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. */
657 /* All existing ofproto_backer instances, indexed by ofproto->up.type. */
658 static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers);
660 static void drop_key_clear(struct dpif_backer *);
661 static struct ofport_dpif *
662 odp_port_to_ofport(const struct dpif_backer *, uint32_t odp_port);
664 static void dpif_stats_update_hit_count(struct ofproto_dpif *ofproto,
666 struct avg_subfacet_rates {
667 double add_rate; /* Moving average of new flows created per minute. */
668 double del_rate; /* Moving average of flows deleted per minute. */
670 static void show_dp_rates(struct ds *ds, const char *heading,
671 const struct avg_subfacet_rates *rates);
672 static void exp_mavg(double *avg, int base, double new);
674 struct ofproto_dpif {
675 struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
677 struct dpif_backer *backer;
679 /* Special OpenFlow rules. */
680 struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */
681 struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */
687 struct netflow *netflow;
688 struct dpif_sflow *sflow;
689 struct dpif_ipfix *ipfix;
690 struct hmap bundles; /* Contains "struct ofbundle"s. */
691 struct mac_learning *ml;
692 struct ofmirror *mirrors[MAX_MIRRORS];
694 bool has_bonded_bundles;
698 struct hmap subfacets;
699 struct governor *governor;
700 long long int consistency_rl;
703 struct table_dpif tables[N_TABLES];
705 /* Support for debugging async flow mods. */
706 struct list completions;
708 bool has_bundle_action; /* True when the first bundle action appears. */
709 struct netdev_stats stats; /* To account packets generated and consumed in
714 long long int stp_last_tick;
716 /* VLAN splinters. */
717 struct hmap realdev_vid_map; /* (realdev,vid) -> vlandev. */
718 struct hmap vlandev_map; /* vlandev -> (realdev,vid). */
721 struct sset ports; /* Set of standard port names. */
722 struct sset ghost_ports; /* Ports with no datapath port. */
723 struct sset port_poll_set; /* Queued names for port_poll() reply. */
724 int port_poll_errno; /* Last errno for port_poll() reply. */
726 /* Per ofproto's dpif stats. */
730 /* Subfacet statistics.
732 * These keep track of the total number of subfacets added and deleted and
733 * flow life span. They are useful for computing the flow rates stats
734 * exposed via "ovs-appctl dpif/show". The goal is to learn about
735 * traffic patterns in ways that we can use later to improve Open vSwitch
736 * performance in new situations. */
737 long long int created; /* Time when it is created. */
738 unsigned int max_n_subfacet; /* Maximum number of flows */
740 /* The average number of subfacets... */
741 struct avg_subfacet_rates hourly; /* ...over the last hour. */
742 struct avg_subfacet_rates daily; /* ...over the last day. */
743 long long int last_minute; /* Last time 'hourly' was updated. */
745 /* Number of subfacets added or deleted since 'last_minute'. */
746 unsigned int subfacet_add_count;
747 unsigned int subfacet_del_count;
749 /* Number of subfacets added or deleted from 'created' to 'last_minute.' */
750 unsigned long long int total_subfacet_add_count;
751 unsigned long long int total_subfacet_del_count;
753 /* Sum of the number of milliseconds that each subfacet existed,
754 * over the subfacets that have been added and then later deleted. */
755 unsigned long long int total_subfacet_life_span;
757 /* Incremented by the number of currently existing subfacets, each
758 * time we pull statistics from the kernel. */
759 unsigned long long int total_subfacet_count;
761 /* Number of times we pull statistics from the kernel. */
762 unsigned long long int n_update_stats;
764 static unsigned long long int avg_subfacet_life_span(
765 const struct ofproto_dpif *);
766 static double avg_subfacet_count(const struct ofproto_dpif *ofproto);
767 static void update_moving_averages(struct ofproto_dpif *ofproto);
768 static void dpif_stats_update_hit_count(struct ofproto_dpif *ofproto,
770 static void update_max_subfacet_count(struct ofproto_dpif *ofproto);
772 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
773 * for debugging the asynchronous flow_mod implementation.) */
776 /* All existing ofproto_dpif instances, indexed by ->up.name. */
777 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
779 static void ofproto_dpif_unixctl_init(void);
781 static struct ofproto_dpif *
782 ofproto_dpif_cast(const struct ofproto *ofproto)
784 ovs_assert(ofproto->ofproto_class == &ofproto_dpif_class);
785 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
788 static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *,
790 static struct ofport_dpif *get_odp_port(const struct ofproto_dpif *,
792 static void ofproto_trace(struct ofproto_dpif *, const struct flow *,
793 const struct ofpbuf *,
794 const struct initial_vals *, struct ds *);
796 /* Packet processing. */
797 static void update_learning_table(struct ofproto_dpif *,
798 const struct flow *, int vlan,
801 #define FLOW_MISS_MAX_BATCH 50
802 static int handle_upcalls(struct dpif_backer *, unsigned int max_batch);
804 /* Flow expiration. */
805 static int expire(struct dpif_backer *);
808 static void send_netflow_active_timeouts(struct ofproto_dpif *);
811 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
812 static size_t compose_sflow_action(const struct ofproto_dpif *,
813 struct ofpbuf *odp_actions,
814 const struct flow *, uint32_t odp_port);
815 static void compose_ipfix_action(const struct ofproto_dpif *,
816 struct ofpbuf *odp_actions,
817 const struct flow *);
818 static void add_mirror_actions(struct xlate_ctx *ctx,
819 const struct flow *flow);
820 /* Global variables. */
821 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
823 /* Initial mappings of port to bridge mappings. */
824 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
826 /* Factory functions. */
829 init(const struct shash *iface_hints)
831 struct shash_node *node;
833 /* Make a local copy, since we don't own 'iface_hints' elements. */
834 SHASH_FOR_EACH(node, iface_hints) {
835 const struct iface_hint *orig_hint = node->data;
836 struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
838 new_hint->br_name = xstrdup(orig_hint->br_name);
839 new_hint->br_type = xstrdup(orig_hint->br_type);
840 new_hint->ofp_port = orig_hint->ofp_port;
842 shash_add(&init_ofp_ports, node->name, new_hint);
847 enumerate_types(struct sset *types)
849 dp_enumerate_types(types);
853 enumerate_names(const char *type, struct sset *names)
855 struct ofproto_dpif *ofproto;
858 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
859 if (strcmp(type, ofproto->up.type)) {
862 sset_add(names, ofproto->up.name);
869 del(const char *type, const char *name)
874 error = dpif_open(name, type, &dpif);
876 error = dpif_delete(dpif);
883 port_open_type(const char *datapath_type, const char *port_type)
885 return dpif_port_open_type(datapath_type, port_type);
888 /* Type functions. */
890 static struct ofproto_dpif *
891 lookup_ofproto_dpif_by_port_name(const char *name)
893 struct ofproto_dpif *ofproto;
895 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
896 if (sset_contains(&ofproto->ports, name)) {
905 type_run(const char *type)
907 static long long int push_timer = LLONG_MIN;
908 struct dpif_backer *backer;
912 backer = shash_find_data(&all_dpif_backers, type);
914 /* This is not necessarily a problem, since backers are only
915 * created on demand. */
919 dpif_run(backer->dpif);
921 /* The most natural place to push facet statistics is when they're pulled
922 * from the datapath. However, when there are many flows in the datapath,
923 * this expensive operation can occur so frequently, that it reduces our
924 * ability to quickly set up flows. To reduce the cost, we push statistics
926 if (time_msec() > push_timer) {
927 push_timer = time_msec() + 2000;
931 if (backer->need_revalidate
932 || !tag_set_is_empty(&backer->revalidate_set)) {
933 struct tag_set revalidate_set = backer->revalidate_set;
934 bool need_revalidate = backer->need_revalidate;
935 struct ofproto_dpif *ofproto;
936 struct simap_node *node;
937 struct simap tmp_backers;
939 /* Handle tunnel garbage collection. */
940 simap_init(&tmp_backers);
941 simap_swap(&backer->tnl_backers, &tmp_backers);
943 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
944 struct ofport_dpif *iter;
946 if (backer != ofproto->backer) {
950 HMAP_FOR_EACH (iter, up.hmap_node, &ofproto->up.ports) {
953 if (!iter->tnl_port) {
957 dp_port = netdev_vport_get_dpif_port(iter->up.netdev);
958 node = simap_find(&tmp_backers, dp_port);
960 simap_put(&backer->tnl_backers, dp_port, node->data);
961 simap_delete(&tmp_backers, node);
962 node = simap_find(&backer->tnl_backers, dp_port);
964 node = simap_find(&backer->tnl_backers, dp_port);
966 uint32_t odp_port = UINT32_MAX;
968 if (!dpif_port_add(backer->dpif, iter->up.netdev,
970 simap_put(&backer->tnl_backers, dp_port, odp_port);
971 node = simap_find(&backer->tnl_backers, dp_port);
976 iter->odp_port = node ? node->data : OVSP_NONE;
977 if (tnl_port_reconfigure(&iter->up, iter->odp_port,
979 backer->need_revalidate = REV_RECONFIGURE;
984 SIMAP_FOR_EACH (node, &tmp_backers) {
985 dpif_port_del(backer->dpif, node->data);
987 simap_destroy(&tmp_backers);
989 switch (backer->need_revalidate) {
990 case REV_RECONFIGURE: COVERAGE_INC(rev_reconfigure); break;
991 case REV_STP: COVERAGE_INC(rev_stp); break;
992 case REV_PORT_TOGGLED: COVERAGE_INC(rev_port_toggled); break;
993 case REV_FLOW_TABLE: COVERAGE_INC(rev_flow_table); break;
994 case REV_INCONSISTENCY: COVERAGE_INC(rev_inconsistency); break;
997 if (backer->need_revalidate) {
998 /* Clear the drop_keys in case we should now be accepting some
999 * formerly dropped flows. */
1000 drop_key_clear(backer);
1003 /* Clear the revalidation flags. */
1004 tag_set_init(&backer->revalidate_set);
1005 backer->need_revalidate = 0;
1007 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1008 struct facet *facet, *next;
1010 if (ofproto->backer != backer) {
1014 HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &ofproto->facets) {
1016 || tag_set_intersects(&revalidate_set, facet->xout.tags)) {
1017 facet_revalidate(facet);
1024 if (timer_expired(&backer->next_expiration)) {
1025 int delay = expire(backer);
1026 timer_set_duration(&backer->next_expiration, delay);
1029 /* Check for port changes in the dpif. */
1030 while ((error = dpif_port_poll(backer->dpif, &devname)) == 0) {
1031 struct ofproto_dpif *ofproto;
1032 struct dpif_port port;
1034 /* Don't report on the datapath's device. */
1035 if (!strcmp(devname, dpif_base_name(backer->dpif))) {
1039 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
1040 &all_ofproto_dpifs) {
1041 if (simap_contains(&ofproto->backer->tnl_backers, devname)) {
1046 ofproto = lookup_ofproto_dpif_by_port_name(devname);
1047 if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
1048 /* The port was removed. If we know the datapath,
1049 * report it through poll_set(). If we don't, it may be
1050 * notifying us of a removal we initiated, so ignore it.
1051 * If there's a pending ENOBUFS, let it stand, since
1052 * everything will be reevaluated. */
1053 if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
1054 sset_add(&ofproto->port_poll_set, devname);
1055 ofproto->port_poll_errno = 0;
1057 } else if (!ofproto) {
1058 /* The port was added, but we don't know with which
1059 * ofproto we should associate it. Delete it. */
1060 dpif_port_del(backer->dpif, port.port_no);
1062 dpif_port_destroy(&port);
1068 if (error != EAGAIN) {
1069 struct ofproto_dpif *ofproto;
1071 /* There was some sort of error, so propagate it to all
1072 * ofprotos that use this backer. */
1073 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
1074 &all_ofproto_dpifs) {
1075 if (ofproto->backer == backer) {
1076 sset_clear(&ofproto->port_poll_set);
1077 ofproto->port_poll_errno = error;
1086 dpif_backer_run_fast(struct dpif_backer *backer, int max_batch)
1090 /* Handle one or more batches of upcalls, until there's nothing left to do
1091 * or until we do a fixed total amount of work.
1093 * We do work in batches because it can be much cheaper to set up a number
1094 * of flows and fire off their patches all at once. We do multiple batches
1095 * because in some cases handling a packet can cause another packet to be
1096 * queued almost immediately as part of the return flow. Both
1097 * optimizations can make major improvements on some benchmarks and
1098 * presumably for real traffic as well. */
1100 while (work < max_batch) {
1101 int retval = handle_upcalls(backer, max_batch - work);
1112 type_run_fast(const char *type)
1114 struct dpif_backer *backer;
1116 backer = shash_find_data(&all_dpif_backers, type);
1118 /* This is not necessarily a problem, since backers are only
1119 * created on demand. */
1123 return dpif_backer_run_fast(backer, FLOW_MISS_MAX_BATCH);
1129 static long long int port_rl = LLONG_MIN;
1130 static unsigned int backer_rl = 0;
1132 if (time_msec() >= port_rl) {
1133 struct ofproto_dpif *ofproto;
1134 struct ofport_dpif *ofport;
1136 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1138 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1139 port_run_fast(ofport);
1142 port_rl = time_msec() + 200;
1145 /* XXX: We have to be careful not to do too much work in this function. If
1146 * we call dpif_backer_run_fast() too often, or with too large a batch,
1147 * performance improves signifcantly, but at a cost. It's possible for the
1148 * number of flows in the datapath to increase without bound, and for poll
1149 * loops to take 10s of seconds. The correct solution to this problem,
1150 * long term, is to separate flow miss handling into it's own thread so it
1151 * isn't affected by revalidations, and expirations. Until then, this is
1152 * the best we can do. */
1153 if (++backer_rl >= 10) {
1154 struct shash_node *node;
1157 SHASH_FOR_EACH (node, &all_dpif_backers) {
1158 dpif_backer_run_fast(node->data, 1);
1164 type_wait(const char *type)
1166 struct dpif_backer *backer;
1168 backer = shash_find_data(&all_dpif_backers, type);
1170 /* This is not necessarily a problem, since backers are only
1171 * created on demand. */
1175 timer_wait(&backer->next_expiration);
1178 /* Basic life-cycle. */
1180 static int add_internal_flows(struct ofproto_dpif *);
1182 static struct ofproto *
1185 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
1186 return &ofproto->up;
1190 dealloc(struct ofproto *ofproto_)
1192 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1197 close_dpif_backer(struct dpif_backer *backer)
1199 struct shash_node *node;
1201 ovs_assert(backer->refcount > 0);
1203 if (--backer->refcount) {
1207 drop_key_clear(backer);
1208 hmap_destroy(&backer->drop_keys);
1210 simap_destroy(&backer->tnl_backers);
1211 hmap_destroy(&backer->odp_to_ofport_map);
1212 node = shash_find(&all_dpif_backers, backer->type);
1214 shash_delete(&all_dpif_backers, node);
1215 dpif_close(backer->dpif);
1220 /* Datapath port slated for removal from datapath. */
1221 struct odp_garbage {
1222 struct list list_node;
1227 open_dpif_backer(const char *type, struct dpif_backer **backerp)
1229 struct dpif_backer *backer;
1230 struct dpif_port_dump port_dump;
1231 struct dpif_port port;
1232 struct shash_node *node;
1233 struct list garbage_list;
1234 struct odp_garbage *garbage, *next;
1240 backer = shash_find_data(&all_dpif_backers, type);
1247 backer_name = xasprintf("ovs-%s", type);
1249 /* Remove any existing datapaths, since we assume we're the only
1250 * userspace controlling the datapath. */
1252 dp_enumerate_names(type, &names);
1253 SSET_FOR_EACH(name, &names) {
1254 struct dpif *old_dpif;
1256 /* Don't remove our backer if it exists. */
1257 if (!strcmp(name, backer_name)) {
1261 if (dpif_open(name, type, &old_dpif)) {
1262 VLOG_WARN("couldn't open old datapath %s to remove it", name);
1264 dpif_delete(old_dpif);
1265 dpif_close(old_dpif);
1268 sset_destroy(&names);
1270 backer = xmalloc(sizeof *backer);
1272 error = dpif_create_and_open(backer_name, type, &backer->dpif);
1275 VLOG_ERR("failed to open datapath of type %s: %s", type,
1281 backer->type = xstrdup(type);
1282 backer->refcount = 1;
1283 hmap_init(&backer->odp_to_ofport_map);
1284 hmap_init(&backer->drop_keys);
1285 timer_set_duration(&backer->next_expiration, 1000);
1286 backer->need_revalidate = 0;
1287 simap_init(&backer->tnl_backers);
1288 tag_set_init(&backer->revalidate_set);
1291 dpif_flow_flush(backer->dpif);
1293 /* Loop through the ports already on the datapath and remove any
1294 * that we don't need anymore. */
1295 list_init(&garbage_list);
1296 dpif_port_dump_start(&port_dump, backer->dpif);
1297 while (dpif_port_dump_next(&port_dump, &port)) {
1298 node = shash_find(&init_ofp_ports, port.name);
1299 if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
1300 garbage = xmalloc(sizeof *garbage);
1301 garbage->odp_port = port.port_no;
1302 list_push_front(&garbage_list, &garbage->list_node);
1305 dpif_port_dump_done(&port_dump);
1307 LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) {
1308 dpif_port_del(backer->dpif, garbage->odp_port);
1309 list_remove(&garbage->list_node);
1313 shash_add(&all_dpif_backers, type, backer);
1315 error = dpif_recv_set(backer->dpif, true);
1317 VLOG_ERR("failed to listen on datapath of type %s: %s",
1318 type, strerror(error));
1319 close_dpif_backer(backer);
1327 construct(struct ofproto *ofproto_)
1329 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1330 struct shash_node *node, *next;
1335 error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1340 max_ports = dpif_get_max_ports(ofproto->backer->dpif);
1341 ofproto_init_max_ports(ofproto_, MIN(max_ports, OFPP_MAX));
1343 ofproto->n_matches = 0;
1345 ofproto->netflow = NULL;
1346 ofproto->sflow = NULL;
1347 ofproto->ipfix = NULL;
1348 ofproto->stp = NULL;
1349 hmap_init(&ofproto->bundles);
1350 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1351 for (i = 0; i < MAX_MIRRORS; i++) {
1352 ofproto->mirrors[i] = NULL;
1354 ofproto->has_bonded_bundles = false;
1356 hmap_init(&ofproto->facets);
1357 hmap_init(&ofproto->subfacets);
1358 ofproto->governor = NULL;
1359 ofproto->consistency_rl = LLONG_MIN;
1361 for (i = 0; i < N_TABLES; i++) {
1362 struct table_dpif *table = &ofproto->tables[i];
1364 table->catchall_table = NULL;
1365 table->other_table = NULL;
1366 table->basis = random_uint32();
1369 list_init(&ofproto->completions);
1371 ofproto_dpif_unixctl_init();
1373 ofproto->has_mirrors = false;
1374 ofproto->has_bundle_action = false;
1376 hmap_init(&ofproto->vlandev_map);
1377 hmap_init(&ofproto->realdev_vid_map);
1379 sset_init(&ofproto->ports);
1380 sset_init(&ofproto->ghost_ports);
1381 sset_init(&ofproto->port_poll_set);
1382 ofproto->port_poll_errno = 0;
1384 SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1385 struct iface_hint *iface_hint = node->data;
1387 if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1388 /* Check if the datapath already has this port. */
1389 if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1390 sset_add(&ofproto->ports, node->name);
1393 free(iface_hint->br_name);
1394 free(iface_hint->br_type);
1396 shash_delete(&init_ofp_ports, node);
1400 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1401 hash_string(ofproto->up.name, 0));
1402 memset(&ofproto->stats, 0, sizeof ofproto->stats);
1404 ofproto_init_tables(ofproto_, N_TABLES);
1405 error = add_internal_flows(ofproto);
1406 ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1409 ofproto->n_missed = 0;
1411 ofproto->max_n_subfacet = 0;
1412 ofproto->created = time_msec();
1413 ofproto->last_minute = ofproto->created;
1414 memset(&ofproto->hourly, 0, sizeof ofproto->hourly);
1415 memset(&ofproto->daily, 0, sizeof ofproto->daily);
1416 ofproto->subfacet_add_count = 0;
1417 ofproto->subfacet_del_count = 0;
1418 ofproto->total_subfacet_add_count = 0;
1419 ofproto->total_subfacet_del_count = 0;
1420 ofproto->total_subfacet_life_span = 0;
1421 ofproto->total_subfacet_count = 0;
1422 ofproto->n_update_stats = 0;
1428 add_internal_flow(struct ofproto_dpif *ofproto, int id,
1429 const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1431 struct ofputil_flow_mod fm;
1434 match_init_catchall(&fm.match);
1436 match_set_reg(&fm.match, 0, id);
1437 fm.new_cookie = htonll(0);
1438 fm.cookie = htonll(0);
1439 fm.cookie_mask = htonll(0);
1440 fm.table_id = TBL_INTERNAL;
1441 fm.command = OFPFC_ADD;
1442 fm.idle_timeout = 0;
1443 fm.hard_timeout = 0;
1447 fm.ofpacts = ofpacts->data;
1448 fm.ofpacts_len = ofpacts->size;
1450 error = ofproto_flow_mod(&ofproto->up, &fm);
1452 VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
1453 id, ofperr_to_string(error));
1457 *rulep = rule_dpif_lookup__(ofproto, &fm.match.flow, TBL_INTERNAL);
1458 ovs_assert(*rulep != NULL);
1464 add_internal_flows(struct ofproto_dpif *ofproto)
1466 struct ofpact_controller *controller;
1467 uint64_t ofpacts_stub[128 / 8];
1468 struct ofpbuf ofpacts;
1472 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1475 controller = ofpact_put_CONTROLLER(&ofpacts);
1476 controller->max_len = UINT16_MAX;
1477 controller->controller_id = 0;
1478 controller->reason = OFPR_NO_MATCH;
1479 ofpact_pad(&ofpacts);
1481 error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule);
1486 ofpbuf_clear(&ofpacts);
1487 error = add_internal_flow(ofproto, id++, &ofpacts,
1488 &ofproto->no_packet_in_rule);
1493 complete_operations(struct ofproto_dpif *ofproto)
1495 struct dpif_completion *c, *next;
1497 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
1498 ofoperation_complete(c->op, 0);
1499 list_remove(&c->list_node);
1505 destruct(struct ofproto *ofproto_)
1507 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1508 struct rule_dpif *rule, *next_rule;
1509 struct oftable *table;
1512 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1513 complete_operations(ofproto);
1515 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1516 struct cls_cursor cursor;
1518 cls_cursor_init(&cursor, &table->cls, NULL);
1519 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1520 ofproto_rule_destroy(&rule->up);
1524 for (i = 0; i < MAX_MIRRORS; i++) {
1525 mirror_destroy(ofproto->mirrors[i]);
1528 netflow_destroy(ofproto->netflow);
1529 dpif_sflow_destroy(ofproto->sflow);
1530 hmap_destroy(&ofproto->bundles);
1531 mac_learning_destroy(ofproto->ml);
1533 hmap_destroy(&ofproto->facets);
1534 hmap_destroy(&ofproto->subfacets);
1535 governor_destroy(ofproto->governor);
1537 hmap_destroy(&ofproto->vlandev_map);
1538 hmap_destroy(&ofproto->realdev_vid_map);
1540 sset_destroy(&ofproto->ports);
1541 sset_destroy(&ofproto->ghost_ports);
1542 sset_destroy(&ofproto->port_poll_set);
1544 close_dpif_backer(ofproto->backer);
1548 run_fast(struct ofproto *ofproto_)
1550 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1551 struct ofport_dpif *ofport;
1553 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1554 port_run_fast(ofport);
1561 run(struct ofproto *ofproto_)
1563 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1564 struct ofport_dpif *ofport;
1565 struct ofbundle *bundle;
1569 complete_operations(ofproto);
1572 error = run_fast(ofproto_);
1577 if (ofproto->netflow) {
1578 if (netflow_run(ofproto->netflow)) {
1579 send_netflow_active_timeouts(ofproto);
1582 if (ofproto->sflow) {
1583 dpif_sflow_run(ofproto->sflow);
1586 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1589 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1594 mac_learning_run(ofproto->ml, &ofproto->backer->revalidate_set);
1596 /* Check the consistency of a random facet, to aid debugging. */
1597 if (time_msec() >= ofproto->consistency_rl
1598 && !hmap_is_empty(&ofproto->facets)
1599 && !ofproto->backer->need_revalidate) {
1600 struct facet *facet;
1602 ofproto->consistency_rl = time_msec() + 250;
1604 facet = CONTAINER_OF(hmap_random_node(&ofproto->facets),
1605 struct facet, hmap_node);
1606 if (!tag_set_intersects(&ofproto->backer->revalidate_set,
1607 facet->xout.tags)) {
1608 if (!facet_check_consistency(facet)) {
1609 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
1614 if (ofproto->governor) {
1617 governor_run(ofproto->governor);
1619 /* If the governor has shrunk to its minimum size and the number of
1620 * subfacets has dwindled, then drop the governor entirely.
