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);
213 /* Initial values of fields of the packet that may be changed during
214 * flow processing and needed later. */
215 struct initial_vals {
216 /* This is the value of vlan_tci in the packet as actually received from
217 * dpif. This is the same as the facet's flow.vlan_tci unless the packet
218 * was received via a VLAN splinter. In that case, this value is 0
219 * (because the packet as actually received from the dpif had no 802.1Q
220 * tag) but the facet's flow.vlan_tci is set to the VLAN that the splinter
223 * This member should be removed when the VLAN splinters feature is no
229 tag_type tags; /* Tags associated with actions. */
230 enum slow_path_reason slow; /* 0 if fast path may be used. */
231 bool has_learn; /* Actions include NXAST_LEARN? */
232 bool has_normal; /* Actions output to OFPP_NORMAL? */
233 bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */
234 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
235 mirror_mask_t mirrors; /* Bitmap of associated mirrors. */
237 uint64_t odp_actions_stub[256 / 8];
238 struct ofpbuf odp_actions;
242 struct ofproto_dpif *ofproto;
244 /* Flow to which the OpenFlow actions apply. xlate_actions() will modify
245 * this flow when actions change header fields. */
248 struct initial_vals initial_vals;
250 /* The packet corresponding to 'flow', or a null pointer if we are
251 * revalidating without a packet to refer to. */
252 const struct ofpbuf *packet;
254 /* Should OFPP_NORMAL update the MAC learning table? Should "learn"
255 * actions update the flow table?
257 * We want to update these tables if we are actually processing a packet,
258 * or if we are accounting for packets that the datapath has processed, but
259 * not if we are just revalidating. */
262 /* The rule initiating translation or NULL. */
263 struct rule_dpif *rule;
265 /* The actions to translate. If 'rule' is not NULL, these may be NULL. */
266 const struct ofpact *ofpacts;
269 /* Union of the set of TCP flags seen so far in this flow. (Used only by
270 * NXAST_FIN_TIMEOUT. Set to zero to avoid updating updating rules'
274 /* If nonnull, flow translation calls this function just before executing a
275 * resubmit or OFPP_TABLE action. In addition, disables logging of traces
276 * when the recursion depth is exceeded.
278 * 'rule' is the rule being submitted into. It will be null if the
279 * resubmit or OFPP_TABLE action didn't find a matching rule.
281 * This is normally null so the client has to set it manually after
282 * calling xlate_in_init(). */
283 void (*resubmit_hook)(struct xlate_ctx *, struct rule_dpif *rule);
285 /* If nonnull, flow translation calls this function to report some
286 * significant decision, e.g. to explain why OFPP_NORMAL translation
287 * dropped a packet. */
288 void (*report_hook)(struct xlate_ctx *, const char *s);
290 /* If nonnull, flow translation credits the specified statistics to each
291 * rule reached through a resubmit or OFPP_TABLE action.
293 * This is normally null so the client has to set it manually after
294 * calling xlate_in_init(). */
295 const struct dpif_flow_stats *resubmit_stats;
298 /* Context used by xlate_actions() and its callees. */
300 struct xlate_in *xin;
301 struct xlate_out *xout;
303 struct ofproto_dpif *ofproto;
305 /* Flow at the last commit. */
306 struct flow base_flow;
308 /* Tunnel IP destination address as received. This is stored separately
309 * as the base_flow.tunnel is cleared on init to reflect the datapath
310 * behavior. Used to make sure not to send tunneled output to ourselves,
311 * which might lead to an infinite loop. This could happen easily
312 * if a tunnel is marked as 'ip_remote=flow', and the flow does not
313 * actually set the tun_dst field. */
314 ovs_be32 orig_tunnel_ip_dst;
316 /* Stack for the push and pop actions. Each stack element is of type
317 * "union mf_subvalue". */
318 union mf_subvalue init_stack[1024 / sizeof(union mf_subvalue)];
321 /* The rule that we are currently translating, or NULL. */
322 struct rule_dpif *rule;
324 int recurse; /* Recursion level, via xlate_table_action. */
325 bool max_resubmit_trigger; /* Recursed too deeply during translation. */
326 uint32_t orig_skb_priority; /* Priority when packet arrived. */
327 uint8_t table_id; /* OpenFlow table ID where flow was found. */
328 uint32_t sflow_n_outputs; /* Number of output ports. */
329 uint32_t sflow_odp_port; /* Output port for composing sFlow action. */
330 uint16_t user_cookie_offset;/* Used for user_action_cookie fixup. */
331 bool exit; /* No further actions should be processed. */
334 static void xlate_in_init(struct xlate_in *, struct ofproto_dpif *,
335 const struct flow *, const struct initial_vals *,
336 struct rule_dpif *, uint8_t tcp_flags,
337 const struct ofpbuf *);
339 static void xlate_out_uninit(struct xlate_out *);
341 static void xlate_actions(struct xlate_in *, struct xlate_out *);
343 static void xlate_actions_for_side_effects(struct xlate_in *);
345 static void xlate_table_action(struct xlate_ctx *, uint16_t in_port,
346 uint8_t table_id, bool may_packet_in);
348 static size_t put_userspace_action(const struct ofproto_dpif *,
349 struct ofpbuf *odp_actions,
351 const union user_action_cookie *,
354 static void compose_slow_path(const struct ofproto_dpif *, const struct flow *,
355 enum slow_path_reason,
356 uint64_t *stub, size_t stub_size,
357 const struct nlattr **actionsp,
358 size_t *actions_lenp);
360 static void xlate_report(struct xlate_ctx *ctx, const char *s);
362 /* A subfacet (see "struct subfacet" below) has three possible installation
365 * - SF_NOT_INSTALLED: Not installed in the datapath. This will only be the
366 * case just after the subfacet is created, just before the subfacet is
367 * destroyed, or if the datapath returns an error when we try to install a
370 * - SF_FAST_PATH: The subfacet's actions are installed in the datapath.
372 * - SF_SLOW_PATH: An action that sends every packet for the subfacet through
373 * ofproto_dpif is installed in the datapath.
376 SF_NOT_INSTALLED, /* No datapath flow for this subfacet. */
377 SF_FAST_PATH, /* Full actions are installed. */
378 SF_SLOW_PATH, /* Send-to-userspace action is installed. */
381 /* A dpif flow and actions associated with a facet.
383 * See also the large comment on struct facet. */
386 struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */
387 struct list list_node; /* In struct facet's 'facets' list. */
388 struct facet *facet; /* Owning facet. */
390 enum odp_key_fitness key_fitness;
394 long long int used; /* Time last used; time created if not used. */
395 long long int created; /* Time created. */
397 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
398 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
400 enum subfacet_path path; /* Installed in datapath? */
402 /* Datapath port the packet arrived on. This is needed to remove
403 * flows for ports that are no longer part of the bridge. Since the
404 * flow definition only has the OpenFlow port number and the port is
405 * no longer part of the bridge, we can't determine the datapath port
406 * number needed to delete the flow from the datapath. */
407 uint32_t odp_in_port;
410 #define SUBFACET_DESTROY_MAX_BATCH 50
412 static struct subfacet *subfacet_create(struct facet *, struct flow_miss *miss,
414 static struct subfacet *subfacet_find(struct ofproto_dpif *,
415 const struct nlattr *key, size_t key_len,
417 static void subfacet_destroy(struct subfacet *);
418 static void subfacet_destroy__(struct subfacet *);
419 static void subfacet_destroy_batch(struct ofproto_dpif *,
420 struct subfacet **, int n);
421 static void subfacet_reset_dp_stats(struct subfacet *,
422 struct dpif_flow_stats *);
423 static void subfacet_update_time(struct subfacet *, long long int used);
424 static void subfacet_update_stats(struct subfacet *,
425 const struct dpif_flow_stats *);
426 static int subfacet_install(struct subfacet *,
427 const struct ofpbuf *odp_actions,
428 struct dpif_flow_stats *);
429 static void subfacet_uninstall(struct subfacet *);
431 /* An exact-match instantiation of an OpenFlow flow.
433 * A facet associates a "struct flow", which represents the Open vSwitch
434 * userspace idea of an exact-match flow, with one or more subfacets. Each
435 * subfacet tracks the datapath's idea of the exact-match flow equivalent to
436 * the facet. When the kernel module (or other dpif implementation) and Open
437 * vSwitch userspace agree on the definition of a flow key, there is exactly
438 * one subfacet per facet. If the dpif implementation supports more-specific
439 * flow matching than userspace, however, a facet can have more than one
440 * subfacet, each of which corresponds to some distinction in flow that
441 * userspace simply doesn't understand.
443 * Flow expiration works in terms of subfacets, so a facet must have at least
444 * one subfacet or it will never expire, leaking memory. */
447 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
448 struct list list_node; /* In owning rule's 'facets' list. */
449 struct rule_dpif *rule; /* Owning rule. */
452 struct list subfacets;
453 long long int used; /* Time last used; time created if not used. */
460 * - Do include packets and bytes sent "by hand", e.g. with
463 * - Do include packets and bytes that were obtained from the datapath
464 * when a subfacet's statistics were reset (e.g. dpif_flow_put() with
465 * DPIF_FP_ZERO_STATS).
467 * - Do not include packets or bytes that can be obtained from the
468 * datapath for any existing subfacet.
470 uint64_t packet_count; /* Number of packets received. */
471 uint64_t byte_count; /* Number of bytes received. */
473 /* Resubmit statistics. */
474 uint64_t prev_packet_count; /* Number of packets from last stats push. */
475 uint64_t prev_byte_count; /* Number of bytes from last stats push. */
476 long long int prev_used; /* Used time from last stats push. */
479 uint64_t accounted_bytes; /* Bytes processed by facet_account(). */
480 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
481 uint8_t tcp_flags; /* TCP flags seen for this 'rule'. */
483 struct xlate_out xout;
485 /* Initial values of the packet that may be needed later. */
486 struct initial_vals initial_vals;
488 /* Storage for a single subfacet, to reduce malloc() time and space
489 * overhead. (A facet always has at least one subfacet and in the common
490 * case has exactly one subfacet. However, 'one_subfacet' may not
491 * always be valid, since it could have been removed after newer
492 * subfacets were pushed onto the 'subfacets' list.) */
493 struct subfacet one_subfacet;
495 long long int learn_rl; /* Rate limiter for facet_learn(). */
498 static struct facet *facet_create(const struct flow_miss *, uint32_t hash);
499 static void facet_remove(struct facet *);
500 static void facet_free(struct facet *);
502 static struct facet *facet_find(struct ofproto_dpif *,
503 const struct flow *, uint32_t hash);
504 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
505 const struct flow *, uint32_t hash);
506 static bool facet_revalidate(struct facet *);
507 static bool facet_check_consistency(struct facet *);
509 static void facet_flush_stats(struct facet *);
511 static void facet_update_time(struct facet *, long long int used);
512 static void facet_reset_counters(struct facet *);
513 static void facet_push_stats(struct facet *);
514 static void facet_learn(struct facet *);
515 static void facet_account(struct facet *);
516 static void push_all_stats(void);
518 static bool facet_is_controller_flow(struct facet *);
521 struct hmap_node odp_port_node; /* In dpif_backer's "odp_to_ofport_map". */
525 struct ofbundle *bundle; /* Bundle that contains this port, if any. */
526 struct list bundle_node; /* In struct ofbundle's "ports" list. */
527 struct cfm *cfm; /* Connectivity Fault Management, if any. */
528 struct bfd *bfd; /* BFD, if any. */
529 tag_type tag; /* Tag associated with this port. */
530 bool may_enable; /* May be enabled in bonds. */
531 long long int carrier_seq; /* Carrier status changes. */
532 struct tnl_port *tnl_port; /* Tunnel handle, or null. */
535 struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */
536 enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */
537 long long int stp_state_entered;
539 struct hmap priorities; /* Map of attached 'priority_to_dscp's. */
541 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
543 * This is deprecated. It is only for compatibility with broken device
544 * drivers in old versions of Linux that do not properly support VLANs when
545 * VLAN devices are not used. When broken device drivers are no longer in
546 * widespread use, we will delete these interfaces. */
547 uint16_t realdev_ofp_port;
551 /* Node in 'ofport_dpif''s 'priorities' map. Used to maintain a map from
552 * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
553 * traffic egressing the 'ofport' with that priority should be marked with. */
554 struct priority_to_dscp {
555 struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'priorities' map. */
556 uint32_t priority; /* Priority of this queue (see struct flow). */
558 uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */
561 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
563 * This is deprecated. It is only for compatibility with broken device drivers
564 * in old versions of Linux that do not properly support VLANs when VLAN
565 * devices are not used. When broken device drivers are no longer in
566 * widespread use, we will delete these interfaces. */
567 struct vlan_splinter {
568 struct hmap_node realdev_vid_node;
569 struct hmap_node vlandev_node;
570 uint16_t realdev_ofp_port;
571 uint16_t vlandev_ofp_port;
575 static uint32_t vsp_realdev_to_vlandev(const struct ofproto_dpif *,
576 uint32_t realdev, ovs_be16 vlan_tci);
577 static bool vsp_adjust_flow(const struct ofproto_dpif *, struct flow *);
578 static void vsp_remove(struct ofport_dpif *);
579 static void vsp_add(struct ofport_dpif *, uint16_t realdev_ofp_port, int vid);
581 static uint32_t ofp_port_to_odp_port(const struct ofproto_dpif *,
583 static uint16_t odp_port_to_ofp_port(const struct ofproto_dpif *,
586 static struct ofport_dpif *
587 ofport_dpif_cast(const struct ofport *ofport)
589 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
592 static void port_run(struct ofport_dpif *);
593 static void port_run_fast(struct ofport_dpif *);
594 static void port_wait(struct ofport_dpif *);
595 static int set_bfd(struct ofport *, const struct smap *);
596 static int set_cfm(struct ofport *, const struct cfm_settings *);
597 static void ofport_clear_priorities(struct ofport_dpif *);
598 static void run_fast_rl(void);
600 struct dpif_completion {
601 struct list list_node;
602 struct ofoperation *op;
605 /* Extra information about a classifier table.
606 * Currently used just for optimized flow revalidation. */
608 /* If either of these is nonnull, then this table has a form that allows
609 * flows to be tagged to avoid revalidating most flows for the most common
610 * kinds of flow table changes. */
611 struct cls_table *catchall_table; /* Table that wildcards all fields. */
612 struct cls_table *other_table; /* Table with any other wildcard set. */
613 uint32_t basis; /* Keeps each table's tags separate. */
616 /* Reasons that we might need to revalidate every facet, and corresponding
619 * A value of 0 means that there is no need to revalidate.
621 * It would be nice to have some cleaner way to integrate with coverage
622 * counters, but with only a few reasons I guess this is good enough for
624 enum revalidate_reason {
625 REV_RECONFIGURE = 1, /* Switch configuration changed. */
626 REV_STP, /* Spanning tree protocol port status change. */
627 REV_PORT_TOGGLED, /* Port enabled or disabled by CFM, LACP, ...*/
628 REV_FLOW_TABLE, /* Flow table changed. */
629 REV_INCONSISTENCY /* Facet self-check failed. */
631 COVERAGE_DEFINE(rev_reconfigure);
632 COVERAGE_DEFINE(rev_stp);
633 COVERAGE_DEFINE(rev_port_toggled);
634 COVERAGE_DEFINE(rev_flow_table);
635 COVERAGE_DEFINE(rev_inconsistency);
637 /* Drop keys are odp flow keys which have drop flows installed in the kernel.
638 * These are datapath flows which have no associated ofproto, if they did we
639 * would use facets. */
641 struct hmap_node hmap_node;
646 /* All datapaths of a given type share a single dpif backer instance. */
651 struct timer next_expiration;
652 struct hmap odp_to_ofport_map; /* ODP port to ofport mapping. */
654 struct simap tnl_backers; /* Set of dpif ports backing tunnels. */
656 /* Facet revalidation flags applying to facets which use this backer. */
657 enum revalidate_reason need_revalidate; /* Revalidate every facet. */
658 struct tag_set revalidate_set; /* Revalidate only matching facets. */
660 struct hmap drop_keys; /* Set of dropped odp keys. */
663 /* All existing ofproto_backer instances, indexed by ofproto->up.type. */
664 static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers);
666 static void drop_key_clear(struct dpif_backer *);
667 static struct ofport_dpif *
668 odp_port_to_ofport(const struct dpif_backer *, uint32_t odp_port);
670 static void dpif_stats_update_hit_count(struct ofproto_dpif *ofproto,
672 struct avg_subfacet_rates {
673 double add_rate; /* Moving average of new flows created per minute. */
674 double del_rate; /* Moving average of flows deleted per minute. */
676 static void show_dp_rates(struct ds *ds, const char *heading,
677 const struct avg_subfacet_rates *rates);
678 static void exp_mavg(double *avg, int base, double new);
680 struct ofproto_dpif {
681 struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
683 struct dpif_backer *backer;
685 /* Special OpenFlow rules. */
686 struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */
687 struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */
693 struct netflow *netflow;
694 struct dpif_sflow *sflow;
695 struct dpif_ipfix *ipfix;
696 struct hmap bundles; /* Contains "struct ofbundle"s. */
697 struct mac_learning *ml;
698 struct ofmirror *mirrors[MAX_MIRRORS];
700 bool has_bonded_bundles;
704 struct hmap subfacets;
705 struct governor *governor;
706 long long int consistency_rl;
709 struct table_dpif tables[N_TABLES];
711 /* Support for debugging async flow mods. */
712 struct list completions;
714 bool has_bundle_action; /* True when the first bundle action appears. */
715 struct netdev_stats stats; /* To account packets generated and consumed in
720 long long int stp_last_tick;
722 /* VLAN splinters. */
723 struct hmap realdev_vid_map; /* (realdev,vid) -> vlandev. */
724 struct hmap vlandev_map; /* vlandev -> (realdev,vid). */
727 struct sset ports; /* Set of standard port names. */
728 struct sset ghost_ports; /* Ports with no datapath port. */
729 struct sset port_poll_set; /* Queued names for port_poll() reply. */
730 int port_poll_errno; /* Last errno for port_poll() reply. */
732 /* Per ofproto's dpif stats. */
736 /* Subfacet statistics.
738 * These keep track of the total number of subfacets added and deleted and
739 * flow life span. They are useful for computing the flow rates stats
740 * exposed via "ovs-appctl dpif/show". The goal is to learn about
741 * traffic patterns in ways that we can use later to improve Open vSwitch
742 * performance in new situations. */
743 long long int created; /* Time when it is created. */
744 unsigned int max_n_subfacet; /* Maximum number of flows */
746 /* The average number of subfacets... */
747 struct avg_subfacet_rates hourly; /* ...over the last hour. */
748 struct avg_subfacet_rates daily; /* ...over the last day. */
749 long long int last_minute; /* Last time 'hourly' was updated. */
751 /* Number of subfacets added or deleted since 'last_minute'. */
752 unsigned int subfacet_add_count;
753 unsigned int subfacet_del_count;
755 /* Number of subfacets added or deleted from 'created' to 'last_minute.' */
756 unsigned long long int total_subfacet_add_count;
757 unsigned long long int total_subfacet_del_count;
759 /* Sum of the number of milliseconds that each subfacet existed,
760 * over the subfacets that have been added and then later deleted. */
761 unsigned long long int total_subfacet_life_span;
763 /* Incremented by the number of currently existing subfacets, each
764 * time we pull statistics from the kernel. */
765 unsigned long long int total_subfacet_count;
767 /* Number of times we pull statistics from the kernel. */
768 unsigned long long int n_update_stats;
770 static unsigned long long int avg_subfacet_life_span(
771 const struct ofproto_dpif *);
772 static double avg_subfacet_count(const struct ofproto_dpif *ofproto);
773 static void update_moving_averages(struct ofproto_dpif *ofproto);
774 static void dpif_stats_update_hit_count(struct ofproto_dpif *ofproto,
776 static void update_max_subfacet_count(struct ofproto_dpif *ofproto);
778 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
779 * for debugging the asynchronous flow_mod implementation.) */
782 /* All existing ofproto_dpif instances, indexed by ->up.name. */
783 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
785 static void ofproto_dpif_unixctl_init(void);
787 static struct ofproto_dpif *
788 ofproto_dpif_cast(const struct ofproto *ofproto)
790 ovs_assert(ofproto->ofproto_class == &ofproto_dpif_class);
791 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
794 static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *,
796 static struct ofport_dpif *get_odp_port(const struct ofproto_dpif *,
798 static void ofproto_trace(struct ofproto_dpif *, const struct flow *,
799 const struct ofpbuf *,
800 const struct initial_vals *, struct ds *);
802 /* Packet processing. */
803 static void update_learning_table(struct ofproto_dpif *,
804 const struct flow *, int vlan,
807 #define FLOW_MISS_MAX_BATCH 50
808 static int handle_upcalls(struct dpif_backer *, unsigned int max_batch);
810 /* Flow expiration. */
811 static int expire(struct dpif_backer *);
814 static void send_netflow_active_timeouts(struct ofproto_dpif *);
817 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
818 static size_t compose_sflow_action(const struct ofproto_dpif *,
819 struct ofpbuf *odp_actions,
820 const struct flow *, uint32_t odp_port);
821 static void compose_ipfix_action(const struct ofproto_dpif *,
822 struct ofpbuf *odp_actions,
823 const struct flow *);
824 static void add_mirror_actions(struct xlate_ctx *ctx,
825 const struct flow *flow);
826 /* Global variables. */
827 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
829 /* Initial mappings of port to bridge mappings. */
830 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
832 /* Factory functions. */
835 init(const struct shash *iface_hints)
837 struct shash_node *node;
839 /* Make a local copy, since we don't own 'iface_hints' elements. */
840 SHASH_FOR_EACH(node, iface_hints) {
841 const struct iface_hint *orig_hint = node->data;
842 struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
844 new_hint->br_name = xstrdup(orig_hint->br_name);
845 new_hint->br_type = xstrdup(orig_hint->br_type);
846 new_hint->ofp_port = orig_hint->ofp_port;
848 shash_add(&init_ofp_ports, node->name, new_hint);
853 enumerate_types(struct sset *types)
855 dp_enumerate_types(types);
859 enumerate_names(const char *type, struct sset *names)
861 struct ofproto_dpif *ofproto;
864 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
865 if (strcmp(type, ofproto->up.type)) {
868 sset_add(names, ofproto->up.name);
875 del(const char *type, const char *name)
880 error = dpif_open(name, type, &dpif);
882 error = dpif_delete(dpif);
889 port_open_type(const char *datapath_type, const char *port_type)
891 return dpif_port_open_type(datapath_type, port_type);
894 /* Type functions. */
896 static struct ofproto_dpif *
897 lookup_ofproto_dpif_by_port_name(const char *name)
899 struct ofproto_dpif *ofproto;
901 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
902 if (sset_contains(&ofproto->ports, name)) {
911 type_run(const char *type)
913 static long long int push_timer = LLONG_MIN;
914 struct dpif_backer *backer;
918 backer = shash_find_data(&all_dpif_backers, type);
920 /* This is not necessarily a problem, since backers are only
921 * created on demand. */
925 dpif_run(backer->dpif);
927 /* The most natural place to push facet statistics is when they're pulled
928 * from the datapath. However, when there are many flows in the datapath,
929 * this expensive operation can occur so frequently, that it reduces our
930 * ability to quickly set up flows. To reduce the cost, we push statistics
932 if (time_msec() > push_timer) {
933 push_timer = time_msec() + 2000;
937 if (backer->need_revalidate
938 || !tag_set_is_empty(&backer->revalidate_set)) {
939 struct tag_set revalidate_set = backer->revalidate_set;
940 bool need_revalidate = backer->need_revalidate;
941 struct ofproto_dpif *ofproto;
942 struct simap_node *node;
943 struct simap tmp_backers;
945 /* Handle tunnel garbage collection. */
946 simap_init(&tmp_backers);
947 simap_swap(&backer->tnl_backers, &tmp_backers);
949 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
950 struct ofport_dpif *iter;
952 if (backer != ofproto->backer) {
956 HMAP_FOR_EACH (iter, up.hmap_node, &ofproto->up.ports) {
959 if (!iter->tnl_port) {
963 dp_port = netdev_vport_get_dpif_port(iter->up.netdev);
964 node = simap_find(&tmp_backers, dp_port);
966 simap_put(&backer->tnl_backers, dp_port, node->data);
967 simap_delete(&tmp_backers, node);
968 node = simap_find(&backer->tnl_backers, dp_port);
970 node = simap_find(&backer->tnl_backers, dp_port);
972 uint32_t odp_port = UINT32_MAX;
974 if (!dpif_port_add(backer->dpif, iter->up.netdev,
976 simap_put(&backer->tnl_backers, dp_port, odp_port);
977 node = simap_find(&backer->tnl_backers, dp_port);
982 iter->odp_port = node ? node->data : OVSP_NONE;
983 if (tnl_port_reconfigure(&iter->up, iter->odp_port,
985 backer->need_revalidate = REV_RECONFIGURE;
990 SIMAP_FOR_EACH (node, &tmp_backers) {
991 dpif_port_del(backer->dpif, node->data);
993 simap_destroy(&tmp_backers);
995 switch (backer->need_revalidate) {
996 case REV_RECONFIGURE: COVERAGE_INC(rev_reconfigure); break;
997 case REV_STP: COVERAGE_INC(rev_stp); break;
998 case REV_PORT_TOGGLED: COVERAGE_INC(rev_port_toggled); break;
999 case REV_FLOW_TABLE: COVERAGE_INC(rev_flow_table); break;
1000 case REV_INCONSISTENCY: COVERAGE_INC(rev_inconsistency); break;
1003 if (backer->need_revalidate) {
1004 /* Clear the drop_keys in case we should now be accepting some
1005 * formerly dropped flows. */
1006 drop_key_clear(backer);
1009 /* Clear the revalidation flags. */
1010 tag_set_init(&backer->revalidate_set);
1011 backer->need_revalidate = 0;
1013 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1014 struct facet *facet, *next;
1016 if (ofproto->backer != backer) {
1020 HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &ofproto->facets) {
1022 || tag_set_intersects(&revalidate_set, facet->xout.tags)) {
1023 facet_revalidate(facet);
1030 if (timer_expired(&backer->next_expiration)) {
1031 int delay = expire(backer);
1032 timer_set_duration(&backer->next_expiration, delay);
1035 /* Check for port changes in the dpif. */
1036 while ((error = dpif_port_poll(backer->dpif, &devname)) == 0) {
1037 struct ofproto_dpif *ofproto;
1038 struct dpif_port port;
1040 /* Don't report on the datapath's device. */
1041 if (!strcmp(devname, dpif_base_name(backer->dpif))) {
1045 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
1046 &all_ofproto_dpifs) {
1047 if (simap_contains(&ofproto->backer->tnl_backers, devname)) {
1052 ofproto = lookup_ofproto_dpif_by_port_name(devname);
1053 if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
1054 /* The port was removed. If we know the datapath,
1055 * report it through poll_set(). If we don't, it may be
1056 * notifying us of a removal we initiated, so ignore it.
1057 * If there's a pending ENOBUFS, let it stand, since
1058 * everything will be reevaluated. */
1059 if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
1060 sset_add(&ofproto->port_poll_set, devname);
1061 ofproto->port_poll_errno = 0;
1063 } else if (!ofproto) {
1064 /* The port was added, but we don't know with which
1065 * ofproto we should associate it. Delete it. */
1066 dpif_port_del(backer->dpif, port.port_no);
1068 dpif_port_destroy(&port);
1074 if (error != EAGAIN) {
1075 struct ofproto_dpif *ofproto;
1077 /* There was some sort of error, so propagate it to all
1078 * ofprotos that use this backer. */
1079 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
1080 &all_ofproto_dpifs) {
1081 if (ofproto->backer == backer) {
1082 sset_clear(&ofproto->port_poll_set);
1083 ofproto->port_poll_errno = error;
1092 dpif_backer_run_fast(struct dpif_backer *backer, int max_batch)
1096 /* Handle one or more batches of upcalls, until there's nothing left to do
1097 * or until we do a fixed total amount of work.
