling into it's own thread so it
1153 * isn't affected by revalidations, and expirations. Until then, this is
1154 * the best we can do. */
1155 if (++backer_rl >= 10) {
1156 struct shash_node *node;
1159 SHASH_FOR_EACH (node, &all_dpif_backers) {
1160 dpif_backer_run_fast(node->data, 1);
1166 type_wait(const char *type)
1168 struct dpif_backer *backer;
1170 backer = shash_find_data(&all_dpif_backers, type);
1172 /* This is not necessarily a problem, since backers are only
1173 * created on demand. */
1177 timer_wait(&backer->next_expiration);
1180 /* Basic life-cycle. */
1182 static int add_internal_flows(struct ofproto_dpif *);
1184 static struct ofproto *
1187 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
1188 return &ofproto->up;
1192 dealloc(struct ofproto *ofproto_)
1194 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1199 close_dpif_backer(struct dpif_backer *backer)
1201 struct shash_node *node;
1203 ovs_assert(backer->refcount > 0);
1205 if (--backer->refcount) {
1209 drop_key_clear(backer);
1210 hmap_destroy(&backer->drop_keys);
1212 simap_destroy(&backer->tnl_backers);
1213 hmap_destroy(&backer->odp_to_ofport_map);
1214 node = shash_find(&all_dpif_backers, backer->type);
1216 shash_delete(&all_dpif_backers, node);
1217 dpif_close(backer->dpif);
1222 /* Datapath port slated for removal from datapath. */
1223 struct odp_garbage {
1224 struct list list_node;
1229 open_dpif_backer(const char *type, struct dpif_backer **backerp)
1231 struct dpif_backer *backer;
1232 struct dpif_port_dump port_dump;
1233 struct dpif_port port;
1234 struct shash_node *node;
1235 struct list garbage_list;
1236 struct odp_garbage *garbage, *next;
1242 backer = shash_find_data(&all_dpif_backers, type);
1249 backer_name = xasprintf("ovs-%s", type);
1251 /* Remove any existing datapaths, since we assume we're the only
1252 * userspace controlling the datapath. */
1254 dp_enumerate_names(type, &names);
1255 SSET_FOR_EACH(name, &names) {
1256 struct dpif *old_dpif;
1258 /* Don't remove our backer if it exists. */
1259 if (!strcmp(name, backer_name)) {
1263 if (dpif_open(name, type, &old_dpif)) {
1264 VLOG_WARN("couldn't open old datapath %s to remove it", name);
1266 dpif_delete(old_dpif);
1267 dpif_close(old_dpif);
1270 sset_destroy(&names);
1272 backer = xmalloc(sizeof *backer);
1274 error = dpif_create_and_open(backer_name, type, &backer->dpif);
1277 VLOG_ERR("failed to open datapath of type %s: %s", type,
1283 backer->type = xstrdup(type);
1284 backer->refcount = 1;
1285 hmap_init(&backer->odp_to_ofport_map);
1286 hmap_init(&backer->drop_keys);
1287 timer_set_duration(&backer->next_expiration, 1000);
1288 backer->need_revalidate = 0;
1289 simap_init(&backer->tnl_backers);
1290 tag_set_init(&backer->revalidate_set);
1293 dpif_flow_flush(backer->dpif);
1295 /* Loop through the ports already on the datapath and remove any
1296 * that we don't need anymore. */
1297 list_init(&garbage_list);
1298 dpif_port_dump_start(&port_dump, backer->dpif);
1299 while (dpif_port_dump_next(&port_dump, &port)) {
1300 node = shash_find(&init_ofp_ports, port.name);
1301 if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
1302 garbage = xmalloc(sizeof *garbage);
1303 garbage->odp_port = port.port_no;
1304 list_push_front(&garbage_list, &garbage->list_node);
1307 dpif_port_dump_done(&port_dump);
1309 LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) {
1310 dpif_port_del(backer->dpif, garbage->odp_port);
1311 list_remove(&garbage->list_node);
1315 shash_add(&all_dpif_backers, type, backer);
1317 error = dpif_recv_set(backer->dpif, true);
1319 VLOG_ERR("failed to listen on datapath of type %s: %s",
1320 type, strerror(error));
1321 close_dpif_backer(backer);
1329 construct(struct ofproto *ofproto_)
1331 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1332 struct shash_node *node, *next;
1337 error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1342 max_ports = dpif_get_max_ports(ofproto->backer->dpif);
1343 ofproto_init_max_ports(ofproto_, MIN(max_ports, OFPP_MAX));
1345 ofproto->n_matches = 0;
1347 ofproto->netflow = NULL;
1348 ofproto->sflow = NULL;
1349 ofproto->ipfix = NULL;
1350 ofproto->stp = NULL;
1351 hmap_init(&ofproto->bundles);
1352 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1353 for (i = 0; i < MAX_MIRRORS; i++) {
1354 ofproto->mirrors[i] = NULL;
1356 ofproto->has_bonded_bundles = false;
1358 hmap_init(&ofproto->facets);
1359 hmap_init(&ofproto->subfacets);
1360 ofproto->governor = NULL;
1361 ofproto->consistency_rl = LLONG_MIN;
1363 for (i = 0; i < N_TABLES; i++) {
1364 struct table_dpif *table = &ofproto->tables[i];
1366 table->catchall_table = NULL;
1367 table->other_table = NULL;
1368 table->basis = random_uint32();
1371 list_init(&ofproto->completions);
1373 ofproto_dpif_unixctl_init();
1375 ofproto->has_mirrors = false;
1376 ofproto->has_bundle_action = false;
1378 hmap_init(&ofproto->vlandev_map);
1379 hmap_init(&ofproto->realdev_vid_map);
1381 sset_init(&ofproto->ports);
1382 sset_init(&ofproto->ghost_ports);
1383 sset_init(&ofproto->port_poll_set);
1384 ofproto->port_poll_errno = 0;
1386 SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1387 struct iface_hint *iface_hint = node->data;
1389 if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1390 /* Check if the datapath already has this port. */
1391 if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1392 sset_add(&ofproto->ports, node->name);
1395 free(iface_hint->br_name);
1396 free(iface_hint->br_type);
1398 shash_delete(&init_ofp_ports, node);
1402 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1403 hash_string(ofproto->up.name, 0));
1404 memset(&ofproto->stats, 0, sizeof ofproto->stats);
1406 ofproto_init_tables(ofproto_, N_TABLES);
1407 error = add_internal_flows(ofproto);
1408 ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1411 ofproto->n_missed = 0;
1413 ofproto->max_n_subfacet = 0;
1414 ofproto->created = time_msec();
1415 ofproto->last_minute = ofproto->created;
1416 memset(&ofproto->hourly, 0, sizeof ofproto->hourly);
1417 memset(&ofproto->daily, 0, sizeof ofproto->daily);
1418 ofproto->subfacet_add_count = 0;
1419 ofproto->subfacet_del_count = 0;
1420 ofproto->total_subfacet_add_count = 0;
1421 ofproto->total_subfacet_del_count = 0;
1422 ofproto->total_subfacet_life_span = 0;
1423 ofproto->total_subfacet_count = 0;
1424 ofproto->n_update_stats = 0;
1430 add_internal_flow(struct ofproto_dpif *ofproto, int id,
1431 const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1433 struct ofputil_flow_mod fm;
1436 match_init_catchall(&fm.match);
1438 match_set_reg(&fm.match, 0, id);
1439 fm.new_cookie = htonll(0);
1440 fm.cookie = htonll(0);
1441 fm.cookie_mask = htonll(0);
1442 fm.table_id = TBL_INTERNAL;
1443 fm.command = OFPFC_ADD;
1444 fm.idle_timeout = 0;
1445 fm.hard_timeout = 0;
1449 fm.ofpacts = ofpacts->data;
1450 fm.ofpacts_len = ofpacts->size;
1452 error = ofproto_flow_mod(&ofproto->up, &fm);
1454 VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
1455 id, ofperr_to_string(error));
1459 *rulep = rule_dpif_lookup__(ofproto, &fm.match.flow, TBL_INTERNAL);
1460 ovs_assert(*rulep != NULL);
1466 add_internal_flows(struct ofproto_dpif *ofproto)
1468 struct ofpact_controller *controller;
1469 uint64_t ofpacts_stub[128 / 8];
1470 struct ofpbuf ofpacts;
1474 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1477 controller = ofpact_put_CONTROLLER(&ofpacts);
1478 controller->max_len = UINT16_MAX;
1479 controller->controller_id = 0;
1480 controller->reason = OFPR_NO_MATCH;
1481 ofpact_pad(&ofpacts);
1483 error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule);
1488 ofpbuf_clear(&ofpacts);
1489 error = add_internal_flow(ofproto, id++, &ofpacts,
1490 &ofproto->no_packet_in_rule);
1495 complete_operations(struct ofproto_dpif *ofproto)
1497 struct dpif_completion *c, *next;
1499 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
1500 ofoperation_complete(c->op, 0);
1501 list_remove(&c->list_node);
1507 destruct(struct ofproto *ofproto_)
1509 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1510 struct rule_dpif *rule, *next_rule;
1511 struct oftable *table;
1514 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1515 complete_operations(ofproto);
1517 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1518 struct cls_cursor cursor;
1520 cls_cursor_init(&cursor, &table->cls, NULL);
1521 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1522 ofproto_rule_destroy(&rule->up);
1526 for (i = 0; i < MAX_MIRRORS; i++) {
1527 mirror_destroy(ofproto->mirrors[i]);
1530 netflow_destroy(ofproto->netflow);
1531 dpif_sflow_destroy(ofproto->sflow);
1532 hmap_destroy(&ofproto->bundles);
1533 mac_learning_destroy(ofproto->ml);
1535 hmap_destroy(&ofproto->facets);
1536 hmap_destroy(&ofproto->subfacets);
1537 governor_destroy(ofproto->governor);
1539 hmap_destroy(&ofproto->vlandev_map);
1540 hmap_destroy(&ofproto->realdev_vid_map);
1542 sset_destroy(&ofproto->ports);
1543 sset_destroy(&ofproto->ghost_ports);
1544 sset_destroy(&ofproto->port_poll_set);
1546 close_dpif_backer(ofproto->backer);
1550 run_fast(struct ofproto *ofproto_)
1552 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1553 struct ofport_dpif *ofport;
1555 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1556 port_run_fast(ofport);
1563 run(struct ofproto *ofproto_)
1565 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1566 struct ofport_dpif *ofport;
1567 struct ofbundle *bundle;
1571 complete_operations(ofproto);
1574 error = run_fast(ofproto_);
1579 if (ofproto->netflow) {
1580 if (netflow_run(ofproto->netflow)) {
1581 send_netflow_active_timeouts(ofproto);
1584 if (ofproto->sflow) {
1585 dpif_sflow_run(ofproto->sflow);
1588 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1591 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1596 mac_learning_run(ofproto->ml, &ofproto->backer->revalidate_set);
1598 /* Check the consistency of a random facet, to aid debugging. */
1599 if (time_msec() >= ofproto->consistency_rl
1600 && !hmap_is_empty(&ofproto->facets)
1601 && !ofproto->backer->need_revalidate) {
1602 struct facet *facet;
1604 ofproto->consistency_rl = time_msec() + 250;
1606 facet = CONTAINER_OF(hmap_random_node(&ofproto->facets),
1607 struct facet, hmap_node);
1608 if (!tag_set_intersects(&ofproto->backer->revalidate_set,
1610 if (!facet_check_consistency(facet)) {
1611 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
1616 if (ofproto->governor) {
1619 governor_run(ofproto->governor);
1621 /* If the governor has shrunk to its minimum size and the number of
1622 * subfacets has dwindled, then drop the governor entirely.
1624 * For hysteresis, the number of subfacets to drop the governor is
1625 * smaller than the number needed to trigger its creation. */
1626 n_subfacets = hmap_count(&ofproto->subfacets);
1627 if (n_subfacets * 4 < ofproto->up.flow_eviction_threshold
1628 && governor_is_idle(ofproto->governor)) {
1629 governor_destroy(ofproto->governor);
1630 ofproto->governor = NULL;
1638 wait(struct ofproto *ofproto_)
1640 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1641 struct ofport_dpif *ofport;
1642 struct ofbundle *bundle;
1644 if (!clogged && !list_is_empty(&ofproto->completions)) {
1645 poll_immediate_wake();
1648 dpif_wait(ofproto->backer->dpif);
1649 dpif_recv_wait(ofproto->backer->dpif);
1650 if (ofproto->sflow) {
1651 dpif_sflow_wait(ofproto->sflow);
1653 if (!tag_set_is_empty(&ofproto->backer->revalidate_set)) {
1654 poll_immediate_wake();
1656 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1659 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1660 bundle_wait(bundle);
1662 if (ofproto->netflow) {
1663 netflow_wait(ofproto->netflow);
1665 mac_learning_wait(ofproto->ml);
1667 if (ofproto->backer->need_revalidate) {
1668 /* Shouldn't happen, but if it does just go around again. */
1669 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1670 poll_immediate_wake();
1672 if (ofproto->governor) {
1673 governor_wait(ofproto->governor);
1678 get_memory_usage(const struct ofproto *ofproto_, struct simap *usage)
1680 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1682 simap_increase(usage, "facets", hmap_count(&ofproto->facets));
1683 simap_increase(usage, "subfacets", hmap_count(&ofproto->subfacets));
1687 flush(struct ofproto *ofproto_)
1689 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1690 struct subfacet *subfacet, *next_subfacet;
1691 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
1695 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
1696 &ofproto->subfacets) {
1697 if (subfacet->path != SF_NOT_INSTALLED) {
1698 batch[n_batch++] = subfacet;
1699 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
1700 subfacet_destroy_batch(ofproto, batch, n_batch);
1704 subfacet_destroy(subfacet);
1709 subfacet_destroy_batch(ofproto, batch, n_batch);
1714 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1715 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1717 *arp_match_ip = true;
1718 *actions = (OFPUTIL_A_OUTPUT |
1719 OFPUTIL_A_SET_VLAN_VID |
1720 OFPUTIL_A_SET_VLAN_PCP |
1721 OFPUTIL_A_STRIP_VLAN |
1722 OFPUTIL_A_SET_DL_SRC |
1723 OFPUTIL_A_SET_DL_DST |
1724 OFPUTIL_A_SET_NW_SRC |
1725 OFPUTIL_A_SET_NW_DST |
1726 OFPUTIL_A_SET_NW_TOS |
1727 OFPUTIL_A_SET_TP_SRC |
1728 OFPUTIL_A_SET_TP_DST |
1733 get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots)
1735 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1736 struct dpif_dp_stats s;
1738 strcpy(ots->name, "classifier");
1740 dpif_get_dp_stats(ofproto->backer->dpif, &s);
1742 ots->lookup_count = htonll(s.n_hit + s.n_missed);
1743 ots->matched_count = htonll(s.n_hit + ofproto->n_matches);
1746 static struct ofport *
1749 struct ofport_dpif *port = xmalloc(sizeof *port);
1754 port_dealloc(struct ofport *port_)
1756 struct ofport_dpif *port = ofport_dpif_cast(port_);
1761 port_construct(struct ofport *port_)
1763 struct ofport_dpif *port = ofport_dpif_cast(port_);
1764 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1765 const struct netdev *netdev = port->up.netdev;
1766 struct dpif_port dpif_port;
1769 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1770 port->bundle = NULL;
1773 port->tag = tag_create_random();
1774 port->may_enable = true;
1775 port->stp_port = NULL;
1776 port->stp_state = STP_DISABLED;
1777 port->tnl_port = NULL;
1778 hmap_init(&port->priorities);
1779 port->realdev_ofp_port = 0;
1780 port->vlandev_vid = 0;
1781 port->carrier_seq = netdev_get_carrier_resets(netdev);
1783 if (netdev_vport_is_patch(netdev)) {
1784 /* By bailing out here, we don't submit the port to the sFlow module
1785 * to be considered for counter polling export. This is correct
1786 * because the patch port represents an interface that sFlow considers
1787 * to be "internal" to the switch as a whole, and therefore not an
1788 * candidate for counter polling. */
1789 port->odp_port = OVSP_NONE;
1793 error = dpif_port_query_by_name(ofproto->backer->dpif,
1794 netdev_vport_get_dpif_port(netdev),
1800 port->odp_port = dpif_port.port_no;
1802 if (netdev_get_tunnel_config(netdev)) {
1803 port->tnl_port = tnl_port_add(&port->up, port->odp_port);
1805 /* Sanity-check that a mapping doesn't already exist. This
1806 * shouldn't happen for non-tunnel ports. */
1807 if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1808 VLOG_ERR("port %s already has an OpenFlow port number",
1810 dpif_port_destroy(&dpif_port);
1814 hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1815 hash_int(port->odp_port, 0));
1817 dpif_port_destroy(&dpif_port);
1819 if (ofproto->sflow) {
1820 dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1827 port_destruct(struct ofport *port_)
1829 struct ofport_dpif *port = ofport_dpif_cast(port_);
1830 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1831 const char *dp_port_name = netdev_vport_get_dpif_port(port->up.netdev);
1832 const char *devname = netdev_get_name(port->up.netdev);
1834 if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1835 /* The underlying device is still there, so delete it. This
1836 * happens when the ofproto is being destroyed, since the caller
1837 * assumes that removal of attached ports will happen as part of
1839 if (!port->tnl_port) {
1840 dpif_port_del(ofproto->backer->dpif, port->odp_port);
1842 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1845 if (port->odp_port != OVSP_NONE && !port->tnl_port) {
1846 hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1849 tnl_port_del(port->tnl_port);
1850 sset_find_and_delete(&ofproto->ports, devname);
1851 sset_find_and_delete(&ofproto->ghost_ports, devname);
1852 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1853 bundle_remove(port_);
1854 set_cfm(port_, NULL);
1855 set_bfd(port_, NULL);
1856 if (ofproto->sflow) {
1857 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1860 ofport_clear_priorities(port);
1861 hmap_destroy(&port->priorities);
1865 port_modified(struct ofport *port_)
1867 struct ofport_dpif *port = ofport_dpif_cast(port_);
1869 if (port->bundle && port->bundle->bond) {
1870 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1875 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1877 struct ofport_dpif *port = ofport_dpif_cast(port_);
1878 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1879 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1881 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1882 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1883 OFPUTIL_PC_NO_PACKET_IN)) {
1884 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1886 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1887 bundle_update(port->bundle);
1893 set_sflow(struct ofproto *ofproto_,
1894 const struct ofproto_sflow_options *sflow_options)
1896 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1897 struct dpif_sflow *ds = ofproto->sflow;
1899 if (sflow_options) {
1901 struct ofport_dpif *ofport;
1903 ds = ofproto->sflow = dpif_sflow_create();
1904 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1905 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1907 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1909 dpif_sflow_set_options(ds, sflow_options);
1912 dpif_sflow_destroy(ds);
1913 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1914 ofproto->sflow = NULL;
1922 struct ofproto *ofproto_,
1923 const struct ofproto_ipfix_bridge_exporter_options *bridge_exporter_options,
1924 const struct ofproto_ipfix_flow_exporter_options *flow_exporters_options,
1925 size_t n_flow_exporters_options)
1927 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1928 struct dpif_ipfix *di = ofproto->ipfix;
1930 if (bridge_exporter_options || flow_exporters_options) {
1932 di = ofproto->ipfix = dpif_ipfix_create();
1934 dpif_ipfix_set_options(
1935 di, bridge_exporter_options, flow_exporters_options,
1936 n_flow_exporters_options);
1939 dpif_ipfix_destroy(di);
1940 ofproto->ipfix = NULL;
1947 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1949 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1956 struct ofproto_dpif *ofproto;
1958 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1959 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1960 ofport->cfm = cfm_create(ofport->up.netdev);
1963 if (cfm_configure(ofport->cfm, s)) {
1969 cfm_destroy(ofport->cfm);
1975 get_cfm_status(const struct ofport *ofport_,
1976 struct ofproto_cfm_status *status)
1978 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1981 status->faults = cfm_get_fault(ofport->cfm);
1982 status->remote_opstate = cfm_get_opup(ofport->cfm);
1983 status->health = cfm_get_health(ofport->cfm);
1984 cfm_get_remote_mpids(ofport->cfm, &status->rmps, &status->n_rmps);
1992 set_bfd(struct ofport *ofport_, const struct smap *cfg)
1994 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
1995 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1999 ofport->bfd = bfd_configure(old, netdev_get_name(ofport->up.netdev), cfg);
2000 if (ofport->bfd != old) {
2001 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2008 get_bfd_status(struct ofport *ofport_, struct smap *smap)
2010 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2013 bfd_get_status(ofport->bfd, smap);
2020 /* Spanning Tree. */
2023 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
2025 struct ofproto_dpif *ofproto = ofproto_;
2026 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
2027 struct ofport_dpif *ofport;
2029 ofport = stp_port_get_aux(sp);
2031 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
2032 ofproto->up.name, port_num);
2034 struct eth_header *eth = pkt->l2;
2036 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
2037 if (eth_addr_is_zero(eth->eth_src)) {
2038 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
2039 "with unknown MAC", ofproto->up.name, port_num);
2041 send_packet(ofport, pkt);
2047 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
2049 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
2051 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2053 /* Only revalidate flows if the configuration changed. */
2054 if (!s != !ofproto->stp) {
2055 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2059 if (!ofproto->stp) {
2060 ofproto->stp = stp_create(ofproto_->name, s->system_id,
2061 send_bpdu_cb, ofproto);
2062 ofproto->stp_last_tick = time_msec();
2065 stp_set_bridge_id(ofproto->stp, s->system_id);
2066 stp_set_bridge_priority(ofproto->stp, s->priority);
2067 stp_set_hello_time(ofproto->stp, s->hello_time);
2068 stp_set_max_age(ofproto->stp, s->max_age);
2069 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
2071 struct ofport *ofport;
2073 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
2074 set_stp_port(ofport, NULL);
2077 stp_destroy(ofproto->stp);
2078 ofproto->stp = NULL;
2085 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
2087 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2091 s->bridge_id = stp_get_bridge_id(ofproto->stp);
2092 s->designated_root = stp_get_designated_root(ofproto->stp);
2093 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
2102 update_stp_port_state(struct ofport_dpif *ofport)
2104 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2105 enum stp_state state;
2107 /* Figure out new state. */
2108 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
2112 if (ofport->stp_state != state) {
2113 enum ofputil_port_state of_state;
2116 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
2117 netdev_get_name(ofport->up.netdev),
2118 stp_state_name(ofport->stp_state),
2119 stp_state_name(state));
2120 if (stp_learn_in_state(ofport->stp_state)
2121 != stp_learn_in_state(state)) {
2122 /* xxx Learning action flows should also be flushed. */
2123 mac_learning_flush(ofproto->ml,
2124 &ofproto->backer->revalidate_set);
2126 fwd_change = stp_forward_in_state(ofport->stp_state)
2127 != stp_forward_in_state(state);
2129 ofproto->backer->need_revalidate = REV_STP;
2130 ofport->stp_state = state;
2131 ofport->stp_state_entered = time_msec();
2133 if (fwd_change && ofport->bundle) {
2134 bundle_update(ofport->bundle);
2137 /* Update the STP state bits in the OpenFlow port description. */
2138 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
2139 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
2140 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
2141 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
2142 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
2144 ofproto_port_set_state(&ofport->up, of_state);
2148 /* Configures STP on 'ofport_' using the settings defined in 's'. The
2149 * caller is responsible for assigning STP port numbers and ensuring
2150 * there are no duplicates. */
2152 set_stp_port(struct ofport *ofport_,
2153 const struct ofproto_port_stp_settings *s)
2155 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2156 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2157 struct stp_port *sp = ofport->stp_port;
2159 if (!s || !s->enable) {
2161 ofport->stp_port = NULL;
2162 stp_port_disable(sp);
2163 update_stp_port_state(ofport);
2166 } else if (sp && stp_port_no(sp) != s->port_num
2167 && ofport == stp_port_get_aux(sp)) {
2168 /* The port-id changed, so disable the old one if it's not
2169 * already in use by another port. */
2170 stp_port_disable(sp);
2173 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
2174 stp_port_enable(sp);
2176 stp_port_set_aux(sp, ofport);
2177 stp_port_set_priority(sp, s->priority);
2178 stp_port_set_path_cost(sp, s->path_cost);
2180 update_stp_port_state(ofport);
2186 get_stp_port_status(struct ofport *ofport_,
2187 struct ofproto_port_stp_status *s)
2189 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2190 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2191 struct stp_port *sp = ofport->stp_port;
2193 if (!ofproto->stp || !sp) {
2199 s->port_id = stp_port_get_id(sp);
2200 s->state = stp_port_get_state(sp);
2201 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
2202 s->role = stp_port_get_role(sp);
2203 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
2209 stp_run(struct ofproto_dpif *ofproto)
2212 long long int now = time_msec();
2213 long long int elapsed = now - ofproto->stp_last_tick;
2214 struct stp_port *sp;
2217 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
2218 ofproto->stp_last_tick = now;
2220 while (stp_get_changed_port(ofproto->stp, &sp)) {
2221 struct ofport_dpif *ofport = stp_port_get_aux(sp);
2224 update_stp_port_state(ofport);
2228 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
2229 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
2235 stp_wait(struct ofproto_dpif *ofproto)
2238 poll_timer_wait(1000);
2242 /* Returns true if STP should process 'flow'. */
2244 stp_should_process_flow(const struct flow *flow)
2246 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
2250 stp_process_packet(const struct ofport_dpif *ofport,
2251 const struct ofpbuf *packet)
2253 struct ofpbuf payload = *packet;
2254 struct eth_header *eth = payload.data;
2255 struct stp_port *sp = ofport->stp_port;
2257 /* Sink packets on ports that have STP disabled when the bridge has
2259 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
2263 /* Trim off padding on payload. */
2264 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
2265 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
2268 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
2269 stp_received_bpdu(sp, payload.data, payload.size);
2273 static struct priority_to_dscp *
2274 get_priority(const struct ofport_dpif *ofport, uint32_t priority)
2276 struct priority_to_dscp *pdscp;
2279 hash = hash_int(priority, 0);
2280 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) {
2281 if (pdscp->priority == priority) {
2289 ofport_clear_priorities(struct ofport_dpif *ofport)
2291 struct priority_to_dscp *pdscp, *next;
2293 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) {
2294 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
2300 set_queues(struct ofport *ofport_,
2301 const struct ofproto_port_queue *qdscp_list,
2304 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2305 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2306 struct hmap new = HMAP_INITIALIZER(&new);
2309 for (i = 0; i < n_qdscp; i++) {
2310 struct priority_to_dscp *pdscp;
2314 dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
2315 if (dpif_queue_to_priority(ofproto->backer->dpif, qdscp_list[i].queue,
2320 pdscp = get_priority(ofport, priority);
2322 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
2324 pdscp = xmalloc(sizeof *pdscp);
2325 pdscp->priority = priority;
2327 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2330 if (pdscp->dscp != dscp) {
2332 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2335 hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0));
2338 if (!hmap_is_empty(&ofport->priorities)) {
2339 ofport_clear_priorities(ofport);
2340 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2343 hmap_swap(&new, &ofport->priorities);
2351 /* Expires all MAC learning entries associated with 'bundle' and forces its
2352 * ofproto to revalidate every flow.