1622 * For hysteresis, the number of subfacets to drop the governor is
1623 * smaller than the number needed to trigger its creation. */
1624 n_subfacets = hmap_count(&ofproto->subfacets);
1625 if (n_subfacets * 4 < ofproto->up.flow_eviction_threshold
1626 && governor_is_idle(ofproto->governor)) {
1627 governor_destroy(ofproto->governor);
1628 ofproto->governor = NULL;
1636 wait(struct ofproto *ofproto_)
1638 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1639 struct ofport_dpif *ofport;
1640 struct ofbundle *bundle;
1642 if (!clogged && !list_is_empty(&ofproto->completions)) {
1643 poll_immediate_wake();
1646 dpif_wait(ofproto->backer->dpif);
1647 dpif_recv_wait(ofproto->backer->dpif);
1648 if (ofproto->sflow) {
1649 dpif_sflow_wait(ofproto->sflow);
1651 if (!tag_set_is_empty(&ofproto->backer->revalidate_set)) {
1652 poll_immediate_wake();
1654 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1657 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1658 bundle_wait(bundle);
1660 if (ofproto->netflow) {
1661 netflow_wait(ofproto->netflow);
1663 mac_learning_wait(ofproto->ml);
1665 if (ofproto->backer->need_revalidate) {
1666 /* Shouldn't happen, but if it does just go around again. */
1667 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1668 poll_immediate_wake();
1670 if (ofproto->governor) {
1671 governor_wait(ofproto->governor);
1676 get_memory_usage(const struct ofproto *ofproto_, struct simap *usage)
1678 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1680 simap_increase(usage, "facets", hmap_count(&ofproto->facets));
1681 simap_increase(usage, "subfacets", hmap_count(&ofproto->subfacets));
1685 flush(struct ofproto *ofproto_)
1687 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1688 struct subfacet *subfacet, *next_subfacet;
1689 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
1693 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
1694 &ofproto->subfacets) {
1695 if (subfacet->path != SF_NOT_INSTALLED) {
1696 batch[n_batch++] = subfacet;
1697 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
1698 subfacet_destroy_batch(ofproto, batch, n_batch);
1702 subfacet_destroy(subfacet);
1707 subfacet_destroy_batch(ofproto, batch, n_batch);
1712 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1713 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1715 *arp_match_ip = true;
1716 *actions = (OFPUTIL_A_OUTPUT |
1717 OFPUTIL_A_SET_VLAN_VID |
1718 OFPUTIL_A_SET_VLAN_PCP |
1719 OFPUTIL_A_STRIP_VLAN |
1720 OFPUTIL_A_SET_DL_SRC |
1721 OFPUTIL_A_SET_DL_DST |
1722 OFPUTIL_A_SET_NW_SRC |
1723 OFPUTIL_A_SET_NW_DST |
1724 OFPUTIL_A_SET_NW_TOS |
1725 OFPUTIL_A_SET_TP_SRC |
1726 OFPUTIL_A_SET_TP_DST |
1731 get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots)
1733 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1734 struct dpif_dp_stats s;
1736 strcpy(ots->name, "classifier");
1738 dpif_get_dp_stats(ofproto->backer->dpif, &s);
1740 ots->lookup_count = htonll(s.n_hit + s.n_missed);
1741 ots->matched_count = htonll(s.n_hit + ofproto->n_matches);
1744 static struct ofport *
1747 struct ofport_dpif *port = xmalloc(sizeof *port);
1752 port_dealloc(struct ofport *port_)
1754 struct ofport_dpif *port = ofport_dpif_cast(port_);
1759 port_construct(struct ofport *port_)
1761 struct ofport_dpif *port = ofport_dpif_cast(port_);
1762 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1763 const struct netdev *netdev = port->up.netdev;
1764 struct dpif_port dpif_port;
1767 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1768 port->bundle = NULL;
1770 port->tag = tag_create_random();
1771 port->may_enable = true;
1772 port->stp_port = NULL;
1773 port->stp_state = STP_DISABLED;
1774 port->tnl_port = NULL;
1775 hmap_init(&port->priorities);
1776 port->realdev_ofp_port = 0;
1777 port->vlandev_vid = 0;
1778 port->carrier_seq = netdev_get_carrier_resets(netdev);
1780 if (netdev_vport_is_patch(netdev)) {
1781 /* By bailing out here, we don't submit the port to the sFlow module
1782 * to be considered for counter polling export. This is correct
1783 * because the patch port represents an interface that sFlow considers
1784 * to be "internal" to the switch as a whole, and therefore not an
1785 * candidate for counter polling. */
1786 port->odp_port = OVSP_NONE;
1790 error = dpif_port_query_by_name(ofproto->backer->dpif,
1791 netdev_vport_get_dpif_port(netdev),
1797 port->odp_port = dpif_port.port_no;
1799 if (netdev_get_tunnel_config(netdev)) {
1800 port->tnl_port = tnl_port_add(&port->up, port->odp_port);
1802 /* Sanity-check that a mapping doesn't already exist. This
1803 * shouldn't happen for non-tunnel ports. */
1804 if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1805 VLOG_ERR("port %s already has an OpenFlow port number",
1807 dpif_port_destroy(&dpif_port);
1811 hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1812 hash_int(port->odp_port, 0));
1814 dpif_port_destroy(&dpif_port);
1816 if (ofproto->sflow) {
1817 dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1824 port_destruct(struct ofport *port_)
1826 struct ofport_dpif *port = ofport_dpif_cast(port_);
1827 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1828 const char *dp_port_name = netdev_vport_get_dpif_port(port->up.netdev);
1829 const char *devname = netdev_get_name(port->up.netdev);
1831 if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1832 /* The underlying device is still there, so delete it. This
1833 * happens when the ofproto is being destroyed, since the caller
1834 * assumes that removal of attached ports will happen as part of
1836 if (!port->tnl_port) {
1837 dpif_port_del(ofproto->backer->dpif, port->odp_port);
1839 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1842 if (port->odp_port != OVSP_NONE && !port->tnl_port) {
1843 hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1846 tnl_port_del(port->tnl_port);
1847 sset_find_and_delete(&ofproto->ports, devname);
1848 sset_find_and_delete(&ofproto->ghost_ports, devname);
1849 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1850 bundle_remove(port_);
1851 set_cfm(port_, NULL);
1852 if (ofproto->sflow) {
1853 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1856 ofport_clear_priorities(port);
1857 hmap_destroy(&port->priorities);
1861 port_modified(struct ofport *port_)
1863 struct ofport_dpif *port = ofport_dpif_cast(port_);
1865 if (port->bundle && port->bundle->bond) {
1866 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1871 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1873 struct ofport_dpif *port = ofport_dpif_cast(port_);
1874 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1875 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1877 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1878 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1879 OFPUTIL_PC_NO_PACKET_IN)) {
1880 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1882 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1883 bundle_update(port->bundle);
1889 set_sflow(struct ofproto *ofproto_,
1890 const struct ofproto_sflow_options *sflow_options)
1892 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1893 struct dpif_sflow *ds = ofproto->sflow;
1895 if (sflow_options) {
1897 struct ofport_dpif *ofport;
1899 ds = ofproto->sflow = dpif_sflow_create();
1900 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1901 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1903 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1905 dpif_sflow_set_options(ds, sflow_options);
1908 dpif_sflow_destroy(ds);
1909 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1910 ofproto->sflow = NULL;
1918 struct ofproto *ofproto_,
1919 const struct ofproto_ipfix_bridge_exporter_options *bridge_exporter_options,
1920 const struct ofproto_ipfix_flow_exporter_options *flow_exporters_options,
1921 size_t n_flow_exporters_options)
1923 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1924 struct dpif_ipfix *di = ofproto->ipfix;
1926 if (bridge_exporter_options || flow_exporters_options) {
1928 di = ofproto->ipfix = dpif_ipfix_create();
1930 dpif_ipfix_set_options(
1931 di, bridge_exporter_options, flow_exporters_options,
1932 n_flow_exporters_options);
1935 dpif_ipfix_destroy(di);
1936 ofproto->ipfix = NULL;
1943 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1945 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1952 struct ofproto_dpif *ofproto;
1954 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1955 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1956 ofport->cfm = cfm_create(ofport->up.netdev);
1959 if (cfm_configure(ofport->cfm, s)) {
1965 cfm_destroy(ofport->cfm);
1971 get_cfm_status(const struct ofport *ofport_,
1972 struct ofproto_cfm_status *status)
1974 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1977 status->faults = cfm_get_fault(ofport->cfm);
1978 status->remote_opstate = cfm_get_opup(ofport->cfm);
1979 status->health = cfm_get_health(ofport->cfm);
1980 cfm_get_remote_mpids(ofport->cfm, &status->rmps, &status->n_rmps);
1987 /* Spanning Tree. */
1990 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
1992 struct ofproto_dpif *ofproto = ofproto_;
1993 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
1994 struct ofport_dpif *ofport;
1996 ofport = stp_port_get_aux(sp);
1998 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
1999 ofproto->up.name, port_num);
2001 struct eth_header *eth = pkt->l2;
2003 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
2004 if (eth_addr_is_zero(eth->eth_src)) {
2005 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
2006 "with unknown MAC", ofproto->up.name, port_num);
2008 send_packet(ofport, pkt);
2014 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
2016 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
2018 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2020 /* Only revalidate flows if the configuration changed. */
2021 if (!s != !ofproto->stp) {
2022 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2026 if (!ofproto->stp) {
2027 ofproto->stp = stp_create(ofproto_->name, s->system_id,
2028 send_bpdu_cb, ofproto);
2029 ofproto->stp_last_tick = time_msec();
2032 stp_set_bridge_id(ofproto->stp, s->system_id);
2033 stp_set_bridge_priority(ofproto->stp, s->priority);
2034 stp_set_hello_time(ofproto->stp, s->hello_time);
2035 stp_set_max_age(ofproto->stp, s->max_age);
2036 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
2038 struct ofport *ofport;
2040 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
2041 set_stp_port(ofport, NULL);
2044 stp_destroy(ofproto->stp);
2045 ofproto->stp = NULL;
2052 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
2054 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2058 s->bridge_id = stp_get_bridge_id(ofproto->stp);
2059 s->designated_root = stp_get_designated_root(ofproto->stp);
2060 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
2069 update_stp_port_state(struct ofport_dpif *ofport)
2071 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2072 enum stp_state state;
2074 /* Figure out new state. */
2075 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
2079 if (ofport->stp_state != state) {
2080 enum ofputil_port_state of_state;
2083 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
2084 netdev_get_name(ofport->up.netdev),
2085 stp_state_name(ofport->stp_state),
2086 stp_state_name(state));
2087 if (stp_learn_in_state(ofport->stp_state)
2088 != stp_learn_in_state(state)) {
2089 /* xxx Learning action flows should also be flushed. */
2090 mac_learning_flush(ofproto->ml,
2091 &ofproto->backer->revalidate_set);
2093 fwd_change = stp_forward_in_state(ofport->stp_state)
2094 != stp_forward_in_state(state);
2096 ofproto->backer->need_revalidate = REV_STP;
2097 ofport->stp_state = state;
2098 ofport->stp_state_entered = time_msec();
2100 if (fwd_change && ofport->bundle) {
2101 bundle_update(ofport->bundle);
2104 /* Update the STP state bits in the OpenFlow port description. */
2105 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
2106 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
2107 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
2108 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
2109 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
2111 ofproto_port_set_state(&ofport->up, of_state);
2115 /* Configures STP on 'ofport_' using the settings defined in 's'. The
2116 * caller is responsible for assigning STP port numbers and ensuring
2117 * there are no duplicates. */
2119 set_stp_port(struct ofport *ofport_,
2120 const struct ofproto_port_stp_settings *s)
2122 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2123 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2124 struct stp_port *sp = ofport->stp_port;
2126 if (!s || !s->enable) {
2128 ofport->stp_port = NULL;
2129 stp_port_disable(sp);
2130 update_stp_port_state(ofport);
2133 } else if (sp && stp_port_no(sp) != s->port_num
2134 && ofport == stp_port_get_aux(sp)) {
2135 /* The port-id changed, so disable the old one if it's not
2136 * already in use by another port. */
2137 stp_port_disable(sp);
2140 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
2141 stp_port_enable(sp);
2143 stp_port_set_aux(sp, ofport);
2144 stp_port_set_priority(sp, s->priority);
2145 stp_port_set_path_cost(sp, s->path_cost);
2147 update_stp_port_state(ofport);
2153 get_stp_port_status(struct ofport *ofport_,
2154 struct ofproto_port_stp_status *s)
2156 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2157 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2158 struct stp_port *sp = ofport->stp_port;
2160 if (!ofproto->stp || !sp) {
2166 s->port_id = stp_port_get_id(sp);
2167 s->state = stp_port_get_state(sp);
2168 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
2169 s->role = stp_port_get_role(sp);
2170 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
2176 stp_run(struct ofproto_dpif *ofproto)
2179 long long int now = time_msec();
2180 long long int elapsed = now - ofproto->stp_last_tick;
2181 struct stp_port *sp;
2184 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
2185 ofproto->stp_last_tick = now;
2187 while (stp_get_changed_port(ofproto->stp, &sp)) {
2188 struct ofport_dpif *ofport = stp_port_get_aux(sp);
2191 update_stp_port_state(ofport);
2195 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
2196 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
2202 stp_wait(struct ofproto_dpif *ofproto)
2205 poll_timer_wait(1000);
2209 /* Returns true if STP should process 'flow'. */
2211 stp_should_process_flow(const struct flow *flow)
2213 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
2217 stp_process_packet(const struct ofport_dpif *ofport,
2218 const struct ofpbuf *packet)
2220 struct ofpbuf payload = *packet;
2221 struct eth_header *eth = payload.data;
2222 struct stp_port *sp = ofport->stp_port;
2224 /* Sink packets on ports that have STP disabled when the bridge has
2226 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
2230 /* Trim off padding on payload. */
2231 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
2232 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
2235 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
2236 stp_received_bpdu(sp, payload.data, payload.size);
2240 static struct priority_to_dscp *
2241 get_priority(const struct ofport_dpif *ofport, uint32_t priority)
2243 struct priority_to_dscp *pdscp;
2246 hash = hash_int(priority, 0);
2247 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) {
2248 if (pdscp->priority == priority) {
2256 ofport_clear_priorities(struct ofport_dpif *ofport)
2258 struct priority_to_dscp *pdscp, *next;
2260 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) {
2261 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
2267 set_queues(struct ofport *ofport_,
2268 const struct ofproto_port_queue *qdscp_list,
2271 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2272 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2273 struct hmap new = HMAP_INITIALIZER(&new);
2276 for (i = 0; i < n_qdscp; i++) {
2277 struct priority_to_dscp *pdscp;
2281 dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
2282 if (dpif_queue_to_priority(ofproto->backer->dpif, qdscp_list[i].queue,
2287 pdscp = get_priority(ofport, priority);
2289 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
2291 pdscp = xmalloc(sizeof *pdscp);
2292 pdscp->priority = priority;
2294 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2297 if (pdscp->dscp != dscp) {
2299 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2302 hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0));
2305 if (!hmap_is_empty(&ofport->priorities)) {
2306 ofport_clear_priorities(ofport);
2307 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2310 hmap_swap(&new, &ofport->priorities);
2318 /* Expires all MAC learning entries associated with 'bundle' and forces its
2319 * ofproto to revalidate every flow.