1099 * We do work in batches because it can be much cheaper to set up a number
1100 * of flows and fire off their patches all at once. We do multiple batches
1101 * because in some cases handling a packet can cause another packet to be
1102 * queued almost immediately as part of the return flow. Both
1103 * optimizations can make major improvements on some benchmarks and
1104 * presumably for real traffic as well. */
1106 while (work < max_batch) {
1107 int retval = handle_upcalls(backer, max_batch - work);
1118 type_run_fast(const char *type)
1120 struct dpif_backer *backer;
1122 backer = shash_find_data(&all_dpif_backers, type);
1124 /* This is not necessarily a problem, since backers are only
1125 * created on demand. */
1129 return dpif_backer_run_fast(backer, FLOW_MISS_MAX_BATCH);
1135 static long long int port_rl = LLONG_MIN;
1136 static unsigned int backer_rl = 0;
1138 if (time_msec() >= port_rl) {
1139 struct ofproto_dpif *ofproto;
1140 struct ofport_dpif *ofport;
1142 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1144 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1145 port_run_fast(ofport);
1148 port_rl = time_msec() + 200;
1151 /* XXX: We have to be careful not to do too much work in this function. If
1152 * we call dpif_backer_run_fast() too often, or with too large a batch,
1153 * performance improves signifcantly, but at a cost. It's possible for the
1154 * number of flows in the datapath to increase without bound, and for poll
1155 * loops to take 10s of seconds. The correct solution to this problem,
1156 * long term, is to separate flow miss handling into it's own thread so it
1157 * isn't affected by revalidations, and expirations. Until then, this is
1158 * the best we can do. */
1159 if (++backer_rl >= 10) {
1160 struct shash_node *node;
1163 SHASH_FOR_EACH (node, &all_dpif_backers) {
1164 dpif_backer_run_fast(node->data, 1);
1170 type_wait(const char *type)
1172 struct dpif_backer *backer;
1174 backer = shash_find_data(&all_dpif_backers, type);
1176 /* This is not necessarily a problem, since backers are only
1177 * created on demand. */
1181 timer_wait(&backer->next_expiration);
1184 /* Basic life-cycle. */
1186 static int add_internal_flows(struct ofproto_dpif *);
1188 static struct ofproto *
1191 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
1192 return &ofproto->up;
1196 dealloc(struct ofproto *ofproto_)
1198 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1203 close_dpif_backer(struct dpif_backer *backer)
1205 struct shash_node *node;
1207 ovs_assert(backer->refcount > 0);
1209 if (--backer->refcount) {
1213 drop_key_clear(backer);
1214 hmap_destroy(&backer->drop_keys);
1216 simap_destroy(&backer->tnl_backers);
1217 hmap_destroy(&backer->odp_to_ofport_map);
1218 node = shash_find(&all_dpif_backers, backer->type);
1220 shash_delete(&all_dpif_backers, node);
1221 dpif_close(backer->dpif);
1226 /* Datapath port slated for removal from datapath. */
1227 struct odp_garbage {
1228 struct list list_node;
1233 open_dpif_backer(const char *type, struct dpif_backer **backerp)
1235 struct dpif_backer *backer;
1236 struct dpif_port_dump port_dump;
1237 struct dpif_port port;
1238 struct shash_node *node;
1239 struct list garbage_list;
1240 struct odp_garbage *garbage, *next;
1246 backer = shash_find_data(&all_dpif_backers, type);
1253 backer_name = xasprintf("ovs-%s", type);
1255 /* Remove any existing datapaths, since we assume we're the only
1256 * userspace controlling the datapath. */
1258 dp_enumerate_names(type, &names);
1259 SSET_FOR_EACH(name, &names) {
1260 struct dpif *old_dpif;
1262 /* Don't remove our backer if it exists. */
1263 if (!strcmp(name, backer_name)) {
1267 if (dpif_open(name, type, &old_dpif)) {
1268 VLOG_WARN("couldn't open old datapath %s to remove it", name);
1270 dpif_delete(old_dpif);
1271 dpif_close(old_dpif);
1274 sset_destroy(&names);
1276 backer = xmalloc(sizeof *backer);
1278 error = dpif_create_and_open(backer_name, type, &backer->dpif);
1281 VLOG_ERR("failed to open datapath of type %s: %s", type,
1287 backer->type = xstrdup(type);
1288 backer->refcount = 1;
1289 hmap_init(&backer->odp_to_ofport_map);
1290 hmap_init(&backer->drop_keys);
1291 timer_set_duration(&backer->next_expiration, 1000);
1292 backer->need_revalidate = 0;
1293 simap_init(&backer->tnl_backers);
1294 tag_set_init(&backer->revalidate_set);
1297 dpif_flow_flush(backer->dpif);
1299 /* Loop through the ports already on the datapath and remove any
1300 * that we don't need anymore. */
1301 list_init(&garbage_list);
1302 dpif_port_dump_start(&port_dump, backer->dpif);
1303 while (dpif_port_dump_next(&port_dump, &port)) {
1304 node = shash_find(&init_ofp_ports, port.name);
1305 if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
1306 garbage = xmalloc(sizeof *garbage);
1307 garbage->odp_port = port.port_no;
1308 list_push_front(&garbage_list, &garbage->list_node);
1311 dpif_port_dump_done(&port_dump);
1313 LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) {
1314 dpif_port_del(backer->dpif, garbage->odp_port);
1315 list_remove(&garbage->list_node);
1319 shash_add(&all_dpif_backers, type, backer);
1321 error = dpif_recv_set(backer->dpif, true);
1323 VLOG_ERR("failed to listen on datapath of type %s: %s",
1324 type, strerror(error));
1325 close_dpif_backer(backer);
1333 construct(struct ofproto *ofproto_)
1335 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1336 struct shash_node *node, *next;
1341 error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1346 max_ports = dpif_get_max_ports(ofproto->backer->dpif);
1347 ofproto_init_max_ports(ofproto_, MIN(max_ports, OFPP_MAX));
1349 ofproto->n_matches = 0;
1351 ofproto->netflow = NULL;
1352 ofproto->sflow = NULL;
1353 ofproto->ipfix = NULL;
1354 ofproto->stp = NULL;
1355 hmap_init(&ofproto->bundles);
1356 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1357 for (i = 0; i < MAX_MIRRORS; i++) {
1358 ofproto->mirrors[i] = NULL;
1360 ofproto->has_bonded_bundles = false;
1362 hmap_init(&ofproto->facets);
1363 hmap_init(&ofproto->subfacets);
1364 ofproto->governor = NULL;
1365 ofproto->consistency_rl = LLONG_MIN;
1367 for (i = 0; i < N_TABLES; i++) {
1368 struct table_dpif *table = &ofproto->tables[i];
1370 table->catchall_table = NULL;
1371 table->other_table = NULL;
1372 table->basis = random_uint32();
1375 list_init(&ofproto->completions);
1377 ofproto_dpif_unixctl_init();
1379 ofproto->has_mirrors = false;
1380 ofproto->has_bundle_action = false;
1382 hmap_init(&ofproto->vlandev_map);
1383 hmap_init(&ofproto->realdev_vid_map);
1385 sset_init(&ofproto->ports);
1386 sset_init(&ofproto->ghost_ports);
1387 sset_init(&ofproto->port_poll_set);
1388 ofproto->port_poll_errno = 0;
1390 SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1391 struct iface_hint *iface_hint = node->data;
1393 if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1394 /* Check if the datapath already has this port. */
1395 if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1396 sset_add(&ofproto->ports, node->name);
1399 free(iface_hint->br_name);
1400 free(iface_hint->br_type);
1402 shash_delete(&init_ofp_ports, node);
1406 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1407 hash_string(ofproto->up.name, 0));
1408 memset(&ofproto->stats, 0, sizeof ofproto->stats);
1410 ofproto_init_tables(ofproto_, N_TABLES);
1411 error = add_internal_flows(ofproto);
1412 ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1415 ofproto->n_missed = 0;
1417 ofproto->max_n_subfacet = 0;
1418 ofproto->created = time_msec();
1419 ofproto->last_minute = ofproto->created;
1420 memset(&ofproto->hourly, 0, sizeof ofproto->hourly);
1421 memset(&ofproto->daily, 0, sizeof ofproto->daily);
1422 ofproto->subfacet_add_count = 0;
1423 ofproto->subfacet_del_count = 0;
1424 ofproto->total_subfacet_add_count = 0;
1425 ofproto->total_subfacet_del_count = 0;
1426 ofproto->total_subfacet_life_span = 0;
1427 ofproto->total_subfacet_count = 0;
1428 ofproto->n_update_stats = 0;
1434 add_internal_flow(struct ofproto_dpif *ofproto, int id,
1435 const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1437 struct ofputil_flow_mod fm;
1440 match_init_catchall(&fm.match);
1442 match_set_reg(&fm.match, 0, id);
1443 fm.new_cookie = htonll(0);
1444 fm.cookie = htonll(0);
1445 fm.cookie_mask = htonll(0);
1446 fm.table_id = TBL_INTERNAL;
1447 fm.command = OFPFC_ADD;
1448 fm.idle_timeout = 0;
1449 fm.hard_timeout = 0;
1453 fm.ofpacts = ofpacts->data;
1454 fm.ofpacts_len = ofpacts->size;
1456 error = ofproto_flow_mod(&ofproto->up, &fm);
1458 VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
1459 id, ofperr_to_string(error));
1463 *rulep = rule_dpif_lookup__(ofproto, &fm.match.flow, TBL_INTERNAL);
1464 ovs_assert(*rulep != NULL);
1470 add_internal_flows(struct ofproto_dpif *ofproto)
1472 struct ofpact_controller *controller;
1473 uint64_t ofpacts_stub[128 / 8];
1474 struct ofpbuf ofpacts;
1478 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1481 controller = ofpact_put_CONTROLLER(&ofpacts);
1482 controller->max_len = UINT16_MAX;
1483 controller->controller_id = 0;
1484 controller->reason = OFPR_NO_MATCH;
1485 ofpact_pad(&ofpacts);
1487 error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule);
1492 ofpbuf_clear(&ofpacts);
1493 error = add_internal_flow(ofproto, id++, &ofpacts,
1494 &ofproto->no_packet_in_rule);
1499 complete_operations(struct ofproto_dpif *ofproto)
1501 struct dpif_completion *c, *next;
1503 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
1504 ofoperation_complete(c->op, 0);
1505 list_remove(&c->list_node);
1511 destruct(struct ofproto *ofproto_)
1513 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1514 struct rule_dpif *rule, *next_rule;
1515 struct oftable *table;
1518 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1519 complete_operations(ofproto);
1521 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1522 struct cls_cursor cursor;
1524 cls_cursor_init(&cursor, &table->cls, NULL);
1525 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1526 ofproto_rule_destroy(&rule->up);
1530 for (i = 0; i < MAX_MIRRORS; i++) {
1531 mirror_destroy(ofproto->mirrors[i]);
1534 netflow_destroy(ofproto->netflow);
1535 dpif_sflow_destroy(ofproto->sflow);
1536 hmap_destroy(&ofproto->bundles);
1537 mac_learning_destroy(ofproto->ml);
1539 hmap_destroy(&ofproto->facets);
1540 hmap_destroy(&ofproto->subfacets);
1541 governor_destroy(ofproto->governor);
1543 hmap_destroy(&ofproto->vlandev_map);
1544 hmap_destroy(&ofproto->realdev_vid_map);
1546 sset_destroy(&ofproto->ports);
1547 sset_destroy(&ofproto->ghost_ports);
1548 sset_destroy(&ofproto->port_poll_set);
1550 close_dpif_backer(ofproto->backer);
1554 run_fast(struct ofproto *ofproto_)
1556 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1557 struct ofport_dpif *ofport;
1559 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1560 port_run_fast(ofport);
1567 run(struct ofproto *ofproto_)
1569 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1570 struct ofport_dpif *ofport;
1571 struct ofbundle *bundle;
1575 complete_operations(ofproto);
1578 error = run_fast(ofproto_);
1583 if (ofproto->netflow) {
1584 if (netflow_run(ofproto->netflow)) {
1585 send_netflow_active_timeouts(ofproto);
1588 if (ofproto->sflow) {
1589 dpif_sflow_run(ofproto->sflow);
1592 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1595 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1600 mac_learning_run(ofproto->ml, &ofproto->backer->revalidate_set);
1602 /* Check the consistency of a random facet, to aid debugging. */
1603 if (time_msec() >= ofproto->consistency_rl
1604 && !hmap_is_empty(&ofproto->facets)
1605 && !ofproto->backer->need_revalidate) {
1606 struct facet *facet;
1608 ofproto->consistency_rl = time_msec() + 250;
1610 facet = CONTAINER_OF(hmap_random_node(&ofproto->facets),
1611 struct facet, hmap_node);
1612 if (!tag_set_intersects(&ofproto->backer->revalidate_set,
1613 facet->xout.tags)) {
1614 if (!facet_check_consistency(facet)) {
1615 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
1620 if (ofproto->governor) {
1623 governor_run(ofproto->governor);
1625 /* If the governor has shrunk to its minimum size and the number of
1626 * subfacets has dwindled, then drop the governor entirely.
1628 * For hysteresis, the number of subfacets to drop the governor is
1629 * smaller than the number needed to trigger its creation. */
1630 n_subfacets = hmap_count(&ofproto->subfacets);
1631 if (n_subfacets * 4 < ofproto->up.flow_eviction_threshold
1632 && governor_is_idle(ofproto->governor)) {
1633 governor_destroy(ofproto->governor);
1634 ofproto->governor = NULL;
1642 wait(struct ofproto *ofproto_)
1644 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1645 struct ofport_dpif *ofport;
1646 struct ofbundle *bundle;
1648 if (!clogged && !list_is_empty(&ofproto->completions)) {
1649 poll_immediate_wake();
1652 dpif_wait(ofproto->backer->dpif);
1653 dpif_recv_wait(ofproto->backer->dpif);
1654 if (ofproto->sflow) {
1655 dpif_sflow_wait(ofproto->sflow);
1657 if (!tag_set_is_empty(&ofproto->backer->revalidate_set)) {
1658 poll_immediate_wake();
1660 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1663 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1664 bundle_wait(bundle);
1666 if (ofproto->netflow) {
1667 netflow_wait(ofproto->netflow);
1669 mac_learning_wait(ofproto->ml);
1671 if (ofproto->backer->need_revalidate) {
1672 /* Shouldn't happen, but if it does just go around again. */
1673 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1674 poll_immediate_wake();
1676 if (ofproto->governor) {
1677 governor_wait(ofproto->governor);
1682 get_memory_usage(const struct ofproto *ofproto_, struct simap *usage)
1684 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1686 simap_increase(usage, "facets", hmap_count(&ofproto->facets));
1687 simap_increase(usage, "subfacets", hmap_count(&ofproto->subfacets));
1691 flush(struct ofproto *ofproto_)
1693 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1694 struct subfacet *subfacet, *next_subfacet;
1695 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
1699 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
1700 &ofproto->subfacets) {
1701 if (subfacet->path != SF_NOT_INSTALLED) {
1702 batch[n_batch++] = subfacet;
1703 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
1704 subfacet_destroy_batch(ofproto, batch, n_batch);
1708 subfacet_destroy(subfacet);
1713 subfacet_destroy_batch(ofproto, batch, n_batch);
1718 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1719 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1721 *arp_match_ip = true;
1722 *actions = (OFPUTIL_A_OUTPUT |
1723 OFPUTIL_A_SET_VLAN_VID |
1724 OFPUTIL_A_SET_VLAN_PCP |
1725 OFPUTIL_A_STRIP_VLAN |
1726 OFPUTIL_A_SET_DL_SRC |
1727 OFPUTIL_A_SET_DL_DST |
1728 OFPUTIL_A_SET_NW_SRC |
1729 OFPUTIL_A_SET_NW_DST |
1730 OFPUTIL_A_SET_NW_TOS |
1731 OFPUTIL_A_SET_TP_SRC |
1732 OFPUTIL_A_SET_TP_DST |
1737 get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots)
1739 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1740 struct dpif_dp_stats s;
1742 strcpy(ots->name, "classifier");
1744 dpif_get_dp_stats(ofproto->backer->dpif, &s);
1746 ots->lookup_count = htonll(s.n_hit + s.n_missed);
1747 ots->matched_count = htonll(s.n_hit + ofproto->n_matches);
1750 static struct ofport *
1753 struct ofport_dpif *port = xmalloc(sizeof *port);
1758 port_dealloc(struct ofport *port_)
1760 struct ofport_dpif *port = ofport_dpif_cast(port_);
1765 port_construct(struct ofport *port_)
1767 struct ofport_dpif *port = ofport_dpif_cast(port_);
1768 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1769 const struct netdev *netdev = port->up.netdev;
1770 struct dpif_port dpif_port;
1773 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1774 port->bundle = NULL;
1777 port->tag = tag_create_random();
1778 port->may_enable = true;
1779 port->stp_port = NULL;
1780 port->stp_state = STP_DISABLED;
1781 port->tnl_port = NULL;
1782 hmap_init(&port->priorities);
1783 port->realdev_ofp_port = 0;
1784 port->vlandev_vid = 0;
1785 port->carrier_seq = netdev_get_carrier_resets(netdev);
1787 if (netdev_vport_is_patch(netdev)) {
1788 /* By bailing out here, we don't submit the port to the sFlow module
1789 * to be considered for counter polling export. This is correct
1790 * because the patch port represents an interface that sFlow considers
1791 * to be "internal" to the switch as a whole, and therefore not an
1792 * candidate for counter polling. */
1793 port->odp_port = OVSP_NONE;
1797 error = dpif_port_query_by_name(ofproto->backer->dpif,
1798 netdev_vport_get_dpif_port(netdev),
1804 port->odp_port = dpif_port.port_no;
1806 if (netdev_get_tunnel_config(netdev)) {
1807 port->tnl_port = tnl_port_add(&port->up, port->odp_port);
1809 /* Sanity-check that a mapping doesn't already exist. This
1810 * shouldn't happen for non-tunnel ports. */
1811 if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1812 VLOG_ERR("port %s already has an OpenFlow port number",
1814 dpif_port_destroy(&dpif_port);
1818 hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1819 hash_int(port->odp_port, 0));
1821 dpif_port_destroy(&dpif_port);
1823 if (ofproto->sflow) {
1824 dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1831 port_destruct(struct ofport *port_)
1833 struct ofport_dpif *port = ofport_dpif_cast(port_);
1834 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1835 const char *dp_port_name = netdev_vport_get_dpif_port(port->up.netdev);
1836 const char *devname = netdev_get_name(port->up.netdev);
1838 if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1839 /* The underlying device is still there, so delete it. This
1840 * happens when the ofproto is being destroyed, since the caller
1841 * assumes that removal of attached ports will happen as part of
1843 if (!port->tnl_port) {
1844 dpif_port_del(ofproto->backer->dpif, port->odp_port);
1846 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1849 if (port->odp_port != OVSP_NONE && !port->tnl_port) {
1850 hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1853 tnl_port_del(port->tnl_port);
1854 sset_find_and_delete(&ofproto->ports, devname);
1855 sset_find_and_delete(&ofproto->ghost_ports, devname);
1856 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1857 bundle_remove(port_);
1858 set_cfm(port_, NULL);
1859 set_bfd(port_, NULL);
1860 if (ofproto->sflow) {
1861 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1864 ofport_clear_priorities(port);
1865 hmap_destroy(&port->priorities);
1869 port_modified(struct ofport *port_)
1871 struct ofport_dpif *port = ofport_dpif_cast(port_);
1873 if (port->bundle && port->bundle->bond) {
1874 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1879 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1881 struct ofport_dpif *port = ofport_dpif_cast(port_);
1882 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1883 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1885 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1886 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1887 OFPUTIL_PC_NO_PACKET_IN)) {
1888 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1890 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1891 bundle_update(port->bundle);
1897 set_sflow(struct ofproto *ofproto_,
1898 const struct ofproto_sflow_options *sflow_options)
1900 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1901 struct dpif_sflow *ds = ofproto->sflow;
1903 if (sflow_options) {
1905 struct ofport_dpif *ofport;
1907 ds = ofproto->sflow = dpif_sflow_create();
1908 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1909 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1911 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1913 dpif_sflow_set_options(ds, sflow_options);
1916 dpif_sflow_destroy(ds);
1917 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1918 ofproto->sflow = NULL;
1926 struct ofproto *ofproto_,
1927 const struct ofproto_ipfix_bridge_exporter_options *bridge_exporter_options,
1928 const struct ofproto_ipfix_flow_exporter_options *flow_exporters_options,
1929 size_t n_flow_exporters_options)
1931 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1932 struct dpif_ipfix *di = ofproto->ipfix;
1934 if (bridge_exporter_options || flow_exporters_options) {
1936 di = ofproto->ipfix = dpif_ipfix_create();
1938 dpif_ipfix_set_options(
1939 di, bridge_exporter_options, flow_exporters_options,
1940 n_flow_exporters_options);
1943 dpif_ipfix_destroy(di);
1944 ofproto->ipfix = NULL;
1951 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1953 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1960 struct ofproto_dpif *ofproto;
1962 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1963 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1964 ofport->cfm = cfm_create(ofport->up.netdev);
1967 if (cfm_configure(ofport->cfm, s)) {
1973 cfm_destroy(ofport->cfm);
1979 get_cfm_status(const struct ofport *ofport_,
1980 struct ofproto_cfm_status *status)
1982 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1985 status->faults = cfm_get_fault(ofport->cfm);
1986 status->remote_opstate = cfm_get_opup(ofport->cfm);
1987 status->health = cfm_get_health(ofport->cfm);
1988 cfm_get_remote_mpids(ofport->cfm, &status->rmps, &status->n_rmps);
1996 set_bfd(struct ofport *ofport_, const struct smap *cfg)
1998 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
1999 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2003 ofport->bfd = bfd_configure(old, netdev_get_name(ofport->up.netdev), cfg);
2004 if (ofport->bfd != old) {
2005 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2012 get_bfd_status(struct ofport *ofport_, struct smap *smap)
2014 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2017 bfd_get_status(ofport->bfd, smap);
2024 /* Spanning Tree. */
2027 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
2029 struct ofproto_dpif *ofproto = ofproto_;
2030 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
2031 struct ofport_dpif *ofport;
2033 ofport = stp_port_get_aux(sp);
2035 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
2036 ofproto->up.name, port_num);
2038 struct eth_header *eth = pkt->l2;
2040 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
2041 if (eth_addr_is_zero(eth->eth_src)) {
2042 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
2043 "with unknown MAC", ofproto->up.name, port_num);
2045 send_packet(ofport, pkt);
2051 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
2053 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
2055 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2057 /* Only revalidate flows if the configuration changed. */
2058 if (!s != !ofproto->stp) {
2059 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2063 if (!ofproto->stp) {
2064 ofproto->stp = stp_create(ofproto_->name, s->system_id,
2065 send_bpdu_cb, ofproto);
2066 ofproto->stp_last_tick = time_msec();
2069 stp_set_bridge_id(ofproto->stp, s->system_id);
2070 stp_set_bridge_priority(ofproto->stp, s->priority);
2071 stp_set_hello_time(ofproto->stp, s->hello_time);
2072 stp_set_max_age(ofproto->stp, s->max_age);
2073 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
2075 struct ofport *ofport;
2077 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
2078 set_stp_port(ofport, NULL);
2081 stp_destroy(ofproto->stp);
2082 ofproto->stp = NULL;
2089 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
2091 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2095 s->bridge_id = stp_get_bridge_id(ofproto->stp);
2096 s->designated_root = stp_get_designated_root(ofproto->stp);
2097 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
2106 update_stp_port_state(struct ofport_dpif *ofport)
2108 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2109 enum stp_state state;
2111 /* Figure out new state. */
2112 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
2116 if (ofport->stp_state != state) {
2117 enum ofputil_port_state of_state;
2120 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
2121 netdev_get_name(ofport->up.netdev),
2122 stp_state_name(ofport->stp_state),
2123 stp_state_name(state));
2124 if (stp_learn_in_state(ofport->stp_state)
2125 != stp_learn_in_state(state)) {
2126 /* xxx Learning action flows should also be flushed. */
2127 mac_learning_flush(ofproto->ml,
2128 &ofproto->backer->revalidate_set);
2130 fwd_change = stp_forward_in_state(ofport->stp_state)
2131 != stp_forward_in_state(state);
2133 ofproto->backer->need_revalidate = REV_STP;
2134 ofport->stp_state = state;
2135 ofport->stp_state_entered = time_msec();
2137 if (fwd_change && ofport->bundle) {
2138 bundle_update(ofport->bundle);
2141 /* Update the STP state bits in the OpenFlow port description. */
2142 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
2143 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
2144 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
2145 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
2146 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
2148 ofproto_port_set_state(&ofport->up, of_state);
2152 /* Configures STP on 'ofport_' using the settings defined in 's'. The
2153 * caller is responsible for assigning STP port numbers and ensuring
2154 * there are no duplicates. */
2156 set_stp_port(struct ofport *ofport_,
2157 const struct ofproto_port_stp_settings *s)
2159 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2160 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2161 struct stp_port *sp = ofport->stp_port;
2163 if (!s || !s->enable) {
2165 ofport->stp_port = NULL;
2166 stp_port_disable(sp);
2167 update_stp_port_state(ofport);
2170 } else if (sp && stp_port_no(sp) != s->port_num
2171 && ofport == stp_port_get_aux(sp)) {
2172 /* The port-id changed, so disable the old one if it's not
2173 * already in use by another port. */
2174 stp_port_disable(sp);
2177 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
2178 stp_port_enable(sp);
2180 stp_port_set_aux(sp, ofport);
2181 stp_port_set_priority(sp, s->priority);
2182 stp_port_set_path_cost(sp, s->path_cost);
2184 update_stp_port_state(ofport);
2190 get_stp_port_status(struct ofport *ofport_,
2191 struct ofproto_port_stp_status *s)
2193 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2194 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2195 struct stp_port *sp = ofport->stp_port;
2197 if (!ofproto->stp || !sp) {
2203 s->port_id = stp_port_get_id(sp);
2204 s->state = stp_port_get_state(sp);
2205 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
2206 s->role = stp_port_get_role(sp);
2207 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
2213 stp_run(struct ofproto_dpif *ofproto)
2216 long long int now = time_msec();
2217 long long int elapsed = now - ofproto->stp_last_tick;
2218 struct stp_port *sp;
2221 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
2222 ofproto->stp_last_tick = now;
2224 while (stp_get_changed_port(ofproto->stp, &sp)) {
2225 struct ofport_dpif *ofport = stp_port_get_aux(sp);
2228 update_stp_port_state(ofport);
2232 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
2233 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
2239 stp_wait(struct ofproto_dpif *ofproto)
2242 poll_timer_wait(1000);
2246 /* Returns true if STP should process 'flow'. */
2248 stp_should_process_flow(const struct flow *flow)
2250 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
2254 stp_process_packet(const struct ofport_dpif *ofport,
2255 const struct ofpbuf *packet)
2257 struct ofpbuf payload = *packet;
2258 struct eth_header *eth = payload.data;
2259 struct stp_port *sp = ofport->stp_port;
2261 /* Sink packets on ports that have STP disabled when the bridge has
2263 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
2267 /* Trim off padding on payload. */
2268 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
2269 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
2272 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
2273 stp_received_bpdu(sp, payload.data, payload.size);
2277 static struct priority_to_dscp *
2278 get_priority(const struct ofport_dpif *ofport, uint32_t priority)
2280 struct priority_to_dscp *pdscp;
2283 hash = hash_int(priority, 0);
2284 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) {
2285 if (pdscp->priority == priority) {
2293 ofport_clear_priorities(struct ofport_dpif *ofport)
2295 struct priority_to_dscp *pdscp, *next;
2297 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) {
2298 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
2304 set_queues(struct ofport *ofport_,
2305 const struct ofproto_port_queue *qdscp_list,
2308 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2309 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2310 struct hmap new = HMAP_INITIALIZER(&new);
2313 for (i = 0; i < n_qdscp; i++) {
2314 struct priority_to_dscp *pdscp;
2318 dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
2319 if (dpif_queue_to_priority(ofproto->backer->dpif, qdscp_list[i].queue,
2324 pdscp = get_priority(ofport, priority);
2326 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
2328 pdscp = xmalloc(sizeof *pdscp);
2329 pdscp->priority = priority;
2331 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2334 if (pdscp->dscp != dscp) {
2336 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2339 hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0));
2342 if (!hmap_is_empty(&ofport->priorities)) {
2343 ofport_clear_priorities(ofport);
2344 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2347 hmap_swap(&new, &ofport->priorities);
2355 /* Expires all MAC learning entries associated with 'bundle' and forces its
2356 * ofproto to revalidate every flow.