2354 * Normally MAC learning entries are removed only from the ofproto associated
2355 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2356 * are removed from every ofproto. When patch ports and SLB bonds are in use
2357 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2358 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2359 * with the host from which it migrated. */
2361 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2363 struct ofproto_dpif *ofproto = bundle->ofproto;
2364 struct mac_learning *ml = ofproto->ml;
2365 struct mac_entry *mac, *next_mac;
2367 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2368 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2369 if (mac->port.p == bundle) {
2371 struct ofproto_dpif *o;
2373 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2375 struct mac_entry *e;
2377 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan,
2380 mac_learning_expire(o->ml, e);
2386 mac_learning_expire(ml, mac);
2391 static struct ofbundle *
2392 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2394 struct ofbundle *bundle;
2396 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2397 &ofproto->bundles) {
2398 if (bundle->aux == aux) {
2405 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
2406 * ones that are found to 'bundles'. */
2408 bundle_lookup_multiple(struct ofproto_dpif *ofproto,
2409 void **auxes, size_t n_auxes,
2410 struct hmapx *bundles)
2414 hmapx_init(bundles);
2415 for (i = 0; i < n_auxes; i++) {
2416 struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
2418 hmapx_add(bundles, bundle);
2424 bundle_update(struct ofbundle *bundle)
2426 struct ofport_dpif *port;
2428 bundle->floodable = true;
2429 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2430 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2431 || !stp_forward_in_state(port->stp_state)) {
2432 bundle->floodable = false;
2439 bundle_del_port(struct ofport_dpif *port)
2441 struct ofbundle *bundle = port->bundle;
2443 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2445 list_remove(&port->bundle_node);
2446 port->bundle = NULL;
2449 lacp_slave_unregister(bundle->lacp, port);
2452 bond_slave_unregister(bundle->bond, port);
2455 bundle_update(bundle);
2459 bundle_add_port(struct ofbundle *bundle, uint16_t ofp_port,
2460 struct lacp_slave_settings *lacp)
2462 struct ofport_dpif *port;
2464 port = get_ofp_port(bundle->ofproto, ofp_port);
2469 if (port->bundle != bundle) {
2470 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2472 bundle_del_port(port);
2475 port->bundle = bundle;
2476 list_push_back(&bundle->ports, &port->bundle_node);
2477 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2478 || !stp_forward_in_state(port->stp_state)) {
2479 bundle->floodable = false;
2483 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2484 lacp_slave_register(bundle->lacp, port, lacp);
2491 bundle_destroy(struct ofbundle *bundle)
2493 struct ofproto_dpif *ofproto;
2494 struct ofport_dpif *port, *next_port;
2501 ofproto = bundle->ofproto;
2502 for (i = 0; i < MAX_MIRRORS; i++) {
2503 struct ofmirror *m = ofproto->mirrors[i];
2505 if (m->out == bundle) {
2507 } else if (hmapx_find_and_delete(&m->srcs, bundle)
2508 || hmapx_find_and_delete(&m->dsts, bundle)) {
2509 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2514 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2515 bundle_del_port(port);
2518 bundle_flush_macs(bundle, true);
2519 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2521 free(bundle->trunks);
2522 lacp_destroy(bundle->lacp);
2523 bond_destroy(bundle->bond);
2528 bundle_set(struct ofproto *ofproto_, void *aux,
2529 const struct ofproto_bundle_settings *s)
2531 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2532 bool need_flush = false;
2533 struct ofport_dpif *port;
2534 struct ofbundle *bundle;
2535 unsigned long *trunks;
2541 bundle_destroy(bundle_lookup(ofproto, aux));
2545 ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2546 ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2548 bundle = bundle_lookup(ofproto, aux);
2550 bundle = xmalloc(sizeof *bundle);
2552 bundle->ofproto = ofproto;
2553 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2554 hash_pointer(aux, 0));
2556 bundle->name = NULL;
2558 list_init(&bundle->ports);
2559 bundle->vlan_mode = PORT_VLAN_TRUNK;
2561 bundle->trunks = NULL;
2562 bundle->use_priority_tags = s->use_priority_tags;
2563 bundle->lacp = NULL;
2564 bundle->bond = NULL;
2566 bundle->floodable = true;
2568 bundle->src_mirrors = 0;
2569 bundle->dst_mirrors = 0;
2570 bundle->mirror_out = 0;
2573 if (!bundle->name || strcmp(s->name, bundle->name)) {
2575 bundle->name = xstrdup(s->name);
2580 if (!bundle->lacp) {
2581 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2582 bundle->lacp = lacp_create();
2584 lacp_configure(bundle->lacp, s->lacp);
2586 lacp_destroy(bundle->lacp);
2587 bundle->lacp = NULL;
2590 /* Update set of ports. */
2592 for (i = 0; i < s->n_slaves; i++) {
2593 if (!bundle_add_port(bundle, s->slaves[i],
2594 s->lacp ? &s->lacp_slaves[i] : NULL)) {
2598 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2599 struct ofport_dpif *next_port;
2601 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2602 for (i = 0; i < s->n_slaves; i++) {
2603 if (s->slaves[i] == port->up.ofp_port) {
2608 bundle_del_port(port);
2612 ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2614 if (list_is_empty(&bundle->ports)) {
2615 bundle_destroy(bundle);
2619 /* Set VLAN tagging mode */
2620 if (s->vlan_mode != bundle->vlan_mode
2621 || s->use_priority_tags != bundle->use_priority_tags) {
2622 bundle->vlan_mode = s->vlan_mode;
2623 bundle->use_priority_tags = s->use_priority_tags;
2628 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2629 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2631 if (vlan != bundle->vlan) {
2632 bundle->vlan = vlan;
2636 /* Get trunked VLANs. */
2637 switch (s->vlan_mode) {
2638 case PORT_VLAN_ACCESS:
2642 case PORT_VLAN_TRUNK:
2643 trunks = CONST_CAST(unsigned long *, s->trunks);
2646 case PORT_VLAN_NATIVE_UNTAGGED:
2647 case PORT_VLAN_NATIVE_TAGGED:
2648 if (vlan != 0 && (!s->trunks
2649 || !bitmap_is_set(s->trunks, vlan)
2650 || bitmap_is_set(s->trunks, 0))) {
2651 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2653 trunks = bitmap_clone(s->trunks, 4096);
2655 trunks = bitmap_allocate1(4096);
2657 bitmap_set1(trunks, vlan);
2658 bitmap_set0(trunks, 0);
2660 trunks = CONST_CAST(unsigned long *, s->trunks);
2667 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2668 free(bundle->trunks);
2669 if (trunks == s->trunks) {
2670 bundle->trunks = vlan_bitmap_clone(trunks);
2672 bundle->trunks = trunks;
2677 if (trunks != s->trunks) {
2682 if (!list_is_short(&bundle->ports)) {
2683 bundle->ofproto->has_bonded_bundles = true;
2685 if (bond_reconfigure(bundle->bond, s->bond)) {
2686 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2689 bundle->bond = bond_create(s->bond);
2690 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2693 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2694 bond_slave_register(bundle->bond, port, port->up.netdev);
2697 bond_destroy(bundle->bond);
2698 bundle->bond = NULL;
2701 /* If we changed something that would affect MAC learning, un-learn
2702 * everything on this port and force flow revalidation. */
2704 bundle_flush_macs(bundle, false);
2711 bundle_remove(struct ofport *port_)
2713 struct ofport_dpif *port = ofport_dpif_cast(port_);
2714 struct ofbundle *bundle = port->bundle;
2717 bundle_del_port(port);
2718 if (list_is_empty(&bundle->ports)) {
2719 bundle_destroy(bundle);
2720 } else if (list_is_short(&bundle->ports)) {
2721 bond_destroy(bundle->bond);
2722 bundle->bond = NULL;
2728 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2730 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2731 struct ofport_dpif *port = port_;
2732 uint8_t ea[ETH_ADDR_LEN];
2735 error = netdev_get_etheraddr(port->up.netdev, ea);
2737 struct ofpbuf packet;
2740 ofpbuf_init(&packet, 0);
2741 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2743 memcpy(packet_pdu, pdu, pdu_size);
2745 send_packet(port, &packet);
2746 ofpbuf_uninit(&packet);
2748 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2749 "%s (%s)", port->bundle->name,
2750 netdev_get_name(port->up.netdev), strerror(error));
2755 bundle_send_learning_packets(struct ofbundle *bundle)
2757 struct ofproto_dpif *ofproto = bundle->ofproto;
2758 int error, n_packets, n_errors;
2759 struct mac_entry *e;
2761 error = n_packets = n_errors = 0;
2762 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2763 if (e->port.p != bundle) {
2764 struct ofpbuf *learning_packet;
2765 struct ofport_dpif *port;
2769 /* The assignment to "port" is unnecessary but makes "grep"ing for
2770 * struct ofport_dpif more effective. */
2771 learning_packet = bond_compose_learning_packet(bundle->bond,
2775 ret = send_packet(port, learning_packet);
2776 ofpbuf_delete(learning_packet);
2786 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2787 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2788 "packets, last error was: %s",
2789 bundle->name, n_errors, n_packets, strerror(error));
2791 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2792 bundle->name, n_packets);
2797 bundle_run(struct ofbundle *bundle)
2800 lacp_run(bundle->lacp, send_pdu_cb);
2803 struct ofport_dpif *port;
2805 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2806 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2809 bond_run(bundle->bond, &bundle->ofproto->backer->revalidate_set,
2810 lacp_status(bundle->lacp));
2811 if (bond_should_send_learning_packets(bundle->bond)) {
2812 bundle_send_learning_packets(bundle);
2818 bundle_wait(struct ofbundle *bundle)
2821 lacp_wait(bundle->lacp);
2824 bond_wait(bundle->bond);
2831 mirror_scan(struct ofproto_dpif *ofproto)
2835 for (idx = 0; idx < MAX_MIRRORS; idx++) {
2836 if (!ofproto->mirrors[idx]) {
2843 static struct ofmirror *
2844 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
2848 for (i = 0; i < MAX_MIRRORS; i++) {
2849 struct ofmirror *mirror = ofproto->mirrors[i];
2850 if (mirror && mirror->aux == aux) {
2858 /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */
2860 mirror_update_dups(struct ofproto_dpif *ofproto)
2864 for (i = 0; i < MAX_MIRRORS; i++) {
2865 struct ofmirror *m = ofproto->mirrors[i];
2868 m->dup_mirrors = MIRROR_MASK_C(1) << i;
2872 for (i = 0; i < MAX_MIRRORS; i++) {
2873 struct ofmirror *m1 = ofproto->mirrors[i];
2880 for (j = i + 1; j < MAX_MIRRORS; j++) {
2881 struct ofmirror *m2 = ofproto->mirrors[j];
2883 if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) {
2884 m1->dup_mirrors |= MIRROR_MASK_C(1) << j;
2885 m2->dup_mirrors |= m1->dup_mirrors;
2892 mirror_set(struct ofproto *ofproto_, void *aux,
2893 const struct ofproto_mirror_settings *s)
2895 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2896 mirror_mask_t mirror_bit;
2897 struct ofbundle *bundle;
2898 struct ofmirror *mirror;
2899 struct ofbundle *out;
2900 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
2901 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
2904 mirror = mirror_lookup(ofproto, aux);
2906 mirror_destroy(mirror);
2912 idx = mirror_scan(ofproto);
2914 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
2916 ofproto->up.name, MAX_MIRRORS, s->name);
2920 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
2921 mirror->ofproto = ofproto;
2924 mirror->out_vlan = -1;
2925 mirror->name = NULL;
2928 if (!mirror->name || strcmp(s->name, mirror->name)) {
2930 mirror->name = xstrdup(s->name);
2933 /* Get the new configuration. */
2934 if (s->out_bundle) {
2935 out = bundle_lookup(ofproto, s->out_bundle);
2937 mirror_destroy(mirror);
2943 out_vlan = s->out_vlan;
2945 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
2946 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
2948 /* If the configuration has not changed, do nothing. */
2949 if (hmapx_equals(&srcs, &mirror->srcs)
2950 && hmapx_equals(&dsts, &mirror->dsts)
2951 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
2952 && mirror->out == out
2953 && mirror->out_vlan == out_vlan)
2955 hmapx_destroy(&srcs);
2956 hmapx_destroy(&dsts);
2960 hmapx_swap(&srcs, &mirror->srcs);
2961 hmapx_destroy(&srcs);
2963 hmapx_swap(&dsts, &mirror->dsts);
2964 hmapx_destroy(&dsts);
2966 free(mirror->vlans);
2967 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
2970 mirror->out_vlan = out_vlan;
2972 /* Update bundles. */
2973 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2974 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
2975 if (hmapx_contains(&mirror->srcs, bundle)) {
2976 bundle->src_mirrors |= mirror_bit;
2978 bundle->src_mirrors &= ~mirror_bit;
2981 if (hmapx_contains(&mirror->dsts, bundle)) {
2982 bundle->dst_mirrors |= mirror_bit;
2984 bundle->dst_mirrors &= ~mirror_bit;
2987 if (mirror->out == bundle) {
2988 bundle->mirror_out |= mirror_bit;
2990 bundle->mirror_out &= ~mirror_bit;
2994 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2995 ofproto->has_mirrors = true;
2996 mac_learning_flush(ofproto->ml,
2997 &ofproto->backer->revalidate_set);
2998 mirror_update_dups(ofproto);
3004 mirror_destroy(struct ofmirror *mirror)
3006 struct ofproto_dpif *ofproto;
3007 mirror_mask_t mirror_bit;
3008 struct ofbundle *bundle;
3015 ofproto = mirror->ofproto;
3016 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3017 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
3019 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
3020 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
3021 bundle->src_mirrors &= ~mirror_bit;
3022 bundle->dst_mirrors &= ~mirror_bit;
3023 bundle->mirror_out &= ~mirror_bit;
3026 hmapx_destroy(&mirror->srcs);
3027 hmapx_destroy(&mirror->dsts);
3028 free(mirror->vlans);
3030 ofproto->mirrors[mirror->idx] = NULL;
3034 mirror_update_dups(ofproto);
3036 ofproto->has_mirrors = false;
3037 for (i = 0; i < MAX_MIRRORS; i++) {
3038 if (ofproto->mirrors[i]) {
3039 ofproto->has_mirrors = true;
3046 mirror_get_stats(struct ofproto *ofproto_, void *aux,
3047 uint64_t *packets, uint64_t *bytes)
3049 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3050 struct ofmirror *mirror = mirror_lookup(ofproto, aux);
3053 *packets = *bytes = UINT64_MAX;
3059 *packets = mirror->packet_count;
3060 *bytes = mirror->byte_count;
3066 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
3068 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3069 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
3070 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
3076 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
3078 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3079 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
3080 return bundle && bundle->mirror_out != 0;
3084 forward_bpdu_changed(struct ofproto *ofproto_)
3086 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3087 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3091 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
3094 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3095 mac_learning_set_idle_time(ofproto->ml, idle_time);
3096 mac_learning_set_max_entries(ofproto->ml, max_entries);
3101 static struct ofport_dpif *
3102 get_ofp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
3104 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
3105 return ofport ? ofport_dpif_cast(ofport) : NULL;
3108 static struct ofport_dpif *
3109 get_odp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
3111 struct ofport_dpif *port = odp_port_to_ofport(ofproto->backer, odp_port);
3112 return port && &ofproto->up == port->up.ofproto ? port : NULL;
3116 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
3117 struct ofproto_port *ofproto_port,
3118 struct dpif_port *dpif_port)
3120 ofproto_port->name = dpif_port->name;
3121 ofproto_port->type = dpif_port->type;
3122 ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
3125 static struct ofport_dpif *
3126 ofport_get_peer(const struct ofport_dpif *ofport_dpif)
3128 const struct ofproto_dpif *ofproto;
3131 peer = netdev_vport_patch_peer(ofport_dpif->up.netdev);
3136 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
3137 struct ofport *ofport;
3139 ofport = shash_find_data(&ofproto->up.port_by_name, peer);
3140 if (ofport && ofport->ofproto->ofproto_class == &ofproto_dpif_class) {
3141 return ofport_dpif_cast(ofport);
3148 port_run_fast(struct ofport_dpif *ofport)
3150 if (ofport->cfm && cfm_should_send_ccm(ofport->cfm)) {
3151 struct ofpbuf packet;
3153 ofpbuf_init(&packet, 0);
3154 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
3155 send_packet(ofport, &packet);
3156 ofpbuf_uninit(&packet);
3159 if (ofport->bfd && bfd_should_send_packet(ofport->bfd)) {
3160 struct ofpbuf packet;
3162 ofpbuf_init(&packet, 0);
3163 bfd_put_packet(ofport->bfd, &packet, ofport->up.pp.hw_addr);
3164 send_packet(ofport, &packet);
3165 ofpbuf_uninit(&packet);
3170 port_run(struct ofport_dpif *ofport)
3172 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
3173 bool carrier_changed = carrier_seq != ofport->carrier_seq;
3174 bool enable = netdev_get_carrier(ofport->up.netdev);
3176 ofport->carrier_seq = carrier_seq;
3178 port_run_fast(ofport);
3180 if (ofport->tnl_port
3181 && tnl_port_reconfigure(&ofport->up, ofport->odp_port,
3182 &ofport->tnl_port)) {
3183 ofproto_dpif_cast(ofport->up.ofproto)->backer->need_revalidate = true;
3187 int cfm_opup = cfm_get_opup(ofport->cfm);
3189 cfm_run(ofport->cfm);
3190 enable = enable && !cfm_get_fault(ofport->cfm);
3192 if (cfm_opup >= 0) {
3193 enable = enable && cfm_opup;
3198 bfd_run(ofport->bfd);
3199 enable = enable && bfd_forwarding(ofport->bfd);
3202 if (ofport->bundle) {
3203 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
3204 if (carrier_changed) {
3205 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
3209 if (ofport->may_enable != enable) {
3210 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3212 if (ofproto->has_bundle_action) {
3213 ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
3217 ofport->may_enable = enable;
3221 port_wait(struct ofport_dpif *ofport)
3224 cfm_wait(ofport->cfm);
3228 bfd_wait(ofport->bfd);
3233 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
3234 struct ofproto_port *ofproto_port)
3236 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3237 struct dpif_port dpif_port;
3240 if (sset_contains(&ofproto->ghost_ports, devname)) {
3241 const char *type = netdev_get_type_from_name(devname);
3243 /* We may be called before ofproto->up.port_by_name is populated with
3244 * the appropriate ofport. For this reason, we must get the name and
3245 * type from the netdev layer directly. */
3247 const struct ofport *ofport;
3249 ofport = shash_find_data(&ofproto->up.port_by_name, devname);
3250 ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
3251 ofproto_port->name = xstrdup(devname);
3252 ofproto_port->type = xstrdup(type);
3258 if (!sset_contains(&ofproto->ports, devname)) {
3261 error = dpif_port_query_by_name(ofproto->backer->dpif,
3262 devname, &dpif_port);
3264 ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
3270 port_add(struct ofproto *ofproto_, struct netdev *netdev)
3272 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3273 const char *dp_port_name = netdev_vport_get_dpif_port(netdev);
3274 const char *devname = netdev_get_name(netdev);
3276 if (netdev_vport_is_patch(netdev)) {
3277 sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
3281 if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
3282 uint32_t port_no = UINT32_MAX;
3285 error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
3289 if (netdev_get_tunnel_config(netdev)) {
3290 simap_put(&ofproto->backer->tnl_backers, dp_port_name, port_no);
3294 if (netdev_get_tunnel_config(netdev)) {
3295 sset_add(&ofproto->ghost_ports, devname);
3297 sset_add(&ofproto->ports, devname);
3303 port_del(struct ofproto *ofproto_, uint16_t ofp_port)
3305 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3306 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
3313 sset_find_and_delete(&ofproto->ghost_ports,
3314 netdev_get_name(ofport->up.netdev));
3315 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3316 if (!ofport->tnl_port) {
3317 error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
3319 /* The caller is going to close ofport->up.netdev. If this is a
3320 * bonded port, then the bond is using that netdev, so remove it
3321 * from the bond. The client will need to reconfigure everything
3322 * after deleting ports, so then the slave will get re-added. */
3323 bundle_remove(&ofport->up);
3330 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
3332 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3337 error = netdev_get_stats(ofport->up.netdev, stats);
3339 if (!error && ofport_->ofp_port == OFPP_LOCAL) {
3340 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3342 /* ofproto->stats.tx_packets represents packets that we created
3343 * internally and sent to some port (e.g. packets sent with
3344 * send_packet()). Account for them as if they had come from
3345 * OFPP_LOCAL and got forwarded. */
3347 if (stats->rx_packets != UINT64_MAX) {
3348 stats->rx_packets += ofproto->stats.tx_packets;
3351 if (stats->rx_bytes != UINT64_MAX) {
3352 stats->rx_bytes += ofproto->stats.tx_bytes;
3355 /* ofproto->stats.rx_packets represents packets that were received on
3356 * some port and we processed internally and dropped (e.g. STP).
3357 * Account for them as if they had been forwarded to OFPP_LOCAL. */
3359 if (stats->tx_packets != UINT64_MAX) {
3360 stats->tx_packets += ofproto->stats.rx_packets;
3363 if (stats->tx_bytes != UINT64_MAX) {
3364 stats->tx_bytes += ofproto->stats.rx_bytes;
3371 struct port_dump_state {
3376 struct ofproto_port port;
3381 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
3383 *statep = xzalloc(sizeof(struct port_dump_state));
3388 port_dump_next(const struct ofproto *ofproto_, void *state_,
3389 struct ofproto_port *port)
3391 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3392 struct port_dump_state *state = state_;
3393 const struct sset *sset;
3394 struct sset_node *node;
3396 if (state->has_port) {
3397 ofproto_port_destroy(&state->port);
3398 state->has_port = false;
3400 sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
3401 while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
3404 error = port_query_by_name(ofproto_, node->name, &state->port);
3406 *port = state->port;
3407 state->has_port = true;
3409 } else if (error != ENODEV) {
3414 if (!state->ghost) {
3415 state->ghost = true;
3418 return port_dump_next(ofproto_, state_, port);
3425 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3427 struct port_dump_state *state = state_;
3429 if (state->has_port) {
3430 ofproto_port_destroy(&state->port);
3437 port_poll(const struct ofproto *ofproto_, char **devnamep)
3439 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3441 if (ofproto->port_poll_errno) {
3442 int error = ofproto->port_poll_errno;
3443 ofproto->port_poll_errno = 0;
3447 if (sset_is_empty(&ofproto->port_poll_set)) {
3451 *devnamep = sset_pop(&ofproto->port_poll_set);
3456 port_poll_wait(const struct ofproto *ofproto_)
3458 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3459 dpif_port_poll_wait(ofproto->backer->dpif);
3463 port_is_lacp_current(const struct ofport *ofport_)
3465 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3466 return (ofport->bundle && ofport->bundle->lacp
3467 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3471 /* Upcall handling. */
3473 /* Flow miss batching.
3475 * Some dpifs implement operations faster when you hand them off in a batch.
3476 * To allow batching, "struct flow_miss" queues the dpif-related work needed
3477 * for a given flow. Each "struct flow_miss" corresponds to sending one or
3478 * more packets, plus possibly installing the flow in the dpif.
3480 * So far we only batch the operations that affect flow setup time the most.