2321 * Normally MAC learning entries are removed only from the ofproto associated
2322 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2323 * are removed from every ofproto. When patch ports and SLB bonds are in use
2324 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2325 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2326 * with the host from which it migrated. */
2328 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2330 struct ofproto_dpif *ofproto = bundle->ofproto;
2331 struct mac_learning *ml = ofproto->ml;
2332 struct mac_entry *mac, *next_mac;
2334 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2335 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2336 if (mac->port.p == bundle) {
2338 struct ofproto_dpif *o;
2340 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2342 struct mac_entry *e;
2344 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan,
2347 mac_learning_expire(o->ml, e);
2353 mac_learning_expire(ml, mac);
2358 static struct ofbundle *
2359 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2361 struct ofbundle *bundle;
2363 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2364 &ofproto->bundles) {
2365 if (bundle->aux == aux) {
2372 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
2373 * ones that are found to 'bundles'. */
2375 bundle_lookup_multiple(struct ofproto_dpif *ofproto,
2376 void **auxes, size_t n_auxes,
2377 struct hmapx *bundles)
2381 hmapx_init(bundles);
2382 for (i = 0; i < n_auxes; i++) {
2383 struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
2385 hmapx_add(bundles, bundle);
2391 bundle_update(struct ofbundle *bundle)
2393 struct ofport_dpif *port;
2395 bundle->floodable = true;
2396 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2397 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2398 || !stp_forward_in_state(port->stp_state)) {
2399 bundle->floodable = false;
2406 bundle_del_port(struct ofport_dpif *port)
2408 struct ofbundle *bundle = port->bundle;
2410 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2412 list_remove(&port->bundle_node);
2413 port->bundle = NULL;
2416 lacp_slave_unregister(bundle->lacp, port);
2419 bond_slave_unregister(bundle->bond, port);
2422 bundle_update(bundle);
2426 bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port,
2427 struct lacp_slave_settings *lacp)
2429 struct ofport_dpif *port;
2431 port = get_ofp_port(bundle->ofproto, ofp_port);
2436 if (port->bundle != bundle) {
2437 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2439 bundle_del_port(port);
2442 port->bundle = bundle;
2443 list_push_back(&bundle->ports, &port->bundle_node);
2444 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2445 || !stp_forward_in_state(port->stp_state)) {
2446 bundle->floodable = false;
2450 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2451 lacp_slave_register(bundle->lacp, port, lacp);
2458 bundle_destroy(struct ofbundle *bundle)
2460 struct ofproto_dpif *ofproto;
2461 struct ofport_dpif *port, *next_port;
2468 ofproto = bundle->ofproto;
2469 for (i = 0; i < MAX_MIRRORS; i++) {
2470 struct ofmirror *m = ofproto->mirrors[i];
2472 if (m->out == bundle) {
2474 } else if (hmapx_find_and_delete(&m->srcs, bundle)
2475 || hmapx_find_and_delete(&m->dsts, bundle)) {
2476 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2481 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2482 bundle_del_port(port);
2485 bundle_flush_macs(bundle, true);
2486 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2488 free(bundle->trunks);
2489 lacp_destroy(bundle->lacp);
2490 bond_destroy(bundle->bond);
2495 bundle_set(struct ofproto *ofproto_, void *aux,
2496 const struct ofproto_bundle_settings *s)
2498 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2499 bool need_flush = false;
2500 struct ofport_dpif *port;
2501 struct ofbundle *bundle;
2502 unsigned long *trunks;
2508 bundle_destroy(bundle_lookup(ofproto, aux));
2512 ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2513 ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2515 bundle = bundle_lookup(ofproto, aux);
2517 bundle = xmalloc(sizeof *bundle);
2519 bundle->ofproto = ofproto;
2520 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2521 hash_pointer(aux, 0));
2523 bundle->name = NULL;
2525 list_init(&bundle->ports);
2526 bundle->vlan_mode = PORT_VLAN_TRUNK;
2528 bundle->trunks = NULL;
2529 bundle->use_priority_tags = s->use_priority_tags;
2530 bundle->lacp = NULL;
2531 bundle->bond = NULL;
2533 bundle->floodable = true;
2535 bundle->src_mirrors = 0;
2536 bundle->dst_mirrors = 0;
2537 bundle->mirror_out = 0;
2540 if (!bundle->name || strcmp(s->name, bundle->name)) {
2542 bundle->name = xstrdup(s->name);
2547 if (!bundle->lacp) {
2548 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2549 bundle->lacp = lacp_create();
2551 lacp_configure(bundle->lacp, s->lacp);
2553 lacp_destroy(bundle->lacp);
2554 bundle->lacp = NULL;
2557 /* Update set of ports. */
2559 for (i = 0; i < s->n_slaves; i++) {
2560 if (!bundle_add_port(bundle, s->slaves[i],
2561 s->lacp ? &s->lacp_slaves[i] : NULL)) {
2565 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2566 struct ofport_dpif *next_port;
2568 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2569 for (i = 0; i < s->n_slaves; i++) {
2570 if (s->slaves[i] == port->up.ofp_port) {
2575 bundle_del_port(port);
2579 ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2581 if (list_is_empty(&bundle->ports)) {
2582 bundle_destroy(bundle);
2586 /* Set VLAN tagging mode */
2587 if (s->vlan_mode != bundle->vlan_mode
2588 || s->use_priority_tags != bundle->use_priority_tags) {
2589 bundle->vlan_mode = s->vlan_mode;
2590 bundle->use_priority_tags = s->use_priority_tags;
2595 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2596 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2598 if (vlan != bundle->vlan) {
2599 bundle->vlan = vlan;
2603 /* Get trunked VLANs. */
2604 switch (s->vlan_mode) {
2605 case PORT_VLAN_ACCESS:
2609 case PORT_VLAN_TRUNK:
2610 trunks = CONST_CAST(unsigned long *, s->trunks);
2613 case PORT_VLAN_NATIVE_UNTAGGED:
2614 case PORT_VLAN_NATIVE_TAGGED:
2615 if (vlan != 0 && (!s->trunks
2616 || !bitmap_is_set(s->trunks, vlan)
2617 || bitmap_is_set(s->trunks, 0))) {
2618 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2620 trunks = bitmap_clone(s->trunks, 4096);
2622 trunks = bitmap_allocate1(4096);
2624 bitmap_set1(trunks, vlan);
2625 bitmap_set0(trunks, 0);
2627 trunks = CONST_CAST(unsigned long *, s->trunks);
2634 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2635 free(bundle->trunks);
2636 if (trunks == s->trunks) {
2637 bundle->trunks = vlan_bitmap_clone(trunks);
2639 bundle->trunks = trunks;
2644 if (trunks != s->trunks) {
2649 if (!list_is_short(&bundle->ports)) {
2650 bundle->ofproto->has_bonded_bundles = true;
2652 if (bond_reconfigure(bundle->bond, s->bond)) {
2653 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2656 bundle->bond = bond_create(s->bond);
2657 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2660 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2661 bond_slave_register(bundle->bond, port, port->up.netdev);
2664 bond_destroy(bundle->bond);
2665 bundle->bond = NULL;
2668 /* If we changed something that would affect MAC learning, un-learn
2669 * everything on this port and force flow revalidation. */
2671 bundle_flush_macs(bundle, false);
2678 bundle_remove(struct ofport *port_)
2680 struct ofport_dpif *port = ofport_dpif_cast(port_);
2681 struct ofbundle *bundle = port->bundle;
2684 bundle_del_port(port);
2685 if (list_is_empty(&bundle->ports)) {
2686 bundle_destroy(bundle);
2687 } else if (list_is_short(&bundle->ports)) {
2688 bond_destroy(bundle->bond);
2689 bundle->bond = NULL;
2695 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2697 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2698 struct ofport_dpif *port = port_;
2699 uint8_t ea[ETH_ADDR_LEN];
2702 error = netdev_get_etheraddr(port->up.netdev, ea);
2704 struct ofpbuf packet;
2707 ofpbuf_init(&packet, 0);
2708 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2710 memcpy(packet_pdu, pdu, pdu_size);
2712 send_packet(port, &packet);
2713 ofpbuf_uninit(&packet);
2715 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2716 "%s (%s)", port->bundle->name,
2717 netdev_get_name(port->up.netdev), strerror(error));
2722 bundle_send_learning_packets(struct ofbundle *bundle)
2724 struct ofproto_dpif *ofproto = bundle->ofproto;
2725 int error, n_packets, n_errors;
2726 struct mac_entry *e;
2728 error = n_packets = n_errors = 0;
2729 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2730 if (e->port.p != bundle) {
2731 struct ofpbuf *learning_packet;
2732 struct ofport_dpif *port;
2736 /* The assignment to "port" is unnecessary but makes "grep"ing for
2737 * struct ofport_dpif more effective. */
2738 learning_packet = bond_compose_learning_packet(bundle->bond,
2742 ret = send_packet(port, learning_packet);
2743 ofpbuf_delete(learning_packet);
2753 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2754 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2755 "packets, last error was: %s",
2756 bundle->name, n_errors, n_packets, strerror(error));
2758 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2759 bundle->name, n_packets);
2764 bundle_run(struct ofbundle *bundle)
2767 lacp_run(bundle->lacp, send_pdu_cb);
2770 struct ofport_dpif *port;
2772 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2773 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2776 bond_run(bundle->bond, &bundle->ofproto->backer->revalidate_set,
2777 lacp_status(bundle->lacp));
2778 if (bond_should_send_learning_packets(bundle->bond)) {
2779 bundle_send_learning_packets(bundle);
2785 bundle_wait(struct ofbundle *bundle)
2788 lacp_wait(bundle->lacp);
2791 bond_wait(bundle->bond);
2798 mirror_scan(struct ofproto_dpif *ofproto)
2802 for (idx = 0; idx < MAX_MIRRORS; idx++) {
2803 if (!ofproto->mirrors[idx]) {
2810 static struct ofmirror *
2811 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
2815 for (i = 0; i < MAX_MIRRORS; i++) {
2816 struct ofmirror *mirror = ofproto->mirrors[i];
2817 if (mirror && mirror->aux == aux) {
2825 /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */
2827 mirror_update_dups(struct ofproto_dpif *ofproto)
2831 for (i = 0; i < MAX_MIRRORS; i++) {
2832 struct ofmirror *m = ofproto->mirrors[i];
2835 m->dup_mirrors = MIRROR_MASK_C(1) << i;
2839 for (i = 0; i < MAX_MIRRORS; i++) {
2840 struct ofmirror *m1 = ofproto->mirrors[i];
2847 for (j = i + 1; j < MAX_MIRRORS; j++) {
2848 struct ofmirror *m2 = ofproto->mirrors[j];
2850 if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) {
2851 m1->dup_mirrors |= MIRROR_MASK_C(1) << j;
2852 m2->dup_mirrors |= m1->dup_mirrors;
2859 mirror_set(struct ofproto *ofproto_, void *aux,
2860 const struct ofproto_mirror_settings *s)
2862 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2863 mirror_mask_t mirror_bit;
2864 struct ofbundle *bundle;
2865 struct ofmirror *mirror;
2866 struct ofbundle *out;
2867 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
2868 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
2871 mirror = mirror_lookup(ofproto, aux);
2873 mirror_destroy(mirror);
2879 idx = mirror_scan(ofproto);
2881 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
2883 ofproto->up.name, MAX_MIRRORS, s->name);
2887 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
2888 mirror->ofproto = ofproto;
2891 mirror->out_vlan = -1;
2892 mirror->name = NULL;
2895 if (!mirror->name || strcmp(s->name, mirror->name)) {
2897 mirror->name = xstrdup(s->name);
2900 /* Get the new configuration. */
2901 if (s->out_bundle) {
2902 out = bundle_lookup(ofproto, s->out_bundle);
2904 mirror_destroy(mirror);
2910 out_vlan = s->out_vlan;
2912 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
2913 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
2915 /* If the configuration has not changed, do nothing. */
2916 if (hmapx_equals(&srcs, &mirror->srcs)
2917 && hmapx_equals(&dsts, &mirror->dsts)
2918 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
2919 && mirror->out == out
2920 && mirror->out_vlan == out_vlan)
2922 hmapx_destroy(&srcs);
2923 hmapx_destroy(&dsts);
2927 hmapx_swap(&srcs, &mirror->srcs);
2928 hmapx_destroy(&srcs);
2930 hmapx_swap(&dsts, &mirror->dsts);
2931 hmapx_destroy(&dsts);
2933 free(mirror->vlans);
2934 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
2937 mirror->out_vlan = out_vlan;
2939 /* Update bundles. */
2940 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2941 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
2942 if (hmapx_contains(&mirror->srcs, bundle)) {
2943 bundle->src_mirrors |= mirror_bit;
2945 bundle->src_mirrors &= ~mirror_bit;
2948 if (hmapx_contains(&mirror->dsts, bundle)) {
2949 bundle->dst_mirrors |= mirror_bit;
2951 bundle->dst_mirrors &= ~mirror_bit;
2954 if (mirror->out == bundle) {
2955 bundle->mirror_out |= mirror_bit;
2957 bundle->mirror_out &= ~mirror_bit;
2961 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2962 ofproto->has_mirrors = true;
2963 mac_learning_flush(ofproto->ml,
2964 &ofproto->backer->revalidate_set);
2965 mirror_update_dups(ofproto);
2971 mirror_destroy(struct ofmirror *mirror)
2973 struct ofproto_dpif *ofproto;
2974 mirror_mask_t mirror_bit;
2975 struct ofbundle *bundle;
2982 ofproto = mirror->ofproto;
2983 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2984 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
2986 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2987 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
2988 bundle->src_mirrors &= ~mirror_bit;
2989 bundle->dst_mirrors &= ~mirror_bit;
2990 bundle->mirror_out &= ~mirror_bit;
2993 hmapx_destroy(&mirror->srcs);
2994 hmapx_destroy(&mirror->dsts);
2995 free(mirror->vlans);
2997 ofproto->mirrors[mirror->idx] = NULL;
3001 mirror_update_dups(ofproto);
3003 ofproto->has_mirrors = false;
3004 for (i = 0; i < MAX_MIRRORS; i++) {
3005 if (ofproto->mirrors[i]) {
3006 ofproto->has_mirrors = true;
3013 mirror_get_stats(struct ofproto *ofproto_, void *aux,
3014 uint64_t *packets, uint64_t *bytes)
3016 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3017 struct ofmirror *mirror = mirror_lookup(ofproto, aux);
3020 *packets = *bytes = UINT64_MAX;
3026 *packets = mirror->packet_count;
3027 *bytes = mirror->byte_count;
3033 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
3035 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3036 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
3037 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
3043 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
3045 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3046 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
3047 return bundle && bundle->mirror_out != 0;
3051 forward_bpdu_changed(struct ofproto *ofproto_)
3053 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3054 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3058 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
3061 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3062 mac_learning_set_idle_time(ofproto->ml, idle_time);
3063 mac_learning_set_max_entries(ofproto->ml, max_entries);
3068 static struct ofport_dpif *
3069 get_ofp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
3071 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
3072 return ofport ? ofport_dpif_cast(ofport) : NULL;
3075 static struct ofport_dpif *
3076 get_odp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
3078 struct ofport_dpif *port = odp_port_to_ofport(ofproto->backer, odp_port);
3079 return port && &ofproto->up == port->up.ofproto ? port : NULL;
3083 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
3084 struct ofproto_port *ofproto_port,
3085 struct dpif_port *dpif_port)
3087 ofproto_port->name = dpif_port->name;
3088 ofproto_port->type = dpif_port->type;
3089 ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
3092 static struct ofport_dpif *
3093 ofport_get_peer(const struct ofport_dpif *ofport_dpif)
3095 const struct ofproto_dpif *ofproto;
3098 peer = netdev_vport_patch_peer(ofport_dpif->up.netdev);
3103 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
3104 struct ofport *ofport;
3106 ofport = shash_find_data(&ofproto->up.port_by_name, peer);
3107 if (ofport && ofport->ofproto->ofproto_class == &ofproto_dpif_class) {
3108 return ofport_dpif_cast(ofport);
3115 port_run_fast(struct ofport_dpif *ofport)
3117 if (ofport->cfm && cfm_should_send_ccm(ofport->cfm)) {
3118 struct ofpbuf packet;
3120 ofpbuf_init(&packet, 0);
3121 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
3122 send_packet(ofport, &packet);
3123 ofpbuf_uninit(&packet);
3128 port_run(struct ofport_dpif *ofport)
3130 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
3131 bool carrier_changed = carrier_seq != ofport->carrier_seq;
3132 bool enable = netdev_get_carrier(ofport->up.netdev);
3134 ofport->carrier_seq = carrier_seq;
3136 port_run_fast(ofport);
3138 if (ofport->tnl_port
3139 && tnl_port_reconfigure(&ofport->up, ofport->odp_port,
3140 &ofport->tnl_port)) {
3141 ofproto_dpif_cast(ofport->up.ofproto)->backer->need_revalidate = true;
3145 int cfm_opup = cfm_get_opup(ofport->cfm);
3147 cfm_run(ofport->cfm);
3148 enable = enable && !cfm_get_fault(ofport->cfm);
3150 if (cfm_opup >= 0) {
3151 enable = enable && cfm_opup;
3155 if (ofport->bundle) {
3156 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
3157 if (carrier_changed) {
3158 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
3162 if (ofport->may_enable != enable) {
3163 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3165 if (ofproto->has_bundle_action) {
3166 ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
3170 ofport->may_enable = enable;
3174 port_wait(struct ofport_dpif *ofport)
3177 cfm_wait(ofport->cfm);
3182 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
3183 struct ofproto_port *ofproto_port)
3185 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3186 struct dpif_port dpif_port;
3189 if (sset_contains(&ofproto->ghost_ports, devname)) {
3190 const char *type = netdev_get_type_from_name(devname);
3192 /* We may be called before ofproto->up.port_by_name is populated with
3193 * the appropriate ofport. For this reason, we must get the name and
3194 * type from the netdev layer directly. */
3196 const struct ofport *ofport;
3198 ofport = shash_find_data(&ofproto->up.port_by_name, devname);
3199 ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
3200 ofproto_port->name = xstrdup(devname);
3201 ofproto_port->type = xstrdup(type);
3207 if (!sset_contains(&ofproto->ports, devname)) {
3210 error = dpif_port_query_by_name(ofproto->backer->dpif,
3211 devname, &dpif_port);
3213 ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
3219 port_add(struct ofproto *ofproto_, struct netdev *netdev)
3221 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3222 const char *dp_port_name = netdev_vport_get_dpif_port(netdev);
3223 const char *devname = netdev_get_name(netdev);
3225 if (netdev_vport_is_patch(netdev)) {
3226 sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
3230 if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
3231 uint32_t port_no = UINT32_MAX;
3234 error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
3238 if (netdev_get_tunnel_config(netdev)) {
3239 simap_put(&ofproto->backer->tnl_backers, dp_port_name, port_no);
3243 if (netdev_get_tunnel_config(netdev)) {
3244 sset_add(&ofproto->ghost_ports, devname);
3246 sset_add(&ofproto->ports, devname);
3252 port_del(struct ofproto *ofproto_, uint16_t ofp_port)
3254 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3255 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
3262 sset_find_and_delete(&ofproto->ghost_ports,
3263 netdev_get_name(ofport->up.netdev));
3264 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3265 if (!ofport->tnl_port) {
3266 error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
3268 /* The caller is going to close ofport->up.netdev. If this is a
3269 * bonded port, then the bond is using that netdev, so remove it
3270 * from the bond. The client will need to reconfigure everything
3271 * after deleting ports, so then the slave will get re-added. */
3272 bundle_remove(&ofport->up);
3279 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
3281 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3286 error = netdev_get_stats(ofport->up.netdev, stats);
3288 if (!error && ofport_->ofp_port == OFPP_LOCAL) {
3289 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3291 /* ofproto->stats.tx_packets represents packets that we created
3292 * internally and sent to some port (e.g. packets sent with
3293 * send_packet()). Account for them as if they had come from
3294 * OFPP_LOCAL and got forwarded. */
3296 if (stats->rx_packets != UINT64_MAX) {
3297 stats->rx_packets += ofproto->stats.tx_packets;
3300 if (stats->rx_bytes != UINT64_MAX) {
3301 stats->rx_bytes += ofproto->stats.tx_bytes;
3304 /* ofproto->stats.rx_packets represents packets that were received on
3305 * some port and we processed internally and dropped (e.g. STP).
3306 * Account for them as if they had been forwarded to OFPP_LOCAL. */
3308 if (stats->tx_packets != UINT64_MAX) {
3309 stats->tx_packets += ofproto->stats.rx_packets;
3312 if (stats->tx_bytes != UINT64_MAX) {
3313 stats->tx_bytes += ofproto->stats.rx_bytes;
3320 /* Account packets for LOCAL port. */
3322 ofproto_update_local_port_stats(const struct ofproto *ofproto_,
3323 size_t tx_size, size_t rx_size)
3325 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3328 ofproto->stats.rx_packets++;
3329 ofproto->stats.rx_bytes += rx_size;
3332 ofproto->stats.tx_packets++;
3333 ofproto->stats.tx_bytes += tx_size;
3337 struct port_dump_state {
3342 struct ofproto_port port;
3347 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
3349 *statep = xzalloc(sizeof(struct port_dump_state));
3354 port_dump_next(const struct ofproto *ofproto_, void *state_,
3355 struct ofproto_port *port)
3357 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3358 struct port_dump_state *state = state_;
3359 const struct sset *sset;
3360 struct sset_node *node;
3362 if (state->has_port) {
3363 ofproto_port_destroy(&state->port);
3364 state->has_port = false;
3366 sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
3367 while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
3370 error = port_query_by_name(ofproto_, node->name, &state->port);
3372 *port = state->port;
3373 state->has_port = true;
3375 } else if (error != ENODEV) {
3380 if (!state->ghost) {
3381 state->ghost = true;
3384 return port_dump_next(ofproto_, state_, port);
3391 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3393 struct port_dump_state *state = state_;
3395 if (state->has_port) {
3396 ofproto_port_destroy(&state->port);
3403 port_poll(const struct ofproto *ofproto_, char **devnamep)
3405 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3407 if (ofproto->port_poll_errno) {
3408 int error = ofproto->port_poll_errno;
3409 ofproto->port_poll_errno = 0;
3413 if (sset_is_empty(&ofproto->port_poll_set)) {
3417 *devnamep = sset_pop(&ofproto->port_poll_set);
3422 port_poll_wait(const struct ofproto *ofproto_)
3424 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3425 dpif_port_poll_wait(ofproto->backer->dpif);
3429 port_is_lacp_current(const struct ofport *ofport_)
3431 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3432 return (ofport->bundle && ofport->bundle->lacp
3433 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3437 /* Upcall handling. */
3439 /* Flow miss batching.
3441 * Some dpifs implement operations faster when you hand them off in a batch.
3442 * To allow batching, "struct flow_miss" queues the dpif-related work needed
3443 * for a given flow. Each "struct flow_miss" corresponds to sending one or
3444 * more packets, plus possibly installing the flow in the dpif.
3446 * So far we only batch the operations that affect flow setup time the most.