2358 * Normally MAC learning entries are removed only from the ofproto associated
2359 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2360 * are removed from every ofproto. When patch ports and SLB bonds are in use
2361 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2362 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2363 * with the host from which it migrated. */
2365 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2367 struct ofproto_dpif *ofproto = bundle->ofproto;
2368 struct mac_learning *ml = ofproto->ml;
2369 struct mac_entry *mac, *next_mac;
2371 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2372 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2373 if (mac->port.p == bundle) {
2375 struct ofproto_dpif *o;
2377 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2379 struct mac_entry *e;
2381 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan,
2384 mac_learning_expire(o->ml, e);
2390 mac_learning_expire(ml, mac);
2395 static struct ofbundle *
2396 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2398 struct ofbundle *bundle;
2400 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2401 &ofproto->bundles) {
2402 if (bundle->aux == aux) {
2409 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
2410 * ones that are found to 'bundles'. */
2412 bundle_lookup_multiple(struct ofproto_dpif *ofproto,
2413 void **auxes, size_t n_auxes,
2414 struct hmapx *bundles)
2418 hmapx_init(bundles);
2419 for (i = 0; i < n_auxes; i++) {
2420 struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
2422 hmapx_add(bundles, bundle);
2428 bundle_update(struct ofbundle *bundle)
2430 struct ofport_dpif *port;
2432 bundle->floodable = true;
2433 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2434 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2435 || !stp_forward_in_state(port->stp_state)) {
2436 bundle->floodable = false;
2443 bundle_del_port(struct ofport_dpif *port)
2445 struct ofbundle *bundle = port->bundle;
2447 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2449 list_remove(&port->bundle_node);
2450 port->bundle = NULL;
2453 lacp_slave_unregister(bundle->lacp, port);
2456 bond_slave_unregister(bundle->bond, port);
2459 bundle_update(bundle);
2463 bundle_add_port(struct ofbundle *bundle, uint16_t ofp_port,
2464 struct lacp_slave_settings *lacp)
2466 struct ofport_dpif *port;
2468 port = get_ofp_port(bundle->ofproto, ofp_port);
2473 if (port->bundle != bundle) {
2474 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2476 bundle_del_port(port);
2479 port->bundle = bundle;
2480 list_push_back(&bundle->ports, &port->bundle_node);
2481 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2482 || !stp_forward_in_state(port->stp_state)) {
2483 bundle->floodable = false;
2487 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2488 lacp_slave_register(bundle->lacp, port, lacp);
2495 bundle_destroy(struct ofbundle *bundle)
2497 struct ofproto_dpif *ofproto;
2498 struct ofport_dpif *port, *next_port;
2505 ofproto = bundle->ofproto;
2506 for (i = 0; i < MAX_MIRRORS; i++) {
2507 struct ofmirror *m = ofproto->mirrors[i];
2509 if (m->out == bundle) {
2511 } else if (hmapx_find_and_delete(&m->srcs, bundle)
2512 || hmapx_find_and_delete(&m->dsts, bundle)) {
2513 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2518 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2519 bundle_del_port(port);
2522 bundle_flush_macs(bundle, true);
2523 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2525 free(bundle->trunks);
2526 lacp_destroy(bundle->lacp);
2527 bond_destroy(bundle->bond);
2532 bundle_set(struct ofproto *ofproto_, void *aux,
2533 const struct ofproto_bundle_settings *s)
2535 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2536 bool need_flush = false;
2537 struct ofport_dpif *port;
2538 struct ofbundle *bundle;
2539 unsigned long *trunks;
2545 bundle_destroy(bundle_lookup(ofproto, aux));
2549 ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2550 ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2552 bundle = bundle_lookup(ofproto, aux);
2554 bundle = xmalloc(sizeof *bundle);
2556 bundle->ofproto = ofproto;
2557 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2558 hash_pointer(aux, 0));
2560 bundle->name = NULL;
2562 list_init(&bundle->ports);
2563 bundle->vlan_mode = PORT_VLAN_TRUNK;
2565 bundle->trunks = NULL;
2566 bundle->use_priority_tags = s->use_priority_tags;
2567 bundle->lacp = NULL;
2568 bundle->bond = NULL;
2570 bundle->floodable = true;
2572 bundle->src_mirrors = 0;
2573 bundle->dst_mirrors = 0;
2574 bundle->mirror_out = 0;
2577 if (!bundle->name || strcmp(s->name, bundle->name)) {
2579 bundle->name = xstrdup(s->name);
2584 if (!bundle->lacp) {
2585 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2586 bundle->lacp = lacp_create();
2588 lacp_configure(bundle->lacp, s->lacp);
2590 lacp_destroy(bundle->lacp);
2591 bundle->lacp = NULL;
2594 /* Update set of ports. */
2596 for (i = 0; i < s->n_slaves; i++) {
2597 if (!bundle_add_port(bundle, s->slaves[i],
2598 s->lacp ? &s->lacp_slaves[i] : NULL)) {
2602 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2603 struct ofport_dpif *next_port;
2605 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2606 for (i = 0; i < s->n_slaves; i++) {
2607 if (s->slaves[i] == port->up.ofp_port) {
2612 bundle_del_port(port);
2616 ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2618 if (list_is_empty(&bundle->ports)) {
2619 bundle_destroy(bundle);
2623 /* Set VLAN tagging mode */
2624 if (s->vlan_mode != bundle->vlan_mode
2625 || s->use_priority_tags != bundle->use_priority_tags) {
2626 bundle->vlan_mode = s->vlan_mode;
2627 bundle->use_priority_tags = s->use_priority_tags;
2632 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2633 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2635 if (vlan != bundle->vlan) {
2636 bundle->vlan = vlan;
2640 /* Get trunked VLANs. */
2641 switch (s->vlan_mode) {
2642 case PORT_VLAN_ACCESS:
2646 case PORT_VLAN_TRUNK:
2647 trunks = CONST_CAST(unsigned long *, s->trunks);
2650 case PORT_VLAN_NATIVE_UNTAGGED:
2651 case PORT_VLAN_NATIVE_TAGGED:
2652 if (vlan != 0 && (!s->trunks
2653 || !bitmap_is_set(s->trunks, vlan)
2654 || bitmap_is_set(s->trunks, 0))) {
2655 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2657 trunks = bitmap_clone(s->trunks, 4096);
2659 trunks = bitmap_allocate1(4096);
2661 bitmap_set1(trunks, vlan);
2662 bitmap_set0(trunks, 0);
2664 trunks = CONST_CAST(unsigned long *, s->trunks);
2671 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2672 free(bundle->trunks);
2673 if (trunks == s->trunks) {
2674 bundle->trunks = vlan_bitmap_clone(trunks);
2676 bundle->trunks = trunks;
2681 if (trunks != s->trunks) {
2686 if (!list_is_short(&bundle->ports)) {
2687 bundle->ofproto->has_bonded_bundles = true;
2689 if (bond_reconfigure(bundle->bond, s->bond)) {
2690 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2693 bundle->bond = bond_create(s->bond);
2694 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2697 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2698 bond_slave_register(bundle->bond, port, port->up.netdev);
2701 bond_destroy(bundle->bond);
2702 bundle->bond = NULL;
2705 /* If we changed something that would affect MAC learning, un-learn
2706 * everything on this port and force flow revalidation. */
2708 bundle_flush_macs(bundle, false);
2715 bundle_remove(struct ofport *port_)
2717 struct ofport_dpif *port = ofport_dpif_cast(port_);
2718 struct ofbundle *bundle = port->bundle;
2721 bundle_del_port(port);
2722 if (list_is_empty(&bundle->ports)) {
2723 bundle_destroy(bundle);
2724 } else if (list_is_short(&bundle->ports)) {
2725 bond_destroy(bundle->bond);
2726 bundle->bond = NULL;
2732 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2734 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2735 struct ofport_dpif *port = port_;
2736 uint8_t ea[ETH_ADDR_LEN];
2739 error = netdev_get_etheraddr(port->up.netdev, ea);
2741 struct ofpbuf packet;
2744 ofpbuf_init(&packet, 0);
2745 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2747 memcpy(packet_pdu, pdu, pdu_size);
2749 send_packet(port, &packet);
2750 ofpbuf_uninit(&packet);
2752 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2753 "%s (%s)", port->bundle->name,
2754 netdev_get_name(port->up.netdev), strerror(error));
2759 bundle_send_learning_packets(struct ofbundle *bundle)
2761 struct ofproto_dpif *ofproto = bundle->ofproto;
2762 int error, n_packets, n_errors;
2763 struct mac_entry *e;
2765 error = n_packets = n_errors = 0;
2766 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2767 if (e->port.p != bundle) {
2768 struct ofpbuf *learning_packet;
2769 struct ofport_dpif *port;
2773 /* The assignment to "port" is unnecessary but makes "grep"ing for
2774 * struct ofport_dpif more effective. */
2775 learning_packet = bond_compose_learning_packet(bundle->bond,
2779 ret = send_packet(port, learning_packet);
2780 ofpbuf_delete(learning_packet);
2790 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2791 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2792 "packets, last error was: %s",
2793 bundle->name, n_errors, n_packets, strerror(error));
2795 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2796 bundle->name, n_packets);
2801 bundle_run(struct ofbundle *bundle)
2804 lacp_run(bundle->lacp, send_pdu_cb);
2807 struct ofport_dpif *port;
2809 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2810 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2813 bond_run(bundle->bond, &bundle->ofproto->backer->revalidate_set,
2814 lacp_status(bundle->lacp));
2815 if (bond_should_send_learning_packets(bundle->bond)) {
2816 bundle_send_learning_packets(bundle);
2822 bundle_wait(struct ofbundle *bundle)
2825 lacp_wait(bundle->lacp);
2828 bond_wait(bundle->bond);
2835 mirror_scan(struct ofproto_dpif *ofproto)
2839 for (idx = 0; idx < MAX_MIRRORS; idx++) {
2840 if (!ofproto->mirrors[idx]) {
2847 static struct ofmirror *
2848 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
2852 for (i = 0; i < MAX_MIRRORS; i++) {
2853 struct ofmirror *mirror = ofproto->mirrors[i];
2854 if (mirror && mirror->aux == aux) {
2862 /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */
2864 mirror_update_dups(struct ofproto_dpif *ofproto)
2868 for (i = 0; i < MAX_MIRRORS; i++) {
2869 struct ofmirror *m = ofproto->mirrors[i];
2872 m->dup_mirrors = MIRROR_MASK_C(1) << i;
2876 for (i = 0; i < MAX_MIRRORS; i++) {
2877 struct ofmirror *m1 = ofproto->mirrors[i];
2884 for (j = i + 1; j < MAX_MIRRORS; j++) {
2885 struct ofmirror *m2 = ofproto->mirrors[j];
2887 if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) {
2888 m1->dup_mirrors |= MIRROR_MASK_C(1) << j;
2889 m2->dup_mirrors |= m1->dup_mirrors;
2896 mirror_set(struct ofproto *ofproto_, void *aux,
2897 const struct ofproto_mirror_settings *s)
2899 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2900 mirror_mask_t mirror_bit;
2901 struct ofbundle *bundle;
2902 struct ofmirror *mirror;
2903 struct ofbundle *out;
2904 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
2905 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
2908 mirror = mirror_lookup(ofproto, aux);
2910 mirror_destroy(mirror);
2916 idx = mirror_scan(ofproto);
2918 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
2920 ofproto->up.name, MAX_MIRRORS, s->name);
2924 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
2925 mirror->ofproto = ofproto;
2928 mirror->out_vlan = -1;
2929 mirror->name = NULL;
2932 if (!mirror->name || strcmp(s->name, mirror->name)) {
2934 mirror->name = xstrdup(s->name);
2937 /* Get the new configuration. */
2938 if (s->out_bundle) {
2939 out = bundle_lookup(ofproto, s->out_bundle);
2941 mirror_destroy(mirror);
2947 out_vlan = s->out_vlan;
2949 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
2950 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
2952 /* If the configuration has not changed, do nothing. */
2953 if (hmapx_equals(&srcs, &mirror->srcs)
2954 && hmapx_equals(&dsts, &mirror->dsts)
2955 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
2956 && mirror->out == out
2957 && mirror->out_vlan == out_vlan)
2959 hmapx_destroy(&srcs);
2960 hmapx_destroy(&dsts);
2964 hmapx_swap(&srcs, &mirror->srcs);
2965 hmapx_destroy(&srcs);
2967 hmapx_swap(&dsts, &mirror->dsts);
2968 hmapx_destroy(&dsts);
2970 free(mirror->vlans);
2971 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
2974 mirror->out_vlan = out_vlan;
2976 /* Update bundles. */
2977 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2978 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
2979 if (hmapx_contains(&mirror->srcs, bundle)) {
2980 bundle->src_mirrors |= mirror_bit;
2982 bundle->src_mirrors &= ~mirror_bit;
2985 if (hmapx_contains(&mirror->dsts, bundle)) {
2986 bundle->dst_mirrors |= mirror_bit;
2988 bundle->dst_mirrors &= ~mirror_bit;
2991 if (mirror->out == bundle) {
2992 bundle->mirror_out |= mirror_bit;
2994 bundle->mirror_out &= ~mirror_bit;
2998 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2999 ofproto->has_mirrors = true;
3000 mac_learning_flush(ofproto->ml,
3001 &ofproto->backer->revalidate_set);
3002 mirror_update_dups(ofproto);
3008 mirror_destroy(struct ofmirror *mirror)
3010 struct ofproto_dpif *ofproto;
3011 mirror_mask_t mirror_bit;
3012 struct ofbundle *bundle;
3019 ofproto = mirror->ofproto;
3020 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3021 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
3023 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
3024 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
3025 bundle->src_mirrors &= ~mirror_bit;
3026 bundle->dst_mirrors &= ~mirror_bit;
3027 bundle->mirror_out &= ~mirror_bit;
3030 hmapx_destroy(&mirror->srcs);
3031 hmapx_destroy(&mirror->dsts);
3032 free(mirror->vlans);
3034 ofproto->mirrors[mirror->idx] = NULL;
3038 mirror_update_dups(ofproto);
3040 ofproto->has_mirrors = false;
3041 for (i = 0; i < MAX_MIRRORS; i++) {
3042 if (ofproto->mirrors[i]) {
3043 ofproto->has_mirrors = true;
3050 mirror_get_stats(struct ofproto *ofproto_, void *aux,
3051 uint64_t *packets, uint64_t *bytes)
3053 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3054 struct ofmirror *mirror = mirror_lookup(ofproto, aux);
3057 *packets = *bytes = UINT64_MAX;
3063 *packets = mirror->packet_count;
3064 *bytes = mirror->byte_count;
3070 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
3072 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3073 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
3074 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
3080 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
3082 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3083 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
3084 return bundle && bundle->mirror_out != 0;
3088 forward_bpdu_changed(struct ofproto *ofproto_)
3090 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3091 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3095 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
3098 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3099 mac_learning_set_idle_time(ofproto->ml, idle_time);
3100 mac_learning_set_max_entries(ofproto->ml, max_entries);
3105 static struct ofport_dpif *
3106 get_ofp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
3108 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
3109 return ofport ? ofport_dpif_cast(ofport) : NULL;
3112 static struct ofport_dpif *
3113 get_odp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
3115 struct ofport_dpif *port = odp_port_to_ofport(ofproto->backer, odp_port);
3116 return port && &ofproto->up == port->up.ofproto ? port : NULL;
3120 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
3121 struct ofproto_port *ofproto_port,
3122 struct dpif_port *dpif_port)
3124 ofproto_port->name = dpif_port->name;
3125 ofproto_port->type = dpif_port->type;
3126 ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
3129 static struct ofport_dpif *
3130 ofport_get_peer(const struct ofport_dpif *ofport_dpif)
3132 const struct ofproto_dpif *ofproto;
3135 peer = netdev_vport_patch_peer(ofport_dpif->up.netdev);
3140 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
3141 struct ofport *ofport;
3143 ofport = shash_find_data(&ofproto->up.port_by_name, peer);
3144 if (ofport && ofport->ofproto->ofproto_class == &ofproto_dpif_class) {
3145 return ofport_dpif_cast(ofport);
3152 port_run_fast(struct ofport_dpif *ofport)
3154 if (ofport->cfm && cfm_should_send_ccm(ofport->cfm)) {
3155 struct ofpbuf packet;
3157 ofpbuf_init(&packet, 0);
3158 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
3159 send_packet(ofport, &packet);
3160 ofpbuf_uninit(&packet);
3163 if (ofport->bfd && bfd_should_send_packet(ofport->bfd)) {
3164 struct ofpbuf packet;
3166 ofpbuf_init(&packet, 0);
3167 bfd_put_packet(ofport->bfd, &packet, ofport->up.pp.hw_addr);
3168 send_packet(ofport, &packet);
3169 ofpbuf_uninit(&packet);
3174 port_run(struct ofport_dpif *ofport)
3176 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
3177 bool carrier_changed = carrier_seq != ofport->carrier_seq;
3178 bool enable = netdev_get_carrier(ofport->up.netdev);
3180 ofport->carrier_seq = carrier_seq;
3182 port_run_fast(ofport);
3184 if (ofport->tnl_port
3185 && tnl_port_reconfigure(&ofport->up, ofport->odp_port,
3186 &ofport->tnl_port)) {
3187 ofproto_dpif_cast(ofport->up.ofproto)->backer->need_revalidate = true;
3191 int cfm_opup = cfm_get_opup(ofport->cfm);
3193 cfm_run(ofport->cfm);
3194 enable = enable && !cfm_get_fault(ofport->cfm);
3196 if (cfm_opup >= 0) {
3197 enable = enable && cfm_opup;
3202 bfd_run(ofport->bfd);
3203 enable = enable && bfd_forwarding(ofport->bfd);
3206 if (ofport->bundle) {
3207 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
3208 if (carrier_changed) {
3209 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
3213 if (ofport->may_enable != enable) {
3214 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3216 if (ofproto->has_bundle_action) {
3217 ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
3221 ofport->may_enable = enable;
3225 port_wait(struct ofport_dpif *ofport)
3228 cfm_wait(ofport->cfm);
3232 bfd_wait(ofport->bfd);
3237 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
3238 struct ofproto_port *ofproto_port)
3240 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3241 struct dpif_port dpif_port;
3244 if (sset_contains(&ofproto->ghost_ports, devname)) {
3245 const char *type = netdev_get_type_from_name(devname);
3247 /* We may be called before ofproto->up.port_by_name is populated with
3248 * the appropriate ofport. For this reason, we must get the name and
3249 * type from the netdev layer directly. */
3251 const struct ofport *ofport;
3253 ofport = shash_find_data(&ofproto->up.port_by_name, devname);
3254 ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
3255 ofproto_port->name = xstrdup(devname);
3256 ofproto_port->type = xstrdup(type);
3262 if (!sset_contains(&ofproto->ports, devname)) {
3265 error = dpif_port_query_by_name(ofproto->backer->dpif,
3266 devname, &dpif_port);
3268 ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
3274 port_add(struct ofproto *ofproto_, struct netdev *netdev)
3276 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3277 const char *dp_port_name = netdev_vport_get_dpif_port(netdev);
3278 const char *devname = netdev_get_name(netdev);
3280 if (netdev_vport_is_patch(netdev)) {
3281 sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
3285 if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
3286 uint32_t port_no = UINT32_MAX;
3289 error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
3293 if (netdev_get_tunnel_config(netdev)) {
3294 simap_put(&ofproto->backer->tnl_backers, dp_port_name, port_no);
3298 if (netdev_get_tunnel_config(netdev)) {
3299 sset_add(&ofproto->ghost_ports, devname);
3301 sset_add(&ofproto->ports, devname);
3307 port_del(struct ofproto *ofproto_, uint16_t ofp_port)
3309 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3310 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
3317 sset_find_and_delete(&ofproto->ghost_ports,
3318 netdev_get_name(ofport->up.netdev));
3319 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3320 if (!ofport->tnl_port) {
3321 error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
3323 /* The caller is going to close ofport->up.netdev. If this is a
3324 * bonded port, then the bond is using that netdev, so remove it
3325 * from the bond. The client will need to reconfigure everything
3326 * after deleting ports, so then the slave will get re-added. */
3327 bundle_remove(&ofport->up);
3334 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
3336 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3341 error = netdev_get_stats(ofport->up.netdev, stats);
3343 if (!error && ofport_->ofp_port == OFPP_LOCAL) {
3344 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3346 /* ofproto->stats.tx_packets represents packets that we created
3347 * internally and sent to some port (e.g. packets sent with
3348 * send_packet()). Account for them as if they had come from
3349 * OFPP_LOCAL and got forwarded. */
3351 if (stats->rx_packets != UINT64_MAX) {
3352 stats->rx_packets += ofproto->stats.tx_packets;
3355 if (stats->rx_bytes != UINT64_MAX) {
3356 stats->rx_bytes += ofproto->stats.tx_bytes;
3359 /* ofproto->stats.rx_packets represents packets that were received on
3360 * some port and we processed internally and dropped (e.g. STP).
3361 * Account for them as if they had been forwarded to OFPP_LOCAL. */
3363 if (stats->tx_packets != UINT64_MAX) {
3364 stats->tx_packets += ofproto->stats.rx_packets;
3367 if (stats->tx_bytes != UINT64_MAX) {
3368 stats->tx_bytes += ofproto->stats.rx_bytes;
3375 struct port_dump_state {
3380 struct ofproto_port port;
3385 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
3387 *statep = xzalloc(sizeof(struct port_dump_state));
3392 port_dump_next(const struct ofproto *ofproto_, void *state_,
3393 struct ofproto_port *port)
3395 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3396 struct port_dump_state *state = state_;
3397 const struct sset *sset;
3398 struct sset_node *node;
3400 if (state->has_port) {
3401 ofproto_port_destroy(&state->port);
3402 state->has_port = false;
3404 sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
3405 while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
3408 error = port_query_by_name(ofproto_, node->name, &state->port);
3410 *port = state->port;
3411 state->has_port = true;
3413 } else if (error != ENODEV) {
3418 if (!state->ghost) {
3419 state->ghost = true;
3422 return port_dump_next(ofproto_, state_, port);
3429 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3431 struct port_dump_state *state = state_;
3433 if (state->has_port) {
3434 ofproto_port_destroy(&state->port);
3441 port_poll(const struct ofproto *ofproto_, char **devnamep)
3443 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3445 if (ofproto->port_poll_errno) {
3446 int error = ofproto->port_poll_errno;
3447 ofproto->port_poll_errno = 0;
3451 if (sset_is_empty(&ofproto->port_poll_set)) {
3455 *devnamep = sset_pop(&ofproto->port_poll_set);
3460 port_poll_wait(const struct ofproto *ofproto_)
3462 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3463 dpif_port_poll_wait(ofproto->backer->dpif);
3467 port_is_lacp_current(const struct ofport *ofport_)
3469 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3470 return (ofport->bundle && ofport->bundle->lacp
3471 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3475 /* Upcall handling. */
3477 /* Flow miss batching.
3479 * Some dpifs implement operations faster when you hand them off in a batch.
3480 * To allow batching, "struct flow_miss" queues the dpif-related work needed
3481 * for a given flow. Each "struct flow_miss" corresponds to sending one or
3482 * more packets, plus possibly installing the flow in the dpif.
3484 * So far we only batch the operations that affect flow setup time the most.
3485 * It's possible to batch more than that, but the benefit might be minimal. */
3487 struct hmap_node hmap_node;
3488 struct ofproto_dpif *ofproto;
3490 enum odp_key_fitness key_fitness;
3491 const struct nlattr *key;
3493 struct initial_vals initial_vals;
3494 struct list packets;
3495 enum dpif_upcall_type upcall_type;
3496 uint32_t odp_in_port;
3499 struct flow_miss_op {
3500 struct dpif_op dpif_op;
3502 uint64_t slow_stub[128 / 8]; /* Buffer for compose_slow_path() */
3503 struct xlate_out xout;
3504 bool xout_garbage; /* 'xout' needs to be uninitialized? */
3507 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
3508 * OpenFlow controller as necessary according to their individual
3509 * configurations. */
3511 send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet,
3512 const struct flow *flow)
3514 struct ofputil_packet_in pin;
3516 pin.packet = packet->data;
3517 pin.packet_len = packet->size;
3518 pin.reason = OFPR_NO_MATCH;
3519 pin.controller_id = 0;
3524 pin.send_len = 0; /* not used for flow table misses */
3526 flow_get_metadata(flow, &pin.fmd);
3528 connmgr_send_packet_in(ofproto->up.connmgr, &pin);
3531 static enum slow_path_reason
3532 process_special(struct ofproto_dpif *ofproto, const struct flow *flow,
3533 const struct ofport_dpif *ofport, const struct ofpbuf *packet)
3537 } else if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) {
3539 cfm_process_heartbeat(ofport->cfm, packet);
3542 } else if (ofport->bfd && bfd_should_process_flow(flow)) {
3544 bfd_process_packet(ofport->bfd, flow, packet);
3547 } else if (ofport->bundle && ofport->bundle->lacp
3548 && flow->dl_type == htons(ETH_TYPE_LACP)) {
3550 lacp_process_packet(ofport->bundle->lacp, ofport, packet);
3553 } else if (ofproto->stp && stp_should_process_flow(flow)) {
3555 stp_process_packet(ofport, packet);
3563 static struct flow_miss *
3564 flow_miss_find(struct hmap *todo, const struct ofproto_dpif *ofproto,
3565 const struct flow *flow, uint32_t hash)
3567 struct flow_miss *miss;
3569 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
3570 if (miss->ofproto == ofproto && flow_equal(&miss->flow, flow)) {
3578 /* Partially Initializes 'op' as an "execute" operation for 'miss' and
3579 * 'packet'. The caller must initialize op->actions and op->actions_len. If
3580 * 'miss' is associated with a subfacet the caller must also initialize the
3581 * returned op->subfacet, and if anything needs to be freed after processing
3582 * the op, the caller must initialize op->garbage also. */
3584 init_flow_miss_execute_op(struct flow_miss *miss, struct ofpbuf *packet,
3585 struct flow_miss_op *op)
3587 if (miss->flow.vlan_tci != miss->initial_vals.vlan_tci) {
3588 /* This packet was received on a VLAN splinter port. We
3589 * added a VLAN to the packet to make the packet resemble
3590 * the flow, but the actions were composed assuming that
3591 * the packet contained no VLAN. So, we must remove the
3592 * VLAN header from the packet before trying to execute the
3594 eth_pop_vlan(packet);
3597 op->xout_garbage = false;
3598 op->dpif_op.type = DPIF_OP_EXECUTE;
3599 op->dpif_op.u.execute.key = miss->key;
3600 op->dpif_op.u.execute.key_len = miss->key_len;
3601 op->dpif_op.u.execute.packet = packet;
3604 /* Helper for handle_flow_miss_without_facet() and
3605 * handle_flow_miss_with_facet(). */
3607 handle_flow_miss_common(struct rule_dpif *rule,
3608 struct ofpbuf *packet, const struct flow *flow)
3610 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3612 ofproto->n_matches++;
3614 if (rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
3616 * Extra-special case for fail-open mode.
3618 * We are in fail-open mode and the packet matched the fail-open
3619 * rule, but we are connected to a controller too. We should send
3620 * the packet up to the controller in the hope that it will try to
3621 * set up a flow and thereby allow us to exit fail-open.
3623 * See the top-level comment in fail-open.c for more information.
3625 send_packet_in_miss(ofproto, packet, flow);
3629 /* Figures out whether a flow that missed in 'ofproto', whose details are in
3630 * 'miss', is likely to be worth tracking in detail in userspace and (usually)
3631 * installing a datapath flow. The answer is usually "yes" (a return value of
3632 * true). However, for short flows the cost of bookkeeping is much higher than
3633 * the benefits, so when the datapath holds a large number of flows we impose
3634 * some heuristics to decide which flows are likely to be worth tracking. */
3636 flow_miss_should_make_facet(struct ofproto_dpif *ofproto,
3637 struct flow_miss *miss, uint32_t hash)
3639 if (!ofproto->governor) {
3642 n_subfacets = hmap_count(&ofproto->subfacets);
3643 if (n_subfacets * 2 <= ofproto->up.flow_eviction_threshold) {
3647 ofproto->governor = governor_create(ofproto->up.name);
3650 return governor_should_install_flow(ofproto->governor, hash,
3651 list_size(&miss->packets));
3654 /* Handles 'miss', which matches 'rule', without creating a facet or subfacet
3655 * or creating any datapath flow. May add an "execute" operation to 'ops' and
3656 * increment '*n_ops'. */
3658 handle_flow_miss_without_facet(struct flow_miss *miss,
3659 struct flow_miss_op *ops, size_t *n_ops)
3661 struct rule_dpif *rule = rule_dpif_lookup(miss->ofproto, &miss->flow);
3662 long long int now = time_msec();
3663 struct ofpbuf *packet;
3664 struct xlate_in xin;
3666 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3667 struct flow_miss_op *op = &ops[*n_ops];
3668 struct dpif_flow_stats stats;
3670 COVERAGE_INC(facet_suppress);
3672 handle_flow_miss_common(rule, packet, &miss->flow);
3674 dpif_flow_stats_extract(&miss->flow, packet, now, &stats);
3675 rule_credit_stats(rule, &stats);
3677 xlate_in_init(&xin, miss->ofproto, &miss->flow, &miss->initial_vals,
3678 rule, stats.tcp_flags, packet);
3679 xin.resubmit_stats = &stats;
3680 xlate_actions(&xin, &op->xout);
3682 if (op->xout.odp_actions.size) {
3683 struct dpif_execute *execute = &op->dpif_op.u.execute;
3685 init_flow_miss_execute_op(miss, packet, op);
3686 execute->actions = op->xout.odp_actions.data;
3687 execute->actions_len = op->xout.odp_actions.size;
3688 op->xout_garbage = true;
3692 xlate_out_uninit(&op->xout);
3697 /* Handles 'miss', which matches 'facet'. May add any required datapath
3698 * operations to 'ops', incrementing '*n_ops' for each new op.
3700 * All of the packets in 'miss' are considered to have arrived at time 'now'.
3701 * This is really important only for new facets: if we just called time_msec()
3702 * here, then the new subfacet or its packets could look (occasionally) as
3703 * though it was used some time after the facet was used. That can make a
3704 * one-packet flow look like it has a nonzero duration, which looks odd in
3705 * e.g. NetFlow statistics. */
3707 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
3709 struct flow_miss_op *ops, size_t *n_ops)
3711 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3712 enum subfacet_path want_path;
3713 struct subfacet *subfacet;
3714 struct ofpbuf *packet;
3716 subfacet = subfacet_create(facet, miss, now);
3717 want_path = subfacet->facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
3719 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3720 struct flow_miss_op *op = &ops[*n_ops];
3721 struct dpif_flow_stats stats;
3723 handle_flow_miss_common(facet->rule, packet, &miss->flow);
3725 if (want_path != SF_FAST_PATH) {
3726 struct xlate_in xin;
3728 xlate_in_init(&xin, ofproto, &facet->flow, &facet->initial_vals,
3729 facet->rule, 0, packet);
3730 xlate_actions_for_side_effects(&xin);
3733 dpif_flow_stats_extract(&facet->flow, packet, now, &stats);
3734 subfacet_update_stats(subfacet, &stats);
3736 if (facet->xout.odp_actions.size) {
3737 struct dpif_execute *execute = &op->dpif_op.u.execute;
3739 init_flow_miss_execute_op(miss, packet, op);
3740 execute->actions = facet->xout.odp_actions.data,
3741 execute->actions_len = facet->xout.odp_actions.size;
3746 if (miss->upcall_type == DPIF_UC_MISS || subfacet->path != want_path) {
3747 struct flow_miss_op *op = &ops[(*n_ops)++];
3748 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
3750 subfacet->path = want_path;
3752 op->xout_garbage = false;
3753 op->dpif_op.type = DPIF_OP_FLOW_PUT;
3754 put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
3755 put->key = miss->key;
3756 put->key_len = miss->key_len;
3757 if (want_path == SF_FAST_PATH) {
3758 put->actions = facet->xout.odp_actions.data;
3759 put->actions_len = facet->xout.odp_actions.size;
3761 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
3762 op->slow_stub, sizeof op->slow_stub,
3763 &put->actions, &put->actions_len);
3769 /* Handles flow miss 'miss'. May add any required datapath operations
3770 * to 'ops', incrementing '*n_ops' for each new op. */
3772 handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops,
3775 struct ofproto_dpif *ofproto = miss->ofproto;
3776 struct facet *facet;
3780 /* The caller must ensure that miss->hmap_node.hash contains
3781 * flow_hash(miss->flow, 0). */
3782 hash = miss->hmap_node.hash;
3784 facet = facet_lookup_valid(ofproto, &miss->flow, hash);
3786 /* There does not exist a bijection between 'struct flow' and datapath
3787 * flow keys with fitness ODP_FIT_TO_LITTLE. This breaks a fundamental
3788 * assumption used throughout the facet and subfacet handling code.