3481 * It's possible to batch more than that, but the benefit might be minimal. */
3483 struct hmap_node hmap_node;
3484 struct ofproto_dpif *ofproto;
3486 enum odp_key_fitness key_fitness;
3487 const struct nlattr *key;
3489 struct initial_vals initial_vals;
3490 struct list packets;
3491 enum dpif_upcall_type upcall_type;
3492 uint32_t odp_in_port;
3495 struct flow_miss_op {
3496 struct dpif_op dpif_op;
3497 void *garbage; /* Pointer to pass to free(), NULL if none. */
3498 uint64_t stub[1024 / 8]; /* Temporary buffer. */
3501 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
3502 * OpenFlow controller as necessary according to their individual
3503 * configurations. */
3505 send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet,
3506 const struct flow *flow)
3508 struct ofputil_packet_in pin;
3510 pin.packet = packet->data;
3511 pin.packet_len = packet->size;
3512 pin.reason = OFPR_NO_MATCH;
3513 pin.controller_id = 0;
3518 pin.send_len = 0; /* not used for flow table misses */
3520 flow_get_metadata(flow, &pin.fmd);
3522 connmgr_send_packet_in(ofproto->up.connmgr, &pin);
3525 static enum slow_path_reason
3526 process_special(struct ofproto_dpif *ofproto, const struct flow *flow,
3527 const struct ofport_dpif *ofport, const struct ofpbuf *packet)
3531 } else if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) {
3533 cfm_process_heartbeat(ofport->cfm, packet);
3536 } else if (ofport->bfd && bfd_should_process_flow(flow)) {
3538 bfd_process_packet(ofport->bfd, flow, packet);
3541 } else if (ofport->bundle && ofport->bundle->lacp
3542 && flow->dl_type == htons(ETH_TYPE_LACP)) {
3544 lacp_process_packet(ofport->bundle->lacp, ofport, packet);
3547 } else if (ofproto->stp && stp_should_process_flow(flow)) {
3549 stp_process_packet(ofport, packet);
3557 static struct flow_miss *
3558 flow_miss_find(struct hmap *todo, const struct ofproto_dpif *ofproto,
3559 const struct flow *flow, uint32_t hash)
3561 struct flow_miss *miss;
3563 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
3564 if (miss->ofproto == ofproto && flow_equal(&miss->flow, flow)) {
3572 /* Partially Initializes 'op' as an "execute" operation for 'miss' and
3573 * 'packet'. The caller must initialize op->actions and op->actions_len. If
3574 * 'miss' is associated with a subfacet the caller must also initialize the
3575 * returned op->subfacet, and if anything needs to be freed after processing
3576 * the op, the caller must initialize op->garbage also. */
3578 init_flow_miss_execute_op(struct flow_miss *miss, struct ofpbuf *packet,
3579 struct flow_miss_op *op)
3581 if (miss->flow.vlan_tci != miss->initial_vals.vlan_tci) {
3582 /* This packet was received on a VLAN splinter port. We
3583 * added a VLAN to the packet to make the packet resemble
3584 * the flow, but the actions were composed assuming that
3585 * the packet contained no VLAN. So, we must remove the
3586 * VLAN header from the packet before trying to execute the
3588 eth_pop_vlan(packet);
3592 op->dpif_op.type = DPIF_OP_EXECUTE;
3593 op->dpif_op.u.execute.key = miss->key;
3594 op->dpif_op.u.execute.key_len = miss->key_len;
3595 op->dpif_op.u.execute.packet = packet;
3598 /* Helper for handle_flow_miss_without_facet() and
3599 * handle_flow_miss_with_facet(). */
3601 handle_flow_miss_common(struct rule_dpif *rule,
3602 struct ofpbuf *packet, const struct flow *flow)
3604 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3606 ofproto->n_matches++;
3608 if (rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
3610 * Extra-special case for fail-open mode.
3612 * We are in fail-open mode and the packet matched the fail-open
3613 * rule, but we are connected to a controller too. We should send
3614 * the packet up to the controller in the hope that it will try to
3615 * set up a flow and thereby allow us to exit fail-open.
3617 * See the top-level comment in fail-open.c for more information.
3619 send_packet_in_miss(ofproto, packet, flow);
3623 /* Figures out whether a flow that missed in 'ofproto', whose details are in
3624 * 'miss', is likely to be worth tracking in detail in userspace and (usually)
3625 * installing a datapath flow. The answer is usually "yes" (a return value of
3626 * true). However, for short flows the cost of bookkeeping is much higher than
3627 * the benefits, so when the datapath holds a large number of flows we impose
3628 * some heuristics to decide which flows are likely to be worth tracking. */
3630 flow_miss_should_make_facet(struct ofproto_dpif *ofproto,
3631 struct flow_miss *miss, uint32_t hash)
3633 if (!ofproto->governor) {
3636 n_subfacets = hmap_count(&ofproto->subfacets);
3637 if (n_subfacets * 2 <= ofproto->up.flow_eviction_threshold) {
3641 ofproto->governor = governor_create(ofproto->up.name);
3644 return governor_should_install_flow(ofproto->governor, hash,
3645 list_size(&miss->packets));
3648 /* Handles 'miss', which matches 'rule', without creating a facet or subfacet
3649 * or creating any datapath flow. May add an "execute" operation to 'ops' and
3650 * increment '*n_ops'. */
3652 handle_flow_miss_without_facet(struct flow_miss *miss,
3653 struct flow_miss_op *ops, size_t *n_ops)
3655 struct rule_dpif *rule = rule_dpif_lookup(miss->ofproto, &miss->flow);
3656 long long int now = time_msec();
3657 struct action_xlate_ctx ctx;
3658 struct ofpbuf *packet;
3660 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3661 struct flow_miss_op *op = &ops[*n_ops];
3662 struct dpif_flow_stats stats;
3663 struct ofpbuf odp_actions;
3665 COVERAGE_INC(facet_suppress);
3667 handle_flow_miss_common(rule, packet, &miss->flow);
3669 ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub);
3671 dpif_flow_stats_extract(&miss->flow, packet, now, &stats);
3672 rule_credit_stats(rule, &stats);
3674 action_xlate_ctx_init(&ctx, miss->ofproto, &miss->flow,
3675 &miss->initial_vals, rule, stats.tcp_flags,
3677 ctx.resubmit_stats = &stats;
3678 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len,
3681 if (odp_actions.size) {
3682 struct dpif_execute *execute = &op->dpif_op.u.execute;
3684 init_flow_miss_execute_op(miss, packet, op);
3685 execute->actions = odp_actions.data;
3686 execute->actions_len = odp_actions.size;
3687 op->garbage = ofpbuf_get_uninit_pointer(&odp_actions);
3691 ofpbuf_uninit(&odp_actions);
3696 /* Handles 'miss', which matches 'facet'. May add any required datapath
3697 * operations to 'ops', incrementing '*n_ops' for each new op.
3699 * All of the packets in 'miss' are considered to have arrived at time 'now'.
3700 * This is really important only for new facets: if we just called time_msec()
3701 * here, then the new subfacet or its packets could look (occasionally) as
3702 * though it was used some time after the facet was used. That can make a
3703 * one-packet flow look like it has a nonzero duration, which looks odd in
3704 * e.g. NetFlow statistics. */
3706 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
3708 struct flow_miss_op *ops, size_t *n_ops)
3710 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3711 enum subfacet_path want_path;
3712 struct subfacet *subfacet;
3713 struct ofpbuf *packet;
3715 subfacet = subfacet_create(facet, miss, now);
3716 want_path = subfacet->facet->slow ? SF_SLOW_PATH : SF_FAST_PATH;
3718 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3719 struct flow_miss_op *op = &ops[*n_ops];
3720 struct dpif_flow_stats stats;
3722 handle_flow_miss_common(facet->rule, packet, &miss->flow);
3724 if (want_path != SF_FAST_PATH) {
3725 struct action_xlate_ctx ctx;
3727 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
3728 &facet->initial_vals, facet->rule, 0,
3730 xlate_actions_for_side_effects(&ctx, facet->rule->up.ofpacts,
3731 facet->rule->up.ofpacts_len);
3734 dpif_flow_stats_extract(&facet->flow, packet, now, &stats);
3735 subfacet_update_stats(subfacet, &stats);
3737 if (facet->odp_actions.size) {
3738 struct dpif_execute *execute = &op->dpif_op.u.execute;
3740 init_flow_miss_execute_op(miss, packet, op);
3741 execute->actions = facet->odp_actions.data,
3742 execute->actions_len = facet->odp_actions.size;
3747 if (miss->upcall_type == DPIF_UC_MISS || subfacet->path != want_path) {
3748 struct flow_miss_op *op = &ops[(*n_ops)++];
3749 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
3751 subfacet->path = want_path;
3754 op->dpif_op.type = DPIF_OP_FLOW_PUT;
3755 put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
3756 put->key = miss->key;
3757 put->key_len = miss->key_len;
3758 if (want_path == SF_FAST_PATH) {
3759 put->actions = facet->odp_actions.data;
3760 put->actions_len = facet->odp_actions.size;
3762 compose_slow_path(ofproto, &facet->flow, facet->slow,
3763 op->stub, sizeof op->stub,
3764 &put->actions, &put->actions_len);
3770 /* Handles flow miss 'miss'. May add any required datapath operations
3771 * to 'ops', incrementing '*n_ops' for each new op. */
3773 handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops,
3776 struct ofproto_dpif *ofproto = miss->ofproto;
3777 struct facet *facet;
3781 /* The caller must ensure that miss->hmap_node.hash contains
3782 * flow_hash(miss->flow, 0). */
3783 hash = miss->hmap_node.hash;
3785 facet = facet_lookup_valid(ofproto, &miss->flow, hash);
3787 /* There does not exist a bijection between 'struct flow' and datapath
3788 * flow keys with fitness ODP_FIT_TO_LITTLE. This breaks a fundamental
3789 * assumption used throughout the facet and subfacet handling code.
3790 * Since we have to handle these misses in userspace anyway, we simply
3791 * skip facet creation, avoiding the problem alltogether. */
3792 if (miss->key_fitness == ODP_FIT_TOO_LITTLE
3793 || !flow_miss_should_make_facet(ofproto, miss, hash)) {
3794 handle_flow_miss_without_facet(miss, ops, n_ops);
3798 facet = facet_create(miss, hash);
3803 handle_flow_miss_with_facet(miss, facet, now, ops, n_ops);
3806 static struct drop_key *
3807 drop_key_lookup(const struct dpif_backer *backer, const struct nlattr *key,
3810 struct drop_key *drop_key;
3812 HMAP_FOR_EACH_WITH_HASH (drop_key, hmap_node, hash_bytes(key, key_len, 0),
3813 &backer->drop_keys) {
3814 if (drop_key->key_len == key_len
3815 && !memcmp(drop_key->key, key, key_len)) {
3823 drop_key_clear(struct dpif_backer *backer)
3825 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3826 struct drop_key *drop_key, *next;
3828 HMAP_FOR_EACH_SAFE (drop_key, next, hmap_node, &backer->drop_keys) {
3831 error = dpif_flow_del(backer->dpif, drop_key->key, drop_key->key_len,
3833 if (error && !VLOG_DROP_WARN(&rl)) {
3834 struct ds ds = DS_EMPTY_INITIALIZER;
3835 odp_flow_key_format(drop_key->key, drop_key->key_len, &ds);
3836 VLOG_WARN("Failed to delete drop key (%s) (%s)", strerror(error),
3841 hmap_remove(&backer->drop_keys, &drop_key->hmap_node);
3842 free(drop_key->key);
3847 /* Given a datpath, packet, and flow metadata ('backer', 'packet', and 'key'
3848 * respectively), populates 'flow' with the result of odp_flow_key_to_flow().
3849 * Optionally, if nonnull, populates 'fitnessp' with the fitness of 'flow' as
3850 * returned by odp_flow_key_to_flow(). Also, optionally populates 'ofproto'
3851 * with the ofproto_dpif, and 'odp_in_port' with the datapath in_port, that
3852 * 'packet' ingressed.
3854 * If 'ofproto' is nonnull, requires 'flow''s in_port to exist. Otherwise sets
3855 * 'flow''s in_port to OFPP_NONE.
3857 * This function does post-processing on data returned from
3858 * odp_flow_key_to_flow() to help make VLAN splinters transparent to the rest
3859 * of the upcall processing logic. In particular, if the extracted in_port is
3860 * a VLAN splinter port, it replaces flow->in_port by the "real" port, sets
3861 * flow->vlan_tci correctly for the VLAN of the VLAN splinter port, and pushes
3862 * a VLAN header onto 'packet' (if it is nonnull).
3864 * Optionally, if 'initial_vals' is nonnull, sets 'initial_vals->vlan_tci'
3865 * to the VLAN TCI with which the packet was really received, that is, the
3866 * actual VLAN TCI extracted by odp_flow_key_to_flow(). (This differs from
3867 * the value returned in flow->vlan_tci only for packets received on
3870 * Similarly, this function also includes some logic to help with tunnels. It
3871 * may modify 'flow' as necessary to make the tunneling implementation
3872 * transparent to the upcall processing logic.
3874 * Returns 0 if successful, ENODEV if the parsed flow has no associated ofport,
3875 * or some other positive errno if there are other problems. */
3877 ofproto_receive(const struct dpif_backer *backer, struct ofpbuf *packet,
3878 const struct nlattr *key, size_t key_len,
3879 struct flow *flow, enum odp_key_fitness *fitnessp,
3880 struct ofproto_dpif **ofproto, uint32_t *odp_in_port,
3881 struct initial_vals *initial_vals)
3883 const struct ofport_dpif *port;
3884 enum odp_key_fitness fitness;
3887 fitness = odp_flow_key_to_flow(key, key_len, flow);
3888 if (fitness == ODP_FIT_ERROR) {
3894 initial_vals->vlan_tci = flow->vlan_tci;
3898 *odp_in_port = flow->in_port;
3901 port = (tnl_port_should_receive(flow)
3902 ? ofport_dpif_cast(tnl_port_receive(flow))
3903 : odp_port_to_ofport(backer, flow->in_port));
3904 flow->in_port = port ? port->up.ofp_port : OFPP_NONE;
3909 /* XXX: Since the tunnel module is not scoped per backer, for a tunnel port
3910 * it's theoretically possible that we'll receive an ofport belonging to an
3911 * entirely different datapath. In practice, this can't happen because no
3912 * platforms has two separate datapaths which each support tunneling. */
3913 ovs_assert(ofproto_dpif_cast(port->up.ofproto)->backer == backer);
3915 if (vsp_adjust_flow(ofproto_dpif_cast(port->up.ofproto), flow)) {
3917 /* Make the packet resemble the flow, so that it gets sent to
3918 * an OpenFlow controller properly, so that it looks correct
3919 * for sFlow, and so that flow_extract() will get the correct
3920 * vlan_tci if it is called on 'packet'.
3922 * The allocated space inside 'packet' probably also contains
3923 * 'key', that is, both 'packet' and 'key' are probably part of
3924 * a struct dpif_upcall (see the large comment on that
3925 * structure definition), so pushing data on 'packet' is in
3926 * general not a good idea since it could overwrite 'key' or
3927 * free it as a side effect. However, it's OK in this special
3928 * case because we know that 'packet' is inside a Netlink
3929 * attribute: pushing 4 bytes will just overwrite the 4-byte
3930 * "struct nlattr", which is fine since we don't need that
3931 * header anymore. */
3932 eth_push_vlan(packet, flow->vlan_tci);
3934 /* We can't reproduce 'key' from 'flow'. */
3935 fitness = fitness == ODP_FIT_PERFECT ? ODP_FIT_TOO_MUCH : fitness;
3940 *ofproto = ofproto_dpif_cast(port->up.ofproto);
3945 *fitnessp = fitness;
3951 handle_miss_upcalls(struct dpif_backer *backer, struct dpif_upcall *upcalls,
3954 struct dpif_upcall *upcall;
3955 struct flow_miss *miss;
3956 struct flow_miss misses[FLOW_MISS_MAX_BATCH];
3957 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
3958 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
3968 /* Construct the to-do list.
3970 * This just amounts to extracting the flow from each packet and sticking
3971 * the packets that have the same flow in the same "flow_miss" structure so
3972 * that we can process them together. */
3975 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
3976 struct flow_miss *miss = &misses[n_misses];
3977 struct flow_miss *existing_miss;
3978 struct ofproto_dpif *ofproto;
3979 uint32_t odp_in_port;
3984 error = ofproto_receive(backer, upcall->packet, upcall->key,
3985 upcall->key_len, &flow, &miss->key_fitness,
3986 &ofproto, &odp_in_port, &miss->initial_vals);
3987 if (error == ENODEV) {
3988 struct drop_key *drop_key;
3990 /* Received packet on port for which we couldn't associate
3991 * an ofproto. This can happen if a port is removed while
3992 * traffic is being received. Print a rate-limited message
3993 * in case it happens frequently. Install a drop flow so
3994 * that future packets of the flow are inexpensively dropped
3996 VLOG_INFO_RL(&rl, "received packet on unassociated port %"PRIu32,
3999 drop_key = drop_key_lookup(backer, upcall->key, upcall->key_len);
4001 drop_key = xmalloc(sizeof *drop_key);
4002 drop_key->key = xmemdup(upcall->key, upcall->key_len);
4003 drop_key->key_len = upcall->key_len;
4005 hmap_insert(&backer->drop_keys, &drop_key->hmap_node,
4006 hash_bytes(drop_key->key, drop_key->key_len, 0));
4007 dpif_flow_put(backer->dpif, DPIF_FP_CREATE | DPIF_FP_MODIFY,
4008 drop_key->key, drop_key->key_len, NULL, 0, NULL);
4016 ofproto->n_missed++;
4017 flow_extract(upcall->packet, flow.skb_priority, flow.skb_mark,
4018 &flow.tunnel, flow.in_port, &miss->flow);
4020 /* Add other packets to a to-do list. */
4021 hash = flow_hash(&miss->flow, 0);
4022 existing_miss = flow_miss_find(&todo, ofproto, &miss->flow, hash);
4023 if (!existing_miss) {
4024 hmap_insert(&todo, &miss->hmap_node, hash);
4025 miss->ofproto = ofproto;
4026 miss->key = upcall->key;
4027 miss->key_len = upcall->key_len;
4028 miss->upcall_type = upcall->type;
4029 miss->odp_in_port = odp_in_port;
4030 list_init(&miss->packets);
4034 miss = existing_miss;
4036 list_push_back(&miss->packets, &upcall->packet->list_node);
4039 /* Process each element in the to-do list, constructing the set of
4040 * operations to batch. */
4042 HMAP_FOR_EACH (miss, hmap_node, &todo) {
4043 handle_flow_miss(miss, flow_miss_ops, &n_ops);
4045 ovs_assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
4047 /* Execute batch. */
4048 for (i = 0; i < n_ops; i++) {
4049 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
4051 dpif_operate(backer->dpif, dpif_ops, n_ops);
4054 for (i = 0; i < n_ops; i++) {
4055 free(flow_miss_ops[i].garbage);
4057 hmap_destroy(&todo);
4060 static enum { SFLOW_UPCALL, MISS_UPCALL, BAD_UPCALL, FLOW_SAMPLE_UPCALL,
4062 classify_upcall(const struct dpif_upcall *upcall)
4064 size_t userdata_len;
4065 union user_action_cookie cookie;
4067 /* First look at the upcall type. */
4068 switch (upcall->type) {
4069 case DPIF_UC_ACTION:
4075 case DPIF_N_UC_TYPES:
4077 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
4081 /* "action" upcalls need a closer look. */
4082 if (!upcall->userdata) {
4083 VLOG_WARN_RL(&rl, "action upcall missing cookie");
4086 userdata_len = nl_attr_get_size(upcall->userdata);
4087 if (userdata_len < sizeof cookie.type
4088 || userdata_len > sizeof cookie) {
4089 VLOG_WARN_RL(&rl, "action upcall cookie has unexpected size %zu",
4093 memset(&cookie, 0, sizeof cookie);
4094 memcpy(&cookie, nl_attr_get(upcall->userdata), userdata_len);
4095 if (userdata_len == sizeof cookie.sflow
4096 && cookie.type == USER_ACTION_COOKIE_SFLOW) {
4097 return SFLOW_UPCALL;
4098 } else if (userdata_len == sizeof cookie.slow_path
4099 && cookie.type == USER_ACTION_COOKIE_SLOW_PATH) {
4101 } else if (userdata_len == sizeof cookie.flow_sample
4102 && cookie.type == USER_ACTION_COOKIE_FLOW_SAMPLE) {
4103 return FLOW_SAMPLE_UPCALL;
4104 } else if (userdata_len == sizeof cookie.ipfix
4105 && cookie.type == USER_ACTION_COOKIE_IPFIX) {
4106 return IPFIX_UPCALL;
4108 VLOG_WARN_RL(&rl, "invalid user cookie of type %"PRIu16
4109 " and size %zu", cookie.type, userdata_len);
4115 handle_sflow_upcall(struct dpif_backer *backer,
4116 const struct dpif_upcall *upcall)
4118 struct ofproto_dpif *ofproto;
4119 union user_action_cookie cookie;
4121 uint32_t odp_in_port;
4123 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
4124 &flow, NULL, &ofproto, &odp_in_port, NULL)
4125 || !ofproto->sflow) {
4129 memset(&cookie, 0, sizeof cookie);
4130 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.sflow);
4131 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
4132 odp_in_port, &cookie);
4136 handle_flow_sample_upcall(struct dpif_backer *backer,
4137 const struct dpif_upcall *upcall)
4139 struct ofproto_dpif *ofproto;
4140 union user_action_cookie cookie;
4143 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
4144 &flow, NULL, &ofproto, NULL, NULL)
4145 || !ofproto->ipfix) {
4149 memset(&cookie, 0, sizeof cookie);
4150 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.flow_sample);
4152 /* The flow reflects exactly the contents of the packet. Sample
4153 * the packet using it. */
4154 dpif_ipfix_flow_sample(ofproto->ipfix, upcall->packet, &flow,
4155 cookie.flow_sample.collector_set_id,
4156 cookie.flow_sample.probability,
4157 cookie.flow_sample.obs_domain_id,
4158 cookie.flow_sample.obs_point_id);
4162 handle_ipfix_upcall(struct dpif_backer *backer,
4163 const struct dpif_upcall *upcall)
4165 struct ofproto_dpif *ofproto;
4168 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
4169 &flow, NULL, &ofproto, NULL, NULL)
4170 || !ofproto->ipfix) {
4174 /* The flow reflects exactly the contents of the packet. Sample
4175 * the packet using it. */
4176 dpif_ipfix_bridge_sample(ofproto->ipfix, upcall->packet, &flow);
4180 handle_upcalls(struct dpif_backer *backer, unsigned int max_batch)
4182 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
4183 struct ofpbuf miss_bufs[FLOW_MISS_MAX_BATCH];
4184 uint64_t miss_buf_stubs[FLOW_MISS_MAX_BATCH][4096 / 8];
4189 ovs_assert(max_batch <= FLOW_MISS_MAX_BATCH);
4192 for (n_processed = 0; n_processed < max_batch; n_processed++) {
4193 struct dpif_upcall *upcall = &misses[n_misses];
4194 struct ofpbuf *buf = &miss_bufs[n_misses];
4197 ofpbuf_use_stub(buf, miss_buf_stubs[n_misses],
4198 sizeof miss_buf_stubs[n_misses]);
4199 error = dpif_recv(backer->dpif, upcall, buf);
4205 switch (classify_upcall(upcall)) {
4207 /* Handle it later. */
4212 handle_sflow_upcall(backer, upcall);
4216 case FLOW_SAMPLE_UPCALL:
4217 handle_flow_sample_upcall(backer, upcall);
4222 handle_ipfix_upcall(backer, upcall);
4232 /* Handle deferred MISS_UPCALL processing. */
4233 handle_miss_upcalls(backer, misses, n_misses);
4234 for (i = 0; i < n_misses; i++) {
4235 ofpbuf_uninit(&miss_bufs[i]);
4241 /* Flow expiration. */
4243 static int subfacet_max_idle(const struct ofproto_dpif *);
4244 static void update_stats(struct dpif_backer *);
4245 static void rule_expire(struct rule_dpif *);
4246 static void expire_subfacets(struct ofproto_dpif *, int dp_max_idle);
4248 /* This function is called periodically by run(). Its job is to collect
4249 * updates for the flows that have been installed into the datapath, most
4250 * importantly when they last were used, and then use that information to
4251 * expire flows that have not been used recently.
4253 * Returns the number of milliseconds after which it should be called again. */
4255 expire(struct dpif_backer *backer)
4257 struct ofproto_dpif *ofproto;
4258 int max_idle = INT32_MAX;
4260 /* Periodically clear out the drop keys in an effort to keep them
4261 * relatively few. */
4262 drop_key_clear(backer);
4264 /* Update stats for each flow in the backer. */
4265 update_stats(backer);
4267 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4268 struct rule *rule, *next_rule;
4271 if (ofproto->backer != backer) {
4275 /* Keep track of the max number of flows per ofproto_dpif. */
4276 update_max_subfacet_count(ofproto);
4278 /* Expire subfacets that have been idle too long. */
4279 dp_max_idle = subfacet_max_idle(ofproto);
4280 expire_subfacets(ofproto, dp_max_idle);
4282 max_idle = MIN(max_idle, dp_max_idle);
4284 /* Expire OpenFlow flows whose idle_timeout or hard_timeout
4286 LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
4287 &ofproto->up.expirable) {
4288 rule_expire(rule_dpif_cast(rule));
4291 /* All outstanding data in existing flows has been accounted, so it's a
4292 * good time to do bond rebalancing. */
4293 if (ofproto->has_bonded_bundles) {
4294 struct ofbundle *bundle;
4296 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
4298 bond_rebalance(bundle->bond, &backer->revalidate_set);
4304 return MIN(max_idle, 1000);
4307 /* Updates flow table statistics given that the datapath just reported 'stats'
4308 * as 'subfacet''s statistics. */
4310 update_subfacet_stats(struct subfacet *subfacet,
4311 const struct dpif_flow_stats *stats)
4313 struct facet *facet = subfacet->facet;
4315 if (stats->n_packets >= subfacet->dp_packet_count) {
4316 uint64_t extra = stats->n_packets - subfacet->dp_packet_count;
4317 facet->packet_count += extra;
4319 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
4322 if (stats->n_bytes >= subfacet->dp_byte_count) {
4323 facet->byte_count += stats->n_bytes - subfacet->dp_byte_count;
4325 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
4328 subfacet->dp_packet_count = stats->n_packets;
4329 subfacet->dp_byte_count = stats->n_bytes;
4331 facet->tcp_flags |= stats->tcp_flags;
4333 subfacet_update_time(subfacet, stats->used);
4334 if (facet->accounted_bytes < facet->byte_count) {
4336 facet_account(facet);
4337 facet->accounted_bytes = facet->byte_count;
4341 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
4342 * about, or a flow that shouldn't be installed but was anyway. Delete it. */
4344 delete_unexpected_flow(struct ofproto_dpif *ofproto,
4345 const struct nlattr *key, size_t key_len)
4347 if (!VLOG_DROP_WARN(&rl)) {
4351 odp_flow_key_format(key, key_len, &s);
4352 VLOG_WARN("unexpected flow on %s: %s", ofproto->up.name, ds_cstr(&s));
4356 COVERAGE_INC(facet_unexpected);
4357 dpif_flow_del(ofproto->backer->dpif, key, key_len, NULL);
4360 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
4362 * This function also pushes statistics updates to rules which each facet
4363 * resubmits into. Generally these statistics will be accurate. However, if a
4364 * facet changes the rule it resubmits into at some time in between
4365 * update_stats() runs, it is possible that statistics accrued to the
4366 * old rule will be incorrectly attributed to the new rule. This could be
4367 * avoided by calling update_stats() whenever rules are created or
4368 * deleted. However, the performance impact of making so many calls to the
4369 * datapath do not justify the benefit of having perfectly accurate statistics.
