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"
25 #include "byte-order.h"
30 #include "dynamic-string.h"
31 #include "fail-open.h"
35 #include "mac-learning.h"
36 #include "meta-flow.h"
37 #include "multipath.h"
38 #include "netdev-vport.h"
45 #include "ofp-actions.h"
46 #include "ofp-parse.h"
47 #include "ofp-print.h"
48 #include "ofproto-dpif-governor.h"
49 #include "ofproto-dpif-ipfix.h"
50 #include "ofproto-dpif-sflow.h"
51 #include "poll-loop.h"
56 #include "unaligned.h"
58 #include "vlan-bitmap.h"
61 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
63 COVERAGE_DEFINE(ofproto_dpif_expired);
64 COVERAGE_DEFINE(ofproto_dpif_xlate);
65 COVERAGE_DEFINE(facet_changed_rule);
66 COVERAGE_DEFINE(facet_revalidate);
67 COVERAGE_DEFINE(facet_unexpected);
68 COVERAGE_DEFINE(facet_suppress);
70 /* Maximum depth of flow table recursion (due to resubmit actions) in a
71 * flow translation. */
72 #define MAX_RESUBMIT_RECURSION 64
74 /* Number of implemented OpenFlow tables. */
75 enum { N_TABLES = 255 };
76 enum { TBL_INTERNAL = N_TABLES - 1 }; /* Used for internal hidden rules. */
77 BUILD_ASSERT_DECL(N_TABLES >= 2 && N_TABLES <= 255);
89 * - Do include packets and bytes from facets that have been deleted or
90 * whose own statistics have been folded into the rule.
92 * - Do include packets and bytes sent "by hand" that were accounted to
93 * the rule without any facet being involved (this is a rare corner
94 * case in rule_execute()).
96 * - Do not include packet or bytes that can be obtained from any facet's
97 * packet_count or byte_count member or that can be obtained from the
98 * datapath by, e.g., dpif_flow_get() for any subfacet.
100 uint64_t packet_count; /* Number of packets received. */
101 uint64_t byte_count; /* Number of bytes received. */
103 tag_type tag; /* Caches rule_calculate_tag() result. */
105 struct list facets; /* List of "struct facet"s. */
108 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
110 return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
113 static struct rule_dpif *rule_dpif_lookup(struct ofproto_dpif *,
114 const struct flow *);
115 static struct rule_dpif *rule_dpif_lookup__(struct ofproto_dpif *,
118 static struct rule_dpif *rule_dpif_miss_rule(struct ofproto_dpif *ofproto,
119 const struct flow *flow);
121 static void rule_credit_stats(struct rule_dpif *,
122 const struct dpif_flow_stats *);
123 static void flow_push_stats(struct facet *, const struct dpif_flow_stats *);
124 static tag_type rule_calculate_tag(const struct flow *,
125 const struct minimask *, uint32_t basis);
126 static void rule_invalidate(const struct rule_dpif *);
128 #define MAX_MIRRORS 32
129 typedef uint32_t mirror_mask_t;
130 #define MIRROR_MASK_C(X) UINT32_C(X)
131 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
133 struct ofproto_dpif *ofproto; /* Owning ofproto. */
134 size_t idx; /* In ofproto's "mirrors" array. */
135 void *aux; /* Key supplied by ofproto's client. */
136 char *name; /* Identifier for log messages. */
138 /* Selection criteria. */
139 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
140 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
141 unsigned long *vlans; /* Bitmap of chosen VLANs, NULL selects all. */
143 /* Output (exactly one of out == NULL and out_vlan == -1 is true). */
144 struct ofbundle *out; /* Output port or NULL. */
145 int out_vlan; /* Output VLAN or -1. */
146 mirror_mask_t dup_mirrors; /* Bitmap of mirrors with the same output. */
149 int64_t packet_count; /* Number of packets sent. */
150 int64_t byte_count; /* Number of bytes sent. */
153 static void mirror_destroy(struct ofmirror *);
154 static void update_mirror_stats(struct ofproto_dpif *ofproto,
155 mirror_mask_t mirrors,
156 uint64_t packets, uint64_t bytes);
159 struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
160 struct ofproto_dpif *ofproto; /* Owning ofproto. */
161 void *aux; /* Key supplied by ofproto's client. */
162 char *name; /* Identifier for log messages. */
165 struct list ports; /* Contains "struct ofport"s. */
166 enum port_vlan_mode vlan_mode; /* VLAN mode */
167 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
168 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
169 * NULL if all VLANs are trunked. */
170 struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */
171 struct bond *bond; /* Nonnull iff more than one port. */
172 bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */
175 bool floodable; /* True if no port has OFPUTIL_PC_NO_FLOOD set. */
177 /* Port mirroring info. */
178 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
179 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
180 mirror_mask_t mirror_out; /* Mirrors that output to this bundle. */
183 static void bundle_remove(struct ofport *);
184 static void bundle_update(struct ofbundle *);
185 static void bundle_destroy(struct ofbundle *);
186 static void bundle_del_port(struct ofport_dpif *);
187 static void bundle_run(struct ofbundle *);
188 static void bundle_wait(struct ofbundle *);
189 static struct ofbundle *lookup_input_bundle(const struct ofproto_dpif *,
190 uint16_t in_port, bool warn,
191 struct ofport_dpif **in_ofportp);
193 /* A controller may use OFPP_NONE as the ingress port to indicate that
194 * it did not arrive on a "real" port. 'ofpp_none_bundle' exists for
195 * when an input bundle is needed for validation (e.g., mirroring or
196 * OFPP_NORMAL processing). It is not connected to an 'ofproto' or have
197 * any 'port' structs, so care must be taken when dealing with it. */
198 static struct ofbundle ofpp_none_bundle = {
200 .vlan_mode = PORT_VLAN_TRUNK
203 static void stp_run(struct ofproto_dpif *ofproto);
204 static void stp_wait(struct ofproto_dpif *ofproto);
205 static int set_stp_port(struct ofport *,
206 const struct ofproto_port_stp_settings *);
208 static bool ofbundle_includes_vlan(const struct ofbundle *, uint16_t vlan);
212 /* Initial values of fields of the packet that may be changed during
213 * flow processing and needed later. */
214 struct initial_vals {
215 /* This is the value of vlan_tci in the packet as actually received from
216 * dpif. This is the same as the facet's flow.vlan_tci unless the packet
217 * was received via a VLAN splinter. In that case, this value is 0
218 * (because the packet as actually received from the dpif had no 802.1Q
219 * tag) but the facet's flow.vlan_tci is set to the VLAN that the splinter
222 * This member should be removed when the VLAN splinters feature is no
226 /* If received on a tunnel, the IP TOS value of the tunnel. */
227 uint8_t tunnel_ip_tos;
231 tag_type tags; /* Tags associated with actions. */
232 enum slow_path_reason slow; /* 0 if fast path may be used. */
233 bool has_learn; /* Actions include NXAST_LEARN? */
234 bool has_normal; /* Actions output to OFPP_NORMAL? */
235 bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */
236 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
237 mirror_mask_t mirrors; /* Bitmap of associated mirrors. */
239 uint64_t odp_actions_stub[256 / 8];
240 struct ofpbuf odp_actions;
244 struct ofproto_dpif *ofproto;
246 /* Flow to which the OpenFlow actions apply. xlate_actions() will modify
247 * this flow when actions change header fields. */
250 struct initial_vals initial_vals;
252 /* The packet corresponding to 'flow', or a null pointer if we are
253 * revalidating without a packet to refer to. */
254 const struct ofpbuf *packet;
256 /* Should OFPP_NORMAL update the MAC learning table? Should "learn"
257 * actions update the flow table?
259 * We want to update these tables if we are actually processing a packet,
260 * or if we are accounting for packets that the datapath has processed, but
261 * not if we are just revalidating. */
264 /* The rule initiating translation or NULL. */
265 struct rule_dpif *rule;
267 /* The actions to translate. If 'rule' is not NULL, these may be NULL. */
268 const struct ofpact *ofpacts;
271 /* Union of the set of TCP flags seen so far in this flow. (Used only by
272 * NXAST_FIN_TIMEOUT. Set to zero to avoid updating updating rules'
276 /* If nonnull, flow translation calls this function just before executing a
277 * resubmit or OFPP_TABLE action. In addition, disables logging of traces
278 * when the recursion depth is exceeded.
280 * 'rule' is the rule being submitted into. It will be null if the
281 * resubmit or OFPP_TABLE action didn't find a matching rule.
283 * This is normally null so the client has to set it manually after
284 * calling xlate_in_init(). */
285 void (*resubmit_hook)(struct xlate_ctx *, struct rule_dpif *rule);
287 /* If nonnull, flow translation calls this function to report some
288 * significant decision, e.g. to explain why OFPP_NORMAL translation
289 * dropped a packet. */
290 void (*report_hook)(struct xlate_ctx *, const char *s);
292 /* If nonnull, flow translation credits the specified statistics to each
293 * rule reached through a resubmit or OFPP_TABLE action.
295 * This is normally null so the client has to set it manually after
296 * calling xlate_in_init(). */
297 const struct dpif_flow_stats *resubmit_stats;
300 /* Context used by xlate_actions() and its callees. */
302 struct xlate_in *xin;
303 struct xlate_out *xout;
305 struct ofproto_dpif *ofproto;
307 /* Flow at the last commit. */
308 struct flow base_flow;
310 /* Stack for the push and pop actions. Each stack element is of type
311 * "union mf_subvalue". */
312 union mf_subvalue init_stack[1024 / sizeof(union mf_subvalue)];
315 /* The rule that we are currently translating, or NULL. */
316 struct rule_dpif *rule;
318 int recurse; /* Recursion level, via xlate_table_action. */
319 bool max_resubmit_trigger; /* Recursed too deeply during translation. */
320 uint32_t orig_skb_priority; /* Priority when packet arrived. */
321 uint8_t table_id; /* OpenFlow table ID where flow was found. */
322 uint32_t sflow_n_outputs; /* Number of output ports. */
323 uint32_t sflow_odp_port; /* Output port for composing sFlow action. */
324 uint16_t user_cookie_offset;/* Used for user_action_cookie fixup. */
325 bool exit; /* No further actions should be processed. */
328 static void xlate_in_init(struct xlate_in *, struct ofproto_dpif *,
329 const struct flow *, const struct initial_vals *,
330 struct rule_dpif *, uint8_t tcp_flags,
331 const struct ofpbuf *);
333 static void xlate_out_uninit(struct xlate_out *);
335 static void xlate_actions(struct xlate_in *, struct xlate_out *);
337 static void xlate_actions_for_side_effects(struct xlate_in *);
339 static void xlate_table_action(struct xlate_ctx *, uint16_t in_port,
340 uint8_t table_id, bool may_packet_in);
342 static size_t put_userspace_action(const struct ofproto_dpif *,
343 struct ofpbuf *odp_actions,
345 const union user_action_cookie *,
348 static void compose_slow_path(const struct ofproto_dpif *, const struct flow *,
349 enum slow_path_reason,
350 uint64_t *stub, size_t stub_size,
351 const struct nlattr **actionsp,
352 size_t *actions_lenp);
354 static void xlate_report(struct xlate_ctx *ctx, const char *s);
356 /* A subfacet (see "struct subfacet" below) has three possible installation
359 * - SF_NOT_INSTALLED: Not installed in the datapath. This will only be the
360 * case just after the subfacet is created, just before the subfacet is
361 * destroyed, or if the datapath returns an error when we try to install a
364 * - SF_FAST_PATH: The subfacet's actions are installed in the datapath.
366 * - SF_SLOW_PATH: An action that sends every packet for the subfacet through
367 * ofproto_dpif is installed in the datapath.
370 SF_NOT_INSTALLED, /* No datapath flow for this subfacet. */
371 SF_FAST_PATH, /* Full actions are installed. */
372 SF_SLOW_PATH, /* Send-to-userspace action is installed. */
375 /* A dpif flow and actions associated with a facet.
377 * See also the large comment on struct facet. */
380 struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */
381 struct list list_node; /* In struct facet's 'facets' list. */
382 struct facet *facet; /* Owning facet. */
384 enum odp_key_fitness key_fitness;
388 long long int used; /* Time last used; time created if not used. */
389 long long int created; /* Time created. */
391 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
392 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
394 enum subfacet_path path; /* Installed in datapath? */
396 /* Datapath port the packet arrived on. This is needed to remove
397 * flows for ports that are no longer part of the bridge. Since the
398 * flow definition only has the OpenFlow port number and the port is
399 * no longer part of the bridge, we can't determine the datapath port
400 * number needed to delete the flow from the datapath. */
401 uint32_t odp_in_port;
404 #define SUBFACET_DESTROY_MAX_BATCH 50
406 static struct subfacet *subfacet_create(struct facet *, struct flow_miss *miss,
408 static struct subfacet *subfacet_find(struct ofproto_dpif *,
409 const struct nlattr *key, size_t key_len,
411 static void subfacet_destroy(struct subfacet *);
412 static void subfacet_destroy__(struct subfacet *);
413 static void subfacet_destroy_batch(struct ofproto_dpif *,
414 struct subfacet **, int n);
415 static void subfacet_reset_dp_stats(struct subfacet *,
416 struct dpif_flow_stats *);
417 static void subfacet_update_time(struct subfacet *, long long int used);
418 static void subfacet_update_stats(struct subfacet *,
419 const struct dpif_flow_stats *);
420 static int subfacet_install(struct subfacet *,
421 const struct ofpbuf *odp_actions,
422 struct dpif_flow_stats *);
423 static void subfacet_uninstall(struct subfacet *);
425 /* An exact-match instantiation of an OpenFlow flow.
427 * A facet associates a "struct flow", which represents the Open vSwitch
428 * userspace idea of an exact-match flow, with one or more subfacets. Each
429 * subfacet tracks the datapath's idea of the exact-match flow equivalent to
430 * the facet. When the kernel module (or other dpif implementation) and Open
431 * vSwitch userspace agree on the definition of a flow key, there is exactly
432 * one subfacet per facet. If the dpif implementation supports more-specific
433 * flow matching than userspace, however, a facet can have more than one
434 * subfacet, each of which corresponds to some distinction in flow that
435 * userspace simply doesn't understand.
437 * Flow expiration works in terms of subfacets, so a facet must have at least
438 * one subfacet or it will never expire, leaking memory. */
441 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
442 struct list list_node; /* In owning rule's 'facets' list. */
443 struct rule_dpif *rule; /* Owning rule. */
446 struct list subfacets;
447 long long int used; /* Time last used; time created if not used. */
454 * - Do include packets and bytes sent "by hand", e.g. with
457 * - Do include packets and bytes that were obtained from the datapath
458 * when a subfacet's statistics were reset (e.g. dpif_flow_put() with
459 * DPIF_FP_ZERO_STATS).
461 * - Do not include packets or bytes that can be obtained from the
462 * datapath for any existing subfacet.
464 uint64_t packet_count; /* Number of packets received. */
465 uint64_t byte_count; /* Number of bytes received. */
467 /* Resubmit statistics. */
468 uint64_t prev_packet_count; /* Number of packets from last stats push. */
469 uint64_t prev_byte_count; /* Number of bytes from last stats push. */
470 long long int prev_used; /* Used time from last stats push. */
473 uint64_t accounted_bytes; /* Bytes processed by facet_account(). */
474 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
475 uint8_t tcp_flags; /* TCP flags seen for this 'rule'. */
477 struct xlate_out xout;
479 /* Initial values of the packet that may be needed later. */
480 struct initial_vals initial_vals;
482 /* Storage for a single subfacet, to reduce malloc() time and space
483 * overhead. (A facet always has at least one subfacet and in the common
484 * case has exactly one subfacet. However, 'one_subfacet' may not
485 * always be valid, since it could have been removed after newer
486 * subfacets were pushed onto the 'subfacets' list.) */
487 struct subfacet one_subfacet;
489 long long int learn_rl; /* Rate limiter for facet_learn(). */
492 static struct facet *facet_create(const struct flow_miss *, uint32_t hash);
493 static void facet_remove(struct facet *);
494 static void facet_free(struct facet *);
496 static struct facet *facet_find(struct ofproto_dpif *,
497 const struct flow *, uint32_t hash);
498 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
499 const struct flow *, uint32_t hash);
500 static bool facet_revalidate(struct facet *);
501 static bool facet_check_consistency(struct facet *);
503 static void facet_flush_stats(struct facet *);
505 static void facet_update_time(struct facet *, long long int used);
506 static void facet_reset_counters(struct facet *);
507 static void facet_push_stats(struct facet *);
508 static void facet_learn(struct facet *);
509 static void facet_account(struct facet *);
510 static void push_all_stats(void);
512 static bool facet_is_controller_flow(struct facet *);
515 struct hmap_node odp_port_node; /* In dpif_backer's "odp_to_ofport_map". */
519 struct ofbundle *bundle; /* Bundle that contains this port, if any. */
520 struct list bundle_node; /* In struct ofbundle's "ports" list. */
521 struct cfm *cfm; /* Connectivity Fault Management, if any. */
522 tag_type tag; /* Tag associated with this port. */
523 bool may_enable; /* May be enabled in bonds. */
524 long long int carrier_seq; /* Carrier status changes. */
525 struct tnl_port *tnl_port; /* Tunnel handle, or null. */
528 struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */
529 enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */
530 long long int stp_state_entered;
532 struct hmap priorities; /* Map of attached 'priority_to_dscp's. */
534 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
536 * This is deprecated. It is only for compatibility with broken device
537 * drivers in old versions of Linux that do not properly support VLANs when
538 * VLAN devices are not used. When broken device drivers are no longer in
539 * widespread use, we will delete these interfaces. */
540 uint16_t realdev_ofp_port;
544 /* Node in 'ofport_dpif''s 'priorities' map. Used to maintain a map from
545 * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
546 * traffic egressing the 'ofport' with that priority should be marked with. */
547 struct priority_to_dscp {
548 struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'priorities' map. */
549 uint32_t priority; /* Priority of this queue (see struct flow). */
551 uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */
554 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
556 * This is deprecated. It is only for compatibility with broken device drivers
557 * in old versions of Linux that do not properly support VLANs when VLAN
558 * devices are not used. When broken device drivers are no longer in
559 * widespread use, we will delete these interfaces. */
560 struct vlan_splinter {
561 struct hmap_node realdev_vid_node;
562 struct hmap_node vlandev_node;
563 uint16_t realdev_ofp_port;
564 uint16_t vlandev_ofp_port;
568 static uint16_t vsp_realdev_to_vlandev(const struct ofproto_dpif *,
569 uint16_t realdev_ofp_port,
571 static bool vsp_adjust_flow(const struct ofproto_dpif *, struct flow *);
572 static void vsp_remove(struct ofport_dpif *);
573 static void vsp_add(struct ofport_dpif *, uint16_t realdev_ofp_port, int vid);
575 static uint32_t ofp_port_to_odp_port(const struct ofproto_dpif *,
577 static uint16_t odp_port_to_ofp_port(const struct ofproto_dpif *,
580 static struct ofport_dpif *
581 ofport_dpif_cast(const struct ofport *ofport)
583 ovs_assert(ofport->ofproto->ofproto_class == &ofproto_dpif_class);
584 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
587 static void port_run(struct ofport_dpif *);
588 static void port_run_fast(struct ofport_dpif *);
589 static void port_wait(struct ofport_dpif *);
590 static int set_cfm(struct ofport *, const struct cfm_settings *);
591 static void ofport_clear_priorities(struct ofport_dpif *);
592 static void run_fast_rl(void);
594 struct dpif_completion {
595 struct list list_node;
596 struct ofoperation *op;
599 /* Extra information about a classifier table.
600 * Currently used just for optimized flow revalidation. */
602 /* If either of these is nonnull, then this table has a form that allows
603 * flows to be tagged to avoid revalidating most flows for the most common
604 * kinds of flow table changes. */
605 struct cls_table *catchall_table; /* Table that wildcards all fields. */
606 struct cls_table *other_table; /* Table with any other wildcard set. */
607 uint32_t basis; /* Keeps each table's tags separate. */
610 /* Reasons that we might need to revalidate every facet, and corresponding
613 * A value of 0 means that there is no need to revalidate.
615 * It would be nice to have some cleaner way to integrate with coverage
616 * counters, but with only a few reasons I guess this is good enough for
618 enum revalidate_reason {
619 REV_RECONFIGURE = 1, /* Switch configuration changed. */
620 REV_STP, /* Spanning tree protocol port status change. */
621 REV_PORT_TOGGLED, /* Port enabled or disabled by CFM, LACP, ...*/
622 REV_FLOW_TABLE, /* Flow table changed. */
623 REV_INCONSISTENCY /* Facet self-check failed. */
625 COVERAGE_DEFINE(rev_reconfigure);
626 COVERAGE_DEFINE(rev_stp);
627 COVERAGE_DEFINE(rev_port_toggled);
628 COVERAGE_DEFINE(rev_flow_table);
629 COVERAGE_DEFINE(rev_inconsistency);
631 /* Drop keys are odp flow keys which have drop flows installed in the kernel.
632 * These are datapath flows which have no associated ofproto, if they did we
633 * would use facets. */
635 struct hmap_node hmap_node;
640 /* All datapaths of a given type share a single dpif backer instance. */
645 struct timer next_expiration;
646 struct hmap odp_to_ofport_map; /* ODP port to ofport mapping. */
648 struct simap tnl_backers; /* Set of dpif ports backing tunnels. */
650 /* Facet revalidation flags applying to facets which use this backer. */
651 enum revalidate_reason need_revalidate; /* Revalidate every facet. */
652 struct tag_set revalidate_set; /* Revalidate only matching facets. */
654 struct hmap drop_keys; /* Set of dropped odp keys. */
655 bool recv_set_enable; /* Enables or disables receiving packets. */
658 /* All existing ofproto_backer instances, indexed by ofproto->up.type. */
659 static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers);
661 static void drop_key_clear(struct dpif_backer *);
662 static struct ofport_dpif *
663 odp_port_to_ofport(const struct dpif_backer *, uint32_t odp_port);
665 static void dpif_stats_update_hit_count(struct ofproto_dpif *ofproto,
667 struct avg_subfacet_rates {
668 double add_rate; /* Moving average of new flows created per minute. */
669 double del_rate; /* Moving average of flows deleted per minute. */
671 static void show_dp_rates(struct ds *ds, const char *heading,
672 const struct avg_subfacet_rates *rates);
673 static void exp_mavg(double *avg, int base, double new);
675 struct ofproto_dpif {
676 struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
678 struct dpif_backer *backer;
680 /* Special OpenFlow rules. */
681 struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */
682 struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */
688 struct netflow *netflow;
689 struct dpif_sflow *sflow;
690 struct dpif_ipfix *ipfix;
691 struct hmap bundles; /* Contains "struct ofbundle"s. */
692 struct mac_learning *ml;
693 struct ofmirror *mirrors[MAX_MIRRORS];
695 bool has_bonded_bundles;
699 struct hmap subfacets;
700 struct governor *governor;
701 long long int consistency_rl;
704 struct table_dpif tables[N_TABLES];
706 /* Support for debugging async flow mods. */
707 struct list completions;
709 bool has_bundle_action; /* True when the first bundle action appears. */
710 struct netdev_stats stats; /* To account packets generated and consumed in
715 long long int stp_last_tick;
717 /* VLAN splinters. */
718 struct hmap realdev_vid_map; /* (realdev,vid) -> vlandev. */
719 struct hmap vlandev_map; /* vlandev -> (realdev,vid). */
722 struct sset ports; /* Set of standard port names. */
723 struct sset ghost_ports; /* Ports with no datapath port. */
724 struct sset port_poll_set; /* Queued names for port_poll() reply. */
725 int port_poll_errno; /* Last errno for port_poll() reply. */
727 /* Per ofproto's dpif stats. */
731 /* Subfacet statistics.
733 * These keep track of the total number of subfacets added and deleted and
734 * flow life span. They are useful for computing the flow rates stats
735 * exposed via "ovs-appctl dpif/show". The goal is to learn about
736 * traffic patterns in ways that we can use later to improve Open vSwitch
737 * performance in new situations. */
738 long long int created; /* Time when it is created. */
739 unsigned int max_n_subfacet; /* Maximum number of flows */
741 /* The average number of subfacets... */
742 struct avg_subfacet_rates hourly; /* ...over the last hour. */
743 struct avg_subfacet_rates daily; /* ...over the last day. */
744 long long int last_minute; /* Last time 'hourly' was updated. */
746 /* Number of subfacets added or deleted since 'last_minute'. */
747 unsigned int subfacet_add_count;
748 unsigned int subfacet_del_count;
750 /* Number of subfacets added or deleted from 'created' to 'last_minute.' */
751 unsigned long long int total_subfacet_add_count;
752 unsigned long long int total_subfacet_del_count;
754 /* Sum of the number of milliseconds that each subfacet existed,
755 * over the subfacets that have been added and then later deleted. */
756 unsigned long long int total_subfacet_life_span;
758 /* Incremented by the number of currently existing subfacets, each
759 * time we pull statistics from the kernel. */
760 unsigned long long int total_subfacet_count;
762 /* Number of times we pull statistics from the kernel. */
763 unsigned long long int n_update_stats;
765 static unsigned long long int avg_subfacet_life_span(
766 const struct ofproto_dpif *);
767 static double avg_subfacet_count(const struct ofproto_dpif *ofproto);
768 static void update_moving_averages(struct ofproto_dpif *ofproto);
769 static void dpif_stats_update_hit_count(struct ofproto_dpif *ofproto,
771 static void update_max_subfacet_count(struct ofproto_dpif *ofproto);
773 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
774 * for debugging the asynchronous flow_mod implementation.) */
777 /* All existing ofproto_dpif instances, indexed by ->up.name. */
778 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
780 static void ofproto_dpif_unixctl_init(void);
782 static struct ofproto_dpif *
783 ofproto_dpif_cast(const struct ofproto *ofproto)
785 ovs_assert(ofproto->ofproto_class == &ofproto_dpif_class);
786 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
789 static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *,
791 static struct ofport_dpif *get_odp_port(const struct ofproto_dpif *,
793 static void ofproto_trace(struct ofproto_dpif *, const struct flow *,
794 const struct ofpbuf *,
795 const struct initial_vals *, struct ds *);
797 /* Packet processing. */
798 static void update_learning_table(struct ofproto_dpif *,
799 const struct flow *, int vlan,
802 #define FLOW_MISS_MAX_BATCH 50
803 static int handle_upcalls(struct dpif_backer *, unsigned int max_batch);
805 /* Flow expiration. */
806 static int expire(struct dpif_backer *);
809 static void send_netflow_active_timeouts(struct ofproto_dpif *);
812 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
813 static size_t compose_sflow_action(const struct ofproto_dpif *,
814 struct ofpbuf *odp_actions,
815 const struct flow *, uint32_t odp_port);
816 static void compose_ipfix_action(const struct ofproto_dpif *,
817 struct ofpbuf *odp_actions,
818 const struct flow *);
819 static void add_mirror_actions(struct xlate_ctx *ctx,
820 const struct flow *flow);
821 /* Global variables. */
822 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
824 /* Initial mappings of port to bridge mappings. */
825 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
827 /* Factory functions. */
830 init(const struct shash *iface_hints)
832 struct shash_node *node;
834 /* Make a local copy, since we don't own 'iface_hints' elements. */
835 SHASH_FOR_EACH(node, iface_hints) {
836 const struct iface_hint *orig_hint = node->data;
837 struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
839 new_hint->br_name = xstrdup(orig_hint->br_name);
840 new_hint->br_type = xstrdup(orig_hint->br_type);
841 new_hint->ofp_port = orig_hint->ofp_port;
843 shash_add(&init_ofp_ports, node->name, new_hint);
848 enumerate_types(struct sset *types)
850 dp_enumerate_types(types);
854 enumerate_names(const char *type, struct sset *names)
856 struct ofproto_dpif *ofproto;
859 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
860 if (strcmp(type, ofproto->up.type)) {
863 sset_add(names, ofproto->up.name);
870 del(const char *type, const char *name)
875 error = dpif_open(name, type, &dpif);
877 error = dpif_delete(dpif);
884 port_open_type(const char *datapath_type, const char *port_type)
886 return dpif_port_open_type(datapath_type, port_type);
889 /* Type functions. */
891 static struct ofproto_dpif *
892 lookup_ofproto_dpif_by_port_name(const char *name)
894 struct ofproto_dpif *ofproto;
896 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
897 if (sset_contains(&ofproto->ports, name)) {
906 type_run(const char *type)
908 static long long int push_timer = LLONG_MIN;
909 struct dpif_backer *backer;
913 backer = shash_find_data(&all_dpif_backers, type);
915 /* This is not necessarily a problem, since backers are only
916 * created on demand. */
920 dpif_run(backer->dpif);
922 /* The most natural place to push facet statistics is when they're pulled
923 * from the datapath. However, when there are many flows in the datapath,
924 * this expensive operation can occur so frequently, that it reduces our
925 * ability to quickly set up flows. To reduce the cost, we push statistics
927 if (time_msec() > push_timer) {
928 push_timer = time_msec() + 2000;
932 /* If vswitchd started with other_config:flow_restore_wait set as "true",
933 * and the configuration has now changed to "false", enable receiving
934 * packets from the datapath. */
935 if (!backer->recv_set_enable && !ofproto_get_flow_restore_wait()) {
936 backer->recv_set_enable = true;
938 error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
940 VLOG_ERR("Failed to enable receiving packets in dpif.");
943 dpif_flow_flush(backer->dpif);
944 backer->need_revalidate = REV_RECONFIGURE;
947 if (backer->need_revalidate
948 || !tag_set_is_empty(&backer->revalidate_set)) {
949 struct tag_set revalidate_set = backer->revalidate_set;
950 bool need_revalidate = backer->need_revalidate;
951 struct ofproto_dpif *ofproto;
952 struct simap_node *node;
953 struct simap tmp_backers;
955 /* Handle tunnel garbage collection. */
956 simap_init(&tmp_backers);
957 simap_swap(&backer->tnl_backers, &tmp_backers);
959 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
960 struct ofport_dpif *iter;
962 if (backer != ofproto->backer) {
966 HMAP_FOR_EACH (iter, up.hmap_node, &ofproto->up.ports) {
969 if (!iter->tnl_port) {
973 dp_port = netdev_vport_get_dpif_port(iter->up.netdev);
974 node = simap_find(&tmp_backers, dp_port);
976 simap_put(&backer->tnl_backers, dp_port, node->data);
977 simap_delete(&tmp_backers, node);
978 node = simap_find(&backer->tnl_backers, dp_port);
980 node = simap_find(&backer->tnl_backers, dp_port);
982 uint32_t odp_port = UINT32_MAX;
984 if (!dpif_port_add(backer->dpif, iter->up.netdev,
986 simap_put(&backer->tnl_backers, dp_port, odp_port);
987 node = simap_find(&backer->tnl_backers, dp_port);
992 iter->odp_port = node ? node->data : OVSP_NONE;
993 if (tnl_port_reconfigure(&iter->up, iter->odp_port,
995 backer->need_revalidate = REV_RECONFIGURE;
1000 SIMAP_FOR_EACH (node, &tmp_backers) {
1001 dpif_port_del(backer->dpif, node->data);
1003 simap_destroy(&tmp_backers);
1005 switch (backer->need_revalidate) {
1006 case REV_RECONFIGURE: COVERAGE_INC(rev_reconfigure); break;
1007 case REV_STP: COVERAGE_INC(rev_stp); break;
1008 case REV_PORT_TOGGLED: COVERAGE_INC(rev_port_toggled); break;
1009 case REV_FLOW_TABLE: COVERAGE_INC(rev_flow_table); break;
1010 case REV_INCONSISTENCY: COVERAGE_INC(rev_inconsistency); break;
1013 if (backer->need_revalidate) {
1014 /* Clear the drop_keys in case we should now be accepting some
1015 * formerly dropped flows. */
1016 drop_key_clear(backer);
1019 /* Clear the revalidation flags. */
1020 tag_set_init(&backer->revalidate_set);
1021 backer->need_revalidate = 0;
1023 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1024 struct facet *facet, *next;
1026 if (ofproto->backer != backer) {
1030 HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &ofproto->facets) {
1032 || tag_set_intersects(&revalidate_set, facet->xout.tags)) {
1033 facet_revalidate(facet);
1040 if (!backer->recv_set_enable) {
1041 /* Wake up before a max of 1000ms. */
1042 timer_set_duration(&backer->next_expiration, 1000);
1043 } else if (timer_expired(&backer->next_expiration)) {
1044 int delay = expire(backer);
1045 timer_set_duration(&backer->next_expiration, delay);
1048 /* Check for port changes in the dpif. */
1049 while ((error = dpif_port_poll(backer->dpif, &devname)) == 0) {
1050 struct ofproto_dpif *ofproto;
1051 struct dpif_port port;
1053 /* Don't report on the datapath's device. */
1054 if (!strcmp(devname, dpif_base_name(backer->dpif))) {
1058 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
1059 &all_ofproto_dpifs) {
1060 if (simap_contains(&ofproto->backer->tnl_backers, devname)) {
1065 ofproto = lookup_ofproto_dpif_by_port_name(devname);
1066 if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
1067 /* The port was removed. If we know the datapath,
1068 * report it through poll_set(). If we don't, it may be
1069 * notifying us of a removal we initiated, so ignore it.