3789 * Since we have to handle these misses in userspace anyway, we simply
3790 * skip facet creation, avoiding the problem alltogether. */
3791 if (miss->key_fitness == ODP_FIT_TOO_LITTLE
3792 || !flow_miss_should_make_facet(ofproto, miss, hash)) {
3793 handle_flow_miss_without_facet(miss, ops, n_ops);
3797 facet = facet_create(miss, hash);
3802 handle_flow_miss_with_facet(miss, facet, now, ops, n_ops);
3805 static struct drop_key *
3806 drop_key_lookup(const struct dpif_backer *backer, const struct nlattr *key,
3809 struct drop_key *drop_key;
3811 HMAP_FOR_EACH_WITH_HASH (drop_key, hmap_node, hash_bytes(key, key_len, 0),
3812 &backer->drop_keys) {
3813 if (drop_key->key_len == key_len
3814 && !memcmp(drop_key->key, key, key_len)) {
3822 drop_key_clear(struct dpif_backer *backer)
3824 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3825 struct drop_key *drop_key, *next;
3827 HMAP_FOR_EACH_SAFE (drop_key, next, hmap_node, &backer->drop_keys) {
3830 error = dpif_flow_del(backer->dpif, drop_key->key, drop_key->key_len,
3832 if (error && !VLOG_DROP_WARN(&rl)) {
3833 struct ds ds = DS_EMPTY_INITIALIZER;
3834 odp_flow_key_format(drop_key->key, drop_key->key_len, &ds);
3835 VLOG_WARN("Failed to delete drop key (%s) (%s)", strerror(error),
3840 hmap_remove(&backer->drop_keys, &drop_key->hmap_node);
3841 free(drop_key->key);
3846 /* Given a datpath, packet, and flow metadata ('backer', 'packet', and 'key'
3847 * respectively), populates 'flow' with the result of odp_flow_key_to_flow().
3848 * Optionally, if nonnull, populates 'fitnessp' with the fitness of 'flow' as
3849 * returned by odp_flow_key_to_flow(). Also, optionally populates 'ofproto'
3850 * with the ofproto_dpif, and 'odp_in_port' with the datapath in_port, that
3851 * 'packet' ingressed.
3853 * If 'ofproto' is nonnull, requires 'flow''s in_port to exist. Otherwise sets
3854 * 'flow''s in_port to OFPP_NONE.
3856 * This function does post-processing on data returned from
3857 * odp_flow_key_to_flow() to help make VLAN splinters transparent to the rest
3858 * of the upcall processing logic. In particular, if the extracted in_port is
3859 * a VLAN splinter port, it replaces flow->in_port by the "real" port, sets
3860 * flow->vlan_tci correctly for the VLAN of the VLAN splinter port, and pushes
3861 * a VLAN header onto 'packet' (if it is nonnull).
3863 * Optionally, if 'initial_vals' is nonnull, sets 'initial_vals->vlan_tci'
3864 * to the VLAN TCI with which the packet was really received, that is, the
3865 * actual VLAN TCI extracted by odp_flow_key_to_flow(). (This differs from
3866 * the value returned in flow->vlan_tci only for packets received on
3869 * Similarly, this function also includes some logic to help with tunnels. It
3870 * may modify 'flow' as necessary to make the tunneling implementation
3871 * transparent to the upcall processing logic.
3873 * Returns 0 if successful, ENODEV if the parsed flow has no associated ofport,
3874 * or some other positive errno if there are other problems. */
3876 ofproto_receive(const struct dpif_backer *backer, struct ofpbuf *packet,
3877 const struct nlattr *key, size_t key_len,
3878 struct flow *flow, enum odp_key_fitness *fitnessp,
3879 struct ofproto_dpif **ofproto, uint32_t *odp_in_port,
3880 struct initial_vals *initial_vals)
3882 const struct ofport_dpif *port;
3883 enum odp_key_fitness fitness;
3886 fitness = odp_flow_key_to_flow(key, key_len, flow);
3887 if (fitness == ODP_FIT_ERROR) {
3893 initial_vals->vlan_tci = flow->vlan_tci;
3897 *odp_in_port = flow->in_port;
3900 port = (tnl_port_should_receive(flow)
3901 ? ofport_dpif_cast(tnl_port_receive(flow))
3902 : odp_port_to_ofport(backer, flow->in_port));
3903 flow->in_port = port ? port->up.ofp_port : OFPP_NONE;
3908 /* XXX: Since the tunnel module is not scoped per backer, for a tunnel port
3909 * it's theoretically possible that we'll receive an ofport belonging to an
3910 * entirely different datapath. In practice, this can't happen because no
3911 * platforms has two separate datapaths which each support tunneling. */
3912 ovs_assert(ofproto_dpif_cast(port->up.ofproto)->backer == backer);
3914 if (vsp_adjust_flow(ofproto_dpif_cast(port->up.ofproto), flow)) {
3916 /* Make the packet resemble the flow, so that it gets sent to
3917 * an OpenFlow controller properly, so that it looks correct
3918 * for sFlow, and so that flow_extract() will get the correct
3919 * vlan_tci if it is called on 'packet'.
3921 * The allocated space inside 'packet' probably also contains
3922 * 'key', that is, both 'packet' and 'key' are probably part of
3923 * a struct dpif_upcall (see the large comment on that
3924 * structure definition), so pushing data on 'packet' is in
3925 * general not a good idea since it could overwrite 'key' or
3926 * free it as a side effect. However, it's OK in this special
3927 * case because we know that 'packet' is inside a Netlink
3928 * attribute: pushing 4 bytes will just overwrite the 4-byte
3929 * "struct nlattr", which is fine since we don't need that
3930 * header anymore. */
3931 eth_push_vlan(packet, flow->vlan_tci);
3933 /* We can't reproduce 'key' from 'flow'. */
3934 fitness = fitness == ODP_FIT_PERFECT ? ODP_FIT_TOO_MUCH : fitness;
3939 *ofproto = ofproto_dpif_cast(port->up.ofproto);
3944 *fitnessp = fitness;
3950 handle_miss_upcalls(struct dpif_backer *backer, struct dpif_upcall *upcalls,
3953 struct dpif_upcall *upcall;
3954 struct flow_miss *miss;
3955 struct flow_miss misses[FLOW_MISS_MAX_BATCH];
3956 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
3957 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
3967 /* Construct the to-do list.
3969 * This just amounts to extracting the flow from each packet and sticking
3970 * the packets that have the same flow in the same "flow_miss" structure so
3971 * that we can process them together. */
3974 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
3975 struct flow_miss *miss = &misses[n_misses];
3976 struct flow_miss *existing_miss;
3977 struct ofproto_dpif *ofproto;
3978 uint32_t odp_in_port;
3983 error = ofproto_receive(backer, upcall->packet, upcall->key,
3984 upcall->key_len, &flow, &miss->key_fitness,
3985 &ofproto, &odp_in_port, &miss->initial_vals);
3986 if (error == ENODEV) {
3987 struct drop_key *drop_key;
3989 /* Received packet on port for which we couldn't associate
3990 * an ofproto. This can happen if a port is removed while
3991 * traffic is being received. Print a rate-limited message
3992 * in case it happens frequently. Install a drop flow so
3993 * that future packets of the flow are inexpensively dropped
3995 VLOG_INFO_RL(&rl, "received packet on unassociated port %"PRIu32,
3998 drop_key = drop_key_lookup(backer, upcall->key, upcall->key_len);
4000 drop_key = xmalloc(sizeof *drop_key);
4001 drop_key->key = xmemdup(upcall->key, upcall->key_len);
4002 drop_key->key_len = upcall->key_len;
4004 hmap_insert(&backer->drop_keys, &drop_key->hmap_node,
4005 hash_bytes(drop_key->key, drop_key->key_len, 0));
4006 dpif_flow_put(backer->dpif, DPIF_FP_CREATE | DPIF_FP_MODIFY,
4007 drop_key->key, drop_key->key_len, NULL, 0, NULL);
4015 ofproto->n_missed++;
4016 flow_extract(upcall->packet, flow.skb_priority, flow.skb_mark,
4017 &flow.tunnel, flow.in_port, &miss->flow);
4019 /* Add other packets to a to-do list. */
4020 hash = flow_hash(&miss->flow, 0);
4021 existing_miss = flow_miss_find(&todo, ofproto, &miss->flow, hash);
4022 if (!existing_miss) {
4023 hmap_insert(&todo, &miss->hmap_node, hash);
4024 miss->ofproto = ofproto;
4025 miss->key = upcall->key;
4026 miss->key_len = upcall->key_len;
4027 miss->upcall_type = upcall->type;
4028 miss->odp_in_port = odp_in_port;
4029 list_init(&miss->packets);
4033 miss = existing_miss;
4035 list_push_back(&miss->packets, &upcall->packet->list_node);
4038 /* Process each element in the to-do list, constructing the set of
4039 * operations to batch. */
4041 HMAP_FOR_EACH (miss, hmap_node, &todo) {
4042 handle_flow_miss(miss, flow_miss_ops, &n_ops);
4044 ovs_assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
4046 /* Execute batch. */
4047 for (i = 0; i < n_ops; i++) {
4048 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
4050 dpif_operate(backer->dpif, dpif_ops, n_ops);
4053 for (i = 0; i < n_ops; i++) {
4054 if (flow_miss_ops[i].xout_garbage) {
4055 xlate_out_uninit(&flow_miss_ops[i].xout);
4058 hmap_destroy(&todo);
4061 static enum { SFLOW_UPCALL, MISS_UPCALL, BAD_UPCALL, FLOW_SAMPLE_UPCALL,
4063 classify_upcall(const struct dpif_upcall *upcall)
4065 size_t userdata_len;
4066 union user_action_cookie cookie;
4068 /* First look at the upcall type. */
4069 switch (upcall->type) {
4070 case DPIF_UC_ACTION:
4076 case DPIF_N_UC_TYPES:
4078 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
4082 /* "action" upcalls need a closer look. */
4083 if (!upcall->userdata) {
4084 VLOG_WARN_RL(&rl, "action upcall missing cookie");
4087 userdata_len = nl_attr_get_size(upcall->userdata);
4088 if (userdata_len < sizeof cookie.type
4089 || userdata_len > sizeof cookie) {
4090 VLOG_WARN_RL(&rl, "action upcall cookie has unexpected size %zu",
4094 memset(&cookie, 0, sizeof cookie);
4095 memcpy(&cookie, nl_attr_get(upcall->userdata), userdata_len);
4096 if (userdata_len == sizeof cookie.sflow
4097 && cookie.type == USER_ACTION_COOKIE_SFLOW) {
4098 return SFLOW_UPCALL;
4099 } else if (userdata_len == sizeof cookie.slow_path
4100 && cookie.type == USER_ACTION_COOKIE_SLOW_PATH) {
4102 } else if (userdata_len == sizeof cookie.flow_sample
4103 && cookie.type == USER_ACTION_COOKIE_FLOW_SAMPLE) {
4104 return FLOW_SAMPLE_UPCALL;
4105 } else if (userdata_len == sizeof cookie.ipfix
4106 && cookie.type == USER_ACTION_COOKIE_IPFIX) {
4107 return IPFIX_UPCALL;
4109 VLOG_WARN_RL(&rl, "invalid user cookie of type %"PRIu16
4110 " and size %zu", cookie.type, userdata_len);
4116 handle_sflow_upcall(struct dpif_backer *backer,
4117 const struct dpif_upcall *upcall)
4119 struct ofproto_dpif *ofproto;
4120 union user_action_cookie cookie;
4122 uint32_t odp_in_port;
4124 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
4125 &flow, NULL, &ofproto, &odp_in_port, NULL)
4126 || !ofproto->sflow) {
4130 memset(&cookie, 0, sizeof cookie);
4131 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.sflow);
4132 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
4133 odp_in_port, &cookie);
4137 handle_flow_sample_upcall(struct dpif_backer *backer,
4138 const struct dpif_upcall *upcall)
4140 struct ofproto_dpif *ofproto;
4141 union user_action_cookie cookie;
4144 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
4145 &flow, NULL, &ofproto, NULL, NULL)
4146 || !ofproto->ipfix) {
4150 memset(&cookie, 0, sizeof cookie);
4151 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.flow_sample);
4153 /* The flow reflects exactly the contents of the packet. Sample
4154 * the packet using it. */
4155 dpif_ipfix_flow_sample(ofproto->ipfix, upcall->packet, &flow,
4156 cookie.flow_sample.collector_set_id,
4157 cookie.flow_sample.probability,
4158 cookie.flow_sample.obs_domain_id,
4159 cookie.flow_sample.obs_point_id);
4163 handle_ipfix_upcall(struct dpif_backer *backer,
4164 const struct dpif_upcall *upcall)
4166 struct ofproto_dpif *ofproto;
4169 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
4170 &flow, NULL, &ofproto, NULL, NULL)
4171 || !ofproto->ipfix) {
4175 /* The flow reflects exactly the contents of the packet. Sample
4176 * the packet using it. */
4177 dpif_ipfix_bridge_sample(ofproto->ipfix, upcall->packet, &flow);
4181 handle_upcalls(struct dpif_backer *backer, unsigned int max_batch)
4183 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
4184 struct ofpbuf miss_bufs[FLOW_MISS_MAX_BATCH];
4185 uint64_t miss_buf_stubs[FLOW_MISS_MAX_BATCH][4096 / 8];
4190 ovs_assert(max_batch <= FLOW_MISS_MAX_BATCH);
4193 for (n_processed = 0; n_processed < max_batch; n_processed++) {
4194 struct dpif_upcall *upcall = &misses[n_misses];
4195 struct ofpbuf *buf = &miss_bufs[n_misses];
4198 ofpbuf_use_stub(buf, miss_buf_stubs[n_misses],
4199 sizeof miss_buf_stubs[n_misses]);
4200 error = dpif_recv(backer->dpif, upcall, buf);
4206 switch (classify_upcall(upcall)) {
4208 /* Handle it later. */
4213 handle_sflow_upcall(backer, upcall);
4217 case FLOW_SAMPLE_UPCALL:
4218 handle_flow_sample_upcall(backer, upcall);
4223 handle_ipfix_upcall(backer, upcall);
4233 /* Handle deferred MISS_UPCALL processing. */
4234 handle_miss_upcalls(backer, misses, n_misses);
4235 for (i = 0; i < n_misses; i++) {
4236 ofpbuf_uninit(&miss_bufs[i]);
4242 /* Flow expiration. */
4244 static int subfacet_max_idle(const struct ofproto_dpif *);
4245 static void update_stats(struct dpif_backer *);
4246 static void rule_expire(struct rule_dpif *);
4247 static void expire_subfacets(struct ofproto_dpif *, int dp_max_idle);
4249 /* This function is called periodically by run(). Its job is to collect
4250 * updates for the flows that have been installed into the datapath, most
4251 * importantly when they last were used, and then use that information to
4252 * expire flows that have not been used recently.
4254 * Returns the number of milliseconds after which it should be called again. */
4256 expire(struct dpif_backer *backer)
4258 struct ofproto_dpif *ofproto;
4259 int max_idle = INT32_MAX;
4261 /* Periodically clear out the drop keys in an effort to keep them
4262 * relatively few. */
4263 drop_key_clear(backer);
4265 /* Update stats for each flow in the backer. */
4266 update_stats(backer);
4268 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4269 struct rule *rule, *next_rule;
4272 if (ofproto->backer != backer) {
4276 /* Keep track of the max number of flows per ofproto_dpif. */
4277 update_max_subfacet_count(ofproto);
4279 /* Expire subfacets that have been idle too long. */
4280 dp_max_idle = subfacet_max_idle(ofproto);
4281 expire_subfacets(ofproto, dp_max_idle);
4283 max_idle = MIN(max_idle, dp_max_idle);
4285 /* Expire OpenFlow flows whose idle_timeout or hard_timeout
4287 LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
4288 &ofproto->up.expirable) {
4289 rule_expire(rule_dpif_cast(rule));
4292 /* All outstanding data in existing flows has been accounted, so it's a
4293 * good time to do bond rebalancing. */
4294 if (ofproto->has_bonded_bundles) {
4295 struct ofbundle *bundle;
4297 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
4299 bond_rebalance(bundle->bond, &backer->revalidate_set);
4305 return MIN(max_idle, 1000);
4308 /* Updates flow table statistics given that the datapath just reported 'stats'
4309 * as 'subfacet''s statistics. */
4311 update_subfacet_stats(struct subfacet *subfacet,
4312 const struct dpif_flow_stats *stats)
4314 struct facet *facet = subfacet->facet;
4316 if (stats->n_packets >= subfacet->dp_packet_count) {
4317 uint64_t extra = stats->n_packets - subfacet->dp_packet_count;
4318 facet->packet_count += extra;
4320 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
4323 if (stats->n_bytes >= subfacet->dp_byte_count) {
4324 facet->byte_count += stats->n_bytes - subfacet->dp_byte_count;
4326 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
4329 subfacet->dp_packet_count = stats->n_packets;
4330 subfacet->dp_byte_count = stats->n_bytes;
4332 facet->tcp_flags |= stats->tcp_flags;
4334 subfacet_update_time(subfacet, stats->used);
4335 if (facet->accounted_bytes < facet->byte_count) {
4337 facet_account(facet);
4338 facet->accounted_bytes = facet->byte_count;
4342 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
4343 * about, or a flow that shouldn't be installed but was anyway. Delete it. */
4345 delete_unexpected_flow(struct ofproto_dpif *ofproto,
4346 const struct nlattr *key, size_t key_len)
4348 if (!VLOG_DROP_WARN(&rl)) {
4352 odp_flow_key_format(key, key_len, &s);
4353 VLOG_WARN("unexpected flow on %s: %s", ofproto->up.name, ds_cstr(&s));
4357 COVERAGE_INC(facet_unexpected);
4358 dpif_flow_del(ofproto->backer->dpif, key, key_len, NULL);
4361 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
4363 * This function also pushes statistics updates to rules which each facet
4364 * resubmits into. Generally these statistics will be accurate. However, if a
4365 * facet changes the rule it resubmits into at some time in between
4366 * update_stats() runs, it is possible that statistics accrued to the
4367 * old rule will be incorrectly attributed to the new rule. This could be
4368 * avoided by calling update_stats() whenever rules are created or
4369 * deleted. However, the performance impact of making so many calls to the
4370 * datapath do not justify the benefit of having perfectly accurate statistics.
4372 * In addition, this function maintains per ofproto flow hit counts. The patch
4373 * port is not treated specially. e.g. A packet ingress from br0 patched into
4374 * br1 will increase the hit count of br0 by 1, however, does not affect
4375 * the hit or miss counts of br1.
4378 update_stats(struct dpif_backer *backer)
4380 const struct dpif_flow_stats *stats;
4381 struct dpif_flow_dump dump;
4382 const struct nlattr *key;
4383 struct ofproto_dpif *ofproto;
4386 dpif_flow_dump_start(&dump, backer->dpif);
4387 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
4389 struct subfacet *subfacet;
4390 struct ofport_dpif *ofport;
4393 if (ofproto_receive(backer, NULL, key, key_len, &flow, NULL, &ofproto,
4398 ofproto->total_subfacet_count += hmap_count(&ofproto->subfacets);
4399 ofproto->n_update_stats++;
4401 ofport = get_ofp_port(ofproto, flow.in_port);
4402 if (ofport && ofport->tnl_port) {
4403 netdev_vport_inc_rx(ofport->up.netdev, stats);
4406 key_hash = odp_flow_key_hash(key, key_len);
4407 subfacet = subfacet_find(ofproto, key, key_len, key_hash);
4408 switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
4410 /* Update ofproto_dpif's hit count. */
4411 if (stats->n_packets > subfacet->dp_packet_count) {
4412 uint64_t delta = stats->n_packets - subfacet->dp_packet_count;
4413 dpif_stats_update_hit_count(ofproto, delta);
4416 update_subfacet_stats(subfacet, stats);
4420 /* Stats are updated per-packet. */
4423 case SF_NOT_INSTALLED:
4425 delete_unexpected_flow(ofproto, key, key_len);
4430 dpif_flow_dump_done(&dump);
4432 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4433 update_moving_averages(ofproto);
4438 /* Calculates and returns the number of milliseconds of idle time after which
4439 * subfacets should expire from the datapath. When a subfacet expires, we fold
4440 * its statistics into its facet, and when a facet's last subfacet expires, we
4441 * fold its statistic into its rule. */
4443 subfacet_max_idle(const struct ofproto_dpif *ofproto)
4446 * Idle time histogram.
4448 * Most of the time a switch has a relatively small number of subfacets.
4449 * When this is the case we might as well keep statistics for all of them
4450 * in userspace and to cache them in the kernel datapath for performance as
4453 * As the number of subfacets increases, the memory required to maintain
4454 * statistics about them in userspace and in the kernel becomes
4455 * significant. However, with a large number of subfacets it is likely
4456 * that only a few of them are "heavy hitters" that consume a large amount
4457 * of bandwidth. At this point, only heavy hitters are worth caching in
4458 * the kernel and maintaining in userspaces; other subfacets we can
4461 * The technique used to compute the idle time is to build a histogram with
4462 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
4463 * that is installed in the kernel gets dropped in the appropriate bucket.
4464 * After the histogram has been built, we compute the cutoff so that only
4465 * the most-recently-used 1% of subfacets (but at least
4466 * ofproto->up.flow_eviction_threshold flows) are kept cached. At least
4467 * the most-recently-used bucket of subfacets is kept, so actually an
4468 * arbitrary number of subfacets can be kept in any given expiration run
4469 * (though the next run will delete most of those unless they receive
4472 * This requires a second pass through the subfacets, in addition to the
4473 * pass made by update_stats(), because the former function never looks at
4474 * uninstallable subfacets.
4476 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
4477 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
4478 int buckets[N_BUCKETS] = { 0 };
4479 int total, subtotal, bucket;
4480 struct subfacet *subfacet;
4484 total = hmap_count(&ofproto->subfacets);
4485 if (total <= ofproto->up.flow_eviction_threshold) {
4486 return N_BUCKETS * BUCKET_WIDTH;
4489 /* Build histogram. */
4491 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
4492 long long int idle = now - subfacet->used;
4493 int bucket = (idle <= 0 ? 0
4494 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
4495 : (unsigned int) idle / BUCKET_WIDTH);
4499 /* Find the first bucket whose flows should be expired. */
4500 subtotal = bucket = 0;
4502 subtotal += buckets[bucket++];
4503 } while (bucket < N_BUCKETS &&
4504 subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100));
4506 if (VLOG_IS_DBG_ENABLED()) {
4510 ds_put_cstr(&s, "keep");
4511 for (i = 0; i < N_BUCKETS; i++) {
4513 ds_put_cstr(&s, ", drop");
4516 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
4519 VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s));
4523 return bucket * BUCKET_WIDTH;
4527 expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle)
4529 /* Cutoff time for most flows. */
4530 long long int normal_cutoff = time_msec() - dp_max_idle;
4532 /* We really want to keep flows for special protocols around, so use a more
4533 * conservative cutoff. */
4534 long long int special_cutoff = time_msec() - 10000;
4536 struct subfacet *subfacet, *next_subfacet;
4537 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
4541 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
4542 &ofproto->subfacets) {
4543 long long int cutoff;
4545 cutoff = (subfacet->facet->xout.slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP
4549 if (subfacet->used < cutoff) {
4550 if (subfacet->path != SF_NOT_INSTALLED) {
4551 batch[n_batch++] = subfacet;
4552 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
4553 subfacet_destroy_batch(ofproto, batch, n_batch);
4557 subfacet_destroy(subfacet);
4563 subfacet_destroy_batch(ofproto, batch, n_batch);
4567 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
4568 * then delete it entirely. */
4570 rule_expire(struct rule_dpif *rule)
4572 struct facet *facet, *next_facet;
4576 if (rule->up.pending) {
4577 /* We'll have to expire it later. */
4581 /* Has 'rule' expired? */
4583 if (rule->up.hard_timeout
4584 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
4585 reason = OFPRR_HARD_TIMEOUT;
4586 } else if (rule->up.idle_timeout
4587 && now > rule->up.used + rule->up.idle_timeout * 1000) {
4588 reason = OFPRR_IDLE_TIMEOUT;
4593 COVERAGE_INC(ofproto_dpif_expired);
4595 /* Update stats. (This is a no-op if the rule expired due to an idle
4596 * timeout, because that only happens when the rule has no facets left.) */
4597 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4598 facet_remove(facet);
4601 /* Get rid of the rule. */
4602 ofproto_rule_expire(&rule->up, reason);
4607 /* Creates and returns a new facet based on 'miss'.
4609 * The caller must already have determined that no facet with an identical
4610 * 'miss->flow' exists in 'miss->ofproto'.
4612 * 'hash' must be the return value of flow_hash(miss->flow, 0).
4614 * The facet will initially have no subfacets. The caller should create (at
4615 * least) one subfacet with subfacet_create(). */
4616 static struct facet *
4617 facet_create(const struct flow_miss *miss, uint32_t hash)
4619 struct ofproto_dpif *ofproto = miss->ofproto;
4620 struct xlate_in xin;
4621 struct facet *facet;
4623 facet = xzalloc(sizeof *facet);
4624 facet->used = time_msec();
4625 facet->flow = miss->flow;
4626 facet->initial_vals = miss->initial_vals;
4627 facet->rule = rule_dpif_lookup(ofproto, &facet->flow);
4628 facet->learn_rl = time_msec() + 500;
4630 hmap_insert(&ofproto->facets, &facet->hmap_node, hash);
4631 list_push_back(&facet->rule->facets, &facet->list_node);
4632 list_init(&facet->subfacets);
4633 netflow_flow_init(&facet->nf_flow);
4634 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
4636 xlate_in_init(&xin, ofproto, &facet->flow, &facet->initial_vals,
4637 facet->rule, 0, NULL);
4638 xin.may_learn = true;
4639 xlate_actions(&xin, &facet->xout);
4640 facet->nf_flow.output_iface = facet->xout.nf_output_iface;
4646 facet_free(struct facet *facet)
4649 xlate_out_uninit(&facet->xout);
4654 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
4655 * 'packet', which arrived on 'in_port'. */
4657 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
4658 const struct nlattr *odp_actions, size_t actions_len,
4659 struct ofpbuf *packet)
4661 struct odputil_keybuf keybuf;
4665 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4666 odp_flow_key_from_flow(&key, flow,
4667 ofp_port_to_odp_port(ofproto, flow->in_port));
4669 error = dpif_execute(ofproto->backer->dpif, key.data, key.size,
4670 odp_actions, actions_len, packet);
4674 /* Remove 'facet' from 'ofproto' and free up the associated memory:
4676 * - If 'facet' was installed in the datapath, uninstalls it and updates its
4677 * rule's statistics, via subfacet_uninstall().
4679 * - Removes 'facet' from its rule and from ofproto->facets.
4682 facet_remove(struct facet *facet)
4684 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4685 struct subfacet *subfacet, *next_subfacet;
4687 ovs_assert(!list_is_empty(&facet->subfacets));
4689 /* First uninstall all of the subfacets to get final statistics. */
4690 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4691 subfacet_uninstall(subfacet);
4694 /* Flush the final stats to the rule.
4696 * This might require us to have at least one subfacet around so that we
4697 * can use its actions for accounting in facet_account(), which is why we
4698 * have uninstalled but not yet destroyed the subfacets. */
4699 facet_flush_stats(facet);
4701 /* Now we're really all done so destroy everything. */
4702 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
4703 &facet->subfacets) {
4704 subfacet_destroy__(subfacet);
4706 hmap_remove(&ofproto->facets, &facet->hmap_node);
4707 list_remove(&facet->list_node);
4711 /* Feed information from 'facet' back into the learning table to keep it in
4712 * sync with what is actually flowing through the datapath. */
4714 facet_learn(struct facet *facet)
4716 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4717 long long int now = time_msec();
4718 struct xlate_in xin;
4720 if (!facet->xout.has_fin_timeout && now < facet->learn_rl) {
4724 facet->learn_rl = now + 500;
4726 if (!facet->xout.has_learn
4727 && !facet->xout.has_normal
4728 && (!facet->xout.has_fin_timeout
4729 || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
4733 xlate_in_init(&xin, ofproto, &facet->flow, &facet->initial_vals,
4734 facet->rule, facet->tcp_flags, NULL);
4735 xin.may_learn = true;
4736 xlate_actions_for_side_effects(&xin);
4740 facet_account(struct facet *facet)
4742 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4743 const struct nlattr *a;
4748 if (!facet->xout.has_normal || !ofproto->has_bonded_bundles) {
4751 n_bytes = facet->byte_count - facet->accounted_bytes;
4753 /* This loop feeds byte counters to bond_account() for rebalancing to use
4754 * as a basis. We also need to track the actual VLAN on which the packet
4755 * is going to be sent to ensure that it matches the one passed to
4756 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
4759 * We use the actions from an arbitrary subfacet because they should all
4760 * be equally valid for our purpose. */
4761 vlan_tci = facet->flow.vlan_tci;
4762 NL_ATTR_FOR_EACH_UNSAFE (a, left, facet->xout.odp_actions.data,
4763 facet->xout.odp_actions.size) {
4764 const struct ovs_action_push_vlan *vlan;
4765 struct ofport_dpif *port;
4767 switch (nl_attr_type(a)) {
4768 case OVS_ACTION_ATTR_OUTPUT:
4769 port = get_odp_port(ofproto, nl_attr_get_u32(a));
4770 if (port && port->bundle && port->bundle->bond) {
4771 bond_account(port->bundle->bond, &facet->flow,
4772 vlan_tci_to_vid(vlan_tci), n_bytes);
4776 case OVS_ACTION_ATTR_POP_VLAN:
4777 vlan_tci = htons(0);
4780 case OVS_ACTION_ATTR_PUSH_VLAN:
4781 vlan = nl_attr_get(a);
4782 vlan_tci = vlan->vlan_tci;
4788 /* Returns true if the only action for 'facet' is to send to the controller.