4371 * In addition, this function maintains per ofproto flow hit counts. The patch
4372 * port is not treated specially. e.g. A packet ingress from br0 patched into
4373 * br1 will increase the hit count of br0 by 1, however, does not affect
4374 * the hit or miss counts of br1.
4377 update_stats(struct dpif_backer *backer)
4379 const struct dpif_flow_stats *stats;
4380 struct dpif_flow_dump dump;
4381 const struct nlattr *key;
4382 struct ofproto_dpif *ofproto;
4385 dpif_flow_dump_start(&dump, backer->dpif);
4386 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
4388 struct subfacet *subfacet;
4389 struct ofport_dpif *ofport;
4392 if (ofproto_receive(backer, NULL, key, key_len, &flow, NULL, &ofproto,
4397 ofproto->total_subfacet_count += hmap_count(&ofproto->subfacets);
4398 ofproto->n_update_stats++;
4400 ofport = get_ofp_port(ofproto, flow.in_port);
4401 if (ofport && ofport->tnl_port) {
4402 netdev_vport_inc_rx(ofport->up.netdev, stats);
4405 key_hash = odp_flow_key_hash(key, key_len);
4406 subfacet = subfacet_find(ofproto, key, key_len, key_hash);
4407 switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
4409 /* Update ofproto_dpif's hit count. */
4410 if (stats->n_packets > subfacet->dp_packet_count) {
4411 uint64_t delta = stats->n_packets - subfacet->dp_packet_count;
4412 dpif_stats_update_hit_count(ofproto, delta);
4415 update_subfacet_stats(subfacet, stats);
4419 /* Stats are updated per-packet. */
4422 case SF_NOT_INSTALLED:
4424 delete_unexpected_flow(ofproto, key, key_len);
4429 dpif_flow_dump_done(&dump);
4431 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4432 update_moving_averages(ofproto);
4437 /* Calculates and returns the number of milliseconds of idle time after which
4438 * subfacets should expire from the datapath. When a subfacet expires, we fold
4439 * its statistics into its facet, and when a facet's last subfacet expires, we
4440 * fold its statistic into its rule. */
4442 subfacet_max_idle(const struct ofproto_dpif *ofproto)
4445 * Idle time histogram.
4447 * Most of the time a switch has a relatively small number of subfacets.
4448 * When this is the case we might as well keep statistics for all of them
4449 * in userspace and to cache them in the kernel datapath for performance as
4452 * As the number of subfacets increases, the memory required to maintain
4453 * statistics about them in userspace and in the kernel becomes
4454 * significant. However, with a large number of subfacets it is likely
4455 * that only a few of them are "heavy hitters" that consume a large amount
4456 * of bandwidth. At this point, only heavy hitters are worth caching in
4457 * the kernel and maintaining in userspaces; other subfacets we can
4460 * The technique used to compute the idle time is to build a histogram with
4461 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
4462 * that is installed in the kernel gets dropped in the appropriate bucket.
4463 * After the histogram has been built, we compute the cutoff so that only
4464 * the most-recently-used 1% of subfacets (but at least
4465 * ofproto->up.flow_eviction_threshold flows) are kept cached. At least
4466 * the most-recently-used bucket of subfacets is kept, so actually an
4467 * arbitrary number of subfacets can be kept in any given expiration run
4468 * (though the next run will delete most of those unless they receive
4471 * This requires a second pass through the subfacets, in addition to the
4472 * pass made by update_stats(), because the former function never looks at
4473 * uninstallable subfacets.
4475 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
4476 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
4477 int buckets[N_BUCKETS] = { 0 };
4478 int total, subtotal, bucket;
4479 struct subfacet *subfacet;
4483 total = hmap_count(&ofproto->subfacets);
4484 if (total <= ofproto->up.flow_eviction_threshold) {
4485 return N_BUCKETS * BUCKET_WIDTH;
4488 /* Build histogram. */
4490 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
4491 long long int idle = now - subfacet->used;
4492 int bucket = (idle <= 0 ? 0
4493 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
4494 : (unsigned int) idle / BUCKET_WIDTH);
4498 /* Find the first bucket whose flows should be expired. */
4499 subtotal = bucket = 0;
4501 subtotal += buckets[bucket++];
4502 } while (bucket < N_BUCKETS &&
4503 subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100));
4505 if (VLOG_IS_DBG_ENABLED()) {
4509 ds_put_cstr(&s, "keep");
4510 for (i = 0; i < N_BUCKETS; i++) {
4512 ds_put_cstr(&s, ", drop");
4515 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
4518 VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s));
4522 return bucket * BUCKET_WIDTH;
4526 expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle)
4528 /* Cutoff time for most flows. */
4529 long long int normal_cutoff = time_msec() - dp_max_idle;
4531 /* We really want to keep flows for special protocols around, so use a more
4532 * conservative cutoff. */
4533 long long int special_cutoff = time_msec() - 10000;
4535 struct subfacet *subfacet, *next_subfacet;
4536 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
4540 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
4541 &ofproto->subfacets) {
4542 long long int cutoff;
4544 cutoff = (subfacet->facet->slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP
4548 if (subfacet->used < cutoff) {
4549 if (subfacet->path != SF_NOT_INSTALLED) {
4550 batch[n_batch++] = subfacet;
4551 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
4552 subfacet_destroy_batch(ofproto, batch, n_batch);
4556 subfacet_destroy(subfacet);
4562 subfacet_destroy_batch(ofproto, batch, n_batch);
4566 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
4567 * then delete it entirely. */
4569 rule_expire(struct rule_dpif *rule)
4571 struct facet *facet, *next_facet;
4575 if (rule->up.pending) {
4576 /* We'll have to expire it later. */
4580 /* Has 'rule' expired? */
4582 if (rule->up.hard_timeout
4583 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
4584 reason = OFPRR_HARD_TIMEOUT;
4585 } else if (rule->up.idle_timeout
4586 && now > rule->up.used + rule->up.idle_timeout * 1000) {
4587 reason = OFPRR_IDLE_TIMEOUT;
4592 COVERAGE_INC(ofproto_dpif_expired);
4594 /* Update stats. (This is a no-op if the rule expired due to an idle
4595 * timeout, because that only happens when the rule has no facets left.) */
4596 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4597 facet_remove(facet);
4600 /* Get rid of the rule. */
4601 ofproto_rule_expire(&rule->up, reason);
4606 /* Creates and returns a new facet based on 'miss'.
4608 * The caller must already have determined that no facet with an identical
4609 * 'miss->flow' exists in 'miss->ofproto'.
4611 * 'hash' must be the return value of flow_hash(miss->flow, 0).
4613 * The facet will initially have no subfacets. The caller should create (at
4614 * least) one subfacet with subfacet_create(). */
4615 static struct facet *
4616 facet_create(const struct flow_miss *miss, uint32_t hash)
4618 struct ofproto_dpif *ofproto = miss->ofproto;
4619 struct action_xlate_ctx ctx;
4620 struct facet *facet;
4622 facet = xzalloc(sizeof *facet);
4623 facet->used = time_msec();
4624 facet->flow = miss->flow;
4625 facet->initial_vals = miss->initial_vals;
4626 facet->rule = rule_dpif_lookup(ofproto, &facet->flow);
4627 facet->learn_rl = time_msec() + 500;
4629 hmap_insert(&ofproto->facets, &facet->hmap_node, hash);
4630 list_push_back(&facet->rule->facets, &facet->list_node);
4631 list_init(&facet->subfacets);
4632 netflow_flow_init(&facet->nf_flow);
4633 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
4635 ofpbuf_use_stub(&facet->odp_actions, &facet->odp_actions_stub,
4636 sizeof facet->odp_actions_stub);
4637 action_xlate_ctx_init(&ctx, ofproto, &facet->flow, &facet->initial_vals,
4638 facet->rule, 0, NULL);
4639 ctx.may_learn = true;
4640 xlate_actions(&ctx, facet->rule->up.ofpacts, facet->rule->up.ofpacts_len,
4641 &facet->odp_actions);
4642 facet->tags = ctx.tags;
4643 facet->has_learn = ctx.has_learn;
4644 facet->has_normal = ctx.has_normal;
4645 facet->has_fin_timeout = ctx.has_fin_timeout;
4646 facet->nf_flow.output_iface = ctx.nf_output_iface;
4647 facet->mirrors = ctx.mirrors;
4648 facet->slow = ctx.slow;
4654 facet_free(struct facet *facet)
4657 ofpbuf_uninit(&facet->odp_actions);
4662 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
4663 * 'packet', which arrived on 'in_port'. */
4665 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
4666 const struct nlattr *odp_actions, size_t actions_len,
4667 struct ofpbuf *packet)
4669 struct odputil_keybuf keybuf;
4673 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4674 odp_flow_key_from_flow(&key, flow,
4675 ofp_port_to_odp_port(ofproto, flow->in_port));
4677 error = dpif_execute(ofproto->backer->dpif, key.data, key.size,
4678 odp_actions, actions_len, packet);
4682 /* Remove 'facet' from 'ofproto' and free up the associated memory:
4684 * - If 'facet' was installed in the datapath, uninstalls it and updates its
4685 * rule's statistics, via subfacet_uninstall().
4687 * - Removes 'facet' from its rule and from ofproto->facets.
4690 facet_remove(struct facet *facet)
4692 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4693 struct subfacet *subfacet, *next_subfacet;
4695 ovs_assert(!list_is_empty(&facet->subfacets));
4697 /* First uninstall all of the subfacets to get final statistics. */
4698 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4699 subfacet_uninstall(subfacet);
4702 /* Flush the final stats to the rule.
4704 * This might require us to have at least one subfacet around so that we
4705 * can use its actions for accounting in facet_account(), which is why we
4706 * have uninstalled but not yet destroyed the subfacets. */
4707 facet_flush_stats(facet);
4709 /* Now we're really all done so destroy everything. */
4710 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
4711 &facet->subfacets) {
4712 subfacet_destroy__(subfacet);
4714 hmap_remove(&ofproto->facets, &facet->hmap_node);
4715 list_remove(&facet->list_node);
4719 /* Feed information from 'facet' back into the learning table to keep it in
4720 * sync with what is actually flowing through the datapath. */
4722 facet_learn(struct facet *facet)
4724 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4725 long long int now = time_msec();
4726 struct action_xlate_ctx ctx;
4728 if (!facet->has_fin_timeout && now < facet->learn_rl) {
4732 facet->learn_rl = now + 500;
4734 if (!facet->has_learn
4735 && !facet->has_normal
4736 && (!facet->has_fin_timeout
4737 || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
4741 action_xlate_ctx_init(&ctx, ofproto, &facet->flow, &facet->initial_vals,
4742 facet->rule, facet->tcp_flags, NULL);
4743 ctx.may_learn = true;
4744 xlate_actions_for_side_effects(&ctx, facet->rule->up.ofpacts,
4745 facet->rule->up.ofpacts_len);
4749 facet_account(struct facet *facet)
4751 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4752 const struct nlattr *a;
4757 if (!facet->has_normal || !ofproto->has_bonded_bundles) {
4760 n_bytes = facet->byte_count - facet->accounted_bytes;
4762 /* This loop feeds byte counters to bond_account() for rebalancing to use
4763 * as a basis. We also need to track the actual VLAN on which the packet
4764 * is going to be sent to ensure that it matches the one passed to
4765 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
4768 * We use the actions from an arbitrary subfacet because they should all
4769 * be equally valid for our purpose. */
4770 vlan_tci = facet->flow.vlan_tci;
4771 NL_ATTR_FOR_EACH_UNSAFE (a, left, facet->odp_actions.data,
4772 facet->odp_actions.size) {
4773 const struct ovs_action_push_vlan *vlan;
4774 struct ofport_dpif *port;
4776 switch (nl_attr_type(a)) {
4777 case OVS_ACTION_ATTR_OUTPUT:
4778 port = get_odp_port(ofproto, nl_attr_get_u32(a));
4779 if (port && port->bundle && port->bundle->bond) {
4780 bond_account(port->bundle->bond, &facet->flow,
4781 vlan_tci_to_vid(vlan_tci), n_bytes);
4785 case OVS_ACTION_ATTR_POP_VLAN:
4786 vlan_tci = htons(0);
4789 case OVS_ACTION_ATTR_PUSH_VLAN:
4790 vlan = nl_attr_get(a);
4791 vlan_tci = vlan->vlan_tci;
4797 /* Returns true if the only action for 'facet' is to send to the controller.
4798 * (We don't report NetFlow expiration messages for such facets because they
4799 * are just part of the control logic for the network, not real traffic). */
4801 facet_is_controller_flow(struct facet *facet)
4804 const struct rule *rule = &facet->rule->up;
4805 const struct ofpact *ofpacts = rule->ofpacts;
4806 size_t ofpacts_len = rule->ofpacts_len;
4808 if (ofpacts_len > 0 &&
4809 ofpacts->type == OFPACT_CONTROLLER &&
4810 ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len)) {
4817 /* Folds all of 'facet''s statistics into its rule. Also updates the
4818 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
4819 * 'facet''s statistics in the datapath should have been zeroed and folded into
4820 * its packet and byte counts before this function is called. */
4822 facet_flush_stats(struct facet *facet)
4824 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4825 struct subfacet *subfacet;
4827 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4828 ovs_assert(!subfacet->dp_byte_count);
4829 ovs_assert(!subfacet->dp_packet_count);
4832 facet_push_stats(facet);
4833 if (facet->accounted_bytes < facet->byte_count) {
4834 facet_account(facet);
4835 facet->accounted_bytes = facet->byte_count;
4838 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
4839 struct ofexpired expired;
4840 expired.flow = facet->flow;
4841 expired.packet_count = facet->packet_count;
4842 expired.byte_count = facet->byte_count;
4843 expired.used = facet->used;
4844 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4847 /* Reset counters to prevent double counting if 'facet' ever gets
4849 facet_reset_counters(facet);
4851 netflow_flow_clear(&facet->nf_flow);
4852 facet->tcp_flags = 0;
4855 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
4856 * Returns it if found, otherwise a null pointer.
4858 * 'hash' must be the return value of flow_hash(flow, 0).
4860 * The returned facet might need revalidation; use facet_lookup_valid()
4861 * instead if that is important. */
4862 static struct facet *
4863 facet_find(struct ofproto_dpif *ofproto,
4864 const struct flow *flow, uint32_t hash)
4866 struct facet *facet;
4868 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, hash, &ofproto->facets) {
4869 if (flow_equal(flow, &facet->flow)) {
4877 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
4878 * Returns it if found, otherwise a null pointer.
4880 * 'hash' must be the return value of flow_hash(flow, 0).
4882 * The returned facet is guaranteed to be valid. */
4883 static struct facet *
4884 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow,
4887 struct facet *facet;
4889 facet = facet_find(ofproto, flow, hash);
4891 && (ofproto->backer->need_revalidate
4892 || tag_set_intersects(&ofproto->backer->revalidate_set,
4894 && !facet_revalidate(facet)) {
4895 facet = facet_find(ofproto, flow, hash);
4902 facet_check_consistency(struct facet *facet)
4904 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
4906 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4908 uint64_t odp_actions_stub[1024 / 8];
4909 struct ofpbuf odp_actions;
4911 struct action_xlate_ctx ctx;
4912 struct rule_dpif *rule;
4915 /* Check the rule for consistency. */
4916 rule = rule_dpif_lookup(ofproto, &facet->flow);
4917 if (rule != facet->rule) {
4918 if (!VLOG_DROP_WARN(&rl)) {
4919 struct ds s = DS_EMPTY_INITIALIZER;
4921 flow_format(&s, &facet->flow);
4922 ds_put_format(&s, ": facet associated with wrong rule (was "
4923 "table=%"PRIu8",", facet->rule->up.table_id);
4924 cls_rule_format(&facet->rule->up.cr, &s);
4925 ds_put_format(&s, ") (should have been table=%"PRIu8",",
4927 cls_rule_format(&rule->up.cr, &s);
4928 ds_put_cstr(&s, ")\n");
4935 /* Check the datapath actions for consistency. */
4936 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
4937 action_xlate_ctx_init(&ctx, ofproto, &facet->flow, &facet->initial_vals,
4939 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, &odp_actions);
4941 ok = ofpbuf_equal(&facet->odp_actions, &odp_actions)
4942 && facet->slow == ctx.slow;
4943 if (!ok && !VLOG_DROP_WARN(&rl)) {
4944 struct ds s = DS_EMPTY_INITIALIZER;
4946 flow_format(&s, &facet->flow);
4947 ds_put_cstr(&s, ": inconsistency in facet");
4949 if (!ofpbuf_equal(&facet->odp_actions, &odp_actions)) {
4950 ds_put_cstr(&s, " (actions were: ");
4951 format_odp_actions(&s, facet->odp_actions.data,
4952 facet->odp_actions.size);
4953 ds_put_cstr(&s, ") (correct actions: ");
4954 format_odp_actions(&s, odp_actions.data, odp_actions.size);
4955 ds_put_cstr(&s, ")");
4958 if (facet->slow != ctx.slow) {
4959 ds_put_format(&s, " slow path incorrect. should be %d", ctx.slow);
4964 ofpbuf_uninit(&odp_actions);
4969 /* Re-searches the classifier for 'facet':
4971 * - If the rule found is different from 'facet''s current rule, moves
4972 * 'facet' to the new rule and recompiles its actions.
4974 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
4975 * where it is and recompiles its actions anyway.
4977 * - If any of 'facet''s subfacets correspond to a new flow according to
4978 * ofproto_receive(), 'facet' is removed.
4980 * Returns true if 'facet' is still valid. False if 'facet' was removed. */
4982 facet_revalidate(struct facet *facet)
4984 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4986 uint64_t odp_actions_stub[1024 / 8];
4987 struct ofpbuf odp_actions;
4989 struct action_xlate_ctx ctx;
4990 struct rule_dpif *new_rule;
4991 struct subfacet *subfacet;
4993 COVERAGE_INC(facet_revalidate);
4995 /* Check that child subfacets still correspond to this facet. Tunnel
4996 * configuration changes could cause a subfacet's OpenFlow in_port to
4998 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4999 struct ofproto_dpif *recv_ofproto;
5000 struct flow recv_flow;
5003 error = ofproto_receive(ofproto->backer, NULL, subfacet->key,
5004 subfacet->key_len, &recv_flow, NULL,
5005 &recv_ofproto, NULL, NULL);
5007 || recv_ofproto != ofproto
5008 || memcmp(&recv_flow, &facet->flow, sizeof recv_flow)) {
5009 facet_remove(facet);
5014 new_rule = rule_dpif_lookup(ofproto, &facet->flow);
5016 /* Calculate new datapath actions.
5018 * We do not modify any 'facet' state yet, because we might need to, e.g.,
5019 * emit a NetFlow expiration and, if so, we need to have the old state
5020 * around to properly compose it. */
5021 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5022 action_xlate_ctx_init(&ctx, ofproto, &facet->flow, &facet->initial_vals,
5024 xlate_actions(&ctx, new_rule->up.ofpacts, new_rule->up.ofpacts_len,
5027 /* A facet's slow path reason should only change under dramatic
5028 * circumstances. Rather than try to update everything, it's simpler to
5029 * remove the facet and start over. */
5030 if (facet->slow != ctx.slow) {
5031 facet_remove(facet);
5032 ofpbuf_uninit(&odp_actions);
5036 if (!ofpbuf_equal(&facet->odp_actions, &odp_actions)) {
5037 LIST_FOR_EACH(subfacet, list_node, &facet->subfacets) {
5038 if (subfacet->path == SF_FAST_PATH) {
5039 struct dpif_flow_stats stats;
5041 subfacet_install(subfacet, &odp_actions, &stats);
5042 subfacet_update_stats(subfacet, &stats);
5046 facet_flush_stats(facet);
5048 ofpbuf_clear(&facet->odp_actions);
5049 ofpbuf_put(&facet->odp_actions, odp_actions.data, odp_actions.size);
5052 /* Update 'facet' now that we've taken care of all the old state. */
5053 facet->tags = ctx.tags;
5054 facet->slow = ctx.slow;
5055 facet->nf_flow.output_iface = ctx.nf_output_iface;
5056 facet->has_learn = ctx.has_learn;
5057 facet->has_normal = ctx.has_normal;
5058 facet->has_fin_timeout = ctx.has_fin_timeout;
5059 facet->mirrors = ctx.mirrors;
5061 if (facet->rule != new_rule) {
5062 COVERAGE_INC(facet_changed_rule);
5063 list_remove(&facet->list_node);
5064 list_push_back(&new_rule->facets, &facet->list_node);
5065 facet->rule = new_rule;
5066 facet->used = new_rule->up.created;
5067 facet->prev_used = facet->used;
5070 ofpbuf_uninit(&odp_actions);
5074 /* Updates 'facet''s used time. Caller is responsible for calling
5075 * facet_push_stats() to update the flows which 'facet' resubmits into. */
5077 facet_update_time(struct facet *facet, long long int used)
5079 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5080 if (used > facet->used) {
5082 ofproto_rule_update_used(&facet->rule->up, used);
5083 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
5088 facet_reset_counters(struct facet *facet)
5090 facet->packet_count = 0;
5091 facet->byte_count = 0;
5092 facet->prev_packet_count = 0;
5093 facet->prev_byte_count = 0;
5094 facet->accounted_bytes = 0;
5098 facet_push_stats(struct facet *facet)
5100 struct dpif_flow_stats stats;
5102 ovs_assert(facet->packet_count >= facet->prev_packet_count);
5103 ovs_assert(facet->byte_count >= facet->prev_byte_count);
5104 ovs_assert(facet->used >= facet->prev_used);
5106 stats.n_packets = facet->packet_count - facet->prev_packet_count;
5107 stats.n_bytes = facet->byte_count - facet->prev_byte_count;
5108 stats.used = facet->used;
5109 stats.tcp_flags = 0;
5111 if (stats.n_packets || stats.n_bytes || facet->used > facet->prev_used) {
5112 facet->prev_packet_count = facet->packet_count;
5113 facet->prev_byte_count = facet->byte_count;
5114 facet->prev_used = facet->used;
5116 rule_credit_stats(facet->rule, &stats);
5117 flow_push_stats(facet, &stats);
5119 update_mirror_stats(ofproto_dpif_cast(facet->rule->up.ofproto),
5120 facet->mirrors, stats.n_packets, stats.n_bytes);
5125 push_all_stats__(bool run_fast)
5127 static long long int rl = LLONG_MIN;
5128 struct ofproto_dpif *ofproto;
5130 if (time_msec() < rl) {
5134 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5135 struct facet *facet;
5137 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
5138 facet_push_stats(facet);
5145 rl = time_msec() + 100;
5149 push_all_stats(void)
5151 push_all_stats__(true);
5155 rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats)
5157 rule->packet_count += stats->n_packets;
5158 rule->byte_count += stats->n_bytes;
5159 ofproto_rule_update_used(&rule->up, stats->used);
5162 /* Pushes flow statistics to the rules which 'facet->flow' resubmits
5163 * into given 'facet->rule''s actions and mirrors. */
5165 flow_push_stats(struct facet *facet, const struct dpif_flow_stats *stats)
5167 struct rule_dpif *rule = facet->rule;
5168 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5169 struct action_xlate_ctx ctx;
5171 ofproto_rule_update_used(&rule->up, stats->used);
5173 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
5174 &facet->initial_vals, rule, 0, NULL);
5175 ctx.resubmit_stats = stats;
5176 xlate_actions_for_side_effects(&ctx, rule->up.ofpacts,
5177 rule->up.ofpacts_len);
5182 static struct subfacet *
5183 subfacet_find(struct ofproto_dpif *ofproto,
5184 const struct nlattr *key, size_t key_len, uint32_t key_hash)
5186 struct subfacet *subfacet;
5188 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
5189 &ofproto->subfacets) {
5190 if (subfacet->key_len == key_len
5191 && !memcmp(key, subfacet->key, key_len)) {
5199 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
5200 * 'key_fitness', 'key', and 'key_len' members in 'miss'. Returns the
5201 * existing subfacet if there is one, otherwise creates and returns a
5203 static struct subfacet *
5204 subfacet_create(struct facet *facet, struct flow_miss *miss,
5207 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5208 enum odp_key_fitness key_fitness = miss->key_fitness;
5209 const struct nlattr *key = miss->key;
5210 size_t key_len = miss->key_len;
5212 struct subfacet *subfacet;
5214 key_hash = odp_flow_key_hash(key, key_len);
5216 if (list_is_empty(&facet->subfacets)) {
5217 subfacet = &facet->one_subfacet;
5219 subfacet = subfacet_find(ofproto, key, key_len, key_hash);
5221 if (subfacet->facet == facet) {
5225 /* This shouldn't happen. */
5226 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
5227 subfacet_destroy(subfacet);
5230 subfacet = xmalloc(sizeof *subfacet);
5233 hmap_insert(&ofproto->subfacets, &subfacet->hmap_node, key_hash);
5234 list_push_back(&facet->subfacets, &subfacet->list_node);
5235 subfacet->facet = facet;
5236 subfacet->key_fitness = key_fitness;
5237 subfacet->key = xmemdup(key, key_len);
5238 subfacet->key_len = key_len;
5239 subfacet->used = now;
5240 subfacet->created = now;
5241 subfacet->dp_packet_count = 0;
5242 subfacet->dp_byte_count = 0;
5243 subfacet->path = SF_NOT_INSTALLED;
5244 subfacet->odp_in_port = miss->odp_in_port;
5246 ofproto->subfacet_add_count++;
5250 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
5251 * its facet within 'ofproto', and frees it. */
5253 subfacet_destroy__(struct subfacet *subfacet)
5255 struct facet *facet = subfacet->facet;
5256 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5258 /* Update ofproto stats before uninstall the subfacet. */
5259 ofproto->subfacet_del_count++;
5260 ofproto->total_subfacet_life_span += (time_msec() - subfacet->created);
5262 subfacet_uninstall(subfacet);
5263 hmap_remove(&ofproto->subfacets, &subfacet->hmap_node);
5264 list_remove(&subfacet->list_node);
5265 free(subfacet->key);
5266 if (subfacet != &facet->one_subfacet) {
5271 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
5272 * last remaining subfacet in its facet destroys the facet too. */
5274 subfacet_destroy(struct subfacet *subfacet)
5276 struct facet *facet = subfacet->facet;
5278 if (list_is_singleton(&facet->subfacets)) {
5279 /* facet_remove() needs at least one subfacet (it will remove it). */
5280 facet_remove(facet);
5282 subfacet_destroy__(subfacet);
5287 subfacet_destroy_batch(struct ofproto_dpif *ofproto,
5288 struct subfacet **subfacets, int n)
5290 struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH];
5291 struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH];
5292 struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH];
5295 for (i = 0; i < n; i++) {
5296 ops[i].type = DPIF_OP_FLOW_DEL;
5297 ops[i].u.flow_del.key = subfacets[i]->key;
5298 ops[i].u.flow_del.key_len = subfacets[i]->key_len;
5299 ops[i].u.flow_del.stats = &stats[i];
5303 dpif_operate(ofproto->backer->dpif, opsp, n);
5304 for (i = 0; i < n; i++) {
5305 subfacet_reset_dp_stats(subfacets[i], &stats[i]);
5306 subfacets[i]->path = SF_NOT_INSTALLED;
5307 subfacet_destroy(subfacets[i]);
5312 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
5313 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
5314 * in the datapath will be zeroed and 'stats' will be updated with traffic new
5315 * since 'subfacet' was last updated.