3447 * It's possible to batch more than that, but the benefit might be minimal. */
3449 struct hmap_node hmap_node;
3450 struct ofproto_dpif *ofproto;
3452 enum odp_key_fitness key_fitness;
3453 const struct nlattr *key;
3455 struct initial_vals initial_vals;
3456 struct list packets;
3457 enum dpif_upcall_type upcall_type;
3458 uint32_t odp_in_port;
3461 struct flow_miss_op {
3462 struct dpif_op dpif_op;
3464 uint64_t slow_stub[128 / 8]; /* Buffer for compose_slow_path() */
3465 struct xlate_out xout;
3466 bool xout_garbage; /* 'xout' needs to be uninitialized? */
3469 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
3470 * OpenFlow controller as necessary according to their individual
3471 * configurations. */
3473 send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet,
3474 const struct flow *flow)
3476 struct ofputil_packet_in pin;
3478 pin.packet = packet->data;
3479 pin.packet_len = packet->size;
3480 pin.reason = OFPR_NO_MATCH;
3481 pin.controller_id = 0;
3486 pin.send_len = 0; /* not used for flow table misses */
3488 flow_get_metadata(flow, &pin.fmd);
3490 connmgr_send_packet_in(ofproto->up.connmgr, &pin);
3493 static enum slow_path_reason
3494 process_special(struct ofproto_dpif *ofproto, const struct flow *flow,
3495 const struct ofport_dpif *ofport, const struct ofpbuf *packet)
3499 } else if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) {
3501 cfm_process_heartbeat(ofport->cfm, packet);
3504 } else if (ofport->bundle && ofport->bundle->lacp
3505 && flow->dl_type == htons(ETH_TYPE_LACP)) {
3507 lacp_process_packet(ofport->bundle->lacp, ofport, packet);
3510 } else if (ofproto->stp && stp_should_process_flow(flow)) {
3512 stp_process_packet(ofport, packet);
3520 static struct flow_miss *
3521 flow_miss_find(struct hmap *todo, const struct ofproto_dpif *ofproto,
3522 const struct flow *flow, uint32_t hash)
3524 struct flow_miss *miss;
3526 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
3527 if (miss->ofproto == ofproto && flow_equal(&miss->flow, flow)) {
3535 /* Partially Initializes 'op' as an "execute" operation for 'miss' and
3536 * 'packet'. The caller must initialize op->actions and op->actions_len. If
3537 * 'miss' is associated with a subfacet the caller must also initialize the
3538 * returned op->subfacet, and if anything needs to be freed after processing
3539 * the op, the caller must initialize op->garbage also. */
3541 init_flow_miss_execute_op(struct flow_miss *miss, struct ofpbuf *packet,
3542 struct flow_miss_op *op)
3544 if (miss->flow.vlan_tci != miss->initial_vals.vlan_tci) {
3545 /* This packet was received on a VLAN splinter port. We
3546 * added a VLAN to the packet to make the packet resemble
3547 * the flow, but the actions were composed assuming that
3548 * the packet contained no VLAN. So, we must remove the
3549 * VLAN header from the packet before trying to execute the
3551 eth_pop_vlan(packet);
3554 op->xout_garbage = false;
3555 op->dpif_op.type = DPIF_OP_EXECUTE;
3556 op->dpif_op.u.execute.key = miss->key;
3557 op->dpif_op.u.execute.key_len = miss->key_len;
3558 op->dpif_op.u.execute.packet = packet;
3561 /* Helper for handle_flow_miss_without_facet() and
3562 * handle_flow_miss_with_facet(). */
3564 handle_flow_miss_common(struct rule_dpif *rule,
3565 struct ofpbuf *packet, const struct flow *flow)
3567 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3569 ofproto->n_matches++;
3571 if (rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
3573 * Extra-special case for fail-open mode.
3575 * We are in fail-open mode and the packet matched the fail-open
3576 * rule, but we are connected to a controller too. We should send
3577 * the packet up to the controller in the hope that it will try to
3578 * set up a flow and thereby allow us to exit fail-open.
3580 * See the top-level comment in fail-open.c for more information.
3582 send_packet_in_miss(ofproto, packet, flow);
3586 /* Figures out whether a flow that missed in 'ofproto', whose details are in
3587 * 'miss', is likely to be worth tracking in detail in userspace and (usually)
3588 * installing a datapath flow. The answer is usually "yes" (a return value of
3589 * true). However, for short flows the cost of bookkeeping is much higher than
3590 * the benefits, so when the datapath holds a large number of flows we impose
3591 * some heuristics to decide which flows are likely to be worth tracking. */
3593 flow_miss_should_make_facet(struct ofproto_dpif *ofproto,
3594 struct flow_miss *miss, uint32_t hash)
3596 if (!ofproto->governor) {
3599 n_subfacets = hmap_count(&ofproto->subfacets);
3600 if (n_subfacets * 2 <= ofproto->up.flow_eviction_threshold) {
3604 ofproto->governor = governor_create(ofproto->up.name);
3607 return governor_should_install_flow(ofproto->governor, hash,
3608 list_size(&miss->packets));
3611 /* Handles 'miss', which matches 'rule', without creating a facet or subfacet
3612 * or creating any datapath flow. May add an "execute" operation to 'ops' and
3613 * increment '*n_ops'. */
3615 handle_flow_miss_without_facet(struct flow_miss *miss,
3616 struct flow_miss_op *ops, size_t *n_ops)
3618 struct rule_dpif *rule = rule_dpif_lookup(miss->ofproto, &miss->flow);
3619 long long int now = time_msec();
3620 struct ofpbuf *packet;
3621 struct xlate_in xin;
3623 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3624 struct flow_miss_op *op = &ops[*n_ops];
3625 struct dpif_flow_stats stats;
3627 COVERAGE_INC(facet_suppress);
3629 handle_flow_miss_common(rule, packet, &miss->flow);
3631 dpif_flow_stats_extract(&miss->flow, packet, now, &stats);
3632 rule_credit_stats(rule, &stats);
3634 xlate_in_init(&xin, miss->ofproto, &miss->flow, &miss->initial_vals,
3635 rule, stats.tcp_flags, packet);
3636 xin.resubmit_stats = &stats;
3637 xlate_actions(&xin, &op->xout);
3639 if (op->xout.odp_actions.size) {
3640 struct dpif_execute *execute = &op->dpif_op.u.execute;
3642 init_flow_miss_execute_op(miss, packet, op);
3643 execute->actions = op->xout.odp_actions.data;
3644 execute->actions_len = op->xout.odp_actions.size;
3645 op->xout_garbage = true;
3649 xlate_out_uninit(&op->xout);
3654 /* Handles 'miss', which matches 'facet'. May add any required datapath
3655 * operations to 'ops', incrementing '*n_ops' for each new op.
3657 * All of the packets in 'miss' are considered to have arrived at time 'now'.
3658 * This is really important only for new facets: if we just called time_msec()
3659 * here, then the new subfacet or its packets could look (occasionally) as
3660 * though it was used some time after the facet was used. That can make a
3661 * one-packet flow look like it has a nonzero duration, which looks odd in
3662 * e.g. NetFlow statistics. */
3664 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
3666 struct flow_miss_op *ops, size_t *n_ops)
3668 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3669 enum subfacet_path want_path;
3670 struct subfacet *subfacet;
3671 struct ofpbuf *packet;
3673 subfacet = subfacet_create(facet, miss, now);
3674 want_path = subfacet->facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
3676 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3677 struct flow_miss_op *op = &ops[*n_ops];
3678 struct dpif_flow_stats stats;
3680 handle_flow_miss_common(facet->rule, packet, &miss->flow);
3682 if (want_path != SF_FAST_PATH) {
3683 struct xlate_in xin;
3685 xlate_in_init(&xin, ofproto, &facet->flow, &facet->initial_vals,
3686 facet->rule, 0, packet);
3687 xlate_actions_for_side_effects(&xin);
3690 dpif_flow_stats_extract(&facet->flow, packet, now, &stats);
3691 subfacet_update_stats(subfacet, &stats);
3693 if (facet->xout.odp_actions.size) {
3694 struct dpif_execute *execute = &op->dpif_op.u.execute;
3696 init_flow_miss_execute_op(miss, packet, op);
3697 execute->actions = facet->xout.odp_actions.data,
3698 execute->actions_len = facet->xout.odp_actions.size;
3703 if (miss->upcall_type == DPIF_UC_MISS || subfacet->path != want_path) {
3704 struct flow_miss_op *op = &ops[(*n_ops)++];
3705 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
3707 subfacet->path = want_path;
3709 op->xout_garbage = false;
3710 op->dpif_op.type = DPIF_OP_FLOW_PUT;
3711 put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
3712 put->key = miss->key;
3713 put->key_len = miss->key_len;
3714 if (want_path == SF_FAST_PATH) {
3715 put->actions = facet->xout.odp_actions.data;
3716 put->actions_len = facet->xout.odp_actions.size;
3718 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
3719 op->slow_stub, sizeof op->slow_stub,
3720 &put->actions, &put->actions_len);
3726 /* Handles flow miss 'miss'. May add any required datapath operations
3727 * to 'ops', incrementing '*n_ops' for each new op. */
3729 handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops,
3732 struct ofproto_dpif *ofproto = miss->ofproto;
3733 struct facet *facet;
3737 /* The caller must ensure that miss->hmap_node.hash contains
3738 * flow_hash(miss->flow, 0). */
3739 hash = miss->hmap_node.hash;
3741 facet = facet_lookup_valid(ofproto, &miss->flow, hash);
3743 /* There does not exist a bijection between 'struct flow' and datapath
3744 * flow keys with fitness ODP_FIT_TO_LITTLE. This breaks a fundamental
3745 * assumption used throughout the facet and subfacet handling code.
3746 * Since we have to handle these misses in userspace anyway, we simply
3747 * skip facet creation, avoiding the problem alltogether. */
3748 if (miss->key_fitness == ODP_FIT_TOO_LITTLE
3749 || !flow_miss_should_make_facet(ofproto, miss, hash)) {
3750 handle_flow_miss_without_facet(miss, ops, n_ops);
3754 facet = facet_create(miss, hash);
3759 handle_flow_miss_with_facet(miss, facet, now, ops, n_ops);
3762 static struct drop_key *
3763 drop_key_lookup(const struct dpif_backer *backer, const struct nlattr *key,
3766 struct drop_key *drop_key;
3768 HMAP_FOR_EACH_WITH_HASH (drop_key, hmap_node, hash_bytes(key, key_len, 0),
3769 &backer->drop_keys) {
3770 if (drop_key->key_len == key_len
3771 && !memcmp(drop_key->key, key, key_len)) {
3779 drop_key_clear(struct dpif_backer *backer)
3781 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3782 struct drop_key *drop_key, *next;
3784 HMAP_FOR_EACH_SAFE (drop_key, next, hmap_node, &backer->drop_keys) {
3787 error = dpif_flow_del(backer->dpif, drop_key->key, drop_key->key_len,
3789 if (error && !VLOG_DROP_WARN(&rl)) {
3790 struct ds ds = DS_EMPTY_INITIALIZER;
3791 odp_flow_key_format(drop_key->key, drop_key->key_len, &ds);
3792 VLOG_WARN("Failed to delete drop key (%s) (%s)", strerror(error),
3797 hmap_remove(&backer->drop_keys, &drop_key->hmap_node);
3798 free(drop_key->key);
3803 /* Given a datpath, packet, and flow metadata ('backer', 'packet', and 'key'
3804 * respectively), populates 'flow' with the result of odp_flow_key_to_flow().
3805 * Optionally, if nonnull, populates 'fitnessp' with the fitness of 'flow' as
3806 * returned by odp_flow_key_to_flow(). Also, optionally populates 'ofproto'
3807 * with the ofproto_dpif, and 'odp_in_port' with the datapath in_port, that
3808 * 'packet' ingressed.
3810 * If 'ofproto' is nonnull, requires 'flow''s in_port to exist. Otherwise sets
3811 * 'flow''s in_port to OFPP_NONE.
3813 * This function does post-processing on data returned from
3814 * odp_flow_key_to_flow() to help make VLAN splinters transparent to the rest
3815 * of the upcall processing logic. In particular, if the extracted in_port is
3816 * a VLAN splinter port, it replaces flow->in_port by the "real" port, sets
3817 * flow->vlan_tci correctly for the VLAN of the VLAN splinter port, and pushes
3818 * a VLAN header onto 'packet' (if it is nonnull).
3820 * Optionally, if 'initial_vals' is nonnull, sets 'initial_vals->vlan_tci'
3821 * to the VLAN TCI with which the packet was really received, that is, the
3822 * actual VLAN TCI extracted by odp_flow_key_to_flow(). (This differs from
3823 * the value returned in flow->vlan_tci only for packets received on
3824 * VLAN splinters.) Also, if received on an IP tunnel, sets
3825 * 'initial_vals->tunnel_ip_tos' to the tunnel's IP TOS.
3827 * Similarly, this function also includes some logic to help with tunnels. It
3828 * may modify 'flow' as necessary to make the tunneling implementation
3829 * transparent to the upcall processing logic.
3831 * Returns 0 if successful, ENODEV if the parsed flow has no associated ofport,
3832 * or some other positive errno if there are other problems. */
3834 ofproto_receive(const struct dpif_backer *backer, struct ofpbuf *packet,
3835 const struct nlattr *key, size_t key_len,
3836 struct flow *flow, enum odp_key_fitness *fitnessp,
3837 struct ofproto_dpif **ofproto, uint32_t *odp_in_port,
3838 struct initial_vals *initial_vals)
3840 const struct ofport_dpif *port;
3841 enum odp_key_fitness fitness;
3844 fitness = odp_flow_key_to_flow(key, key_len, flow);
3845 if (fitness == ODP_FIT_ERROR) {
3851 initial_vals->vlan_tci = flow->vlan_tci;
3852 initial_vals->tunnel_ip_tos = flow->tunnel.ip_tos;
3856 *odp_in_port = flow->in_port;
3859 if (tnl_port_should_receive(flow)) {
3860 const struct ofport *ofport = tnl_port_receive(flow);
3862 flow->in_port = OFPP_NONE;
3865 port = ofport_dpif_cast(ofport);
3867 /* We can't reproduce 'key' from 'flow'. */
3868 fitness = fitness == ODP_FIT_PERFECT ? ODP_FIT_TOO_MUCH : fitness;
3870 /* XXX: Since the tunnel module is not scoped per backer, it's
3871 * theoretically possible that we'll receive an ofport belonging to an
3872 * entirely different datapath. In practice, this can't happen because
3873 * no platforms has two separate datapaths which each support
3875 ovs_assert(ofproto_dpif_cast(port->up.ofproto)->backer == backer);
3877 port = odp_port_to_ofport(backer, flow->in_port);
3879 flow->in_port = OFPP_NONE;
3883 flow->in_port = port->up.ofp_port;
3884 if (vsp_adjust_flow(ofproto_dpif_cast(port->up.ofproto), flow)) {
3886 /* Make the packet resemble the flow, so that it gets sent to
3887 * an OpenFlow controller properly, so that it looks correct
3888 * for sFlow, and so that flow_extract() will get the correct
3889 * vlan_tci if it is called on 'packet'.
3891 * The allocated space inside 'packet' probably also contains
3892 * 'key', that is, both 'packet' and 'key' are probably part of
3893 * a struct dpif_upcall (see the large comment on that
3894 * structure definition), so pushing data on 'packet' is in
3895 * general not a good idea since it could overwrite 'key' or
3896 * free it as a side effect. However, it's OK in this special
3897 * case because we know that 'packet' is inside a Netlink
3898 * attribute: pushing 4 bytes will just overwrite the 4-byte
3899 * "struct nlattr", which is fine since we don't need that
3900 * header anymore. */
3901 eth_push_vlan(packet, flow->vlan_tci);
3903 /* We can't reproduce 'key' from 'flow'. */
3904 fitness = fitness == ODP_FIT_PERFECT ? ODP_FIT_TOO_MUCH : fitness;
3910 *ofproto = ofproto_dpif_cast(port->up.ofproto);
3915 *fitnessp = fitness;
3921 handle_miss_upcalls(struct dpif_backer *backer, struct dpif_upcall *upcalls,
3924 struct dpif_upcall *upcall;
3925 struct flow_miss *miss;
3926 struct flow_miss misses[FLOW_MISS_MAX_BATCH];
3927 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
3928 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
3938 /* Construct the to-do list.
3940 * This just amounts to extracting the flow from each packet and sticking
3941 * the packets that have the same flow in the same "flow_miss" structure so
3942 * that we can process them together. */
3945 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
3946 struct flow_miss *miss = &misses[n_misses];
3947 struct flow_miss *existing_miss;
3948 struct ofproto_dpif *ofproto;
3949 uint32_t odp_in_port;
3954 error = ofproto_receive(backer, upcall->packet, upcall->key,
3955 upcall->key_len, &flow, &miss->key_fitness,
3956 &ofproto, &odp_in_port, &miss->initial_vals);
3957 if (error == ENODEV) {
3958 struct drop_key *drop_key;
3960 /* Received packet on port for which we couldn't associate
3961 * an ofproto. This can happen if a port is removed while
3962 * traffic is being received. Print a rate-limited message
3963 * in case it happens frequently. Install a drop flow so
3964 * that future packets of the flow are inexpensively dropped
3966 VLOG_INFO_RL(&rl, "received packet on unassociated port %"PRIu32,
3969 drop_key = drop_key_lookup(backer, upcall->key, upcall->key_len);
3971 drop_key = xmalloc(sizeof *drop_key);
3972 drop_key->key = xmemdup(upcall->key, upcall->key_len);
3973 drop_key->key_len = upcall->key_len;
3975 hmap_insert(&backer->drop_keys, &drop_key->hmap_node,
3976 hash_bytes(drop_key->key, drop_key->key_len, 0));
3977 dpif_flow_put(backer->dpif, DPIF_FP_CREATE | DPIF_FP_MODIFY,
3978 drop_key->key, drop_key->key_len, NULL, 0, NULL);
3986 ofproto->n_missed++;
3987 flow_extract(upcall->packet, flow.skb_priority, flow.skb_mark,
3988 &flow.tunnel, flow.in_port, &miss->flow);
3990 /* Add other packets to a to-do list. */
3991 hash = flow_hash(&miss->flow, 0);
3992 existing_miss = flow_miss_find(&todo, ofproto, &miss->flow, hash);
3993 if (!existing_miss) {
3994 hmap_insert(&todo, &miss->hmap_node, hash);
3995 miss->ofproto = ofproto;
3996 miss->key = upcall->key;
3997 miss->key_len = upcall->key_len;
3998 miss->upcall_type = upcall->type;
3999 miss->odp_in_port = odp_in_port;
4000 list_init(&miss->packets);
4004 miss = existing_miss;
4006 list_push_back(&miss->packets, &upcall->packet->list_node);
4009 /* Process each element in the to-do list, constructing the set of
4010 * operations to batch. */
4012 HMAP_FOR_EACH (miss, hmap_node, &todo) {
4013 handle_flow_miss(miss, flow_miss_ops, &n_ops);
4015 ovs_assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
4017 /* Execute batch. */
4018 for (i = 0; i < n_ops; i++) {
4019 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
4021 dpif_operate(backer->dpif, dpif_ops, n_ops);
4024 for (i = 0; i < n_ops; i++) {
4025 if (flow_miss_ops[i].xout_garbage) {
4026 xlate_out_uninit(&flow_miss_ops[i].xout);
4029 hmap_destroy(&todo);
4032 static enum { SFLOW_UPCALL, MISS_UPCALL, BAD_UPCALL, FLOW_SAMPLE_UPCALL,
4034 classify_upcall(const struct dpif_upcall *upcall)
4036 size_t userdata_len;
4037 union user_action_cookie cookie;
4039 /* First look at the upcall type. */
4040 switch (upcall->type) {
4041 case DPIF_UC_ACTION:
4047 case DPIF_N_UC_TYPES:
4049 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
4053 /* "action" upcalls need a closer look. */
4054 if (!upcall->userdata) {
4055 VLOG_WARN_RL(&rl, "action upcall missing cookie");
4058 userdata_len = nl_attr_get_size(upcall->userdata);
4059 if (userdata_len < sizeof cookie.type
4060 || userdata_len > sizeof cookie) {
4061 VLOG_WARN_RL(&rl, "action upcall cookie has unexpected size %zu",
4065 memset(&cookie, 0, sizeof cookie);
4066 memcpy(&cookie, nl_attr_get(upcall->userdata), userdata_len);
4067 if (userdata_len == sizeof cookie.sflow
4068 && cookie.type == USER_ACTION_COOKIE_SFLOW) {
4069 return SFLOW_UPCALL;
4070 } else if (userdata_len == sizeof cookie.slow_path
4071 && cookie.type == USER_ACTION_COOKIE_SLOW_PATH) {
4073 } else if (userdata_len == sizeof cookie.flow_sample
4074 && cookie.type == USER_ACTION_COOKIE_FLOW_SAMPLE) {
4075 return FLOW_SAMPLE_UPCALL;
4076 } else if (userdata_len == sizeof cookie.ipfix
4077 && cookie.type == USER_ACTION_COOKIE_IPFIX) {
4078 return IPFIX_UPCALL;
4080 VLOG_WARN_RL(&rl, "invalid user cookie of type %"PRIu16
4081 " and size %zu", cookie.type, userdata_len);
4087 handle_sflow_upcall(struct dpif_backer *backer,
4088 const struct dpif_upcall *upcall)
4090 struct ofproto_dpif *ofproto;
4091 union user_action_cookie cookie;
4093 uint32_t odp_in_port;
4095 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
4096 &flow, NULL, &ofproto, &odp_in_port, NULL)
4097 || !ofproto->sflow) {
4101 memset(&cookie, 0, sizeof cookie);
4102 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.sflow);
4103 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
4104 odp_in_port, &cookie);
4108 handle_flow_sample_upcall(struct dpif_backer *backer,
4109 const struct dpif_upcall *upcall)
4111 struct ofproto_dpif *ofproto;
4112 union user_action_cookie cookie;
4115 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
4116 &flow, NULL, &ofproto, NULL, NULL)
4117 || !ofproto->ipfix) {
4121 memset(&cookie, 0, sizeof cookie);
4122 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.flow_sample);
4124 /* The flow reflects exactly the contents of the packet. Sample
4125 * the packet using it. */
4126 dpif_ipfix_flow_sample(ofproto->ipfix, upcall->packet, &flow,
4127 cookie.flow_sample.collector_set_id,
4128 cookie.flow_sample.probability,
4129 cookie.flow_sample.obs_domain_id,
4130 cookie.flow_sample.obs_point_id);
4134 handle_ipfix_upcall(struct dpif_backer *backer,
4135 const struct dpif_upcall *upcall)
4137 struct ofproto_dpif *ofproto;
4140 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
4141 &flow, NULL, &ofproto, NULL, NULL)
4142 || !ofproto->ipfix) {
4146 /* The flow reflects exactly the contents of the packet. Sample
4147 * the packet using it. */
4148 dpif_ipfix_bridge_sample(ofproto->ipfix, upcall->packet, &flow);
4152 handle_upcalls(struct dpif_backer *backer, unsigned int max_batch)
4154 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
4155 struct ofpbuf miss_bufs[FLOW_MISS_MAX_BATCH];
4156 uint64_t miss_buf_stubs[FLOW_MISS_MAX_BATCH][4096 / 8];
4161 ovs_assert(max_batch <= FLOW_MISS_MAX_BATCH);
4164 for (n_processed = 0; n_processed < max_batch; n_processed++) {
4165 struct dpif_upcall *upcall = &misses[n_misses];
4166 struct ofpbuf *buf = &miss_bufs[n_misses];
4169 ofpbuf_use_stub(buf, miss_buf_stubs[n_misses],
4170 sizeof miss_buf_stubs[n_misses]);
4171 error = dpif_recv(backer->dpif, upcall, buf);
4177 switch (classify_upcall(upcall)) {
4179 /* Handle it later. */
4184 handle_sflow_upcall(backer, upcall);
4188 case FLOW_SAMPLE_UPCALL:
4189 handle_flow_sample_upcall(backer, upcall);
4194 handle_ipfix_upcall(backer, upcall);
4204 /* Handle deferred MISS_UPCALL processing. */
4205 handle_miss_upcalls(backer, misses, n_misses);
4206 for (i = 0; i < n_misses; i++) {
4207 ofpbuf_uninit(&miss_bufs[i]);
4213 /* Flow expiration. */
4215 static int subfacet_max_idle(const struct ofproto_dpif *);
4216 static void update_stats(struct dpif_backer *);
4217 static void rule_expire(struct rule_dpif *);
4218 static void expire_subfacets(struct ofproto_dpif *, int dp_max_idle);
4220 /* This function is called periodically by run(). Its job is to collect
4221 * updates for the flows that have been installed into the datapath, most
4222 * importantly when they last were used, and then use that information to
4223 * expire flows that have not been used recently.