4789 * (We don't report NetFlow expiration messages for such facets because they
4790 * are just part of the control logic for the network, not real traffic). */
4792 facet_is_controller_flow(struct facet *facet)
4795 const struct rule *rule = &facet->rule->up;
4796 const struct ofpact *ofpacts = rule->ofpacts;
4797 size_t ofpacts_len = rule->ofpacts_len;
4799 if (ofpacts_len > 0 &&
4800 ofpacts->type == OFPACT_CONTROLLER &&
4801 ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len)) {
4808 /* Folds all of 'facet''s statistics into its rule. Also updates the
4809 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
4810 * 'facet''s statistics in the datapath should have been zeroed and folded into
4811 * its packet and byte counts before this function is called. */
4813 facet_flush_stats(struct facet *facet)
4815 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4816 struct subfacet *subfacet;
4818 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4819 ovs_assert(!subfacet->dp_byte_count);
4820 ovs_assert(!subfacet->dp_packet_count);
4823 facet_push_stats(facet);
4824 if (facet->accounted_bytes < facet->byte_count) {
4825 facet_account(facet);
4826 facet->accounted_bytes = facet->byte_count;
4829 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
4830 struct ofexpired expired;
4831 expired.flow = facet->flow;
4832 expired.packet_count = facet->packet_count;
4833 expired.byte_count = facet->byte_count;
4834 expired.used = facet->used;
4835 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4838 /* Reset counters to prevent double counting if 'facet' ever gets
4840 facet_reset_counters(facet);
4842 netflow_flow_clear(&facet->nf_flow);
4843 facet->tcp_flags = 0;
4846 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
4847 * Returns it if found, otherwise a null pointer.
4849 * 'hash' must be the return value of flow_hash(flow, 0).
4851 * The returned facet might need revalidation; use facet_lookup_valid()
4852 * instead if that is important. */
4853 static struct facet *
4854 facet_find(struct ofproto_dpif *ofproto,
4855 const struct flow *flow, uint32_t hash)
4857 struct facet *facet;
4859 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, hash, &ofproto->facets) {
4860 if (flow_equal(flow, &facet->flow)) {
4868 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
4869 * Returns it if found, otherwise a null pointer.
4871 * 'hash' must be the return value of flow_hash(flow, 0).
4873 * The returned facet is guaranteed to be valid. */
4874 static struct facet *
4875 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow,
4878 struct facet *facet;
4880 facet = facet_find(ofproto, flow, hash);
4882 && (ofproto->backer->need_revalidate
4883 || tag_set_intersects(&ofproto->backer->revalidate_set,
4885 && !facet_revalidate(facet)) {
4886 facet_revalidate(facet);
4888 /* facet_revalidate() may have destroyed 'facet'. */
4889 facet = facet_find(ofproto, flow, hash);
4896 facet_check_consistency(struct facet *facet)
4898 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
4900 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4902 struct xlate_out xout;
4903 struct xlate_in xin;
4905 struct rule_dpif *rule;
4908 /* Check the rule for consistency. */
4909 rule = rule_dpif_lookup(ofproto, &facet->flow);
4910 if (rule != facet->rule) {
4911 if (!VLOG_DROP_WARN(&rl)) {
4912 struct ds s = DS_EMPTY_INITIALIZER;
4914 flow_format(&s, &facet->flow);
4915 ds_put_format(&s, ": facet associated with wrong rule (was "
4916 "table=%"PRIu8",", facet->rule->up.table_id);
4917 cls_rule_format(&facet->rule->up.cr, &s);
4918 ds_put_format(&s, ") (should have been table=%"PRIu8",",
4920 cls_rule_format(&rule->up.cr, &s);
4921 ds_put_cstr(&s, ")\n");
4928 /* Check the datapath actions for consistency. */
4929 xlate_in_init(&xin, ofproto, &facet->flow, &facet->initial_vals, rule,
4931 xlate_actions(&xin, &xout);
4933 ok = ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)
4934 && facet->xout.slow == xout.slow;
4935 if (!ok && !VLOG_DROP_WARN(&rl)) {
4936 struct ds s = DS_EMPTY_INITIALIZER;
4938 flow_format(&s, &facet->flow);
4939 ds_put_cstr(&s, ": inconsistency in facet");
4941 if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4942 ds_put_cstr(&s, " (actions were: ");
4943 format_odp_actions(&s, facet->xout.odp_actions.data,
4944 facet->xout.odp_actions.size);
4945 ds_put_cstr(&s, ") (correct actions: ");
4946 format_odp_actions(&s, xout.odp_actions.data,
4947 xout.odp_actions.size);
4948 ds_put_cstr(&s, ")");
4951 if (facet->xout.slow != xout.slow) {
4952 ds_put_format(&s, " slow path incorrect. should be %d", xout.slow);
4957 xlate_out_uninit(&xout);
4962 /* Re-searches the classifier for 'facet':
4964 * - If the rule found is different from 'facet''s current rule, moves
4965 * 'facet' to the new rule and recompiles its actions.
4967 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
4968 * where it is and recompiles its actions anyway.
4970 * - If any of 'facet''s subfacets correspond to a new flow according to
4971 * ofproto_receive(), 'facet' is removed.
4973 * Returns true if 'facet' is still valid. False if 'facet' was removed. */
4975 facet_revalidate(struct facet *facet)
4977 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4978 struct rule_dpif *new_rule;
4979 struct subfacet *subfacet;
4980 struct xlate_out xout;
4981 struct xlate_in xin;
4983 COVERAGE_INC(facet_revalidate);
4985 /* Check that child subfacets still correspond to this facet. Tunnel
4986 * configuration changes could cause a subfacet's OpenFlow in_port to
4988 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4989 struct ofproto_dpif *recv_ofproto;
4990 struct flow recv_flow;
4993 error = ofproto_receive(ofproto->backer, NULL, subfacet->key,
4994 subfacet->key_len, &recv_flow, NULL,
4995 &recv_ofproto, NULL, NULL);
4997 || recv_ofproto != ofproto
4998 || memcmp(&recv_flow, &facet->flow, sizeof recv_flow)) {
4999 facet_remove(facet);
5004 new_rule = rule_dpif_lookup(ofproto, &facet->flow);
5006 /* Calculate new datapath actions.
5008 * We do not modify any 'facet' state yet, because we might need to, e.g.,
5009 * emit a NetFlow expiration and, if so, we need to have the old state
5010 * around to properly compose it. */
5011 xlate_in_init(&xin, ofproto, &facet->flow, &facet->initial_vals, new_rule,
5013 xlate_actions(&xin, &xout);
5015 /* A facet's slow path reason should only change under dramatic
5016 * circumstances. Rather than try to update everything, it's simpler to
5017 * remove the facet and start over. */
5018 if (facet->xout.slow != xout.slow) {
5019 facet_remove(facet);
5020 xlate_out_uninit(&xout);
5024 if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
5025 LIST_FOR_EACH(subfacet, list_node, &facet->subfacets) {
5026 if (subfacet->path == SF_FAST_PATH) {
5027 struct dpif_flow_stats stats;
5029 subfacet_install(subfacet, &xout.odp_actions, &stats);
5030 subfacet_update_stats(subfacet, &stats);
5034 facet_flush_stats(facet);
5036 ofpbuf_clear(&facet->xout.odp_actions);
5037 ofpbuf_put(&facet->xout.odp_actions, xout.odp_actions.data,
5038 xout.odp_actions.size);
5041 /* Update 'facet' now that we've taken care of all the old state. */
5042 facet->xout.tags = xout.tags;
5043 facet->xout.slow = xout.slow;
5044 facet->xout.has_learn = xout.has_learn;
5045 facet->xout.has_normal = xout.has_normal;
5046 facet->xout.has_fin_timeout = xout.has_fin_timeout;
5047 facet->xout.nf_output_iface = xout.nf_output_iface;
5048 facet->xout.mirrors = xout.mirrors;
5049 facet->nf_flow.output_iface = facet->xout.nf_output_iface;
5051 if (facet->rule != new_rule) {
5052 COVERAGE_INC(facet_changed_rule);
5053 list_remove(&facet->list_node);
5054 list_push_back(&new_rule->facets, &facet->list_node);
5055 facet->rule = new_rule;
5056 facet->used = new_rule->up.created;
5057 facet->prev_used = facet->used;
5060 xlate_out_uninit(&xout);
5064 /* Updates 'facet''s used time. Caller is responsible for calling
5065 * facet_push_stats() to update the flows which 'facet' resubmits into. */
5067 facet_update_time(struct facet *facet, long long int used)
5069 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5070 if (used > facet->used) {
5072 ofproto_rule_update_used(&facet->rule->up, used);
5073 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
5078 facet_reset_counters(struct facet *facet)
5080 facet->packet_count = 0;
5081 facet->byte_count = 0;
5082 facet->prev_packet_count = 0;
5083 facet->prev_byte_count = 0;
5084 facet->accounted_bytes = 0;
5088 facet_push_stats(struct facet *facet)
5090 struct dpif_flow_stats stats;
5092 ovs_assert(facet->packet_count >= facet->prev_packet_count);
5093 ovs_assert(facet->byte_count >= facet->prev_byte_count);
5094 ovs_assert(facet->used >= facet->prev_used);
5096 stats.n_packets = facet->packet_count - facet->prev_packet_count;
5097 stats.n_bytes = facet->byte_count - facet->prev_byte_count;
5098 stats.used = facet->used;
5099 stats.tcp_flags = 0;
5101 if (stats.n_packets || stats.n_bytes || facet->used > facet->prev_used) {
5102 facet->prev_packet_count = facet->packet_count;
5103 facet->prev_byte_count = facet->byte_count;
5104 facet->prev_used = facet->used;
5106 rule_credit_stats(facet->rule, &stats);
5107 flow_push_stats(facet, &stats);
5109 update_mirror_stats(ofproto_dpif_cast(facet->rule->up.ofproto),
5110 facet->xout.mirrors, stats.n_packets,
5116 push_all_stats__(bool run_fast)
5118 static long long int rl = LLONG_MIN;
5119 struct ofproto_dpif *ofproto;
5121 if (time_msec() < rl) {
5125 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5126 struct facet *facet;
5128 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
5129 facet_push_stats(facet);
5136 rl = time_msec() + 100;
5140 push_all_stats(void)
5142 push_all_stats__(true);
5146 rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats)
5148 rule->packet_count += stats->n_packets;
5149 rule->byte_count += stats->n_bytes;
5150 ofproto_rule_update_used(&rule->up, stats->used);
5153 /* Pushes flow statistics to the rules which 'facet->flow' resubmits
5154 * into given 'facet->rule''s actions and mirrors. */
5156 flow_push_stats(struct facet *facet, const struct dpif_flow_stats *stats)
5158 struct rule_dpif *rule = facet->rule;
5159 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5160 struct xlate_in xin;
5162 ofproto_rule_update_used(&rule->up, stats->used);
5164 xlate_in_init(&xin, ofproto, &facet->flow, &facet->initial_vals, rule,
5166 xin.resubmit_stats = stats;
5167 xlate_actions_for_side_effects(&xin);
5172 static struct subfacet *
5173 subfacet_find(struct ofproto_dpif *ofproto,
5174 const struct nlattr *key, size_t key_len, uint32_t key_hash)
5176 struct subfacet *subfacet;
5178 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
5179 &ofproto->subfacets) {
5180 if (subfacet->key_len == key_len
5181 && !memcmp(key, subfacet->key, key_len)) {
5189 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
5190 * 'key_fitness', 'key', and 'key_len' members in 'miss'. Returns the
5191 * existing subfacet if there is one, otherwise creates and returns a
5193 static struct subfacet *
5194 subfacet_create(struct facet *facet, struct flow_miss *miss,
5197 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5198 enum odp_key_fitness key_fitness = miss->key_fitness;
5199 const struct nlattr *key = miss->key;
5200 size_t key_len = miss->key_len;
5202 struct subfacet *subfacet;
5204 key_hash = odp_flow_key_hash(key, key_len);
5206 if (list_is_empty(&facet->subfacets)) {
5207 subfacet = &facet->one_subfacet;
5209 subfacet = subfacet_find(ofproto, key, key_len, key_hash);
5211 if (subfacet->facet == facet) {
5215 /* This shouldn't happen. */
5216 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
5217 subfacet_destroy(subfacet);
5220 subfacet = xmalloc(sizeof *subfacet);
5223 hmap_insert(&ofproto->subfacets, &subfacet->hmap_node, key_hash);
5224 list_push_back(&facet->subfacets, &subfacet->list_node);
5225 subfacet->facet = facet;
5226 subfacet->key_fitness = key_fitness;
5227 subfacet->key = xmemdup(key, key_len);
5228 subfacet->key_len = key_len;
5229 subfacet->used = now;
5230 subfacet->created = now;
5231 subfacet->dp_packet_count = 0;
5232 subfacet->dp_byte_count = 0;
5233 subfacet->path = SF_NOT_INSTALLED;
5234 subfacet->odp_in_port = miss->odp_in_port;
5236 ofproto->subfacet_add_count++;
5240 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
5241 * its facet within 'ofproto', and frees it. */
5243 subfacet_destroy__(struct subfacet *subfacet)
5245 struct facet *facet = subfacet->facet;
5246 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5248 /* Update ofproto stats before uninstall the subfacet. */
5249 ofproto->subfacet_del_count++;
5250 ofproto->total_subfacet_life_span += (time_msec() - subfacet->created);
5252 subfacet_uninstall(subfacet);
5253 hmap_remove(&ofproto->subfacets, &subfacet->hmap_node);
5254 list_remove(&subfacet->list_node);
5255 free(subfacet->key);
5256 if (subfacet != &facet->one_subfacet) {
5261 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
5262 * last remaining subfacet in its facet destroys the facet too. */
5264 subfacet_destroy(struct subfacet *subfacet)
5266 struct facet *facet = subfacet->facet;
5268 if (list_is_singleton(&facet->subfacets)) {
5269 /* facet_remove() needs at least one subfacet (it will remove it). */
5270 facet_remove(facet);
5272 subfacet_destroy__(subfacet);
5277 subfacet_destroy_batch(struct ofproto_dpif *ofproto,
5278 struct subfacet **subfacets, int n)
5280 struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH];
5281 struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH];
5282 struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH];
5285 for (i = 0; i < n; i++) {
5286 ops[i].type = DPIF_OP_FLOW_DEL;
5287 ops[i].u.flow_del.key = subfacets[i]->key;
5288 ops[i].u.flow_del.key_len = subfacets[i]->key_len;
5289 ops[i].u.flow_del.stats = &stats[i];
5293 dpif_operate(ofproto->backer->dpif, opsp, n);
5294 for (i = 0; i < n; i++) {
5295 subfacet_reset_dp_stats(subfacets[i], &stats[i]);
5296 subfacets[i]->path = SF_NOT_INSTALLED;
5297 subfacet_destroy(subfacets[i]);
5302 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
5303 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
5304 * in the datapath will be zeroed and 'stats' will be updated with traffic new
5305 * since 'subfacet' was last updated.
5307 * Returns 0 if successful, otherwise a positive errno value. */
5309 subfacet_install(struct subfacet *subfacet, const struct ofpbuf *odp_actions,
5310 struct dpif_flow_stats *stats)
5312 struct facet *facet = subfacet->facet;
5313 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5314 enum subfacet_path path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
5315 const struct nlattr *actions = odp_actions->data;
5316 size_t actions_len = odp_actions->size;
5318 uint64_t slow_path_stub[128 / 8];
5319 enum dpif_flow_put_flags flags;
5322 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
5324 flags |= DPIF_FP_ZERO_STATS;
5327 if (path == SF_SLOW_PATH) {
5328 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
5329 slow_path_stub, sizeof slow_path_stub,
5330 &actions, &actions_len);
5333 ret = dpif_flow_put(ofproto->backer->dpif, flags, subfacet->key,
5334 subfacet->key_len, actions, actions_len, stats);
5337 subfacet_reset_dp_stats(subfacet, stats);
5341 subfacet->path = path;
5346 /* If 'subfacet' is installed in the datapath, uninstalls it. */
5348 subfacet_uninstall(struct subfacet *subfacet)
5350 if (subfacet->path != SF_NOT_INSTALLED) {
5351 struct rule_dpif *rule = subfacet->facet->rule;
5352 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5353 struct dpif_flow_stats stats;
5356 error = dpif_flow_del(ofproto->backer->dpif, subfacet->key,
5357 subfacet->key_len, &stats);
5358 subfacet_reset_dp_stats(subfacet, &stats);
5360 subfacet_update_stats(subfacet, &stats);
5362 subfacet->path = SF_NOT_INSTALLED;
5364 ovs_assert(subfacet->dp_packet_count == 0);
5365 ovs_assert(subfacet->dp_byte_count == 0);
5369 /* Resets 'subfacet''s datapath statistics counters. This should be called
5370 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
5371 * non-null, it should contain the statistics returned by dpif when 'subfacet'
5372 * was reset in the datapath. 'stats' will be modified to include only
5373 * statistics new since 'subfacet' was last updated. */
5375 subfacet_reset_dp_stats(struct subfacet *subfacet,
5376 struct dpif_flow_stats *stats)
5379 && subfacet->dp_packet_count <= stats->n_packets
5380 && subfacet->dp_byte_count <= stats->n_bytes) {
5381 stats->n_packets -= subfacet->dp_packet_count;
5382 stats->n_bytes -= subfacet->dp_byte_count;
5385 subfacet->dp_packet_count = 0;
5386 subfacet->dp_byte_count = 0;
5389 /* Updates 'subfacet''s used time. The caller is responsible for calling
5390 * facet_push_stats() to update the flows which 'subfacet' resubmits into. */
5392 subfacet_update_time(struct subfacet *subfacet, long long int used)
5394 if (used > subfacet->used) {
5395 subfacet->used = used;
5396 facet_update_time(subfacet->facet, used);
5400 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
5402 * Because of the meaning of a subfacet's counters, it only makes sense to do
5403 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
5404 * represents a packet that was sent by hand or if it represents statistics
5405 * that have been cleared out of the datapath. */
5407 subfacet_update_stats(struct subfacet *subfacet,
5408 const struct dpif_flow_stats *stats)
5410 if (stats->n_packets || stats->used > subfacet->used) {
5411 struct facet *facet = subfacet->facet;
5413 subfacet_update_time(subfacet, stats->used);
5414 facet->packet_count += stats->n_packets;
5415 facet->byte_count += stats->n_bytes;
5416 facet->tcp_flags |= stats->tcp_flags;
5417 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
5423 static struct rule_dpif *
5424 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow)
5426 struct rule_dpif *rule;
5428 rule = rule_dpif_lookup__(ofproto, flow, 0);
5433 return rule_dpif_miss_rule(ofproto, flow);
5436 static struct rule_dpif *
5437 rule_dpif_lookup__(struct ofproto_dpif *ofproto, const struct flow *flow,
5440 struct cls_rule *cls_rule;
5441 struct classifier *cls;
5443 if (table_id >= N_TABLES) {
5447 cls = &ofproto->up.tables[table_id].cls;
5448 if (flow->nw_frag & FLOW_NW_FRAG_ANY
5449 && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
5450 /* For OFPC_NORMAL frag_handling, we must pretend that transport ports
5451 * are unavailable. */
5452 struct flow ofpc_normal_flow = *flow;
5453 ofpc_normal_flow.tp_src = htons(0);
5454 ofpc_normal_flow.tp_dst = htons(0);
5455 cls_rule = classifier_lookup(cls, &ofpc_normal_flow);
5457 cls_rule = classifier_lookup(cls, flow);
5459 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
5462 static struct rule_dpif *
5463 rule_dpif_miss_rule(struct ofproto_dpif *ofproto, const struct flow *flow)
5465 struct ofport_dpif *port;
5467 port = get_ofp_port(ofproto, flow->in_port);
5469 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, flow->in_port);
5470 return ofproto->miss_rule;
5473 if (port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN) {
5474 return ofproto->no_packet_in_rule;
5476 return ofproto->miss_rule;
5480 complete_operation(struct rule_dpif *rule)
5482 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5484 rule_invalidate(rule);
5486 struct dpif_completion *c = xmalloc(sizeof *c);
5487 c->op = rule->up.pending;
5488 list_push_back(&ofproto->completions, &c->list_node);
5490 ofoperation_complete(rule->up.pending, 0);
5494 static struct rule *
5497 struct rule_dpif *rule = xmalloc(sizeof *rule);
5502 rule_dealloc(struct rule *rule_)
5504 struct rule_dpif *rule = rule_dpif_cast(rule_);
5509 rule_construct(struct rule *rule_)
5511 struct rule_dpif *rule = rule_dpif_cast(rule_);
5512 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5513 struct rule_dpif *victim;
5516 rule->packet_count = 0;
5517 rule->byte_count = 0;
5519 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
5520 if (victim && !list_is_empty(&victim->facets)) {
5521 struct facet *facet;
5523 rule->facets = victim->facets;
5524 list_moved(&rule->facets);
5525 LIST_FOR_EACH (facet, list_node, &rule->facets) {
5526 /* XXX: We're only clearing our local counters here. It's possible
5527 * that quite a few packets are unaccounted for in the datapath
5528 * statistics. These will be accounted to the new rule instead of
5529 * cleared as required. This could be fixed by clearing out the
5530 * datapath statistics for this facet, but currently it doesn't
5532 facet_reset_counters(facet);
5536 /* Must avoid list_moved() in this case. */
5537 list_init(&rule->facets);
5540 table_id = rule->up.table_id;
5542 rule->tag = victim->tag;
5543 } else if (table_id == 0) {
5548 miniflow_expand(&rule->up.cr.match.flow, &flow);
5549 rule->tag = rule_calculate_tag(&flow, &rule->up.cr.match.mask,
5550 ofproto->tables[table_id].basis);
5553 complete_operation(rule);
5558 rule_destruct(struct rule *rule_)
5560 struct rule_dpif *rule = rule_dpif_cast(rule_);
5561 struct facet *facet, *next_facet;
5563 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
5564 facet_revalidate(facet);
5567 complete_operation(rule);
5571 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
5573 struct rule_dpif *rule = rule_dpif_cast(rule_);
5575 /* push_all_stats() can handle flow misses which, when using the learn
5576 * action, can cause rules to be added and deleted. This can corrupt our
5577 * caller's datastructures which assume that rule_get_stats() doesn't have
5578 * an impact on the flow table. To be safe, we disable miss handling. */
5579 push_all_stats__(false);
5581 /* Start from historical data for 'rule' itself that are no longer tracked
5582 * in facets. This counts, for example, facets that have expired. */
5583 *packets = rule->packet_count;
5584 *bytes = rule->byte_count;
5588 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
5589 struct ofpbuf *packet)
5591 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5592 struct initial_vals initial_vals;
5593 struct dpif_flow_stats stats;
5594 struct xlate_out xout;
5595 struct xlate_in xin;
5597 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
5598 rule_credit_stats(rule, &stats);
5600 initial_vals.vlan_tci = flow->vlan_tci;
5601 xlate_in_init(&xin, ofproto, flow, &initial_vals, rule, stats.tcp_flags,
5603 xin.resubmit_stats = &stats;
5604 xlate_actions(&xin, &xout);
5606 execute_odp_actions(ofproto, flow, xout.odp_actions.data,
5607 xout.odp_actions.size, packet);
5609 xlate_out_uninit(&xout);
5613 rule_execute(struct rule *rule, const struct flow *flow,
5614 struct ofpbuf *packet)
5616 rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
5617 ofpbuf_delete(packet);
5622 rule_modify_actions(struct rule *rule_)
5624 struct rule_dpif *rule = rule_dpif_cast(rule_);
5626 complete_operation(rule);
5629 /* Sends 'packet' out 'ofport'.
5630 * May modify 'packet'.
5631 * Returns 0 if successful, otherwise a positive errno value. */
5633 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
5635 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
5636 uint64_t odp_actions_stub[1024 / 8];
5637 struct ofpbuf key, odp_actions;
5638 struct dpif_flow_stats stats;
5639 struct odputil_keybuf keybuf;
5640 struct ofpact_output output;
5641 struct xlate_out xout;
5642 struct xlate_in xin;
5646 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5647 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5649 /* Use OFPP_NONE as the in_port to avoid special packet processing. */
5650 flow_extract(packet, 0, 0, NULL, OFPP_NONE, &flow);
5651 odp_flow_key_from_flow(&key, &flow, ofp_port_to_odp_port(ofproto,
5653 dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
5655 ofpact_init(&output.ofpact, OFPACT_OUTPUT, sizeof output);
5656 output.port = ofport->up.ofp_port;
5659 xlate_in_init(&xin, ofproto, &flow, NULL, NULL, 0, packet);
5660 xin.ofpacts_len = sizeof output;
5661 xin.ofpacts = &output.ofpact;
5662 xin.resubmit_stats = &stats;
5663 xlate_actions(&xin, &xout);
5665 error = dpif_execute(ofproto->backer->dpif,
5667 xout.odp_actions.data, xout.odp_actions.size,
5669 xlate_out_uninit(&xout);
5672 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %s (%s)",
5673 ofproto->up.name, netdev_get_name(ofport->up.netdev),
5677 ofproto->stats.tx_packets++;
5678 ofproto->stats.tx_bytes += packet->size;
5682 /* OpenFlow to datapath action translation. */
5684 static bool may_receive(const struct ofport_dpif *, struct xlate_ctx *);
5685 static void do_xlate_actions(const struct ofpact *, size_t ofpacts_len,
5686 struct xlate_ctx *);
5687 static void xlate_normal(struct xlate_ctx *);
5689 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
5690 * The action will state 'slow' as the reason that the action is in the slow
5691 * path. (This is purely informational: it allows a human viewing "ovs-dpctl
5692 * dump-flows" output to see why a flow is in the slow path.)
5694 * The 'stub_size' bytes in 'stub' will be used to store the action.
5695 * 'stub_size' must be large enough for the action.
5697 * The action and its size will be stored in '*actionsp' and '*actions_lenp',
5700 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
5701 enum slow_path_reason slow,
5702 uint64_t *stub, size_t stub_size,
5703 const struct nlattr **actionsp, size_t *actions_lenp)
5705 union user_action_cookie cookie;
5708 cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
5709 cookie.slow_path.unused = 0;
5710 cookie.slow_path.reason = slow;
5712 ofpbuf_use_stack(&buf, stub, stub_size);
5713 if (slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP | SLOW_STP)) {
5714 uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif, UINT32_MAX);
5715 odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
5717 put_userspace_action(ofproto, &buf, flow, &cookie,
5718 sizeof cookie.slow_path);
5720 *actionsp = buf.data;
5721 *actions_lenp = buf.size;
5725 put_userspace_action(const struct ofproto_dpif *ofproto,
5726 struct ofpbuf *odp_actions,
5727 const struct flow *flow,
5728 const union user_action_cookie *cookie,
5729 const size_t cookie_size)
5733 pid = dpif_port_get_pid(ofproto->backer->dpif,
5734 ofp_port_to_odp_port(ofproto, flow->in_port));
5736 return odp_put_userspace_action(pid, cookie, cookie_size, odp_actions);
5739 /* Compose SAMPLE action for sFlow or IPFIX. The given probability is
5740 * the number of packets out of UINT32_MAX to sample. The given
5741 * cookie is passed back in the callback for each sampled packet.