5317 * Returns 0 if successful, otherwise a positive errno value. */
5319 subfacet_install(struct subfacet *subfacet, const struct ofpbuf *odp_actions,
5320 struct dpif_flow_stats *stats)
5322 struct facet *facet = subfacet->facet;
5323 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5324 enum subfacet_path path = facet->slow ? SF_SLOW_PATH : SF_FAST_PATH;
5325 const struct nlattr *actions = odp_actions->data;
5326 size_t actions_len = odp_actions->size;
5328 uint64_t slow_path_stub[128 / 8];
5329 enum dpif_flow_put_flags flags;
5332 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
5334 flags |= DPIF_FP_ZERO_STATS;
5337 if (path == SF_SLOW_PATH) {
5338 compose_slow_path(ofproto, &facet->flow, facet->slow,
5339 slow_path_stub, sizeof slow_path_stub,
5340 &actions, &actions_len);
5343 ret = dpif_flow_put(ofproto->backer->dpif, flags, subfacet->key,
5344 subfacet->key_len, actions, actions_len, stats);
5347 subfacet_reset_dp_stats(subfacet, stats);
5351 subfacet->path = path;
5356 /* If 'subfacet' is installed in the datapath, uninstalls it. */
5358 subfacet_uninstall(struct subfacet *subfacet)
5360 if (subfacet->path != SF_NOT_INSTALLED) {
5361 struct rule_dpif *rule = subfacet->facet->rule;
5362 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5363 struct dpif_flow_stats stats;
5366 error = dpif_flow_del(ofproto->backer->dpif, subfacet->key,
5367 subfacet->key_len, &stats);
5368 subfacet_reset_dp_stats(subfacet, &stats);
5370 subfacet_update_stats(subfacet, &stats);
5372 subfacet->path = SF_NOT_INSTALLED;
5374 ovs_assert(subfacet->dp_packet_count == 0);
5375 ovs_assert(subfacet->dp_byte_count == 0);
5379 /* Resets 'subfacet''s datapath statistics counters. This should be called
5380 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
5381 * non-null, it should contain the statistics returned by dpif when 'subfacet'
5382 * was reset in the datapath. 'stats' will be modified to include only
5383 * statistics new since 'subfacet' was last updated. */
5385 subfacet_reset_dp_stats(struct subfacet *subfacet,
5386 struct dpif_flow_stats *stats)
5389 && subfacet->dp_packet_count <= stats->n_packets
5390 && subfacet->dp_byte_count <= stats->n_bytes) {
5391 stats->n_packets -= subfacet->dp_packet_count;
5392 stats->n_bytes -= subfacet->dp_byte_count;
5395 subfacet->dp_packet_count = 0;
5396 subfacet->dp_byte_count = 0;
5399 /* Updates 'subfacet''s used time. The caller is responsible for calling
5400 * facet_push_stats() to update the flows which 'subfacet' resubmits into. */
5402 subfacet_update_time(struct subfacet *subfacet, long long int used)
5404 if (used > subfacet->used) {
5405 subfacet->used = used;
5406 facet_update_time(subfacet->facet, used);
5410 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
5412 * Because of the meaning of a subfacet's counters, it only makes sense to do
5413 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
5414 * represents a packet that was sent by hand or if it represents statistics
5415 * that have been cleared out of the datapath. */
5417 subfacet_update_stats(struct subfacet *subfacet,
5418 const struct dpif_flow_stats *stats)
5420 if (stats->n_packets || stats->used > subfacet->used) {
5421 struct facet *facet = subfacet->facet;
5423 subfacet_update_time(subfacet, stats->used);
5424 facet->packet_count += stats->n_packets;
5425 facet->byte_count += stats->n_bytes;
5426 facet->tcp_flags |= stats->tcp_flags;
5427 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
5433 static struct rule_dpif *
5434 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow)
5436 struct rule_dpif *rule;
5438 rule = rule_dpif_lookup__(ofproto, flow, 0);
5443 return rule_dpif_miss_rule(ofproto, flow);
5446 static struct rule_dpif *
5447 rule_dpif_lookup__(struct ofproto_dpif *ofproto, const struct flow *flow,
5450 struct cls_rule *cls_rule;
5451 struct classifier *cls;
5453 if (table_id >= N_TABLES) {
5457 cls = &ofproto->up.tables[table_id].cls;
5458 if (flow->nw_frag & FLOW_NW_FRAG_ANY
5459 && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
5460 /* For OFPC_NORMAL frag_handling, we must pretend that transport ports
5461 * are unavailable. */
5462 struct flow ofpc_normal_flow = *flow;
5463 ofpc_normal_flow.tp_src = htons(0);
5464 ofpc_normal_flow.tp_dst = htons(0);
5465 cls_rule = classifier_lookup(cls, &ofpc_normal_flow);
5467 cls_rule = classifier_lookup(cls, flow);
5469 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
5472 static struct rule_dpif *
5473 rule_dpif_miss_rule(struct ofproto_dpif *ofproto, const struct flow *flow)
5475 struct ofport_dpif *port;
5477 port = get_ofp_port(ofproto, flow->in_port);
5479 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, flow->in_port);
5480 return ofproto->miss_rule;
5483 if (port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN) {
5484 return ofproto->no_packet_in_rule;
5486 return ofproto->miss_rule;
5490 complete_operation(struct rule_dpif *rule)
5492 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5494 rule_invalidate(rule);
5496 struct dpif_completion *c = xmalloc(sizeof *c);
5497 c->op = rule->up.pending;
5498 list_push_back(&ofproto->completions, &c->list_node);
5500 ofoperation_complete(rule->up.pending, 0);
5504 static struct rule *
5507 struct rule_dpif *rule = xmalloc(sizeof *rule);
5512 rule_dealloc(struct rule *rule_)
5514 struct rule_dpif *rule = rule_dpif_cast(rule_);
5519 rule_construct(struct rule *rule_)
5521 struct rule_dpif *rule = rule_dpif_cast(rule_);
5522 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5523 struct rule_dpif *victim;
5526 rule->packet_count = 0;
5527 rule->byte_count = 0;
5529 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
5530 if (victim && !list_is_empty(&victim->facets)) {
5531 struct facet *facet;
5533 rule->facets = victim->facets;
5534 list_moved(&rule->facets);
5535 LIST_FOR_EACH (facet, list_node, &rule->facets) {
5536 /* XXX: We're only clearing our local counters here. It's possible
5537 * that quite a few packets are unaccounted for in the datapath
5538 * statistics. These will be accounted to the new rule instead of
5539 * cleared as required. This could be fixed by clearing out the
5540 * datapath statistics for this facet, but currently it doesn't
5542 facet_reset_counters(facet);
5546 /* Must avoid list_moved() in this case. */
5547 list_init(&rule->facets);
5550 table_id = rule->up.table_id;
5552 rule->tag = victim->tag;
5553 } else if (table_id == 0) {
5558 miniflow_expand(&rule->up.cr.match.flow, &flow);
5559 rule->tag = rule_calculate_tag(&flow, &rule->up.cr.match.mask,
5560 ofproto->tables[table_id].basis);
5563 complete_operation(rule);
5568 rule_destruct(struct rule *rule_)
5570 struct rule_dpif *rule = rule_dpif_cast(rule_);
5571 struct facet *facet, *next_facet;
5573 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
5574 facet_revalidate(facet);
5577 complete_operation(rule);
5581 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
5583 struct rule_dpif *rule = rule_dpif_cast(rule_);
5585 /* push_all_stats() can handle flow misses which, when using the learn
5586 * action, can cause rules to be added and deleted. This can corrupt our
5587 * caller's datastructures which assume that rule_get_stats() doesn't have
5588 * an impact on the flow table. To be safe, we disable miss handling. */
5589 push_all_stats__(false);
5591 /* Start from historical data for 'rule' itself that are no longer tracked
5592 * in facets. This counts, for example, facets that have expired. */
5593 *packets = rule->packet_count;
5594 *bytes = rule->byte_count;
5598 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
5599 struct ofpbuf *packet)
5601 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5602 struct initial_vals initial_vals;
5603 struct dpif_flow_stats stats;
5604 struct action_xlate_ctx ctx;
5605 uint64_t odp_actions_stub[1024 / 8];
5606 struct ofpbuf odp_actions;
5608 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
5609 rule_credit_stats(rule, &stats);
5611 initial_vals.vlan_tci = flow->vlan_tci;
5612 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5613 action_xlate_ctx_init(&ctx, ofproto, flow, &initial_vals,
5614 rule, stats.tcp_flags, packet);
5615 ctx.resubmit_stats = &stats;
5616 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, &odp_actions);
5618 execute_odp_actions(ofproto, flow, odp_actions.data,
5619 odp_actions.size, packet);
5621 ofpbuf_uninit(&odp_actions);
5625 rule_execute(struct rule *rule, const struct flow *flow,
5626 struct ofpbuf *packet)
5628 rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
5629 ofpbuf_delete(packet);
5634 rule_modify_actions(struct rule *rule_)
5636 struct rule_dpif *rule = rule_dpif_cast(rule_);
5638 complete_operation(rule);
5641 /* Sends 'packet' out 'ofport'.
5642 * May modify 'packet'.
5643 * Returns 0 if successful, otherwise a positive errno value. */
5645 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
5647 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
5648 uint64_t odp_actions_stub[1024 / 8];
5649 struct ofpbuf key, odp_actions;
5650 struct dpif_flow_stats stats;
5651 struct odputil_keybuf keybuf;
5652 struct ofpact_output output;
5653 struct action_xlate_ctx ctx;
5657 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5658 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5660 /* Use OFPP_NONE as the in_port to avoid special packet processing. */
5661 flow_extract(packet, 0, 0, NULL, OFPP_NONE, &flow);
5662 odp_flow_key_from_flow(&key, &flow, ofp_port_to_odp_port(ofproto,
5664 dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
5666 ofpact_init(&output.ofpact, OFPACT_OUTPUT, sizeof output);
5667 output.port = ofport->up.ofp_port;
5670 action_xlate_ctx_init(&ctx, ofproto, &flow, NULL, NULL, 0, packet);
5671 ctx.resubmit_stats = &stats;
5672 xlate_actions(&ctx, &output.ofpact, sizeof output, &odp_actions);
5674 error = dpif_execute(ofproto->backer->dpif,
5676 odp_actions.data, odp_actions.size,
5678 ofpbuf_uninit(&odp_actions);
5681 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %s (%s)",
5682 ofproto->up.name, netdev_get_name(ofport->up.netdev),
5686 ofproto->stats.tx_packets++;
5687 ofproto->stats.tx_bytes += packet->size;
5691 /* OpenFlow to datapath action translation. */
5693 static bool may_receive(const struct ofport_dpif *, struct action_xlate_ctx *);
5694 static void do_xlate_actions(const struct ofpact *, size_t ofpacts_len,
5695 struct action_xlate_ctx *);
5696 static void xlate_normal(struct action_xlate_ctx *);
5698 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
5699 * The action will state 'slow' as the reason that the action is in the slow
5700 * path. (This is purely informational: it allows a human viewing "ovs-dpctl
5701 * dump-flows" output to see why a flow is in the slow path.)
5703 * The 'stub_size' bytes in 'stub' will be used to store the action.
5704 * 'stub_size' must be large enough for the action.
5706 * The action and its size will be stored in '*actionsp' and '*actions_lenp',
5709 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
5710 enum slow_path_reason slow,
5711 uint64_t *stub, size_t stub_size,
5712 const struct nlattr **actionsp, size_t *actions_lenp)
5714 union user_action_cookie cookie;
5717 cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
5718 cookie.slow_path.unused = 0;
5719 cookie.slow_path.reason = slow;
5721 ofpbuf_use_stack(&buf, stub, stub_size);
5722 if (slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP | SLOW_STP)) {
5723 uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif, UINT32_MAX);
5724 odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
5726 put_userspace_action(ofproto, &buf, flow, &cookie,
5727 sizeof cookie.slow_path);
5729 *actionsp = buf.data;
5730 *actions_lenp = buf.size;
5734 put_userspace_action(const struct ofproto_dpif *ofproto,
5735 struct ofpbuf *odp_actions,
5736 const struct flow *flow,
5737 const union user_action_cookie *cookie,
5738 const size_t cookie_size)
5742 pid = dpif_port_get_pid(ofproto->backer->dpif,
5743 ofp_port_to_odp_port(ofproto, flow->in_port));
5745 return odp_put_userspace_action(pid, cookie, cookie_size, odp_actions);
5748 /* Compose SAMPLE action for sFlow or IPFIX. The given probability is
5749 * the number of packets out of UINT32_MAX to sample. The given
5750 * cookie is passed back in the callback for each sampled packet.
5753 compose_sample_action(const struct ofproto_dpif *ofproto,
5754 struct ofpbuf *odp_actions,
5755 const struct flow *flow,
5756 const uint32_t probability,
5757 const union user_action_cookie *cookie,
5758 const size_t cookie_size)
5760 size_t sample_offset, actions_offset;
5763 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
5765 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
5767 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
5768 cookie_offset = put_userspace_action(ofproto, odp_actions, flow, cookie,
5771 nl_msg_end_nested(odp_actions, actions_offset);
5772 nl_msg_end_nested(odp_actions, sample_offset);
5773 return cookie_offset;
5777 compose_sflow_cookie(const struct ofproto_dpif *ofproto,
5778 ovs_be16 vlan_tci, uint32_t odp_port,
5779 unsigned int n_outputs, union user_action_cookie *cookie)
5783 cookie->type = USER_ACTION_COOKIE_SFLOW;
5784 cookie->sflow.vlan_tci = vlan_tci;
5786 /* See http://www.sflow.org/sflow_version_5.txt (search for "Input/output
5787 * port information") for the interpretation of cookie->output. */
5788 switch (n_outputs) {
5790 /* 0x40000000 | 256 means "packet dropped for unknown reason". */
5791 cookie->sflow.output = 0x40000000 | 256;
5795 ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port);
5797 cookie->sflow.output = ifindex;
5802 /* 0x80000000 means "multiple output ports. */
5803 cookie->sflow.output = 0x80000000 | n_outputs;
5808 /* Compose SAMPLE action for sFlow bridge sampling. */
5810 compose_sflow_action(const struct ofproto_dpif *ofproto,
5811 struct ofpbuf *odp_actions,
5812 const struct flow *flow,
5815 uint32_t probability;
5816 union user_action_cookie cookie;
5818 if (!ofproto->sflow || flow->in_port == OFPP_NONE) {
5822 probability = dpif_sflow_get_probability(ofproto->sflow);
5823 compose_sflow_cookie(ofproto, htons(0), odp_port,
5824 odp_port == OVSP_NONE ? 0 : 1, &cookie);
5826 return compose_sample_action(ofproto, odp_actions, flow, probability,
5827 &cookie, sizeof cookie.sflow);
5831 compose_flow_sample_cookie(uint16_t probability, uint32_t collector_set_id,
5832 uint32_t obs_domain_id, uint32_t obs_point_id,
5833 union user_action_cookie *cookie)
5835 cookie->type = USER_ACTION_COOKIE_FLOW_SAMPLE;
5836 cookie->flow_sample.probability = probability;
5837 cookie->flow_sample.collector_set_id = collector_set_id;
5838 cookie->flow_sample.obs_domain_id = obs_domain_id;
5839 cookie->flow_sample.obs_point_id = obs_point_id;
5843 compose_ipfix_cookie(union user_action_cookie *cookie)
5845 cookie->type = USER_ACTION_COOKIE_IPFIX;
5848 /* Compose SAMPLE action for IPFIX bridge sampling. */
5850 compose_ipfix_action(const struct ofproto_dpif *ofproto,
5851 struct ofpbuf *odp_actions,
5852 const struct flow *flow)
5854 uint32_t probability;
5855 union user_action_cookie cookie;
5857 if (!ofproto->ipfix || flow->in_port == OFPP_NONE) {
5861 probability = dpif_ipfix_get_bridge_exporter_probability(ofproto->ipfix);
5862 compose_ipfix_cookie(&cookie);
5864 compose_sample_action(ofproto, odp_actions, flow, probability,
5865 &cookie, sizeof cookie.ipfix);
5868 /* SAMPLE action for sFlow must be first action in any given list of
5869 * actions. At this point we do not have all information required to
5870 * build it. So try to build sample action as complete as possible. */
5872 add_sflow_action(struct action_xlate_ctx *ctx)
5874 ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto,
5876 &ctx->flow, OVSP_NONE);
5877 ctx->sflow_odp_port = 0;
5878 ctx->sflow_n_outputs = 0;
5881 /* SAMPLE action for IPFIX must be 1st or 2nd action in any given list
5882 * of actions, eventually after the SAMPLE action for sFlow. */
5884 add_ipfix_action(struct action_xlate_ctx *ctx)
5886 compose_ipfix_action(ctx->ofproto, ctx->odp_actions, &ctx->flow);
5889 /* Fix SAMPLE action according to data collected while composing ODP actions.
5890 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
5891 * USERSPACE action's user-cookie which is required for sflow. */
5893 fix_sflow_action(struct action_xlate_ctx *ctx)
5895 const struct flow *base = &ctx->base_flow;
5896 union user_action_cookie *cookie;
5898 if (!ctx->user_cookie_offset) {
5902 cookie = ofpbuf_at(ctx->odp_actions, ctx->user_cookie_offset,
5903 sizeof cookie->sflow);
5904 ovs_assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
5906 compose_sflow_cookie(ctx->ofproto, base->vlan_tci,
5907 ctx->sflow_odp_port, ctx->sflow_n_outputs, cookie);
5911 compose_output_action__(struct action_xlate_ctx *ctx, uint16_t ofp_port,
5914 const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
5915 ovs_be16 flow_vlan_tci;
5916 uint32_t flow_skb_mark;
5917 uint8_t flow_nw_tos;
5918 struct priority_to_dscp *pdscp;
5919 uint32_t out_port, odp_port;
5921 /* If 'struct flow' gets additional metadata, we'll need to zero it out
5922 * before traversing a patch port. */
5923 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 20);
5926 xlate_report(ctx, "Nonexistent output port");
5928 } else if (ofport->up.pp.config & OFPUTIL_PC_NO_FWD) {
5929 xlate_report(ctx, "OFPPC_NO_FWD set, skipping output");
5931 } else if (check_stp && !stp_forward_in_state(ofport->stp_state)) {
5932 xlate_report(ctx, "STP not in forwarding state, skipping output");
5936 if (netdev_vport_is_patch(ofport->up.netdev)) {
5937 struct ofport_dpif *peer = ofport_get_peer(ofport);
5938 struct flow old_flow = ctx->flow;
5939 const struct ofproto_dpif *peer_ofproto;
5940 enum slow_path_reason special;
5941 struct ofport_dpif *in_port;
5944 xlate_report(ctx, "Nonexistent patch port peer");
5948 peer_ofproto = ofproto_dpif_cast(peer->up.ofproto);
5949 if (peer_ofproto->backer != ctx->ofproto->backer) {
5950 xlate_report(ctx, "Patch port peer on a different datapath");
5954 ctx->ofproto = ofproto_dpif_cast(peer->up.ofproto);
5955 ctx->flow.in_port = peer->up.ofp_port;
5956 ctx->flow.metadata = htonll(0);
5957 memset(&ctx->flow.tunnel, 0, sizeof ctx->flow.tunnel);
5958 memset(ctx->flow.regs, 0, sizeof ctx->flow.regs);
5960 in_port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
5961 special = process_special(ctx->ofproto, &ctx->flow, in_port,
5964 ctx->slow = special;
5965 } else if (!in_port || may_receive(in_port, ctx)) {
5966 if (!in_port || stp_forward_in_state(in_port->stp_state)) {
5967 xlate_table_action(ctx, ctx->flow.in_port, 0, true);
5969 /* Forwarding is disabled by STP. Let OFPP_NORMAL and the
5970 * learning action look at the packet, then drop it. */
5971 struct flow old_base_flow = ctx->base_flow;
5972 size_t old_size = ctx->odp_actions->size;
5973 xlate_table_action(ctx, ctx->flow.in_port, 0, true);
5974 ctx->base_flow = old_base_flow;
5975 ctx->odp_actions->size = old_size;
5979 ctx->flow = old_flow;
5980 ctx->ofproto = ofproto_dpif_cast(ofport->up.ofproto);
5982 if (ctx->resubmit_stats) {
5983 netdev_vport_inc_tx(ofport->up.netdev, ctx->resubmit_stats);
5984 netdev_vport_inc_rx(peer->up.netdev, ctx->resubmit_stats);
5990 flow_vlan_tci = ctx->flow.vlan_tci;
5991 flow_skb_mark = ctx->flow.skb_mark;
5992 flow_nw_tos = ctx->flow.nw_tos;
5994 pdscp = get_priority(ofport, ctx->flow.skb_priority);
5996 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
5997 ctx->flow.nw_tos |= pdscp->dscp;
6000 if (ofport->tnl_port) {
6001 /* Save tunnel metadata so that changes made due to
6002 * the Logical (tunnel) Port are not visible for any further
6003 * matches, while explicit set actions on tunnel metadata are.
6005 struct flow_tnl flow_tnl = ctx->flow.tunnel;
6006 odp_port = tnl_port_send(ofport->tnl_port, &ctx->flow);
6007 if (odp_port == OVSP_NONE) {
6008 xlate_report(ctx, "Tunneling decided against output");
6009 goto out; /* restore flow_nw_tos */
6011 if (ctx->flow.tunnel.ip_dst == ctx->orig_tunnel_ip_dst) {
6012 xlate_report(ctx, "Not tunneling to our own address");
6013 goto out; /* restore flow_nw_tos */
6015 if (ctx->resubmit_stats) {
6016 netdev_vport_inc_tx(ofport->up.netdev, ctx->resubmit_stats);
6018 out_port = odp_port;
6019 commit_odp_tunnel_action(&ctx->flow, &ctx->base_flow,
6021 ctx->flow.tunnel = flow_tnl; /* Restore tunnel metadata */
6023 odp_port = ofport->odp_port;
6024 out_port = vsp_realdev_to_vlandev(ctx->ofproto, odp_port,
6025 ctx->flow.vlan_tci);
6026 if (out_port != odp_port) {
6027 ctx->flow.vlan_tci = htons(0);
6029 ctx->flow.skb_mark &= ~IPSEC_MARK;
6031 commit_odp_actions(&ctx->flow, &ctx->base_flow, ctx->odp_actions);
6032 nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_OUTPUT, out_port);
6034 ctx->sflow_odp_port = odp_port;
6035 ctx->sflow_n_outputs++;
6036 ctx->nf_output_iface = ofp_port;
6039 ctx->flow.vlan_tci = flow_vlan_tci;
6040 ctx->flow.skb_mark = flow_skb_mark;
6042 ctx->flow.nw_tos = flow_nw_tos;
6046 compose_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port)
6048 compose_output_action__(ctx, ofp_port, true);
6052 tag_the_flow(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
6054 struct ofproto_dpif *ofproto = ctx->ofproto;
6055 uint8_t table_id = ctx->table_id;
6057 if (table_id > 0 && table_id < N_TABLES) {
6058 struct table_dpif *table = &ofproto->tables[table_id];
6059 if (table->other_table) {
6060 ctx->tags |= (rule && rule->tag
6062 : rule_calculate_tag(&ctx->flow,
6063 &table->other_table->mask,
6069 /* Common rule processing in one place to avoid duplicating code. */
6070 static struct rule_dpif *
6071 ctx_rule_hooks(struct action_xlate_ctx *ctx, struct rule_dpif *rule,
6074 if (ctx->resubmit_hook) {
6075 ctx->resubmit_hook(ctx, rule);
6077 if (rule == NULL && may_packet_in) {
6079 * check if table configuration flags
6080 * OFPTC_TABLE_MISS_CONTROLLER, default.