1070 * If there's a pending ENOBUFS, let it stand, since
1071 * everything will be reevaluated. */
1072 if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
1073 sset_add(&ofproto->port_poll_set, devname);
1074 ofproto->port_poll_errno = 0;
1076 } else if (!ofproto) {
1077 /* The port was added, but we don't know with which
1078 * ofproto we should associate it. Delete it. */
1079 dpif_port_del(backer->dpif, port.port_no);
1081 dpif_port_destroy(&port);
1087 if (error != EAGAIN) {
1088 struct ofproto_dpif *ofproto;
1090 /* There was some sort of error, so propagate it to all
1091 * ofprotos that use this backer. */
1092 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
1093 &all_ofproto_dpifs) {
1094 if (ofproto->backer == backer) {
1095 sset_clear(&ofproto->port_poll_set);
1096 ofproto->port_poll_errno = error;
1105 dpif_backer_run_fast(struct dpif_backer *backer, int max_batch)
1109 /* If recv_set_enable is false, we should not handle upcalls. */
1110 if (!backer->recv_set_enable) {
1114 /* Handle one or more batches of upcalls, until there's nothing left to do
1115 * or until we do a fixed total amount of work.
1117 * We do work in batches because it can be much cheaper to set up a number
1118 * of flows and fire off their patches all at once. We do multiple batches
1119 * because in some cases handling a packet can cause another packet to be
1120 * queued almost immediately as part of the return flow. Both
1121 * optimizations can make major improvements on some benchmarks and
1122 * presumably for real traffic as well. */
1124 while (work < max_batch) {
1125 int retval = handle_upcalls(backer, max_batch - work);
1136 type_run_fast(const char *type)
1138 struct dpif_backer *backer;
1140 backer = shash_find_data(&all_dpif_backers, type);
1142 /* This is not necessarily a problem, since backers are only
1143 * created on demand. */
1147 return dpif_backer_run_fast(backer, FLOW_MISS_MAX_BATCH);
1153 static long long int port_rl = LLONG_MIN;
1154 static unsigned int backer_rl = 0;
1156 if (time_msec() >= port_rl) {
1157 struct ofproto_dpif *ofproto;
1158 struct ofport_dpif *ofport;
1160 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1162 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1163 port_run_fast(ofport);
1166 port_rl = time_msec() + 200;
1169 /* XXX: We have to be careful not to do too much work in this function. If
1170 * we call dpif_backer_run_fast() too often, or with too large a batch,
1171 * performance improves signifcantly, but at a cost. It's possible for the
1172 * number of flows in the datapath to increase without bound, and for poll
1173 * loops to take 10s of seconds. The correct solution to this problem,
1174 * long term, is to separate flow miss handling into it's own thread so it
1175 * isn't affected by revalidations, and expirations. Until then, this is
1176 * the best we can do. */
1177 if (++backer_rl >= 10) {
1178 struct shash_node *node;
1181 SHASH_FOR_EACH (node, &all_dpif_backers) {
1182 dpif_backer_run_fast(node->data, 1);
1188 type_wait(const char *type)
1190 struct dpif_backer *backer;
1192 backer = shash_find_data(&all_dpif_backers, type);
1194 /* This is not necessarily a problem, since backers are only
1195 * created on demand. */
1199 timer_wait(&backer->next_expiration);
1202 /* Basic life-cycle. */
1204 static int add_internal_flows(struct ofproto_dpif *);
1206 static struct ofproto *
1209 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
1210 return &ofproto->up;
1214 dealloc(struct ofproto *ofproto_)
1216 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1221 close_dpif_backer(struct dpif_backer *backer)
1223 struct shash_node *node;
1225 ovs_assert(backer->refcount > 0);
1227 if (--backer->refcount) {
1231 drop_key_clear(backer);
1232 hmap_destroy(&backer->drop_keys);
1234 simap_destroy(&backer->tnl_backers);
1235 hmap_destroy(&backer->odp_to_ofport_map);
1236 node = shash_find(&all_dpif_backers, backer->type);
1238 shash_delete(&all_dpif_backers, node);
1239 dpif_close(backer->dpif);
1244 /* Datapath port slated for removal from datapath. */
1245 struct odp_garbage {
1246 struct list list_node;
1251 open_dpif_backer(const char *type, struct dpif_backer **backerp)
1253 struct dpif_backer *backer;
1254 struct dpif_port_dump port_dump;
1255 struct dpif_port port;
1256 struct shash_node *node;
1257 struct list garbage_list;
1258 struct odp_garbage *garbage, *next;
1264 backer = shash_find_data(&all_dpif_backers, type);
1271 backer_name = xasprintf("ovs-%s", type);
1273 /* Remove any existing datapaths, since we assume we're the only
1274 * userspace controlling the datapath. */
1276 dp_enumerate_names(type, &names);
1277 SSET_FOR_EACH(name, &names) {
1278 struct dpif *old_dpif;
1280 /* Don't remove our backer if it exists. */
1281 if (!strcmp(name, backer_name)) {
1285 if (dpif_open(name, type, &old_dpif)) {
1286 VLOG_WARN("couldn't open old datapath %s to remove it", name);
1288 dpif_delete(old_dpif);
1289 dpif_close(old_dpif);
1292 sset_destroy(&names);
1294 backer = xmalloc(sizeof *backer);
1296 error = dpif_create_and_open(backer_name, type, &backer->dpif);
1299 VLOG_ERR("failed to open datapath of type %s: %s", type,
1305 backer->type = xstrdup(type);
1306 backer->refcount = 1;
1307 hmap_init(&backer->odp_to_ofport_map);
1308 hmap_init(&backer->drop_keys);
1309 timer_set_duration(&backer->next_expiration, 1000);
1310 backer->need_revalidate = 0;
1311 simap_init(&backer->tnl_backers);
1312 tag_set_init(&backer->revalidate_set);
1313 backer->recv_set_enable = !ofproto_get_flow_restore_wait();
1316 if (backer->recv_set_enable) {
1317 dpif_flow_flush(backer->dpif);
1320 /* Loop through the ports already on the datapath and remove any
1321 * that we don't need anymore. */
1322 list_init(&garbage_list);
1323 dpif_port_dump_start(&port_dump, backer->dpif);
1324 while (dpif_port_dump_next(&port_dump, &port)) {
1325 node = shash_find(&init_ofp_ports, port.name);
1326 if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
1327 garbage = xmalloc(sizeof *garbage);
1328 garbage->odp_port = port.port_no;
1329 list_push_front(&garbage_list, &garbage->list_node);
1332 dpif_port_dump_done(&port_dump);
1334 LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) {
1335 dpif_port_del(backer->dpif, garbage->odp_port);
1336 list_remove(&garbage->list_node);
1340 shash_add(&all_dpif_backers, type, backer);
1342 error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
1344 VLOG_ERR("failed to listen on datapath of type %s: %s",
1345 type, strerror(error));
1346 close_dpif_backer(backer);
1354 construct(struct ofproto *ofproto_)
1356 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1357 struct shash_node *node, *next;
1362 error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1367 max_ports = dpif_get_max_ports(ofproto->backer->dpif);
1368 ofproto_init_max_ports(ofproto_, MIN(max_ports, OFPP_MAX));
1370 ofproto->n_matches = 0;
1372 ofproto->netflow = NULL;
1373 ofproto->sflow = NULL;
1374 ofproto->ipfix = NULL;
1375 ofproto->stp = NULL;
1376 hmap_init(&ofproto->bundles);
1377 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1378 for (i = 0; i < MAX_MIRRORS; i++) {
1379 ofproto->mirrors[i] = NULL;
1381 ofproto->has_bonded_bundles = false;
1383 hmap_init(&ofproto->facets);
1384 hmap_init(&ofproto->subfacets);
1385 ofproto->governor = NULL;
1386 ofproto->consistency_rl = LLONG_MIN;
1388 for (i = 0; i < N_TABLES; i++) {
1389 struct table_dpif *table = &ofproto->tables[i];
1391 table->catchall_table = NULL;
1392 table->other_table = NULL;
1393 table->basis = random_uint32();
1396 list_init(&ofproto->completions);
1398 ofproto_dpif_unixctl_init();
1400 ofproto->has_mirrors = false;
1401 ofproto->has_bundle_action = false;
1403 hmap_init(&ofproto->vlandev_map);
1404 hmap_init(&ofproto->realdev_vid_map);
1406 sset_init(&ofproto->ports);
1407 sset_init(&ofproto->ghost_ports);
1408 sset_init(&ofproto->port_poll_set);
1409 ofproto->port_poll_errno = 0;
1411 SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1412 struct iface_hint *iface_hint = node->data;
1414 if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1415 /* Check if the datapath already has this port. */
1416 if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1417 sset_add(&ofproto->ports, node->name);
1420 free(iface_hint->br_name);
1421 free(iface_hint->br_type);
1423 shash_delete(&init_ofp_ports, node);
1427 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1428 hash_string(ofproto->up.name, 0));
1429 memset(&ofproto->stats, 0, sizeof ofproto->stats);
1431 ofproto_init_tables(ofproto_, N_TABLES);
1432 error = add_internal_flows(ofproto);
1433 ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1436 ofproto->n_missed = 0;
1438 ofproto->max_n_subfacet = 0;
1439 ofproto->created = time_msec();
1440 ofproto->last_minute = ofproto->created;
1441 memset(&ofproto->hourly, 0, sizeof ofproto->hourly);
1442 memset(&ofproto->daily, 0, sizeof ofproto->daily);
1443 ofproto->subfacet_add_count = 0;
1444 ofproto->subfacet_del_count = 0;
1445 ofproto->total_subfacet_add_count = 0;
1446 ofproto->total_subfacet_del_count = 0;
1447 ofproto->total_subfacet_life_span = 0;
1448 ofproto->total_subfacet_count = 0;
1449 ofproto->n_update_stats = 0;
1455 add_internal_flow(struct ofproto_dpif *ofproto, int id,
1456 const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1458 struct ofputil_flow_mod fm;
1461 match_init_catchall(&fm.match);
1463 match_set_reg(&fm.match, 0, id);
1464 fm.new_cookie = htonll(0);
1465 fm.cookie = htonll(0);
1466 fm.cookie_mask = htonll(0);
1467 fm.table_id = TBL_INTERNAL;
1468 fm.command = OFPFC_ADD;
1469 fm.idle_timeout = 0;
1470 fm.hard_timeout = 0;
1474 fm.ofpacts = ofpacts->data;
1475 fm.ofpacts_len = ofpacts->size;
1477 error = ofproto_flow_mod(&ofproto->up, &fm);
1479 VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
1480 id, ofperr_to_string(error));
1484 *rulep = rule_dpif_lookup__(ofproto, &fm.match.flow, TBL_INTERNAL);
1485 ovs_assert(*rulep != NULL);
1491 add_internal_flows(struct ofproto_dpif *ofproto)
1493 struct ofpact_controller *controller;
1494 uint64_t ofpacts_stub[128 / 8];
1495 struct ofpbuf ofpacts;
1499 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1502 controller = ofpact_put_CONTROLLER(&ofpacts);
1503 controller->max_len = UINT16_MAX;
1504 controller->controller_id = 0;
1505 controller->reason = OFPR_NO_MATCH;
1506 ofpact_pad(&ofpacts);
1508 error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule);
1513 ofpbuf_clear(&ofpacts);
1514 error = add_internal_flow(ofproto, id++, &ofpacts,
1515 &ofproto->no_packet_in_rule);
1520 complete_operations(struct ofproto_dpif *ofproto)
1522 struct dpif_completion *c, *next;
1524 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
1525 ofoperation_complete(c->op, 0);
1526 list_remove(&c->list_node);
1532 destruct(struct ofproto *ofproto_)
1534 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1535 struct rule_dpif *rule, *next_rule;
1536 struct oftable *table;
1539 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1540 complete_operations(ofproto);
1542 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1543 struct cls_cursor cursor;
1545 cls_cursor_init(&cursor, &table->cls, NULL);
1546 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1547 ofproto_rule_destroy(&rule->up);
1551 for (i = 0; i < MAX_MIRRORS; i++) {
1552 mirror_destroy(ofproto->mirrors[i]);
1555 netflow_destroy(ofproto->netflow);
1556 dpif_sflow_destroy(ofproto->sflow);
1557 hmap_destroy(&ofproto->bundles);
1558 mac_learning_destroy(ofproto->ml);
1560 hmap_destroy(&ofproto->facets);
1561 hmap_destroy(&ofproto->subfacets);
1562 governor_destroy(ofproto->governor);
1564 hmap_destroy(&ofproto->vlandev_map);
1565 hmap_destroy(&ofproto->realdev_vid_map);
1567 sset_destroy(&ofproto->ports);
1568 sset_destroy(&ofproto->ghost_ports);
1569 sset_destroy(&ofproto->port_poll_set);
1571 close_dpif_backer(ofproto->backer);
1575 run_fast(struct ofproto *ofproto_)
1577 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1578 struct ofport_dpif *ofport;
1580 /* Do not perform any periodic activity required by 'ofproto' while
1581 * waiting for flow restore to complete. */
1582 if (ofproto_get_flow_restore_wait()) {
1586 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1587 port_run_fast(ofport);
1594 run(struct ofproto *ofproto_)
1596 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1597 struct ofport_dpif *ofport;
1598 struct ofbundle *bundle;
1602 complete_operations(ofproto);
1605 /* Do not perform any periodic activity below required by 'ofproto' while
1606 * waiting for flow restore to complete. */
1607 if (ofproto_get_flow_restore_wait()) {
1611 error = run_fast(ofproto_);
1616 if (ofproto->netflow) {
1617 if (netflow_run(ofproto->netflow)) {
1618 send_netflow_active_timeouts(ofproto);
1621 if (ofproto->sflow) {
1622 dpif_sflow_run(ofproto->sflow);
1625 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1628 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1633 mac_learning_run(ofproto->ml, &ofproto->backer->revalidate_set);
1635 /* Check the consistency of a random facet, to aid debugging. */
1636 if (time_msec() >= ofproto->consistency_rl
1637 && !hmap_is_empty(&ofproto->facets)
1638 && !ofproto->backer->need_revalidate) {
1639 struct facet *facet;
1641 ofproto->consistency_rl = time_msec() + 250;
1643 facet = CONTAINER_OF(hmap_random_node(&ofproto->facets),
1644 struct facet, hmap_node);
1645 if (!tag_set_intersects(&ofproto->backer->revalidate_set,
1646 facet->xout.tags)) {
1647 if (!facet_check_consistency(facet)) {
1648 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
1653 if (ofproto->governor) {
1656 governor_run(ofproto->governor);
1658 /* If the governor has shrunk to its minimum size and the number of
1659 * subfacets has dwindled, then drop the governor entirely.
1661 * For hysteresis, the number of subfacets to drop the governor is
1662 * smaller than the number needed to trigger its creation. */
1663 n_subfacets = hmap_count(&ofproto->subfacets);
1664 if (n_subfacets * 4 < ofproto->up.flow_eviction_threshold
1665 && governor_is_idle(ofproto->governor)) {
1666 governor_destroy(ofproto->governor);
1667 ofproto->governor = NULL;
1675 wait(struct ofproto *ofproto_)
1677 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1678 struct ofport_dpif *ofport;
1679 struct ofbundle *bundle;
1681 if (!clogged && !list_is_empty(&ofproto->completions)) {
1682 poll_immediate_wake();
1685 if (ofproto_get_flow_restore_wait()) {
1689 dpif_wait(ofproto->backer->dpif);
1690 dpif_recv_wait(ofproto->backer->dpif);
1691 if (ofproto->sflow) {
1692 dpif_sflow_wait(ofproto->sflow);
1694 if (!tag_set_is_empty(&ofproto->backer->revalidate_set)) {
1695 poll_immediate_wake();
1697 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1700 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1701 bundle_wait(bundle);
1703 if (ofproto->netflow) {
1704 netflow_wait(ofproto->netflow);
1706 mac_learning_wait(ofproto->ml);
1708 if (ofproto->backer->need_revalidate) {
1709 /* Shouldn't happen, but if it does just go around again. */
1710 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1711 poll_immediate_wake();
1713 if (ofproto->governor) {
1714 governor_wait(ofproto->governor);
1719 get_memory_usage(const struct ofproto *ofproto_, struct simap *usage)
1721 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1723 simap_increase(usage, "facets", hmap_count(&ofproto->facets));
1724 simap_increase(usage, "subfacets", hmap_count(&ofproto->subfacets));
1728 flush(struct ofproto *ofproto_)
1730 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1731 struct subfacet *subfacet, *next_subfacet;
1732 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
1736 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
1737 &ofproto->subfacets) {
1738 if (subfacet->path != SF_NOT_INSTALLED) {
1739 batch[n_batch++] = subfacet;
1740 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
1741 subfacet_destroy_batch(ofproto, batch, n_batch);
1745 subfacet_destroy(subfacet);
1750 subfacet_destroy_batch(ofproto, batch, n_batch);
1755 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1756 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1758 *arp_match_ip = true;
1759 *actions = (OFPUTIL_A_OUTPUT |
1760 OFPUTIL_A_SET_VLAN_VID |
1761 OFPUTIL_A_SET_VLAN_PCP |
1762 OFPUTIL_A_STRIP_VLAN |
1763 OFPUTIL_A_SET_DL_SRC |
1764 OFPUTIL_A_SET_DL_DST |
1765 OFPUTIL_A_SET_NW_SRC |
1766 OFPUTIL_A_SET_NW_DST |
1767 OFPUTIL_A_SET_NW_TOS |
1768 OFPUTIL_A_SET_TP_SRC |
1769 OFPUTIL_A_SET_TP_DST |
1774 get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots)
1776 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1777 struct dpif_dp_stats s;
1779 strcpy(ots->name, "classifier");
1781 dpif_get_dp_stats(ofproto->backer->dpif, &s);
1783 ots->lookup_count = htonll(s.n_hit + s.n_missed);
1784 ots->matched_count = htonll(s.n_hit + ofproto->n_matches);
1787 static struct ofport *
1790 struct ofport_dpif *port = xmalloc(sizeof *port);
1795 port_dealloc(struct ofport *port_)
1797 struct ofport_dpif *port = ofport_dpif_cast(port_);
1802 port_construct(struct ofport *port_)
1804 struct ofport_dpif *port = ofport_dpif_cast(port_);
1805 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1806 const struct netdev *netdev = port->up.netdev;
1807 struct dpif_port dpif_port;
1810 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1811 port->bundle = NULL;
1813 port->tag = tag_create_random();
1814 port->may_enable = true;
1815 port->stp_port = NULL;
1816 port->stp_state = STP_DISABLED;
1817 port->tnl_port = NULL;
1818 hmap_init(&port->priorities);
1819 port->realdev_ofp_port = 0;
1820 port->vlandev_vid = 0;
1821 port->carrier_seq = netdev_get_carrier_resets(netdev);
1823 if (netdev_vport_is_patch(netdev)) {
1824 /* By bailing out here, we don't submit the port to the sFlow module
1825 * to be considered for counter polling export. This is correct
1826 * because the patch port represents an interface that sFlow considers
1827 * to be "internal" to the switch as a whole, and therefore not an
1828 * candidate for counter polling. */
1829 port->odp_port = OVSP_NONE;
1833 error = dpif_port_query_by_name(ofproto->backer->dpif,
1834 netdev_vport_get_dpif_port(netdev),
1840 port->odp_port = dpif_port.port_no;
1842 if (netdev_get_tunnel_config(netdev)) {
1843 port->tnl_port = tnl_port_add(&port->up, port->odp_port);
1845 /* Sanity-check that a mapping doesn't already exist. This
1846 * shouldn't happen for non-tunnel ports. */
1847 if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1848 VLOG_ERR("port %s already has an OpenFlow port number",
1850 dpif_port_destroy(&dpif_port);
1854 hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1855 hash_int(port->odp_port, 0));
1857 dpif_port_destroy(&dpif_port);
1859 if (ofproto->sflow) {
1860 dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1867 port_destruct(struct ofport *port_)
1869 struct ofport_dpif *port = ofport_dpif_cast(port_);
1870 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1871 const char *dp_port_name = netdev_vport_get_dpif_port(port->up.netdev);
1872 const char *devname = netdev_get_name(port->up.netdev);
1874 if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1875 /* The underlying device is still there, so delete it. This
1876 * happens when the ofproto is being destroyed, since the caller
1877 * assumes that removal of attached ports will happen as part of
1879 if (!port->tnl_port) {
1880 dpif_port_del(ofproto->backer->dpif, port->odp_port);
1882 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1885 if (port->odp_port != OVSP_NONE && !port->tnl_port) {
1886 hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1889 tnl_port_del(port->tnl_port);
1890 sset_find_and_delete(&ofproto->ports, devname);
1891 sset_find_and_delete(&ofproto->ghost_ports, devname);
1892 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1893 bundle_remove(port_);
1894 set_cfm(port_, NULL);
1895 if (ofproto->sflow) {
1896 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1899 ofport_clear_priorities(port);
1900 hmap_destroy(&port->priorities);
1904 port_modified(struct ofport *port_)
1906 struct ofport_dpif *port = ofport_dpif_cast(port_);
1908 if (port->bundle && port->bundle->bond) {
1909 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1913 cfm_set_netdev(port->cfm, port->up.netdev);
1918 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1920 struct ofport_dpif *port = ofport_dpif_cast(port_);
1921 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1922 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1924 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1925 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1926 OFPUTIL_PC_NO_PACKET_IN)) {
1927 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1929 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1930 bundle_update(port->bundle);
1936 set_sflow(struct ofproto *ofproto_,
1937 const struct ofproto_sflow_options *sflow_options)
1939 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1940 struct dpif_sflow *ds = ofproto->sflow;
1942 if (sflow_options) {
1944 struct ofport_dpif *ofport;
1946 ds = ofproto->sflow = dpif_sflow_create();
1947 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1948 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1950 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1952 dpif_sflow_set_options(ds, sflow_options);
1955 dpif_sflow_destroy(ds);
1956 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1957 ofproto->sflow = NULL;
1965 struct ofproto *ofproto_,
1966 const struct ofproto_ipfix_bridge_exporter_options *bridge_exporter_options,
1967 const struct ofproto_ipfix_flow_exporter_options *flow_exporters_options,
1968 size_t n_flow_exporters_options)
1970 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1971 struct dpif_ipfix *di = ofproto->ipfix;
1973 if (bridge_exporter_options || flow_exporters_options) {
1975 di = ofproto->ipfix = dpif_ipfix_create();
1977 dpif_ipfix_set_options(
1978 di, bridge_exporter_options, flow_exporters_options,
1979 n_flow_exporters_options);
1982 dpif_ipfix_destroy(di);
1983 ofproto->ipfix = NULL;
1990 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1992 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1999 struct ofproto_dpif *ofproto;
2001 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2002 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2003 ofport->cfm = cfm_create(ofport->up.netdev);
2006 if (cfm_configure(ofport->cfm, s)) {
2012 cfm_destroy(ofport->cfm);
2018 get_cfm_status(const struct ofport *ofport_,
2019 struct ofproto_cfm_status *status)
2021 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2024 status->faults = cfm_get_fault(ofport->cfm);
2025 status->remote_opstate = cfm_get_opup(ofport->cfm);
2026 status->health = cfm_get_health(ofport->cfm);
2027 cfm_get_remote_mpids(ofport->cfm, &status->rmps, &status->n_rmps);
2034 /* Spanning Tree. */
2037 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
2039 struct ofproto_dpif *ofproto = ofproto_;
2040 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
2041 struct ofport_dpif *ofport;
2043 ofport = stp_port_get_aux(sp);
2045 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
2046 ofproto->up.name, port_num);
2048 struct eth_header *eth = pkt->l2;
2050 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
2051 if (eth_addr_is_zero(eth->eth_src)) {
2052 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
2053 "with unknown MAC", ofproto->up.name, port_num);
2055 send_packet(ofport, pkt);
2061 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
2063 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
2065 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2067 /* Only revalidate flows if the configuration changed. */
2068 if (!s != !ofproto->stp) {
2069 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2073 if (!ofproto->stp) {
2074 ofproto->stp = stp_create(ofproto_->name, s->system_id,
2075 send_bpdu_cb, ofproto);
2076 ofproto->stp_last_tick = time_msec();
2079 stp_set_bridge_id(ofproto->stp, s->system_id);
2080 stp_set_bridge_priority(ofproto->stp, s->priority);
2081 stp_set_hello_time(ofproto->stp, s->hello_time);
2082 stp_set_max_age(ofproto->stp, s->max_age);
2083 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
2085 struct ofport *ofport;
2087 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
2088 set_stp_port(ofport, NULL);
2091 stp_destroy(ofproto->stp);
2092 ofproto->stp = NULL;
2099 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
2101 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2105 s->bridge_id = stp_get_bridge_id(ofproto->stp);
2106 s->designated_root = stp_get_designated_root(ofproto->stp);
2107 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
2116 update_stp_port_state(struct ofport_dpif *ofport)
2118 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2119 enum stp_state state;
2121 /* Figure out new state. */
2122 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
2126 if (ofport->stp_state != state) {
2127 enum ofputil_port_state of_state;
2130 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
2131 netdev_get_name(ofport->up.netdev),
2132 stp_state_name(ofport->stp_state),
2133 stp_state_name(state));
2134 if (stp_learn_in_state(ofport->stp_state)
2135 != stp_learn_in_state(state)) {
2136 /* xxx Learning action flows should also be flushed. */
2137 mac_learning_flush(ofproto->ml,
2138 &ofproto->backer->revalidate_set);
2140 fwd_change = stp_forward_in_state(ofport->stp_state)
2141 != stp_forward_in_state(state);
2143 ofproto->backer->need_revalidate = REV_STP;
2144 ofport->stp_state = state;
2145 ofport->stp_state_entered = time_msec();
2147 if (fwd_change && ofport->bundle) {
2148 bundle_update(ofport->bundle);
2151 /* Update the STP state bits in the OpenFlow port description. */
2152 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
2153 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
2154 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
2155 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
2156 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
2158 ofproto_port_set_state(&ofport->up, of_state);
2162 /* Configures STP on 'ofport_' using the settings defined in 's'. The
2163 * caller is responsible for assigning STP port numbers and ensuring
2164 * there are no duplicates. */
2166 set_stp_port(struct ofport *ofport_,
2167 const struct ofproto_port_stp_settings *s)
2169 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2170 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2171 struct stp_port *sp = ofport->stp_port;
2173 if (!s || !s->enable) {
2175 ofport->stp_port = NULL;
2176 stp_port_disable(sp);
2177 update_stp_port_state(ofport);
2180 } else if (sp && stp_port_no(sp) != s->port_num
2181 && ofport == stp_port_get_aux(sp)) {
2182 /* The port-id changed, so disable the old one if it's not
2183 * already in use by another port. */
2184 stp_port_disable(sp);
2187 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
2188 stp_port_enable(sp);
2190 stp_port_set_aux(sp, ofport);
2191 stp_port_set_priority(sp, s->priority);
2192 stp_port_set_path_cost(sp, s->path_cost);
2194 update_stp_port_state(ofport);
2200 get_stp_port_status(struct ofport *ofport_,
2201 struct ofproto_port_stp_status *s)
2203 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2204 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2205 struct stp_port *sp = ofport->stp_port;
2207 if (!ofproto->stp || !sp) {
2213 s->port_id = stp_port_get_id(sp);
2214 s->state = stp_port_get_state(sp);
2215 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
2216 s->role = stp_port_get_role(sp);
2217 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
2223 stp_run(struct ofproto_dpif *ofproto)
2226 long long int now = time_msec();
2227 long long int elapsed = now - ofproto->stp_last_tick;
2228 struct stp_port *sp;
2231 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
2232 ofproto->stp_last_tick = now;
2234 while (stp_get_changed_port(ofproto->stp, &sp)) {
2235 struct ofport_dpif *ofport = stp_port_get_aux(sp);
2238 update_stp_port_state(ofport);
2242 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
2243 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
2249 stp_wait(struct ofproto_dpif *ofproto)
2252 poll_timer_wait(1000);
2256 /* Returns true if STP should process 'flow'. */
2258 stp_should_process_flow(const struct flow *flow)
2260 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
2264 stp_process_packet(const struct ofport_dpif *ofport,
2265 const struct ofpbuf *packet)
2267 struct ofpbuf payload = *packet;
2268 struct eth_header *eth = payload.data;
2269 struct stp_port *sp = ofport->stp_port;
2271 /* Sink packets on ports that have STP disabled when the bridge has
2273 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
2277 /* Trim off padding on payload. */
2278 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
2279 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
2282 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
2283 stp_received_bpdu(sp, payload.data, payload.size);
2287 static struct priority_to_dscp *
2288 get_priority(const struct ofport_dpif *ofport, uint32_t priority)
2290 struct priority_to_dscp *pdscp;
2293 hash = hash_int(priority, 0);
2294 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) {
2295 if (pdscp->priority == priority) {
2303 ofport_clear_priorities(struct ofport_dpif *ofport)
2305 struct priority_to_dscp *pdscp, *next;
2307 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) {
2308 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
2314 set_queues(struct ofport *ofport_,
2315 const struct ofproto_port_queue *qdscp_list,
2318 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2319 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2320 struct hmap new = HMAP_INITIALIZER(&new);
2323 for (i = 0; i < n_qdscp; i++) {
2324 struct priority_to_dscp *pdscp;
2328 dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
2329 if (dpif_queue_to_priority(ofproto->backer->dpif, qdscp_list[i].queue,
2334 pdscp = get_priority(ofport, priority);
2336 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
2338 pdscp = xmalloc(sizeof *pdscp);
2339 pdscp->priority = priority;
2341 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2344 if (pdscp->dscp != dscp) {
2346 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2349 hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0));
2352 if (!hmap_is_empty(&ofport->priorities)) {
2353 ofport_clear_priorities(ofport);
2354 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2357 hmap_swap(&new, &ofport->priorities);
2365 /* Expires all MAC learning entries associated with 'bundle' and forces its
2366 * ofproto to revalidate every flow.