5744 compose_sample_action(const struct ofproto_dpif *ofproto,
5745 struct ofpbuf *odp_actions,
5746 const struct flow *flow,
5747 const uint32_t probability,
5748 const union user_action_cookie *cookie,
5749 const size_t cookie_size)
5751 size_t sample_offset, actions_offset;
5754 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
5756 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
5758 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
5759 cookie_offset = put_userspace_action(ofproto, odp_actions, flow, cookie,
5762 nl_msg_end_nested(odp_actions, actions_offset);
5763 nl_msg_end_nested(odp_actions, sample_offset);
5764 return cookie_offset;
5768 compose_sflow_cookie(const struct ofproto_dpif *ofproto,
5769 ovs_be16 vlan_tci, uint32_t odp_port,
5770 unsigned int n_outputs, union user_action_cookie *cookie)
5774 cookie->type = USER_ACTION_COOKIE_SFLOW;
5775 cookie->sflow.vlan_tci = vlan_tci;
5777 /* See http://www.sflow.org/sflow_version_5.txt (search for "Input/output
5778 * port information") for the interpretation of cookie->output. */
5779 switch (n_outputs) {
5781 /* 0x40000000 | 256 means "packet dropped for unknown reason". */
5782 cookie->sflow.output = 0x40000000 | 256;
5786 ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port);
5788 cookie->sflow.output = ifindex;
5793 /* 0x80000000 means "multiple output ports. */
5794 cookie->sflow.output = 0x80000000 | n_outputs;
5799 /* Compose SAMPLE action for sFlow bridge sampling. */
5801 compose_sflow_action(const struct ofproto_dpif *ofproto,
5802 struct ofpbuf *odp_actions,
5803 const struct flow *flow,
5806 uint32_t probability;
5807 union user_action_cookie cookie;
5809 if (!ofproto->sflow || flow->in_port == OFPP_NONE) {
5813 probability = dpif_sflow_get_probability(ofproto->sflow);
5814 compose_sflow_cookie(ofproto, htons(0), odp_port,
5815 odp_port == OVSP_NONE ? 0 : 1, &cookie);
5817 return compose_sample_action(ofproto, odp_actions, flow, probability,
5818 &cookie, sizeof cookie.sflow);
5822 compose_flow_sample_cookie(uint16_t probability, uint32_t collector_set_id,
5823 uint32_t obs_domain_id, uint32_t obs_point_id,
5824 union user_action_cookie *cookie)
5826 cookie->type = USER_ACTION_COOKIE_FLOW_SAMPLE;
5827 cookie->flow_sample.probability = probability;
5828 cookie->flow_sample.collector_set_id = collector_set_id;
5829 cookie->flow_sample.obs_domain_id = obs_domain_id;
5830 cookie->flow_sample.obs_point_id = obs_point_id;
5834 compose_ipfix_cookie(union user_action_cookie *cookie)
5836 cookie->type = USER_ACTION_COOKIE_IPFIX;
5839 /* Compose SAMPLE action for IPFIX bridge sampling. */
5841 compose_ipfix_action(const struct ofproto_dpif *ofproto,
5842 struct ofpbuf *odp_actions,
5843 const struct flow *flow)
5845 uint32_t probability;
5846 union user_action_cookie cookie;
5848 if (!ofproto->ipfix || flow->in_port == OFPP_NONE) {
5852 probability = dpif_ipfix_get_bridge_exporter_probability(ofproto->ipfix);
5853 compose_ipfix_cookie(&cookie);
5855 compose_sample_action(ofproto, odp_actions, flow, probability,
5856 &cookie, sizeof cookie.ipfix);
5859 /* SAMPLE action for sFlow must be first action in any given list of
5860 * actions. At this point we do not have all information required to
5861 * build it. So try to build sample action as complete as possible. */
5863 add_sflow_action(struct xlate_ctx *ctx)
5865 ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto,
5866 &ctx->xout->odp_actions,
5867 &ctx->xin->flow, OVSP_NONE);
5868 ctx->sflow_odp_port = 0;
5869 ctx->sflow_n_outputs = 0;
5872 /* SAMPLE action for IPFIX must be 1st or 2nd action in any given list
5873 * of actions, eventually after the SAMPLE action for sFlow. */
5875 add_ipfix_action(struct xlate_ctx *ctx)
5877 compose_ipfix_action(ctx->ofproto, &ctx->xout->odp_actions,
5881 /* Fix SAMPLE action according to data collected while composing ODP actions.
5882 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
5883 * USERSPACE action's user-cookie which is required for sflow. */
5885 fix_sflow_action(struct xlate_ctx *ctx)
5887 const struct flow *base = &ctx->base_flow;
5888 union user_action_cookie *cookie;
5890 if (!ctx->user_cookie_offset) {
5894 cookie = ofpbuf_at(&ctx->xout->odp_actions, ctx->user_cookie_offset,
5895 sizeof cookie->sflow);
5896 ovs_assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
5898 compose_sflow_cookie(ctx->ofproto, base->vlan_tci,
5899 ctx->sflow_odp_port, ctx->sflow_n_outputs, cookie);
5903 compose_output_action__(struct xlate_ctx *ctx, uint16_t ofp_port,
5906 const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
5907 ovs_be16 flow_vlan_tci;
5908 uint32_t flow_skb_mark;
5909 uint8_t flow_nw_tos;
5910 struct priority_to_dscp *pdscp;
5911 uint32_t out_port, odp_port;
5913 /* If 'struct flow' gets additional metadata, we'll need to zero it out
5914 * before traversing a patch port. */
5915 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 20);
5918 xlate_report(ctx, "Nonexistent output port");
5920 } else if (ofport->up.pp.config & OFPUTIL_PC_NO_FWD) {
5921 xlate_report(ctx, "OFPPC_NO_FWD set, skipping output");
5923 } else if (check_stp && !stp_forward_in_state(ofport->stp_state)) {
5924 xlate_report(ctx, "STP not in forwarding state, skipping output");
5928 if (netdev_vport_is_patch(ofport->up.netdev)) {
5929 struct ofport_dpif *peer = ofport_get_peer(ofport);
5930 struct flow old_flow = ctx->xin->flow;
5931 const struct ofproto_dpif *peer_ofproto;
5932 enum slow_path_reason special;
5933 struct ofport_dpif *in_port;
5936 xlate_report(ctx, "Nonexistent patch port peer");
5940 peer_ofproto = ofproto_dpif_cast(peer->up.ofproto);
5941 if (peer_ofproto->backer != ctx->ofproto->backer) {
5942 xlate_report(ctx, "Patch port peer on a different datapath");
5946 ctx->ofproto = ofproto_dpif_cast(peer->up.ofproto);
5947 ctx->xin->flow.in_port = peer->up.ofp_port;
5948 ctx->xin->flow.metadata = htonll(0);
5949 memset(&ctx->xin->flow.tunnel, 0, sizeof ctx->xin->flow.tunnel);
5950 memset(ctx->xin->flow.regs, 0, sizeof ctx->xin->flow.regs);
5952 in_port = get_ofp_port(ctx->ofproto, ctx->xin->flow.in_port);
5953 special = process_special(ctx->ofproto, &ctx->xin->flow, in_port,
5956 ctx->xout->slow = special;
5957 } else if (!in_port || may_receive(in_port, ctx)) {
5958 if (!in_port || stp_forward_in_state(in_port->stp_state)) {
5959 xlate_table_action(ctx, ctx->xin->flow.in_port, 0, true);
5961 /* Forwarding is disabled by STP. Let OFPP_NORMAL and the
5962 * learning action look at the packet, then drop it. */
5963 struct flow old_base_flow = ctx->base_flow;
5964 size_t old_size = ctx->xout->odp_actions.size;
5965 xlate_table_action(ctx, ctx->xin->flow.in_port, 0, true);
5966 ctx->base_flow = old_base_flow;
5967 ctx->xout->odp_actions.size = old_size;
5971 ctx->xin->flow = old_flow;
5972 ctx->ofproto = ofproto_dpif_cast(ofport->up.ofproto);
5974 if (ctx->xin->resubmit_stats) {
5975 netdev_vport_inc_tx(ofport->up.netdev, ctx->xin->resubmit_stats);
5976 netdev_vport_inc_rx(peer->up.netdev, ctx->xin->resubmit_stats);
5982 flow_vlan_tci = ctx->xin->flow.vlan_tci;
5983 flow_skb_mark = ctx->xin->flow.skb_mark;
5984 flow_nw_tos = ctx->xin->flow.nw_tos;
5986 pdscp = get_priority(ofport, ctx->xin->flow.skb_priority);
5988 ctx->xin->flow.nw_tos &= ~IP_DSCP_MASK;
5989 ctx->xin->flow.nw_tos |= pdscp->dscp;
5992 if (ofport->tnl_port) {
5993 /* Save tunnel metadata so that changes made due to
5994 * the Logical (tunnel) Port are not visible for any further
5995 * matches, while explicit set actions on tunnel metadata are.
5997 struct flow_tnl flow_tnl = ctx->xin->flow.tunnel;
5998 odp_port = tnl_port_send(ofport->tnl_port, &ctx->xin->flow);
5999 if (odp_port == OVSP_NONE) {
6000 xlate_report(ctx, "Tunneling decided against output");
6001 goto out; /* restore flow_nw_tos */
6003 if (ctx->xin->flow.tunnel.ip_dst == ctx->orig_tunnel_ip_dst) {
6004 xlate_report(ctx, "Not tunneling to our own address");
6005 goto out; /* restore flow_nw_tos */
6007 if (ctx->xin->resubmit_stats) {
6008 netdev_vport_inc_tx(ofport->up.netdev, ctx->xin->resubmit_stats);
6010 out_port = odp_port;
6011 commit_odp_tunnel_action(&ctx->xin->flow, &ctx->base_flow,
6012 &ctx->xout->odp_actions);
6013 ctx->xin->flow.tunnel = flow_tnl; /* Restore tunnel metadata */
6015 odp_port = ofport->odp_port;
6016 out_port = vsp_realdev_to_vlandev(ctx->ofproto, odp_port,
6017 ctx->xin->flow.vlan_tci);
6018 if (out_port != odp_port) {
6019 ctx->xin->flow.vlan_tci = htons(0);
6021 ctx->xin->flow.skb_mark &= ~IPSEC_MARK;
6023 commit_odp_actions(&ctx->xin->flow, &ctx->base_flow,
6024 &ctx->xout->odp_actions);
6025 nl_msg_put_u32(&ctx->xout->odp_actions, OVS_ACTION_ATTR_OUTPUT, out_port);
6027 ctx->sflow_odp_port = odp_port;
6028 ctx->sflow_n_outputs++;
6029 ctx->xout->nf_output_iface = ofp_port;
6032 ctx->xin->flow.vlan_tci = flow_vlan_tci;
6033 ctx->xin->flow.skb_mark = flow_skb_mark;
6035 ctx->xin->flow.nw_tos = flow_nw_tos;
6039 compose_output_action(struct xlate_ctx *ctx, uint16_t ofp_port)
6041 compose_output_action__(ctx, ofp_port, true);
6045 tag_the_flow(struct xlate_ctx *ctx, struct rule_dpif *rule)
6047 struct ofproto_dpif *ofproto = ctx->ofproto;
6048 uint8_t table_id = ctx->table_id;
6050 if (table_id > 0 && table_id < N_TABLES) {
6051 struct table_dpif *table = &ofproto->tables[table_id];
6052 if (table->other_table) {
6053 ctx->xout->tags |= (rule && rule->tag
6055 : rule_calculate_tag(&ctx->xin->flow,
6056 &table->other_table->mask,
6062 /* Common rule processing in one place to avoid duplicating code. */
6063 static struct rule_dpif *
6064 ctx_rule_hooks(struct xlate_ctx *ctx, struct rule_dpif *rule,
6067 if (ctx->xin->resubmit_hook) {
6068 ctx->xin->resubmit_hook(ctx, rule);
6070 if (rule == NULL && may_packet_in) {
6072 * check if table configuration flags
6073 * OFPTC_TABLE_MISS_CONTROLLER, default.
6074 * OFPTC_TABLE_MISS_CONTINUE,
6075 * OFPTC_TABLE_MISS_DROP
6076 * When OF1.0, OFPTC_TABLE_MISS_CONTINUE is used. What to do?
6078 rule = rule_dpif_miss_rule(ctx->ofproto, &ctx->xin->flow);
6080 if (rule && ctx->xin->resubmit_stats) {
6081 rule_credit_stats(rule, ctx->xin->resubmit_stats);
6087 xlate_table_action(struct xlate_ctx *ctx,
6088 uint16_t in_port, uint8_t table_id, bool may_packet_in)
6090 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
6091 struct rule_dpif *rule;
6092 uint16_t old_in_port = ctx->xin->flow.in_port;
6093 uint8_t old_table_id = ctx->table_id;
6095 ctx->table_id = table_id;
6097 /* Look up a flow with 'in_port' as the input port. */
6098 ctx->xin->flow.in_port = in_port;
6099 rule = rule_dpif_lookup__(ctx->ofproto, &ctx->xin->flow, table_id);
6101 tag_the_flow(ctx, rule);
6103 /* Restore the original input port. Otherwise OFPP_NORMAL and
6104 * OFPP_IN_PORT will have surprising behavior. */
6105 ctx->xin->flow.in_port = old_in_port;
6107 rule = ctx_rule_hooks(ctx, rule, may_packet_in);
6110 struct rule_dpif *old_rule = ctx->rule;
6114 do_xlate_actions(rule->up.ofpacts, rule->up.ofpacts_len, ctx);
6115 ctx->rule = old_rule;
6119 ctx->table_id = old_table_id;
6121 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
6123 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
6124 MAX_RESUBMIT_RECURSION);
6125 ctx->max_resubmit_trigger = true;
6130 xlate_ofpact_resubmit(struct xlate_ctx *ctx,
6131 const struct ofpact_resubmit *resubmit)
6136 in_port = resubmit->in_port;
6137 if (in_port == OFPP_IN_PORT) {
6138 in_port = ctx->xin->flow.in_port;
6141 table_id = resubmit->table_id;
6142 if (table_id == 255) {
6143 table_id = ctx->table_id;
6146 xlate_table_action(ctx, in_port, table_id, false);
6150 flood_packets(struct xlate_ctx *ctx, bool all)
6152 struct ofport_dpif *ofport;
6154 HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) {
6155 uint16_t ofp_port = ofport->up.ofp_port;
6157 if (ofp_port == ctx->xin->flow.in_port) {
6162 compose_output_action__(ctx, ofp_port, false);
6163 } else if (!(ofport->up.pp.config & OFPUTIL_PC_NO_FLOOD)) {
6164 compose_output_action(ctx, ofp_port);
6168 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
6172 execute_controller_action(struct xlate_ctx *ctx, int len,
6173 enum ofp_packet_in_reason reason,
6174 uint16_t controller_id)
6176 struct ofputil_packet_in pin;
6177 struct ofpbuf *packet;
6179 ovs_assert(!ctx->xout->slow || ctx->xout->slow == SLOW_CONTROLLER);
6180 ctx->xout->slow = SLOW_CONTROLLER;
6181 if (!ctx->xin->packet) {
6185 packet = ofpbuf_clone(ctx->xin->packet);
6187 if (packet->l2 && packet->l3) {
6188 struct eth_header *eh;
6189 uint16_t mpls_depth;
6191 eth_pop_vlan(packet);
6194 memcpy(eh->eth_src, ctx->xin->flow.dl_src, sizeof eh->eth_src);
6195 memcpy(eh->eth_dst, ctx->xin->flow.dl_dst, sizeof eh->eth_dst);
6197 if (ctx->xin->flow.vlan_tci & htons(VLAN_CFI)) {
6198 eth_push_vlan(packet, ctx->xin->flow.vlan_tci);
6201 mpls_depth = eth_mpls_depth(packet);
6203 if (mpls_depth < ctx->xin->flow.mpls_depth) {
6204 push_mpls(packet, ctx->xin->flow.dl_type, ctx->xin->flow.mpls_lse);
6205 } else if (mpls_depth > ctx->xin->flow.mpls_depth) {
6206 pop_mpls(packet, ctx->xin->flow.dl_type);
6207 } else if (mpls_depth) {
6208 set_mpls_lse(packet, ctx->xin->flow.mpls_lse);
6212 if (ctx->xin->flow.dl_type == htons(ETH_TYPE_IP)) {
6213 packet_set_ipv4(packet, ctx->xin->flow.nw_src,
6214 ctx->xin->flow.nw_dst, ctx->xin->flow.nw_tos,
6215 ctx->xin->flow.nw_ttl);
6219 if (ctx->xin->flow.nw_proto == IPPROTO_TCP) {
6220 packet_set_tcp_port(packet, ctx->xin->flow.tp_src,
6221 ctx->xin->flow.tp_dst);
6222 } else if (ctx->xin->flow.nw_proto == IPPROTO_UDP) {
6223 packet_set_udp_port(packet, ctx->xin->flow.tp_src,
6224 ctx->xin->flow.tp_dst);
6230 pin.packet = packet->data;
6231 pin.packet_len = packet->size;
6232 pin.reason = reason;
6233 pin.controller_id = controller_id;
6234 pin.table_id = ctx->table_id;
6235 pin.cookie = ctx->rule ? ctx->rule->up.flow_cookie : 0;
6238 flow_get_metadata(&ctx->xin->flow, &pin.fmd);
6240 connmgr_send_packet_in(ctx->ofproto->up.connmgr, &pin);
6241 ofpbuf_delete(packet);
6245 execute_mpls_push_action(struct xlate_ctx *ctx, ovs_be16 eth_type)
6247 ovs_assert(eth_type_mpls(eth_type));
6249 if (ctx->base_flow.mpls_depth) {
6250 ctx->xin->flow.mpls_lse &= ~htonl(MPLS_BOS_MASK);
6251 ctx->xin->flow.mpls_depth++;
6256 if (ctx->xin->flow.dl_type == htons(ETH_TYPE_IPV6)) {
6257 label = htonl(0x2); /* IPV6 Explicit Null. */
6259 label = htonl(0x0); /* IPV4 Explicit Null. */
6261 tc = (ctx->xin->flow.nw_tos & IP_DSCP_MASK) >> 2;
6262 ttl = ctx->xin->flow.nw_ttl ? ctx->xin->flow.nw_ttl : 0x40;
6263 ctx->xin->flow.mpls_lse = set_mpls_lse_values(ttl, tc, 1, label);
6264 ctx->xin->flow.mpls_depth = 1;
6266 ctx->xin->flow.dl_type = eth_type;
6270 execute_mpls_pop_action(struct xlate_ctx *ctx, ovs_be16 eth_type)
6272 ovs_assert(eth_type_mpls(ctx->xin->flow.dl_type));
6273 ovs_assert(!eth_type_mpls(eth_type));
6275 if (ctx->xin->flow.mpls_depth) {
6276 ctx->xin->flow.mpls_depth--;
6277 ctx->xin->flow.mpls_lse = htonl(0);
6278 if (!ctx->xin->flow.mpls_depth) {
6279 ctx->xin->flow.dl_type = eth_type;
6285 compose_dec_ttl(struct xlate_ctx *ctx, struct ofpact_cnt_ids *ids)
6287 if (ctx->xin->flow.dl_type != htons(ETH_TYPE_IP) &&
6288 ctx->xin->flow.dl_type != htons(ETH_TYPE_IPV6)) {
6292 if (ctx->xin->flow.nw_ttl > 1) {
6293 ctx->xin->flow.nw_ttl--;
6298 for (i = 0; i < ids->n_controllers; i++) {
6299 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL,
6303 /* Stop processing for current table. */
6309 execute_set_mpls_ttl_action(struct xlate_ctx *ctx, uint8_t ttl)
6311 if (!eth_type_mpls(ctx->xin->flow.dl_type)) {
6315 set_mpls_lse_ttl(&ctx->xin->flow.mpls_lse, ttl);
6320 execute_dec_mpls_ttl_action(struct xlate_ctx *ctx)
6322 uint8_t ttl = mpls_lse_to_ttl(ctx->xin->flow.mpls_lse);
6324 if (!eth_type_mpls(ctx->xin->flow.dl_type)) {
6330 set_mpls_lse_ttl(&ctx->xin->flow.mpls_lse, ttl);
6333 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL, 0);
6335 /* Stop processing for current table. */
6341 xlate_output_action(struct xlate_ctx *ctx,
6342 uint16_t port, uint16_t max_len, bool may_packet_in)
6344 uint16_t prev_nf_output_iface = ctx->xout->nf_output_iface;
6346 ctx->xout->nf_output_iface = NF_OUT_DROP;
6350 compose_output_action(ctx, ctx->xin->flow.in_port);
6353 xlate_table_action(ctx, ctx->xin->flow.in_port, 0, may_packet_in);
6359 flood_packets(ctx, false);
6362 flood_packets(ctx, true);
6364 case OFPP_CONTROLLER:
6365 execute_controller_action(ctx, max_len, OFPR_ACTION, 0);
6371 if (port != ctx->xin->flow.in_port) {
6372 compose_output_action(ctx, port);
6374 xlate_report(ctx, "skipping output to input port");
6379 if (prev_nf_output_iface == NF_OUT_FLOOD) {
6380 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
6381 } else if (ctx->xout->nf_output_iface == NF_OUT_DROP) {
6382 ctx->xout->nf_output_iface = prev_nf_output_iface;
6383 } else if (prev_nf_output_iface != NF_OUT_DROP &&
6384 ctx->xout->nf_output_iface != NF_OUT_FLOOD) {
6385 ctx->xout->nf_output_iface = NF_OUT_MULTI;
6390 xlate_output_reg_action(struct xlate_ctx *ctx,
6391 const struct ofpact_output_reg *or)
6393 uint64_t port = mf_get_subfield(&or->src, &ctx->xin->flow);
6394 if (port <= UINT16_MAX) {
6395 xlate_output_action(ctx, port, or->max_len, false);
6400 xlate_enqueue_action(struct xlate_ctx *ctx,
6401 const struct ofpact_enqueue *enqueue)
6403 uint16_t ofp_port = enqueue->port;
6404 uint32_t queue_id = enqueue->queue;
6405 uint32_t flow_priority, priority;
6408 /* Translate queue to priority. */
6409 error = dpif_queue_to_priority(ctx->ofproto->backer->dpif,
6410 queue_id, &priority);
6412 /* Fall back to ordinary output action. */
6413 xlate_output_action(ctx, enqueue->port, 0, false);
6417 /* Check output port. */
6418 if (ofp_port == OFPP_IN_PORT) {
6419 ofp_port = ctx->xin->flow.in_port;
6420 } else if (ofp_port == ctx->xin->flow.in_port) {
6424 /* Add datapath actions. */
6425 flow_priority = ctx->xin->flow.skb_priority;
6426 ctx->xin->flow.skb_priority = priority;
6427 compose_output_action(ctx, ofp_port);
6428 ctx->xin->flow.skb_priority = flow_priority;
6430 /* Update NetFlow output port. */
6431 if (ctx->xout->nf_output_iface == NF_OUT_DROP) {
6432 ctx->xout->nf_output_iface = ofp_port;
6433 } else if (ctx->xout->nf_output_iface != NF_OUT_FLOOD) {
6434 ctx->xout->nf_output_iface = NF_OUT_MULTI;
6439 xlate_set_queue_action(struct xlate_ctx *ctx, uint32_t queue_id)
6441 uint32_t skb_priority;
6443 if (!dpif_queue_to_priority(ctx->ofproto->backer->dpif,
6444 queue_id, &skb_priority)) {
6445 ctx->xin->flow.skb_priority = skb_priority;
6447 /* Couldn't translate queue to a priority. Nothing to do. A warning
6448 * has already been logged. */
6453 slave_enabled_cb(uint16_t ofp_port, void *ofproto_)
6455 struct ofproto_dpif *ofproto = ofproto_;
6456 struct ofport_dpif *port;
6466 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
6469 port = get_ofp_port(ofproto, ofp_port);
6470 return port ? port->may_enable : false;
6475 xlate_bundle_action(struct xlate_ctx *ctx,
6476 const struct ofpact_bundle *bundle)
6480 port = bundle_execute(bundle, &ctx->xin->flow, slave_enabled_cb,
6482 if (bundle->dst.field) {
6483 nxm_reg_load(&bundle->dst, port, &ctx->xin->flow);
6485 xlate_output_action(ctx, port, 0, false);
6490 xlate_learn_action(struct xlate_ctx *ctx,
6491 const struct ofpact_learn *learn)
6493 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
6494 struct ofputil_flow_mod fm;
6495 uint64_t ofpacts_stub[1024 / 8];
6496 struct ofpbuf ofpacts;
6499 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
6500 learn_execute(learn, &ctx->xin->flow, &fm, &ofpacts);
6502 error = ofproto_flow_mod(&ctx->ofproto->up, &fm);
6503 if (error && !VLOG_DROP_WARN(&rl)) {
6504 VLOG_WARN("learning action failed to modify flow table (%s)",
6505 ofperr_get_name(error));
6508 ofpbuf_uninit(&ofpacts);
6511 /* Reduces '*timeout' to no more than 'max'. A value of zero in either case
6512 * means "infinite". */
6514 reduce_timeout(uint16_t max, uint16_t *timeout)
6516 if (max && (!*timeout || *timeout > max)) {
6522 xlate_fin_timeout(struct xlate_ctx *ctx,
6523 const struct ofpact_fin_timeout *oft)
6525 if (ctx->xin->tcp_flags & (TCP_FIN | TCP_RST) && ctx->rule) {
6526 struct rule_dpif *rule = ctx->rule;
6528 reduce_timeout(oft->fin_idle_timeout, &rule->up.idle_timeout);
6529 reduce_timeout(oft->fin_hard_timeout, &rule->up.hard_timeout);
6534 xlate_sample_action(struct xlate_ctx *ctx,
6535 const struct ofpact_sample *os)
6537 union user_action_cookie cookie;
6538 /* Scale the probability from 16-bit to 32-bit while representing
6539 * the same percentage. */
6540 uint32_t probability = (os->probability << 16) | os->probability;
6542 commit_odp_actions(&ctx->xin->flow, &ctx->base_flow,
6543 &ctx->xout->odp_actions);
6545 compose_flow_sample_cookie(os->probability, os->collector_set_id,
6546 os->obs_domain_id, os->obs_point_id, &cookie);
6547 compose_sample_action(ctx->ofproto, &ctx->xout->odp_actions, &ctx->xin->flow,
6548 probability, &cookie, sizeof cookie.flow_sample);
6552 may_receive(const struct ofport_dpif *port, struct xlate_ctx *ctx)
6554 if (port->up.pp.config & (eth_addr_equals(ctx->xin->flow.dl_dst,
6556 ? OFPUTIL_PC_NO_RECV_STP
6557 : OFPUTIL_PC_NO_RECV)) {
6561 /* Only drop packets here if both forwarding and learning are
6562 * disabled. If just learning is enabled, we need to have
6563 * OFPP_NORMAL and the learning action have a look at the packet
6564 * before we can drop it. */
6565 if (!stp_forward_in_state(port->stp_state)
6566 && !stp_learn_in_state(port->stp_state)) {
6574 tunnel_ecn_ok(struct xlate_ctx *ctx)
6576 if (is_ip_any(&ctx->base_flow)
6577 && (ctx->xin->flow.tunnel.ip_tos & IP_ECN_MASK) == IP_ECN_CE) {
6578 if ((ctx->base_flow.nw_tos & IP_ECN_MASK) == IP_ECN_NOT_ECT) {
6579 VLOG_WARN_RL(&rl, "dropping tunnel packet marked ECN CE"
6580 " but is not ECN capable");
6583 /* Set the ECN CE value in the tunneled packet. */
6584 ctx->xin->flow.nw_tos |= IP_ECN_CE;
6592 do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len,
6593 struct xlate_ctx *ctx)
6595 bool was_evictable = true;
6596 const struct ofpact *a;
6599 /* Don't let the rule we're working on get evicted underneath us. */
6600 was_evictable = ctx->rule->up.evictable;
6601 ctx->rule->up.evictable = false;
6604 do_xlate_actions_again:
6605 OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) {
6606 struct ofpact_controller *controller;
6607 const struct ofpact_metadata *metadata;
6615 xlate_output_action(ctx, ofpact_get_OUTPUT(a)->port,
6616 ofpact_get_OUTPUT(a)->max_len, true);
6619 case OFPACT_CONTROLLER:
6620 controller = ofpact_get_CONTROLLER(a);
6621 execute_controller_action(ctx, controller->max_len,
6623 controller->controller_id);
6626 case OFPACT_ENQUEUE:
6627 xlate_enqueue_action(ctx, ofpact_get_ENQUEUE(a));
6630 case OFPACT_SET_VLAN_VID:
6631 ctx->xin->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
6632 ctx->xin->flow.vlan_tci |=
6633 (htons(ofpact_get_SET_VLAN_VID(a)->vlan_vid)
6637 case OFPACT_SET_VLAN_PCP:
6638 ctx->xin->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
6639 ctx->xin->flow.vlan_tci |=
6640 htons((ofpact_get_SET_VLAN_PCP(a)->vlan_pcp << VLAN_PCP_SHIFT)
6644 case OFPACT_STRIP_VLAN:
6645 ctx->xin->flow.vlan_tci = htons(0);
6648 case OFPACT_PUSH_VLAN:
6649 /* XXX 802.1AD(QinQ) */
6650 ctx->xin->flow.vlan_tci = htons(VLAN_CFI);
6653 case OFPACT_SET_ETH_SRC:
6654 memcpy(ctx->xin->flow.dl_src, ofpact_get_SET_ETH_SRC(a)->mac,
6658 case OFPACT_SET_ETH_DST:
6659 memcpy(ctx->xin->flow.dl_dst, ofpact_get_SET_ETH_DST(a)->mac,
6663 case OFPACT_SET_IPV4_SRC:
6664 if (ctx->xin->flow.dl_type == htons(ETH_TYPE_IP)) {
6665 ctx->xin->flow.nw_src = ofpact_get_SET_IPV4_SRC(a)->ipv4;
6669 case OFPACT_SET_IPV4_DST:
6670 if (ctx->xin->flow.dl_type == htons(ETH_TYPE_IP)) {
6671 ctx->xin->flow.nw_dst = ofpact_get_SET_IPV4_DST(a)->ipv4;
6675 case OFPACT_SET_IPV4_DSCP:
6676 /* OpenFlow 1.0 only supports IPv4. */
6677 if (ctx->xin->flow.dl_type == htons(ETH_TYPE_IP)) {
6678 ctx->xin->flow.nw_tos &= ~IP_DSCP_MASK;
6679 ctx->xin->flow.nw_tos |= ofpact_get_SET_IPV4_DSCP(a)->dscp;
6683 case OFPACT_SET_L4_SRC_PORT:
6684 if (is_ip_any(&ctx->xin->flow)) {
6685 ctx->xin->flow.tp_src =
6686 htons(ofpact_get_SET_L4_SRC_PORT(a)->port);
6690 case OFPACT_SET_L4_DST_PORT:
6691 if (is_ip_any(&ctx->xin->flow)) {
6692 ctx->xin->flow.tp_dst =
6693 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->xin->flow.tunnel.tun_id =
6703 htonll(ofpact_get_SET_TUNNEL(a)->tun_id);
6706 case OFPACT_SET_QUEUE:
6707 xlate_set_queue_action(ctx, ofpact_get_SET_QUEUE(a)->queue_id);
6710 case OFPACT_POP_QUEUE:
6711 ctx->xin->flow.skb_priority = ctx->orig_skb_priority;
6714 case OFPACT_REG_MOVE:
6715 nxm_execute_reg_move(ofpact_get_REG_MOVE(a), &ctx->xin->flow);
6718 case OFPACT_REG_LOAD:
6719 nxm_execute_reg_load(ofpact_get_REG_LOAD(a), &ctx->xin->flow);
6722 case OFPACT_STACK_PUSH:
6723 nxm_execute_stack_push(ofpact_get_STACK_PUSH(a), &ctx->xin->flow,
6727 case OFPACT_STACK_POP:
6728 nxm_execute_stack_pop(ofpact_get_STACK_POP(a), &ctx->xin->flow,
6732 case OFPACT_PUSH_MPLS:
6733 execute_mpls_push_action(ctx, ofpact_get_PUSH_MPLS(a)->ethertype);
6736 case OFPACT_POP_MPLS:
6737 execute_mpls_pop_action(ctx, ofpact_get_POP_MPLS(a)->ethertype);
6740 case OFPACT_SET_MPLS_TTL:
6741 if (execute_set_mpls_ttl_action(ctx,
6742 ofpact_get_SET_MPLS_TTL(a)->ttl)) {
6747 case OFPACT_DEC_MPLS_TTL:
6748 if (execute_dec_mpls_ttl_action(ctx)) {
6753 case OFPACT_DEC_TTL:
6754 if (compose_dec_ttl(ctx, ofpact_get_DEC_TTL(a))) {
6760 /* Nothing to do. */
6763 case OFPACT_MULTIPATH:
6764 multipath_execute(ofpact_get_MULTIPATH(a), &ctx->xin->flow);
6768 ctx->ofproto->has_bundle_action = true;
6769 xlate_bundle_action(ctx, ofpact_get_BUNDLE(a));
6772 case OFPACT_OUTPUT_REG:
6773 xlate_output_reg_action(ctx, ofpact_get_OUTPUT_REG(a));
6777 ctx->xout->has_learn = true;
6778 if (ctx->xin->may_learn) {
6779 xlate_learn_action(ctx, ofpact_get_LEARN(a));
6787 case OFPACT_FIN_TIMEOUT:
6788 ctx->xout->has_fin_timeout = true;
6789 xlate_fin_timeout(ctx, ofpact_get_FIN_TIMEOUT(a));
6792 case OFPACT_CLEAR_ACTIONS:
6794 * Nothing to do because writa-actions is not supported for now.