6081 * OFPTC_TABLE_MISS_CONTINUE,
6082 * OFPTC_TABLE_MISS_DROP
6083 * When OF1.0, OFPTC_TABLE_MISS_CONTINUE is used. What to do?
6085 rule = rule_dpif_miss_rule(ctx->ofproto, &ctx->flow);
6087 if (rule && ctx->resubmit_stats) {
6088 rule_credit_stats(rule, ctx->resubmit_stats);
6094 xlate_table_action(struct action_xlate_ctx *ctx,
6095 uint16_t in_port, uint8_t table_id, bool may_packet_in)
6097 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
6098 struct rule_dpif *rule;
6099 uint16_t old_in_port = ctx->flow.in_port;
6100 uint8_t old_table_id = ctx->table_id;
6102 ctx->table_id = table_id;
6104 /* Look up a flow with 'in_port' as the input port. */
6105 ctx->flow.in_port = in_port;
6106 rule = rule_dpif_lookup__(ctx->ofproto, &ctx->flow, table_id);
6108 tag_the_flow(ctx, rule);
6110 /* Restore the original input port. Otherwise OFPP_NORMAL and
6111 * OFPP_IN_PORT will have surprising behavior. */
6112 ctx->flow.in_port = old_in_port;
6114 rule = ctx_rule_hooks(ctx, rule, may_packet_in);
6117 struct rule_dpif *old_rule = ctx->rule;
6121 do_xlate_actions(rule->up.ofpacts, rule->up.ofpacts_len, ctx);
6122 ctx->rule = old_rule;
6126 ctx->table_id = old_table_id;
6128 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
6130 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
6131 MAX_RESUBMIT_RECURSION);
6132 ctx->max_resubmit_trigger = true;
6137 xlate_ofpact_resubmit(struct action_xlate_ctx *ctx,
6138 const struct ofpact_resubmit *resubmit)
6143 in_port = resubmit->in_port;
6144 if (in_port == OFPP_IN_PORT) {
6145 in_port = ctx->flow.in_port;
6148 table_id = resubmit->table_id;
6149 if (table_id == 255) {
6150 table_id = ctx->table_id;
6153 xlate_table_action(ctx, in_port, table_id, false);
6157 flood_packets(struct action_xlate_ctx *ctx, bool all)
6159 struct ofport_dpif *ofport;
6161 HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) {
6162 uint16_t ofp_port = ofport->up.ofp_port;
6164 if (ofp_port == ctx->flow.in_port) {
6169 compose_output_action__(ctx, ofp_port, false);
6170 } else if (!(ofport->up.pp.config & OFPUTIL_PC_NO_FLOOD)) {
6171 compose_output_action(ctx, ofp_port);
6175 ctx->nf_output_iface = NF_OUT_FLOOD;
6179 execute_controller_action(struct action_xlate_ctx *ctx, int len,
6180 enum ofp_packet_in_reason reason,
6181 uint16_t controller_id)
6183 struct ofputil_packet_in pin;
6184 struct ofpbuf *packet;
6186 ovs_assert(!ctx->slow || ctx->slow == SLOW_CONTROLLER);
6187 ctx->slow = SLOW_CONTROLLER;
6192 packet = ofpbuf_clone(ctx->packet);
6194 if (packet->l2 && packet->l3) {
6195 struct eth_header *eh;
6196 uint16_t mpls_depth;
6198 eth_pop_vlan(packet);
6201 memcpy(eh->eth_src, ctx->flow.dl_src, sizeof eh->eth_src);
6202 memcpy(eh->eth_dst, ctx->flow.dl_dst, sizeof eh->eth_dst);
6204 if (ctx->flow.vlan_tci & htons(VLAN_CFI)) {
6205 eth_push_vlan(packet, ctx->flow.vlan_tci);
6208 mpls_depth = eth_mpls_depth(packet);
6210 if (mpls_depth < ctx->flow.mpls_depth) {
6211 push_mpls(packet, ctx->flow.dl_type, ctx->flow.mpls_lse);
6212 } else if (mpls_depth > ctx->flow.mpls_depth) {
6213 pop_mpls(packet, ctx->flow.dl_type);
6214 } else if (mpls_depth) {
6215 set_mpls_lse(packet, ctx->flow.mpls_lse);
6219 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
6220 packet_set_ipv4(packet, ctx->flow.nw_src, ctx->flow.nw_dst,
6221 ctx->flow.nw_tos, ctx->flow.nw_ttl);
6225 if (ctx->flow.nw_proto == IPPROTO_TCP) {
6226 packet_set_tcp_port(packet, ctx->flow.tp_src,
6228 } else if (ctx->flow.nw_proto == IPPROTO_UDP) {
6229 packet_set_udp_port(packet, ctx->flow.tp_src,
6236 pin.packet = packet->data;
6237 pin.packet_len = packet->size;
6238 pin.reason = reason;
6239 pin.controller_id = controller_id;
6240 pin.table_id = ctx->table_id;
6241 pin.cookie = ctx->rule ? ctx->rule->up.flow_cookie : 0;
6244 flow_get_metadata(&ctx->flow, &pin.fmd);
6246 connmgr_send_packet_in(ctx->ofproto->up.connmgr, &pin);
6247 ofpbuf_delete(packet);
6251 execute_mpls_push_action(struct action_xlate_ctx *ctx, ovs_be16 eth_type)
6253 ovs_assert(eth_type_mpls(eth_type));
6255 if (ctx->base_flow.mpls_depth) {
6256 ctx->flow.mpls_lse &= ~htonl(MPLS_BOS_MASK);
6257 ctx->flow.mpls_depth++;
6262 if (ctx->flow.dl_type == htons(ETH_TYPE_IPV6)) {
6263 label = htonl(0x2); /* IPV6 Explicit Null. */
6265 label = htonl(0x0); /* IPV4 Explicit Null. */
6267 tc = (ctx->flow.nw_tos & IP_DSCP_MASK) >> 2;
6268 ttl = ctx->flow.nw_ttl ? ctx->flow.nw_ttl : 0x40;
6269 ctx->flow.mpls_lse = set_mpls_lse_values(ttl, tc, 1, label);
6270 ctx->flow.mpls_depth = 1;
6272 ctx->flow.dl_type = eth_type;
6276 execute_mpls_pop_action(struct action_xlate_ctx *ctx, ovs_be16 eth_type)
6278 ovs_assert(eth_type_mpls(ctx->flow.dl_type));
6279 ovs_assert(!eth_type_mpls(eth_type));
6281 if (ctx->flow.mpls_depth) {
6282 ctx->flow.mpls_depth--;
6283 ctx->flow.mpls_lse = htonl(0);
6284 if (!ctx->flow.mpls_depth) {
6285 ctx->flow.dl_type = eth_type;
6291 compose_dec_ttl(struct action_xlate_ctx *ctx, struct ofpact_cnt_ids *ids)
6293 if (ctx->flow.dl_type != htons(ETH_TYPE_IP) &&
6294 ctx->flow.dl_type != htons(ETH_TYPE_IPV6)) {
6298 if (ctx->flow.nw_ttl > 1) {
6304 for (i = 0; i < ids->n_controllers; i++) {
6305 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL,
6309 /* Stop processing for current table. */
6315 execute_set_mpls_ttl_action(struct action_xlate_ctx *ctx, uint8_t ttl)
6317 if (!eth_type_mpls(ctx->flow.dl_type)) {
6321 set_mpls_lse_ttl(&ctx->flow.mpls_lse, ttl);
6326 execute_dec_mpls_ttl_action(struct action_xlate_ctx *ctx)
6328 uint8_t ttl = mpls_lse_to_ttl(ctx->flow.mpls_lse);
6330 if (!eth_type_mpls(ctx->flow.dl_type)) {
6336 set_mpls_lse_ttl(&ctx->flow.mpls_lse, ttl);
6339 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL, 0);
6341 /* Stop processing for current table. */
6347 xlate_output_action(struct action_xlate_ctx *ctx,
6348 uint16_t port, uint16_t max_len, bool may_packet_in)
6350 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
6352 ctx->nf_output_iface = NF_OUT_DROP;
6356 compose_output_action(ctx, ctx->flow.in_port);
6359 xlate_table_action(ctx, ctx->flow.in_port, 0, may_packet_in);
6365 flood_packets(ctx, false);
6368 flood_packets(ctx, true);
6370 case OFPP_CONTROLLER:
6371 execute_controller_action(ctx, max_len, OFPR_ACTION, 0);
6377 if (port != ctx->flow.in_port) {
6378 compose_output_action(ctx, port);
6380 xlate_report(ctx, "skipping output to input port");
6385 if (prev_nf_output_iface == NF_OUT_FLOOD) {
6386 ctx->nf_output_iface = NF_OUT_FLOOD;
6387 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
6388 ctx->nf_output_iface = prev_nf_output_iface;
6389 } else if (prev_nf_output_iface != NF_OUT_DROP &&
6390 ctx->nf_output_iface != NF_OUT_FLOOD) {
6391 ctx->nf_output_iface = NF_OUT_MULTI;
6396 xlate_output_reg_action(struct action_xlate_ctx *ctx,
6397 const struct ofpact_output_reg *or)
6399 uint64_t port = mf_get_subfield(&or->src, &ctx->flow);
6400 if (port <= UINT16_MAX) {
6401 xlate_output_action(ctx, port, or->max_len, false);
6406 xlate_enqueue_action(struct action_xlate_ctx *ctx,
6407 const struct ofpact_enqueue *enqueue)
6409 uint16_t ofp_port = enqueue->port;
6410 uint32_t queue_id = enqueue->queue;
6411 uint32_t flow_priority, priority;
6414 /* Translate queue to priority. */
6415 error = dpif_queue_to_priority(ctx->ofproto->backer->dpif,
6416 queue_id, &priority);
6418 /* Fall back to ordinary output action. */
6419 xlate_output_action(ctx, enqueue->port, 0, false);
6423 /* Check output port. */
6424 if (ofp_port == OFPP_IN_PORT) {
6425 ofp_port = ctx->flow.in_port;
6426 } else if (ofp_port == ctx->flow.in_port) {
6430 /* Add datapath actions. */
6431 flow_priority = ctx->flow.skb_priority;
6432 ctx->flow.skb_priority = priority;
6433 compose_output_action(ctx, ofp_port);
6434 ctx->flow.skb_priority = flow_priority;
6436 /* Update NetFlow output port. */
6437 if (ctx->nf_output_iface == NF_OUT_DROP) {
6438 ctx->nf_output_iface = ofp_port;
6439 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
6440 ctx->nf_output_iface = NF_OUT_MULTI;
6445 xlate_set_queue_action(struct action_xlate_ctx *ctx, uint32_t queue_id)
6447 uint32_t skb_priority;
6449 if (!dpif_queue_to_priority(ctx->ofproto->backer->dpif,
6450 queue_id, &skb_priority)) {
6451 ctx->flow.skb_priority = skb_priority;
6453 /* Couldn't translate queue to a priority. Nothing to do. A warning
6454 * has already been logged. */
6459 slave_enabled_cb(uint16_t ofp_port, void *ofproto_)
6461 struct ofproto_dpif *ofproto = ofproto_;
6462 struct ofport_dpif *port;
6472 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
6475 port = get_ofp_port(ofproto, ofp_port);
6476 return port ? port->may_enable : false;
6481 xlate_bundle_action(struct action_xlate_ctx *ctx,
6482 const struct ofpact_bundle *bundle)
6486 port = bundle_execute(bundle, &ctx->flow, slave_enabled_cb, ctx->ofproto);
6487 if (bundle->dst.field) {
6488 nxm_reg_load(&bundle->dst, port, &ctx->flow);
6490 xlate_output_action(ctx, port, 0, false);
6495 xlate_learn_action(struct action_xlate_ctx *ctx,
6496 const struct ofpact_learn *learn)
6498 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
6499 struct ofputil_flow_mod fm;
6500 uint64_t ofpacts_stub[1024 / 8];
6501 struct ofpbuf ofpacts;
6504 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
6505 learn_execute(learn, &ctx->flow, &fm, &ofpacts);
6507 error = ofproto_flow_mod(&ctx->ofproto->up, &fm);
6508 if (error && !VLOG_DROP_WARN(&rl)) {
6509 VLOG_WARN("learning action failed to modify flow table (%s)",
6510 ofperr_get_name(error));
6513 ofpbuf_uninit(&ofpacts);
6516 /* Reduces '*timeout' to no more than 'max'. A value of zero in either case
6517 * means "infinite". */
6519 reduce_timeout(uint16_t max, uint16_t *timeout)
6521 if (max && (!*timeout || *timeout > max)) {
6527 xlate_fin_timeout(struct action_xlate_ctx *ctx,
6528 const struct ofpact_fin_timeout *oft)
6530 if (ctx->tcp_flags & (TCP_FIN | TCP_RST) && ctx->rule) {
6531 struct rule_dpif *rule = ctx->rule;
6533 reduce_timeout(oft->fin_idle_timeout, &rule->up.idle_timeout);
6534 reduce_timeout(oft->fin_hard_timeout, &rule->up.hard_timeout);
6539 xlate_sample_action(struct action_xlate_ctx *ctx,
6540 const struct ofpact_sample *os)
6542 union user_action_cookie cookie;
6543 /* Scale the probability from 16-bit to 32-bit while representing
6544 * the same percentage. */
6545 uint32_t probability = (os->probability << 16) | os->probability;
6547 commit_odp_actions(&ctx->flow, &ctx->base_flow, ctx->odp_actions);
6549 compose_flow_sample_cookie(os->probability, os->collector_set_id,
6550 os->obs_domain_id, os->obs_point_id, &cookie);
6551 compose_sample_action(ctx->ofproto, ctx->odp_actions, &ctx->flow,
6552 probability, &cookie, sizeof cookie.flow_sample);
6556 may_receive(const struct ofport_dpif *port, struct action_xlate_ctx *ctx)
6558 if (port->up.pp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
6559 ? OFPUTIL_PC_NO_RECV_STP
6560 : OFPUTIL_PC_NO_RECV)) {
6564 /* Only drop packets here if both forwarding and learning are
6565 * disabled. If just learning is enabled, we need to have
6566 * OFPP_NORMAL and the learning action have a look at the packet
6567 * before we can drop it. */
6568 if (!stp_forward_in_state(port->stp_state)
6569 && !stp_learn_in_state(port->stp_state)) {
6577 tunnel_ecn_ok(struct action_xlate_ctx *ctx)
6579 if (is_ip_any(&ctx->base_flow)
6580 && (ctx->flow.tunnel.ip_tos & IP_ECN_MASK) == IP_ECN_CE) {
6581 if ((ctx->base_flow.nw_tos & IP_ECN_MASK) == IP_ECN_NOT_ECT) {
6582 VLOG_WARN_RL(&rl, "dropping tunnel packet marked ECN CE"
6583 " but is not ECN capable");
6586 /* Set the ECN CE value in the tunneled packet. */
6587 ctx->flow.nw_tos |= IP_ECN_CE;
6595 do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len,
6596 struct action_xlate_ctx *ctx)
6598 bool was_evictable = true;
6599 const struct ofpact *a;
6602 /* Don't let the rule we're working on get evicted underneath us. */
6603 was_evictable = ctx->rule->up.evictable;
6604 ctx->rule->up.evictable = false;
6607 do_xlate_actions_again:
6608 OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) {
6609 struct ofpact_controller *controller;
6610 const struct ofpact_metadata *metadata;
6618 xlate_output_action(ctx, ofpact_get_OUTPUT(a)->port,
6619 ofpact_get_OUTPUT(a)->max_len, true);
6622 case OFPACT_CONTROLLER:
6623 controller = ofpact_get_CONTROLLER(a);
6624 execute_controller_action(ctx, controller->max_len,
6626 controller->controller_id);
6629 case OFPACT_ENQUEUE:
6630 xlate_enqueue_action(ctx, ofpact_get_ENQUEUE(a));
6633 case OFPACT_SET_VLAN_VID:
6634 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
6635 ctx->flow.vlan_tci |= (htons(ofpact_get_SET_VLAN_VID(a)->vlan_vid)
6639 case OFPACT_SET_VLAN_PCP:
6640 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
6641 ctx->flow.vlan_tci |= htons((ofpact_get_SET_VLAN_PCP(a)->vlan_pcp
6646 case OFPACT_STRIP_VLAN:
6647 ctx->flow.vlan_tci = htons(0);
6650 case OFPACT_PUSH_VLAN:
6651 /* XXX 802.1AD(QinQ) */
6652 ctx->flow.vlan_tci = htons(VLAN_CFI);
6655 case OFPACT_SET_ETH_SRC:
6656 memcpy(ctx->flow.dl_src, ofpact_get_SET_ETH_SRC(a)->mac,
6660 case OFPACT_SET_ETH_DST:
6661 memcpy(ctx->flow.dl_dst, ofpact_get_SET_ETH_DST(a)->mac,
6665 case OFPACT_SET_IPV4_SRC:
6666 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
6667 ctx->flow.nw_src = ofpact_get_SET_IPV4_SRC(a)->ipv4;
6671 case OFPACT_SET_IPV4_DST:
6672 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
6673 ctx->flow.nw_dst = ofpact_get_SET_IPV4_DST(a)->ipv4;
6677 case OFPACT_SET_IPV4_DSCP:
6678 /* OpenFlow 1.0 only supports IPv4. */
6679 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
6680 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
6681 ctx->flow.nw_tos |= ofpact_get_SET_IPV4_DSCP(a)->dscp;
6685 case OFPACT_SET_L4_SRC_PORT:
6686 if (is_ip_any(&ctx->flow)) {
6687 ctx->flow.tp_src = htons(ofpact_get_SET_L4_SRC_PORT(a)->port);
6691 case OFPACT_SET_L4_DST_PORT:
6692 if (is_ip_any(&ctx->flow)) {
6693 ctx->flow.tp_dst = htons(ofpact_get_SET_L4_DST_PORT(a)->port);
6697 case OFPACT_RESUBMIT:
6698 xlate_ofpact_resubmit(ctx, ofpact_get_RESUBMIT(a));
6701 case OFPACT_SET_TUNNEL:
6702 ctx->flow.tunnel.tun_id = htonll(ofpact_get_SET_TUNNEL(a)->tun_id);
6705 case OFPACT_SET_QUEUE:
6706 xlate_set_queue_action(ctx, ofpact_get_SET_QUEUE(a)->queue_id);
6709 case OFPACT_POP_QUEUE:
6710 ctx->flow.skb_priority = ctx->orig_skb_priority;
6713 case OFPACT_REG_MOVE:
6714 nxm_execute_reg_move(ofpact_get_REG_MOVE(a), &ctx->flow);
6717 case OFPACT_REG_LOAD:
6718 nxm_execute_reg_load(ofpact_get_REG_LOAD(a), &ctx->flow);
6721 case OFPACT_STACK_PUSH:
6722 nxm_execute_stack_push(ofpact_get_STACK_PUSH(a), &ctx->flow,
6726 case OFPACT_STACK_POP:
6727 nxm_execute_stack_pop(ofpact_get_STACK_POP(a), &ctx->flow,
6731 case OFPACT_PUSH_MPLS:
6732 execute_mpls_push_action(ctx, ofpact_get_PUSH_MPLS(a)->ethertype);
6735 case OFPACT_POP_MPLS:
6736 execute_mpls_pop_action(ctx, ofpact_get_POP_MPLS(a)->ethertype);
6739 case OFPACT_SET_MPLS_TTL:
6740 if (execute_set_mpls_ttl_action(ctx, ofpact_get_SET_MPLS_TTL(a)->ttl)) {
6745 case OFPACT_DEC_MPLS_TTL:
6746 if (execute_dec_mpls_ttl_action(ctx)) {
6751 case OFPACT_DEC_TTL:
6752 if (compose_dec_ttl(ctx, ofpact_get_DEC_TTL(a))) {
6758 /* Nothing to do. */
6761 case OFPACT_MULTIPATH:
6762 multipath_execute(ofpact_get_MULTIPATH(a), &ctx->flow);
6766 ctx->ofproto->has_bundle_action = true;
6767 xlate_bundle_action(ctx, ofpact_get_BUNDLE(a));
6770 case OFPACT_OUTPUT_REG:
6771 xlate_output_reg_action(ctx, ofpact_get_OUTPUT_REG(a));
6775 ctx->has_learn = true;
6776 if (ctx->may_learn) {
6777 xlate_learn_action(ctx, ofpact_get_LEARN(a));
6785 case OFPACT_FIN_TIMEOUT:
6786 ctx->has_fin_timeout = true;
6787 xlate_fin_timeout(ctx, ofpact_get_FIN_TIMEOUT(a));
6790 case OFPACT_CLEAR_ACTIONS:
6792 * Nothing to do because writa-actions is not supported for now.
6793 * When writa-actions is supported, clear-actions also must
6794 * be supported at the same time.
6798 case OFPACT_WRITE_METADATA:
6799 metadata = ofpact_get_WRITE_METADATA(a);
6800 ctx->flow.metadata &= ~metadata->mask;
6801 ctx->flow.metadata |= metadata->metadata & metadata->mask;
6804 case OFPACT_GOTO_TABLE: {
6805 /* It is assumed that goto-table is the last action. */
6806 struct ofpact_goto_table *ogt = ofpact_get_GOTO_TABLE(a);
6807 struct rule_dpif *rule;
6809 ovs_assert(ctx->table_id < ogt->table_id);
6811 ctx->table_id = ogt->table_id;
6813 /* Look up a flow from the new table. */
6814 rule = rule_dpif_lookup__(ctx->ofproto, &ctx->flow, ctx->table_id);
6816 tag_the_flow(ctx, rule);
6818 rule = ctx_rule_hooks(ctx, rule, true);
6822 ctx->rule->up.evictable = was_evictable;
6825 was_evictable = rule->up.evictable;
6826 rule->up.evictable = false;
6828 /* Tail recursion removal. */
6829 ofpacts = rule->up.ofpacts;
6830 ofpacts_len = rule->up.ofpacts_len;
6831 goto do_xlate_actions_again;
6837 xlate_sample_action(ctx, ofpact_get_SAMPLE(a));
6844 ctx->rule->up.evictable = was_evictable;
6849 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
6850 struct ofproto_dpif *ofproto, const struct flow *flow,
6851 const struct initial_vals *initial_vals,
6852 struct rule_dpif *rule,
6853 uint8_t tcp_flags, const struct ofpbuf *packet)
6855 /* Flow initialization rules:
6856 * - 'base_flow' must match the kernel's view of the packet at the
6857 * time that action processing starts. 'flow' represents any
6858 * transformations we wish to make through actions.
6859 * - By default 'base_flow' and 'flow' are the same since the input
6860 * packet matches the output before any actions are applied.
6861 * - When using VLAN splinters, 'base_flow''s VLAN is set to the value
6862 * of the received packet as seen by the kernel. If we later output
6863 * to another device without any modifications this will cause us to
6864 * insert a new tag since the original one was stripped off by the
6866 * - Tunnel metadata as received is retained in 'flow'. This allows
6867 * tunnel metadata matching also in later tables.
6868 * Since a kernel action for setting the tunnel metadata will only be
6869 * generated with actual tunnel output, changing the tunnel metadata
6870 * values in 'flow' (such as tun_id) will only have effect with a later
6871 * tunnel output action.