2368 * Normally MAC learning entries are removed only from the ofproto associated
2369 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2370 * are removed from every ofproto. When patch ports and SLB bonds are in use
2371 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2372 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2373 * with the host from which it migrated. */
2375 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2377 struct ofproto_dpif *ofproto = bundle->ofproto;
2378 struct mac_learning *ml = ofproto->ml;
2379 struct mac_entry *mac, *next_mac;
2381 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2382 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2383 if (mac->port.p == bundle) {
2385 struct ofproto_dpif *o;
2387 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2389 struct mac_entry *e;
2391 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan,
2394 mac_learning_expire(o->ml, e);
2400 mac_learning_expire(ml, mac);
2405 static struct ofbundle *
2406 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2408 struct ofbundle *bundle;
2410 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2411 &ofproto->bundles) {
2412 if (bundle->aux == aux) {
2419 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
2420 * ones that are found to 'bundles'. */
2422 bundle_lookup_multiple(struct ofproto_dpif *ofproto,
2423 void **auxes, size_t n_auxes,
2424 struct hmapx *bundles)
2428 hmapx_init(bundles);
2429 for (i = 0; i < n_auxes; i++) {
2430 struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
2432 hmapx_add(bundles, bundle);
2438 bundle_update(struct ofbundle *bundle)
2440 struct ofport_dpif *port;
2442 bundle->floodable = true;
2443 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2444 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2445 || !stp_forward_in_state(port->stp_state)) {
2446 bundle->floodable = false;
2453 bundle_del_port(struct ofport_dpif *port)
2455 struct ofbundle *bundle = port->bundle;
2457 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2459 list_remove(&port->bundle_node);
2460 port->bundle = NULL;
2463 lacp_slave_unregister(bundle->lacp, port);
2466 bond_slave_unregister(bundle->bond, port);
2469 bundle_update(bundle);
2473 bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port,
2474 struct lacp_slave_settings *lacp)
2476 struct ofport_dpif *port;
2478 port = get_ofp_port(bundle->ofproto, ofp_port);
2483 if (port->bundle != bundle) {
2484 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2486 bundle_del_port(port);
2489 port->bundle = bundle;
2490 list_push_back(&bundle->ports, &port->bundle_node);
2491 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2492 || !stp_forward_in_state(port->stp_state)) {
2493 bundle->floodable = false;
2497 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2498 lacp_slave_register(bundle->lacp, port, lacp);
2505 bundle_destroy(struct ofbundle *bundle)
2507 struct ofproto_dpif *ofproto;
2508 struct ofport_dpif *port, *next_port;
2515 ofproto = bundle->ofproto;
2516 for (i = 0; i < MAX_MIRRORS; i++) {
2517 struct ofmirror *m = ofproto->mirrors[i];
2519 if (m->out == bundle) {
2521 } else if (hmapx_find_and_delete(&m->srcs, bundle)
2522 || hmapx_find_and_delete(&m->dsts, bundle)) {
2523 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2528 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2529 bundle_del_port(port);
2532 bundle_flush_macs(bundle, true);
2533 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2535 free(bundle->trunks);
2536 lacp_destroy(bundle->lacp);
2537 bond_destroy(bundle->bond);
2542 bundle_set(struct ofproto *ofproto_, void *aux,
2543 const struct ofproto_bundle_settings *s)
2545 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2546 bool need_flush = false;
2547 struct ofport_dpif *port;
2548 struct ofbundle *bundle;
2549 unsigned long *trunks;
2555 bundle_destroy(bundle_lookup(ofproto, aux));
2559 ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2560 ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2562 bundle = bundle_lookup(ofproto, aux);
2564 bundle = xmalloc(sizeof *bundle);
2566 bundle->ofproto = ofproto;
2567 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2568 hash_pointer(aux, 0));
2570 bundle->name = NULL;
2572 list_init(&bundle->ports);
2573 bundle->vlan_mode = PORT_VLAN_TRUNK;
2575 bundle->trunks = NULL;
2576 bundle->use_priority_tags = s->use_priority_tags;
2577 bundle->lacp = NULL;
2578 bundle->bond = NULL;
2580 bundle->floodable = true;
2582 bundle->src_mirrors = 0;
2583 bundle->dst_mirrors = 0;
2584 bundle->mirror_out = 0;
2587 if (!bundle->name || strcmp(s->name, bundle->name)) {
2589 bundle->name = xstrdup(s->name);
2594 if (!bundle->lacp) {
2595 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2596 bundle->lacp = lacp_create();
2598 lacp_configure(bundle->lacp, s->lacp);
2600 lacp_destroy(bundle->lacp);
2601 bundle->lacp = NULL;
2604 /* Update set of ports. */
2606 for (i = 0; i < s->n_slaves; i++) {
2607 if (!bundle_add_port(bundle, s->slaves[i],
2608 s->lacp ? &s->lacp_slaves[i] : NULL)) {
2612 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2613 struct ofport_dpif *next_port;
2615 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2616 for (i = 0; i < s->n_slaves; i++) {
2617 if (s->slaves[i] == port->up.ofp_port) {
2622 bundle_del_port(port);
2626 ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2628 if (list_is_empty(&bundle->ports)) {
2629 bundle_destroy(bundle);
2633 /* Set VLAN tagging mode */
2634 if (s->vlan_mode != bundle->vlan_mode
2635 || s->use_priority_tags != bundle->use_priority_tags) {
2636 bundle->vlan_mode = s->vlan_mode;
2637 bundle->use_priority_tags = s->use_priority_tags;
2642 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2643 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2645 if (vlan != bundle->vlan) {
2646 bundle->vlan = vlan;
2650 /* Get trunked VLANs. */
2651 switch (s->vlan_mode) {
2652 case PORT_VLAN_ACCESS:
2656 case PORT_VLAN_TRUNK:
2657 trunks = CONST_CAST(unsigned long *, s->trunks);
2660 case PORT_VLAN_NATIVE_UNTAGGED:
2661 case PORT_VLAN_NATIVE_TAGGED:
2662 if (vlan != 0 && (!s->trunks
2663 || !bitmap_is_set(s->trunks, vlan)
2664 || bitmap_is_set(s->trunks, 0))) {
2665 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2667 trunks = bitmap_clone(s->trunks, 4096);
2669 trunks = bitmap_allocate1(4096);
2671 bitmap_set1(trunks, vlan);
2672 bitmap_set0(trunks, 0);
2674 trunks = CONST_CAST(unsigned long *, s->trunks);
2681 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2682 free(bundle->trunks);
2683 if (trunks == s->trunks) {
2684 bundle->trunks = vlan_bitmap_clone(trunks);
2686 bundle->trunks = trunks;
2691 if (trunks != s->trunks) {
2696 if (!list_is_short(&bundle->ports)) {
2697 bundle->ofproto->has_bonded_bundles = true;
2699 if (bond_reconfigure(bundle->bond, s->bond)) {
2700 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2703 bundle->bond = bond_create(s->bond);
2704 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2707 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2708 bond_slave_register(bundle->bond, port, port->up.netdev);
2711 bond_destroy(bundle->bond);
2712 bundle->bond = NULL;
2715 /* If we changed something that would affect MAC learning, un-learn
2716 * everything on this port and force flow revalidation. */
2718 bundle_flush_macs(bundle, false);
2725 bundle_remove(struct ofport *port_)
2727 struct ofport_dpif *port = ofport_dpif_cast(port_);
2728 struct ofbundle *bundle = port->bundle;
2731 bundle_del_port(port);
2732 if (list_is_empty(&bundle->ports)) {
2733 bundle_destroy(bundle);
2734 } else if (list_is_short(&bundle->ports)) {
2735 bond_destroy(bundle->bond);
2736 bundle->bond = NULL;
2742 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2744 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2745 struct ofport_dpif *port = port_;
2746 uint8_t ea[ETH_ADDR_LEN];
2749 error = netdev_get_etheraddr(port->up.netdev, ea);
2751 struct ofpbuf packet;
2754 ofpbuf_init(&packet, 0);
2755 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2757 memcpy(packet_pdu, pdu, pdu_size);
2759 send_packet(port, &packet);
2760 ofpbuf_uninit(&packet);
2762 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2763 "%s (%s)", port->bundle->name,
2764 netdev_get_name(port->up.netdev), strerror(error));
2769 bundle_send_learning_packets(struct ofbundle *bundle)
2771 struct ofproto_dpif *ofproto = bundle->ofproto;
2772 int error, n_packets, n_errors;
2773 struct mac_entry *e;
2775 error = n_packets = n_errors = 0;
2776 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2777 if (e->port.p != bundle) {
2778 struct ofpbuf *learning_packet;
2779 struct ofport_dpif *port;
2783 /* The assignment to "port" is unnecessary but makes "grep"ing for
2784 * struct ofport_dpif more effective. */
2785 learning_packet = bond_compose_learning_packet(bundle->bond,
2789 ret = send_packet(port, learning_packet);
2790 ofpbuf_delete(learning_packet);
2800 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2801 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2802 "packets, last error was: %s",
2803 bundle->name, n_errors, n_packets, strerror(error));
2805 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2806 bundle->name, n_packets);
2811 bundle_run(struct ofbundle *bundle)
2814 lacp_run(bundle->lacp, send_pdu_cb);
2817 struct ofport_dpif *port;
2819 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2820 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2823 bond_run(bundle->bond, &bundle->ofproto->backer->revalidate_set,
2824 lacp_status(bundle->lacp));
2825 if (bond_should_send_learning_packets(bundle->bond)) {
2826 bundle_send_learning_packets(bundle);
2832 bundle_wait(struct ofbundle *bundle)
2835 lacp_wait(bundle->lacp);
2838 bond_wait(bundle->bond);
2845 mirror_scan(struct ofproto_dpif *ofproto)
2849 for (idx = 0; idx < MAX_MIRRORS; idx++) {
2850 if (!ofproto->mirrors[idx]) {
2857 static struct ofmirror *
2858 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
2862 for (i = 0; i < MAX_MIRRORS; i++) {
2863 struct ofmirror *mirror = ofproto->mirrors[i];
2864 if (mirror && mirror->aux == aux) {
2872 /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */
2874 mirror_update_dups(struct ofproto_dpif *ofproto)
2878 for (i = 0; i < MAX_MIRRORS; i++) {
2879 struct ofmirror *m = ofproto->mirrors[i];
2882 m->dup_mirrors = MIRROR_MASK_C(1) << i;
2886 for (i = 0; i < MAX_MIRRORS; i++) {
2887 struct ofmirror *m1 = ofproto->mirrors[i];
2894 for (j = i + 1; j < MAX_MIRRORS; j++) {
2895 struct ofmirror *m2 = ofproto->mirrors[j];
2897 if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) {
2898 m1->dup_mirrors |= MIRROR_MASK_C(1) << j;
2899 m2->dup_mirrors |= m1->dup_mirrors;
2906 mirror_set(struct ofproto *ofproto_, void *aux,
2907 const struct ofproto_mirror_settings *s)
2909 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2910 mirror_mask_t mirror_bit;
2911 struct ofbundle *bundle;
2912 struct ofmirror *mirror;
2913 struct ofbundle *out;
2914 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
2915 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
2918 mirror = mirror_lookup(ofproto, aux);
2920 mirror_destroy(mirror);
2926 idx = mirror_scan(ofproto);
2928 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
2930 ofproto->up.name, MAX_MIRRORS, s->name);
2934 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
2935 mirror->ofproto = ofproto;
2938 mirror->out_vlan = -1;
2939 mirror->name = NULL;
2942 if (!mirror->name || strcmp(s->name, mirror->name)) {
2944 mirror->name = xstrdup(s->name);
2947 /* Get the new configuration. */
2948 if (s->out_bundle) {
2949 out = bundle_lookup(ofproto, s->out_bundle);
2951 mirror_destroy(mirror);
2957 out_vlan = s->out_vlan;
2959 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
2960 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
2962 /* If the configuration has not changed, do nothing. */
2963 if (hmapx_equals(&srcs, &mirror->srcs)
2964 && hmapx_equals(&dsts, &mirror->dsts)
2965 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
2966 && mirror->out == out
2967 && mirror->out_vlan == out_vlan)
2969 hmapx_destroy(&srcs);
2970 hmapx_destroy(&dsts);
2974 hmapx_swap(&srcs, &mirror->srcs);
2975 hmapx_destroy(&srcs);
2977 hmapx_swap(&dsts, &mirror->dsts);
2978 hmapx_destroy(&dsts);
2980 free(mirror->vlans);
2981 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
2984 mirror->out_vlan = out_vlan;
2986 /* Update bundles. */
2987 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2988 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
2989 if (hmapx_contains(&mirror->srcs, bundle)) {
2990 bundle->src_mirrors |= mirror_bit;
2992 bundle->src_mirrors &= ~mirror_bit;
2995 if (hmapx_contains(&mirror->dsts, bundle)) {
2996 bundle->dst_mirrors |= mirror_bit;
2998 bundle->dst_mirrors &= ~mirror_bit;
3001 if (mirror->out == bundle) {
3002 bundle->mirror_out |= mirror_bit;
3004 bundle->mirror_out &= ~mirror_bit;
3008 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3009 ofproto->has_mirrors = true;
3010 mac_learning_flush(ofproto->ml,
3011 &ofproto->backer->revalidate_set);
3012 mirror_update_dups(ofproto);
3018 mirror_destroy(struct ofmirror *mirror)
3020 struct ofproto_dpif *ofproto;
3021 mirror_mask_t mirror_bit;
3022 struct ofbundle *bundle;
3029 ofproto = mirror->ofproto;
3030 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3031 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
3033 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
3034 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
3035 bundle->src_mirrors &= ~mirror_bit;
3036 bundle->dst_mirrors &= ~mirror_bit;
3037 bundle->mirror_out &= ~mirror_bit;
3040 hmapx_destroy(&mirror->srcs);
3041 hmapx_destroy(&mirror->dsts);
3042 free(mirror->vlans);
3044 ofproto->mirrors[mirror->idx] = NULL;
3048 mirror_update_dups(ofproto);
3050 ofproto->has_mirrors = false;
3051 for (i = 0; i < MAX_MIRRORS; i++) {
3052 if (ofproto->mirrors[i]) {
3053 ofproto->has_mirrors = true;
3060 mirror_get_stats(struct ofproto *ofproto_, void *aux,
3061 uint64_t *packets, uint64_t *bytes)
3063 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3064 struct ofmirror *mirror = mirror_lookup(ofproto, aux);
3067 *packets = *bytes = UINT64_MAX;
3073 *packets = mirror->packet_count;
3074 *bytes = mirror->byte_count;
3080 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
3082 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3083 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
3084 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
3090 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
3092 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3093 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
3094 return bundle && bundle->mirror_out != 0;
3098 forward_bpdu_changed(struct ofproto *ofproto_)
3100 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3101 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3105 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
3108 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3109 mac_learning_set_idle_time(ofproto->ml, idle_time);
3110 mac_learning_set_max_entries(ofproto->ml, max_entries);
3115 static struct ofport_dpif *
3116 get_ofp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
3118 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
3119 return ofport ? ofport_dpif_cast(ofport) : NULL;
3122 static struct ofport_dpif *
3123 get_odp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
3125 struct ofport_dpif *port = odp_port_to_ofport(ofproto->backer, odp_port);
3126 return port && &ofproto->up == port->up.ofproto ? port : NULL;
3130 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
3131 struct ofproto_port *ofproto_port,
3132 struct dpif_port *dpif_port)
3134 ofproto_port->name = dpif_port->name;
3135 ofproto_port->type = dpif_port->type;
3136 ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
3139 static struct ofport_dpif *
3140 ofport_get_peer(const struct ofport_dpif *ofport_dpif)
3142 const struct ofproto_dpif *ofproto;
3145 peer = netdev_vport_patch_peer(ofport_dpif->up.netdev);
3150 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
3151 struct ofport *ofport;
3153 ofport = shash_find_data(&ofproto->up.port_by_name, peer);
3154 if (ofport && ofport->ofproto->ofproto_class == &ofproto_dpif_class) {
3155 return ofport_dpif_cast(ofport);
3162 port_run_fast(struct ofport_dpif *ofport)
3164 if (ofport->cfm && cfm_should_send_ccm(ofport->cfm)) {
3165 struct ofpbuf packet;
3167 ofpbuf_init(&packet, 0);
3168 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
3169 send_packet(ofport, &packet);
3170 ofpbuf_uninit(&packet);
3175 port_run(struct ofport_dpif *ofport)
3177 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
3178 bool carrier_changed = carrier_seq != ofport->carrier_seq;
3179 bool enable = netdev_get_carrier(ofport->up.netdev);
3181 ofport->carrier_seq = carrier_seq;
3183 port_run_fast(ofport);
3185 if (ofport->tnl_port
3186 && tnl_port_reconfigure(&ofport->up, ofport->odp_port,
3187 &ofport->tnl_port)) {
3188 ofproto_dpif_cast(ofport->up.ofproto)->backer->need_revalidate = true;
3192 int cfm_opup = cfm_get_opup(ofport->cfm);
3194 cfm_run(ofport->cfm);
3195 enable = enable && !cfm_get_fault(ofport->cfm);
3197 if (cfm_opup >= 0) {
3198 enable = enable && cfm_opup;
3202 if (ofport->bundle) {
3203 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
3204 if (carrier_changed) {
3205 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
3209 if (ofport->may_enable != enable) {
3210 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3212 if (ofproto->has_bundle_action) {
3213 ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
3217 ofport->may_enable = enable;
3221 port_wait(struct ofport_dpif *ofport)
3224 cfm_wait(ofport->cfm);
3229 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
3230 struct ofproto_port *ofproto_port)
3232 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3233 struct dpif_port dpif_port;
3236 if (sset_contains(&ofproto->ghost_ports, devname)) {
3237 const char *type = netdev_get_type_from_name(devname);
3239 /* We may be called before ofproto->up.port_by_name is populated with
3240 * the appropriate ofport. For this reason, we must get the name and
3241 * type from the netdev layer directly. */
3243 const struct ofport *ofport;
3245 ofport = shash_find_data(&ofproto->up.port_by_name, devname);
3246 ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
3247 ofproto_port->name = xstrdup(devname);
3248 ofproto_port->type = xstrdup(type);
3254 if (!sset_contains(&ofproto->ports, devname)) {
3257 error = dpif_port_query_by_name(ofproto->backer->dpif,
3258 devname, &dpif_port);
3260 ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
3266 port_add(struct ofproto *ofproto_, struct netdev *netdev)
3268 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3269 const char *dp_port_name = netdev_vport_get_dpif_port(netdev);
3270 const char *devname = netdev_get_name(netdev);
3272 if (netdev_vport_is_patch(netdev)) {
3273 sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
3277 if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
3278 uint32_t port_no = UINT32_MAX;
3281 error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
3285 if (netdev_get_tunnel_config(netdev)) {
3286 simap_put(&ofproto->backer->tnl_backers, dp_port_name, port_no);
3290 if (netdev_get_tunnel_config(netdev)) {
3291 sset_add(&ofproto->ghost_ports, devname);
3293 sset_add(&ofproto->ports, devname);
3299 port_del(struct ofproto *ofproto_, uint16_t ofp_port)
3301 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3302 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
3309 sset_find_and_delete(&ofproto->ghost_ports,
3310 netdev_get_name(ofport->up.netdev));
3311 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3312 if (!ofport->tnl_port) {
3313 error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
3315 /* The caller is going to close ofport->up.netdev. If this is a
3316 * bonded port, then the bond is using that netdev, so remove it
3317 * from the bond. The client will need to reconfigure everything
3318 * after deleting ports, so then the slave will get re-added. */
3319 bundle_remove(&ofport->up);
3326 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
3328 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3333 error = netdev_get_stats(ofport->up.netdev, stats);
3335 if (!error && ofport_->ofp_port == OFPP_LOCAL) {
3336 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3338 /* ofproto->stats.tx_packets represents packets that we created
3339 * internally and sent to some port (e.g. packets sent with
3340 * send_packet()). Account for them as if they had come from
3341 * OFPP_LOCAL and got forwarded. */
3343 if (stats->rx_packets != UINT64_MAX) {
3344 stats->rx_packets += ofproto->stats.tx_packets;
3347 if (stats->rx_bytes != UINT64_MAX) {
3348 stats->rx_bytes += ofproto->stats.tx_bytes;
3351 /* ofproto->stats.rx_packets represents packets that were received on
3352 * some port and we processed internally and dropped (e.g. STP).
3353 * Account for them as if they had been forwarded to OFPP_LOCAL. */
3355 if (stats->tx_packets != UINT64_MAX) {
3356 stats->tx_packets += ofproto->stats.rx_packets;
3359 if (stats->tx_bytes != UINT64_MAX) {
3360 stats->tx_bytes += ofproto->stats.rx_bytes;
3367 /* Account packets for LOCAL port. */
3369 ofproto_update_local_port_stats(const struct ofproto *ofproto_,
3370 size_t tx_size, size_t rx_size)
3372 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3375 ofproto->stats.rx_packets++;
3376 ofproto->stats.rx_bytes += rx_size;
3379 ofproto->stats.tx_packets++;
3380 ofproto->stats.tx_bytes += tx_size;
3384 struct port_dump_state {
3389 struct ofproto_port port;
3394 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
3396 *statep = xzalloc(sizeof(struct port_dump_state));
3401 port_dump_next(const struct ofproto *ofproto_, void *state_,
3402 struct ofproto_port *port)
3404 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3405 struct port_dump_state *state = state_;
3406 const struct sset *sset;
3407 struct sset_node *node;
3409 if (state->has_port) {
3410 ofproto_port_destroy(&state->port);
3411 state->has_port = false;
3413 sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
3414 while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
3417 error = port_query_by_name(ofproto_, node->name, &state->port);
3419 *port = state->port;
3420 state->has_port = true;
3422 } else if (error != ENODEV) {
3427 if (!state->ghost) {
3428 state->ghost = true;
3431 return port_dump_next(ofproto_, state_, port);
3438 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3440 struct port_dump_state *state = state_;
3442 if (state->has_port) {
3443 ofproto_port_destroy(&state->port);
3450 port_poll(const struct ofproto *ofproto_, char **devnamep)
3452 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3454 if (ofproto->port_poll_errno) {
3455 int error = ofproto->port_poll_errno;
3456 ofproto->port_poll_errno = 0;
3460 if (sset_is_empty(&ofproto->port_poll_set)) {
3464 *devnamep = sset_pop(&ofproto->port_poll_set);
3469 port_poll_wait(const struct ofproto *ofproto_)
3471 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3472 dpif_port_poll_wait(ofproto->backer->dpif);
3476 port_is_lacp_current(const struct ofport *ofport_)
3478 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3479 return (ofport->bundle && ofport->bundle->lacp
3480 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3484 /* Upcall handling. */
3486 /* Flow miss batching.
3488 * Some dpifs implement operations faster when you hand them off in a batch.
3489 * To allow batching, "struct flow_miss" queues the dpif-related work needed
3490 * for a given flow. Each "struct flow_miss" corresponds to sending one or
3491 * more packets, plus possibly installing the flow in the dpif.
3493 * So far we only batch the operations that affect flow setup time the most.
3494 * It's possible to batch more than that, but the benefit might be minimal. */
3496 struct hmap_node hmap_node;
3497 struct ofproto_dpif *ofproto;
3499 enum odp_key_fitness key_fitness;
3500 const struct nlattr *key;
3502 struct initial_vals initial_vals;
3503 struct list packets;
3504 enum dpif_upcall_type upcall_type;
3505 uint32_t odp_in_port;
3508 struct flow_miss_op {
3509 struct dpif_op dpif_op;
3511 uint64_t slow_stub[128 / 8]; /* Buffer for compose_slow_path() */
3512 struct xlate_out xout;
3513 bool xout_garbage; /* 'xout' needs to be uninitialized? */
3516 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
3517 * OpenFlow controller as necessary according to their individual
3518 * configurations. */
3520 send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet,
3521 const struct flow *flow)
3523 struct ofputil_packet_in pin;
3525 pin.packet = packet->data;
3526 pin.packet_len = packet->size;
3527 pin.reason = OFPR_NO_MATCH;
3528 pin.controller_id = 0;
3533 pin.send_len = 0; /* not used for flow table misses */
3535 flow_get_metadata(flow, &pin.fmd);
3537 connmgr_send_packet_in(ofproto->up.connmgr, &pin);
3540 static enum slow_path_reason
3541 process_special(struct ofproto_dpif *ofproto, const struct flow *flow,
3542 const struct ofport_dpif *ofport, const struct ofpbuf *packet)
3546 } else if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) {
3548 cfm_process_heartbeat(ofport->cfm, packet);
3551 } else if (ofport->bundle && ofport->bundle->lacp
3552 && flow->dl_type == htons(ETH_TYPE_LACP)) {
3554 lacp_process_packet(ofport->bundle->lacp, ofport, packet);
3557 } else if (ofproto->stp && stp_should_process_flow(flow)) {
3559 stp_process_packet(ofport, packet);
3567 static struct flow_miss *
3568 flow_miss_find(struct hmap *todo, const struct ofproto_dpif *ofproto,
3569 const struct flow *flow, uint32_t hash)
3571 struct flow_miss *miss;
3573 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
3574 if (miss->ofproto == ofproto && flow_equal(&miss->flow, flow)) {
3582 /* Partially Initializes 'op' as an "execute" operation for 'miss' and
3583 * 'packet'. The caller must initialize op->actions and op->actions_len. If
3584 * 'miss' is associated with a subfacet the caller must also initialize the
3585 * returned op->subfacet, and if anything needs to be freed after processing
3586 * the op, the caller must initialize op->garbage also. */
3588 init_flow_miss_execute_op(struct flow_miss *miss, struct ofpbuf *packet,
3589 struct flow_miss_op *op)
3591 if (miss->flow.vlan_tci != miss->initial_vals.vlan_tci) {
3592 /* This packet was received on a VLAN splinter port. We
3593 * added a VLAN to the packet to make the packet resemble
3594 * the flow, but the actions were composed assuming that
3595 * the packet contained no VLAN. So, we must remove the
3596 * VLAN header from the packet before trying to execute the
3598 eth_pop_vlan(packet);
3601 op->xout_garbage = false;
3602 op->dpif_op.type = DPIF_OP_EXECUTE;
3603 op->dpif_op.u.execute.key = miss->key;
3604 op->dpif_op.u.execute.key_len = miss->key_len;
3605 op->dpif_op.u.execute.packet = packet;
3608 /* Helper for handle_flow_miss_without_facet() and
3609 * handle_flow_miss_with_facet(). */
3611 handle_flow_miss_common(struct rule_dpif *rule,
3612 struct ofpbuf *packet, const struct flow *flow)
3614 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3616 ofproto->n_matches++;
3618 if (rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
3620 * Extra-special case for fail-open mode.
3622 * We are in fail-open mode and the packet matched the fail-open
3623 * rule, but we are connected to a controller too. We should send
3624 * the packet up to the controller in the hope that it will try to
3625 * set up a flow and thereby allow us to exit fail-open.
3627 * See the top-level comment in fail-open.c for more information.
3629 send_packet_in_miss(ofproto, packet, flow);
3633 /* Figures out whether a flow that missed in 'ofproto', whose details are in
3634 * 'miss', is likely to be worth tracking in detail in userspace and (usually)
3635 * installing a datapath flow. The answer is usually "yes" (a return value of
3636 * true). However, for short flows the cost of bookkeeping is much higher than
3637 * the benefits, so when the datapath holds a large number of flows we impose
3638 * some heuristics to decide which flows are likely to be worth tracking. */
3640 flow_miss_should_make_facet(struct ofproto_dpif *ofproto,
3641 struct flow_miss *miss, uint32_t hash)
3643 if (!ofproto->governor) {
3646 n_subfacets = hmap_count(&ofproto->subfacets);
3647 if (n_subfacets * 2 <= ofproto->up.flow_eviction_threshold) {
3651 ofproto->governor = governor_create(ofproto->up.name);
3654 return governor_should_install_flow(ofproto->governor, hash,
3655 list_size(&miss->packets));
3658 /* Handles 'miss', which matches 'rule', without creating a facet or subfacet
3659 * or creating any datapath flow. May add an "execute" operation to 'ops' and
3660 * increment '*n_ops'. */
3662 handle_flow_miss_without_facet(struct flow_miss *miss,
3663 struct flow_miss_op *ops, size_t *n_ops)
3665 struct rule_dpif *rule = rule_dpif_lookup(miss->ofproto, &miss->flow);
3666 long long int now = time_msec();
3667 struct ofpbuf *packet;
3668 struct xlate_in xin;
3670 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3671 struct flow_miss_op *op = &ops[*n_ops];
3672 struct dpif_flow_stats stats;
3674 COVERAGE_INC(facet_suppress);
3676 handle_flow_miss_common(rule, packet, &miss->flow);
3678 dpif_flow_stats_extract(&miss->flow, packet, now, &stats);
3679 rule_credit_stats(rule, &stats);
3681 xlate_in_init(&xin, miss->ofproto, &miss->flow, &miss->initial_vals,
3682 rule, stats.tcp_flags, packet);
3683 xin.resubmit_stats = &stats;
3684 xlate_actions(&xin, &op->xout);
3686 if (op->xout.odp_actions.size) {
3687 struct dpif_execute *execute = &op->dpif_op.u.execute;
3689 init_flow_miss_execute_op(miss, packet, op);
3690 execute->actions = op->xout.odp_actions.data;
3691 execute->actions_len = op->xout.odp_actions.size;
3692 op->xout_garbage = true;
3696 xlate_out_uninit(&op->xout);
3701 /* Handles 'miss', which matches 'facet'. May add any required datapath
3702 * operations to 'ops', incrementing '*n_ops' for each new op.