4225 * Returns the number of milliseconds after which it should be called again. */
4227 expire(struct dpif_backer *backer)
4229 struct ofproto_dpif *ofproto;
4230 int max_idle = INT32_MAX;
4232 /* Periodically clear out the drop keys in an effort to keep them
4233 * relatively few. */
4234 drop_key_clear(backer);
4236 /* Update stats for each flow in the backer. */
4237 update_stats(backer);
4239 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4240 struct rule *rule, *next_rule;
4243 if (ofproto->backer != backer) {
4247 /* Keep track of the max number of flows per ofproto_dpif. */
4248 update_max_subfacet_count(ofproto);
4250 /* Expire subfacets that have been idle too long. */
4251 dp_max_idle = subfacet_max_idle(ofproto);
4252 expire_subfacets(ofproto, dp_max_idle);
4254 max_idle = MIN(max_idle, dp_max_idle);
4256 /* Expire OpenFlow flows whose idle_timeout or hard_timeout
4258 LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
4259 &ofproto->up.expirable) {
4260 rule_expire(rule_dpif_cast(rule));
4263 /* All outstanding data in existing flows has been accounted, so it's a
4264 * good time to do bond rebalancing. */
4265 if (ofproto->has_bonded_bundles) {
4266 struct ofbundle *bundle;
4268 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
4270 bond_rebalance(bundle->bond, &backer->revalidate_set);
4276 return MIN(max_idle, 1000);
4279 /* Updates flow table statistics given that the datapath just reported 'stats'
4280 * as 'subfacet''s statistics. */
4282 update_subfacet_stats(struct subfacet *subfacet,
4283 const struct dpif_flow_stats *stats)
4285 struct facet *facet = subfacet->facet;
4287 if (stats->n_packets >= subfacet->dp_packet_count) {
4288 uint64_t extra = stats->n_packets - subfacet->dp_packet_count;
4289 facet->packet_count += extra;
4291 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
4294 if (stats->n_bytes >= subfacet->dp_byte_count) {
4295 facet->byte_count += stats->n_bytes - subfacet->dp_byte_count;
4297 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
4300 subfacet->dp_packet_count = stats->n_packets;
4301 subfacet->dp_byte_count = stats->n_bytes;
4303 facet->tcp_flags |= stats->tcp_flags;
4305 subfacet_update_time(subfacet, stats->used);
4306 if (facet->accounted_bytes < facet->byte_count) {
4308 facet_account(facet);
4309 facet->accounted_bytes = facet->byte_count;
4313 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
4314 * about, or a flow that shouldn't be installed but was anyway. Delete it. */
4316 delete_unexpected_flow(struct ofproto_dpif *ofproto,
4317 const struct nlattr *key, size_t key_len)
4319 if (!VLOG_DROP_WARN(&rl)) {
4323 odp_flow_key_format(key, key_len, &s);
4324 VLOG_WARN("unexpected flow on %s: %s", ofproto->up.name, ds_cstr(&s));
4328 COVERAGE_INC(facet_unexpected);
4329 dpif_flow_del(ofproto->backer->dpif, key, key_len, NULL);
4332 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
4334 * This function also pushes statistics updates to rules which each facet
4335 * resubmits into. Generally these statistics will be accurate. However, if a
4336 * facet changes the rule it resubmits into at some time in between
4337 * update_stats() runs, it is possible that statistics accrued to the
4338 * old rule will be incorrectly attributed to the new rule. This could be
4339 * avoided by calling update_stats() whenever rules are created or
4340 * deleted. However, the performance impact of making so many calls to the
4341 * datapath do not justify the benefit of having perfectly accurate statistics.
4343 * In addition, this function maintains per ofproto flow hit counts. The patch
4344 * port is not treated specially. e.g. A packet ingress from br0 patched into
4345 * br1 will increase the hit count of br0 by 1, however, does not affect
4346 * the hit or miss counts of br1.
4349 update_stats(struct dpif_backer *backer)
4351 const struct dpif_flow_stats *stats;
4352 struct dpif_flow_dump dump;
4353 const struct nlattr *key;
4354 struct ofproto_dpif *ofproto;
4357 dpif_flow_dump_start(&dump, backer->dpif);
4358 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
4360 struct subfacet *subfacet;
4361 struct ofport_dpif *ofport;
4364 if (ofproto_receive(backer, NULL, key, key_len, &flow, NULL, &ofproto,
4369 ofproto->total_subfacet_count += hmap_count(&ofproto->subfacets);
4370 ofproto->n_update_stats++;
4372 ofport = get_ofp_port(ofproto, flow.in_port);
4373 if (ofport && ofport->tnl_port) {
4374 netdev_vport_inc_rx(ofport->up.netdev, stats);
4377 key_hash = odp_flow_key_hash(key, key_len);
4378 subfacet = subfacet_find(ofproto, key, key_len, key_hash);
4379 switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
4381 /* Update ofproto_dpif's hit count. */
4382 if (stats->n_packets > subfacet->dp_packet_count) {
4383 uint64_t delta = stats->n_packets - subfacet->dp_packet_count;
4384 dpif_stats_update_hit_count(ofproto, delta);
4387 update_subfacet_stats(subfacet, stats);
4391 /* Stats are updated per-packet. */
4394 case SF_NOT_INSTALLED:
4396 delete_unexpected_flow(ofproto, key, key_len);
4401 dpif_flow_dump_done(&dump);
4403 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4404 update_moving_averages(ofproto);
4409 /* Calculates and returns the number of milliseconds of idle time after which
4410 * subfacets should expire from the datapath. When a subfacet expires, we fold
4411 * its statistics into its facet, and when a facet's last subfacet expires, we
4412 * fold its statistic into its rule. */
4414 subfacet_max_idle(const struct ofproto_dpif *ofproto)
4417 * Idle time histogram.
4419 * Most of the time a switch has a relatively small number of subfacets.
4420 * When this is the case we might as well keep statistics for all of them
4421 * in userspace and to cache them in the kernel datapath for performance as
4424 * As the number of subfacets increases, the memory required to maintain
4425 * statistics about them in userspace and in the kernel becomes
4426 * significant. However, with a large number of subfacets it is likely
4427 * that only a few of them are "heavy hitters" that consume a large amount
4428 * of bandwidth. At this point, only heavy hitters are worth caching in
4429 * the kernel and maintaining in userspaces; other subfacets we can
4432 * The technique used to compute the idle time is to build a histogram with
4433 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
4434 * that is installed in the kernel gets dropped in the appropriate bucket.
4435 * After the histogram has been built, we compute the cutoff so that only
4436 * the most-recently-used 1% of subfacets (but at least
4437 * ofproto->up.flow_eviction_threshold flows) are kept cached. At least
4438 * the most-recently-used bucket of subfacets is kept, so actually an
4439 * arbitrary number of subfacets can be kept in any given expiration run
4440 * (though the next run will delete most of those unless they receive
4443 * This requires a second pass through the subfacets, in addition to the
4444 * pass made by update_stats(), because the former function never looks at
4445 * uninstallable subfacets.
4447 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
4448 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
4449 int buckets[N_BUCKETS] = { 0 };
4450 int total, subtotal, bucket;
4451 struct subfacet *subfacet;
4455 total = hmap_count(&ofproto->subfacets);
4456 if (total <= ofproto->up.flow_eviction_threshold) {
4457 return N_BUCKETS * BUCKET_WIDTH;
4460 /* Build histogram. */
4462 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
4463 long long int idle = now - subfacet->used;
4464 int bucket = (idle <= 0 ? 0
4465 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
4466 : (unsigned int) idle / BUCKET_WIDTH);
4470 /* Find the first bucket whose flows should be expired. */
4471 subtotal = bucket = 0;
4473 subtotal += buckets[bucket++];
4474 } while (bucket < N_BUCKETS &&
4475 subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100));
4477 if (VLOG_IS_DBG_ENABLED()) {
4481 ds_put_cstr(&s, "keep");
4482 for (i = 0; i < N_BUCKETS; i++) {
4484 ds_put_cstr(&s, ", drop");
4487 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
4490 VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s));
4494 return bucket * BUCKET_WIDTH;
4498 expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle)
4500 /* Cutoff time for most flows. */
4501 long long int normal_cutoff = time_msec() - dp_max_idle;
4503 /* We really want to keep flows for special protocols around, so use a more
4504 * conservative cutoff. */
4505 long long int special_cutoff = time_msec() - 10000;
4507 struct subfacet *subfacet, *next_subfacet;
4508 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
4512 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
4513 &ofproto->subfacets) {
4514 long long int cutoff;
4516 cutoff = (subfacet->facet->xout.slow & (SLOW_CFM | SLOW_LACP
4520 if (subfacet->used < cutoff) {
4521 if (subfacet->path != SF_NOT_INSTALLED) {
4522 batch[n_batch++] = subfacet;
4523 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
4524 subfacet_destroy_batch(ofproto, batch, n_batch);
4528 subfacet_destroy(subfacet);
4534 subfacet_destroy_batch(ofproto, batch, n_batch);
4538 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
4539 * then delete it entirely. */
4541 rule_expire(struct rule_dpif *rule)
4543 struct facet *facet, *next_facet;
4547 if (rule->up.pending) {
4548 /* We'll have to expire it later. */
4552 /* Has 'rule' expired? */
4554 if (rule->up.hard_timeout
4555 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
4556 reason = OFPRR_HARD_TIMEOUT;
4557 } else if (rule->up.idle_timeout
4558 && now > rule->up.used + rule->up.idle_timeout * 1000) {
4559 reason = OFPRR_IDLE_TIMEOUT;
4564 COVERAGE_INC(ofproto_dpif_expired);
4566 /* Update stats. (This is a no-op if the rule expired due to an idle
4567 * timeout, because that only happens when the rule has no facets left.) */
4568 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4569 facet_remove(facet);
4572 /* Get rid of the rule. */
4573 ofproto_rule_expire(&rule->up, reason);
4578 /* Creates and returns a new facet based on 'miss'.
4580 * The caller must already have determined that no facet with an identical
4581 * 'miss->flow' exists in 'miss->ofproto'.
4583 * 'hash' must be the return value of flow_hash(miss->flow, 0).
4585 * The facet will initially have no subfacets. The caller should create (at
4586 * least) one subfacet with subfacet_create(). */
4587 static struct facet *
4588 facet_create(const struct flow_miss *miss, uint32_t hash)
4590 struct ofproto_dpif *ofproto = miss->ofproto;
4591 struct xlate_in xin;
4592 struct facet *facet;
4594 facet = xzalloc(sizeof *facet);
4595 facet->used = time_msec();
4596 facet->flow = miss->flow;
4597 facet->initial_vals = miss->initial_vals;
4598 facet->rule = rule_dpif_lookup(ofproto, &facet->flow);
4599 facet->learn_rl = time_msec() + 500;
4601 hmap_insert(&ofproto->facets, &facet->hmap_node, hash);
4602 list_push_back(&facet->rule->facets, &facet->list_node);
4603 list_init(&facet->subfacets);
4604 netflow_flow_init(&facet->nf_flow);
4605 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
4607 xlate_in_init(&xin, ofproto, &facet->flow, &facet->initial_vals,
4608 facet->rule, 0, NULL);
4609 xin.may_learn = true;
4610 xlate_actions(&xin, &facet->xout);
4611 facet->nf_flow.output_iface = facet->xout.nf_output_iface;
4617 facet_free(struct facet *facet)
4620 xlate_out_uninit(&facet->xout);
4625 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
4626 * 'packet', which arrived on 'in_port'. */
4628 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
4629 const struct nlattr *odp_actions, size_t actions_len,
4630 struct ofpbuf *packet)
4632 struct odputil_keybuf keybuf;
4636 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4637 odp_flow_key_from_flow(&key, flow,
4638 ofp_port_to_odp_port(ofproto, flow->in_port));
4640 error = dpif_execute(ofproto->backer->dpif, key.data, key.size,
4641 odp_actions, actions_len, packet);
4645 /* Remove 'facet' from 'ofproto' and free up the associated memory:
4647 * - If 'facet' was installed in the datapath, uninstalls it and updates its
4648 * rule's statistics, via subfacet_uninstall().
4650 * - Removes 'facet' from its rule and from ofproto->facets.
4653 facet_remove(struct facet *facet)
4655 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4656 struct subfacet *subfacet, *next_subfacet;
4658 ovs_assert(!list_is_empty(&facet->subfacets));
4660 /* First uninstall all of the subfacets to get final statistics. */
4661 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4662 subfacet_uninstall(subfacet);
4665 /* Flush the final stats to the rule.
4667 * This might require us to have at least one subfacet around so that we
4668 * can use its actions for accounting in facet_account(), which is why we
4669 * have uninstalled but not yet destroyed the subfacets. */
4670 facet_flush_stats(facet);
4672 /* Now we're really all done so destroy everything. */
4673 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
4674 &facet->subfacets) {
4675 subfacet_destroy__(subfacet);
4677 hmap_remove(&ofproto->facets, &facet->hmap_node);
4678 list_remove(&facet->list_node);
4682 /* Feed information from 'facet' back into the learning table to keep it in
4683 * sync with what is actually flowing through the datapath. */
4685 facet_learn(struct facet *facet)
4687 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4688 long long int now = time_msec();
4689 struct xlate_in xin;
4691 if (!facet->xout.has_fin_timeout && now < facet->learn_rl) {
4695 facet->learn_rl = now + 500;
4697 if (!facet->xout.has_learn
4698 && !facet->xout.has_normal
4699 && (!facet->xout.has_fin_timeout
4700 || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
4704 xlate_in_init(&xin, ofproto, &facet->flow, &facet->initial_vals,
4705 facet->rule, facet->tcp_flags, NULL);
4706 xin.may_learn = true;
4707 xlate_actions_for_side_effects(&xin);
4711 facet_account(struct facet *facet)
4713 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4714 const struct nlattr *a;
4719 if (!facet->xout.has_normal || !ofproto->has_bonded_bundles) {
4722 n_bytes = facet->byte_count - facet->accounted_bytes;
4724 /* This loop feeds byte counters to bond_account() for rebalancing to use
4725 * as a basis. We also need to track the actual VLAN on which the packet
4726 * is going to be sent to ensure that it matches the one passed to
4727 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
4730 * We use the actions from an arbitrary subfacet because they should all
4731 * be equally valid for our purpose. */
4732 vlan_tci = facet->flow.vlan_tci;
4733 NL_ATTR_FOR_EACH_UNSAFE (a, left, facet->xout.odp_actions.data,
4734 facet->xout.odp_actions.size) {
4735 const struct ovs_action_push_vlan *vlan;
4736 struct ofport_dpif *port;
4738 switch (nl_attr_type(a)) {
4739 case OVS_ACTION_ATTR_OUTPUT:
4740 port = get_odp_port(ofproto, nl_attr_get_u32(a));
4741 if (port && port->bundle && port->bundle->bond) {
4742 bond_account(port->bundle->bond, &facet->flow,
4743 vlan_tci_to_vid(vlan_tci), n_bytes);
4747 case OVS_ACTION_ATTR_POP_VLAN:
4748 vlan_tci = htons(0);
4751 case OVS_ACTION_ATTR_PUSH_VLAN:
4752 vlan = nl_attr_get(a);
4753 vlan_tci = vlan->vlan_tci;
4759 /* Returns true if the only action for 'facet' is to send to the controller.
4760 * (We don't report NetFlow expiration messages for such facets because they
4761 * are just part of the control logic for the network, not real traffic). */
4763 facet_is_controller_flow(struct facet *facet)
4766 const struct rule *rule = &facet->rule->up;
4767 const struct ofpact *ofpacts = rule->ofpacts;
4768 size_t ofpacts_len = rule->ofpacts_len;
4770 if (ofpacts_len > 0 &&
4771 ofpacts->type == OFPACT_CONTROLLER &&
4772 ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len)) {
4779 /* Folds all of 'facet''s statistics into its rule. Also updates the
4780 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
4781 * 'facet''s statistics in the datapath should have been zeroed and folded into
4782 * its packet and byte counts before this function is called. */
4784 facet_flush_stats(struct facet *facet)
4786 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4787 struct subfacet *subfacet;
4789 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4790 ovs_assert(!subfacet->dp_byte_count);
4791 ovs_assert(!subfacet->dp_packet_count);
4794 facet_push_stats(facet);
4795 if (facet->accounted_bytes < facet->byte_count) {
4796 facet_account(facet);
4797 facet->accounted_bytes = facet->byte_count;
4800 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
4801 struct ofexpired expired;
4802 expired.flow = facet->flow;
4803 expired.packet_count = facet->packet_count;
4804 expired.byte_count = facet->byte_count;
4805 expired.used = facet->used;
4806 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4809 facet->rule->packet_count += facet->packet_count;
4810 facet->rule->byte_count += facet->byte_count;
4812 /* Reset counters to prevent double counting if 'facet' ever gets
4814 facet_reset_counters(facet);
4816 netflow_flow_clear(&facet->nf_flow);
4817 facet->tcp_flags = 0;
4820 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
4821 * Returns it if found, otherwise a null pointer.
4823 * 'hash' must be the return value of flow_hash(flow, 0).
4825 * The returned facet might need revalidation; use facet_lookup_valid()
4826 * instead if that is important. */
4827 static struct facet *
4828 facet_find(struct ofproto_dpif *ofproto,
4829 const struct flow *flow, uint32_t hash)
4831 struct facet *facet;
4833 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, hash, &ofproto->facets) {
4834 if (flow_equal(flow, &facet->flow)) {
4842 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
4843 * Returns it if found, otherwise a null pointer.
4845 * 'hash' must be the return value of flow_hash(flow, 0).
4847 * The returned facet is guaranteed to be valid. */
4848 static struct facet *
4849 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow,
4852 struct facet *facet;
4854 facet = facet_find(ofproto, flow, hash);
4856 && (ofproto->backer->need_revalidate
4857 || tag_set_intersects(&ofproto->backer->revalidate_set,
4859 && !facet_revalidate(facet)) {
4867 facet_check_consistency(struct facet *facet)
4869 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
4871 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4873 struct xlate_out xout;
4874 struct xlate_in xin;
4876 struct rule_dpif *rule;
4879 /* Check the rule for consistency. */
4880 rule = rule_dpif_lookup(ofproto, &facet->flow);
4881 if (rule != facet->rule) {
4882 if (!VLOG_DROP_WARN(&rl)) {
4883 struct ds s = DS_EMPTY_INITIALIZER;
4885 flow_format(&s, &facet->flow);
4886 ds_put_format(&s, ": facet associated with wrong rule (was "
4887 "table=%"PRIu8",", facet->rule->up.table_id);
4888 cls_rule_format(&facet->rule->up.cr, &s);
4889 ds_put_format(&s, ") (should have been table=%"PRIu8",",
4891 cls_rule_format(&rule->up.cr, &s);
4892 ds_put_cstr(&s, ")\n");
4899 /* Check the datapath actions for consistency. */
4900 xlate_in_init(&xin, ofproto, &facet->flow, &facet->initial_vals, rule,
4902 xlate_actions(&xin, &xout);
4904 ok = ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)
4905 && facet->xout.slow == xout.slow;
4906 if (!ok && !VLOG_DROP_WARN(&rl)) {
4907 struct ds s = DS_EMPTY_INITIALIZER;
4909 flow_format(&s, &facet->flow);
4910 ds_put_cstr(&s, ": inconsistency in facet");
4912 if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4913 ds_put_cstr(&s, " (actions were: ");
4914 format_odp_actions(&s, facet->xout.odp_actions.data,
4915 facet->xout.odp_actions.size);
4916 ds_put_cstr(&s, ") (correct actions: ");
4917 format_odp_actions(&s, xout.odp_actions.data,
4918 xout.odp_actions.size);
4919 ds_put_cstr(&s, ")");
4922 if (facet->xout.slow != xout.slow) {
4923 ds_put_format(&s, " slow path incorrect. should be %d", xout.slow);
4928 xlate_out_uninit(&xout);
4933 /* Re-searches the classifier for 'facet':
4935 * - If the rule found is different from 'facet''s current rule, moves
4936 * 'facet' to the new rule and recompiles its actions.