6872 * - Tunnel 'base_flow' is completely cleared since that is what the
6873 * kernel does. If we wish to maintain the original values an action
6874 * needs to be generated. */
6876 ctx->ofproto = ofproto;
6878 ctx->base_flow = ctx->flow;
6879 memset(&ctx->base_flow.tunnel, 0, sizeof ctx->base_flow.tunnel);
6880 ctx->orig_tunnel_ip_dst = flow->tunnel.ip_dst;
6882 ctx->packet = packet;
6883 ctx->may_learn = packet != NULL;
6884 ctx->tcp_flags = tcp_flags;
6885 ctx->resubmit_hook = NULL;
6886 ctx->report_hook = NULL;
6887 ctx->resubmit_stats = NULL;
6890 ctx->base_flow.vlan_tci = initial_vals->vlan_tci;
6894 /* Translates the 'ofpacts_len' bytes of "struct ofpacts" starting at 'ofpacts'
6895 * into datapath actions in 'odp_actions', using 'ctx'. */
6897 xlate_actions(struct action_xlate_ctx *ctx,
6898 const struct ofpact *ofpacts, size_t ofpacts_len,
6899 struct ofpbuf *odp_actions)
6901 /* Normally false. Set to true if we ever hit MAX_RESUBMIT_RECURSION, so
6902 * that in the future we always keep a copy of the original flow for
6903 * tracing purposes. */
6904 static bool hit_resubmit_limit;
6906 enum slow_path_reason special;
6907 struct ofport_dpif *in_port;
6908 struct flow orig_flow;
6910 COVERAGE_INC(ofproto_dpif_xlate);
6912 ofpbuf_clear(odp_actions);
6913 ofpbuf_reserve(odp_actions, NL_A_U32_SIZE);
6915 ctx->odp_actions = odp_actions;
6918 ctx->has_learn = false;
6919 ctx->has_normal = false;
6920 ctx->has_fin_timeout = false;
6921 ctx->nf_output_iface = NF_OUT_DROP;
6924 ctx->max_resubmit_trigger = false;
6925 ctx->orig_skb_priority = ctx->flow.skb_priority;
6929 ofpbuf_use_stub(&ctx->stack, ctx->init_stack, sizeof ctx->init_stack);
6931 if (ctx->ofproto->has_mirrors || hit_resubmit_limit) {
6932 /* Do this conditionally because the copy is expensive enough that it
6933 * shows up in profiles. */
6934 orig_flow = ctx->flow;
6937 if (ctx->flow.nw_frag & FLOW_NW_FRAG_ANY) {
6938 switch (ctx->ofproto->up.frag_handling) {
6939 case OFPC_FRAG_NORMAL:
6940 /* We must pretend that transport ports are unavailable. */
6941 ctx->flow.tp_src = ctx->base_flow.tp_src = htons(0);
6942 ctx->flow.tp_dst = ctx->base_flow.tp_dst = htons(0);
6945 case OFPC_FRAG_DROP:
6948 case OFPC_FRAG_REASM:
6951 case OFPC_FRAG_NX_MATCH:
6952 /* Nothing to do. */
6955 case OFPC_INVALID_TTL_TO_CONTROLLER:
6960 in_port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
6961 special = process_special(ctx->ofproto, &ctx->flow, in_port, ctx->packet);
6963 ctx->slow = special;
6965 static struct vlog_rate_limit trace_rl = VLOG_RATE_LIMIT_INIT(1, 1);
6966 struct initial_vals initial_vals;
6967 size_t sample_actions_len;
6968 uint32_t local_odp_port;
6970 initial_vals.vlan_tci = ctx->base_flow.vlan_tci;
6972 add_sflow_action(ctx);
6973 add_ipfix_action(ctx);
6974 sample_actions_len = ctx->odp_actions->size;
6976 if (tunnel_ecn_ok(ctx) && (!in_port || may_receive(in_port, ctx))) {
6977 do_xlate_actions(ofpacts, ofpacts_len, ctx);
6979 /* We've let OFPP_NORMAL and the learning action look at the
6980 * packet, so drop it now if forwarding is disabled. */
6981 if (in_port && !stp_forward_in_state(in_port->stp_state)) {
6982 ctx->odp_actions->size = sample_actions_len;
6986 if (ctx->max_resubmit_trigger && !ctx->resubmit_hook) {
6987 if (!hit_resubmit_limit) {
6988 /* We didn't record the original flow. Make sure we do from
6990 hit_resubmit_limit = true;
6991 } else if (!VLOG_DROP_ERR(&trace_rl)) {
6992 struct ds ds = DS_EMPTY_INITIALIZER;
6994 ofproto_trace(ctx->ofproto, &orig_flow, ctx->packet,
6995 &initial_vals, &ds);
6996 VLOG_ERR("Trace triggered by excessive resubmit "
6997 "recursion:\n%s", ds_cstr(&ds));
7002 local_odp_port = ofp_port_to_odp_port(ctx->ofproto, OFPP_LOCAL);
7003 if (!connmgr_must_output_local(ctx->ofproto->up.connmgr, &ctx->flow,
7005 ctx->odp_actions->data,
7006 ctx->odp_actions->size)) {
7007 compose_output_action(ctx, OFPP_LOCAL);
7009 if (ctx->ofproto->has_mirrors) {
7010 add_mirror_actions(ctx, &orig_flow);
7012 fix_sflow_action(ctx);
7015 ofpbuf_uninit(&ctx->stack);
7018 /* Translates the 'ofpacts_len' bytes of "struct ofpact"s starting at 'ofpacts'
7019 * into datapath actions, using 'ctx', and discards the datapath actions. */
7021 xlate_actions_for_side_effects(struct action_xlate_ctx *ctx,
7022 const struct ofpact *ofpacts,
7025 uint64_t odp_actions_stub[1024 / 8];
7026 struct ofpbuf odp_actions;
7028 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
7029 xlate_actions(ctx, ofpacts, ofpacts_len, &odp_actions);
7030 ofpbuf_uninit(&odp_actions);
7034 xlate_report(struct action_xlate_ctx *ctx, const char *s)
7036 if (ctx->report_hook) {
7037 ctx->report_hook(ctx, s);
7041 /* OFPP_NORMAL implementation. */
7043 static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *);
7045 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
7046 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle',
7047 * the bundle on which the packet was received, returns the VLAN to which the
7050 * Both 'vid' and the return value are in the range 0...4095. */
7052 input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid)
7054 switch (in_bundle->vlan_mode) {
7055 case PORT_VLAN_ACCESS:
7056 return in_bundle->vlan;
7059 case PORT_VLAN_TRUNK:
7062 case PORT_VLAN_NATIVE_UNTAGGED:
7063 case PORT_VLAN_NATIVE_TAGGED:
7064 return vid ? vid : in_bundle->vlan;
7071 /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'.
7072 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
7075 * 'vid' should be the VID obtained from the 802.1Q header that was received as
7076 * part of a packet (specify 0 if there was no 802.1Q header), in the range
7079 input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn)
7081 /* Allow any VID on the OFPP_NONE port. */
7082 if (in_bundle == &ofpp_none_bundle) {
7086 switch (in_bundle->vlan_mode) {
7087 case PORT_VLAN_ACCESS:
7090 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7091 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
7092 "packet received on port %s configured as VLAN "
7093 "%"PRIu16" access port",
7094 in_bundle->ofproto->up.name, vid,
7095 in_bundle->name, in_bundle->vlan);
7101 case PORT_VLAN_NATIVE_UNTAGGED:
7102 case PORT_VLAN_NATIVE_TAGGED:
7104 /* Port must always carry its native VLAN. */
7108 case PORT_VLAN_TRUNK:
7109 if (!ofbundle_includes_vlan(in_bundle, vid)) {
7111 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7112 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet "
7113 "received on port %s not configured for trunking "
7115 in_bundle->ofproto->up.name, vid,
7116 in_bundle->name, vid);
7128 /* Given 'vlan', the VLAN that a packet belongs to, and
7129 * 'out_bundle', a bundle on which the packet is to be output, returns the VID
7130 * that should be included in the 802.1Q header. (If the return value is 0,
7131 * then the 802.1Q header should only be included in the packet if there is a
7134 * Both 'vlan' and the return value are in the range 0...4095. */
7136 output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan)
7138 switch (out_bundle->vlan_mode) {
7139 case PORT_VLAN_ACCESS:
7142 case PORT_VLAN_TRUNK:
7143 case PORT_VLAN_NATIVE_TAGGED:
7146 case PORT_VLAN_NATIVE_UNTAGGED:
7147 return vlan == out_bundle->vlan ? 0 : vlan;
7155 output_normal(struct action_xlate_ctx *ctx, const struct ofbundle *out_bundle,
7158 struct ofport_dpif *port;
7160 ovs_be16 tci, old_tci;
7162 vid = output_vlan_to_vid(out_bundle, vlan);
7163 if (!out_bundle->bond) {
7164 port = ofbundle_get_a_port(out_bundle);
7166 port = bond_choose_output_slave(out_bundle->bond, &ctx->flow,
7169 /* No slaves enabled, so drop packet. */
7174 old_tci = ctx->flow.vlan_tci;
7176 if (tci || out_bundle->use_priority_tags) {
7177 tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK);
7179 tci |= htons(VLAN_CFI);
7182 ctx->flow.vlan_tci = tci;
7184 compose_output_action(ctx, port->up.ofp_port);
7185 ctx->flow.vlan_tci = old_tci;
7189 mirror_mask_ffs(mirror_mask_t mask)
7191 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
7196 ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
7198 return (bundle->vlan_mode != PORT_VLAN_ACCESS
7199 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
7203 ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan)
7205 return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan);
7208 /* Returns an arbitrary interface within 'bundle'. */
7209 static struct ofport_dpif *
7210 ofbundle_get_a_port(const struct ofbundle *bundle)
7212 return CONTAINER_OF(list_front(&bundle->ports),
7213 struct ofport_dpif, bundle_node);
7217 vlan_is_mirrored(const struct ofmirror *m, int vlan)
7219 return !m->vlans || bitmap_is_set(m->vlans, vlan);
7223 add_mirror_actions(struct action_xlate_ctx *ctx, const struct flow *orig_flow)
7225 struct ofproto_dpif *ofproto = ctx->ofproto;
7226 mirror_mask_t mirrors;
7227 struct ofbundle *in_bundle;
7230 const struct nlattr *a;
7233 in_bundle = lookup_input_bundle(ctx->ofproto, orig_flow->in_port,
7234 ctx->packet != NULL, NULL);
7238 mirrors = in_bundle->src_mirrors;
7240 /* Drop frames on bundles reserved for mirroring. */
7241 if (in_bundle->mirror_out) {
7242 if (ctx->packet != NULL) {
7243 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7244 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
7245 "%s, which is reserved exclusively for mirroring",
7246 ctx->ofproto->up.name, in_bundle->name);
7252 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
7253 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
7256 vlan = input_vid_to_vlan(in_bundle, vid);
7258 /* Look at the output ports to check for destination selections. */
7260 NL_ATTR_FOR_EACH (a, left, ctx->odp_actions->data,
7261 ctx->odp_actions->size) {
7262 enum ovs_action_attr type = nl_attr_type(a);
7263 struct ofport_dpif *ofport;
7265 if (type != OVS_ACTION_ATTR_OUTPUT) {
7269 ofport = get_odp_port(ofproto, nl_attr_get_u32(a));
7270 if (ofport && ofport->bundle) {
7271 mirrors |= ofport->bundle->dst_mirrors;
7279 /* Restore the original packet before adding the mirror actions. */
7280 ctx->flow = *orig_flow;
7285 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
7287 if (!vlan_is_mirrored(m, vlan)) {
7288 mirrors = zero_rightmost_1bit(mirrors);
7292 mirrors &= ~m->dup_mirrors;
7293 ctx->mirrors |= m->dup_mirrors;
7295 output_normal(ctx, m->out, vlan);
7296 } else if (vlan != m->out_vlan
7297 && !eth_addr_is_reserved(orig_flow->dl_dst)) {
7298 struct ofbundle *bundle;
7300 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
7301 if (ofbundle_includes_vlan(bundle, m->out_vlan)
7302 && !bundle->mirror_out) {
7303 output_normal(ctx, bundle, m->out_vlan);
7311 update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors,
7312 uint64_t packets, uint64_t bytes)
7318 for (; mirrors; mirrors = zero_rightmost_1bit(mirrors)) {
7321 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
7324 /* In normal circumstances 'm' will not be NULL. However,
7325 * if mirrors are reconfigured, we can temporarily get out
7326 * of sync in facet_revalidate(). We could "correct" the
7327 * mirror list before reaching here, but doing that would
7328 * not properly account the traffic stats we've currently
7329 * accumulated for previous mirror configuration. */
7333 m->packet_count += packets;
7334 m->byte_count += bytes;
7338 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
7339 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
7340 * indicate this; newer upstream kernels use gratuitous ARP requests. */
7342 is_gratuitous_arp(const struct flow *flow)
7344 return (flow->dl_type == htons(ETH_TYPE_ARP)
7345 && eth_addr_is_broadcast(flow->dl_dst)
7346 && (flow->nw_proto == ARP_OP_REPLY
7347 || (flow->nw_proto == ARP_OP_REQUEST
7348 && flow->nw_src == flow->nw_dst)));
7352 update_learning_table(struct ofproto_dpif *ofproto,
7353 const struct flow *flow, int vlan,
7354 struct ofbundle *in_bundle)
7356 struct mac_entry *mac;
7358 /* Don't learn the OFPP_NONE port. */
7359 if (in_bundle == &ofpp_none_bundle) {
7363 if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
7367 mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan);
7368 if (is_gratuitous_arp(flow)) {
7369 /* We don't want to learn from gratuitous ARP packets that are
7370 * reflected back over bond slaves so we lock the learning table. */
7371 if (!in_bundle->bond) {
7372 mac_entry_set_grat_arp_lock(mac);
7373 } else if (mac_entry_is_grat_arp_locked(mac)) {
7378 if (mac_entry_is_new(mac) || mac->port.p != in_bundle) {
7379 /* The log messages here could actually be useful in debugging,
7380 * so keep the rate limit relatively high. */
7381 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
7382 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
7383 "on port %s in VLAN %d",
7384 ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src),
7385 in_bundle->name, vlan);
7387 mac->port.p = in_bundle;
7388 tag_set_add(&ofproto->backer->revalidate_set,
7389 mac_learning_changed(ofproto->ml, mac));
7393 static struct ofbundle *
7394 lookup_input_bundle(const struct ofproto_dpif *ofproto, uint16_t in_port,
7395 bool warn, struct ofport_dpif **in_ofportp)
7397 struct ofport_dpif *ofport;
7399 /* Find the port and bundle for the received packet. */
7400 ofport = get_ofp_port(ofproto, in_port);
7402 *in_ofportp = ofport;
7404 if (ofport && ofport->bundle) {
7405 return ofport->bundle;
7408 /* Special-case OFPP_NONE, which a controller may use as the ingress
7409 * port for traffic that it is sourcing. */
7410 if (in_port == OFPP_NONE) {
7411 return &ofpp_none_bundle;
7414 /* Odd. A few possible reasons here:
7416 * - We deleted a port but there are still a few packets queued up
7419 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
7420 * we don't know about.
7422 * - The ofproto client didn't configure the port as part of a bundle.
7423 * This is particularly likely to happen if a packet was received on the
7424 * port after it was created, but before the client had a chance to
7425 * configure its bundle.
7428 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7430 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
7431 "port %"PRIu16, ofproto->up.name, in_port);
7436 /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or
7437 * dropped. Returns true if they may be forwarded, false if they should be
7440 * 'in_port' must be the ofport_dpif that corresponds to flow->in_port.
7441 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
7443 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
7444 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
7445 * checked by input_vid_is_valid().
7447 * May also add tags to '*tags', although the current implementation only does
7448 * so in one special case.
7451 is_admissible(struct action_xlate_ctx *ctx, struct ofport_dpif *in_port,
7454 struct ofproto_dpif *ofproto = ctx->ofproto;
7455 struct flow *flow = &ctx->flow;
7456 struct ofbundle *in_bundle = in_port->bundle;
7458 /* Drop frames for reserved multicast addresses
7459 * only if forward_bpdu option is absent. */
7460 if (!ofproto->up.forward_bpdu && eth_addr_is_reserved(flow->dl_dst)) {
7461 xlate_report(ctx, "packet has reserved destination MAC, dropping");
7465 if (in_bundle->bond) {
7466 struct mac_entry *mac;
7468 switch (bond_check_admissibility(in_bundle->bond, in_port,
7469 flow->dl_dst, &ctx->tags)) {
7474 xlate_report(ctx, "bonding refused admissibility, dropping");
7477 case BV_DROP_IF_MOVED:
7478 mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL);
7479 if (mac && mac->port.p != in_bundle &&
7480 (!is_gratuitous_arp(flow)
7481 || mac_entry_is_grat_arp_locked(mac))) {
7482 xlate_report(ctx, "SLB bond thinks this packet looped back, "
7494 xlate_normal(struct action_xlate_ctx *ctx)
7496 struct ofport_dpif *in_port;
7497 struct ofbundle *in_bundle;
7498 struct mac_entry *mac;
7502 ctx->has_normal = true;
7504 in_bundle = lookup_input_bundle(ctx->ofproto, ctx->flow.in_port,
7505 ctx->packet != NULL, &in_port);
7507 xlate_report(ctx, "no input bundle, dropping");
7511 /* Drop malformed frames. */
7512 if (ctx->flow.dl_type == htons(ETH_TYPE_VLAN) &&
7513 !(ctx->flow.vlan_tci & htons(VLAN_CFI))) {
7514 if (ctx->packet != NULL) {
7515 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7516 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
7517 "VLAN tag received on port %s",
7518 ctx->ofproto->up.name, in_bundle->name);
7520 xlate_report(ctx, "partial VLAN tag, dropping");
7524 /* Drop frames on bundles reserved for mirroring. */
7525 if (in_bundle->mirror_out) {
7526 if (ctx->packet != NULL) {
7527 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7528 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
7529 "%s, which is reserved exclusively for mirroring",
7530 ctx->ofproto->up.name, in_bundle->name);
7532 xlate_report(ctx, "input port is mirror output port, dropping");
7537 vid = vlan_tci_to_vid(ctx->flow.vlan_tci);
7538 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
7539 xlate_report(ctx, "disallowed VLAN VID for this input port, dropping");
7542 vlan = input_vid_to_vlan(in_bundle, vid);
7544 /* Check other admissibility requirements. */
7545 if (in_port && !is_admissible(ctx, in_port, vlan)) {
7549 /* Learn source MAC. */
7550 if (ctx->may_learn) {
7551 update_learning_table(ctx->ofproto, &ctx->flow, vlan, in_bundle);
7554 /* Determine output bundle. */
7555 mac = mac_learning_lookup(ctx->ofproto->ml, ctx->flow.dl_dst, vlan,
7558 if (mac->port.p != in_bundle) {
7559 xlate_report(ctx, "forwarding to learned port");
7560 output_normal(ctx, mac->port.p, vlan);
7562 xlate_report(ctx, "learned port is input port, dropping");
7565 struct ofbundle *bundle;
7567 xlate_report(ctx, "no learned MAC for destination, flooding");
7568 HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
7569 if (bundle != in_bundle
7570 && ofbundle_includes_vlan(bundle, vlan)
7571 && bundle->floodable
7572 && !bundle->mirror_out) {
7573 output_normal(ctx, bundle, vlan);
7576 ctx->nf_output_iface = NF_OUT_FLOOD;
7580 /* Optimized flow revalidation.
7582 * It's a difficult problem, in general, to tell which facets need to have
7583 * their actions recalculated whenever the OpenFlow flow table changes. We
7584 * don't try to solve that general problem: for most kinds of OpenFlow flow
7585 * table changes, we recalculate the actions for every facet. This is
7586 * relatively expensive, but it's good enough if the OpenFlow flow table
7587 * doesn't change very often.
7589 * However, we can expect one particular kind of OpenFlow flow table change to
7590 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
7591 * of CPU on revalidating every facet whenever MAC learning modifies the flow
7592 * table, we add a special case that applies to flow tables in which every rule
7593 * has the same form (that is, the same wildcards), except that the table is
7594 * also allowed to have a single "catch-all" flow that matches all packets. We
7595 * optimize this case by tagging all of the facets that resubmit into the table
7596 * and invalidating the same tag whenever a flow changes in that table. The
7597 * end result is that we revalidate just the facets that need it (and sometimes
7598 * a few more, but not all of the facets or even all of the facets that
7599 * resubmit to the table modified by MAC learning). */
7601 /* Calculates the tag to use for 'flow' and mask 'mask' when it is inserted
7602 * into an OpenFlow table with the given 'basis'. */
7604 rule_calculate_tag(const struct flow *flow, const struct minimask *mask,
7607 if (minimask_is_catchall(mask)) {
7610 uint32_t hash = flow_hash_in_minimask(flow, mask, secret);
7611 return tag_create_deterministic(hash);
7615 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
7616 * taggability of that table.
7618 * This function must be called after *each* change to a flow table. If you
7619 * skip calling it on some changes then the pointer comparisons at the end can
7620 * be invalid if you get unlucky. For example, if a flow removal causes a
7621 * cls_table to be destroyed and then a flow insertion causes a cls_table with
7622 * different wildcards to be created with the same address, then this function
7623 * will incorrectly skip revalidation. */
7625 table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
7627 struct table_dpif *table = &ofproto->tables[table_id];
7628 const struct oftable *oftable = &ofproto->up.tables[table_id];
7629 struct cls_table *catchall, *other;
7630 struct cls_table *t;
7632 catchall = other = NULL;
7634 switch (hmap_count(&oftable->cls.tables)) {
7636 /* We could tag this OpenFlow table but it would make the logic a
7637 * little harder and it's a corner case that doesn't seem worth it
7643 HMAP_FOR_EACH (t, hmap_node, &oftable->cls.tables) {
7644 if (cls_table_is_catchall(t)) {
7646 } else if (!other) {
7649 /* Indicate that we can't tag this by setting both tables to
7650 * NULL. (We know that 'catchall' is already NULL.) */
7657 /* Can't tag this table. */
7661 if (table->catchall_table != catchall || table->other_table != other) {
7662 table->catchall_table = catchall;
7663 table->other_table = other;
7664 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
7668 /* Given 'rule' that has changed in some way (either it is a rule being
7669 * inserted, a rule being deleted, or a rule whose actions are being
7670 * modified), marks facets for revalidation to ensure that packets will be
7671 * forwarded correctly according to the new state of the flow table.