3704 * All of the packets in 'miss' are considered to have arrived at time 'now'.
3705 * This is really important only for new facets: if we just called time_msec()
3706 * here, then the new subfacet or its packets could look (occasionally) as
3707 * though it was used some time after the facet was used. That can make a
3708 * one-packet flow look like it has a nonzero duration, which looks odd in
3709 * e.g. NetFlow statistics. */
3711 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
3713 struct flow_miss_op *ops, size_t *n_ops)
3715 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3716 enum subfacet_path want_path;
3717 struct subfacet *subfacet;
3718 struct ofpbuf *packet;
3720 subfacet = subfacet_create(facet, miss, now);
3721 want_path = subfacet->facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
3723 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3724 struct flow_miss_op *op = &ops[*n_ops];
3725 struct dpif_flow_stats stats;
3727 handle_flow_miss_common(facet->rule, packet, &miss->flow);
3729 if (want_path != SF_FAST_PATH) {
3730 struct xlate_in xin;
3732 xlate_in_init(&xin, ofproto, &facet->flow, &facet->initial_vals,
3733 facet->rule, 0, packet);
3734 xlate_actions_for_side_effects(&xin);
3737 dpif_flow_stats_extract(&facet->flow, packet, now, &stats);
3738 subfacet_update_stats(subfacet, &stats);
3740 if (facet->xout.odp_actions.size) {
3741 struct dpif_execute *execute = &op->dpif_op.u.execute;
3743 init_flow_miss_execute_op(miss, packet, op);
3744 execute->actions = facet->xout.odp_actions.data,
3745 execute->actions_len = facet->xout.odp_actions.size;
3750 if (miss->upcall_type == DPIF_UC_MISS || subfacet->path != want_path) {
3751 struct flow_miss_op *op = &ops[(*n_ops)++];
3752 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
3754 subfacet->path = want_path;
3756 op->xout_garbage = false;
3757 op->dpif_op.type = DPIF_OP_FLOW_PUT;
3758 put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
3759 put->key = miss->key;
3760 put->key_len = miss->key_len;
3761 if (want_path == SF_FAST_PATH) {
3762 put->actions = facet->xout.odp_actions.data;
3763 put->actions_len = facet->xout.odp_actions.size;
3765 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
3766 op->slow_stub, sizeof op->slow_stub,
3767 &put->actions, &put->actions_len);
3773 /* Handles flow miss 'miss'. May add any required datapath operations
3774 * to 'ops', incrementing '*n_ops' for each new op. */
3776 handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops,
3779 struct ofproto_dpif *ofproto = miss->ofproto;
3780 struct facet *facet;
3784 /* The caller must ensure that miss->hmap_node.hash contains
3785 * flow_hash(miss->flow, 0). */
3786 hash = miss->hmap_node.hash;
3788 facet = facet_lookup_valid(ofproto, &miss->flow, hash);
3790 /* There does not exist a bijection between 'struct flow' and datapath
3791 * flow keys with fitness ODP_FIT_TO_LITTLE. This breaks a fundamental
3792 * assumption used throughout the facet and subfacet handling code.
3793 * Since we have to handle these misses in userspace anyway, we simply
3794 * skip facet creation, avoiding the problem alltogether. */
3795 if (miss->key_fitness == ODP_FIT_TOO_LITTLE
3796 || !flow_miss_should_make_facet(ofproto, miss, hash)) {
3797 handle_flow_miss_without_facet(miss, ops, n_ops);
3801 facet = facet_create(miss, hash);
3806 handle_flow_miss_with_facet(miss, facet, now, ops, n_ops);
3809 static struct drop_key *
3810 drop_key_lookup(const struct dpif_backer *backer, const struct nlattr *key,
3813 struct drop_key *drop_key;
3815 HMAP_FOR_EACH_WITH_HASH (drop_key, hmap_node, hash_bytes(key, key_len, 0),
3816 &backer->drop_keys) {
3817 if (drop_key->key_len == key_len
3818 && !memcmp(drop_key->key, key, key_len)) {
3826 drop_key_clear(struct dpif_backer *backer)
3828 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3829 struct drop_key *drop_key, *next;
3831 HMAP_FOR_EACH_SAFE (drop_key, next, hmap_node, &backer->drop_keys) {
3834 error = dpif_flow_del(backer->dpif, drop_key->key, drop_key->key_len,
3836 if (error && !VLOG_DROP_WARN(&rl)) {
3837 struct ds ds = DS_EMPTY_INITIALIZER;
3838 odp_flow_key_format(drop_key->key, drop_key->key_len, &ds);
3839 VLOG_WARN("Failed to delete drop key (%s) (%s)", strerror(error),
3844 hmap_remove(&backer->drop_keys, &drop_key->hmap_node);
3845 free(drop_key->key);
3850 /* Given a datpath, packet, and flow metadata ('backer', 'packet', and 'key'
3851 * respectively), populates 'flow' with the result of odp_flow_key_to_flow().
3852 * Optionally, if nonnull, populates 'fitnessp' with the fitness of 'flow' as
3853 * returned by odp_flow_key_to_flow(). Also, optionally populates 'ofproto'
3854 * with the ofproto_dpif, and 'odp_in_port' with the datapath in_port, that
3855 * 'packet' ingressed.
3857 * If 'ofproto' is nonnull, requires 'flow''s in_port to exist. Otherwise sets
3858 * 'flow''s in_port to OFPP_NONE.
3860 * This function does post-processing on data returned from
3861 * odp_flow_key_to_flow() to help make VLAN splinters transparent to the rest
3862 * of the upcall processing logic. In particular, if the extracted in_port is
3863 * a VLAN splinter port, it replaces flow->in_port by the "real" port, sets
3864 * flow->vlan_tci correctly for the VLAN of the VLAN splinter port, and pushes
3865 * a VLAN header onto 'packet' (if it is nonnull).
3867 * Optionally, if 'initial_vals' is nonnull, sets 'initial_vals->vlan_tci'
3868 * to the VLAN TCI with which the packet was really received, that is, the
3869 * actual VLAN TCI extracted by odp_flow_key_to_flow(). (This differs from
3870 * the value returned in flow->vlan_tci only for packets received on
3871 * VLAN splinters.) Also, if received on an IP tunnel, sets
3872 * 'initial_vals->tunnel_ip_tos' to the tunnel's IP TOS.
3874 * Similarly, this function also includes some logic to help with tunnels. It
3875 * may modify 'flow' as necessary to make the tunneling implementation
3876 * transparent to the upcall processing logic.
3878 * Returns 0 if successful, ENODEV if the parsed flow has no associated ofport,
3879 * or some other positive errno if there are other problems. */
3881 ofproto_receive(const struct dpif_backer *backer, struct ofpbuf *packet,
3882 const struct nlattr *key, size_t key_len,
3883 struct flow *flow, enum odp_key_fitness *fitnessp,
3884 struct ofproto_dpif **ofproto, uint32_t *odp_in_port,
3885 struct initial_vals *initial_vals)
3887 const struct ofport_dpif *port;
3888 enum odp_key_fitness fitness;
3891 fitness = odp_flow_key_to_flow(key, key_len, flow);
3892 if (fitness == ODP_FIT_ERROR) {
3898 initial_vals->vlan_tci = flow->vlan_tci;
3899 initial_vals->tunnel_ip_tos = flow->tunnel.ip_tos;
3903 *odp_in_port = flow->in_port;
3906 if (tnl_port_should_receive(flow)) {
3907 const struct ofport *ofport = tnl_port_receive(flow);
3909 flow->in_port = OFPP_NONE;
3912 port = ofport_dpif_cast(ofport);
3914 /* We can't reproduce 'key' from 'flow'. */
3915 fitness = fitness == ODP_FIT_PERFECT ? ODP_FIT_TOO_MUCH : fitness;
3917 /* XXX: Since the tunnel module is not scoped per backer, it's
3918 * theoretically possible that we'll receive an ofport belonging to an
3919 * entirely different datapath. In practice, this can't happen because
3920 * no platforms has two separate datapaths which each support
3922 ovs_assert(ofproto_dpif_cast(port->up.ofproto)->backer == backer);
3924 port = odp_port_to_ofport(backer, flow->in_port);
3926 flow->in_port = OFPP_NONE;
3930 flow->in_port = port->up.ofp_port;
3931 if (vsp_adjust_flow(ofproto_dpif_cast(port->up.ofproto), flow)) {
3933 /* Make the packet resemble the flow, so that it gets sent to
3934 * an OpenFlow controller properly, so that it looks correct
3935 * for sFlow, and so that flow_extract() will get the correct
3936 * vlan_tci if it is called on 'packet'.
3938 * The allocated space inside 'packet' probably also contains
3939 * 'key', that is, both 'packet' and 'key' are probably part of
3940 * a struct dpif_upcall (see the large comment on that
3941 * structure definition), so pushing data on 'packet' is in
3942 * general not a good idea since it could overwrite 'key' or
3943 * free it as a side effect. However, it's OK in this special
3944 * case because we know that 'packet' is inside a Netlink
3945 * attribute: pushing 4 bytes will just overwrite the 4-byte
3946 * "struct nlattr", which is fine since we don't need that
3947 * header anymore. */
3948 eth_push_vlan(packet, flow->vlan_tci);
3950 /* We can't reproduce 'key' from 'flow'. */
3951 fitness = fitness == ODP_FIT_PERFECT ? ODP_FIT_TOO_MUCH : fitness;
3957 *ofproto = ofproto_dpif_cast(port->up.ofproto);
3962 *fitnessp = fitness;
3968 handle_miss_upcalls(struct dpif_backer *backer, struct dpif_upcall *upcalls,
3971 struct dpif_upcall *upcall;
3972 struct flow_miss *miss;
3973 struct flow_miss misses[FLOW_MISS_MAX_BATCH];
3974 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
3975 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
3985 /* Construct the to-do list.
3987 * This just amounts to extracting the flow from each packet and sticking
3988 * the packets that have the same flow in the same "flow_miss" structure so
3989 * that we can process them together. */
3992 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
3993 struct flow_miss *miss = &misses[n_misses];
3994 struct flow_miss *existing_miss;
3995 struct ofproto_dpif *ofproto;
3996 uint32_t odp_in_port;
4001 error = ofproto_receive(backer, upcall->packet, upcall->key,
4002 upcall->key_len, &flow, &miss->key_fitness,
4003 &ofproto, &odp_in_port, &miss->initial_vals);
4004 if (error == ENODEV) {
4005 struct drop_key *drop_key;
4007 /* Received packet on port for which we couldn't associate
4008 * an ofproto. This can happen if a port is removed while
4009 * traffic is being received. Print a rate-limited message
4010 * in case it happens frequently. Install a drop flow so
4011 * that future packets of the flow are inexpensively dropped
4013 VLOG_INFO_RL(&rl, "received packet on unassociated port %"PRIu32,
4016 drop_key = drop_key_lookup(backer, upcall->key, upcall->key_len);
4018 drop_key = xmalloc(sizeof *drop_key);
4019 drop_key->key = xmemdup(upcall->key, upcall->key_len);
4020 drop_key->key_len = upcall->key_len;
4022 hmap_insert(&backer->drop_keys, &drop_key->hmap_node,
4023 hash_bytes(drop_key->key, drop_key->key_len, 0));
4024 dpif_flow_put(backer->dpif, DPIF_FP_CREATE | DPIF_FP_MODIFY,
4025 drop_key->key, drop_key->key_len, NULL, 0, NULL);
4033 ofproto->n_missed++;
4034 flow_extract(upcall->packet, flow.skb_priority, flow.skb_mark,
4035 &flow.tunnel, flow.in_port, &miss->flow);
4037 /* Add other packets to a to-do list. */
4038 hash = flow_hash(&miss->flow, 0);
4039 existing_miss = flow_miss_find(&todo, ofproto, &miss->flow, hash);
4040 if (!existing_miss) {
4041 hmap_insert(&todo, &miss->hmap_node, hash);
4042 miss->ofproto = ofproto;
4043 miss->key = upcall->key;
4044 miss->key_len = upcall->key_len;
4045 miss->upcall_type = upcall->type;
4046 miss->odp_in_port = odp_in_port;
4047 list_init(&miss->packets);
4051 miss = existing_miss;
4053 list_push_back(&miss->packets, &upcall->packet->list_node);
4056 /* Process each element in the to-do list, constructing the set of
4057 * operations to batch. */
4059 HMAP_FOR_EACH (miss, hmap_node, &todo) {
4060 handle_flow_miss(miss, flow_miss_ops, &n_ops);
4062 ovs_assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
4064 /* Execute batch. */
4065 for (i = 0; i < n_ops; i++) {
4066 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
4068 dpif_operate(backer->dpif, dpif_ops, n_ops);
4071 for (i = 0; i < n_ops; i++) {
4072 if (flow_miss_ops[i].xout_garbage) {
4073 xlate_out_uninit(&flow_miss_ops[i].xout);
4076 hmap_destroy(&todo);
4079 static enum { SFLOW_UPCALL, MISS_UPCALL, BAD_UPCALL, FLOW_SAMPLE_UPCALL,
4081 classify_upcall(const struct dpif_upcall *upcall)
4083 size_t userdata_len;
4084 union user_action_cookie cookie;
4086 /* First look at the upcall type. */
4087 switch (upcall->type) {
4088 case DPIF_UC_ACTION:
4094 case DPIF_N_UC_TYPES:
4096 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
4100 /* "action" upcalls need a closer look. */
4101 if (!upcall->userdata) {
4102 VLOG_WARN_RL(&rl, "action upcall missing cookie");
4105 userdata_len = nl_attr_get_size(upcall->userdata);
4106 if (userdata_len < sizeof cookie.type
4107 || userdata_len > sizeof cookie) {
4108 VLOG_WARN_RL(&rl, "action upcall cookie has unexpected size %zu",
4112 memset(&cookie, 0, sizeof cookie);
4113 memcpy(&cookie, nl_attr_get(upcall->userdata), userdata_len);
4114 if (userdata_len == sizeof cookie.sflow
4115 && cookie.type == USER_ACTION_COOKIE_SFLOW) {
4116 return SFLOW_UPCALL;
4117 } else if (userdata_len == sizeof cookie.slow_path
4118 && cookie.type == USER_ACTION_COOKIE_SLOW_PATH) {
4120 } else if (userdata_len == sizeof cookie.flow_sample
4121 && cookie.type == USER_ACTION_COOKIE_FLOW_SAMPLE) {
4122 return FLOW_SAMPLE_UPCALL;
4123 } else if (userdata_len == sizeof cookie.ipfix
4124 && cookie.type == USER_ACTION_COOKIE_IPFIX) {
4125 return IPFIX_UPCALL;
4127 VLOG_WARN_RL(&rl, "invalid user cookie of type %"PRIu16
4128 " and size %zu", cookie.type, userdata_len);
4134 handle_sflow_upcall(struct dpif_backer *backer,
4135 const struct dpif_upcall *upcall)
4137 struct ofproto_dpif *ofproto;
4138 union user_action_cookie cookie;
4140 uint32_t odp_in_port;
4142 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
4143 &flow, NULL, &ofproto, &odp_in_port, NULL)
4144 || !ofproto->sflow) {
4148 memset(&cookie, 0, sizeof cookie);
4149 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.sflow);
4150 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
4151 odp_in_port, &cookie);
4155 handle_flow_sample_upcall(struct dpif_backer *backer,
4156 const struct dpif_upcall *upcall)
4158 struct ofproto_dpif *ofproto;
4159 union user_action_cookie cookie;
4162 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
4163 &flow, NULL, &ofproto, NULL, NULL)
4164 || !ofproto->ipfix) {
4168 memset(&cookie, 0, sizeof cookie);
4169 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.flow_sample);
4171 /* The flow reflects exactly the contents of the packet. Sample
4172 * the packet using it. */
4173 dpif_ipfix_flow_sample(ofproto->ipfix, upcall->packet, &flow,
4174 cookie.flow_sample.collector_set_id,
4175 cookie.flow_sample.probability,
4176 cookie.flow_sample.obs_domain_id,
4177 cookie.flow_sample.obs_point_id);
4181 handle_ipfix_upcall(struct dpif_backer *backer,
4182 const struct dpif_upcall *upcall)
4184 struct ofproto_dpif *ofproto;
4187 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
4188 &flow, NULL, &ofproto, NULL, NULL)
4189 || !ofproto->ipfix) {
4193 /* The flow reflects exactly the contents of the packet. Sample
4194 * the packet using it. */
4195 dpif_ipfix_bridge_sample(ofproto->ipfix, upcall->packet, &flow);
4199 handle_upcalls(struct dpif_backer *backer, unsigned int max_batch)
4201 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
4202 struct ofpbuf miss_bufs[FLOW_MISS_MAX_BATCH];
4203 uint64_t miss_buf_stubs[FLOW_MISS_MAX_BATCH][4096 / 8];
4208 ovs_assert(max_batch <= FLOW_MISS_MAX_BATCH);
4211 for (n_processed = 0; n_processed < max_batch; n_processed++) {
4212 struct dpif_upcall *upcall = &misses[n_misses];
4213 struct ofpbuf *buf = &miss_bufs[n_misses];
4216 ofpbuf_use_stub(buf, miss_buf_stubs[n_misses],
4217 sizeof miss_buf_stubs[n_misses]);
4218 error = dpif_recv(backer->dpif, upcall, buf);
4224 switch (classify_upcall(upcall)) {
4226 /* Handle it later. */
4231 handle_sflow_upcall(backer, upcall);
4235 case FLOW_SAMPLE_UPCALL:
4236 handle_flow_sample_upcall(backer, upcall);
4241 handle_ipfix_upcall(backer, upcall);
4251 /* Handle deferred MISS_UPCALL processing. */
4252 handle_miss_upcalls(backer, misses, n_misses);
4253 for (i = 0; i < n_misses; i++) {
4254 ofpbuf_uninit(&miss_bufs[i]);
4260 /* Flow expiration. */
4262 static int subfacet_max_idle(const struct ofproto_dpif *);
4263 static void update_stats(struct dpif_backer *);
4264 static void rule_expire(struct rule_dpif *);
4265 static void expire_subfacets(struct ofproto_dpif *, int dp_max_idle);
4267 /* This function is called periodically by run(). Its job is to collect
4268 * updates for the flows that have been installed into the datapath, most
4269 * importantly when they last were used, and then use that information to
4270 * expire flows that have not been used recently.
4272 * Returns the number of milliseconds after which it should be called again. */
4274 expire(struct dpif_backer *backer)
4276 struct ofproto_dpif *ofproto;
4277 int max_idle = INT32_MAX;
4279 /* Periodically clear out the drop keys in an effort to keep them
4280 * relatively few. */
4281 drop_key_clear(backer);
4283 /* Update stats for each flow in the backer. */
4284 update_stats(backer);
4286 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4287 struct rule *rule, *next_rule;
4290 if (ofproto->backer != backer) {
4294 /* Keep track of the max number of flows per ofproto_dpif. */
4295 update_max_subfacet_count(ofproto);
4297 /* Expire subfacets that have been idle too long. */
4298 dp_max_idle = subfacet_max_idle(ofproto);
4299 expire_subfacets(ofproto, dp_max_idle);
4301 max_idle = MIN(max_idle, dp_max_idle);
4303 /* Expire OpenFlow flows whose idle_timeout or hard_timeout
4305 LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
4306 &ofproto->up.expirable) {
4307 rule_expire(rule_dpif_cast(rule));
4310 /* All outstanding data in existing flows has been accounted, so it's a
4311 * good time to do bond rebalancing. */
4312 if (ofproto->has_bonded_bundles) {
4313 struct ofbundle *bundle;
4315 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
4317 bond_rebalance(bundle->bond, &backer->revalidate_set);
4323 return MIN(max_idle, 1000);
4326 /* Updates flow table statistics given that the datapath just reported 'stats'
4327 * as 'subfacet''s statistics. */
4329 update_subfacet_stats(struct subfacet *subfacet,
4330 const struct dpif_flow_stats *stats)
4332 struct facet *facet = subfacet->facet;
4334 if (stats->n_packets >= subfacet->dp_packet_count) {
4335 uint64_t extra = stats->n_packets - subfacet->dp_packet_count;
4336 facet->packet_count += extra;
4338 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
4341 if (stats->n_bytes >= subfacet->dp_byte_count) {
4342 facet->byte_count += stats->n_bytes - subfacet->dp_byte_count;
4344 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
4347 subfacet->dp_packet_count = stats->n_packets;
4348 subfacet->dp_byte_count = stats->n_bytes;
4350 facet->tcp_flags |= stats->tcp_flags;
4352 subfacet_update_time(subfacet, stats->used);
4353 if (facet->accounted_bytes < facet->byte_count) {
4355 facet_account(facet);
4356 facet->accounted_bytes = facet->byte_count;
4360 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
4361 * about, or a flow that shouldn't be installed but was anyway. Delete it. */
4363 delete_unexpected_flow(struct ofproto_dpif *ofproto,
4364 const struct nlattr *key, size_t key_len)
4366 if (!VLOG_DROP_WARN(&rl)) {
4370 odp_flow_key_format(key, key_len, &s);
4371 VLOG_WARN("unexpected flow on %s: %s", ofproto->up.name, ds_cstr(&s));
4375 COVERAGE_INC(facet_unexpected);
4376 dpif_flow_del(ofproto->backer->dpif, key, key_len, NULL);
4379 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
4381 * This function also pushes statistics updates to rules which each facet
4382 * resubmits into. Generally these statistics will be accurate. However, if a
4383 * facet changes the rule it resubmits into at some time in between
4384 * update_stats() runs, it is possible that statistics accrued to the
4385 * old rule will be incorrectly attributed to the new rule. This could be
4386 * avoided by calling update_stats() whenever rules are created or
4387 * deleted. However, the performance impact of making so many calls to the
4388 * datapath do not justify the benefit of having perfectly accurate statistics.
4390 * In addition, this function maintains per ofproto flow hit counts. The patch
4391 * port is not treated specially. e.g. A packet ingress from br0 patched into
4392 * br1 will increase the hit count of br0 by 1, however, does not affect
4393 * the hit or miss counts of br1.
4396 update_stats(struct dpif_backer *backer)
4398 const struct dpif_flow_stats *stats;
4399 struct dpif_flow_dump dump;
4400 const struct nlattr *key;
4401 struct ofproto_dpif *ofproto;
4404 dpif_flow_dump_start(&dump, backer->dpif);
4405 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
4407 struct subfacet *subfacet;
4408 struct ofport_dpif *ofport;
4411 if (ofproto_receive(backer, NULL, key, key_len, &flow, NULL, &ofproto,
4416 ofproto->total_subfacet_count += hmap_count(&ofproto->subfacets);
4417 ofproto->n_update_stats++;
4419 ofport = get_ofp_port(ofproto, flow.in_port);
4420 if (ofport && ofport->tnl_port) {
4421 netdev_vport_inc_rx(ofport->up.netdev, stats);
4424 key_hash = odp_flow_key_hash(key, key_len);
4425 subfacet = subfacet_find(ofproto, key, key_len, key_hash);
4426 switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
4428 /* Update ofproto_dpif's hit count. */
4429 if (stats->n_packets > subfacet->dp_packet_count) {
4430 uint64_t delta = stats->n_packets - subfacet->dp_packet_count;
4431 dpif_stats_update_hit_count(ofproto, delta);
4434 update_subfacet_stats(subfacet, stats);
4438 /* Stats are updated per-packet. */
4441 case SF_NOT_INSTALLED:
4443 delete_unexpected_flow(ofproto, key, key_len);
4448 dpif_flow_dump_done(&dump);
4450 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4451 update_moving_averages(ofproto);
4456 /* Calculates and returns the number of milliseconds of idle time after which
4457 * subfacets should expire from the datapath. When a subfacet expires, we fold
4458 * its statistics into its facet, and when a facet's last subfacet expires, we
4459 * fold its statistic into its rule. */
4461 subfacet_max_idle(const struct ofproto_dpif *ofproto)
4464 * Idle time histogram.
4466 * Most of the time a switch has a relatively small number of subfacets.
4467 * When this is the case we might as well keep statistics for all of them
4468 * in userspace and to cache them in the kernel datapath for performance as
4471 * As the number of subfacets increases, the memory required to maintain
4472 * statistics about them in userspace and in the kernel becomes
4473 * significant. However, with a large number of subfacets it is likely
4474 * that only a few of them are "heavy hitters" that consume a large amount
4475 * of bandwidth. At this point, only heavy hitters are worth caching in
4476 * the kernel and maintaining in userspaces; other subfacets we can
4479 * The technique used to compute the idle time is to build a histogram with
4480 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
4481 * that is installed in the kernel gets dropped in the appropriate bucket.
4482 * After the histogram has been built, we compute the cutoff so that only
4483 * the most-recently-used 1% of subfacets (but at least
4484 * ofproto->up.flow_eviction_threshold flows) are kept cached. At least
4485 * the most-recently-used bucket of subfacets is kept, so actually an
4486 * arbitrary number of subfacets can be kept in any given expiration run
4487 * (though the next run will delete most of those unless they receive
4490 * This requires a second pass through the subfacets, in addition to the
4491 * pass made by update_stats(), because the former function never looks at
4492 * uninstallable subfacets.
4494 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
4495 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
4496 int buckets[N_BUCKETS] = { 0 };
4497 int total, subtotal, bucket;
4498 struct subfacet *subfacet;
4502 total = hmap_count(&ofproto->subfacets);
4503 if (total <= ofproto->up.flow_eviction_threshold) {
4504 return N_BUCKETS * BUCKET_WIDTH;
4507 /* Build histogram. */
4509 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
4510 long long int idle = now - subfacet->used;
4511 int bucket = (idle <= 0 ? 0
4512 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
4513 : (unsigned int) idle / BUCKET_WIDTH);
4517 /* Find the first bucket whose flows should be expired. */
4518 subtotal = bucket = 0;
4520 subtotal += buckets[bucket++];
4521 } while (bucket < N_BUCKETS &&
4522 subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100));
4524 if (VLOG_IS_DBG_ENABLED()) {
4528 ds_put_cstr(&s, "keep");
4529 for (i = 0; i < N_BUCKETS; i++) {
4531 ds_put_cstr(&s, ", drop");
4534 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
4537 VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s));
4541 return bucket * BUCKET_WIDTH;
4545 expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle)
4547 /* Cutoff time for most flows. */
4548 long long int normal_cutoff = time_msec() - dp_max_idle;
4550 /* We really want to keep flows for special protocols around, so use a more
4551 * conservative cutoff. */
4552 long long int special_cutoff = time_msec() - 10000;
4554 struct subfacet *subfacet, *next_subfacet;
4555 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
4559 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
4560 &ofproto->subfacets) {
4561 long long int cutoff;
4563 cutoff = (subfacet->facet->xout.slow & (SLOW_CFM | SLOW_LACP
4567 if (subfacet->used < cutoff) {
4568 if (subfacet->path != SF_NOT_INSTALLED) {
4569 batch[n_batch++] = subfacet;
4570 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
4571 subfacet_destroy_batch(ofproto, batch, n_batch);
4575 subfacet_destroy(subfacet);
4581 subfacet_destroy_batch(ofproto, batch, n_batch);
4585 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
4586 * then delete it entirely. */
4588 rule_expire(struct rule_dpif *rule)
4590 struct facet *facet, *next_facet;
4594 if (rule->up.pending) {
4595 /* We'll have to expire it later. */
4599 /* Has 'rule' expired? */
4601 if (rule->up.hard_timeout
4602 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
4603 reason = OFPRR_HARD_TIMEOUT;
4604 } else if (rule->up.idle_timeout
4605 && now > rule->up.used + rule->up.idle_timeout * 1000) {
4606 reason = OFPRR_IDLE_TIMEOUT;
4611 COVERAGE_INC(ofproto_dpif_expired);
4613 /* Update stats. (This is a no-op if the rule expired due to an idle
4614 * timeout, because that only happens when the rule has no facets left.) */
4615 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4616 facet_remove(facet);
4619 /* Get rid of the rule. */
4620 ofproto_rule_expire(&rule->up, reason);
4625 /* Creates and returns a new facet based on 'miss'.
4627 * The caller must already have determined that no facet with an identical
4628 * 'miss->flow' exists in 'miss->ofproto'.
4630 * 'hash' must be the return value of flow_hash(miss->flow, 0).
4632 * The facet will initially have no subfacets. The caller should create (at
4633 * least) one subfacet with subfacet_create(). */
4634 static struct facet *
4635 facet_create(const struct flow_miss *miss, uint32_t hash)
4637 struct ofproto_dpif *ofproto = miss->ofproto;
4638 struct xlate_in xin;
4639 struct facet *facet;
4641 facet = xzalloc(sizeof *facet);
4642 facet->used = time_msec();
4643 facet->flow = miss->flow;
4644 facet->initial_vals = miss->initial_vals;
4645 facet->rule = rule_dpif_lookup(ofproto, &facet->flow);
4646 facet->learn_rl = time_msec() + 500;
4648 hmap_insert(&ofproto->facets, &facet->hmap_node, hash);
4649 list_push_back(&facet->rule->facets, &facet->list_node);
4650 list_init(&facet->subfacets);
4651 netflow_flow_init(&facet->nf_flow);
4652 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
4654 xlate_in_init(&xin, ofproto, &facet->flow, &facet->initial_vals,
4655 facet->rule, 0, NULL);
4656 xin.may_learn = true;
4657 xlate_actions(&xin, &facet->xout);
4658 facet->nf_flow.output_iface = facet->xout.nf_output_iface;
4664 facet_free(struct facet *facet)
4667 xlate_out_uninit(&facet->xout);
4672 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
4673 * 'packet', which arrived on 'in_port'. */
4675 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
4676 const struct nlattr *odp_actions, size_t actions_len,
4677 struct ofpbuf *packet)
4679 struct odputil_keybuf keybuf;
4683 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4684 odp_flow_key_from_flow(&key, flow,
4685 ofp_port_to_odp_port(ofproto, flow->in_port));
4687 error = dpif_execute(ofproto->backer->dpif, key.data, key.size,
4688 odp_actions, actions_len, packet);
4692 /* Remove 'facet' from 'ofproto' and free up the associated memory:
4694 * - If 'facet' was installed in the datapath, uninstalls it and updates its
4695 * rule's statistics, via subfacet_uninstall().