4938 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
4939 * where it is and recompiles its actions anyway.
4941 * - If any of 'facet''s subfacets correspond to a new flow according to
4942 * ofproto_receive(), 'facet' is removed.
4944 * Returns true if 'facet' is still valid. False if 'facet' was removed. */
4946 facet_revalidate(struct facet *facet)
4948 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4949 struct rule_dpif *new_rule;
4950 struct subfacet *subfacet;
4951 struct xlate_out xout;
4952 struct xlate_in xin;
4954 COVERAGE_INC(facet_revalidate);
4956 /* Check that child subfacets still correspond to this facet. Tunnel
4957 * configuration changes could cause a subfacet's OpenFlow in_port to
4959 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4960 struct ofproto_dpif *recv_ofproto;
4961 struct flow recv_flow;
4964 error = ofproto_receive(ofproto->backer, NULL, subfacet->key,
4965 subfacet->key_len, &recv_flow, NULL,
4966 &recv_ofproto, NULL, NULL);
4968 || recv_ofproto != ofproto
4969 || memcmp(&recv_flow, &facet->flow, sizeof recv_flow)) {
4970 facet_remove(facet);
4975 new_rule = rule_dpif_lookup(ofproto, &facet->flow);
4977 /* Calculate new datapath actions.
4979 * We do not modify any 'facet' state yet, because we might need to, e.g.,
4980 * emit a NetFlow expiration and, if so, we need to have the old state
4981 * around to properly compose it. */
4982 xlate_in_init(&xin, ofproto, &facet->flow, &facet->initial_vals, new_rule,
4984 xlate_actions(&xin, &xout);
4986 /* A facet's slow path reason should only change under dramatic
4987 * circumstances. Rather than try to update everything, it's simpler to
4988 * remove the facet and start over. */
4989 if (facet->xout.slow != xout.slow) {
4990 facet_remove(facet);
4991 xlate_out_uninit(&xout);
4995 if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4996 LIST_FOR_EACH(subfacet, list_node, &facet->subfacets) {
4997 if (subfacet->path == SF_FAST_PATH) {
4998 struct dpif_flow_stats stats;
5000 subfacet_install(subfacet, &xout.odp_actions, &stats);
5001 subfacet_update_stats(subfacet, &stats);
5005 facet_flush_stats(facet);
5007 ofpbuf_clear(&facet->xout.odp_actions);
5008 ofpbuf_put(&facet->xout.odp_actions, xout.odp_actions.data,
5009 xout.odp_actions.size);
5012 /* Update 'facet' now that we've taken care of all the old state. */
5013 facet->xout.tags = xout.tags;
5014 facet->xout.slow = xout.slow;
5015 facet->xout.has_learn = xout.has_learn;
5016 facet->xout.has_normal = xout.has_normal;
5017 facet->xout.has_fin_timeout = xout.has_fin_timeout;
5018 facet->xout.nf_output_iface = xout.nf_output_iface;
5019 facet->xout.mirrors = xout.mirrors;
5020 facet->nf_flow.output_iface = facet->xout.nf_output_iface;
5022 if (facet->rule != new_rule) {
5023 COVERAGE_INC(facet_changed_rule);
5024 list_remove(&facet->list_node);
5025 list_push_back(&new_rule->facets, &facet->list_node);
5026 facet->rule = new_rule;
5027 facet->used = new_rule->up.created;
5028 facet->prev_used = facet->used;
5031 xlate_out_uninit(&xout);
5035 /* Updates 'facet''s used time. Caller is responsible for calling
5036 * facet_push_stats() to update the flows which 'facet' resubmits into. */
5038 facet_update_time(struct facet *facet, long long int used)
5040 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5041 if (used > facet->used) {
5043 ofproto_rule_update_used(&facet->rule->up, used);
5044 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
5049 facet_reset_counters(struct facet *facet)
5051 facet->packet_count = 0;
5052 facet->byte_count = 0;
5053 facet->prev_packet_count = 0;
5054 facet->prev_byte_count = 0;
5055 facet->accounted_bytes = 0;
5059 facet_push_stats(struct facet *facet)
5061 struct dpif_flow_stats stats;
5063 ovs_assert(facet->packet_count >= facet->prev_packet_count);
5064 ovs_assert(facet->byte_count >= facet->prev_byte_count);
5065 ovs_assert(facet->used >= facet->prev_used);
5067 stats.n_packets = facet->packet_count - facet->prev_packet_count;
5068 stats.n_bytes = facet->byte_count - facet->prev_byte_count;
5069 stats.used = facet->used;
5070 stats.tcp_flags = 0;
5072 if (stats.n_packets || stats.n_bytes || facet->used > facet->prev_used) {
5073 facet->prev_packet_count = facet->packet_count;
5074 facet->prev_byte_count = facet->byte_count;
5075 facet->prev_used = facet->used;
5077 flow_push_stats(facet, &stats);
5079 update_mirror_stats(ofproto_dpif_cast(facet->rule->up.ofproto),
5080 facet->xout.mirrors, stats.n_packets,
5086 push_all_stats__(bool run_fast)
5088 static long long int rl = LLONG_MIN;
5089 struct ofproto_dpif *ofproto;
5091 if (time_msec() < rl) {
5095 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5096 struct facet *facet;
5098 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
5099 facet_push_stats(facet);
5106 rl = time_msec() + 100;
5110 push_all_stats(void)
5112 push_all_stats__(true);
5116 rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats)
5118 rule->packet_count += stats->n_packets;
5119 rule->byte_count += stats->n_bytes;
5120 ofproto_rule_update_used(&rule->up, stats->used);
5123 /* Pushes flow statistics to the rules which 'facet->flow' resubmits
5124 * into given 'facet->rule''s actions and mirrors. */
5126 flow_push_stats(struct facet *facet, const struct dpif_flow_stats *stats)
5128 struct rule_dpif *rule = facet->rule;
5129 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5130 struct xlate_in xin;
5132 ofproto_rule_update_used(&rule->up, stats->used);
5134 xlate_in_init(&xin, ofproto, &facet->flow, &facet->initial_vals, rule,
5136 xin.resubmit_stats = stats;
5137 xlate_actions_for_side_effects(&xin);
5142 static struct subfacet *
5143 subfacet_find(struct ofproto_dpif *ofproto,
5144 const struct nlattr *key, size_t key_len, uint32_t key_hash)
5146 struct subfacet *subfacet;
5148 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
5149 &ofproto->subfacets) {
5150 if (subfacet->key_len == key_len
5151 && !memcmp(key, subfacet->key, key_len)) {
5159 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
5160 * 'key_fitness', 'key', and 'key_len' members in 'miss'. Returns the
5161 * existing subfacet if there is one, otherwise creates and returns a
5163 static struct subfacet *
5164 subfacet_create(struct facet *facet, struct flow_miss *miss,
5167 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5168 enum odp_key_fitness key_fitness = miss->key_fitness;
5169 const struct nlattr *key = miss->key;
5170 size_t key_len = miss->key_len;
5172 struct subfacet *subfacet;
5174 key_hash = odp_flow_key_hash(key, key_len);
5176 if (list_is_empty(&facet->subfacets)) {
5177 subfacet = &facet->one_subfacet;
5179 subfacet = subfacet_find(ofproto, key, key_len, key_hash);
5181 if (subfacet->facet == facet) {
5185 /* This shouldn't happen. */
5186 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
5187 subfacet_destroy(subfacet);
5190 subfacet = xmalloc(sizeof *subfacet);
5193 hmap_insert(&ofproto->subfacets, &subfacet->hmap_node, key_hash);
5194 list_push_back(&facet->subfacets, &subfacet->list_node);
5195 subfacet->facet = facet;
5196 subfacet->key_fitness = key_fitness;
5197 subfacet->key = xmemdup(key, key_len);
5198 subfacet->key_len = key_len;
5199 subfacet->used = now;
5200 subfacet->created = now;
5201 subfacet->dp_packet_count = 0;
5202 subfacet->dp_byte_count = 0;
5203 subfacet->path = SF_NOT_INSTALLED;
5204 subfacet->odp_in_port = miss->odp_in_port;
5206 ofproto->subfacet_add_count++;
5210 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
5211 * its facet within 'ofproto', and frees it. */
5213 subfacet_destroy__(struct subfacet *subfacet)
5215 struct facet *facet = subfacet->facet;
5216 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5218 /* Update ofproto stats before uninstall the subfacet. */
5219 ofproto->subfacet_del_count++;
5220 ofproto->total_subfacet_life_span += (time_msec() - subfacet->created);
5222 subfacet_uninstall(subfacet);
5223 hmap_remove(&ofproto->subfacets, &subfacet->hmap_node);
5224 list_remove(&subfacet->list_node);
5225 free(subfacet->key);
5226 if (subfacet != &facet->one_subfacet) {
5231 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
5232 * last remaining subfacet in its facet destroys the facet too. */
5234 subfacet_destroy(struct subfacet *subfacet)
5236 struct facet *facet = subfacet->facet;
5238 if (list_is_singleton(&facet->subfacets)) {
5239 /* facet_remove() needs at least one subfacet (it will remove it). */
5240 facet_remove(facet);
5242 subfacet_destroy__(subfacet);
5247 subfacet_destroy_batch(struct ofproto_dpif *ofproto,
5248 struct subfacet **subfacets, int n)
5250 struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH];
5251 struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH];
5252 struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH];
5255 for (i = 0; i < n; i++) {
5256 ops[i].type = DPIF_OP_FLOW_DEL;
5257 ops[i].u.flow_del.key = subfacets[i]->key;
5258 ops[i].u.flow_del.key_len = subfacets[i]->key_len;
5259 ops[i].u.flow_del.stats = &stats[i];
5263 dpif_operate(ofproto->backer->dpif, opsp, n);
5264 for (i = 0; i < n; i++) {
5265 subfacet_reset_dp_stats(subfacets[i], &stats[i]);
5266 subfacets[i]->path = SF_NOT_INSTALLED;
5267 subfacet_destroy(subfacets[i]);
5272 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
5273 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
5274 * in the datapath will be zeroed and 'stats' will be updated with traffic new
5275 * since 'subfacet' was last updated.
5277 * Returns 0 if successful, otherwise a positive errno value. */
5279 subfacet_install(struct subfacet *subfacet, const struct ofpbuf *odp_actions,
5280 struct dpif_flow_stats *stats)
5282 struct facet *facet = subfacet->facet;
5283 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5284 enum subfacet_path path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
5285 const struct nlattr *actions = odp_actions->data;
5286 size_t actions_len = odp_actions->size;
5288 uint64_t slow_path_stub[128 / 8];
5289 enum dpif_flow_put_flags flags;
5292 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
5294 flags |= DPIF_FP_ZERO_STATS;
5297 if (path == SF_SLOW_PATH) {
5298 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
5299 slow_path_stub, sizeof slow_path_stub,
5300 &actions, &actions_len);
5303 ret = dpif_flow_put(ofproto->backer->dpif, flags, subfacet->key,
5304 subfacet->key_len, actions, actions_len, stats);
5307 subfacet_reset_dp_stats(subfacet, stats);
5311 subfacet->path = path;
5316 /* If 'subfacet' is installed in the datapath, uninstalls it. */
5318 subfacet_uninstall(struct subfacet *subfacet)
5320 if (subfacet->path != SF_NOT_INSTALLED) {
5321 struct rule_dpif *rule = subfacet->facet->rule;
5322 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5323 struct dpif_flow_stats stats;
5326 error = dpif_flow_del(ofproto->backer->dpif, subfacet->key,
5327 subfacet->key_len, &stats);
5328 subfacet_reset_dp_stats(subfacet, &stats);
5330 subfacet_update_stats(subfacet, &stats);
5332 subfacet->path = SF_NOT_INSTALLED;
5334 ovs_assert(subfacet->dp_packet_count == 0);
5335 ovs_assert(subfacet->dp_byte_count == 0);
5339 /* Resets 'subfacet''s datapath statistics counters. This should be called
5340 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
5341 * non-null, it should contain the statistics returned by dpif when 'subfacet'
5342 * was reset in the datapath. 'stats' will be modified to include only
5343 * statistics new since 'subfacet' was last updated. */
5345 subfacet_reset_dp_stats(struct subfacet *subfacet,
5346 struct dpif_flow_stats *stats)
5349 && subfacet->dp_packet_count <= stats->n_packets
5350 && subfacet->dp_byte_count <= stats->n_bytes) {
5351 stats->n_packets -= subfacet->dp_packet_count;
5352 stats->n_bytes -= subfacet->dp_byte_count;
5355 subfacet->dp_packet_count = 0;
5356 subfacet->dp_byte_count = 0;
5359 /* Updates 'subfacet''s used time. The caller is responsible for calling
5360 * facet_push_stats() to update the flows which 'subfacet' resubmits into. */
5362 subfacet_update_time(struct subfacet *subfacet, long long int used)
5364 if (used > subfacet->used) {
5365 subfacet->used = used;
5366 facet_update_time(subfacet->facet, used);
5370 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
5372 * Because of the meaning of a subfacet's counters, it only makes sense to do
5373 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
5374 * represents a packet that was sent by hand or if it represents statistics
5375 * that have been cleared out of the datapath. */
5377 subfacet_update_stats(struct subfacet *subfacet,
5378 const struct dpif_flow_stats *stats)
5380 if (stats->n_packets || stats->used > subfacet->used) {
5381 struct facet *facet = subfacet->facet;
5383 subfacet_update_time(subfacet, stats->used);
5384 facet->packet_count += stats->n_packets;
5385 facet->byte_count += stats->n_bytes;
5386 facet->tcp_flags |= stats->tcp_flags;
5387 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
5393 static struct rule_dpif *
5394 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow)
5396 struct rule_dpif *rule;
5398 rule = rule_dpif_lookup__(ofproto, flow, 0);
5403 return rule_dpif_miss_rule(ofproto, flow);
5406 static struct rule_dpif *
5407 rule_dpif_lookup__(struct ofproto_dpif *ofproto, const struct flow *flow,
5410 struct cls_rule *cls_rule;
5411 struct classifier *cls;
5413 if (table_id >= N_TABLES) {
5417 cls = &ofproto->up.tables[table_id].cls;
5418 if (flow->nw_frag & FLOW_NW_FRAG_ANY
5419 && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
5420 /* For OFPC_NORMAL frag_handling, we must pretend that transport ports
5421 * are unavailable. */
5422 struct flow ofpc_normal_flow = *flow;
5423 ofpc_normal_flow.tp_src = htons(0);
5424 ofpc_normal_flow.tp_dst = htons(0);
5425 cls_rule = classifier_lookup(cls, &ofpc_normal_flow);
5427 cls_rule = classifier_lookup(cls, flow);
5429 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
5432 static struct rule_dpif *
5433 rule_dpif_miss_rule(struct ofproto_dpif *ofproto, const struct flow *flow)
5435 struct ofport_dpif *port;
5437 port = get_ofp_port(ofproto, flow->in_port);
5439 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, flow->in_port);
5440 return ofproto->miss_rule;
5443 if (port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN) {
5444 return ofproto->no_packet_in_rule;
5446 return ofproto->miss_rule;
5450 complete_operation(struct rule_dpif *rule)
5452 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5454 rule_invalidate(rule);
5456 struct dpif_completion *c = xmalloc(sizeof *c);
5457 c->op = rule->up.pending;
5458 list_push_back(&ofproto->completions, &c->list_node);
5460 ofoperation_complete(rule->up.pending, 0);
5464 static struct rule *
5467 struct rule_dpif *rule = xmalloc(sizeof *rule);
5472 rule_dealloc(struct rule *rule_)
5474 struct rule_dpif *rule = rule_dpif_cast(rule_);
5479 rule_construct(struct rule *rule_)
5481 struct rule_dpif *rule = rule_dpif_cast(rule_);
5482 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5483 struct rule_dpif *victim;
5486 rule->packet_count = 0;
5487 rule->byte_count = 0;
5489 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
5490 if (victim && !list_is_empty(&victim->facets)) {
5491 struct facet *facet;
5493 rule->facets = victim->facets;
5494 list_moved(&rule->facets);
5495 LIST_FOR_EACH (facet, list_node, &rule->facets) {
5496 /* XXX: We're only clearing our local counters here. It's possible
5497 * that quite a few packets are unaccounted for in the datapath
5498 * statistics. These will be accounted to the new rule instead of
5499 * cleared as required. This could be fixed by clearing out the
5500 * datapath statistics for this facet, but currently it doesn't
5502 facet_reset_counters(facet);
5506 /* Must avoid list_moved() in this case. */
5507 list_init(&rule->facets);
5510 table_id = rule->up.table_id;
5512 rule->tag = victim->tag;
5513 } else if (table_id == 0) {
5518 miniflow_expand(&rule->up.cr.match.flow, &flow);
5519 rule->tag = rule_calculate_tag(&flow, &rule->up.cr.match.mask,
5520 ofproto->tables[table_id].basis);
5523 complete_operation(rule);
5528 rule_destruct(struct rule *rule_)
5530 struct rule_dpif *rule = rule_dpif_cast(rule_);
5531 struct facet *facet, *next_facet;
5533 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
5534 facet_revalidate(facet);
5537 complete_operation(rule);
5541 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
5543 struct rule_dpif *rule = rule_dpif_cast(rule_);
5544 struct facet *facet;
5546 /* push_all_stats() can handle flow misses which, when using the learn
5547 * action, can cause rules to be added and deleted. This can corrupt our
5548 * caller's datastructures which assume that rule_get_stats() doesn't have
5549 * an impact on the flow table. To be safe, we disable miss handling. */
5550 push_all_stats__(false);
5552 /* Start from historical data for 'rule' itself that are no longer tracked
5553 * in facets. This counts, for example, facets that have expired. */
5554 *packets = rule->packet_count;
5555 *bytes = rule->byte_count;
5557 /* Add any statistics that are tracked by facets. This includes
5558 * statistical data recently updated by ofproto_update_stats() as well as
5559 * stats for packets that were executed "by hand" via dpif_execute(). */
5560 LIST_FOR_EACH (facet, list_node, &rule->facets) {
5561 *packets += facet->packet_count;
5562 *bytes += facet->byte_count;
5567 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
5568 struct ofpbuf *packet)
5570 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5571 struct initial_vals initial_vals;
5572 struct dpif_flow_stats stats;
5573 struct xlate_out xout;
5574 struct xlate_in xin;
5576 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
5577 rule_credit_stats(rule, &stats);
5579 initial_vals.vlan_tci = flow->vlan_tci;
5580 initial_vals.tunnel_ip_tos = flow->tunnel.ip_tos;
5581 xlate_in_init(&xin, ofproto, flow, &initial_vals, rule, stats.tcp_flags,
5583 xin.resubmit_stats = &stats;
5584 xlate_actions(&xin, &xout);
5586 execute_odp_actions(ofproto, flow, xout.odp_actions.data,
5587 xout.odp_actions.size, packet);
5589 xlate_out_uninit(&xout);
5593 rule_execute(struct rule *rule, const struct flow *flow,
5594 struct ofpbuf *packet)
5596 rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
5597 ofpbuf_delete(packet);
5602 rule_modify_actions(struct rule *rule_)
5604 struct rule_dpif *rule = rule_dpif_cast(rule_);
5606 complete_operation(rule);
5609 /* Sends 'packet' out 'ofport'.
5610 * May modify 'packet'.