6795 * When writa-actions is supported, clear-actions also must
6796 * be supported at the same time.
6800 case OFPACT_WRITE_METADATA:
6801 metadata = ofpact_get_WRITE_METADATA(a);
6802 ctx->xin->flow.metadata &= ~metadata->mask;
6803 ctx->xin->flow.metadata |= metadata->metadata & metadata->mask;
6806 case OFPACT_GOTO_TABLE: {
6807 /* It is assumed that goto-table is the last action. */
6808 struct ofpact_goto_table *ogt = ofpact_get_GOTO_TABLE(a);
6809 struct rule_dpif *rule;
6811 ovs_assert(ctx->table_id < ogt->table_id);
6813 ctx->table_id = ogt->table_id;
6815 /* Look up a flow from the new table. */
6816 rule = rule_dpif_lookup__(ctx->ofproto, &ctx->xin->flow, ctx->table_id);
6818 tag_the_flow(ctx, rule);
6820 rule = ctx_rule_hooks(ctx, rule, true);
6824 ctx->rule->up.evictable = was_evictable;
6827 was_evictable = rule->up.evictable;
6828 rule->up.evictable = false;
6830 /* Tail recursion removal. */
6831 ofpacts = rule->up.ofpacts;
6832 ofpacts_len = rule->up.ofpacts_len;
6833 goto do_xlate_actions_again;
6839 xlate_sample_action(ctx, ofpact_get_SAMPLE(a));
6846 ctx->rule->up.evictable = was_evictable;
6851 xlate_in_init(struct xlate_in *xin, struct ofproto_dpif *ofproto,
6852 const struct flow *flow,
6853 const struct initial_vals *initial_vals,
6854 struct rule_dpif *rule, uint8_t tcp_flags,
6855 const struct ofpbuf *packet)
6857 xin->ofproto = ofproto;
6859 xin->packet = packet;
6860 xin->may_learn = packet != NULL;
6862 xin->ofpacts = NULL;
6863 xin->ofpacts_len = 0;
6864 xin->tcp_flags = tcp_flags;
6865 xin->resubmit_hook = NULL;
6866 xin->report_hook = NULL;
6867 xin->resubmit_stats = NULL;
6870 xin->initial_vals = *initial_vals;
6872 xin->initial_vals.vlan_tci = xin->flow.vlan_tci;
6877 xlate_out_uninit(struct xlate_out *xout)
6880 ofpbuf_uninit(&xout->odp_actions);
6884 /* Translates the 'ofpacts_len' bytes of "struct ofpacts" starting at 'ofpacts'
6885 * into datapath actions in 'odp_actions', using 'ctx'. */
6887 xlate_actions(struct xlate_in *xin, struct xlate_out *xout)
6889 /* Normally false. Set to true if we ever hit MAX_RESUBMIT_RECURSION, so
6890 * that in the future we always keep a copy of the original flow for
6891 * tracing purposes. */
6892 static bool hit_resubmit_limit;
6894 enum slow_path_reason special;
6895 const struct ofpact *ofpacts;
6896 struct ofport_dpif *in_port;
6897 struct flow orig_flow;
6898 struct xlate_ctx ctx;
6901 COVERAGE_INC(ofproto_dpif_xlate);
6903 /* Flow initialization rules:
6904 * - 'base_flow' must match the kernel's view of the packet at the
6905 * time that action processing starts. 'flow' represents any
6906 * transformations we wish to make through actions.
6907 * - By default 'base_flow' and 'flow' are the same since the input
6908 * packet matches the output before any actions are applied.
6909 * - When using VLAN splinters, 'base_flow''s VLAN is set to the value
6910 * of the received packet as seen by the kernel. If we later output
6911 * to another device without any modifications this will cause us to
6912 * insert a new tag since the original one was stripped off by the
6914 * - Tunnel metadata as received is retained in 'flow'. This allows
6915 * tunnel metadata matching also in later tables.
6916 * Since a kernel action for setting the tunnel metadata will only be
6917 * generated with actual tunnel output, changing the tunnel metadata
6918 * values in 'flow' (such as tun_id) will only have effect with a later
6919 * tunnel output action.
6920 * - Tunnel 'base_flow' is completely cleared since that is what the
6921 * kernel does. If we wish to maintain the original values an action
6922 * needs to be generated. */
6927 ctx.ofproto = xin->ofproto;
6928 ctx.rule = xin->rule;
6930 ctx.base_flow = ctx.xin->flow;
6931 ctx.base_flow.vlan_tci = xin->initial_vals.vlan_tci;
6932 memset(&ctx.base_flow.tunnel, 0, sizeof ctx.base_flow.tunnel);
6933 ctx.orig_tunnel_ip_dst = ctx.xin->flow.tunnel.ip_dst;
6937 ctx.xout->has_learn = false;
6938 ctx.xout->has_normal = false;
6939 ctx.xout->has_fin_timeout = false;
6940 ctx.xout->nf_output_iface = NF_OUT_DROP;
6941 ctx.xout->mirrors = 0;
6943 ofpbuf_use_stub(&ctx.xout->odp_actions, ctx.xout->odp_actions_stub,
6944 sizeof ctx.xout->odp_actions_stub);
6945 ofpbuf_reserve(&ctx.xout->odp_actions, NL_A_U32_SIZE);
6948 ctx.max_resubmit_trigger = false;
6949 ctx.orig_skb_priority = ctx.xin->flow.skb_priority;
6954 ofpacts = xin->ofpacts;
6955 ofpacts_len = xin->ofpacts_len;
6956 } else if (xin->rule) {
6957 ofpacts = xin->rule->up.ofpacts;
6958 ofpacts_len = xin->rule->up.ofpacts_len;
6963 ofpbuf_use_stub(&ctx.stack, ctx.init_stack, sizeof ctx.init_stack);
6965 if (ctx.ofproto->has_mirrors || hit_resubmit_limit) {
6966 /* Do this conditionally because the copy is expensive enough that it
6967 * shows up in profiles. */
6968 orig_flow = ctx.xin->flow;
6971 if (ctx.xin->flow.nw_frag & FLOW_NW_FRAG_ANY) {
6972 switch (ctx.ofproto->up.frag_handling) {
6973 case OFPC_FRAG_NORMAL:
6974 /* We must pretend that transport ports are unavailable. */
6975 ctx.xin->flow.tp_src = ctx.base_flow.tp_src = htons(0);
6976 ctx.xin->flow.tp_dst = ctx.base_flow.tp_dst = htons(0);
6979 case OFPC_FRAG_DROP:
6982 case OFPC_FRAG_REASM:
6985 case OFPC_FRAG_NX_MATCH:
6986 /* Nothing to do. */
6989 case OFPC_INVALID_TTL_TO_CONTROLLER:
6994 in_port = get_ofp_port(ctx.ofproto, ctx.xin->flow.in_port);
6995 special = process_special(ctx.ofproto, &ctx.xin->flow, in_port,
6998 ctx.xout->slow = special;
7000 static struct vlog_rate_limit trace_rl = VLOG_RATE_LIMIT_INIT(1, 1);
7001 struct initial_vals initial_vals;
7002 size_t sample_actions_len;
7003 uint32_t local_odp_port;
7005 initial_vals.vlan_tci = ctx.base_flow.vlan_tci;
7007 add_sflow_action(&ctx);
7008 add_ipfix_action(&ctx);
7009 sample_actions_len = ctx.xout->odp_actions.size;
7011 if (tunnel_ecn_ok(&ctx) && (!in_port || may_receive(in_port, &ctx))) {
7012 do_xlate_actions(ofpacts, ofpacts_len, &ctx);
7014 /* We've let OFPP_NORMAL and the learning action look at the
7015 * packet, so drop it now if forwarding is disabled. */
7016 if (in_port && !stp_forward_in_state(in_port->stp_state)) {
7017 ctx.xout->odp_actions.size = sample_actions_len;
7021 if (ctx.max_resubmit_trigger && !ctx.xin->resubmit_hook) {
7022 if (!hit_resubmit_limit) {
7023 /* We didn't record the original flow. Make sure we do from
7025 hit_resubmit_limit = true;
7026 } else if (!VLOG_DROP_ERR(&trace_rl)) {
7027 struct ds ds = DS_EMPTY_INITIALIZER;
7029 ofproto_trace(ctx.ofproto, &orig_flow, ctx.xin->packet,
7030 &initial_vals, &ds);
7031 VLOG_ERR("Trace triggered by excessive resubmit "
7032 "recursion:\n%s", ds_cstr(&ds));
7037 local_odp_port = ofp_port_to_odp_port(ctx.ofproto, OFPP_LOCAL);
7038 if (!connmgr_must_output_local(ctx.ofproto->up.connmgr, &ctx.xin->flow,
7040 ctx.xout->odp_actions.data,
7041 ctx.xout->odp_actions.size)) {
7042 compose_output_action(&ctx, OFPP_LOCAL);
7044 if (ctx.ofproto->has_mirrors) {
7045 add_mirror_actions(&ctx, &orig_flow);
7047 fix_sflow_action(&ctx);
7050 ofpbuf_uninit(&ctx.stack);
7053 /* Translates the 'ofpacts_len' bytes of "struct ofpact"s starting at 'ofpacts'
7054 * into datapath actions, using 'ctx', and discards the datapath actions. */
7056 xlate_actions_for_side_effects(struct xlate_in *xin)
7058 struct xlate_out xout;
7060 xlate_actions(xin, &xout);
7061 xlate_out_uninit(&xout);
7065 xlate_report(struct xlate_ctx *ctx, const char *s)
7067 if (ctx->xin->report_hook) {
7068 ctx->xin->report_hook(ctx, s);
7072 /* OFPP_NORMAL implementation. */
7074 static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *);
7076 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
7077 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle',
7078 * the bundle on which the packet was received, returns the VLAN to which the
7081 * Both 'vid' and the return value are in the range 0...4095. */
7083 input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid)
7085 switch (in_bundle->vlan_mode) {
7086 case PORT_VLAN_ACCESS:
7087 return in_bundle->vlan;
7090 case PORT_VLAN_TRUNK:
7093 case PORT_VLAN_NATIVE_UNTAGGED:
7094 case PORT_VLAN_NATIVE_TAGGED:
7095 return vid ? vid : in_bundle->vlan;
7102 /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'.
7103 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
7106 * 'vid' should be the VID obtained from the 802.1Q header that was received as
7107 * part of a packet (specify 0 if there was no 802.1Q header), in the range
7110 input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn)
7112 /* Allow any VID on the OFPP_NONE port. */
7113 if (in_bundle == &ofpp_none_bundle) {
7117 switch (in_bundle->vlan_mode) {
7118 case PORT_VLAN_ACCESS:
7121 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7122 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
7123 "packet received on port %s configured as VLAN "
7124 "%"PRIu16" access port",
7125 in_bundle->ofproto->up.name, vid,
7126 in_bundle->name, in_bundle->vlan);
7132 case PORT_VLAN_NATIVE_UNTAGGED:
7133 case PORT_VLAN_NATIVE_TAGGED:
7135 /* Port must always carry its native VLAN. */
7139 case PORT_VLAN_TRUNK:
7140 if (!ofbundle_includes_vlan(in_bundle, vid)) {
7142 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7143 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet "
7144 "received on port %s not configured for trunking "
7146 in_bundle->ofproto->up.name, vid,
7147 in_bundle->name, vid);
7159 /* Given 'vlan', the VLAN that a packet belongs to, and
7160 * 'out_bundle', a bundle on which the packet is to be output, returns the VID
7161 * that should be included in the 802.1Q header. (If the return value is 0,
7162 * then the 802.1Q header should only be included in the packet if there is a
7165 * Both 'vlan' and the return value are in the range 0...4095. */
7167 output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan)
7169 switch (out_bundle->vlan_mode) {
7170 case PORT_VLAN_ACCESS:
7173 case PORT_VLAN_TRUNK:
7174 case PORT_VLAN_NATIVE_TAGGED:
7177 case PORT_VLAN_NATIVE_UNTAGGED:
7178 return vlan == out_bundle->vlan ? 0 : vlan;
7186 output_normal(struct xlate_ctx *ctx, const struct ofbundle *out_bundle,
7189 struct ofport_dpif *port;
7191 ovs_be16 tci, old_tci;
7193 vid = output_vlan_to_vid(out_bundle, vlan);
7194 if (!out_bundle->bond) {
7195 port = ofbundle_get_a_port(out_bundle);
7197 port = bond_choose_output_slave(out_bundle->bond, &ctx->xin->flow,
7198 vid, &ctx->xout->tags);
7200 /* No slaves enabled, so drop packet. */
7205 old_tci = ctx->xin->flow.vlan_tci;
7207 if (tci || out_bundle->use_priority_tags) {
7208 tci |= ctx->xin->flow.vlan_tci & htons(VLAN_PCP_MASK);
7210 tci |= htons(VLAN_CFI);
7213 ctx->xin->flow.vlan_tci = tci;
7215 compose_output_action(ctx, port->up.ofp_port);
7216 ctx->xin->flow.vlan_tci = old_tci;
7220 mirror_mask_ffs(mirror_mask_t mask)
7222 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
7227 ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
7229 return (bundle->vlan_mode != PORT_VLAN_ACCESS
7230 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
7234 ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan)
7236 return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan);
7239 /* Returns an arbitrary interface within 'bundle'. */
7240 static struct ofport_dpif *
7241 ofbundle_get_a_port(const struct ofbundle *bundle)
7243 return CONTAINER_OF(list_front(&bundle->ports),
7244 struct ofport_dpif, bundle_node);
7248 vlan_is_mirrored(const struct ofmirror *m, int vlan)
7250 return !m->vlans || bitmap_is_set(m->vlans, vlan);
7254 add_mirror_actions(struct xlate_ctx *ctx, const struct flow *orig_flow)
7256 struct ofproto_dpif *ofproto = ctx->ofproto;
7257 mirror_mask_t mirrors;
7258 struct ofbundle *in_bundle;
7261 const struct nlattr *a;
7264 in_bundle = lookup_input_bundle(ctx->ofproto, orig_flow->in_port,
7265 ctx->xin->packet != NULL, NULL);
7269 mirrors = in_bundle->src_mirrors;
7271 /* Drop frames on bundles reserved for mirroring. */
7272 if (in_bundle->mirror_out) {
7273 if (ctx->xin->packet != NULL) {
7274 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7275 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
7276 "%s, which is reserved exclusively for mirroring",
7277 ctx->ofproto->up.name, in_bundle->name);
7283 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
7284 if (!input_vid_is_valid(vid, in_bundle, ctx->xin->packet != NULL)) {
7287 vlan = input_vid_to_vlan(in_bundle, vid);
7289 /* Look at the output ports to check for destination selections. */
7291 NL_ATTR_FOR_EACH (a, left, ctx->xout->odp_actions.data,
7292 ctx->xout->odp_actions.size) {
7293 enum ovs_action_attr type = nl_attr_type(a);
7294 struct ofport_dpif *ofport;
7296 if (type != OVS_ACTION_ATTR_OUTPUT) {
7300 ofport = get_odp_port(ofproto, nl_attr_get_u32(a));
7301 if (ofport && ofport->bundle) {
7302 mirrors |= ofport->bundle->dst_mirrors;
7310 /* Restore the original packet before adding the mirror actions. */
7311 ctx->xin->flow = *orig_flow;
7316 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
7318 if (!vlan_is_mirrored(m, vlan)) {
7319 mirrors = zero_rightmost_1bit(mirrors);
7323 mirrors &= ~m->dup_mirrors;
7324 ctx->xout->mirrors |= m->dup_mirrors;
7326 output_normal(ctx, m->out, vlan);
7327 } else if (vlan != m->out_vlan
7328 && !eth_addr_is_reserved(orig_flow->dl_dst)) {
7329 struct ofbundle *bundle;
7331 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
7332 if (ofbundle_includes_vlan(bundle, m->out_vlan)
7333 && !bundle->mirror_out) {
7334 output_normal(ctx, bundle, m->out_vlan);
7342 update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors,
7343 uint64_t packets, uint64_t bytes)
7349 for (; mirrors; mirrors = zero_rightmost_1bit(mirrors)) {
7352 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
7355 /* In normal circumstances 'm' will not be NULL. However,
7356 * if mirrors are reconfigured, we can temporarily get out
7357 * of sync in facet_revalidate(). We could "correct" the
7358 * mirror list before reaching here, but doing that would
7359 * not properly account the traffic stats we've currently
7360 * accumulated for previous mirror configuration. */
7364 m->packet_count += packets;
7365 m->byte_count += bytes;
7369 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
7370 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
7371 * indicate this; newer upstream kernels use gratuitous ARP requests. */
7373 is_gratuitous_arp(const struct flow *flow)
7375 return (flow->dl_type == htons(ETH_TYPE_ARP)
7376 && eth_addr_is_broadcast(flow->dl_dst)
7377 && (flow->nw_proto == ARP_OP_REPLY
7378 || (flow->nw_proto == ARP_OP_REQUEST
7379 && flow->nw_src == flow->nw_dst)));
7383 update_learning_table(struct ofproto_dpif *ofproto,
7384 const struct flow *flow, int vlan,
7385 struct ofbundle *in_bundle)
7387 struct mac_entry *mac;
7389 /* Don't learn the OFPP_NONE port. */
7390 if (in_bundle == &ofpp_none_bundle) {
7394 if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
7398 mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan);
7399 if (is_gratuitous_arp(flow)) {
7400 /* We don't want to learn from gratuitous ARP packets that are
7401 * reflected back over bond slaves so we lock the learning table. */
7402 if (!in_bundle->bond) {
7403 mac_entry_set_grat_arp_lock(mac);
7404 } else if (mac_entry_is_grat_arp_locked(mac)) {
7409 if (mac_entry_is_new(mac) || mac->port.p != in_bundle) {
7410 /* The log messages here could actually be useful in debugging,
7411 * so keep the rate limit relatively high. */
7412 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
7413 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
7414 "on port %s in VLAN %d",
7415 ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src),
7416 in_bundle->name, vlan);
7418 mac->port.p = in_bundle;
7419 tag_set_add(&ofproto->backer->revalidate_set,
7420 mac_learning_changed(ofproto->ml, mac));
7424 static struct ofbundle *
7425 lookup_input_bundle(const struct ofproto_dpif *ofproto, uint16_t in_port,
7426 bool warn, struct ofport_dpif **in_ofportp)
7428 struct ofport_dpif *ofport;
7430 /* Find the port and bundle for the received packet. */
7431 ofport = get_ofp_port(ofproto, in_port);
7433 *in_ofportp = ofport;
7435 if (ofport && ofport->bundle) {
7436 return ofport->bundle;
7439 /* Special-case OFPP_NONE, which a controller may use as the ingress
7440 * port for traffic that it is sourcing. */
7441 if (in_port == OFPP_NONE) {
7442 return &ofpp_none_bundle;
7445 /* Odd. A few possible reasons here:
7447 * - We deleted a port but there are still a few packets queued up
7450 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
7451 * we don't know about.
7453 * - The ofproto client didn't configure the port as part of a bundle.
7454 * This is particularly likely to happen if a packet was received on the
7455 * port after it was created, but before the client had a chance to
7456 * configure its bundle.
7459 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7461 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
7462 "port %"PRIu16, ofproto->up.name, in_port);
7467 /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or
7468 * dropped. Returns true if they may be forwarded, false if they should be
7471 * 'in_port' must be the ofport_dpif that corresponds to flow->in_port.
7472 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
7474 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
7475 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
7476 * checked by input_vid_is_valid().
7478 * May also add tags to '*tags', although the current implementation only does
7479 * so in one special case.
7482 is_admissible(struct xlate_ctx *ctx, struct ofport_dpif *in_port,
7485 struct ofproto_dpif *ofproto = ctx->ofproto;
7486 struct flow *flow = &ctx->xin->flow;
7487 struct ofbundle *in_bundle = in_port->bundle;
7489 /* Drop frames for reserved multicast addresses
7490 * only if forward_bpdu option is absent. */
7491 if (!ofproto->up.forward_bpdu && eth_addr_is_reserved(flow->dl_dst)) {
7492 xlate_report(ctx, "packet has reserved destination MAC, dropping");
7496 if (in_bundle->bond) {
7497 struct mac_entry *mac;
7499 switch (bond_check_admissibility(in_bundle->bond, in_port,
7500 flow->dl_dst, &ctx->xout->tags)) {
7505 xlate_report(ctx, "bonding refused admissibility, dropping");
7508 case BV_DROP_IF_MOVED:
7509 mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL);
7510 if (mac && mac->port.p != in_bundle &&
7511 (!is_gratuitous_arp(flow)
7512 || mac_entry_is_grat_arp_locked(mac))) {
7513 xlate_report(ctx, "SLB bond thinks this packet looped back, "
7525 xlate_normal(struct xlate_ctx *ctx)
7527 struct ofport_dpif *in_port;
7528 struct ofbundle *in_bundle;
7529 struct mac_entry *mac;
7533 ctx->xout->has_normal = true;
7535 in_bundle = lookup_input_bundle(ctx->ofproto, ctx->xin->flow.in_port,
7536 ctx->xin->packet != NULL, &in_port);
7538 xlate_report(ctx, "no input bundle, dropping");
7542 /* Drop malformed frames. */
7543 if (ctx->xin->flow.dl_type == htons(ETH_TYPE_VLAN) &&
7544 !(ctx->xin->flow.vlan_tci & htons(VLAN_CFI))) {
7545 if (ctx->xin->packet != NULL) {
7546 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7547 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
7548 "VLAN tag received on port %s",
7549 ctx->ofproto->up.name, in_bundle->name);
7551 xlate_report(ctx, "partial VLAN tag, dropping");
7555 /* Drop frames on bundles reserved for mirroring. */
7556 if (in_bundle->mirror_out) {
7557 if (ctx->xin->packet != NULL) {
7558 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7559 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
7560 "%s, which is reserved exclusively for mirroring",
7561 ctx->ofproto->up.name, in_bundle->name);
7563 xlate_report(ctx, "input port is mirror output port, dropping");
7568 vid = vlan_tci_to_vid(ctx->xin->flow.vlan_tci);
7569 if (!input_vid_is_valid(vid, in_bundle, ctx->xin->packet != NULL)) {
7570 xlate_report(ctx, "disallowed VLAN VID for this input port, dropping");
7573 vlan = input_vid_to_vlan(in_bundle, vid);
7575 /* Check other admissibility requirements. */
7576 if (in_port && !is_admissible(ctx, in_port, vlan)) {
7580 /* Learn source MAC. */
7581 if (ctx->xin->may_learn) {
7582 update_learning_table(ctx->ofproto, &ctx->xin->flow, vlan, in_bundle);
7585 /* Determine output bundle. */
7586 mac = mac_learning_lookup(ctx->ofproto->ml, ctx->xin->flow.dl_dst, vlan,
7589 if (mac->port.p != in_bundle) {
7590 xlate_report(ctx, "forwarding to learned port");
7591 output_normal(ctx, mac->port.p, vlan);
7593 xlate_report(ctx, "learned port is input port, dropping");
7596 struct ofbundle *bundle;
7598 xlate_report(ctx, "no learned MAC for destination, flooding");
7599 HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
7600 if (bundle != in_bundle
7601 && ofbundle_includes_vlan(bundle, vlan)
7602 && bundle->floodable
7603 && !bundle->mirror_out) {
7604 output_normal(ctx, bundle, vlan);
7607 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
7611 /* Optimized flow revalidation.
7613 * It's a difficult problem, in general, to tell which facets need to have
7614 * their actions recalculated whenever the OpenFlow flow table changes. We
7615 * don't try to solve that general problem: for most kinds of OpenFlow flow
7616 * table changes, we recalculate the actions for every facet. This is
7617 * relatively expensive, but it's good enough if the OpenFlow flow table
7618 * doesn't change very often.
7620 * However, we can expect one particular kind of OpenFlow flow table change to
7621 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
7622 * of CPU on revalidating every facet whenever MAC learning modifies the flow
7623 * table, we add a special case that applies to flow tables in which every rule
7624 * has the same form (that is, the same wildcards), except that the table is
7625 * also allowed to have a single "catch-all" flow that matches all packets. We
7626 * optimize this case by tagging all of the facets that resubmit into the table
7627 * and invalidating the same tag whenever a flow changes in that table. The
7628 * end result is that we revalidate just the facets that need it (and sometimes
7629 * a few more, but not all of the facets or even all of the facets that
7630 * resubmit to the table modified by MAC learning). */
7632 /* Calculates the tag to use for 'flow' and mask 'mask' when it is inserted
7633 * into an OpenFlow table with the given 'basis'. */
7635 rule_calculate_tag(const struct flow *flow, const struct minimask *mask,
7638 if (minimask_is_catchall(mask)) {
7641 uint32_t hash = flow_hash_in_minimask(flow, mask, secret);
7642 return tag_create_deterministic(hash);
7646 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
7647 * taggability of that table.
7649 * This function must be called after *each* change to a flow table. If you
7650 * skip calling it on some changes then the pointer comparisons at the end can
7651 * be invalid if you get unlucky. For example, if a flow removal causes a
7652 * cls_table to be destroyed and then a flow insertion causes a cls_table with
7653 * different wildcards to be created with the same address, then this function
7654 * will incorrectly skip revalidation. */
7656 table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
7658 struct table_dpif *table = &ofproto->tables[table_id];
7659 const struct oftable *oftable = &ofproto->up.tables[table_id];
7660 struct cls_table *catchall, *other;
7661 struct cls_table *t;
7663 catchall = other = NULL;
7665 switch (hmap_count(&oftable->cls.tables)) {
7667 /* We could tag this OpenFlow table but it would make the logic a
7668 * little harder and it's a corner case that doesn't seem worth it
7674 HMAP_FOR_EACH (t, hmap_node, &oftable->cls.tables) {
7675 if (cls_table_is_catchall(t)) {
7677 } else if (!other) {
7680 /* Indicate that we can't tag this by setting both tables to
7681 * NULL. (We know that 'catchall' is already NULL.) */
7688 /* Can't tag this table. */
7692 if (table->catchall_table != catchall || table->other_table != other) {
7693 table->catchall_table = catchall;
7694 table->other_table = other;
7695 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
7699 /* Given 'rule' that has changed in some way (either it is a rule being
7700 * inserted, a rule being deleted, or a rule whose actions are being
7701 * modified), marks facets for revalidation to ensure that packets will be
7702 * forwarded correctly according to the new state of the flow table.