4697 * - Removes 'facet' from its rule and from ofproto->facets.
4700 facet_remove(struct facet *facet)
4702 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4703 struct subfacet *subfacet, *next_subfacet;
4705 ovs_assert(!list_is_empty(&facet->subfacets));
4707 /* First uninstall all of the subfacets to get final statistics. */
4708 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4709 subfacet_uninstall(subfacet);
4712 /* Flush the final stats to the rule.
4714 * This might require us to have at least one subfacet around so that we
4715 * can use its actions for accounting in facet_account(), which is why we
4716 * have uninstalled but not yet destroyed the subfacets. */
4717 facet_flush_stats(facet);
4719 /* Now we're really all done so destroy everything. */
4720 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
4721 &facet->subfacets) {
4722 subfacet_destroy__(subfacet);
4724 hmap_remove(&ofproto->facets, &facet->hmap_node);
4725 list_remove(&facet->list_node);
4729 /* Feed information from 'facet' back into the learning table to keep it in
4730 * sync with what is actually flowing through the datapath. */
4732 facet_learn(struct facet *facet)
4734 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4735 long long int now = time_msec();
4736 struct xlate_in xin;
4738 if (!facet->xout.has_fin_timeout && now < facet->learn_rl) {
4742 facet->learn_rl = now + 500;
4744 if (!facet->xout.has_learn
4745 && !facet->xout.has_normal
4746 && (!facet->xout.has_fin_timeout
4747 || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
4751 xlate_in_init(&xin, ofproto, &facet->flow, &facet->initial_vals,
4752 facet->rule, facet->tcp_flags, NULL);
4753 xin.may_learn = true;
4754 xlate_actions_for_side_effects(&xin);
4758 facet_account(struct facet *facet)
4760 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4761 const struct nlattr *a;
4766 if (!facet->xout.has_normal || !ofproto->has_bonded_bundles) {
4769 n_bytes = facet->byte_count - facet->accounted_bytes;
4771 /* This loop feeds byte counters to bond_account() for rebalancing to use
4772 * as a basis. We also need to track the actual VLAN on which the packet
4773 * is going to be sent to ensure that it matches the one passed to
4774 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
4777 * We use the actions from an arbitrary subfacet because they should all
4778 * be equally valid for our purpose. */
4779 vlan_tci = facet->flow.vlan_tci;
4780 NL_ATTR_FOR_EACH_UNSAFE (a, left, facet->xout.odp_actions.data,
4781 facet->xout.odp_actions.size) {
4782 const struct ovs_action_push_vlan *vlan;
4783 struct ofport_dpif *port;
4785 switch (nl_attr_type(a)) {
4786 case OVS_ACTION_ATTR_OUTPUT:
4787 port = get_odp_port(ofproto, nl_attr_get_u32(a));
4788 if (port && port->bundle && port->bundle->bond) {
4789 bond_account(port->bundle->bond, &facet->flow,
4790 vlan_tci_to_vid(vlan_tci), n_bytes);
4794 case OVS_ACTION_ATTR_POP_VLAN:
4795 vlan_tci = htons(0);
4798 case OVS_ACTION_ATTR_PUSH_VLAN:
4799 vlan = nl_attr_get(a);
4800 vlan_tci = vlan->vlan_tci;
4806 /* Returns true if the only action for 'facet' is to send to the controller.
4807 * (We don't report NetFlow expiration messages for such facets because they
4808 * are just part of the control logic for the network, not real traffic). */
4810 facet_is_controller_flow(struct facet *facet)
4813 const struct rule *rule = &facet->rule->up;
4814 const struct ofpact *ofpacts = rule->ofpacts;
4815 size_t ofpacts_len = rule->ofpacts_len;
4817 if (ofpacts_len > 0 &&
4818 ofpacts->type == OFPACT_CONTROLLER &&
4819 ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len)) {
4826 /* Folds all of 'facet''s statistics into its rule. Also updates the
4827 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
4828 * 'facet''s statistics in the datapath should have been zeroed and folded into
4829 * its packet and byte counts before this function is called. */
4831 facet_flush_stats(struct facet *facet)
4833 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4834 struct subfacet *subfacet;
4836 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4837 ovs_assert(!subfacet->dp_byte_count);
4838 ovs_assert(!subfacet->dp_packet_count);
4841 facet_push_stats(facet);
4842 if (facet->accounted_bytes < facet->byte_count) {
4843 facet_account(facet);
4844 facet->accounted_bytes = facet->byte_count;
4847 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
4848 struct ofexpired expired;
4849 expired.flow = facet->flow;
4850 expired.packet_count = facet->packet_count;
4851 expired.byte_count = facet->byte_count;
4852 expired.used = facet->used;
4853 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4856 facet->rule->packet_count += facet->packet_count;
4857 facet->rule->byte_count += facet->byte_count;
4859 /* Reset counters to prevent double counting if 'facet' ever gets
4861 facet_reset_counters(facet);
4863 netflow_flow_clear(&facet->nf_flow);
4864 facet->tcp_flags = 0;
4867 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
4868 * Returns it if found, otherwise a null pointer.
4870 * 'hash' must be the return value of flow_hash(flow, 0).
4872 * The returned facet might need revalidation; use facet_lookup_valid()
4873 * instead if that is important. */
4874 static struct facet *
4875 facet_find(struct ofproto_dpif *ofproto,
4876 const struct flow *flow, uint32_t hash)
4878 struct facet *facet;
4880 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, hash, &ofproto->facets) {
4881 if (flow_equal(flow, &facet->flow)) {
4889 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
4890 * Returns it if found, otherwise a null pointer.
4892 * 'hash' must be the return value of flow_hash(flow, 0).
4894 * The returned facet is guaranteed to be valid. */
4895 static struct facet *
4896 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow,
4899 struct facet *facet;
4901 facet = facet_find(ofproto, flow, hash);
4903 && (ofproto->backer->need_revalidate
4904 || tag_set_intersects(&ofproto->backer->revalidate_set,
4906 && !facet_revalidate(facet)) {
4914 facet_check_consistency(struct facet *facet)
4916 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
4918 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4920 struct xlate_out xout;
4921 struct xlate_in xin;
4923 struct rule_dpif *rule;
4926 /* Check the rule for consistency. */
4927 rule = rule_dpif_lookup(ofproto, &facet->flow);
4928 if (rule != facet->rule) {
4929 if (!VLOG_DROP_WARN(&rl)) {
4930 struct ds s = DS_EMPTY_INITIALIZER;
4932 flow_format(&s, &facet->flow);
4933 ds_put_format(&s, ": facet associated with wrong rule (was "
4934 "table=%"PRIu8",", facet->rule->up.table_id);
4935 cls_rule_format(&facet->rule->up.cr, &s);
4936 ds_put_format(&s, ") (should have been table=%"PRIu8",",
4938 cls_rule_format(&rule->up.cr, &s);
4939 ds_put_cstr(&s, ")\n");
4946 /* Check the datapath actions for consistency. */
4947 xlate_in_init(&xin, ofproto, &facet->flow, &facet->initial_vals, rule,
4949 xlate_actions(&xin, &xout);
4951 ok = ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)
4952 && facet->xout.slow == xout.slow;
4953 if (!ok && !VLOG_DROP_WARN(&rl)) {
4954 struct ds s = DS_EMPTY_INITIALIZER;
4956 flow_format(&s, &facet->flow);
4957 ds_put_cstr(&s, ": inconsistency in facet");
4959 if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4960 ds_put_cstr(&s, " (actions were: ");
4961 format_odp_actions(&s, facet->xout.odp_actions.data,
4962 facet->xout.odp_actions.size);
4963 ds_put_cstr(&s, ") (correct actions: ");
4964 format_odp_actions(&s, xout.odp_actions.data,
4965 xout.odp_actions.size);
4966 ds_put_cstr(&s, ")");
4969 if (facet->xout.slow != xout.slow) {
4970 ds_put_format(&s, " slow path incorrect. should be %d", xout.slow);
4975 xlate_out_uninit(&xout);
4980 /* Re-searches the classifier for 'facet':
4982 * - If the rule found is different from 'facet''s current rule, moves
4983 * 'facet' to the new rule and recompiles its actions.
4985 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
4986 * where it is and recompiles its actions anyway.
4988 * - If any of 'facet''s subfacets correspond to a new flow according to
4989 * ofproto_receive(), 'facet' is removed.
4991 * Returns true if 'facet' is still valid. False if 'facet' was removed. */
4993 facet_revalidate(struct facet *facet)
4995 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4996 struct rule_dpif *new_rule;
4997 struct subfacet *subfacet;
4998 struct xlate_out xout;
4999 struct xlate_in xin;
5001 COVERAGE_INC(facet_revalidate);
5003 /* Check that child subfacets still correspond to this facet. Tunnel
5004 * configuration changes could cause a subfacet's OpenFlow in_port to
5006 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
5007 struct ofproto_dpif *recv_ofproto;
5008 struct flow recv_flow;
5011 error = ofproto_receive(ofproto->backer, NULL, subfacet->key,
5012 subfacet->key_len, &recv_flow, NULL,
5013 &recv_ofproto, NULL, NULL);
5015 || recv_ofproto != ofproto
5016 || memcmp(&recv_flow, &facet->flow, sizeof recv_flow)) {
5017 facet_remove(facet);
5022 new_rule = rule_dpif_lookup(ofproto, &facet->flow);
5024 /* Calculate new datapath actions.
5026 * We do not modify any 'facet' state yet, because we might need to, e.g.,
5027 * emit a NetFlow expiration and, if so, we need to have the old state
5028 * around to properly compose it. */
5029 xlate_in_init(&xin, ofproto, &facet->flow, &facet->initial_vals, new_rule,
5031 xlate_actions(&xin, &xout);
5033 /* A facet's slow path reason should only change under dramatic
5034 * circumstances. Rather than try to update everything, it's simpler to
5035 * remove the facet and start over. */
5036 if (facet->xout.slow != xout.slow) {
5037 facet_remove(facet);
5038 xlate_out_uninit(&xout);
5042 if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
5043 LIST_FOR_EACH(subfacet, list_node, &facet->subfacets) {
5044 if (subfacet->path == SF_FAST_PATH) {
5045 struct dpif_flow_stats stats;
5047 subfacet_install(subfacet, &xout.odp_actions, &stats);
5048 subfacet_update_stats(subfacet, &stats);
5052 facet_flush_stats(facet);
5054 ofpbuf_clear(&facet->xout.odp_actions);
5055 ofpbuf_put(&facet->xout.odp_actions, xout.odp_actions.data,
5056 xout.odp_actions.size);
5059 /* Update 'facet' now that we've taken care of all the old state. */
5060 facet->xout.tags = xout.tags;
5061 facet->xout.slow = xout.slow;
5062 facet->xout.has_learn = xout.has_learn;
5063 facet->xout.has_normal = xout.has_normal;
5064 facet->xout.has_fin_timeout = xout.has_fin_timeout;
5065 facet->xout.nf_output_iface = xout.nf_output_iface;
5066 facet->xout.mirrors = xout.mirrors;
5067 facet->nf_flow.output_iface = facet->xout.nf_output_iface;
5069 if (facet->rule != new_rule) {
5070 COVERAGE_INC(facet_changed_rule);
5071 list_remove(&facet->list_node);
5072 list_push_back(&new_rule->facets, &facet->list_node);
5073 facet->rule = new_rule;
5074 facet->used = new_rule->up.created;
5075 facet->prev_used = facet->used;
5078 xlate_out_uninit(&xout);
5082 /* Updates 'facet''s used time. Caller is responsible for calling
5083 * facet_push_stats() to update the flows which 'facet' resubmits into. */
5085 facet_update_time(struct facet *facet, long long int used)
5087 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5088 if (used > facet->used) {
5090 ofproto_rule_update_used(&facet->rule->up, used);
5091 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
5096 facet_reset_counters(struct facet *facet)
5098 facet->packet_count = 0;
5099 facet->byte_count = 0;
5100 facet->prev_packet_count = 0;
5101 facet->prev_byte_count = 0;
5102 facet->accounted_bytes = 0;
5106 facet_push_stats(struct facet *facet)
5108 struct dpif_flow_stats stats;
5110 ovs_assert(facet->packet_count >= facet->prev_packet_count);
5111 ovs_assert(facet->byte_count >= facet->prev_byte_count);
5112 ovs_assert(facet->used >= facet->prev_used);
5114 stats.n_packets = facet->packet_count - facet->prev_packet_count;
5115 stats.n_bytes = facet->byte_count - facet->prev_byte_count;
5116 stats.used = facet->used;
5117 stats.tcp_flags = 0;
5119 if (stats.n_packets || stats.n_bytes || facet->used > facet->prev_used) {
5120 facet->prev_packet_count = facet->packet_count;
5121 facet->prev_byte_count = facet->byte_count;
5122 facet->prev_used = facet->used;
5124 flow_push_stats(facet, &stats);
5126 update_mirror_stats(ofproto_dpif_cast(facet->rule->up.ofproto),
5127 facet->xout.mirrors, stats.n_packets,
5133 push_all_stats__(bool run_fast)
5135 static long long int rl = LLONG_MIN;
5136 struct ofproto_dpif *ofproto;
5138 if (time_msec() < rl) {
5142 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5143 struct facet *facet;
5145 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
5146 facet_push_stats(facet);
5153 rl = time_msec() + 100;
5157 push_all_stats(void)
5159 push_all_stats__(true);
5163 rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats)
5165 rule->packet_count += stats->n_packets;
5166 rule->byte_count += stats->n_bytes;
5167 ofproto_rule_update_used(&rule->up, stats->used);
5170 /* Pushes flow statistics to the rules which 'facet->flow' resubmits
5171 * into given 'facet->rule''s actions and mirrors. */
5173 flow_push_stats(struct facet *facet, const struct dpif_flow_stats *stats)
5175 struct rule_dpif *rule = facet->rule;
5176 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5177 struct xlate_in xin;
5179 ofproto_rule_update_used(&rule->up, stats->used);
5181 xlate_in_init(&xin, ofproto, &facet->flow, &facet->initial_vals, rule,
5183 xin.resubmit_stats = stats;
5184 xlate_actions_for_side_effects(&xin);
5189 static struct subfacet *
5190 subfacet_find(struct ofproto_dpif *ofproto,
5191 const struct nlattr *key, size_t key_len, uint32_t key_hash)
5193 struct subfacet *subfacet;
5195 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
5196 &ofproto->subfacets) {
5197 if (subfacet->key_len == key_len
5198 && !memcmp(key, subfacet->key, key_len)) {
5206 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
5207 * 'key_fitness', 'key', and 'key_len' members in 'miss'. Returns the
5208 * existing subfacet if there is one, otherwise creates and returns a
5210 static struct subfacet *
5211 subfacet_create(struct facet *facet, struct flow_miss *miss,
5214 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5215 enum odp_key_fitness key_fitness = miss->key_fitness;
5216 const struct nlattr *key = miss->key;
5217 size_t key_len = miss->key_len;
5219 struct subfacet *subfacet;
5221 key_hash = odp_flow_key_hash(key, key_len);
5223 if (list_is_empty(&facet->subfacets)) {
5224 subfacet = &facet->one_subfacet;
5226 subfacet = subfacet_find(ofproto, key, key_len, key_hash);
5228 if (subfacet->facet == facet) {
5232 /* This shouldn't happen. */
5233 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
5234 subfacet_destroy(subfacet);
5237 subfacet = xmalloc(sizeof *subfacet);
5240 hmap_insert(&ofproto->subfacets, &subfacet->hmap_node, key_hash);
5241 list_push_back(&facet->subfacets, &subfacet->list_node);
5242 subfacet->facet = facet;
5243 subfacet->key_fitness = key_fitness;
5244 subfacet->key = xmemdup(key, key_len);
5245 subfacet->key_len = key_len;
5246 subfacet->used = now;
5247 subfacet->created = now;
5248 subfacet->dp_packet_count = 0;
5249 subfacet->dp_byte_count = 0;
5250 subfacet->path = SF_NOT_INSTALLED;
5251 subfacet->odp_in_port = miss->odp_in_port;
5253 ofproto->subfacet_add_count++;
5257 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
5258 * its facet within 'ofproto', and frees it. */
5260 subfacet_destroy__(struct subfacet *subfacet)
5262 struct facet *facet = subfacet->facet;
5263 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5265 /* Update ofproto stats before uninstall the subfacet. */
5266 ofproto->subfacet_del_count++;
5267 ofproto->total_subfacet_life_span += (time_msec() - subfacet->created);
5269 subfacet_uninstall(subfacet);
5270 hmap_remove(&ofproto->subfacets, &subfacet->hmap_node);
5271 list_remove(&subfacet->list_node);
5272 free(subfacet->key);
5273 if (subfacet != &facet->one_subfacet) {
5278 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
5279 * last remaining subfacet in its facet destroys the facet too. */
5281 subfacet_destroy(struct subfacet *subfacet)
5283 struct facet *facet = subfacet->facet;
5285 if (list_is_singleton(&facet->subfacets)) {
5286 /* facet_remove() needs at least one subfacet (it will remove it). */
5287 facet_remove(facet);
5289 subfacet_destroy__(subfacet);
5294 subfacet_destroy_batch(struct ofproto_dpif *ofproto,
5295 struct subfacet **subfacets, int n)
5297 struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH];
5298 struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH];
5299 struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH];
5302 for (i = 0; i < n; i++) {
5303 ops[i].type = DPIF_OP_FLOW_DEL;
5304 ops[i].u.flow_del.key = subfacets[i]->key;
5305 ops[i].u.flow_del.key_len = subfacets[i]->key_len;
5306 ops[i].u.flow_del.stats = &stats[i];
5310 dpif_operate(ofproto->backer->dpif, opsp, n);
5311 for (i = 0; i < n; i++) {
5312 subfacet_reset_dp_stats(subfacets[i], &stats[i]);
5313 subfacets[i]->path = SF_NOT_INSTALLED;
5314 subfacet_destroy(subfacets[i]);
5319 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
5320 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
5321 * in the datapath will be zeroed and 'stats' will be updated with traffic new
5322 * since 'subfacet' was last updated.
5324 * Returns 0 if successful, otherwise a positive errno value. */
5326 subfacet_install(struct subfacet *subfacet, const struct ofpbuf *odp_actions,
5327 struct dpif_flow_stats *stats)
5329 struct facet *facet = subfacet->facet;
5330 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5331 enum subfacet_path path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
5332 const struct nlattr *actions = odp_actions->data;
5333 size_t actions_len = odp_actions->size;
5335 uint64_t slow_path_stub[128 / 8];
5336 enum dpif_flow_put_flags flags;
5339 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
5341 flags |= DPIF_FP_ZERO_STATS;
5344 if (path == SF_SLOW_PATH) {
5345 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
5346 slow_path_stub, sizeof slow_path_stub,
5347 &actions, &actions_len);
5350 ret = dpif_flow_put(ofproto->backer->dpif, flags, subfacet->key,
5351 subfacet->key_len, actions, actions_len, stats);
5354 subfacet_reset_dp_stats(subfacet, stats);
5358 subfacet->path = path;
5363 /* If 'subfacet' is installed in the datapath, uninstalls it. */
5365 subfacet_uninstall(struct subfacet *subfacet)
5367 if (subfacet->path != SF_NOT_INSTALLED) {
5368 struct rule_dpif *rule = subfacet->facet->rule;
5369 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5370 struct dpif_flow_stats stats;
5373 error = dpif_flow_del(ofproto->backer->dpif, subfacet->key,
5374 subfacet->key_len, &stats);
5375 subfacet_reset_dp_stats(subfacet, &stats);
5377 subfacet_update_stats(subfacet, &stats);
5379 subfacet->path = SF_NOT_INSTALLED;
5381 ovs_assert(subfacet->dp_packet_count == 0);
5382 ovs_assert(subfacet->dp_byte_count == 0);
5386 /* Resets 'subfacet''s datapath statistics counters. This should be called
5387 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
5388 * non-null, it should contain the statistics returned by dpif when 'subfacet'
5389 * was reset in the datapath. 'stats' will be modified to include only
5390 * statistics new since 'subfacet' was last updated. */
5392 subfacet_reset_dp_stats(struct subfacet *subfacet,
5393 struct dpif_flow_stats *stats)
5396 && subfacet->dp_packet_count <= stats->n_packets
5397 && subfacet->dp_byte_count <= stats->n_bytes) {
5398 stats->n_packets -= subfacet->dp_packet_count;
5399 stats->n_bytes -= subfacet->dp_byte_count;
5402 subfacet->dp_packet_count = 0;
5403 subfacet->dp_byte_count = 0;
5406 /* Updates 'subfacet''s used time. The caller is responsible for calling
5407 * facet_push_stats() to update the flows which 'subfacet' resubmits into. */
5409 subfacet_update_time(struct subfacet *subfacet, long long int used)
5411 if (used > subfacet->used) {
5412 subfacet->used = used;
5413 facet_update_time(subfacet->facet, used);
5417 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
5419 * Because of the meaning of a subfacet's counters, it only makes sense to do
5420 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
5421 * represents a packet that was sent by hand or if it represents statistics
5422 * that have been cleared out of the datapath. */
5424 subfacet_update_stats(struct subfacet *subfacet,
5425 const struct dpif_flow_stats *stats)
5427 if (stats->n_packets || stats->used > subfacet->used) {
5428 struct facet *facet = subfacet->facet;
5430 subfacet_update_time(subfacet, stats->used);
5431 facet->packet_count += stats->n_packets;
5432 facet->byte_count += stats->n_bytes;
5433 facet->tcp_flags |= stats->tcp_flags;
5434 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
5440 static struct rule_dpif *
5441 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow)
5443 struct rule_dpif *rule;
5445 rule = rule_dpif_lookup__(ofproto, flow, 0);
5450 return rule_dpif_miss_rule(ofproto, flow);
5453 static struct rule_dpif *
5454 rule_dpif_lookup__(struct ofproto_dpif *ofproto, const struct flow *flow,
5457 struct cls_rule *cls_rule;
5458 struct classifier *cls;
5460 if (table_id >= N_TABLES) {
5464 cls = &ofproto->up.tables[table_id].cls;
5465 if (flow->nw_frag & FLOW_NW_FRAG_ANY
5466 && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
5467 /* For OFPC_NORMAL frag_handling, we must pretend that transport ports
5468 * are unavailable. */
5469 struct flow ofpc_normal_flow = *flow;
5470 ofpc_normal_flow.tp_src = htons(0);
5471 ofpc_normal_flow.tp_dst = htons(0);
5472 cls_rule = classifier_lookup(cls, &ofpc_normal_flow);
5474 cls_rule = classifier_lookup(cls, flow);
5476 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
5479 static struct rule_dpif *
5480 rule_dpif_miss_rule(struct ofproto_dpif *ofproto, const struct flow *flow)
5482 struct ofport_dpif *port;
5484 port = get_ofp_port(ofproto, flow->in_port);
5486 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, flow->in_port);
5487 return ofproto->miss_rule;
5490 if (port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN) {
5491 return ofproto->no_packet_in_rule;
5493 return ofproto->miss_rule;
5497 complete_operation(struct rule_dpif *rule)
5499 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5501 rule_invalidate(rule);
5503 struct dpif_completion *c = xmalloc(sizeof *c);
5504 c->op = rule->up.pending;
5505 list_push_back(&ofproto->completions, &c->list_node);
5507 ofoperation_complete(rule->up.pending, 0);
5511 static struct rule *
5514 struct rule_dpif *rule = xmalloc(sizeof *rule);
5519 rule_dealloc(struct rule *rule_)
5521 struct rule_dpif *rule = rule_dpif_cast(rule_);
5526 rule_construct(struct rule *rule_)
5528 struct rule_dpif *rule = rule_dpif_cast(rule_);
5529 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5530 struct rule_dpif *victim;
5533 rule->packet_count = 0;
5534 rule->byte_count = 0;
5536 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
5537 if (victim && !list_is_empty(&victim->facets)) {
5538 struct facet *facet;
5540 rule->facets = victim->facets;
5541 list_moved(&rule->facets);
5542 LIST_FOR_EACH (facet, list_node, &rule->facets) {
5543 /* XXX: We're only clearing our local counters here. It's possible
5544 * that quite a few packets are unaccounted for in the datapath
5545 * statistics. These will be accounted to the new rule instead of
5546 * cleared as required. This could be fixed by clearing out the
5547 * datapath statistics for this facet, but currently it doesn't
5549 facet_reset_counters(facet);
5553 /* Must avoid list_moved() in this case. */
5554 list_init(&rule->facets);
5557 table_id = rule->up.table_id;
5559 rule->tag = victim->tag;
5560 } else if (table_id == 0) {
5565 miniflow_expand(&rule->up.cr.match.flow, &flow);
5566 rule->tag = rule_calculate_tag(&flow, &rule->up.cr.match.mask,
5567 ofproto->tables[table_id].basis);
5570 complete_operation(rule);
5575 rule_destruct(struct rule *rule_)
5577 struct rule_dpif *rule = rule_dpif_cast(rule_);
5578 struct facet *facet, *next_facet;
5580 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
5581 facet_revalidate(facet);
5584 complete_operation(rule);
5588 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
5590 struct rule_dpif *rule = rule_dpif_cast(rule_);
5591 struct facet *facet;
5593 /* push_all_stats() can handle flow misses which, when using the learn
5594 * action, can cause rules to be added and deleted. This can corrupt our
5595 * caller's datastructures which assume that rule_get_stats() doesn't have
5596 * an impact on the flow table. To be safe, we disable miss handling. */
5597 push_all_stats__(false);
5599 /* Start from historical data for 'rule' itself that are no longer tracked
5600 * in facets. This counts, for example, facets that have expired. */
5601 *packets = rule->packet_count;
5602 *bytes = rule->byte_count;
5604 /* Add any statistics that are tracked by facets. This includes
5605 * statistical data recently updated by ofproto_update_stats() as well as
5606 * stats for packets that were executed "by hand" via dpif_execute(). */
5607 LIST_FOR_EACH (facet, list_node, &rule->facets) {
5608 *packets += facet->packet_count;
5609 *bytes += facet->byte_count;
5614 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
5615 struct ofpbuf *packet)
5617 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5618 struct initial_vals initial_vals;
5619 struct dpif_flow_stats stats;
5620 struct xlate_out xout;
5621 struct xlate_in xin;
5623 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
5624 rule_credit_stats(rule, &stats);
5626 initial_vals.vlan_tci = flow->vlan_tci;
5627 initial_vals.tunnel_ip_tos = flow->tunnel.ip_tos;
5628 xlate_in_init(&xin, ofproto, flow, &initial_vals, rule, stats.tcp_flags,
5630 xin.resubmit_stats = &stats;
5631 xlate_actions(&xin, &xout);
5633 execute_odp_actions(ofproto, flow, xout.odp_actions.data,
5634 xout.odp_actions.size, packet);
5636 xlate_out_uninit(&xout);
5640 rule_execute(struct rule *rule, const struct flow *flow,
5641 struct ofpbuf *packet)
5643 rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
5644 ofpbuf_delete(packet);
5649 rule_modify_actions(struct rule *rule_)
5651 struct rule_dpif *rule = rule_dpif_cast(rule_);
5653 complete_operation(rule);
5656 /* Sends 'packet' out 'ofport'.
5657 * May modify 'packet'.
5658 * Returns 0 if successful, otherwise a positive errno value. */
5660 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
5662 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
5663 uint64_t odp_actions_stub[1024 / 8];
5664 struct ofpbuf key, odp_actions;
5665 struct odputil_keybuf keybuf;
5670 flow_extract(packet, 0, 0, NULL, OFPP_LOCAL, &flow);
5671 if (netdev_vport_is_patch(ofport->up.netdev)) {
5672 struct ofproto_dpif *peer_ofproto;
5673 struct dpif_flow_stats stats;
5674 struct ofport_dpif *peer;
5675 struct rule_dpif *rule;
5677 peer = ofport_get_peer(ofport);
5682 dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
5683 netdev_vport_inc_tx(ofport->up.netdev, &stats);
5684 netdev_vport_inc_rx(peer->up.netdev, &stats);
5686 flow.in_port = peer->up.ofp_port;
5687 peer_ofproto = ofproto_dpif_cast(peer->up.ofproto);
5688 rule = rule_dpif_lookup(peer_ofproto, &flow);
5689 rule_dpif_execute(rule, &flow, packet);
5694 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5696 if (ofport->tnl_port) {
5697 struct dpif_flow_stats stats;
5699 odp_port = tnl_port_send(ofport->tnl_port, &flow);
5700 if (odp_port == OVSP_NONE) {
5704 dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
5705 netdev_vport_inc_tx(ofport->up.netdev, &stats);
5706 odp_put_tunnel_action(&flow.tunnel, &odp_actions);
5707 odp_put_skb_mark_action(flow.skb_mark, &odp_actions);
5709 odp_port = vsp_realdev_to_vlandev(ofproto, ofport->odp_port,
5711 if (odp_port != ofport->odp_port) {
5712 eth_pop_vlan(packet);
5713 flow.vlan_tci = htons(0);
5717 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5718 odp_flow_key_from_flow(&key, &flow,
5719 ofp_port_to_odp_port(ofproto, flow.in_port));
5721 compose_sflow_action(ofproto, &odp_actions, &flow, odp_port);
5722 compose_ipfix_action(ofproto, &odp_actions, &flow);
5724 nl_msg_put_u32(&odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port);
5725 error = dpif_execute(ofproto->backer->dpif,
5727 odp_actions.data, odp_actions.size,
5729 ofpbuf_uninit(&odp_actions);
5732 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)",
5733 ofproto->up.name, odp_port, strerror(error));
5735 ofproto_update_local_port_stats(ofport->up.ofproto, packet->size, 0);
5739 /* OpenFlow to datapath action translation. */
5741 static bool may_receive(const struct ofport_dpif *, struct xlate_ctx *);
5742 static void do_xlate_actions(const struct ofpact *, size_t ofpacts_len,
5743 struct xlate_ctx *);
5744 static void xlate_normal(struct xlate_ctx *);
5746 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
5747 * The action will state 'slow' as the reason that the action is in the slow
5748 * path. (This is purely informational: it allows a human viewing "ovs-dpctl
5749 * dump-flows" output to see why a flow is in the slow path.)
5751 * The 'stub_size' bytes in 'stub' will be used to store the action.
5752 * 'stub_size' must be large enough for the action.