5611 * Returns 0 if successful, otherwise a positive errno value. */
5613 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
5615 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
5616 uint64_t odp_actions_stub[1024 / 8];
5617 struct ofpbuf key, odp_actions;
5618 struct odputil_keybuf keybuf;
5623 flow_extract(packet, 0, 0, NULL, OFPP_LOCAL, &flow);
5624 if (netdev_vport_is_patch(ofport->up.netdev)) {
5625 struct ofproto_dpif *peer_ofproto;
5626 struct dpif_flow_stats stats;
5627 struct ofport_dpif *peer;
5628 struct rule_dpif *rule;
5630 peer = ofport_get_peer(ofport);
5635 dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
5636 netdev_vport_inc_tx(ofport->up.netdev, &stats);
5637 netdev_vport_inc_rx(peer->up.netdev, &stats);
5639 flow.in_port = peer->up.ofp_port;
5640 peer_ofproto = ofproto_dpif_cast(peer->up.ofproto);
5641 rule = rule_dpif_lookup(peer_ofproto, &flow);
5642 rule_dpif_execute(rule, &flow, packet);
5647 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5649 if (ofport->tnl_port) {
5650 struct dpif_flow_stats stats;
5652 odp_port = tnl_port_send(ofport->tnl_port, &flow);
5653 if (odp_port == OVSP_NONE) {
5657 dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
5658 netdev_vport_inc_tx(ofport->up.netdev, &stats);
5659 odp_put_tunnel_action(&flow.tunnel, &odp_actions);
5660 odp_put_skb_mark_action(flow.skb_mark, &odp_actions);
5662 odp_port = vsp_realdev_to_vlandev(ofproto, ofport->odp_port,
5664 if (odp_port != ofport->odp_port) {
5665 eth_pop_vlan(packet);
5666 flow.vlan_tci = htons(0);
5670 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5671 odp_flow_key_from_flow(&key, &flow,
5672 ofp_port_to_odp_port(ofproto, flow.in_port));
5674 compose_sflow_action(ofproto, &odp_actions, &flow, odp_port);
5675 compose_ipfix_action(ofproto, &odp_actions, &flow);
5677 nl_msg_put_u32(&odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port);
5678 error = dpif_execute(ofproto->backer->dpif,
5680 odp_actions.data, odp_actions.size,
5682 ofpbuf_uninit(&odp_actions);
5685 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)",
5686 ofproto->up.name, odp_port, strerror(error));
5688 ofproto_update_local_port_stats(ofport->up.ofproto, packet->size, 0);
5692 /* OpenFlow to datapath action translation. */
5694 static bool may_receive(const struct ofport_dpif *, struct xlate_ctx *);
5695 static void do_xlate_actions(const struct ofpact *, size_t ofpacts_len,
5696 struct xlate_ctx *);
5697 static void xlate_normal(struct xlate_ctx *);
5699 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
5700 * The action will state 'slow' as the reason that the action is in the slow
5701 * path. (This is purely informational: it allows a human viewing "ovs-dpctl
5702 * dump-flows" output to see why a flow is in the slow path.)
5704 * The 'stub_size' bytes in 'stub' will be used to store the action.
5705 * 'stub_size' must be large enough for the action.
5707 * The action and its size will be stored in '*actionsp' and '*actions_lenp',
5710 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
5711 enum slow_path_reason slow,
5712 uint64_t *stub, size_t stub_size,
5713 const struct nlattr **actionsp, size_t *actions_lenp)
5715 union user_action_cookie cookie;
5718 cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
5719 cookie.slow_path.unused = 0;
5720 cookie.slow_path.reason = slow;
5722 ofpbuf_use_stack(&buf, stub, stub_size);
5723 if (slow & (SLOW_CFM | SLOW_LACP | SLOW_STP)) {
5724 uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif, UINT32_MAX);
5725 odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
5727 put_userspace_action(ofproto, &buf, flow, &cookie,
5728 sizeof cookie.slow_path);
5730 *actionsp = buf.data;
5731 *actions_lenp = buf.size;
5735 put_userspace_action(const struct ofproto_dpif *ofproto,
5736 struct ofpbuf *odp_actions,
5737 const struct flow *flow,
5738 const union user_action_cookie *cookie,
5739 const size_t cookie_size)
5743 pid = dpif_port_get_pid(ofproto->backer->dpif,
5744 ofp_port_to_odp_port(ofproto, flow->in_port));
5746 return odp_put_userspace_action(pid, cookie, cookie_size, odp_actions);
5749 /* Compose SAMPLE action for sFlow or IPFIX. The given probability is
5750 * the number of packets out of UINT32_MAX to sample. The given
5751 * cookie is passed back in the callback for each sampled packet.
5754 compose_sample_action(const struct ofproto_dpif *ofproto,
5755 struct ofpbuf *odp_actions,
5756 const struct flow *flow,
5757 const uint32_t probability,
5758 const union user_action_cookie *cookie,
5759 const size_t cookie_size)
5761 size_t sample_offset, actions_offset;
5764 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
5766 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
5768 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
5769 cookie_offset = put_userspace_action(ofproto, odp_actions, flow, cookie,
5772 nl_msg_end_nested(odp_actions, actions_offset);
5773 nl_msg_end_nested(odp_actions, sample_offset);
5774 return cookie_offset;
5778 compose_sflow_cookie(const struct ofproto_dpif *ofproto,
5779 ovs_be16 vlan_tci, uint32_t odp_port,
5780 unsigned int n_outputs, union user_action_cookie *cookie)
5784 cookie->type = USER_ACTION_COOKIE_SFLOW;
5785 cookie->sflow.vlan_tci = vlan_tci;
5787 /* See http://www.sflow.org/sflow_version_5.txt (search for "Input/output
5788 * port information") for the interpretation of cookie->output. */
5789 switch (n_outputs) {
5791 /* 0x40000000 | 256 means "packet dropped for unknown reason". */
5792 cookie->sflow.output = 0x40000000 | 256;
5796 ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port);
5798 cookie->sflow.output = ifindex;
5803 /* 0x80000000 means "multiple output ports. */
5804 cookie->sflow.output = 0x80000000 | n_outputs;
5809 /* Compose SAMPLE action for sFlow bridge sampling. */
5811 compose_sflow_action(const struct ofproto_dpif *ofproto,
5812 struct ofpbuf *odp_actions,
5813 const struct flow *flow,
5816 uint32_t probability;
5817 union user_action_cookie cookie;
5819 if (!ofproto->sflow || flow->in_port == OFPP_NONE) {
5823 probability = dpif_sflow_get_probability(ofproto->sflow);
5824 compose_sflow_cookie(ofproto, htons(0), odp_port,
5825 odp_port == OVSP_NONE ? 0 : 1, &cookie);
5827 return compose_sample_action(ofproto, odp_actions, flow, probability,
5828 &cookie, sizeof cookie.sflow);
5832 compose_flow_sample_cookie(uint16_t probability, uint32_t collector_set_id,
5833 uint32_t obs_domain_id, uint32_t obs_point_id,
5834 union user_action_cookie *cookie)
5836 cookie->type = USER_ACTION_COOKIE_FLOW_SAMPLE;
5837 cookie->flow_sample.probability = probability;
5838 cookie->flow_sample.collector_set_id = collector_set_id;
5839 cookie->flow_sample.obs_domain_id = obs_domain_id;
5840 cookie->flow_sample.obs_point_id = obs_point_id;
5844 compose_ipfix_cookie(union user_action_cookie *cookie)
5846 cookie->type = USER_ACTION_COOKIE_IPFIX;
5849 /* Compose SAMPLE action for IPFIX bridge sampling. */
5851 compose_ipfix_action(const struct ofproto_dpif *ofproto,
5852 struct ofpbuf *odp_actions,
5853 const struct flow *flow)
5855 uint32_t probability;
5856 union user_action_cookie cookie;
5858 if (!ofproto->ipfix || flow->in_port == OFPP_NONE) {
5862 probability = dpif_ipfix_get_bridge_exporter_probability(ofproto->ipfix);
5863 compose_ipfix_cookie(&cookie);
5865 compose_sample_action(ofproto, odp_actions, flow, probability,
5866 &cookie, sizeof cookie.ipfix);
5869 /* SAMPLE action for sFlow must be first action in any given list of
5870 * actions. At this point we do not have all information required to
5871 * build it. So try to build sample action as complete as possible. */
5873 add_sflow_action(struct xlate_ctx *ctx)
5875 ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto,
5876 &ctx->xout->odp_actions,
5877 &ctx->xin->flow, OVSP_NONE);
5878 ctx->sflow_odp_port = 0;
5879 ctx->sflow_n_outputs = 0;
5882 /* SAMPLE action for IPFIX must be 1st or 2nd action in any given list
5883 * of actions, eventually after the SAMPLE action for sFlow. */
5885 add_ipfix_action(struct xlate_ctx *ctx)
5887 compose_ipfix_action(ctx->ofproto, &ctx->xout->odp_actions,
5891 /* Fix SAMPLE action according to data collected while composing ODP actions.
5892 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
5893 * USERSPACE action's user-cookie which is required for sflow. */
5895 fix_sflow_action(struct xlate_ctx *ctx)
5897 const struct flow *base = &ctx->base_flow;
5898 union user_action_cookie *cookie;
5900 if (!ctx->user_cookie_offset) {
5904 cookie = ofpbuf_at(&ctx->xout->odp_actions, ctx->user_cookie_offset,
5905 sizeof cookie->sflow);
5906 ovs_assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
5908 compose_sflow_cookie(ctx->ofproto, base->vlan_tci,
5909 ctx->sflow_odp_port, ctx->sflow_n_outputs, cookie);
5913 compose_output_action__(struct xlate_ctx *ctx, uint16_t ofp_port,
5916 const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
5917 ovs_be16 flow_vlan_tci;
5918 uint32_t flow_skb_mark;
5919 uint8_t flow_nw_tos;
5920 struct priority_to_dscp *pdscp;
5921 uint32_t out_port, odp_port;
5923 /* If 'struct flow' gets additional metadata, we'll need to zero it out
5924 * before traversing a patch port. */
5925 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 20);
5928 xlate_report(ctx, "Nonexistent output port");
5930 } else if (ofport->up.pp.config & OFPUTIL_PC_NO_FWD) {
5931 xlate_report(ctx, "OFPPC_NO_FWD set, skipping output");
5933 } else if (check_stp && !stp_forward_in_state(ofport->stp_state)) {
5934 xlate_report(ctx, "STP not in forwarding state, skipping output");
5938 if (netdev_vport_is_patch(ofport->up.netdev)) {
5939 struct ofport_dpif *peer = ofport_get_peer(ofport);
5940 struct flow old_flow = ctx->xin->flow;
5941 const struct ofproto_dpif *peer_ofproto;
5942 enum slow_path_reason special;
5943 struct ofport_dpif *in_port;
5946 xlate_report(ctx, "Nonexistent patch port peer");
5950 peer_ofproto = ofproto_dpif_cast(peer->up.ofproto);
5951 if (peer_ofproto->backer != ctx->ofproto->backer) {
5952 xlate_report(ctx, "Patch port peer on a different datapath");
5956 ctx->ofproto = ofproto_dpif_cast(peer->up.ofproto);
5957 ctx->xin->flow.in_port = peer->up.ofp_port;
5958 ctx->xin->flow.metadata = htonll(0);
5959 memset(&ctx->xin->flow.tunnel, 0, sizeof ctx->xin->flow.tunnel);
5960 memset(ctx->xin->flow.regs, 0, sizeof ctx->xin->flow.regs);
5962 in_port = get_ofp_port(ctx->ofproto, ctx->xin->flow.in_port);
5963 special = process_special(ctx->ofproto, &ctx->xin->flow, in_port,
5966 ctx->xout->slow = special;
5967 } else if (!in_port || may_receive(in_port, ctx)) {
5968 if (!in_port || stp_forward_in_state(in_port->stp_state)) {
5969 xlate_table_action(ctx, ctx->xin->flow.in_port, 0, true);
5971 /* Forwarding is disabled by STP. Let OFPP_NORMAL and the
5972 * learning action look at the packet, then drop it. */
5973 struct flow old_base_flow = ctx->base_flow;
5974 size_t old_size = ctx->xout->odp_actions.size;
5975 xlate_table_action(ctx, ctx->xin->flow.in_port, 0, true);
5976 ctx->base_flow = old_base_flow;
5977 ctx->xout->odp_actions.size = old_size;
5981 ctx->xin->flow = old_flow;
5982 ctx->ofproto = ofproto_dpif_cast(ofport->up.ofproto);
5984 if (ctx->xin->resubmit_stats) {
5985 netdev_vport_inc_tx(ofport->up.netdev, ctx->xin->resubmit_stats);
5986 netdev_vport_inc_rx(peer->up.netdev, ctx->xin->resubmit_stats);
5992 flow_vlan_tci = ctx->xin->flow.vlan_tci;
5993 flow_skb_mark = ctx->xin->flow.skb_mark;
5994 flow_nw_tos = ctx->xin->flow.nw_tos;
5996 pdscp = get_priority(ofport, ctx->xin->flow.skb_priority);
5998 ctx->xin->flow.nw_tos &= ~IP_DSCP_MASK;
5999 ctx->xin->flow.nw_tos |= pdscp->dscp;
6002 if (ofport->tnl_port) {
6003 /* Save tunnel metadata so that changes made due to
6004 * the Logical (tunnel) Port are not visible for any further
6005 * matches, while explicit set actions on tunnel metadata are.
6007 struct flow_tnl flow_tnl = ctx->xin->flow.tunnel;
6008 odp_port = tnl_port_send(ofport->tnl_port, &ctx->xin->flow);
6009 if (odp_port == OVSP_NONE) {
6010 xlate_report(ctx, "Tunneling decided against output");
6011 goto out; /* restore flow_nw_tos */
6014 if (ctx->xin->resubmit_stats) {
6015 netdev_vport_inc_tx(ofport->up.netdev, ctx->xin->resubmit_stats);
6017 out_port = odp_port;
6018 commit_odp_tunnel_action(&ctx->xin->flow, &ctx->base_flow,
6019 &ctx->xout->odp_actions);
6020 ctx->xin->flow.tunnel = flow_tnl; /* Restore tunnel metadata */
6022 uint16_t vlandev_port;
6023 odp_port = ofport->odp_port;
6024 vlandev_port = vsp_realdev_to_vlandev(ctx->ofproto, ofp_port,
6025 ctx->xin->flow.vlan_tci);
6026 if (vlandev_port == ofp_port) {
6027 out_port = odp_port;
6029 out_port = ofp_port_to_odp_port(ctx->ofproto, vlandev_port);
6030 ctx->xin->flow.vlan_tci = htons(0);
6032 ctx->xin->flow.skb_mark &= ~IPSEC_MARK;
6034 commit_odp_actions(&ctx->xin->flow, &ctx->base_flow,
6035 &ctx->xout->odp_actions);
6036 nl_msg_put_u32(&ctx->xout->odp_actions, OVS_ACTION_ATTR_OUTPUT, out_port);
6038 ctx->sflow_odp_port = odp_port;
6039 ctx->sflow_n_outputs++;
6040 ctx->xout->nf_output_iface = ofp_port;
6043 ctx->xin->flow.vlan_tci = flow_vlan_tci;
6044 ctx->xin->flow.skb_mark = flow_skb_mark;
6046 ctx->xin->flow.nw_tos = flow_nw_tos;
6050 compose_output_action(struct xlate_ctx *ctx, uint16_t ofp_port)
6052 compose_output_action__(ctx, ofp_port, true);
6056 tag_the_flow(struct xlate_ctx *ctx, struct rule_dpif *rule)
6058 struct ofproto_dpif *ofproto = ctx->ofproto;
6059 uint8_t table_id = ctx->table_id;
6061 if (table_id > 0 && table_id < N_TABLES) {
6062 struct table_dpif *table = &ofproto->tables[table_id];
6063 if (table->other_table) {
6064 ctx->xout->tags |= (rule && rule->tag
6066 : rule_calculate_tag(&ctx->xin->flow,
6067 &table->other_table->mask,
6073 /* Common rule processing in one place to avoid duplicating code. */
6074 static struct rule_dpif *
6075 ctx_rule_hooks(struct xlate_ctx *ctx, struct rule_dpif *rule,
6078 if (ctx->xin->resubmit_hook) {
6079 ctx->xin->resubmit_hook(ctx, rule);
6081 if (rule == NULL && may_packet_in) {
6083 * check if table configuration flags
6084 * OFPTC_TABLE_MISS_CONTROLLER, default.
6085 * OFPTC_TABLE_MISS_CONTINUE,
6086 * OFPTC_TABLE_MISS_DROP
6087 * When OF1.0, OFPTC_TABLE_MISS_CONTINUE is used. What to do?