7704 * This function must be called after *each* change to a flow table. See
7705 * the comment on table_update_taggable() for more information. */
7707 rule_invalidate(const struct rule_dpif *rule)
7709 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
7711 table_update_taggable(ofproto, rule->up.table_id);
7713 if (!ofproto->backer->need_revalidate) {
7714 struct table_dpif *table = &ofproto->tables[rule->up.table_id];
7716 if (table->other_table && rule->tag) {
7717 tag_set_add(&ofproto->backer->revalidate_set, rule->tag);
7719 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
7725 set_frag_handling(struct ofproto *ofproto_,
7726 enum ofp_config_flags frag_handling)
7728 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7729 if (frag_handling != OFPC_FRAG_REASM) {
7730 ofproto->backer->need_revalidate = REV_RECONFIGURE;
7738 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
7739 const struct flow *flow,
7740 const struct ofpact *ofpacts, size_t ofpacts_len)
7742 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7743 struct initial_vals initial_vals;
7744 struct odputil_keybuf keybuf;
7745 struct dpif_flow_stats stats;
7746 struct xlate_out xout;
7747 struct xlate_in xin;
7751 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
7752 odp_flow_key_from_flow(&key, flow,
7753 ofp_port_to_odp_port(ofproto, flow->in_port));
7755 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
7757 initial_vals.vlan_tci = flow->vlan_tci;
7758 xlate_in_init(&xin, ofproto, flow, &initial_vals, NULL, stats.tcp_flags,
7760 xin.resubmit_stats = &stats;
7761 xin.ofpacts_len = ofpacts_len;
7762 xin.ofpacts = ofpacts;
7764 xlate_actions(&xin, &xout);
7765 dpif_execute(ofproto->backer->dpif, key.data, key.size,
7766 xout.odp_actions.data, xout.odp_actions.size, packet);
7767 xlate_out_uninit(&xout);
7775 set_netflow(struct ofproto *ofproto_,
7776 const struct netflow_options *netflow_options)
7778 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7780 if (netflow_options) {
7781 if (!ofproto->netflow) {
7782 ofproto->netflow = netflow_create();
7784 return netflow_set_options(ofproto->netflow, netflow_options);
7786 netflow_destroy(ofproto->netflow);
7787 ofproto->netflow = NULL;
7793 get_netflow_ids(const struct ofproto *ofproto_,
7794 uint8_t *engine_type, uint8_t *engine_id)
7796 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7798 dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
7802 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
7804 if (!facet_is_controller_flow(facet) &&
7805 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
7806 struct subfacet *subfacet;
7807 struct ofexpired expired;
7809 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
7810 if (subfacet->path == SF_FAST_PATH) {
7811 struct dpif_flow_stats stats;
7813 subfacet_install(subfacet, &facet->xout.odp_actions, &stats);
7814 subfacet_update_stats(subfacet, &stats);
7818 expired.flow = facet->flow;
7819 expired.packet_count = facet->packet_count;
7820 expired.byte_count = facet->byte_count;
7821 expired.used = facet->used;
7822 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
7827 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
7829 struct facet *facet;
7831 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
7832 send_active_timeout(ofproto, facet);
7836 static struct ofproto_dpif *
7837 ofproto_dpif_lookup(const char *name)
7839 struct ofproto_dpif *ofproto;
7841 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
7842 hash_string(name, 0), &all_ofproto_dpifs) {
7843 if (!strcmp(ofproto->up.name, name)) {
7851 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
7852 const char *argv[], void *aux OVS_UNUSED)
7854 struct ofproto_dpif *ofproto;
7857 ofproto = ofproto_dpif_lookup(argv[1]);
7859 unixctl_command_reply_error(conn, "no such bridge");
7862 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
7864 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
7865 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
7869 unixctl_command_reply(conn, "table successfully flushed");
7873 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
7874 const char *argv[], void *aux OVS_UNUSED)
7876 struct ds ds = DS_EMPTY_INITIALIZER;
7877 const struct ofproto_dpif *ofproto;
7878 const struct mac_entry *e;
7880 ofproto = ofproto_dpif_lookup(argv[1]);
7882 unixctl_command_reply_error(conn, "no such bridge");
7886 ds_put_cstr(&ds, " port VLAN MAC Age\n");
7887 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
7888 struct ofbundle *bundle = e->port.p;
7889 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
7890 ofbundle_get_a_port(bundle)->odp_port,
7891 e->vlan, ETH_ADDR_ARGS(e->mac),
7892 mac_entry_age(ofproto->ml, e));
7894 unixctl_command_reply(conn, ds_cstr(&ds));
7899 struct xlate_out xout;
7900 struct xlate_in xin;
7906 trace_format_rule(struct ds *result, uint8_t table_id, int level,
7907 const struct rule_dpif *rule)
7909 ds_put_char_multiple(result, '\t', level);
7911 ds_put_cstr(result, "No match\n");
7915 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
7916 table_id, ntohll(rule->up.flow_cookie));
7917 cls_rule_format(&rule->up.cr, result);
7918 ds_put_char(result, '\n');
7920 ds_put_char_multiple(result, '\t', level);
7921 ds_put_cstr(result, "OpenFlow ");
7922 ofpacts_format(rule->up.ofpacts, rule->up.ofpacts_len, result);
7923 ds_put_char(result, '\n');
7927 trace_format_flow(struct ds *result, int level, const char *title,
7928 struct trace_ctx *trace)
7930 ds_put_char_multiple(result, '\t', level);
7931 ds_put_format(result, "%s: ", title);
7932 if (flow_equal(&trace->xin.flow, &trace->flow)) {
7933 ds_put_cstr(result, "unchanged");
7935 flow_format(result, &trace->xin.flow);
7936 trace->flow = trace->xin.flow;
7938 ds_put_char(result, '\n');
7942 trace_format_regs(struct ds *result, int level, const char *title,
7943 struct trace_ctx *trace)
7947 ds_put_char_multiple(result, '\t', level);
7948 ds_put_format(result, "%s:", title);
7949 for (i = 0; i < FLOW_N_REGS; i++) {
7950 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
7952 ds_put_char(result, '\n');
7956 trace_format_odp(struct ds *result, int level, const char *title,
7957 struct trace_ctx *trace)
7959 struct ofpbuf *odp_actions = &trace->xout.odp_actions;
7961 ds_put_char_multiple(result, '\t', level);
7962 ds_put_format(result, "%s: ", title);
7963 format_odp_actions(result, odp_actions->data, odp_actions->size);
7964 ds_put_char(result, '\n');
7968 trace_resubmit(struct xlate_ctx *ctx, struct rule_dpif *rule)
7970 struct trace_ctx *trace = CONTAINER_OF(ctx->xin, struct trace_ctx, xin);
7971 struct ds *result = trace->result;
7973 ds_put_char(result, '\n');
7974 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
7975 trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace);
7976 trace_format_odp(result, ctx->recurse + 1, "Resubmitted odp", trace);
7977 trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule);
7981 trace_report(struct xlate_ctx *ctx, const char *s)
7983 struct trace_ctx *trace = CONTAINER_OF(ctx->xin, struct trace_ctx, xin);
7984 struct ds *result = trace->result;
7986 ds_put_char_multiple(result, '\t', ctx->recurse);
7987 ds_put_cstr(result, s);
7988 ds_put_char(result, '\n');
7992 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
7993 void *aux OVS_UNUSED)
7995 const struct dpif_backer *backer;
7996 struct ofproto_dpif *ofproto;
7997 struct ofpbuf odp_key;
7998 struct ofpbuf *packet;
7999 struct initial_vals initial_vals;
8007 ofpbuf_init(&odp_key, 0);
8009 /* Handle "-generate" or a hex string as the last argument. */
8010 if (!strcmp(argv[argc - 1], "-generate")) {
8011 packet = ofpbuf_new(0);
8014 const char *error = eth_from_hex(argv[argc - 1], &packet);
8017 } else if (argc == 4) {
8018 /* The 3-argument form must end in "-generate' or a hex string. */
8019 unixctl_command_reply_error(conn, error);
8024 /* Parse the flow and determine whether a datapath or
8025 * bridge is specified. If function odp_flow_key_from_string()
8026 * returns 0, the flow is a odp_flow. If function
8027 * parse_ofp_exact_flow() returns 0, the flow is a br_flow. */
8028 if (!odp_flow_key_from_string(argv[argc - 1], NULL, &odp_key)) {
8029 /* If the odp_flow is the second argument,
8030 * the datapath name is the first argument. */
8032 const char *dp_type;
8033 if (!strncmp(argv[1], "ovs-", 4)) {
8034 dp_type = argv[1] + 4;
8038 backer = shash_find_data(&all_dpif_backers, dp_type);
8040 unixctl_command_reply_error(conn, "Cannot find datapath "
8045 /* No datapath name specified, so there should be only one
8047 struct shash_node *node;
8048 if (shash_count(&all_dpif_backers) != 1) {
8049 unixctl_command_reply_error(conn, "Must specify datapath "
8050 "name, there is more than one type of datapath");
8053 node = shash_first(&all_dpif_backers);
8054 backer = node->data;
8057 /* Extract the ofproto_dpif object from the ofproto_receive()
8059 if (ofproto_receive(backer, NULL, odp_key.data,
8060 odp_key.size, &flow, NULL, &ofproto, NULL,
8062 unixctl_command_reply_error(conn, "Invalid datapath flow");
8065 ds_put_format(&result, "Bridge: %s\n", ofproto->up.name);
8066 } else if (!parse_ofp_exact_flow(&flow, argv[argc - 1])) {
8068 unixctl_command_reply_error(conn, "Must specify bridge name");
8072 ofproto = ofproto_dpif_lookup(argv[1]);
8074 unixctl_command_reply_error(conn, "Unknown bridge name");
8077 initial_vals.vlan_tci = flow.vlan_tci;
8079 unixctl_command_reply_error(conn, "Bad flow syntax");
8083 /* Generate a packet, if requested. */
8085 if (!packet->size) {
8086 flow_compose(packet, &flow);
8088 ds_put_cstr(&result, "Packet: ");
8089 s = ofp_packet_to_string(packet->data, packet->size);
8090 ds_put_cstr(&result, s);
8093 /* Use the metadata from the flow and the packet argument
8094 * to reconstruct the flow. */
8095 flow_extract(packet, flow.skb_priority, flow.skb_mark, NULL,
8096 flow.in_port, &flow);
8097 initial_vals.vlan_tci = flow.vlan_tci;
8101 ofproto_trace(ofproto, &flow, packet, &initial_vals, &result);
8102 unixctl_command_reply(conn, ds_cstr(&result));
8105 ds_destroy(&result);
8106 ofpbuf_delete(packet);
8107 ofpbuf_uninit(&odp_key);
8111 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
8112 const struct ofpbuf *packet,
8113 const struct initial_vals *initial_vals, struct ds *ds)
8115 struct rule_dpif *rule;
8117 ds_put_cstr(ds, "Flow: ");
8118 flow_format(ds, flow);
8119 ds_put_char(ds, '\n');
8121 rule = rule_dpif_lookup(ofproto, flow);
8123 trace_format_rule(ds, 0, 0, rule);
8124 if (rule == ofproto->miss_rule) {
8125 ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
8126 } else if (rule == ofproto->no_packet_in_rule) {
8127 ds_put_cstr(ds, "\nNo match, packets dropped because "
8128 "OFPPC_NO_PACKET_IN is set on in_port.\n");
8132 uint64_t odp_actions_stub[1024 / 8];
8133 struct ofpbuf odp_actions;
8135 struct trace_ctx trace;
8138 tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
8141 ofpbuf_use_stub(&odp_actions,
8142 odp_actions_stub, sizeof odp_actions_stub);
8143 xlate_in_init(&trace.xin, ofproto, flow, initial_vals, rule, tcp_flags,
8145 trace.xin.resubmit_hook = trace_resubmit;
8146 trace.xin.report_hook = trace_report;
8147 xlate_actions(&trace.xin, &trace.xout);
8149 ds_put_char(ds, '\n');
8150 trace_format_flow(ds, 0, "Final flow", &trace);
8151 ds_put_cstr(ds, "Datapath actions: ");
8152 format_odp_actions(ds, trace.xout.odp_actions.data,
8153 trace.xout.odp_actions.size);
8155 if (trace.xout.slow) {
8156 ds_put_cstr(ds, "\nThis flow is handled by the userspace "
8157 "slow path because it:");
8158 switch (trace.xout.slow) {
8160 ds_put_cstr(ds, "\n\t- Consists of CFM packets.");
8163 ds_put_cstr(ds, "\n\t- Consists of LACP packets.");
8166 ds_put_cstr(ds, "\n\t- Consists of STP packets.");
8169 ds_put_cstr(ds, "\n\t- Consists of BFD packets.");
8171 case SLOW_CONTROLLER:
8172 ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages "
8173 "to the OpenFlow controller.");
8180 xlate_out_uninit(&trace.xout);
8185 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
8186 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
8189 unixctl_command_reply(conn, NULL);
8193 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
8194 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
8197 unixctl_command_reply(conn, NULL);
8200 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
8201 * 'reply' describing the results. */
8203 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
8205 struct facet *facet;
8209 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
8210 if (!facet_check_consistency(facet)) {
8215 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
8219 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
8220 ofproto->up.name, errors);
8222 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
8227 ofproto_dpif_self_check(struct unixctl_conn *conn,
8228 int argc, const char *argv[], void *aux OVS_UNUSED)
8230 struct ds reply = DS_EMPTY_INITIALIZER;
8231 struct ofproto_dpif *ofproto;
8234 ofproto = ofproto_dpif_lookup(argv[1]);
8236 unixctl_command_reply_error(conn, "Unknown ofproto (use "
8237 "ofproto/list for help)");
8240 ofproto_dpif_self_check__(ofproto, &reply);
8242 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
8243 ofproto_dpif_self_check__(ofproto, &reply);
8247 unixctl_command_reply(conn, ds_cstr(&reply));
8251 /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list
8252 * of the 'ofproto_shash' nodes. It is the responsibility of the caller
8253 * to destroy 'ofproto_shash' and free the returned value. */
8254 static const struct shash_node **
8255 get_ofprotos(struct shash *ofproto_shash)
8257 const struct ofproto_dpif *ofproto;
8259 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
8260 char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
8261 shash_add_nocopy(ofproto_shash, name, ofproto);
8264 return shash_sort(ofproto_shash);
8268 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
8269 const char *argv[] OVS_UNUSED,
8270 void *aux OVS_UNUSED)
8272 struct ds ds = DS_EMPTY_INITIALIZER;
8273 struct shash ofproto_shash;
8274 const struct shash_node **sorted_ofprotos;
8277 shash_init(&ofproto_shash);
8278 sorted_ofprotos = get_ofprotos(&ofproto_shash);
8279 for (i = 0; i < shash_count(&ofproto_shash); i++) {
8280 const struct shash_node *node = sorted_ofprotos[i];
8281 ds_put_format(&ds, "%s\n", node->name);
8284 shash_destroy(&ofproto_shash);
8285 free(sorted_ofprotos);
8287 unixctl_command_reply(conn, ds_cstr(&ds));
8292 show_dp_format(const struct ofproto_dpif *ofproto, struct ds *ds)
8294 const struct shash_node **ports;
8296 struct avg_subfacet_rates lifetime;
8297 unsigned long long int minutes;
8298 const int min_ms = 60 * 1000; /* milliseconds in one minute. */
8300 minutes = (time_msec() - ofproto->created) / min_ms;
8303 lifetime.add_rate = (double)ofproto->total_subfacet_add_count
8305 lifetime.del_rate = (double)ofproto->total_subfacet_del_count
8308 lifetime.add_rate = 0.0;
8309 lifetime.del_rate = 0.0;
8312 ds_put_format(ds, "%s (%s):\n", ofproto->up.name,
8313 dpif_name(ofproto->backer->dpif));
8315 "\tlookups: hit:%"PRIu64" missed:%"PRIu64"\n",
8316 ofproto->n_hit, ofproto->n_missed);
8317 ds_put_format(ds, "\tflows: cur: %zu, avg: %5.3f, max: %d,"
8318 " life span: %llu(ms)\n",
8319 hmap_count(&ofproto->subfacets),
8320 avg_subfacet_count(ofproto),
8321 ofproto->max_n_subfacet,
8322 avg_subfacet_life_span(ofproto));
8323 if (minutes >= 60) {
8324 show_dp_rates(ds, "\t\thourly avg:", &ofproto->hourly);
8326 if (minutes >= 60 * 24) {
8327 show_dp_rates(ds, "\t\tdaily avg:", &ofproto->daily);
8329 show_dp_rates(ds, "\t\toverall avg:", &lifetime);
8331 ports = shash_sort(&ofproto->up.port_by_name);
8332 for (i = 0; i < shash_count(&ofproto->up.port_by_name); i++) {
8333 const struct shash_node *node = ports[i];
8334 struct ofport *ofport = node->data;
8335 const char *name = netdev_get_name(ofport->netdev);
8336 const char *type = netdev_get_type(ofport->netdev);
8339 ds_put_format(ds, "\t%s %u/", name, ofport->ofp_port);
8341 odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
8342 if (odp_port != OVSP_NONE) {
8343 ds_put_format(ds, "%"PRIu32":", odp_port);
8345 ds_put_cstr(ds, "none:");
8348 if (strcmp(type, "system")) {
8349 struct netdev *netdev;
8352 ds_put_format(ds, " (%s", type);
8354 error = netdev_open(name, type, &netdev);
8359 error = netdev_get_config(netdev, &config);
8361 const struct smap_node **nodes;
8364 nodes = smap_sort(&config);
8365 for (i = 0; i < smap_count(&config); i++) {
8366 const struct smap_node *node = nodes[i];
8367 ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
8368 node->key, node->value);
8372 smap_destroy(&config);
8374 netdev_close(netdev);
8376 ds_put_char(ds, ')');
8378 ds_put_char(ds, '\n');
8384 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc,
8385 const char *argv[], void *aux OVS_UNUSED)
8387 struct ds ds = DS_EMPTY_INITIALIZER;
8388 const struct ofproto_dpif *ofproto;
8392 for (i = 1; i < argc; i++) {
8393 ofproto = ofproto_dpif_lookup(argv[i]);
8395 ds_put_format(&ds, "Unknown bridge %s (use dpif/dump-dps "
8396 "for help)", argv[i]);
8397 unixctl_command_reply_error(conn, ds_cstr(&ds));
8400 show_dp_format(ofproto, &ds);
8403 struct shash ofproto_shash;
8404 const struct shash_node **sorted_ofprotos;
8407 shash_init(&ofproto_shash);
8408 sorted_ofprotos = get_ofprotos(&ofproto_shash);
8409 for (i = 0; i < shash_count(&ofproto_shash); i++) {
8410 const struct shash_node *node = sorted_ofprotos[i];
8411 show_dp_format(node->data, &ds);
8414 shash_destroy(&ofproto_shash);
8415 free(sorted_ofprotos);
8418 unixctl_command_reply(conn, ds_cstr(&ds));
8423 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
8424 int argc OVS_UNUSED, const char *argv[],
8425 void *aux OVS_UNUSED)
8427 struct ds ds = DS_EMPTY_INITIALIZER;
8428 const struct ofproto_dpif *ofproto;
8429 struct subfacet *subfacet;
8431 ofproto = ofproto_dpif_lookup(argv[1]);
8433 unixctl_command_reply_error(conn, "no such bridge");
8437 update_stats(ofproto->backer);
8439 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
8440 struct facet *facet = subfacet->facet;
8442 odp_flow_key_format(subfacet->key, subfacet->key_len, &ds);
8444 ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:",
8445 subfacet->dp_packet_count, subfacet->dp_byte_count);
8446 if (subfacet->used) {
8447 ds_put_format(&ds, "%.3fs",
8448 (time_msec() - subfacet->used) / 1000.0);
8450 ds_put_format(&ds, "never");
8452 if (subfacet->facet->tcp_flags) {
8453 ds_put_cstr(&ds, ", flags:");
8454 packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags);
8457 ds_put_cstr(&ds, ", actions:");
8458 if (facet->xout.slow) {
8459 uint64_t slow_path_stub[128 / 8];
8460 const struct nlattr *actions;
8463 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
8464 slow_path_stub, sizeof slow_path_stub,
8465 &actions, &actions_len);
8466 format_odp_actions(&ds, actions, actions_len);
8468 format_odp_actions(&ds, facet->xout.odp_actions.data,
8469 facet->xout.odp_actions.size);
8471 ds_put_char(&ds, '\n');
8474 unixctl_command_reply(conn, ds_cstr(&ds));
8479 ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn,
8480 int argc OVS_UNUSED, const char *argv[],
8481 void *aux OVS_UNUSED)
8483 struct ds ds = DS_EMPTY_INITIALIZER;
8484 struct ofproto_dpif *ofproto;
8486 ofproto = ofproto_dpif_lookup(argv[1]);
8488 unixctl_command_reply_error(conn, "no such bridge");
8492 flush(&ofproto->up);
8494 unixctl_command_reply(conn, ds_cstr(&ds));
8499 ofproto_dpif_unixctl_init(void)
8501 static bool registered;
8507 unixctl_command_register(
8509 "[dp_name]|bridge odp_flow|br_flow [-generate|packet]",
8510 1, 3, ofproto_unixctl_trace, NULL);
8511 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
8512 ofproto_unixctl_fdb_flush, NULL);
8513 unixctl_command_register("fdb/show", "bridge", 1, 1,
8514 ofproto_unixctl_fdb_show, NULL);
8515 unixctl_command_register("ofproto/clog", "", 0, 0,
8516 ofproto_dpif_clog, NULL);
8517 unixctl_command_register("ofproto/unclog", "", 0, 0,
8518 ofproto_dpif_unclog, NULL);
8519 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
8520 ofproto_dpif_self_check, NULL);
8521 unixctl_command_register("dpif/dump-dps", "", 0, 0,
8522 ofproto_unixctl_dpif_dump_dps, NULL);
8523 unixctl_command_register("dpif/show", "[bridge]", 0, INT_MAX,
8524 ofproto_unixctl_dpif_show, NULL);
8525 unixctl_command_register("dpif/dump-flows", "bridge", 1, 1,
8526 ofproto_unixctl_dpif_dump_flows, NULL);
8527 unixctl_command_register("dpif/del-flows", "bridge", 1, 1,
8528 ofproto_unixctl_dpif_del_flows, NULL);
8531 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
8533 * This is deprecated. It is only for compatibility with broken device drivers
8534 * in old versions of Linux that do not properly support VLANs when VLAN
8535 * devices are not used. When broken device drivers are no longer in
8536 * widespread use, we will delete these interfaces. */
8539 set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid)
8541 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
8542 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
8544 if (realdev_ofp_port == ofport->realdev_ofp_port
8545 && vid == ofport->vlandev_vid) {
8549 ofproto->backer->need_revalidate = REV_RECONFIGURE;
8551 if (ofport->realdev_ofp_port) {
8554 if (realdev_ofp_port && ofport->bundle) {
8555 /* vlandevs are enslaved to their realdevs, so they are not allowed to
8556 * themselves be part of a bundle. */
8557 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
8560 ofport->realdev_ofp_port = realdev_ofp_port;
8561 ofport->vlandev_vid = vid;
8563 if (realdev_ofp_port) {
8564 vsp_add(ofport, realdev_ofp_port, vid);
8571 hash_realdev_vid(uint16_t realdev_ofp_port, int vid)
8573 return hash_2words(realdev_ofp_port, vid);
8576 /* Returns the ODP port number of the Linux VLAN device that corresponds to
8577 * 'vlan_tci' on the network device with port number 'realdev_odp_port' in
8578 * 'ofproto'. For example, given 'realdev_odp_port' of eth0 and 'vlan_tci' 9,
8579 * it would return the port number of eth0.9.
8581 * Unless VLAN splinters are enabled for port 'realdev_odp_port', this
8582 * function just returns its 'realdev_odp_port' argument. */
8584 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
8585 uint32_t realdev_odp_port, ovs_be16 vlan_tci)
8587 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
8588 uint16_t realdev_ofp_port;
8589 int vid = vlan_tci_to_vid(vlan_tci);
8590 const struct vlan_splinter *vsp;
8592 realdev_ofp_port = odp_port_to_ofp_port(ofproto, realdev_odp_port);
8593 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
8594 hash_realdev_vid(realdev_ofp_port, vid),
8595 &ofproto->realdev_vid_map) {
8596 if (vsp->realdev_ofp_port == realdev_ofp_port
8597 && vsp->vid == vid) {
8598 return ofp_port_to_odp_port(ofproto, vsp->vlandev_ofp_port);
8602 return realdev_odp_port;
8605 static struct vlan_splinter *
8606 vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port)
8608 struct vlan_splinter *vsp;
8610 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, hash_int(vlandev_ofp_port, 0),
8611 &ofproto->vlandev_map) {
8612 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
8620 /* Returns the OpenFlow port number of the "real" device underlying the Linux
8621 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
8622 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
8623 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
8624 * eth0 and store 9 in '*vid'.
8626 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
8627 * VLAN device. Unless VLAN splinters are enabled, this is what this function
8630 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
8631 uint16_t vlandev_ofp_port, int *vid)
8633 if (!hmap_is_empty(&ofproto->vlandev_map)) {
8634 const struct vlan_splinter *vsp;
8636 vsp = vlandev_find(ofproto, vlandev_ofp_port);
8641 return vsp->realdev_ofp_port;
8647 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
8648 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
8649 * 'flow->in_port' to the "real" device backing the VLAN device, sets
8650 * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is
8651 * always the case unless VLAN splinters are enabled), returns false without
8652 * making any changes. */
8654 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
8659 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port, &vid);
8664 /* Cause the flow to be processed as if it came in on the real device with
8665 * the VLAN device's VLAN ID. */
8666 flow->in_port = realdev;
8667 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
8672 vsp_remove(struct ofport_dpif *port)
8674 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
8675 struct vlan_splinter *vsp;
8677 vsp = vlandev_find(ofproto, port->up.ofp_port);
8679 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
8680 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
8683 port->realdev_ofp_port = 0;
8685 VLOG_ERR("missing vlan device record");
8690 vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid)
8692 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
8694 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
8695 && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid))
8696 == realdev_ofp_port)) {
8697 struct vlan_splinter *vsp;
8699 vsp = xmalloc(sizeof *vsp);
8700 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
8701 hash_int(port->up.ofp_port, 0));
8702 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
8703 hash_realdev_vid(realdev_ofp_port, vid));
8704 vsp->realdev_ofp_port = realdev_ofp_port;
8705 vsp->vlandev_ofp_port = port->up.ofp_port;
8708 port->realdev_ofp_port = realdev_ofp_port;
8710 VLOG_ERR("duplicate vlan device record");
8715 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
8717 const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
8718 return ofport ? ofport->odp_port : OVSP_NONE;
8721 static struct ofport_dpif *
8722 odp_port_to_ofport(const struct dpif_backer *backer, uint32_t odp_port)
8724 struct ofport_dpif *port;
8726 HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node,
8727 hash_int(odp_port, 0),
8728 &backer->odp_to_ofport_map) {
8729 if (port->odp_port == odp_port) {
8738 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
8740 struct ofport_dpif *port;
8742 port = odp_port_to_ofport(ofproto->backer, odp_port);
8743 if (port && &ofproto->up == port->up.ofproto) {
8744 return port->up.ofp_port;
8749 static unsigned long long int
8750 avg_subfacet_life_span(const struct ofproto_dpif *ofproto)
8752 unsigned long long int dc;
8753 unsigned long long int avg;
8755 dc = ofproto->total_subfacet_del_count + ofproto->subfacet_del_count;
8756 avg = dc ? ofproto->total_subfacet_life_span / dc : 0;
8762 avg_subfacet_count(const struct ofproto_dpif *ofproto)
8766 if (ofproto->n_update_stats) {
8767 avg_c = (double)ofproto->total_subfacet_count
8768 / ofproto->n_update_stats;
8775 show_dp_rates(struct ds *ds, const char *heading,
8776 const struct avg_subfacet_rates *rates)
8778 ds_put_format(ds, "%s add rate: %5.3f/min, del rate: %5.3f/min\n",
8779 heading, rates->add_rate, rates->del_rate);
8783 update_max_subfacet_count(struct ofproto_dpif *ofproto)
8785 ofproto->max_n_subfacet = MAX(ofproto->max_n_subfacet,
8786 hmap_count(&ofproto->subfacets));
8789 /* Compute exponentially weighted moving average, adding 'new' as the newest,
8790 * most heavily weighted element. 'base' designates the rate of decay: after
8791 * 'base' further updates, 'new''s weight in the EWMA decays to about 1/e
8794 exp_mavg(double *avg, int base, double new)
8796 *avg = (*avg * (base - 1) + new) / base;
8800 update_moving_averages(struct ofproto_dpif *ofproto)
8802 const int min_ms = 60 * 1000; /* milliseconds in one minute. */
8804 /* Update hourly averages on the minute boundaries. */
8805 if (time_msec() - ofproto->last_minute >= min_ms) {
8806 exp_mavg(&ofproto->hourly.add_rate, 60, ofproto->subfacet_add_count);
8807 exp_mavg(&ofproto->hourly.del_rate, 60, ofproto->subfacet_del_count);
8809 /* Update daily averages on the hour boundaries. */
8810 if ((ofproto->last_minute - ofproto->created) / min_ms % 60 == 59) {
8811 exp_mavg(&ofproto->daily.add_rate, 24, ofproto->hourly.add_rate);
8812 exp_mavg(&ofproto->daily.del_rate, 24, ofproto->hourly.del_rate);
8815 ofproto->total_subfacet_add_count += ofproto->subfacet_add_count;
8816 ofproto->total_subfacet_del_count += ofproto->subfacet_del_count;
8817 ofproto->subfacet_add_count = 0;
8818 ofproto->subfacet_del_count = 0;
8819 ofproto->last_minute += min_ms;
8824 dpif_stats_update_hit_count(struct ofproto_dpif *ofproto, uint64_t delta)
8826 ofproto->n_hit += delta;
8829 const struct ofproto_class ofproto_dpif_class = {
8864 port_is_lacp_current,
8865 NULL, /* rule_choose_table */
8872 rule_modify_actions,
8886 get_stp_port_status,
8893 is_mirror_output_bundle,
8894 forward_bpdu_changed,
8895 set_mac_table_config,