7673 * This function must be called after *each* change to a flow table. See
7674 * the comment on table_update_taggable() for more information. */
7676 rule_invalidate(const struct rule_dpif *rule)
7678 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
7680 table_update_taggable(ofproto, rule->up.table_id);
7682 if (!ofproto->backer->need_revalidate) {
7683 struct table_dpif *table = &ofproto->tables[rule->up.table_id];
7685 if (table->other_table && rule->tag) {
7686 tag_set_add(&ofproto->backer->revalidate_set, rule->tag);
7688 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
7694 set_frag_handling(struct ofproto *ofproto_,
7695 enum ofp_config_flags frag_handling)
7697 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7698 if (frag_handling != OFPC_FRAG_REASM) {
7699 ofproto->backer->need_revalidate = REV_RECONFIGURE;
7707 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
7708 const struct flow *flow,
7709 const struct ofpact *ofpacts, size_t ofpacts_len)
7711 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7712 struct initial_vals initial_vals;
7713 struct odputil_keybuf keybuf;
7714 struct dpif_flow_stats stats;
7718 struct action_xlate_ctx ctx;
7719 uint64_t odp_actions_stub[1024 / 8];
7720 struct ofpbuf odp_actions;
7722 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
7723 odp_flow_key_from_flow(&key, flow,
7724 ofp_port_to_odp_port(ofproto, flow->in_port));
7726 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
7728 initial_vals.vlan_tci = flow->vlan_tci;
7729 action_xlate_ctx_init(&ctx, ofproto, flow, &initial_vals, NULL,
7730 packet_get_tcp_flags(packet, flow), packet);
7731 ctx.resubmit_stats = &stats;
7733 ofpbuf_use_stub(&odp_actions,
7734 odp_actions_stub, sizeof odp_actions_stub);
7735 xlate_actions(&ctx, ofpacts, ofpacts_len, &odp_actions);
7736 dpif_execute(ofproto->backer->dpif, key.data, key.size,
7737 odp_actions.data, odp_actions.size, packet);
7738 ofpbuf_uninit(&odp_actions);
7746 set_netflow(struct ofproto *ofproto_,
7747 const struct netflow_options *netflow_options)
7749 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7751 if (netflow_options) {
7752 if (!ofproto->netflow) {
7753 ofproto->netflow = netflow_create();
7755 return netflow_set_options(ofproto->netflow, netflow_options);
7757 netflow_destroy(ofproto->netflow);
7758 ofproto->netflow = NULL;
7764 get_netflow_ids(const struct ofproto *ofproto_,
7765 uint8_t *engine_type, uint8_t *engine_id)
7767 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7769 dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
7773 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
7775 if (!facet_is_controller_flow(facet) &&
7776 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
7777 struct subfacet *subfacet;
7778 struct ofexpired expired;
7780 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
7781 if (subfacet->path == SF_FAST_PATH) {
7782 struct dpif_flow_stats stats;
7784 subfacet_install(subfacet, &facet->odp_actions, &stats);
7785 subfacet_update_stats(subfacet, &stats);
7789 expired.flow = facet->flow;
7790 expired.packet_count = facet->packet_count;
7791 expired.byte_count = facet->byte_count;
7792 expired.used = facet->used;
7793 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
7798 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
7800 struct facet *facet;
7802 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
7803 send_active_timeout(ofproto, facet);
7807 static struct ofproto_dpif *
7808 ofproto_dpif_lookup(const char *name)
7810 struct ofproto_dpif *ofproto;
7812 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
7813 hash_string(name, 0), &all_ofproto_dpifs) {
7814 if (!strcmp(ofproto->up.name, name)) {
7822 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
7823 const char *argv[], void *aux OVS_UNUSED)
7825 struct ofproto_dpif *ofproto;
7828 ofproto = ofproto_dpif_lookup(argv[1]);
7830 unixctl_command_reply_error(conn, "no such bridge");
7833 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
7835 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
7836 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
7840 unixctl_command_reply(conn, "table successfully flushed");
7844 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
7845 const char *argv[], void *aux OVS_UNUSED)
7847 struct ds ds = DS_EMPTY_INITIALIZER;
7848 const struct ofproto_dpif *ofproto;
7849 const struct mac_entry *e;
7851 ofproto = ofproto_dpif_lookup(argv[1]);
7853 unixctl_command_reply_error(conn, "no such bridge");
7857 ds_put_cstr(&ds, " port VLAN MAC Age\n");
7858 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
7859 struct ofbundle *bundle = e->port.p;
7860 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
7861 ofbundle_get_a_port(bundle)->odp_port,
7862 e->vlan, ETH_ADDR_ARGS(e->mac),
7863 mac_entry_age(ofproto->ml, e));
7865 unixctl_command_reply(conn, ds_cstr(&ds));
7870 struct action_xlate_ctx ctx;
7876 trace_format_rule(struct ds *result, uint8_t table_id, int level,
7877 const struct rule_dpif *rule)
7879 ds_put_char_multiple(result, '\t', level);
7881 ds_put_cstr(result, "No match\n");
7885 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
7886 table_id, ntohll(rule->up.flow_cookie));
7887 cls_rule_format(&rule->up.cr, result);
7888 ds_put_char(result, '\n');
7890 ds_put_char_multiple(result, '\t', level);
7891 ds_put_cstr(result, "OpenFlow ");
7892 ofpacts_format(rule->up.ofpacts, rule->up.ofpacts_len, result);
7893 ds_put_char(result, '\n');
7897 trace_format_flow(struct ds *result, int level, const char *title,
7898 struct trace_ctx *trace)
7900 ds_put_char_multiple(result, '\t', level);
7901 ds_put_format(result, "%s: ", title);
7902 if (flow_equal(&trace->ctx.flow, &trace->flow)) {
7903 ds_put_cstr(result, "unchanged");
7905 flow_format(result, &trace->ctx.flow);
7906 trace->flow = trace->ctx.flow;
7908 ds_put_char(result, '\n');
7912 trace_format_regs(struct ds *result, int level, const char *title,
7913 struct trace_ctx *trace)
7917 ds_put_char_multiple(result, '\t', level);
7918 ds_put_format(result, "%s:", title);
7919 for (i = 0; i < FLOW_N_REGS; i++) {
7920 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
7922 ds_put_char(result, '\n');
7926 trace_format_odp(struct ds *result, int level, const char *title,
7927 struct trace_ctx *trace)
7929 struct ofpbuf *odp_actions = trace->ctx.odp_actions;
7931 ds_put_char_multiple(result, '\t', level);
7932 ds_put_format(result, "%s: ", title);
7933 format_odp_actions(result, odp_actions->data, odp_actions->size);
7934 ds_put_char(result, '\n');
7938 trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
7940 struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx);
7941 struct ds *result = trace->result;
7943 ds_put_char(result, '\n');
7944 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
7945 trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace);
7946 trace_format_odp(result, ctx->recurse + 1, "Resubmitted odp", trace);
7947 trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule);
7951 trace_report(struct action_xlate_ctx *ctx, const char *s)
7953 struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx);
7954 struct ds *result = trace->result;
7956 ds_put_char_multiple(result, '\t', ctx->recurse);
7957 ds_put_cstr(result, s);
7958 ds_put_char(result, '\n');
7962 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
7963 void *aux OVS_UNUSED)
7965 const struct dpif_backer *backer;
7966 struct ofproto_dpif *ofproto;
7967 struct ofpbuf odp_key;
7968 struct ofpbuf *packet;
7969 struct initial_vals initial_vals;
7977 ofpbuf_init(&odp_key, 0);
7979 /* Handle "-generate" or a hex string as the last argument. */
7980 if (!strcmp(argv[argc - 1], "-generate")) {
7981 packet = ofpbuf_new(0);
7984 const char *error = eth_from_hex(argv[argc - 1], &packet);
7987 } else if (argc == 4) {
7988 /* The 3-argument form must end in "-generate' or a hex string. */
7989 unixctl_command_reply_error(conn, error);
7994 /* Parse the flow and determine whether a datapath or
7995 * bridge is specified. If function odp_flow_key_from_string()
7996 * returns 0, the flow is a odp_flow. If function
7997 * parse_ofp_exact_flow() returns 0, the flow is a br_flow. */
7998 if (!odp_flow_key_from_string(argv[argc - 1], NULL, &odp_key)) {
7999 /* If the odp_flow is the second argument,
8000 * the datapath name is the first argument. */
8002 const char *dp_type;
8003 if (!strncmp(argv[1], "ovs-", 4)) {
8004 dp_type = argv[1] + 4;
8008 backer = shash_find_data(&all_dpif_backers, dp_type);
8010 unixctl_command_reply_error(conn, "Cannot find datapath "
8015 /* No datapath name specified, so there should be only one
8017 struct shash_node *node;
8018 if (shash_count(&all_dpif_backers) != 1) {
8019 unixctl_command_reply_error(conn, "Must specify datapath "
8020 "name, there is more than one type of datapath");
8023 node = shash_first(&all_dpif_backers);
8024 backer = node->data;
8027 /* Extract the ofproto_dpif object from the ofproto_receive()
8029 if (ofproto_receive(backer, NULL, odp_key.data,
8030 odp_key.size, &flow, NULL, &ofproto, NULL,
8032 unixctl_command_reply_error(conn, "Invalid datapath flow");
8035 ds_put_format(&result, "Bridge: %s\n", ofproto->up.name);
8036 } else if (!parse_ofp_exact_flow(&flow, argv[argc - 1])) {
8038 unixctl_command_reply_error(conn, "Must specify bridge name");
8042 ofproto = ofproto_dpif_lookup(argv[1]);
8044 unixctl_command_reply_error(conn, "Unknown bridge name");
8047 initial_vals.vlan_tci = flow.vlan_tci;
8049 unixctl_command_reply_error(conn, "Bad flow syntax");
8053 /* Generate a packet, if requested. */
8055 if (!packet->size) {
8056 flow_compose(packet, &flow);
8058 ds_put_cstr(&result, "Packet: ");
8059 s = ofp_packet_to_string(packet->data, packet->size);
8060 ds_put_cstr(&result, s);
8063 /* Use the metadata from the flow and the packet argument
8064 * to reconstruct the flow. */
8065 flow_extract(packet, flow.skb_priority, flow.skb_mark, NULL,
8066 flow.in_port, &flow);
8067 initial_vals.vlan_tci = flow.vlan_tci;
8071 ofproto_trace(ofproto, &flow, packet, &initial_vals, &result);
8072 unixctl_command_reply(conn, ds_cstr(&result));
8075 ds_destroy(&result);
8076 ofpbuf_delete(packet);
8077 ofpbuf_uninit(&odp_key);
8081 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
8082 const struct ofpbuf *packet,
8083 const struct initial_vals *initial_vals, struct ds *ds)
8085 struct rule_dpif *rule;
8087 ds_put_cstr(ds, "Flow: ");
8088 flow_format(ds, flow);
8089 ds_put_char(ds, '\n');
8091 rule = rule_dpif_lookup(ofproto, flow);
8093 trace_format_rule(ds, 0, 0, rule);
8094 if (rule == ofproto->miss_rule) {
8095 ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
8096 } else if (rule == ofproto->no_packet_in_rule) {
8097 ds_put_cstr(ds, "\nNo match, packets dropped because "
8098 "OFPPC_NO_PACKET_IN is set on in_port.\n");
8102 uint64_t odp_actions_stub[1024 / 8];
8103 struct ofpbuf odp_actions;
8105 struct trace_ctx trace;
8108 tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
8111 ofpbuf_use_stub(&odp_actions,
8112 odp_actions_stub, sizeof odp_actions_stub);
8113 action_xlate_ctx_init(&trace.ctx, ofproto, flow, initial_vals,
8114 rule, tcp_flags, packet);
8115 trace.ctx.resubmit_hook = trace_resubmit;
8116 trace.ctx.report_hook = trace_report;
8117 xlate_actions(&trace.ctx, rule->up.ofpacts, rule->up.ofpacts_len,
8120 ds_put_char(ds, '\n');
8121 trace_format_flow(ds, 0, "Final flow", &trace);
8122 ds_put_cstr(ds, "Datapath actions: ");
8123 format_odp_actions(ds, odp_actions.data, odp_actions.size);
8124 ofpbuf_uninit(&odp_actions);
8126 if (trace.ctx.slow) {
8127 ds_put_cstr(ds, "\nThis flow is handled by the userspace "
8128 "slow path because it:");
8129 switch (trace.ctx.slow) {
8131 ds_put_cstr(ds, "\n\t- Consists of CFM packets.");
8134 ds_put_cstr(ds, "\n\t- Consists of LACP packets.");
8137 ds_put_cstr(ds, "\n\t- Consists of STP packets.");
8140 ds_put_cstr(ds, "\n\t- Consists of BFD packets.");
8142 case SLOW_CONTROLLER:
8143 ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages "
8144 "to the OpenFlow controller.");
8154 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
8155 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
8158 unixctl_command_reply(conn, NULL);
8162 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
8163 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
8166 unixctl_command_reply(conn, NULL);
8169 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
8170 * 'reply' describing the results. */
8172 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
8174 struct facet *facet;
8178 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
8179 if (!facet_check_consistency(facet)) {
8184 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
8188 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
8189 ofproto->up.name, errors);
8191 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
8196 ofproto_dpif_self_check(struct unixctl_conn *conn,
8197 int argc, const char *argv[], void *aux OVS_UNUSED)
8199 struct ds reply = DS_EMPTY_INITIALIZER;
8200 struct ofproto_dpif *ofproto;
8203 ofproto = ofproto_dpif_lookup(argv[1]);
8205 unixctl_command_reply_error(conn, "Unknown ofproto (use "
8206 "ofproto/list for help)");
8209 ofproto_dpif_self_check__(ofproto, &reply);
8211 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
8212 ofproto_dpif_self_check__(ofproto, &reply);
8216 unixctl_command_reply(conn, ds_cstr(&reply));
8220 /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list
8221 * of the 'ofproto_shash' nodes. It is the responsibility of the caller
8222 * to destroy 'ofproto_shash' and free the returned value. */
8223 static const struct shash_node **
8224 get_ofprotos(struct shash *ofproto_shash)
8226 const struct ofproto_dpif *ofproto;
8228 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
8229 char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
8230 shash_add_nocopy(ofproto_shash, name, ofproto);
8233 return shash_sort(ofproto_shash);
8237 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
8238 const char *argv[] OVS_UNUSED,
8239 void *aux OVS_UNUSED)
8241 struct ds ds = DS_EMPTY_INITIALIZER;
8242 struct shash ofproto_shash;
8243 const struct shash_node **sorted_ofprotos;
8246 shash_init(&ofproto_shash);
8247 sorted_ofprotos = get_ofprotos(&ofproto_shash);
8248 for (i = 0; i < shash_count(&ofproto_shash); i++) {
8249 const struct shash_node *node = sorted_ofprotos[i];
8250 ds_put_format(&ds, "%s\n", node->name);
8253 shash_destroy(&ofproto_shash);
8254 free(sorted_ofprotos);
8256 unixctl_command_reply(conn, ds_cstr(&ds));
8261 show_dp_format(const struct ofproto_dpif *ofproto, struct ds *ds)
8263 const struct shash_node **ports;
8265 struct avg_subfacet_rates lifetime;
8266 unsigned long long int minutes;
8267 const int min_ms = 60 * 1000; /* milliseconds in one minute. */
8269 minutes = (time_msec() - ofproto->created) / min_ms;
8272 lifetime.add_rate = (double)ofproto->total_subfacet_add_count
8274 lifetime.del_rate = (double)ofproto->total_subfacet_del_count
8277 lifetime.add_rate = 0.0;
8278 lifetime.del_rate = 0.0;
8281 ds_put_format(ds, "%s (%s):\n", ofproto->up.name,
8282 dpif_name(ofproto->backer->dpif));
8284 "\tlookups: hit:%"PRIu64" missed:%"PRIu64"\n",
8285 ofproto->n_hit, ofproto->n_missed);
8286 ds_put_format(ds, "\tflows: cur: %zu, avg: %5.3f, max: %d,"
8287 " life span: %llu(ms)\n",
8288 hmap_count(&ofproto->subfacets),
8289 avg_subfacet_count(ofproto),
8290 ofproto->max_n_subfacet,
8291 avg_subfacet_life_span(ofproto));
8292 if (minutes >= 60) {
8293 show_dp_rates(ds, "\t\thourly avg:", &ofproto->hourly);
8295 if (minutes >= 60 * 24) {
8296 show_dp_rates(ds, "\t\tdaily avg:", &ofproto->daily);
8298 show_dp_rates(ds, "\t\toverall avg:", &lifetime);
8300 ports = shash_sort(&ofproto->up.port_by_name);
8301 for (i = 0; i < shash_count(&ofproto->up.port_by_name); i++) {
8302 const struct shash_node *node = ports[i];
8303 struct ofport *ofport = node->data;
8304 const char *name = netdev_get_name(ofport->netdev);
8305 const char *type = netdev_get_type(ofport->netdev);
8308 ds_put_format(ds, "\t%s %u/", name, ofport->ofp_port);
8310 odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
8311 if (odp_port != OVSP_NONE) {
8312 ds_put_format(ds, "%"PRIu32":", odp_port);
8314 ds_put_cstr(ds, "none:");
8317 if (strcmp(type, "system")) {
8318 struct netdev *netdev;
8321 ds_put_format(ds, " (%s", type);
8323 error = netdev_open(name, type, &netdev);
8328 error = netdev_get_config(netdev, &config);
8330 const struct smap_node **nodes;
8333 nodes = smap_sort(&config);
8334 for (i = 0; i < smap_count(&config); i++) {
8335 const struct smap_node *node = nodes[i];
8336 ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
8337 node->key, node->value);
8341 smap_destroy(&config);
8343 netdev_close(netdev);
8345 ds_put_char(ds, ')');
8347 ds_put_char(ds, '\n');
8353 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc,
8354 const char *argv[], void *aux OVS_UNUSED)
8356 struct ds ds = DS_EMPTY_INITIALIZER;
8357 const struct ofproto_dpif *ofproto;
8361 for (i = 1; i < argc; i++) {
8362 ofproto = ofproto_dpif_lookup(argv[i]);
8364 ds_put_format(&ds, "Unknown bridge %s (use dpif/dump-dps "
8365 "for help)", argv[i]);
8366 unixctl_command_reply_error(conn, ds_cstr(&ds));
8369 show_dp_format(ofproto, &ds);
8372 struct shash ofproto_shash;
8373 const struct shash_node **sorted_ofprotos;
8376 shash_init(&ofproto_shash);
8377 sorted_ofprotos = get_ofprotos(&ofproto_shash);
8378 for (i = 0; i < shash_count(&ofproto_shash); i++) {
8379 const struct shash_node *node = sorted_ofprotos[i];
8380 show_dp_format(node->data, &ds);
8383 shash_destroy(&ofproto_shash);
8384 free(sorted_ofprotos);
8387 unixctl_command_reply(conn, ds_cstr(&ds));
8392 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
8393 int argc OVS_UNUSED, const char *argv[],
8394 void *aux OVS_UNUSED)
8396 struct ds ds = DS_EMPTY_INITIALIZER;
8397 const struct ofproto_dpif *ofproto;
8398 struct subfacet *subfacet;
8400 ofproto = ofproto_dpif_lookup(argv[1]);
8402 unixctl_command_reply_error(conn, "no such bridge");
8406 update_stats(ofproto->backer);
8408 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
8409 struct facet *facet = subfacet->facet;
8411 odp_flow_key_format(subfacet->key, subfacet->key_len, &ds);
8413 ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:",
8414 subfacet->dp_packet_count, subfacet->dp_byte_count);
8415 if (subfacet->used) {
8416 ds_put_format(&ds, "%.3fs",
8417 (time_msec() - subfacet->used) / 1000.0);
8419 ds_put_format(&ds, "never");
8421 if (subfacet->facet->tcp_flags) {
8422 ds_put_cstr(&ds, ", flags:");
8423 packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags);
8426 ds_put_cstr(&ds, ", actions:");
8428 uint64_t slow_path_stub[128 / 8];
8429 const struct nlattr *actions;
8432 compose_slow_path(ofproto, &facet->flow, facet->slow,
8433 slow_path_stub, sizeof slow_path_stub,
8434 &actions, &actions_len);
8435 format_odp_actions(&ds, actions, actions_len);
8437 format_odp_actions(&ds, facet->odp_actions.data,
8438 facet->odp_actions.size);
8440 ds_put_char(&ds, '\n');
8443 unixctl_command_reply(conn, ds_cstr(&ds));
8448 ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn,
8449 int argc OVS_UNUSED, const char *argv[],
8450 void *aux OVS_UNUSED)
8452 struct ds ds = DS_EMPTY_INITIALIZER;
8453 struct ofproto_dpif *ofproto;
8455 ofproto = ofproto_dpif_lookup(argv[1]);
8457 unixctl_command_reply_error(conn, "no such bridge");
8461 flush(&ofproto->up);
8463 unixctl_command_reply(conn, ds_cstr(&ds));
8468 ofproto_dpif_unixctl_init(void)
8470 static bool registered;
8476 unixctl_command_register(
8478 "[dp_name]|bridge odp_flow|br_flow [-generate|packet]",
8479 1, 3, ofproto_unixctl_trace, NULL);
8480 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
8481 ofproto_unixctl_fdb_flush, NULL);
8482 unixctl_command_register("fdb/show", "bridge", 1, 1,
8483 ofproto_unixctl_fdb_show, NULL);
8484 unixctl_command_register("ofproto/clog", "", 0, 0,
8485 ofproto_dpif_clog, NULL);
8486 unixctl_command_register("ofproto/unclog", "", 0, 0,
8487 ofproto_dpif_unclog, NULL);
8488 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
8489 ofproto_dpif_self_check, NULL);
8490 unixctl_command_register("dpif/dump-dps", "", 0, 0,
8491 ofproto_unixctl_dpif_dump_dps, NULL);
8492 unixctl_command_register("dpif/show", "[bridge]", 0, INT_MAX,
8493 ofproto_unixctl_dpif_show, NULL);
8494 unixctl_command_register("dpif/dump-flows", "bridge", 1, 1,
8495 ofproto_unixctl_dpif_dump_flows, NULL);
8496 unixctl_command_register("dpif/del-flows", "bridge", 1, 1,
8497 ofproto_unixctl_dpif_del_flows, NULL);
8500 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
8502 * This is deprecated. It is only for compatibility with broken device drivers
8503 * in old versions of Linux that do not properly support VLANs when VLAN
8504 * devices are not used. When broken device drivers are no longer in
8505 * widespread use, we will delete these interfaces. */
8508 set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid)
8510 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
8511 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
8513 if (realdev_ofp_port == ofport->realdev_ofp_port
8514 && vid == ofport->vlandev_vid) {
8518 ofproto->backer->need_revalidate = REV_RECONFIGURE;
8520 if (ofport->realdev_ofp_port) {
8523 if (realdev_ofp_port && ofport->bundle) {
8524 /* vlandevs are enslaved to their realdevs, so they are not allowed to
8525 * themselves be part of a bundle. */
8526 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
8529 ofport->realdev_ofp_port = realdev_ofp_port;
8530 ofport->vlandev_vid = vid;
8532 if (realdev_ofp_port) {
8533 vsp_add(ofport, realdev_ofp_port, vid);
8540 hash_realdev_vid(uint16_t realdev_ofp_port, int vid)
8542 return hash_2words(realdev_ofp_port, vid);
8545 /* Returns the ODP port number of the Linux VLAN device that corresponds to
8546 * 'vlan_tci' on the network device with port number 'realdev_odp_port' in
8547 * 'ofproto'. For example, given 'realdev_odp_port' of eth0 and 'vlan_tci' 9,
8548 * it would return the port number of eth0.9.
8550 * Unless VLAN splinters are enabled for port 'realdev_odp_port', this
8551 * function just returns its 'realdev_odp_port' argument. */
8553 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
8554 uint32_t realdev_odp_port, ovs_be16 vlan_tci)
8556 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
8557 uint16_t realdev_ofp_port;
8558 int vid = vlan_tci_to_vid(vlan_tci);
8559 const struct vlan_splinter *vsp;
8561 realdev_ofp_port = odp_port_to_ofp_port(ofproto, realdev_odp_port);
8562 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
8563 hash_realdev_vid(realdev_ofp_port, vid),
8564 &ofproto->realdev_vid_map) {
8565 if (vsp->realdev_ofp_port == realdev_ofp_port
8566 && vsp->vid == vid) {
8567 return ofp_port_to_odp_port(ofproto, vsp->vlandev_ofp_port);
8571 return realdev_odp_port;
8574 static struct vlan_splinter *
8575 vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port)
8577 struct vlan_splinter *vsp;
8579 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, hash_int(vlandev_ofp_port, 0),
8580 &ofproto->vlandev_map) {
8581 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
8589 /* Returns the OpenFlow port number of the "real" device underlying the Linux
8590 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
8591 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
8592 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
8593 * eth0 and store 9 in '*vid'.
8595 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
8596 * VLAN device. Unless VLAN splinters are enabled, this is what this function
8599 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
8600 uint16_t vlandev_ofp_port, int *vid)
8602 if (!hmap_is_empty(&ofproto->vlandev_map)) {
8603 const struct vlan_splinter *vsp;
8605 vsp = vlandev_find(ofproto, vlandev_ofp_port);
8610 return vsp->realdev_ofp_port;
8616 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
8617 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
8618 * 'flow->in_port' to the "real" device backing the VLAN device, sets
8619 * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is
8620 * always the case unless VLAN splinters are enabled), returns false without
8621 * making any changes. */
8623 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
8628 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port, &vid);
8633 /* Cause the flow to be processed as if it came in on the real device with
8634 * the VLAN device's VLAN ID. */
8635 flow->in_port = realdev;
8636 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
8641 vsp_remove(struct ofport_dpif *port)
8643 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
8644 struct vlan_splinter *vsp;
8646 vsp = vlandev_find(ofproto, port->up.ofp_port);
8648 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
8649 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
8652 port->realdev_ofp_port = 0;
8654 VLOG_ERR("missing vlan device record");
8659 vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid)
8661 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
8663 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
8664 && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid))
8665 == realdev_ofp_port)) {
8666 struct vlan_splinter *vsp;
8668 vsp = xmalloc(sizeof *vsp);
8669 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
8670 hash_int(port->up.ofp_port, 0));
8671 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
8672 hash_realdev_vid(realdev_ofp_port, vid));
8673 vsp->realdev_ofp_port = realdev_ofp_port;
8674 vsp->vlandev_ofp_port = port->up.ofp_port;
8677 port->realdev_ofp_port = realdev_ofp_port;
8679 VLOG_ERR("duplicate vlan device record");
8684 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
8686 const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
8687 return ofport ? ofport->odp_port : OVSP_NONE;
8690 static struct ofport_dpif *
8691 odp_port_to_ofport(const struct dpif_backer *backer, uint32_t odp_port)
8693 struct ofport_dpif *port;
8695 HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node,
8696 hash_int(odp_port, 0),
8697 &backer->odp_to_ofport_map) {
8698 if (port->odp_port == odp_port) {
8707 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
8709 struct ofport_dpif *port;
8711 port = odp_port_to_ofport(ofproto->backer, odp_port);
8712 if (port && &ofproto->up == port->up.ofproto) {
8713 return port->up.ofp_port;
8718 static unsigned long long int
8719 avg_subfacet_life_span(const struct ofproto_dpif *ofproto)
8721 unsigned long long int dc;
8722 unsigned long long int avg;
8724 dc = ofproto->total_subfacet_del_count + ofproto->subfacet_del_count;
8725 avg = dc ? ofproto->total_subfacet_life_span / dc : 0;
8731 avg_subfacet_count(const struct ofproto_dpif *ofproto)
8735 if (ofproto->n_update_stats) {
8736 avg_c = (double)ofproto->total_subfacet_count
8737 / ofproto->n_update_stats;
8744 show_dp_rates(struct ds *ds, const char *heading,
8745 const struct avg_subfacet_rates *rates)
8747 ds_put_format(ds, "%s add rate: %5.3f/min, del rate: %5.3f/min\n",
8748 heading, rates->add_rate, rates->del_rate);
8752 update_max_subfacet_count(struct ofproto_dpif *ofproto)
8754 ofproto->max_n_subfacet = MAX(ofproto->max_n_subfacet,
8755 hmap_count(&ofproto->subfacets));
8758 /* Compute exponentially weighted moving average, adding 'new' as the newest,
8759 * most heavily weighted element. 'base' designates the rate of decay: after
8760 * 'base' further updates, 'new''s weight in the EWMA decays to about 1/e
8763 exp_mavg(double *avg, int base, double new)
8765 *avg = (*avg * (base - 1) + new) / base;
8769 update_moving_averages(struct ofproto_dpif *ofproto)
8771 const int min_ms = 60 * 1000; /* milliseconds in one minute. */
8773 /* Update hourly averages on the minute boundaries. */
8774 if (time_msec() - ofproto->last_minute >= min_ms) {
8775 exp_mavg(&ofproto->hourly.add_rate, 60, ofproto->subfacet_add_count);
8776 exp_mavg(&ofproto->hourly.del_rate, 60, ofproto->subfacet_del_count);
8778 /* Update daily averages on the hour boundaries. */
8779 if ((ofproto->last_minute - ofproto->created) / min_ms % 60 == 59) {
8780 exp_mavg(&ofproto->daily.add_rate, 24, ofproto->hourly.add_rate);
8781 exp_mavg(&ofproto->daily.del_rate, 24, ofproto->hourly.del_rate);
8784 ofproto->total_subfacet_add_count += ofproto->subfacet_add_count;
8785 ofproto->total_subfacet_del_count += ofproto->subfacet_del_count;
8786 ofproto->subfacet_add_count = 0;
8787 ofproto->subfacet_del_count = 0;
8788 ofproto->last_minute += min_ms;
8793 dpif_stats_update_hit_count(struct ofproto_dpif *ofproto, uint64_t delta)
8795 ofproto->n_hit += delta;
8798 const struct ofproto_class ofproto_dpif_class = {
8833 port_is_lacp_current,
8834 NULL, /* rule_choose_table */
8841 rule_modify_actions,
8855 get_stp_port_status,
8862 is_mirror_output_bundle,
8863 forward_bpdu_changed,
8864 set_mac_table_config,