5754 * The action and its size will be stored in '*actionsp' and '*actions_lenp',
5757 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
5758 enum slow_path_reason slow,
5759 uint64_t *stub, size_t stub_size,
5760 const struct nlattr **actionsp, size_t *actions_lenp)
5762 union user_action_cookie cookie;
5765 cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
5766 cookie.slow_path.unused = 0;
5767 cookie.slow_path.reason = slow;
5769 ofpbuf_use_stack(&buf, stub, stub_size);
5770 if (slow & (SLOW_CFM | SLOW_LACP | SLOW_STP)) {
5771 uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif, UINT32_MAX);
5772 odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
5774 put_userspace_action(ofproto, &buf, flow, &cookie,
5775 sizeof cookie.slow_path);
5777 *actionsp = buf.data;
5778 *actions_lenp = buf.size;
5782 put_userspace_action(const struct ofproto_dpif *ofproto,
5783 struct ofpbuf *odp_actions,
5784 const struct flow *flow,
5785 const union user_action_cookie *cookie,
5786 const size_t cookie_size)
5790 pid = dpif_port_get_pid(ofproto->backer->dpif,
5791 ofp_port_to_odp_port(ofproto, flow->in_port));
5793 return odp_put_userspace_action(pid, cookie, cookie_size, odp_actions);
5796 /* Compose SAMPLE action for sFlow or IPFIX. The given probability is
5797 * the number of packets out of UINT32_MAX to sample. The given
5798 * cookie is passed back in the callback for each sampled packet.
5801 compose_sample_action(const struct ofproto_dpif *ofproto,
5802 struct ofpbuf *odp_actions,
5803 const struct flow *flow,
5804 const uint32_t probability,
5805 const union user_action_cookie *cookie,
5806 const size_t cookie_size)
5808 size_t sample_offset, actions_offset;
5811 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
5813 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
5815 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
5816 cookie_offset = put_userspace_action(ofproto, odp_actions, flow, cookie,
5819 nl_msg_end_nested(odp_actions, actions_offset);
5820 nl_msg_end_nested(odp_actions, sample_offset);
5821 return cookie_offset;
5825 compose_sflow_cookie(const struct ofproto_dpif *ofproto,
5826 ovs_be16 vlan_tci, uint32_t odp_port,
5827 unsigned int n_outputs, union user_action_cookie *cookie)
5831 cookie->type = USER_ACTION_COOKIE_SFLOW;
5832 cookie->sflow.vlan_tci = vlan_tci;
5834 /* See http://www.sflow.org/sflow_version_5.txt (search for "Input/output
5835 * port information") for the interpretation of cookie->output. */
5836 switch (n_outputs) {
5838 /* 0x40000000 | 256 means "packet dropped for unknown reason". */
5839 cookie->sflow.output = 0x40000000 | 256;
5843 ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port);
5845 cookie->sflow.output = ifindex;
5850 /* 0x80000000 means "multiple output ports. */
5851 cookie->sflow.output = 0x80000000 | n_outputs;
5856 /* Compose SAMPLE action for sFlow bridge sampling. */
5858 compose_sflow_action(const struct ofproto_dpif *ofproto,
5859 struct ofpbuf *odp_actions,
5860 const struct flow *flow,
5863 uint32_t probability;
5864 union user_action_cookie cookie;
5866 if (!ofproto->sflow || flow->in_port == OFPP_NONE) {
5870 probability = dpif_sflow_get_probability(ofproto->sflow);
5871 compose_sflow_cookie(ofproto, htons(0), odp_port,
5872 odp_port == OVSP_NONE ? 0 : 1, &cookie);
5874 return compose_sample_action(ofproto, odp_actions, flow, probability,
5875 &cookie, sizeof cookie.sflow);
5879 compose_flow_sample_cookie(uint16_t probability, uint32_t collector_set_id,
5880 uint32_t obs_domain_id, uint32_t obs_point_id,
5881 union user_action_cookie *cookie)
5883 cookie->type = USER_ACTION_COOKIE_FLOW_SAMPLE;
5884 cookie->flow_sample.probability = probability;
5885 cookie->flow_sample.collector_set_id = collector_set_id;
5886 cookie->flow_sample.obs_domain_id = obs_domain_id;
5887 cookie->flow_sample.obs_point_id = obs_point_id;
5891 compose_ipfix_cookie(union user_action_cookie *cookie)
5893 cookie->type = USER_ACTION_COOKIE_IPFIX;
5896 /* Compose SAMPLE action for IPFIX bridge sampling. */
5898 compose_ipfix_action(const struct ofproto_dpif *ofproto,
5899 struct ofpbuf *odp_actions,
5900 const struct flow *flow)
5902 uint32_t probability;
5903 union user_action_cookie cookie;
5905 if (!ofproto->ipfix || flow->in_port == OFPP_NONE) {
5909 probability = dpif_ipfix_get_bridge_exporter_probability(ofproto->ipfix);
5910 compose_ipfix_cookie(&cookie);
5912 compose_sample_action(ofproto, odp_actions, flow, probability,
5913 &cookie, sizeof cookie.ipfix);
5916 /* SAMPLE action for sFlow must be first action in any given list of
5917 * actions. At this point we do not have all information required to
5918 * build it. So try to build sample action as complete as possible. */
5920 add_sflow_action(struct xlate_ctx *ctx)
5922 ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto,
5923 &ctx->xout->odp_actions,
5924 &ctx->xin->flow, OVSP_NONE);
5925 ctx->sflow_odp_port = 0;
5926 ctx->sflow_n_outputs = 0;
5929 /* SAMPLE action for IPFIX must be 1st or 2nd action in any given list
5930 * of actions, eventually after the SAMPLE action for sFlow. */
5932 add_ipfix_action(struct xlate_ctx *ctx)
5934 compose_ipfix_action(ctx->ofproto, &ctx->xout->odp_actions,
5938 /* Fix SAMPLE action according to data collected while composing ODP actions.
5939 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
5940 * USERSPACE action's user-cookie which is required for sflow. */
5942 fix_sflow_action(struct xlate_ctx *ctx)
5944 const struct flow *base = &ctx->base_flow;
5945 union user_action_cookie *cookie;
5947 if (!ctx->user_cookie_offset) {
5951 cookie = ofpbuf_at(&ctx->xout->odp_actions, ctx->user_cookie_offset,
5952 sizeof cookie->sflow);
5953 ovs_assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
5955 compose_sflow_cookie(ctx->ofproto, base->vlan_tci,
5956 ctx->sflow_odp_port, ctx->sflow_n_outputs, cookie);
5960 compose_output_action__(struct xlate_ctx *ctx, uint16_t ofp_port,
5963 const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
5964 ovs_be16 flow_vlan_tci;
5965 uint32_t flow_skb_mark;
5966 uint8_t flow_nw_tos;
5967 struct priority_to_dscp *pdscp;
5968 uint32_t out_port, odp_port;
5970 /* If 'struct flow' gets additional metadata, we'll need to zero it out
5971 * before traversing a patch port. */
5972 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 20);
5975 xlate_report(ctx, "Nonexistent output port");
5977 } else if (ofport->up.pp.config & OFPUTIL_PC_NO_FWD) {
5978 xlate_report(ctx, "OFPPC_NO_FWD set, skipping output");
5980 } else if (check_stp && !stp_forward_in_state(ofport->stp_state)) {
5981 xlate_report(ctx, "STP not in forwarding state, skipping output");
5985 if (netdev_vport_is_patch(ofport->up.netdev)) {
5986 struct ofport_dpif *peer = ofport_get_peer(ofport);
5987 struct flow old_flow = ctx->xin->flow;
5988 const struct ofproto_dpif *peer_ofproto;
5989 enum slow_path_reason special;
5990 struct ofport_dpif *in_port;
5993 xlate_report(ctx, "Nonexistent patch port peer");
5997 peer_ofproto = ofproto_dpif_cast(peer->up.ofproto);
5998 if (peer_ofproto->backer != ctx->ofproto->backer) {
5999 xlate_report(ctx, "Patch port peer on a different datapath");
6003 ctx->ofproto = ofproto_dpif_cast(peer->up.ofproto);
6004 ctx->xin->flow.in_port = peer->up.ofp_port;
6005 ctx->xin->flow.metadata = htonll(0);
6006 memset(&ctx->xin->flow.tunnel, 0, sizeof ctx->xin->flow.tunnel);
6007 memset(ctx->xin->flow.regs, 0, sizeof ctx->xin->flow.regs);
6009 in_port = get_ofp_port(ctx->ofproto, ctx->xin->flow.in_port);
6010 special = process_special(ctx->ofproto, &ctx->xin->flow, in_port,
6013 ctx->xout->slow = special;
6014 } else if (!in_port || may_receive(in_port, ctx)) {
6015 if (!in_port || stp_forward_in_state(in_port->stp_state)) {
6016 xlate_table_action(ctx, ctx->xin->flow.in_port, 0, true);
6018 /* Forwarding is disabled by STP. Let OFPP_NORMAL and the
6019 * learning action look at the packet, then drop it. */
6020 struct flow old_base_flow = ctx->base_flow;
6021 size_t old_size = ctx->xout->odp_actions.size;
6022 xlate_table_action(ctx, ctx->xin->flow.in_port, 0, true);
6023 ctx->base_flow = old_base_flow;
6024 ctx->xout->odp_actions.size = old_size;
6028 ctx->xin->flow = old_flow;
6029 ctx->ofproto = ofproto_dpif_cast(ofport->up.ofproto);
6031 if (ctx->xin->resubmit_stats) {
6032 netdev_vport_inc_tx(ofport->up.netdev, ctx->xin->resubmit_stats);
6033 netdev_vport_inc_rx(peer->up.netdev, ctx->xin->resubmit_stats);
6039 flow_vlan_tci = ctx->xin->flow.vlan_tci;
6040 flow_skb_mark = ctx->xin->flow.skb_mark;
6041 flow_nw_tos = ctx->xin->flow.nw_tos;
6043 pdscp = get_priority(ofport, ctx->xin->flow.skb_priority);
6045 ctx->xin->flow.nw_tos &= ~IP_DSCP_MASK;
6046 ctx->xin->flow.nw_tos |= pdscp->dscp;
6049 if (ofport->tnl_port) {
6050 /* Save tunnel metadata so that changes made due to
6051 * the Logical (tunnel) Port are not visible for any further
6052 * matches, while explicit set actions on tunnel metadata are.
6054 struct flow_tnl flow_tnl = ctx->xin->flow.tunnel;
6055 odp_port = tnl_port_send(ofport->tnl_port, &ctx->xin->flow);
6056 if (odp_port == OVSP_NONE) {
6057 xlate_report(ctx, "Tunneling decided against output");
6058 goto out; /* restore flow_nw_tos */
6061 if (ctx->xin->resubmit_stats) {
6062 netdev_vport_inc_tx(ofport->up.netdev, ctx->xin->resubmit_stats);
6064 out_port = odp_port;
6065 commit_odp_tunnel_action(&ctx->xin->flow, &ctx->base_flow,
6066 &ctx->xout->odp_actions);
6067 ctx->xin->flow.tunnel = flow_tnl; /* Restore tunnel metadata */
6069 uint16_t vlandev_port;
6070 odp_port = ofport->odp_port;
6071 vlandev_port = vsp_realdev_to_vlandev(ctx->ofproto, ofp_port,
6072 ctx->xin->flow.vlan_tci);
6073 if (vlandev_port == ofp_port) {
6074 out_port = odp_port;
6076 out_port = ofp_port_to_odp_port(ctx->ofproto, vlandev_port);
6077 ctx->xin->flow.vlan_tci = htons(0);
6079 ctx->xin->flow.skb_mark &= ~IPSEC_MARK;
6081 commit_odp_actions(&ctx->xin->flow, &ctx->base_flow,
6082 &ctx->xout->odp_actions);
6083 nl_msg_put_u32(&ctx->xout->odp_actions, OVS_ACTION_ATTR_OUTPUT, out_port);
6085 ctx->sflow_odp_port = odp_port;
6086 ctx->sflow_n_outputs++;
6087 ctx->xout->nf_output_iface = ofp_port;
6090 ctx->xin->flow.vlan_tci = flow_vlan_tci;
6091 ctx->xin->flow.skb_mark = flow_skb_mark;
6093 ctx->xin->flow.nw_tos = flow_nw_tos;
6097 compose_output_action(struct xlate_ctx *ctx, uint16_t ofp_port)
6099 compose_output_action__(ctx, ofp_port, true);
6103 tag_the_flow(struct xlate_ctx *ctx, struct rule_dpif *rule)
6105 struct ofproto_dpif *ofproto = ctx->ofproto;
6106 uint8_t table_id = ctx->table_id;
6108 if (table_id > 0 && table_id < N_TABLES) {
6109 struct table_dpif *table = &ofproto->tables[table_id];
6110 if (table->other_table) {
6111 ctx->xout->tags |= (rule && rule->tag
6113 : rule_calculate_tag(&ctx->xin->flow,
6114 &table->other_table->mask,
6120 /* Common rule processing in one place to avoid duplicating code. */
6121 static struct rule_dpif *
6122 ctx_rule_hooks(struct xlate_ctx *ctx, struct rule_dpif *rule,
6125 if (ctx->xin->resubmit_hook) {
6126 ctx->xin->resubmit_hook(ctx, rule);
6128 if (rule == NULL && may_packet_in) {
6130 * check if table configuration flags
6131 * OFPTC_TABLE_MISS_CONTROLLER, default.
6132 * OFPTC_TABLE_MISS_CONTINUE,
6133 * OFPTC_TABLE_MISS_DROP
6134 * When OF1.0, OFPTC_TABLE_MISS_CONTINUE is used. What to do?
6136 rule = rule_dpif_miss_rule(ctx->ofproto, &ctx->xin->flow);
6138 if (rule && ctx->xin->resubmit_stats) {
6139 rule_credit_stats(rule, ctx->xin->resubmit_stats);
6145 xlate_table_action(struct xlate_ctx *ctx,
6146 uint16_t in_port, uint8_t table_id, bool may_packet_in)
6148 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
6149 struct rule_dpif *rule;
6150 uint16_t old_in_port = ctx->xin->flow.in_port;
6151 uint8_t old_table_id = ctx->table_id;
6153 ctx->table_id = table_id;
6155 /* Look up a flow with 'in_port' as the input port. */
6156 ctx->xin->flow.in_port = in_port;
6157 rule = rule_dpif_lookup__(ctx->ofproto, &ctx->xin->flow, table_id);
6159 tag_the_flow(ctx, rule);
6161 /* Restore the original input port. Otherwise OFPP_NORMAL and
6162 * OFPP_IN_PORT will have surprising behavior. */
6163 ctx->xin->flow.in_port = old_in_port;
6165 rule = ctx_rule_hooks(ctx, rule, may_packet_in);
6168 struct rule_dpif *old_rule = ctx->rule;
6172 do_xlate_actions(rule->up.ofpacts, rule->up.ofpacts_len, ctx);
6173 ctx->rule = old_rule;
6177 ctx->table_id = old_table_id;
6179 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
6181 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
6182 MAX_RESUBMIT_RECURSION);
6183 ctx->max_resubmit_trigger = true;
6188 xlate_ofpact_resubmit(struct xlate_ctx *ctx,
6189 const struct ofpact_resubmit *resubmit)
6194 in_port = resubmit->in_port;
6195 if (in_port == OFPP_IN_PORT) {
6196 in_port = ctx->xin->flow.in_port;
6199 table_id = resubmit->table_id;
6200 if (table_id == 255) {
6201 table_id = ctx->table_id;
6204 xlate_table_action(ctx, in_port, table_id, false);
6208 flood_packets(struct xlate_ctx *ctx, bool all)
6210 struct ofport_dpif *ofport;
6212 HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) {
6213 uint16_t ofp_port = ofport->up.ofp_port;
6215 if (ofp_port == ctx->xin->flow.in_port) {
6220 compose_output_action__(ctx, ofp_port, false);
6221 } else if (!(ofport->up.pp.config & OFPUTIL_PC_NO_FLOOD)) {
6222 compose_output_action(ctx, ofp_port);
6226 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
6230 execute_controller_action(struct xlate_ctx *ctx, int len,
6231 enum ofp_packet_in_reason reason,
6232 uint16_t controller_id)
6234 struct ofputil_packet_in pin;
6235 struct ofpbuf *packet;
6237 ovs_assert(!ctx->xout->slow || ctx->xout->slow == SLOW_CONTROLLER);
6238 ctx->xout->slow = SLOW_CONTROLLER;
6239 if (!ctx->xin->packet) {
6243 packet = ofpbuf_clone(ctx->xin->packet);
6245 if (packet->l2 && packet->l3) {
6246 struct eth_header *eh;
6247 uint16_t mpls_depth;
6249 eth_pop_vlan(packet);
6252 memcpy(eh->eth_src, ctx->xin->flow.dl_src, sizeof eh->eth_src);
6253 memcpy(eh->eth_dst, ctx->xin->flow.dl_dst, sizeof eh->eth_dst);
6255 if (ctx->xin->flow.vlan_tci & htons(VLAN_CFI)) {
6256 eth_push_vlan(packet, ctx->xin->flow.vlan_tci);
6259 mpls_depth = eth_mpls_depth(packet);
6261 if (mpls_depth < ctx->xin->flow.mpls_depth) {
6262 push_mpls(packet, ctx->xin->flow.dl_type, ctx->xin->flow.mpls_lse);
6263 } else if (mpls_depth > ctx->xin->flow.mpls_depth) {
6264 pop_mpls(packet, ctx->xin->flow.dl_type);
6265 } else if (mpls_depth) {
6266 set_mpls_lse(packet, ctx->xin->flow.mpls_lse);
6270 if (ctx->xin->flow.dl_type == htons(ETH_TYPE_IP)) {
6271 packet_set_ipv4(packet, ctx->xin->flow.nw_src,
6272 ctx->xin->flow.nw_dst, ctx->xin->flow.nw_tos,
6273 ctx->xin->flow.nw_ttl);
6277 if (ctx->xin->flow.nw_proto == IPPROTO_TCP) {
6278 packet_set_tcp_port(packet, ctx->xin->flow.tp_src,
6279 ctx->xin->flow.tp_dst);
6280 } else if (ctx->xin->flow.nw_proto == IPPROTO_UDP) {
6281 packet_set_udp_port(packet, ctx->xin->flow.tp_src,
6282 ctx->xin->flow.tp_dst);
6288 pin.packet = packet->data;
6289 pin.packet_len = packet->size;
6290 pin.reason = reason;
6291 pin.controller_id = controller_id;
6292 pin.table_id = ctx->table_id;
6293 pin.cookie = ctx->rule ? ctx->rule->up.flow_cookie : 0;
6296 flow_get_metadata(&ctx->xin->flow, &pin.fmd);
6298 connmgr_send_packet_in(ctx->ofproto->up.connmgr, &pin);
6299 ofpbuf_delete(packet);
6303 execute_mpls_push_action(struct xlate_ctx *ctx, ovs_be16 eth_type)
6305 ovs_assert(eth_type_mpls(eth_type));
6307 if (ctx->base_flow.mpls_depth) {
6308 ctx->xin->flow.mpls_lse &= ~htonl(MPLS_BOS_MASK);
6309 ctx->xin->flow.mpls_depth++;
6314 if (ctx->xin->flow.dl_type == htons(ETH_TYPE_IPV6)) {
6315 label = htonl(0x2); /* IPV6 Explicit Null. */
6317 label = htonl(0x0); /* IPV4 Explicit Null. */
6319 tc = (ctx->xin->flow.nw_tos & IP_DSCP_MASK) >> 2;
6320 ttl = ctx->xin->flow.nw_ttl ? ctx->xin->flow.nw_ttl : 0x40;
6321 ctx->xin->flow.mpls_lse = set_mpls_lse_values(ttl, tc, 1, label);
6322 ctx->xin->flow.mpls_depth = 1;
6324 ctx->xin->flow.dl_type = eth_type;
6328 execute_mpls_pop_action(struct xlate_ctx *ctx, ovs_be16 eth_type)
6330 ovs_assert(eth_type_mpls(ctx->xin->flow.dl_type));
6331 ovs_assert(!eth_type_mpls(eth_type));
6333 if (ctx->xin->flow.mpls_depth) {
6334 ctx->xin->flow.mpls_depth--;
6335 ctx->xin->flow.mpls_lse = htonl(0);
6336 if (!ctx->xin->flow.mpls_depth) {
6337 ctx->xin->flow.dl_type = eth_type;
6343 compose_dec_ttl(struct xlate_ctx *ctx, struct ofpact_cnt_ids *ids)
6345 if (ctx->xin->flow.dl_type != htons(ETH_TYPE_IP) &&
6346 ctx->xin->flow.dl_type != htons(ETH_TYPE_IPV6)) {
6350 if (ctx->xin->flow.nw_ttl > 1) {
6351 ctx->xin->flow.nw_ttl--;
6356 for (i = 0; i < ids->n_controllers; i++) {
6357 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL,
6361 /* Stop processing for current table. */
6367 execute_set_mpls_ttl_action(struct xlate_ctx *ctx, uint8_t ttl)
6369 if (!eth_type_mpls(ctx->xin->flow.dl_type)) {
6373 set_mpls_lse_ttl(&ctx->xin->flow.mpls_lse, ttl);
6378 execute_dec_mpls_ttl_action(struct xlate_ctx *ctx)
6380 uint8_t ttl = mpls_lse_to_ttl(ctx->xin->flow.mpls_lse);
6382 if (!eth_type_mpls(ctx->xin->flow.dl_type)) {
6388 set_mpls_lse_ttl(&ctx->xin->flow.mpls_lse, ttl);
6391 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL, 0);
6393 /* Stop processing for current table. */
6399 xlate_output_action(struct xlate_ctx *ctx,
6400 uint16_t port, uint16_t max_len, bool may_packet_in)
6402 uint16_t prev_nf_output_iface = ctx->xout->nf_output_iface;
6404 ctx->xout->nf_output_iface = NF_OUT_DROP;
6408 compose_output_action(ctx, ctx->xin->flow.in_port);
6411 xlate_table_action(ctx, ctx->xin->flow.in_port, 0, may_packet_in);
6417 flood_packets(ctx, false);
6420 flood_packets(ctx, true);
6422 case OFPP_CONTROLLER:
6423 execute_controller_action(ctx, max_len, OFPR_ACTION, 0);
6429 if (port != ctx->xin->flow.in_port) {
6430 compose_output_action(ctx, port);
6432 xlate_report(ctx, "skipping output to input port");
6437 if (prev_nf_output_iface == NF_OUT_FLOOD) {
6438 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
6439 } else if (ctx->xout->nf_output_iface == NF_OUT_DROP) {
6440 ctx->xout->nf_output_iface = prev_nf_output_iface;
6441 } else if (prev_nf_output_iface != NF_OUT_DROP &&
6442 ctx->xout->nf_output_iface != NF_OUT_FLOOD) {
6443 ctx->xout->nf_output_iface = NF_OUT_MULTI;
6448 xlate_output_reg_action(struct xlate_ctx *ctx,
6449 const struct ofpact_output_reg *or)
6451 uint64_t port = mf_get_subfield(&or->src, &ctx->xin->flow);
6452 if (port <= UINT16_MAX) {
6453 xlate_output_action(ctx, port, or->max_len, false);
6458 xlate_enqueue_action(struct xlate_ctx *ctx,
6459 const struct ofpact_enqueue *enqueue)
6461 uint16_t ofp_port = enqueue->port;
6462 uint32_t queue_id = enqueue->queue;
6463 uint32_t flow_priority, priority;
6466 /* Translate queue to priority. */
6467 error = dpif_queue_to_priority(ctx->ofproto->backer->dpif,
6468 queue_id, &priority);
6470 /* Fall back to ordinary output action. */
6471 xlate_output_action(ctx, enqueue->port, 0, false);
6475 /* Check output port. */
6476 if (ofp_port == OFPP_IN_PORT) {
6477 ofp_port = ctx->xin->flow.in_port;
6478 } else if (ofp_port == ctx->xin->flow.in_port) {
6482 /* Add datapath actions. */
6483 flow_priority = ctx->xin->flow.skb_priority;
6484 ctx->xin->flow.skb_priority = priority;
6485 compose_output_action(ctx, ofp_port);
6486 ctx->xin->flow.skb_priority = flow_priority;
6488 /* Update NetFlow output port. */
6489 if (ctx->xout->nf_output_iface == NF_OUT_DROP) {
6490 ctx->xout->nf_output_iface = ofp_port;
6491 } else if (ctx->xout->nf_output_iface != NF_OUT_FLOOD) {
6492 ctx->xout->nf_output_iface = NF_OUT_MULTI;
6497 xlate_set_queue_action(struct xlate_ctx *ctx, uint32_t queue_id)
6499 uint32_t skb_priority;
6501 if (!dpif_queue_to_priority(ctx->ofproto->backer->dpif,
6502 queue_id, &skb_priority)) {
6503 ctx->xin->flow.skb_priority = skb_priority;
6505 /* Couldn't translate queue to a priority. Nothing to do. A warning
6506 * has already been logged. */
6511 slave_enabled_cb(uint16_t ofp_port, void *ofproto_)
6513 struct ofproto_dpif *ofproto = ofproto_;
6514 struct ofport_dpif *port;
6524 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
6527 port = get_ofp_port(ofproto, ofp_port);
6528 return port ? port->may_enable : false;
6533 xlate_bundle_action(struct xlate_ctx *ctx,
6534 const struct ofpact_bundle *bundle)
6538 port = bundle_execute(bundle, &ctx->xin->flow, slave_enabled_cb,
6540 if (bundle->dst.field) {
6541 nxm_reg_load(&bundle->dst, port, &ctx->xin->flow);
6543 xlate_output_action(ctx, port, 0, false);
6548 xlate_learn_action(struct xlate_ctx *ctx,
6549 const struct ofpact_learn *learn)
6551 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
6552 struct ofputil_flow_mod fm;
6553 uint64_t ofpacts_stub[1024 / 8];
6554 struct ofpbuf ofpacts;
6557 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
6558 learn_execute(learn, &ctx->xin->flow, &fm, &ofpacts);
6560 error = ofproto_flow_mod(&ctx->ofproto->up, &fm);
6561 if (error && !VLOG_DROP_WARN(&rl)) {
6562 VLOG_WARN("learning action failed to modify flow table (%s)",
6563 ofperr_get_name(error));
6566 ofpbuf_uninit(&ofpacts);
6569 /* Reduces '*timeout' to no more than 'max'. A value of zero in either case
6570 * means "infinite". */
6572 reduce_timeout(uint16_t max, uint16_t *timeout)
6574 if (max && (!*timeout || *timeout > max)) {
6580 xlate_fin_timeout(struct xlate_ctx *ctx,
6581 const struct ofpact_fin_timeout *oft)
6583 if (ctx->xin->tcp_flags & (TCP_FIN | TCP_RST) && ctx->rule) {
6584 struct rule_dpif *rule = ctx->rule;
6586 reduce_timeout(oft->fin_idle_timeout, &rule->up.idle_timeout);
6587 reduce_timeout(oft->fin_hard_timeout, &rule->up.hard_timeout);
6592 xlate_sample_action(struct xlate_ctx *ctx,
6593 const struct ofpact_sample *os)
6595 union user_action_cookie cookie;
6596 /* Scale the probability from 16-bit to 32-bit while representing
6597 * the same percentage. */
6598 uint32_t probability = (os->probability << 16) | os->probability;
6600 commit_odp_actions(&ctx->xin->flow, &ctx->base_flow,
6601 &ctx->xout->odp_actions);
6603 compose_flow_sample_cookie(os->probability, os->collector_set_id,
6604 os->obs_domain_id, os->obs_point_id, &cookie);
6605 compose_sample_action(ctx->ofproto, &ctx->xout->odp_actions, &ctx->xin->flow,
6606 probability, &cookie, sizeof cookie.flow_sample);
6610 may_receive(const struct ofport_dpif *port, struct xlate_ctx *ctx)
6612 if (port->up.pp.config & (eth_addr_equals(ctx->xin->flow.dl_dst,
6614 ? OFPUTIL_PC_NO_RECV_STP
6615 : OFPUTIL_PC_NO_RECV)) {
6619 /* Only drop packets here if both forwarding and learning are
6620 * disabled. If just learning is enabled, we need to have
6621 * OFPP_NORMAL and the learning action have a look at the packet
6622 * before we can drop it. */
6623 if (!stp_forward_in_state(port->stp_state)
6624 && !stp_learn_in_state(port->stp_state)) {
6632 tunnel_ecn_ok(struct xlate_ctx *ctx)
6634 if (is_ip_any(&ctx->base_flow)
6635 && (ctx->base_flow.tunnel.ip_tos & IP_ECN_MASK) == IP_ECN_CE) {
6636 if ((ctx->base_flow.nw_tos & IP_ECN_MASK) == IP_ECN_NOT_ECT) {
6637 VLOG_WARN_RL(&rl, "dropping tunnel packet marked ECN CE"
6638 " but is not ECN capable");
6641 /* Set the ECN CE value in the tunneled packet. */
6642 ctx->xin->flow.nw_tos |= IP_ECN_CE;
6650 do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len,
6651 struct xlate_ctx *ctx)
6653 bool was_evictable = true;
6654 const struct ofpact *a;
6657 /* Don't let the rule we're working on get evicted underneath us. */
6658 was_evictable = ctx->rule->up.evictable;
6659 ctx->rule->up.evictable = false;
6662 do_xlate_actions_again:
6663 OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) {
6664 struct ofpact_controller *controller;
6665 const struct ofpact_metadata *metadata;
6673 xlate_output_action(ctx, ofpact_get_OUTPUT(a)->port,
6674 ofpact_get_OUTPUT(a)->max_len, true);
6677 case OFPACT_CONTROLLER:
6678 controller = ofpact_get_CONTROLLER(a);
6679 execute_controller_action(ctx, controller->max_len,
6681 controller->controller_id);
6684 case OFPACT_ENQUEUE:
6685 xlate_enqueue_action(ctx, ofpact_get_ENQUEUE(a));
6688 case OFPACT_SET_VLAN_VID:
6689 ctx->xin->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
6690 ctx->xin->flow.vlan_tci |=
6691 (htons(ofpact_get_SET_VLAN_VID(a)->vlan_vid)
6695 case OFPACT_SET_VLAN_PCP:
6696 ctx->xin->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
6697 ctx->xin->flow.vlan_tci |=
6698 htons((ofpact_get_SET_VLAN_PCP(a)->vlan_pcp << VLAN_PCP_SHIFT)
6702 case OFPACT_STRIP_VLAN:
6703 ctx->xin->flow.vlan_tci = htons(0);
6706 case OFPACT_PUSH_VLAN:
6707 /* XXX 802.1AD(QinQ) */
6708 ctx->xin->flow.vlan_tci = htons(VLAN_CFI);
6711 case OFPACT_SET_ETH_SRC:
6712 memcpy(ctx->xin->flow.dl_src, ofpact_get_SET_ETH_SRC(a)->mac,
6716 case OFPACT_SET_ETH_DST:
6717 memcpy(ctx->xin->flow.dl_dst, ofpact_get_SET_ETH_DST(a)->mac,
6721 case OFPACT_SET_IPV4_SRC:
6722 if (ctx->xin->flow.dl_type == htons(ETH_TYPE_IP)) {
6723 ctx->xin->flow.nw_src = ofpact_get_SET_IPV4_SRC(a)->ipv4;
6727 case OFPACT_SET_IPV4_DST:
6728 if (ctx->xin->flow.dl_type == htons(ETH_TYPE_IP)) {
6729 ctx->xin->flow.nw_dst = ofpact_get_SET_IPV4_DST(a)->ipv4;
6733 case OFPACT_SET_IPV4_DSCP:
6734 /* OpenFlow 1.0 only supports IPv4. */
6735 if (ctx->xin->flow.dl_type == htons(ETH_TYPE_IP)) {
6736 ctx->xin->flow.nw_tos &= ~IP_DSCP_MASK;
6737 ctx->xin->flow.nw_tos |= ofpact_get_SET_IPV4_DSCP(a)->dscp;
6741 case OFPACT_SET_L4_SRC_PORT:
6742 if (is_ip_any(&ctx->xin->flow)) {
6743 ctx->xin->flow.tp_src =
6744 htons(ofpact_get_SET_L4_SRC_PORT(a)->port);
6748 case OFPACT_SET_L4_DST_PORT:
6749 if (is_ip_any(&ctx->xin->flow)) {
6750 ctx->xin->flow.tp_dst =
6751 htons(ofpact_get_SET_L4_DST_PORT(a)->port);
6755 case OFPACT_RESUBMIT:
6756 xlate_ofpact_resubmit(ctx, ofpact_get_RESUBMIT(a));
6759 case OFPACT_SET_TUNNEL:
6760 ctx->xin->flow.tunnel.tun_id =
6761 htonll(ofpact_get_SET_TUNNEL(a)->tun_id);
6764 case OFPACT_SET_QUEUE:
6765 xlate_set_queue_action(ctx, ofpact_get_SET_QUEUE(a)->queue_id);
6768 case OFPACT_POP_QUEUE:
6769 ctx->xin->flow.skb_priority = ctx->orig_skb_priority;
6772 case OFPACT_REG_MOVE:
6773 nxm_execute_reg_move(ofpact_get_REG_MOVE(a), &ctx->xin->flow);
6776 case OFPACT_REG_LOAD:
6777 nxm_execute_reg_load(ofpact_get_REG_LOAD(a), &ctx->xin->flow);
6780 case OFPACT_STACK_PUSH:
6781 nxm_execute_stack_push(ofpact_get_STACK_PUSH(a), &ctx->xin->flow,
6785 case OFPACT_STACK_POP:
6786 nxm_execute_stack_pop(ofpact_get_STACK_POP(a), &ctx->xin->flow,
6790 case OFPACT_PUSH_MPLS:
6791 execute_mpls_push_action(ctx, ofpact_get_PUSH_MPLS(a)->ethertype);
6794 case OFPACT_POP_MPLS:
6795 execute_mpls_pop_action(ctx, ofpact_get_POP_MPLS(a)->ethertype);
6798 case OFPACT_SET_MPLS_TTL:
6799 if (execute_set_mpls_ttl_action(ctx,
6800 ofpact_get_SET_MPLS_TTL(a)->ttl)) {
6805 case OFPACT_DEC_MPLS_TTL:
6806 if (execute_dec_mpls_ttl_action(ctx)) {
6811 case OFPACT_DEC_TTL:
6812 if (compose_dec_ttl(ctx, ofpact_get_DEC_TTL(a))) {
6818 /* Nothing to do. */
6821 case OFPACT_MULTIPATH:
6822 multipath_execute(ofpact_get_MULTIPATH(a), &ctx->xin->flow);
6826 ctx->ofproto->has_bundle_action = true;
6827 xlate_bundle_action(ctx, ofpact_get_BUNDLE(a));
6830 case OFPACT_OUTPUT_REG:
6831 xlate_output_reg_action(ctx, ofpact_get_OUTPUT_REG(a));
6835 ctx->xout->has_learn = true;
6836 if (ctx->xin->may_learn) {
6837 xlate_learn_action(ctx, ofpact_get_LEARN(a));
6845 case OFPACT_FIN_TIMEOUT:
6846 ctx->xout->has_fin_timeout = true;
6847 xlate_fin_timeout(ctx, ofpact_get_FIN_TIMEOUT(a));
6850 case OFPACT_CLEAR_ACTIONS:
6852 * Nothing to do because writa-actions is not supported for now.
6853 * When writa-actions is supported, clear-actions also must
6854 * be supported at the same time.
6858 case OFPACT_WRITE_METADATA:
6859 metadata = ofpact_get_WRITE_METADATA(a);
6860 ctx->xin->flow.metadata &= ~metadata->mask;
6861 ctx->xin->flow.metadata |= metadata->metadata & metadata->mask;
6864 case OFPACT_GOTO_TABLE: {
6865 /* It is assumed that goto-table is the last action. */
6866 struct ofpact_goto_table *ogt = ofpact_get_GOTO_TABLE(a);
6867 struct rule_dpif *rule;
6869 ovs_assert(ctx->table_id < ogt->table_id);
6871 ctx->table_id = ogt->table_id;
6873 /* Look up a flow from the new table. */
6874 rule = rule_dpif_lookup__(ctx->ofproto, &ctx->xin->flow, ctx->table_id);
6876 tag_the_flow(ctx, rule);
6878 rule = ctx_rule_hooks(ctx, rule, true);
6882 ctx->rule->up.evictable = was_evictable;
6885 was_evictable = rule->up.evictable;
6886 rule->up.evictable = false;
6888 /* Tail recursion removal. */
6889 ofpacts = rule->up.ofpacts;
6890 ofpacts_len = rule->up.ofpacts_len;
6891 goto do_xlate_actions_again;
6897 xlate_sample_action(ctx, ofpact_get_SAMPLE(a));
6904 ctx->rule->up.evictable = was_evictable;
6909 xlate_in_init(struct xlate_in *xin, struct ofproto_dpif *ofproto,
6910 const struct flow *flow,
6911 const struct initial_vals *initial_vals,
6912 struct rule_dpif *rule, uint8_t tcp_flags,
6913 const struct ofpbuf *packet)
6915 xin->ofproto = ofproto;
6917 xin->packet = packet;
6918 xin->may_learn = packet != NULL;
6920 xin->ofpacts = NULL;
6921 xin->ofpacts_len = 0;
6922 xin->tcp_flags = tcp_flags;
6923 xin->resubmit_hook = NULL;
6924 xin->report_hook = NULL;
6925 xin->resubmit_stats = NULL;
6928 xin->initial_vals = *initial_vals;
6930 xin->initial_vals.vlan_tci = xin->flow.vlan_tci;
6931 xin->initial_vals.tunnel_ip_tos = xin->flow.tunnel.ip_tos;
6936 xlate_out_uninit(struct xlate_out *xout)
6939 ofpbuf_uninit(&xout->odp_actions);
6943 /* Translates the 'ofpacts_len' bytes of "struct ofpacts" starting at 'ofpacts'
6944 * into datapath actions in 'odp_actions', using 'ctx'. */
6946 xlate_actions(struct xlate_in *xin, struct xlate_out *xout)
6948 /* Normally false. Set to true if we ever hit MAX_RESUBMIT_RECURSION, so
6949 * that in the future we always keep a copy of the original flow for
6950 * tracing purposes. */
6951 static bool hit_resubmit_limit;
6953 enum slow_path_reason special;
6954 const struct ofpact *ofpacts;
6955 struct ofport_dpif *in_port;
6956 struct flow orig_flow;
6957 struct xlate_ctx ctx;
6960 COVERAGE_INC(ofproto_dpif_xlate);
6962 /* Flow initialization rules:
6963 * - 'base_flow' must match the kernel's view of the packet at the
6964 * time that action processing starts. 'flow' represents any
6965 * transformations we wish to make through actions.
6966 * - By default 'base_flow' and 'flow' are the same since the input
6967 * packet matches the output before any actions are applied.
6968 * - When using VLAN splinters, 'base_flow''s VLAN is set to the value
6969 * of the received packet as seen by the kernel. If we later output
6970 * to another device without any modifications this will cause us to
6971 * insert a new tag since the original one was stripped off by the
6973 * - Tunnel 'flow' is largely cleared when transitioning between
6974 * the input and output stages since it does not make sense to output
6975 * a packet with the exact headers that it was received with (i.e.
6976 * the destination IP is us). The one exception is the tun_id, which
6977 * is preserved to allow use in later resubmit lookups and loads into
6979 * - Tunnel 'base_flow' is completely cleared since that is what the
6980 * kernel does. If we wish to maintain the original values an action
6981 * needs to be generated. */
6986 ctx.ofproto = xin->ofproto;
6987 ctx.rule = xin->rule;
6989 ctx.base_flow = ctx.xin->flow;
6990 memset(&ctx.base_flow.tunnel, 0, sizeof ctx.base_flow.tunnel);
6991 ctx.base_flow.vlan_tci = xin->initial_vals.vlan_tci;
6992 ctx.base_flow.tunnel.ip_tos = xin->initial_vals.tunnel_ip_tos;
6996 ctx.xout->has_learn = false;
6997 ctx.xout->has_normal = false;
6998 ctx.xout->has_fin_timeout = false;
6999 ctx.xout->nf_output_iface = NF_OUT_DROP;
7000 ctx.xout->mirrors = 0;
7002 ofpbuf_use_stub(&ctx.xout->odp_actions, ctx.xout->odp_actions_stub,
7003 sizeof ctx.xout->odp_actions_stub);
7004 ofpbuf_reserve(&ctx.xout->odp_actions, NL_A_U32_SIZE);
7007 ctx.max_resubmit_trigger = false;
7008 ctx.orig_skb_priority = ctx.xin->flow.skb_priority;
7013 ofpacts = xin->ofpacts;
7014 ofpacts_len = xin->ofpacts_len;
7015 } else if (xin->rule) {
7016 ofpacts = xin->rule->up.ofpacts;
7017 ofpacts_len = xin->rule->up.ofpacts_len;
7022 ofpbuf_use_stub(&ctx.stack, ctx.init_stack, sizeof ctx.init_stack);
7024 if (ctx.ofproto->has_mirrors || hit_resubmit_limit) {
7025 /* Do this conditionally because the copy is expensive enough that it
7026 * shows up in profiles. */
7027 orig_flow = ctx.xin->flow;
7030 if (ctx.xin->flow.nw_frag & FLOW_NW_FRAG_ANY) {
7031 switch (ctx.ofproto->up.frag_handling) {
7032 case OFPC_FRAG_NORMAL:
7033 /* We must pretend that transport ports are unavailable. */
7034 ctx.xin->flow.tp_src = ctx.base_flow.tp_src = htons(0);
7035 ctx.xin->flow.tp_dst = ctx.base_flow.tp_dst = htons(0);
7038 case OFPC_FRAG_DROP:
7041 case OFPC_FRAG_REASM:
7044 case OFPC_FRAG_NX_MATCH:
7045 /* Nothing to do. */
7048 case OFPC_INVALID_TTL_TO_CONTROLLER:
7053 in_port = get_ofp_port(ctx.ofproto, ctx.xin->flow.in_port);
7054 special = process_special(ctx.ofproto, &ctx.xin->flow, in_port,
7057 ctx.xout->slow = special;
7059 static struct vlog_rate_limit trace_rl = VLOG_RATE_LIMIT_INIT(1, 1);
7060 struct initial_vals initial_vals;
7061 size_t sample_actions_len;
7062 uint32_t local_odp_port;
7064 initial_vals.vlan_tci = ctx.base_flow.vlan_tci;
7065 initial_vals.tunnel_ip_tos = ctx.base_flow.tunnel.ip_tos;
7067 add_sflow_action(&ctx);
7068 add_ipfix_action(&ctx);
7069 sample_actions_len = ctx.xout->odp_actions.size;
7071 if (tunnel_ecn_ok(&ctx) && (!in_port || may_receive(in_port, &ctx))) {
7072 do_xlate_actions(ofpacts, ofpacts_len, &ctx);
7074 /* We've let OFPP_NORMAL and the learning action look at the
7075 * packet, so drop it now if forwarding is disabled. */
7076 if (in_port && !stp_forward_in_state(in_port->stp_state)) {
7077 ctx.xout->odp_actions.size = sample_actions_len;
7081 if (ctx.max_resubmit_trigger && !ctx.xin->resubmit_hook) {
7082 if (!hit_resubmit_limit) {
7083 /* We didn't record the original flow. Make sure we do from
7085 hit_resubmit_limit = true;
7086 } else if (!VLOG_DROP_ERR(&trace_rl)) {
7087 struct ds ds = DS_EMPTY_INITIALIZER;
7089 ofproto_trace(ctx.ofproto, &orig_flow, ctx.xin->packet,
7090 &initial_vals, &ds);
7091 VLOG_ERR("Trace triggered by excessive resubmit "
7092 "recursion:\n%s", ds_cstr(&ds));
7097 local_odp_port = ofp_port_to_odp_port(ctx.ofproto, OFPP_LOCAL);
7098 if (!connmgr_must_output_local(ctx.ofproto->up.connmgr, &ctx.xin->flow,
7100 ctx.xout->odp_actions.data,
7101 ctx.xout->odp_actions.size)) {
7102 compose_output_action(&ctx, OFPP_LOCAL);
7104 if (ctx.ofproto->has_mirrors) {
7105 add_mirror_actions(&ctx, &orig_flow);
7107 fix_sflow_action(&ctx);
7110 ofpbuf_uninit(&ctx.stack);
7113 /* Translates the 'ofpacts_len' bytes of "struct ofpact"s starting at 'ofpacts'
7114 * into datapath actions, using 'ctx', and discards the datapath actions. */
7116 xlate_actions_for_side_effects(struct xlate_in *xin)
7118 struct xlate_out xout;
7120 xlate_actions(xin, &xout);
7121 xlate_out_uninit(&xout);
7125 xlate_report(struct xlate_ctx *ctx, const char *s)
7127 if (ctx->xin->report_hook) {
7128 ctx->xin->report_hook(ctx, s);
7132 /* OFPP_NORMAL implementation. */
7134 static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *);
7136 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
7137 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle',
7138 * the bundle on which the packet was received, returns the VLAN to which the
7141 * Both 'vid' and the return value are in the range 0...4095. */
7143 input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid)
7145 switch (in_bundle->vlan_mode) {
7146 case PORT_VLAN_ACCESS:
7147 return in_bundle->vlan;
7150 case PORT_VLAN_TRUNK:
7153 case PORT_VLAN_NATIVE_UNTAGGED:
7154 case PORT_VLAN_NATIVE_TAGGED:
7155 return vid ? vid : in_bundle->vlan;
7162 /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'.
7163 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
7166 * 'vid' should be the VID obtained from the 802.1Q header that was received as
7167 * part of a packet (specify 0 if there was no 802.1Q header), in the range
7170 input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn)
7172 /* Allow any VID on the OFPP_NONE port. */
7173 if (in_bundle == &ofpp_none_bundle) {
7177 switch (in_bundle->vlan_mode) {
7178 case PORT_VLAN_ACCESS:
7181 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7182 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
7183 "packet received on port %s configured as VLAN "
7184 "%"PRIu16" access port",
7185 in_bundle->ofproto->up.name, vid,
7186 in_bundle->name, in_bundle->vlan);
7192 case PORT_VLAN_NATIVE_UNTAGGED:
7193 case PORT_VLAN_NATIVE_TAGGED:
7195 /* Port must always carry its native VLAN. */
7199 case PORT_VLAN_TRUNK:
7200 if (!ofbundle_includes_vlan(in_bundle, vid)) {
7202 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7203 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet "
7204 "received on port %s not configured for trunking "
7206 in_bundle->ofproto->up.name, vid,
7207 in_bundle->name, vid);
7219 /* Given 'vlan', the VLAN that a packet belongs to, and
7220 * 'out_bundle', a bundle on which the packet is to be output, returns the VID
7221 * that should be included in the 802.1Q header. (If the return value is 0,
7222 * then the 802.1Q header should only be included in the packet if there is a
7225 * Both 'vlan' and the return value are in the range 0...4095. */
7227 output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan)
7229 switch (out_bundle->vlan_mode) {
7230 case PORT_VLAN_ACCESS:
7233 case PORT_VLAN_TRUNK:
7234 case PORT_VLAN_NATIVE_TAGGED:
7237 case PORT_VLAN_NATIVE_UNTAGGED:
7238 return vlan == out_bundle->vlan ? 0 : vlan;
7246 output_normal(struct xlate_ctx *ctx, const struct ofbundle *out_bundle,
7249 struct ofport_dpif *port;
7251 ovs_be16 tci, old_tci;
7253 vid = output_vlan_to_vid(out_bundle, vlan);
7254 if (!out_bundle->bond) {
7255 port = ofbundle_get_a_port(out_bundle);
7257 port = bond_choose_output_slave(out_bundle->bond, &ctx->xin->flow,
7258 vid, &ctx->xout->tags);
7260 /* No slaves enabled, so drop packet. */
7265 old_tci = ctx->xin->flow.vlan_tci;
7267 if (tci || out_bundle->use_priority_tags) {
7268 tci |= ctx->xin->flow.vlan_tci & htons(VLAN_PCP_MASK);
7270 tci |= htons(VLAN_CFI);
7273 ctx->xin->flow.vlan_tci = tci;
7275 compose_output_action(ctx, port->up.ofp_port);
7276 ctx->xin->flow.vlan_tci = old_tci;
7280 mirror_mask_ffs(mirror_mask_t mask)
7282 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
7287 ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
7289 return (bundle->vlan_mode != PORT_VLAN_ACCESS
7290 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
7294 ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan)
7296 return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan);
7299 /* Returns an arbitrary interface within 'bundle'. */
7300 static struct ofport_dpif *
7301 ofbundle_get_a_port(const struct ofbundle *bundle)
7303 return CONTAINER_OF(list_front(&bundle->ports),
7304 struct ofport_dpif, bundle_node);
7308 vlan_is_mirrored(const struct ofmirror *m, int vlan)
7310 return !m->vlans || bitmap_is_set(m->vlans, vlan);
7314 add_mirror_actions(struct xlate_ctx *ctx, const struct flow *orig_flow)
7316 struct ofproto_dpif *ofproto = ctx->ofproto;
7317 mirror_mask_t mirrors;
7318 struct ofbundle *in_bundle;
7321 const struct nlattr *a;
7324 in_bundle = lookup_input_bundle(ctx->ofproto, orig_flow->in_port,
7325 ctx->xin->packet != NULL, NULL);
7329 mirrors = in_bundle->src_mirrors;
7331 /* Drop frames on bundles reserved for mirroring. */
7332 if (in_bundle->mirror_out) {
7333 if (ctx->xin->packet != NULL) {
7334 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7335 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
7336 "%s, which is reserved exclusively for mirroring",
7337 ctx->ofproto->up.name, in_bundle->name);
7343 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
7344 if (!input_vid_is_valid(vid, in_bundle, ctx->xin->packet != NULL)) {
7347 vlan = input_vid_to_vlan(in_bundle, vid);
7349 /* Look at the output ports to check for destination selections. */
7351 NL_ATTR_FOR_EACH (a, left, ctx->xout->odp_actions.data,
7352 ctx->xout->odp_actions.size) {
7353 enum ovs_action_attr type = nl_attr_type(a);
7354 struct ofport_dpif *ofport;
7356 if (type != OVS_ACTION_ATTR_OUTPUT) {
7360 ofport = get_odp_port(ofproto, nl_attr_get_u32(a));
7361 if (ofport && ofport->bundle) {
7362 mirrors |= ofport->bundle->dst_mirrors;
7370 /* Restore the original packet before adding the mirror actions. */
7371 ctx->xin->flow = *orig_flow;
7376 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
7378 if (!vlan_is_mirrored(m, vlan)) {
7379 mirrors = zero_rightmost_1bit(mirrors);
7383 mirrors &= ~m->dup_mirrors;
7384 ctx->xout->mirrors |= m->dup_mirrors;
7386 output_normal(ctx, m->out, vlan);
7387 } else if (vlan != m->out_vlan
7388 && !eth_addr_is_reserved(orig_flow->dl_dst)) {
7389 struct ofbundle *bundle;
7391 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
7392 if (ofbundle_includes_vlan(bundle, m->out_vlan)
7393 && !bundle->mirror_out) {
7394 output_normal(ctx, bundle, m->out_vlan);
7402 update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors,
7403 uint64_t packets, uint64_t bytes)
7409 for (; mirrors; mirrors = zero_rightmost_1bit(mirrors)) {
7412 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
7415 /* In normal circumstances 'm' will not be NULL. However,
7416 * if mirrors are reconfigured, we can temporarily get out
7417 * of sync in facet_revalidate(). We could "correct" the
7418 * mirror list before reaching here, but doing that would
7419 * not properly account the traffic stats we've currently
7420 * accumulated for previous mirror configuration. */
7424 m->packet_count += packets;
7425 m->byte_count += bytes;
7429 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
7430 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
7431 * indicate this; newer upstream kernels use gratuitous ARP requests. */
7433 is_gratuitous_arp(const struct flow *flow)
7435 return (flow->dl_type == htons(ETH_TYPE_ARP)
7436 && eth_addr_is_broadcast(flow->dl_dst)
7437 && (flow->nw_proto == ARP_OP_REPLY
7438 || (flow->nw_proto == ARP_OP_REQUEST
7439 && flow->nw_src == flow->nw_dst)));
7443 update_learning_table(struct ofproto_dpif *ofproto,
7444 const struct flow *flow, int vlan,
7445 struct ofbundle *in_bundle)
7447 struct mac_entry *mac;
7449 /* Don't learn the OFPP_NONE port. */
7450 if (in_bundle == &ofpp_none_bundle) {
7454 if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
7458 mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan);
7459 if (is_gratuitous_arp(flow)) {
7460 /* We don't want to learn from gratuitous ARP packets that are
7461 * reflected back over bond slaves so we lock the learning table. */
7462 if (!in_bundle->bond) {
7463 mac_entry_set_grat_arp_lock(mac);
7464 } else if (mac_entry_is_grat_arp_locked(mac)) {
7469 if (mac_entry_is_new(mac) || mac->port.p != in_bundle) {
7470 /* The log messages here could actually be useful in debugging,
7471 * so keep the rate limit relatively high. */
7472 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
7473 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
7474 "on port %s in VLAN %d",
7475 ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src),
7476 in_bundle->name, vlan);
7478 mac->port.p = in_bundle;
7479 tag_set_add(&ofproto->backer->revalidate_set,
7480 mac_learning_changed(ofproto->ml, mac));
7484 static struct ofbundle *
7485 lookup_input_bundle(const struct ofproto_dpif *ofproto, uint16_t in_port,
7486 bool warn, struct ofport_dpif **in_ofportp)
7488 struct ofport_dpif *ofport;
7490 /* Find the port and bundle for the received packet. */
7491 ofport = get_ofp_port(ofproto, in_port);
7493 *in_ofportp = ofport;
7495 if (ofport && ofport->bundle) {
7496 return ofport->bundle;
7499 /* Special-case OFPP_NONE, which a controller may use as the ingress
7500 * port for traffic that it is sourcing. */
7501 if (in_port == OFPP_NONE) {
7502 return &ofpp_none_bundle;
7505 /* Odd. A few possible reasons here:
7507 * - We deleted a port but there are still a few packets queued up
7510 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
7511 * we don't know about.
7513 * - The ofproto client didn't configure the port as part of a bundle.
7514 * This is particularly likely to happen if a packet was received on the
7515 * port after it was created, but before the client had a chance to
7516 * configure its bundle.
7519 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7521 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
7522 "port %"PRIu16, ofproto->up.name, in_port);
7527 /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or
7528 * dropped. Returns true if they may be forwarded, false if they should be
7531 * 'in_port' must be the ofport_dpif that corresponds to flow->in_port.
7532 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
7534 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
7535 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
7536 * checked by input_vid_is_valid().
7538 * May also add tags to '*tags', although the current implementation only does
7539 * so in one special case.
7542 is_admissible(struct xlate_ctx *ctx, struct ofport_dpif *in_port,
7545 struct ofproto_dpif *ofproto = ctx->ofproto;
7546 struct flow *flow = &ctx->xin->flow;
7547 struct ofbundle *in_bundle = in_port->bundle;
7549 /* Drop frames for reserved multicast addresses
7550 * only if forward_bpdu option is absent. */
7551 if (!ofproto->up.forward_bpdu && eth_addr_is_reserved(flow->dl_dst)) {
7552 xlate_report(ctx, "packet has reserved destination MAC, dropping");
7556 if (in_bundle->bond) {
7557 struct mac_entry *mac;
7559 switch (bond_check_admissibility(in_bundle->bond, in_port,
7560 flow->dl_dst, &ctx->xout->tags)) {
7565 xlate_report(ctx, "bonding refused admissibility, dropping");
7568 case BV_DROP_IF_MOVED:
7569 mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL);
7570 if (mac && mac->port.p != in_bundle &&
7571 (!is_gratuitous_arp(flow)
7572 || mac_entry_is_grat_arp_locked(mac))) {
7573 xlate_report(ctx, "SLB bond thinks this packet looped back, "
7585 xlate_normal(struct xlate_ctx *ctx)
7587 struct ofport_dpif *in_port;
7588 struct ofbundle *in_bundle;
7589 struct mac_entry *mac;
7593 ctx->xout->has_normal = true;
7595 in_bundle = lookup_input_bundle(ctx->ofproto, ctx->xin->flow.in_port,
7596 ctx->xin->packet != NULL, &in_port);
7598 xlate_report(ctx, "no input bundle, dropping");
7602 /* Drop malformed frames. */
7603 if (ctx->xin->flow.dl_type == htons(ETH_TYPE_VLAN) &&
7604 !(ctx->xin->flow.vlan_tci & htons(VLAN_CFI))) {
7605 if (ctx->xin->packet != NULL) {
7606 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7607 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
7608 "VLAN tag received on port %s",
7609 ctx->ofproto->up.name, in_bundle->name);
7611 xlate_report(ctx, "partial VLAN tag, dropping");
7615 /* Drop frames on bundles reserved for mirroring. */
7616 if (in_bundle->mirror_out) {
7617 if (ctx->xin->packet != NULL) {
7618 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7619 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
7620 "%s, which is reserved exclusively for mirroring",
7621 ctx->ofproto->up.name, in_bundle->name);
7623 xlate_report(ctx, "input port is mirror output port, dropping");
7628 vid = vlan_tci_to_vid(ctx->xin->flow.vlan_tci);
7629 if (!input_vid_is_valid(vid, in_bundle, ctx->xin->packet != NULL)) {
7630 xlate_report(ctx, "disallowed VLAN VID for this input port, dropping");
7633 vlan = input_vid_to_vlan(in_bundle, vid);
7635 /* Check other admissibility requirements. */
7636 if (in_port && !is_admissible(ctx, in_port, vlan)) {
7640 /* Learn source MAC. */
7641 if (ctx->xin->may_learn) {
7642 update_learning_table(ctx->ofproto, &ctx->xin->flow, vlan, in_bundle);
7645 /* Determine output bundle. */
7646 mac = mac_learning_lookup(ctx->ofproto->ml, ctx->xin->flow.dl_dst, vlan,
7649 if (mac->port.p != in_bundle) {
7650 xlate_report(ctx, "forwarding to learned port");
7651 output_normal(ctx, mac->port.p, vlan);
7653 xlate_report(ctx, "learned port is input port, dropping");
7656 struct ofbundle *bundle;
7658 xlate_report(ctx, "no learned MAC for destination, flooding");
7659 HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
7660 if (bundle != in_bundle
7661 && ofbundle_includes_vlan(bundle, vlan)
7662 && bundle->floodable
7663 && !bundle->mirror_out) {
7664 output_normal(ctx, bundle, vlan);
7667 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
7671 /* Optimized flow revalidation.
7673 * It's a difficult problem, in general, to tell which facets need to have
7674 * their actions recalculated whenever the OpenFlow flow table changes. We
7675 * don't try to solve that general problem: for most kinds of OpenFlow flow
7676 * table changes, we recalculate the actions for every facet. This is
7677 * relatively expensive, but it's good enough if the OpenFlow flow table
7678 * doesn't change very often.
7680 * However, we can expect one particular kind of OpenFlow flow table change to
7681 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
7682 * of CPU on revalidating every facet whenever MAC learning modifies the flow
7683 * table, we add a special case that applies to flow tables in which every rule
7684 * has the same form (that is, the same wildcards), except that the table is
7685 * also allowed to have a single "catch-all" flow that matches all packets. We
7686 * optimize this case by tagging all of the facets that resubmit into the table
7687 * and invalidating the same tag whenever a flow changes in that table. The
7688 * end result is that we revalidate just the facets that need it (and sometimes
7689 * a few more, but not all of the facets or even all of the facets that
7690 * resubmit to the table modified by MAC learning). */
7692 /* Calculates the tag to use for 'flow' and mask 'mask' when it is inserted
7693 * into an OpenFlow table with the given 'basis'. */
7695 rule_calculate_tag(const struct flow *flow, const struct minimask *mask,
7698 if (minimask_is_catchall(mask)) {
7701 uint32_t hash = flow_hash_in_minimask(flow, mask, secret);
7702 return tag_create_deterministic(hash);
7706 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
7707 * taggability of that table.
7709 * This function must be called after *each* change to a flow table. If you
7710 * skip calling it on some changes then the pointer comparisons at the end can
7711 * be invalid if you get unlucky. For example, if a flow removal causes a
7712 * cls_table to be destroyed and then a flow insertion causes a cls_table with
7713 * different wildcards to be created with the same address, then this function
7714 * will incorrectly skip revalidation. */
7716 table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
7718 struct table_dpif *table = &ofproto->tables[table_id];
7719 const struct oftable *oftable = &ofproto->up.tables[table_id];
7720 struct cls_table *catchall, *other;
7721 struct cls_table *t;
7723 catchall = other = NULL;
7725 switch (hmap_count(&oftable->cls.tables)) {
7727 /* We could tag this OpenFlow table but it would make the logic a
7728 * little harder and it's a corner case that doesn't seem worth it
7734 HMAP_FOR_EACH (t, hmap_node, &oftable->cls.tables) {
7735 if (cls_table_is_catchall(t)) {
7737 } else if (!other) {
7740 /* Indicate that we can't tag this by setting both tables to
7741 * NULL. (We know that 'catchall' is already NULL.) */
7748 /* Can't tag this table. */
7752 if (table->catchall_table != catchall || table->other_table != other) {
7753 table->catchall_table = catchall;
7754 table->other_table = other;
7755 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
7759 /* Given 'rule' that has changed in some way (either it is a rule being
7760 * inserted, a rule being deleted, or a rule whose actions are being
7761 * modified), marks facets for revalidation to ensure that packets will be
7762 * forwarded correctly according to the new state of the flow table.
7764 * This function must be called after *each* change to a flow table. See
7765 * the comment on table_update_taggable() for more information. */
7767 rule_invalidate(const struct rule_dpif *rule)
7769 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
7771 table_update_taggable(ofproto, rule->up.table_id);
7773 if (!ofproto->backer->need_revalidate) {
7774 struct table_dpif *table = &ofproto->tables[rule->up.table_id];
7776 if (table->other_table && rule->tag) {
7777 tag_set_add(&ofproto->backer->revalidate_set, rule->tag);
7779 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
7785 set_frag_handling(struct ofproto *ofproto_,
7786 enum ofp_config_flags frag_handling)
7788 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7789 if (frag_handling != OFPC_FRAG_REASM) {
7790 ofproto->backer->need_revalidate = REV_RECONFIGURE;
7798 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
7799 const struct flow *flow,
7800 const struct ofpact *ofpacts, size_t ofpacts_len)
7802 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7803 struct initial_vals initial_vals;
7804 struct odputil_keybuf keybuf;
7805 struct dpif_flow_stats stats;
7806 struct xlate_out xout;
7807 struct xlate_in xin;
7811 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
7812 odp_flow_key_from_flow(&key, flow,
7813 ofp_port_to_odp_port(ofproto, flow->in_port));
7815 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
7817 initial_vals.vlan_tci = flow->vlan_tci;
7818 initial_vals.tunnel_ip_tos = 0;
7819 xlate_in_init(&xin, ofproto, flow, &initial_vals, NULL, stats.tcp_flags,
7821 xin.resubmit_stats = &stats;
7822 xin.ofpacts_len = ofpacts_len;
7823 xin.ofpacts = ofpacts;
7825 xlate_actions(&xin, &xout);
7826 dpif_execute(ofproto->backer->dpif, key.data, key.size,
7827 xout.odp_actions.data, xout.odp_actions.size, packet);
7828 xlate_out_uninit(&xout);
7836 set_netflow(struct ofproto *ofproto_,
7837 const struct netflow_options *netflow_options)
7839 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7841 if (netflow_options) {
7842 if (!ofproto->netflow) {
7843 ofproto->netflow = netflow_create();
7845 return netflow_set_options(ofproto->netflow, netflow_options);
7847 netflow_destroy(ofproto->netflow);
7848 ofproto->netflow = NULL;
7854 get_netflow_ids(const struct ofproto *ofproto_,
7855 uint8_t *engine_type, uint8_t *engine_id)
7857 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7859 dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
7863 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
7865 if (!facet_is_controller_flow(facet) &&
7866 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
7867 struct subfacet *subfacet;
7868 struct ofexpired expired;
7870 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
7871 if (subfacet->path == SF_FAST_PATH) {
7872 struct dpif_flow_stats stats;
7874 subfacet_install(subfacet, &facet->xout.odp_actions, &stats);
7875 subfacet_update_stats(subfacet, &stats);
7879 expired.flow = facet->flow;
7880 expired.packet_count = facet->packet_count;
7881 expired.byte_count = facet->byte_count;
7882 expired.used = facet->used;
7883 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
7888 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
7890 struct facet *facet;
7892 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
7893 send_active_timeout(ofproto, facet);
7897 static struct ofproto_dpif *
7898 ofproto_dpif_lookup(const char *name)
7900 struct ofproto_dpif *ofproto;
7902 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
7903 hash_string(name, 0), &all_ofproto_dpifs) {
7904 if (!strcmp(ofproto->up.name, name)) {
7912 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
7913 const char *argv[], void *aux OVS_UNUSED)
7915 struct ofproto_dpif *ofproto;
7918 ofproto = ofproto_dpif_lookup(argv[1]);
7920 unixctl_command_reply_error(conn, "no such bridge");
7923 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
7925 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
7926 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
7930 unixctl_command_reply(conn, "table successfully flushed");
7934 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
7935 const char *argv[], void *aux OVS_UNUSED)
7937 struct ds ds = DS_EMPTY_INITIALIZER;
7938 const struct ofproto_dpif *ofproto;
7939 const struct mac_entry *e;
7941 ofproto = ofproto_dpif_lookup(argv[1]);
7943 unixctl_command_reply_error(conn, "no such bridge");
7947 ds_put_cstr(&ds, " port VLAN MAC Age\n");
7948 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
7949 struct ofbundle *bundle = e->port.p;
7950 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
7951 ofbundle_get_a_port(bundle)->odp_port,
7952 e->vlan, ETH_ADDR_ARGS(e->mac),
7953 mac_entry_age(ofproto->ml, e));
7955 unixctl_command_reply(conn, ds_cstr(&ds));
7960 struct xlate_out xout;
7961 struct xlate_in xin;
7967 trace_format_rule(struct ds *result, uint8_t table_id, int level,
7968 const struct rule_dpif *rule)
7970 ds_put_char_multiple(result, '\t', level);
7972 ds_put_cstr(result, "No match\n");
7976 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
7977 table_id, ntohll(rule->up.flow_cookie));
7978 cls_rule_format(&rule->up.cr, result);
7979 ds_put_char(result, '\n');
7981 ds_put_char_multiple(result, '\t', level);
7982 ds_put_cstr(result, "OpenFlow ");
7983 ofpacts_format(rule->up.ofpacts, rule->up.ofpacts_len, result);
7984 ds_put_char(result, '\n');
7988 trace_format_flow(struct ds *result, int level, const char *title,
7989 struct trace_ctx *trace)
7991 ds_put_char_multiple(result, '\t', level);
7992 ds_put_format(result, "%s: ", title);
7993 if (flow_equal(&trace->xin.flow, &trace->flow)) {
7994 ds_put_cstr(result, "unchanged");
7996 flow_format(result, &trace->xin.flow);
7997 trace->flow = trace->xin.flow;
7999 ds_put_char(result, '\n');
8003 trace_format_regs(struct ds *result, int level, const char *title,
8004 struct trace_ctx *trace)
8008 ds_put_char_multiple(result, '\t', level);
8009 ds_put_format(result, "%s:", title);
8010 for (i = 0; i < FLOW_N_REGS; i++) {
8011 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
8013 ds_put_char(result, '\n');
8017 trace_format_odp(struct ds *result, int level, const char *title,
8018 struct trace_ctx *trace)
8020 struct ofpbuf *odp_actions = &trace->xout.odp_actions;
8022 ds_put_char_multiple(result, '\t', level);
8023 ds_put_format(result, "%s: ", title);
8024 format_odp_actions(result, odp_actions->data, odp_actions->size);
8025 ds_put_char(result, '\n');
8029 trace_resubmit(struct xlate_ctx *ctx, struct rule_dpif *rule)
8031 struct trace_ctx *trace = CONTAINER_OF(ctx->xin, struct trace_ctx, xin);
8032 struct ds *result = trace->result;
8034 ds_put_char(result, '\n');
8035 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
8036 trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace);
8037 trace_format_odp(result, ctx->recurse + 1, "Resubmitted odp", trace);
8038 trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule);
8042 trace_report(struct xlate_ctx *ctx, const char *s)
8044 struct trace_ctx *trace = CONTAINER_OF(ctx->xin, struct trace_ctx, xin);
8045 struct ds *result = trace->result;
8047 ds_put_char_multiple(result, '\t', ctx->recurse);
8048 ds_put_cstr(result, s);
8049 ds_put_char(result, '\n');
8053 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
8054 void *aux OVS_UNUSED)
8056 const char *dpname = argv[1];
8057 struct ofproto_dpif *ofproto;
8058 struct ofpbuf odp_key;
8059 struct ofpbuf *packet;
8060 struct initial_vals initial_vals;
8066 ofpbuf_init(&odp_key, 0);
8069 ofproto = ofproto_dpif_lookup(dpname);
8071 unixctl_command_reply_error(conn, "Unknown ofproto (use ofproto/list "
8075 if (argc == 3 || (argc == 4 && !strcmp(argv[3], "-generate"))) {
8076 /* ofproto/trace dpname flow [-generate] */
8077 const char *flow_s = argv[2];
8078 const char *generate_s = argv[3];
8080 /* Allow 'flow_s' to be either a datapath flow or an OpenFlow-like
8081 * flow. We guess which type it is based on whether 'flow_s' contains
8082 * an '(', since a datapath flow always contains '(') but an
8083 * OpenFlow-like flow should not (in fact it's allowed but I believe
8084 * that's not documented anywhere).
8086 * An alternative would be to try to parse 'flow_s' both ways, but then
8087 * it would be tricky giving a sensible error message. After all, do
8088 * you just say "syntax error" or do you present both error messages?
8089 * Both choices seem lousy. */
8090 if (strchr(flow_s, '(')) {
8093 /* Convert string to datapath key. */
8094 ofpbuf_init(&odp_key, 0);
8095 error = odp_flow_key_from_string(flow_s, NULL, &odp_key);
8097 unixctl_command_reply_error(conn, "Bad flow syntax");
8101 /* The user might have specified the wrong ofproto but within the
8102 * same backer. That's OK, ofproto_receive() can find the right
8104 if (ofproto_receive(ofproto->backer, NULL, odp_key.data,
8105 odp_key.size, &flow, NULL, &ofproto, NULL,
8107 unixctl_command_reply_error(conn, "Invalid flow");
8110 ds_put_format(&result, "Bridge: %s\n", ofproto->up.name);
8114 error_s = parse_ofp_exact_flow(&flow, argv[2]);
8116 unixctl_command_reply_error(conn, error_s);
8121 initial_vals.vlan_tci = flow.vlan_tci;
8122 initial_vals.tunnel_ip_tos = flow.tunnel.ip_tos;
8125 /* Generate a packet, if requested. */
8127 packet = ofpbuf_new(0);
8128 flow_compose(packet, &flow);
8130 } else if (argc == 7) {
8131 /* ofproto/trace dpname priority tun_id in_port mark packet */
8132 const char *priority_s = argv[2];
8133 const char *tun_id_s = argv[3];
8134 const char *in_port_s = argv[4];
8135 const char *mark_s = argv[5];
8136 const char *packet_s = argv[6];
8137 uint32_t in_port = atoi(in_port_s);
8138 ovs_be64 tun_id = htonll(strtoull(tun_id_s, NULL, 0));
8139 uint32_t priority = atoi(priority_s);
8140 uint32_t mark = atoi(mark_s);
8143 msg = eth_from_hex(packet_s, &packet);
8145 unixctl_command_reply_error(conn, msg);
8149 ds_put_cstr(&result, "Packet: ");
8150 s = ofp_packet_to_string(packet->data, packet->size);
8151 ds_put_cstr(&result, s);
8154 flow_extract(packet, priority, mark, NULL, in_port, &flow);
8155 flow.tunnel.tun_id = tun_id;
8156 initial_vals.vlan_tci = flow.vlan_tci;
8157 initial_vals.tunnel_ip_tos = flow.tunnel.ip_tos;
8159 unixctl_command_reply_error(conn, "Bad command syntax");
8163 ofproto_trace(ofproto, &flow, packet, &initial_vals, &result);
8164 unixctl_command_reply(conn, ds_cstr(&result));
8167 ds_destroy(&result);
8168 ofpbuf_delete(packet);
8169 ofpbuf_uninit(&odp_key);
8173 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
8174 const struct ofpbuf *packet,
8175 const struct initial_vals *initial_vals, struct ds *ds)
8177 struct rule_dpif *rule;
8179 ds_put_cstr(ds, "Flow: ");
8180 flow_format(ds, flow);
8181 ds_put_char(ds, '\n');
8183 rule = rule_dpif_lookup(ofproto, flow);
8185 trace_format_rule(ds, 0, 0, rule);
8186 if (rule == ofproto->miss_rule) {
8187 ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
8188 } else if (rule == ofproto->no_packet_in_rule) {
8189 ds_put_cstr(ds, "\nNo match, packets dropped because "
8190 "OFPPC_NO_PACKET_IN is set on in_port.\n");
8194 uint64_t odp_actions_stub[1024 / 8];
8195 struct ofpbuf odp_actions;
8197 struct trace_ctx trace;
8200 tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
8203 ofpbuf_use_stub(&odp_actions,
8204 odp_actions_stub, sizeof odp_actions_stub);
8205 xlate_in_init(&trace.xin, ofproto, flow, initial_vals, rule, tcp_flags,
8207 trace.xin.resubmit_hook = trace_resubmit;
8208 trace.xin.report_hook = trace_report;
8209 xlate_actions(&trace.xin, &trace.xout);
8211 ds_put_char(ds, '\n');
8212 trace_format_flow(ds, 0, "Final flow", &trace);
8213 ds_put_cstr(ds, "Datapath actions: ");
8214 format_odp_actions(ds, trace.xout.odp_actions.data,
8215 trace.xout.odp_actions.size);
8217 if (trace.xout.slow) {
8218 ds_put_cstr(ds, "\nThis flow is handled by the userspace "
8219 "slow path because it:");
8220 switch (trace.xout.slow) {
8222 ds_put_cstr(ds, "\n\t- Consists of CFM packets.");
8225 ds_put_cstr(ds, "\n\t- Consists of LACP packets.");
8228 ds_put_cstr(ds, "\n\t- Consists of STP packets.");
8230 case SLOW_CONTROLLER:
8231 ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages "
8232 "to the OpenFlow controller.");
8239 xlate_out_uninit(&trace.xout);
8244 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
8245 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
8248 unixctl_command_reply(conn, NULL);
8252 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
8253 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
8256 unixctl_command_reply(conn, NULL);
8259 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
8260 * 'reply' describing the results. */
8262 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
8264 struct facet *facet;
8268 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
8269 if (!facet_check_consistency(facet)) {
8274 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
8278 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
8279 ofproto->up.name, errors);
8281 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
8286 ofproto_dpif_self_check(struct unixctl_conn *conn,
8287 int argc, const char *argv[], void *aux OVS_UNUSED)
8289 struct ds reply = DS_EMPTY_INITIALIZER;
8290 struct ofproto_dpif *ofproto;
8293 ofproto = ofproto_dpif_lookup(argv[1]);
8295 unixctl_command_reply_error(conn, "Unknown ofproto (use "
8296 "ofproto/list for help)");
8299 ofproto_dpif_self_check__(ofproto, &reply);
8301 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
8302 ofproto_dpif_self_check__(ofproto, &reply);
8306 unixctl_command_reply(conn, ds_cstr(&reply));
8310 /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list
8311 * of the 'ofproto_shash' nodes. It is the responsibility of the caller
8312 * to destroy 'ofproto_shash' and free the returned value. */
8313 static const struct shash_node **
8314 get_ofprotos(struct shash *ofproto_shash)
8316 const struct ofproto_dpif *ofproto;
8318 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
8319 char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
8320 shash_add_nocopy(ofproto_shash, name, ofproto);
8323 return shash_sort(ofproto_shash);
8327 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
8328 const char *argv[] OVS_UNUSED,
8329 void *aux OVS_UNUSED)
8331 struct ds ds = DS_EMPTY_INITIALIZER;
8332 struct shash ofproto_shash;
8333 const struct shash_node **sorted_ofprotos;
8336 shash_init(&ofproto_shash);
8337 sorted_ofprotos = get_ofprotos(&ofproto_shash);
8338 for (i = 0; i < shash_count(&ofproto_shash); i++) {
8339 const struct shash_node *node = sorted_ofprotos[i];
8340 ds_put_format(&ds, "%s\n", node->name);
8343 shash_destroy(&ofproto_shash);
8344 free(sorted_ofprotos);
8346 unixctl_command_reply(conn, ds_cstr(&ds));
8351 show_dp_format(const struct ofproto_dpif *ofproto, struct ds *ds)
8353 const struct shash_node **ports;
8355 struct avg_subfacet_rates lifetime;
8356 unsigned long long int minutes;
8357 const int min_ms = 60 * 1000; /* milliseconds in one minute. */
8359 minutes = (time_msec() - ofproto->created) / min_ms;
8362 lifetime.add_rate = (double)ofproto->total_subfacet_add_count
8364 lifetime.del_rate = (double)ofproto->total_subfacet_del_count
8367 lifetime.add_rate = 0.0;
8368 lifetime.del_rate = 0.0;
8371 ds_put_format(ds, "%s (%s):\n", ofproto->up.name,
8372 dpif_name(ofproto->backer->dpif));
8374 "\tlookups: hit:%"PRIu64" missed:%"PRIu64"\n",
8375 ofproto->n_hit, ofproto->n_missed);
8376 ds_put_format(ds, "\tflows: cur: %zu, avg: %5.3f, max: %d,"
8377 " life span: %llu(ms)\n",
8378 hmap_count(&ofproto->subfacets),
8379 avg_subfacet_count(ofproto),
8380 ofproto->max_n_subfacet,
8381 avg_subfacet_life_span(ofproto));
8382 if (minutes >= 60) {
8383 show_dp_rates(ds, "\t\thourly avg:", &ofproto->hourly);
8385 if (minutes >= 60 * 24) {
8386 show_dp_rates(ds, "\t\tdaily avg:", &ofproto->daily);
8388 show_dp_rates(ds, "\t\toverall avg:", &lifetime);
8390 ports = shash_sort(&ofproto->up.port_by_name);
8391 for (i = 0; i < shash_count(&ofproto->up.port_by_name); i++) {
8392 const struct shash_node *node = ports[i];
8393 struct ofport *ofport = node->data;
8394 const char *name = netdev_get_name(ofport->netdev);
8395 const char *type = netdev_get_type(ofport->netdev);
8398 ds_put_format(ds, "\t%s %u/", name, ofport->ofp_port);
8400 odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
8401 if (odp_port != OVSP_NONE) {
8402 ds_put_format(ds, "%"PRIu32":", odp_port);
8404 ds_put_cstr(ds, "none:");
8407 if (strcmp(type, "system")) {
8408 struct netdev *netdev;
8411 ds_put_format(ds, " (%s", type);
8413 error = netdev_open(name, type, &netdev);
8418 error = netdev_get_config(netdev, &config);
8420 const struct smap_node **nodes;
8423 nodes = smap_sort(&config);
8424 for (i = 0; i < smap_count(&config); i++) {
8425 const struct smap_node *node = nodes[i];
8426 ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
8427 node->key, node->value);
8431 smap_destroy(&config);
8433 netdev_close(netdev);
8435 ds_put_char(ds, ')');
8437 ds_put_char(ds, '\n');
8443 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc,
8444 const char *argv[], void *aux OVS_UNUSED)
8446 struct ds ds = DS_EMPTY_INITIALIZER;
8447 const struct ofproto_dpif *ofproto;
8451 for (i = 1; i < argc; i++) {
8452 ofproto = ofproto_dpif_lookup(argv[i]);
8454 ds_put_format(&ds, "Unknown bridge %s (use dpif/dump-dps "
8455 "for help)", argv[i]);
8456 unixctl_command_reply_error(conn, ds_cstr(&ds));
8459 show_dp_format(ofproto, &ds);
8462 struct shash ofproto_shash;
8463 const struct shash_node **sorted_ofprotos;
8466 shash_init(&ofproto_shash);
8467 sorted_ofprotos = get_ofprotos(&ofproto_shash);
8468 for (i = 0; i < shash_count(&ofproto_shash); i++) {
8469 const struct shash_node *node = sorted_ofprotos[i];
8470 show_dp_format(node->data, &ds);
8473 shash_destroy(&ofproto_shash);
8474 free(sorted_ofprotos);
8477 unixctl_command_reply(conn, ds_cstr(&ds));
8482 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
8483 int argc OVS_UNUSED, const char *argv[],
8484 void *aux OVS_UNUSED)
8486 struct ds ds = DS_EMPTY_INITIALIZER;
8487 const struct ofproto_dpif *ofproto;
8488 struct subfacet *subfacet;
8490 ofproto = ofproto_dpif_lookup(argv[1]);
8492 unixctl_command_reply_error(conn, "no such bridge");
8496 update_stats(ofproto->backer);
8498 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
8499 struct facet *facet = subfacet->facet;
8501 odp_flow_key_format(subfacet->key, subfacet->key_len, &ds);
8503 ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:",
8504 subfacet->dp_packet_count, subfacet->dp_byte_count);
8505 if (subfacet->used) {
8506 ds_put_format(&ds, "%.3fs",
8507 (time_msec() - subfacet->used) / 1000.0);
8509 ds_put_format(&ds, "never");
8511 if (subfacet->facet->tcp_flags) {
8512 ds_put_cstr(&ds, ", flags:");
8513 packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags);
8516 ds_put_cstr(&ds, ", actions:");
8517 if (facet->xout.slow) {
8518 uint64_t slow_path_stub[128 / 8];
8519 const struct nlattr *actions;
8522 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
8523 slow_path_stub, sizeof slow_path_stub,
8524 &actions, &actions_len);
8525 format_odp_actions(&ds, actions, actions_len);
8527 format_odp_actions(&ds, facet->xout.odp_actions.data,
8528 facet->xout.odp_actions.size);
8530 ds_put_char(&ds, '\n');
8533 unixctl_command_reply(conn, ds_cstr(&ds));
8538 ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn,
8539 int argc OVS_UNUSED, const char *argv[],
8540 void *aux OVS_UNUSED)
8542 struct ds ds = DS_EMPTY_INITIALIZER;
8543 struct ofproto_dpif *ofproto;
8545 ofproto = ofproto_dpif_lookup(argv[1]);
8547 unixctl_command_reply_error(conn, "no such bridge");
8551 flush(&ofproto->up);
8553 unixctl_command_reply(conn, ds_cstr(&ds));
8558 ofproto_dpif_unixctl_init(void)
8560 static bool registered;
8566 unixctl_command_register(
8568 "bridge {priority tun_id in_port mark packet | odp_flow [-generate]}",
8569 2, 6, ofproto_unixctl_trace, NULL);
8570 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
8571 ofproto_unixctl_fdb_flush, NULL);
8572 unixctl_command_register("fdb/show", "bridge", 1, 1,
8573 ofproto_unixctl_fdb_show, NULL);
8574 unixctl_command_register("ofproto/clog", "", 0, 0,
8575 ofproto_dpif_clog, NULL);
8576 unixctl_command_register("ofproto/unclog", "", 0, 0,
8577 ofproto_dpif_unclog, NULL);
8578 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
8579 ofproto_dpif_self_check, NULL);
8580 unixctl_command_register("dpif/dump-dps", "", 0, 0,
8581 ofproto_unixctl_dpif_dump_dps, NULL);
8582 unixctl_command_register("dpif/show", "[bridge]", 0, INT_MAX,
8583 ofproto_unixctl_dpif_show, NULL);
8584 unixctl_command_register("dpif/dump-flows", "bridge", 1, 1,
8585 ofproto_unixctl_dpif_dump_flows, NULL);
8586 unixctl_command_register("dpif/del-flows", "bridge", 1, 1,
8587 ofproto_unixctl_dpif_del_flows, NULL);
8590 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
8592 * This is deprecated. It is only for compatibility with broken device drivers
8593 * in old versions of Linux that do not properly support VLANs when VLAN
8594 * devices are not used. When broken device drivers are no longer in
8595 * widespread use, we will delete these interfaces. */
8598 set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid)
8600 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
8601 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
8603 if (realdev_ofp_port == ofport->realdev_ofp_port
8604 && vid == ofport->vlandev_vid) {
8608 ofproto->backer->need_revalidate = REV_RECONFIGURE;
8610 if (ofport->realdev_ofp_port) {
8613 if (realdev_ofp_port && ofport->bundle) {
8614 /* vlandevs are enslaved to their realdevs, so they are not allowed to
8615 * themselves be part of a bundle. */
8616 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
8619 ofport->realdev_ofp_port = realdev_ofp_port;
8620 ofport->vlandev_vid = vid;
8622 if (realdev_ofp_port) {
8623 vsp_add(ofport, realdev_ofp_port, vid);
8630 hash_realdev_vid(uint16_t realdev_ofp_port, int vid)
8632 return hash_2words(realdev_ofp_port, vid);
8635 /* Returns the OFP port number of the Linux VLAN device that corresponds to
8636 * 'vlan_tci' on the network device with port number 'realdev_ofp_port' in
8637 * 'struct ofport_dpif'. For example, given 'realdev_ofp_port' of eth0 and
8638 * 'vlan_tci' 9, it would return the port number of eth0.9.
8640 * Unless VLAN splinters are enabled for port 'realdev_ofp_port', this
8641 * function just returns its 'realdev_ofp_port' argument. */
8643 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
8644 uint16_t realdev_ofp_port, ovs_be16 vlan_tci)
8646 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
8647 int vid = vlan_tci_to_vid(vlan_tci);
8648 const struct vlan_splinter *vsp;
8650 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
8651 hash_realdev_vid(realdev_ofp_port, vid),
8652 &ofproto->realdev_vid_map) {
8653 if (vsp->realdev_ofp_port == realdev_ofp_port
8654 && vsp->vid == vid) {
8655 return vsp->vlandev_ofp_port;
8659 return realdev_ofp_port;
8662 static struct vlan_splinter *
8663 vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port)
8665 struct vlan_splinter *vsp;
8667 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, hash_int(vlandev_ofp_port, 0),
8668 &ofproto->vlandev_map) {
8669 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
8677 /* Returns the OpenFlow port number of the "real" device underlying the Linux
8678 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
8679 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
8680 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
8681 * eth0 and store 9 in '*vid'.
8683 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
8684 * VLAN device. Unless VLAN splinters are enabled, this is what this function
8687 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
8688 uint16_t vlandev_ofp_port, int *vid)
8690 if (!hmap_is_empty(&ofproto->vlandev_map)) {
8691 const struct vlan_splinter *vsp;
8693 vsp = vlandev_find(ofproto, vlandev_ofp_port);
8698 return vsp->realdev_ofp_port;
8704 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
8705 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
8706 * 'flow->in_port' to the "real" device backing the VLAN device, sets
8707 * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is
8708 * always the case unless VLAN splinters are enabled), returns false without
8709 * making any changes. */
8711 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
8716 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port, &vid);
8721 /* Cause the flow to be processed as if it came in on the real device with
8722 * the VLAN device's VLAN ID. */
8723 flow->in_port = realdev;
8724 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
8729 vsp_remove(struct ofport_dpif *port)
8731 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
8732 struct vlan_splinter *vsp;
8734 vsp = vlandev_find(ofproto, port->up.ofp_port);
8736 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
8737 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
8740 port->realdev_ofp_port = 0;
8742 VLOG_ERR("missing vlan device record");
8747 vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid)
8749 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
8751 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
8752 && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid))
8753 == realdev_ofp_port)) {
8754 struct vlan_splinter *vsp;
8756 vsp = xmalloc(sizeof *vsp);
8757 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
8758 hash_int(port->up.ofp_port, 0));
8759 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
8760 hash_realdev_vid(realdev_ofp_port, vid));
8761 vsp->realdev_ofp_port = realdev_ofp_port;
8762 vsp->vlandev_ofp_port = port->up.ofp_port;
8765 port->realdev_ofp_port = realdev_ofp_port;
8767 VLOG_ERR("duplicate vlan device record");
8772 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
8774 const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
8775 return ofport ? ofport->odp_port : OVSP_NONE;
8778 static struct ofport_dpif *
8779 odp_port_to_ofport(const struct dpif_backer *backer, uint32_t odp_port)
8781 struct ofport_dpif *port;
8783 HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node,
8784 hash_int(odp_port, 0),
8785 &backer->odp_to_ofport_map) {
8786 if (port->odp_port == odp_port) {
8795 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
8797 struct ofport_dpif *port;
8799 port = odp_port_to_ofport(ofproto->backer, odp_port);
8800 if (port && &ofproto->up == port->up.ofproto) {
8801 return port->up.ofp_port;
8806 static unsigned long long int
8807 avg_subfacet_life_span(const struct ofproto_dpif *ofproto)
8809 unsigned long long int dc;
8810 unsigned long long int avg;
8812 dc = ofproto->total_subfacet_del_count + ofproto->subfacet_del_count;
8813 avg = dc ? ofproto->total_subfacet_life_span / dc : 0;
8819 avg_subfacet_count(const struct ofproto_dpif *ofproto)
8823 if (ofproto->n_update_stats) {
8824 avg_c = (double)ofproto->total_subfacet_count
8825 / ofproto->n_update_stats;
8832 show_dp_rates(struct ds *ds, const char *heading,
8833 const struct avg_subfacet_rates *rates)
8835 ds_put_format(ds, "%s add rate: %5.3f/min, del rate: %5.3f/min\n",
8836 heading, rates->add_rate, rates->del_rate);
8840 update_max_subfacet_count(struct ofproto_dpif *ofproto)
8842 ofproto->max_n_subfacet = MAX(ofproto->max_n_subfacet,
8843 hmap_count(&ofproto->subfacets));
8846 /* Compute exponentially weighted moving average, adding 'new' as the newest,
8847 * most heavily weighted element. 'base' designates the rate of decay: after
8848 * 'base' further updates, 'new''s weight in the EWMA decays to about 1/e
8851 exp_mavg(double *avg, int base, double new)
8853 *avg = (*avg * (base - 1) + new) / base;
8857 update_moving_averages(struct ofproto_dpif *ofproto)
8859 const int min_ms = 60 * 1000; /* milliseconds in one minute. */
8861 /* Update hourly averages on the minute boundaries. */
8862 if (time_msec() - ofproto->last_minute >= min_ms) {
8863 exp_mavg(&ofproto->hourly.add_rate, 60, ofproto->subfacet_add_count);
8864 exp_mavg(&ofproto->hourly.del_rate, 60, ofproto->subfacet_del_count);
8866 /* Update daily averages on the hour boundaries. */
8867 if ((ofproto->last_minute - ofproto->created) / min_ms % 60 == 59) {
8868 exp_mavg(&ofproto->daily.add_rate, 24, ofproto->hourly.add_rate);
8869 exp_mavg(&ofproto->daily.del_rate, 24, ofproto->hourly.del_rate);
8872 ofproto->total_subfacet_add_count += ofproto->subfacet_add_count;
8873 ofproto->total_subfacet_del_count += ofproto->subfacet_del_count;
8874 ofproto->subfacet_add_count = 0;
8875 ofproto->subfacet_del_count = 0;
8876 ofproto->last_minute += min_ms;
8881 dpif_stats_update_hit_count(struct ofproto_dpif *ofproto, uint64_t delta)
8883 ofproto->n_hit += delta;
8886 const struct ofproto_class ofproto_dpif_class = {
8921 port_is_lacp_current,
8922 NULL, /* rule_choose_table */
8929 rule_modify_actions,
8941 get_stp_port_status,
8948 is_mirror_output_bundle,
8949 forward_bpdu_changed,
8950 set_mac_table_config,