6089 rule = rule_dpif_miss_rule(ctx->ofproto, &ctx->xin->flow);
6091 if (rule && ctx->xin->resubmit_stats) {
6092 rule_credit_stats(rule, ctx->xin->resubmit_stats);
6098 xlate_table_action(struct xlate_ctx *ctx,
6099 uint16_t in_port, uint8_t table_id, bool may_packet_in)
6101 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
6102 struct rule_dpif *rule;
6103 uint16_t old_in_port = ctx->xin->flow.in_port;
6104 uint8_t old_table_id = ctx->table_id;
6106 ctx->table_id = table_id;
6108 /* Look up a flow with 'in_port' as the input port. */
6109 ctx->xin->flow.in_port = in_port;
6110 rule = rule_dpif_lookup__(ctx->ofproto, &ctx->xin->flow, table_id);
6112 tag_the_flow(ctx, rule);
6114 /* Restore the original input port. Otherwise OFPP_NORMAL and
6115 * OFPP_IN_PORT will have surprising behavior. */
6116 ctx->xin->flow.in_port = old_in_port;
6118 rule = ctx_rule_hooks(ctx, rule, may_packet_in);
6121 struct rule_dpif *old_rule = ctx->rule;
6125 do_xlate_actions(rule->up.ofpacts, rule->up.ofpacts_len, ctx);
6126 ctx->rule = old_rule;
6130 ctx->table_id = old_table_id;
6132 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
6134 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
6135 MAX_RESUBMIT_RECURSION);
6136 ctx->max_resubmit_trigger = true;
6141 xlate_ofpact_resubmit(struct xlate_ctx *ctx,
6142 const struct ofpact_resubmit *resubmit)
6147 in_port = resubmit->in_port;
6148 if (in_port == OFPP_IN_PORT) {
6149 in_port = ctx->xin->flow.in_port;
6152 table_id = resubmit->table_id;
6153 if (table_id == 255) {
6154 table_id = ctx->table_id;
6157 xlate_table_action(ctx, in_port, table_id, false);
6161 flood_packets(struct xlate_ctx *ctx, bool all)
6163 struct ofport_dpif *ofport;
6165 HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) {
6166 uint16_t ofp_port = ofport->up.ofp_port;
6168 if (ofp_port == ctx->xin->flow.in_port) {
6173 compose_output_action__(ctx, ofp_port, false);
6174 } else if (!(ofport->up.pp.config & OFPUTIL_PC_NO_FLOOD)) {
6175 compose_output_action(ctx, ofp_port);
6179 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
6183 execute_controller_action(struct xlate_ctx *ctx, int len,
6184 enum ofp_packet_in_reason reason,
6185 uint16_t controller_id)
6187 struct ofputil_packet_in pin;
6188 struct ofpbuf *packet;
6190 ovs_assert(!ctx->xout->slow || ctx->xout->slow == SLOW_CONTROLLER);
6191 ctx->xout->slow = SLOW_CONTROLLER;
6192 if (!ctx->xin->packet) {
6196 packet = ofpbuf_clone(ctx->xin->packet);
6198 if (packet->l2 && packet->l3) {
6199 struct eth_header *eh;
6200 uint16_t mpls_depth;
6202 eth_pop_vlan(packet);
6205 memcpy(eh->eth_src, ctx->xin->flow.dl_src, sizeof eh->eth_src);
6206 memcpy(eh->eth_dst, ctx->xin->flow.dl_dst, sizeof eh->eth_dst);
6208 if (ctx->xin->flow.vlan_tci & htons(VLAN_CFI)) {
6209 eth_push_vlan(packet, ctx->xin->flow.vlan_tci);
6212 mpls_depth = eth_mpls_depth(packet);
6214 if (mpls_depth < ctx->xin->flow.mpls_depth) {
6215 push_mpls(packet, ctx->xin->flow.dl_type, ctx->xin->flow.mpls_lse);
6216 } else if (mpls_depth > ctx->xin->flow.mpls_depth) {
6217 pop_mpls(packet, ctx->xin->flow.dl_type);
6218 } else if (mpls_depth) {
6219 set_mpls_lse(packet, ctx->xin->flow.mpls_lse);
6223 if (ctx->xin->flow.dl_type == htons(ETH_TYPE_IP)) {
6224 packet_set_ipv4(packet, ctx->xin->flow.nw_src,
6225 ctx->xin->flow.nw_dst, ctx->xin->flow.nw_tos,
6226 ctx->xin->flow.nw_ttl);
6230 if (ctx->xin->flow.nw_proto == IPPROTO_TCP) {
6231 packet_set_tcp_port(packet, ctx->xin->flow.tp_src,
6232 ctx->xin->flow.tp_dst);
6233 } else if (ctx->xin->flow.nw_proto == IPPROTO_UDP) {
6234 packet_set_udp_port(packet, ctx->xin->flow.tp_src,
6235 ctx->xin->flow.tp_dst);
6241 pin.packet = packet->data;
6242 pin.packet_len = packet->size;
6243 pin.reason = reason;
6244 pin.controller_id = controller_id;
6245 pin.table_id = ctx->table_id;
6246 pin.cookie = ctx->rule ? ctx->rule->up.flow_cookie : 0;
6249 flow_get_metadata(&ctx->xin->flow, &pin.fmd);
6251 connmgr_send_packet_in(ctx->ofproto->up.connmgr, &pin);
6252 ofpbuf_delete(packet);
6256 execute_mpls_push_action(struct xlate_ctx *ctx, ovs_be16 eth_type)
6258 ovs_assert(eth_type_mpls(eth_type));
6260 if (ctx->base_flow.mpls_depth) {
6261 ctx->xin->flow.mpls_lse &= ~htonl(MPLS_BOS_MASK);
6262 ctx->xin->flow.mpls_depth++;
6267 if (ctx->xin->flow.dl_type == htons(ETH_TYPE_IPV6)) {
6268 label = htonl(0x2); /* IPV6 Explicit Null. */
6270 label = htonl(0x0); /* IPV4 Explicit Null. */
6272 tc = (ctx->xin->flow.nw_tos & IP_DSCP_MASK) >> 2;
6273 ttl = ctx->xin->flow.nw_ttl ? ctx->xin->flow.nw_ttl : 0x40;
6274 ctx->xin->flow.mpls_lse = set_mpls_lse_values(ttl, tc, 1, label);
6275 ctx->xin->flow.mpls_depth = 1;
6277 ctx->xin->flow.dl_type = eth_type;
6281 execute_mpls_pop_action(struct xlate_ctx *ctx, ovs_be16 eth_type)
6283 ovs_assert(eth_type_mpls(ctx->xin->flow.dl_type));
6284 ovs_assert(!eth_type_mpls(eth_type));
6286 if (ctx->xin->flow.mpls_depth) {
6287 ctx->xin->flow.mpls_depth--;
6288 ctx->xin->flow.mpls_lse = htonl(0);
6289 if (!ctx->xin->flow.mpls_depth) {
6290 ctx->xin->flow.dl_type = eth_type;
6296 compose_dec_ttl(struct xlate_ctx *ctx, struct ofpact_cnt_ids *ids)
6298 if (ctx->xin->flow.dl_type != htons(ETH_TYPE_IP) &&
6299 ctx->xin->flow.dl_type != htons(ETH_TYPE_IPV6)) {
6303 if (ctx->xin->flow.nw_ttl > 1) {
6304 ctx->xin->flow.nw_ttl--;
6309 for (i = 0; i < ids->n_controllers; i++) {
6310 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL,
6314 /* Stop processing for current table. */
6320 execute_set_mpls_ttl_action(struct xlate_ctx *ctx, uint8_t ttl)
6322 if (!eth_type_mpls(ctx->xin->flow.dl_type)) {
6326 set_mpls_lse_ttl(&ctx->xin->flow.mpls_lse, ttl);
6331 execute_dec_mpls_ttl_action(struct xlate_ctx *ctx)
6333 uint8_t ttl = mpls_lse_to_ttl(ctx->xin->flow.mpls_lse);
6335 if (!eth_type_mpls(ctx->xin->flow.dl_type)) {
6341 set_mpls_lse_ttl(&ctx->xin->flow.mpls_lse, ttl);
6344 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL, 0);
6346 /* Stop processing for current table. */
6352 xlate_output_action(struct xlate_ctx *ctx,
6353 uint16_t port, uint16_t max_len, bool may_packet_in)
6355 uint16_t prev_nf_output_iface = ctx->xout->nf_output_iface;
6357 ctx->xout->nf_output_iface = NF_OUT_DROP;
6361 compose_output_action(ctx, ctx->xin->flow.in_port);
6364 xlate_table_action(ctx, ctx->xin->flow.in_port, 0, may_packet_in);
6370 flood_packets(ctx, false);
6373 flood_packets(ctx, true);
6375 case OFPP_CONTROLLER:
6376 execute_controller_action(ctx, max_len, OFPR_ACTION, 0);
6382 if (port != ctx->xin->flow.in_port) {
6383 compose_output_action(ctx, port);
6385 xlate_report(ctx, "skipping output to input port");
6390 if (prev_nf_output_iface == NF_OUT_FLOOD) {
6391 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
6392 } else if (ctx->xout->nf_output_iface == NF_OUT_DROP) {
6393 ctx->xout->nf_output_iface = prev_nf_output_iface;
6394 } else if (prev_nf_output_iface != NF_OUT_DROP &&
6395 ctx->xout->nf_output_iface != NF_OUT_FLOOD) {
6396 ctx->xout->nf_output_iface = NF_OUT_MULTI;
6401 xlate_output_reg_action(struct xlate_ctx *ctx,
6402 const struct ofpact_output_reg *or)
6404 uint64_t port = mf_get_subfield(&or->src, &ctx->xin->flow);
6405 if (port <= UINT16_MAX) {
6406 xlate_output_action(ctx, port, or->max_len, false);
6411 xlate_enqueue_action(struct xlate_ctx *ctx,
6412 const struct ofpact_enqueue *enqueue)
6414 uint16_t ofp_port = enqueue->port;
6415 uint32_t queue_id = enqueue->queue;
6416 uint32_t flow_priority, priority;
6419 /* Translate queue to priority. */
6420 error = dpif_queue_to_priority(ctx->ofproto->backer->dpif,
6421 queue_id, &priority);
6423 /* Fall back to ordinary output action. */
6424 xlate_output_action(ctx, enqueue->port, 0, false);
6428 /* Check output port. */
6429 if (ofp_port == OFPP_IN_PORT) {
6430 ofp_port = ctx->xin->flow.in_port;
6431 } else if (ofp_port == ctx->xin->flow.in_port) {
6435 /* Add datapath actions. */
6436 flow_priority = ctx->xin->flow.skb_priority;
6437 ctx->xin->flow.skb_priority = priority;
6438 compose_output_action(ctx, ofp_port);
6439 ctx->xin->flow.skb_priority = flow_priority;
6441 /* Update NetFlow output port. */
6442 if (ctx->xout->nf_output_iface == NF_OUT_DROP) {
6443 ctx->xout->nf_output_iface = ofp_port;
6444 } else if (ctx->xout->nf_output_iface != NF_OUT_FLOOD) {
6445 ctx->xout->nf_output_iface = NF_OUT_MULTI;
6450 xlate_set_queue_action(struct xlate_ctx *ctx, uint32_t queue_id)
6452 uint32_t skb_priority;
6454 if (!dpif_queue_to_priority(ctx->ofproto->backer->dpif,
6455 queue_id, &skb_priority)) {
6456 ctx->xin->flow.skb_priority = skb_priority;
6458 /* Couldn't translate queue to a priority. Nothing to do. A warning
6459 * has already been logged. */
6464 slave_enabled_cb(uint16_t ofp_port, void *ofproto_)
6466 struct ofproto_dpif *ofproto = ofproto_;
6467 struct ofport_dpif *port;
6477 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
6480 port = get_ofp_port(ofproto, ofp_port);
6481 return port ? port->may_enable : false;
6486 xlate_bundle_action(struct xlate_ctx *ctx,
6487 const struct ofpact_bundle *bundle)
6491 port = bundle_execute(bundle, &ctx->xin->flow, slave_enabled_cb,
6493 if (bundle->dst.field) {
6494 nxm_reg_load(&bundle->dst, port, &ctx->xin->flow);
6496 xlate_output_action(ctx, port, 0, false);
6501 xlate_learn_action(struct xlate_ctx *ctx,
6502 const struct ofpact_learn *learn)
6504 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
6505 struct ofputil_flow_mod fm;
6506 uint64_t ofpacts_stub[1024 / 8];
6507 struct ofpbuf ofpacts;
6510 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
6511 learn_execute(learn, &ctx->xin->flow, &fm, &ofpacts);
6513 error = ofproto_flow_mod(&ctx->ofproto->up, &fm);
6514 if (error && !VLOG_DROP_WARN(&rl)) {
6515 VLOG_WARN("learning action failed to modify flow table (%s)",
6516 ofperr_get_name(error));
6519 ofpbuf_uninit(&ofpacts);
6522 /* Reduces '*timeout' to no more than 'max'. A value of zero in either case
6523 * means "infinite". */
6525 reduce_timeout(uint16_t max, uint16_t *timeout)
6527 if (max && (!*timeout || *timeout > max)) {
6533 xlate_fin_timeout(struct xlate_ctx *ctx,
6534 const struct ofpact_fin_timeout *oft)
6536 if (ctx->xin->tcp_flags & (TCP_FIN | TCP_RST) && ctx->rule) {
6537 struct rule_dpif *rule = ctx->rule;
6539 reduce_timeout(oft->fin_idle_timeout, &rule->up.idle_timeout);
6540 reduce_timeout(oft->fin_hard_timeout, &rule->up.hard_timeout);
6545 xlate_sample_action(struct xlate_ctx *ctx,
6546 const struct ofpact_sample *os)
6548 union user_action_cookie cookie;
6549 /* Scale the probability from 16-bit to 32-bit while representing
6550 * the same percentage. */
6551 uint32_t probability = (os->probability << 16) | os->probability;
6553 commit_odp_actions(&ctx->xin->flow, &ctx->base_flow,
6554 &ctx->xout->odp_actions);
6556 compose_flow_sample_cookie(os->probability, os->collector_set_id,
6557 os->obs_domain_id, os->obs_point_id, &cookie);
6558 compose_sample_action(ctx->ofproto, &ctx->xout->odp_actions, &ctx->xin->flow,
6559 probability, &cookie, sizeof cookie.flow_sample);
6563 may_receive(const struct ofport_dpif *port, struct xlate_ctx *ctx)
6565 if (port->up.pp.config & (eth_addr_equals(ctx->xin->flow.dl_dst,
6567 ? OFPUTIL_PC_NO_RECV_STP
6568 : OFPUTIL_PC_NO_RECV)) {
6572 /* Only drop packets here if both forwarding and learning are
6573 * disabled. If just learning is enabled, we need to have
6574 * OFPP_NORMAL and the learning action have a look at the packet
6575 * before we can drop it. */
6576 if (!stp_forward_in_state(port->stp_state)
6577 && !stp_learn_in_state(port->stp_state)) {
6585 tunnel_ecn_ok(struct xlate_ctx *ctx)
6587 if (is_ip_any(&ctx->base_flow)
6588 && (ctx->base_flow.tunnel.ip_tos & IP_ECN_MASK) == IP_ECN_CE) {
6589 if ((ctx->base_flow.nw_tos & IP_ECN_MASK) == IP_ECN_NOT_ECT) {
6590 VLOG_WARN_RL(&rl, "dropping tunnel packet marked ECN CE"
6591 " but is not ECN capable");
6594 /* Set the ECN CE value in the tunneled packet. */
6595 ctx->xin->flow.nw_tos |= IP_ECN_CE;
6603 do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len,
6604 struct xlate_ctx *ctx)
6606 bool was_evictable = true;
6607 const struct ofpact *a;
6610 /* Don't let the rule we're working on get evicted underneath us. */
6611 was_evictable = ctx->rule->up.evictable;
6612 ctx->rule->up.evictable = false;
6615 do_xlate_actions_again:
6616 OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) {
6617 struct ofpact_controller *controller;
6618 const struct ofpact_metadata *metadata;
6626 xlate_output_action(ctx, ofpact_get_OUTPUT(a)->port,
6627 ofpact_get_OUTPUT(a)->max_len, true);
6630 case OFPACT_CONTROLLER:
6631 controller = ofpact_get_CONTROLLER(a);
6632 execute_controller_action(ctx, controller->max_len,
6634 controller->controller_id);
6637 case OFPACT_ENQUEUE:
6638 xlate_enqueue_action(ctx, ofpact_get_ENQUEUE(a));
6641 case OFPACT_SET_VLAN_VID:
6642 ctx->xin->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
6643 ctx->xin->flow.vlan_tci |=
6644 (htons(ofpact_get_SET_VLAN_VID(a)->vlan_vid)
6648 case OFPACT_SET_VLAN_PCP:
6649 ctx->xin->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
6650 ctx->xin->flow.vlan_tci |=
6651 htons((ofpact_get_SET_VLAN_PCP(a)->vlan_pcp << VLAN_PCP_SHIFT)
6655 case OFPACT_STRIP_VLAN:
6656 ctx->xin->flow.vlan_tci = htons(0);
6659 case OFPACT_PUSH_VLAN:
6660 /* XXX 802.1AD(QinQ) */
6661 ctx->xin->flow.vlan_tci = htons(VLAN_CFI);
6664 case OFPACT_SET_ETH_SRC:
6665 memcpy(ctx->xin->flow.dl_src, ofpact_get_SET_ETH_SRC(a)->mac,
6669 case OFPACT_SET_ETH_DST:
6670 memcpy(ctx->xin->flow.dl_dst, ofpact_get_SET_ETH_DST(a)->mac,
6674 case OFPACT_SET_IPV4_SRC:
6675 if (ctx->xin->flow.dl_type == htons(ETH_TYPE_IP)) {
6676 ctx->xin->flow.nw_src = ofpact_get_SET_IPV4_SRC(a)->ipv4;
6680 case OFPACT_SET_IPV4_DST:
6681 if (ctx->xin->flow.dl_type == htons(ETH_TYPE_IP)) {
6682 ctx->xin->flow.nw_dst = ofpact_get_SET_IPV4_DST(a)->ipv4;
6686 case OFPACT_SET_IPV4_DSCP:
6687 /* OpenFlow 1.0 only supports IPv4. */
6688 if (ctx->xin->flow.dl_type == htons(ETH_TYPE_IP)) {
6689 ctx->xin->flow.nw_tos &= ~IP_DSCP_MASK;
6690 ctx->xin->flow.nw_tos |= ofpact_get_SET_IPV4_DSCP(a)->dscp;
6694 case OFPACT_SET_L4_SRC_PORT:
6695 if (is_ip_any(&ctx->xin->flow)) {
6696 ctx->xin->flow.tp_src =
6697 htons(ofpact_get_SET_L4_SRC_PORT(a)->port);
6701 case OFPACT_SET_L4_DST_PORT:
6702 if (is_ip_any(&ctx->xin->flow)) {
6703 ctx->xin->flow.tp_dst =
6704 htons(ofpact_get_SET_L4_DST_PORT(a)->port);
6708 case OFPACT_RESUBMIT:
6709 xlate_ofpact_resubmit(ctx, ofpact_get_RESUBMIT(a));
6712 case OFPACT_SET_TUNNEL:
6713 ctx->xin->flow.tunnel.tun_id =
6714 htonll(ofpact_get_SET_TUNNEL(a)->tun_id);
6717 case OFPACT_SET_QUEUE:
6718 xlate_set_queue_action(ctx, ofpact_get_SET_QUEUE(a)->queue_id);
6721 case OFPACT_POP_QUEUE:
6722 ctx->xin->flow.skb_priority = ctx->orig_skb_priority;
6725 case OFPACT_REG_MOVE:
6726 nxm_execute_reg_move(ofpact_get_REG_MOVE(a), &ctx->xin->flow);
6729 case OFPACT_REG_LOAD:
6730 nxm_execute_reg_load(ofpact_get_REG_LOAD(a), &ctx->xin->flow);
6733 case OFPACT_STACK_PUSH:
6734 nxm_execute_stack_push(ofpact_get_STACK_PUSH(a), &ctx->xin->flow,
6738 case OFPACT_STACK_POP:
6739 nxm_execute_stack_pop(ofpact_get_STACK_POP(a), &ctx->xin->flow,
6743 case OFPACT_PUSH_MPLS:
6744 execute_mpls_push_action(ctx, ofpact_get_PUSH_MPLS(a)->ethertype);
6747 case OFPACT_POP_MPLS:
6748 execute_mpls_pop_action(ctx, ofpact_get_POP_MPLS(a)->ethertype);
6751 case OFPACT_SET_MPLS_TTL:
6752 if (execute_set_mpls_ttl_action(ctx,
6753 ofpact_get_SET_MPLS_TTL(a)->ttl)) {
6758 case OFPACT_DEC_MPLS_TTL:
6759 if (execute_dec_mpls_ttl_action(ctx)) {
6764 case OFPACT_DEC_TTL:
6765 if (compose_dec_ttl(ctx, ofpact_get_DEC_TTL(a))) {
6771 /* Nothing to do. */
6774 case OFPACT_MULTIPATH:
6775 multipath_execute(ofpact_get_MULTIPATH(a), &ctx->xin->flow);
6779 ctx->ofproto->has_bundle_action = true;
6780 xlate_bundle_action(ctx, ofpact_get_BUNDLE(a));
6783 case OFPACT_OUTPUT_REG:
6784 xlate_output_reg_action(ctx, ofpact_get_OUTPUT_REG(a));
6788 ctx->xout->has_learn = true;
6789 if (ctx->xin->may_learn) {
6790 xlate_learn_action(ctx, ofpact_get_LEARN(a));
6798 case OFPACT_FIN_TIMEOUT:
6799 ctx->xout->has_fin_timeout = true;
6800 xlate_fin_timeout(ctx, ofpact_get_FIN_TIMEOUT(a));
6803 case OFPACT_CLEAR_ACTIONS:
6805 * Nothing to do because writa-actions is not supported for now.
6806 * When writa-actions is supported, clear-actions also must
6807 * be supported at the same time.
6811 case OFPACT_WRITE_METADATA:
6812 metadata = ofpact_get_WRITE_METADATA(a);
6813 ctx->xin->flow.metadata &= ~metadata->mask;
6814 ctx->xin->flow.metadata |= metadata->metadata & metadata->mask;
6817 case OFPACT_GOTO_TABLE: {
6818 /* It is assumed that goto-table is the last action. */
6819 struct ofpact_goto_table *ogt = ofpact_get_GOTO_TABLE(a);
6820 struct rule_dpif *rule;
6822 ovs_assert(ctx->table_id < ogt->table_id);
6824 ctx->table_id = ogt->table_id;
6826 /* Look up a flow from the new table. */
6827 rule = rule_dpif_lookup__(ctx->ofproto, &ctx->xin->flow, ctx->table_id);
6829 tag_the_flow(ctx, rule);
6831 rule = ctx_rule_hooks(ctx, rule, true);
6835 ctx->rule->up.evictable = was_evictable;
6838 was_evictable = rule->up.evictable;
6839 rule->up.evictable = false;
6841 /* Tail recursion removal. */
6842 ofpacts = rule->up.ofpacts;
6843 ofpacts_len = rule->up.ofpacts_len;
6844 goto do_xlate_actions_again;
6850 xlate_sample_action(ctx, ofpact_get_SAMPLE(a));
6857 ctx->rule->up.evictable = was_evictable;
6862 xlate_in_init(struct xlate_in *xin, struct ofproto_dpif *ofproto,
6863 const struct flow *flow,
6864 const struct initial_vals *initial_vals,
6865 struct rule_dpif *rule, uint8_t tcp_flags,
6866 const struct ofpbuf *packet)
6868 xin->ofproto = ofproto;
6870 xin->packet = packet;
6871 xin->may_learn = packet != NULL;
6873 xin->ofpacts = NULL;
6874 xin->ofpacts_len = 0;
6875 xin->tcp_flags = tcp_flags;
6876 xin->resubmit_hook = NULL;
6877 xin->report_hook = NULL;
6878 xin->resubmit_stats = NULL;
6881 xin->initial_vals = *initial_vals;
6883 xin->initial_vals.vlan_tci = xin->flow.vlan_tci;
6884 xin->initial_vals.tunnel_ip_tos = xin->flow.tunnel.ip_tos;
6889 xlate_out_uninit(struct xlate_out *xout)
6892 ofpbuf_uninit(&xout->odp_actions);
6896 /* Translates the 'ofpacts_len' bytes of "struct ofpacts" starting at 'ofpacts'
6897 * into datapath actions in 'odp_actions', using 'ctx'. */
6899 xlate_actions(struct xlate_in *xin, struct xlate_out *xout)
6901 /* Normally false. Set to true if we ever hit MAX_RESUBMIT_RECURSION, so
6902 * that in the future we always keep a copy of the original flow for
6903 * tracing purposes. */
6904 static bool hit_resubmit_limit;
6906 enum slow_path_reason special;
6907 const struct ofpact *ofpacts;
6908 struct ofport_dpif *in_port;
6909 struct flow orig_flow;
6910 struct xlate_ctx ctx;
6913 COVERAGE_INC(ofproto_dpif_xlate);
6915 /* Flow initialization rules:
6916 * - 'base_flow' must match the kernel's view of the packet at the
6917 * time that action processing starts. 'flow' represents any
6918 * transformations we wish to make through actions.
6919 * - By default 'base_flow' and 'flow' are the same since the input
6920 * packet matches the output before any actions are applied.
6921 * - When using VLAN splinters, 'base_flow''s VLAN is set to the value
6922 * of the received packet as seen by the kernel. If we later output
6923 * to another device without any modifications this will cause us to
6924 * insert a new tag since the original one was stripped off by the
6926 * - Tunnel 'flow' is largely cleared when transitioning between
6927 * the input and output stages since it does not make sense to output
6928 * a packet with the exact headers that it was received with (i.e.
6929 * the destination IP is us). The one exception is the tun_id, which
6930 * is preserved to allow use in later resubmit lookups and loads into
6932 * - Tunnel 'base_flow' is completely cleared since that is what the
6933 * kernel does. If we wish to maintain the original values an action
6934 * needs to be generated. */
6939 ctx.ofproto = xin->ofproto;
6940 ctx.rule = xin->rule;
6942 ctx.base_flow = ctx.xin->flow;
6943 memset(&ctx.base_flow.tunnel, 0, sizeof ctx.base_flow.tunnel);
6944 ctx.base_flow.vlan_tci = xin->initial_vals.vlan_tci;
6945 ctx.base_flow.tunnel.ip_tos = xin->initial_vals.tunnel_ip_tos;
6949 ctx.xout->has_learn = false;
6950 ctx.xout->has_normal = false;
6951 ctx.xout->has_fin_timeout = false;
6952 ctx.xout->nf_output_iface = NF_OUT_DROP;
6953 ctx.xout->mirrors = 0;
6955 ofpbuf_use_stub(&ctx.xout->odp_actions, ctx.xout->odp_actions_stub,
6956 sizeof ctx.xout->odp_actions_stub);
6957 ofpbuf_reserve(&ctx.xout->odp_actions, NL_A_U32_SIZE);
6960 ctx.max_resubmit_trigger = false;
6961 ctx.orig_skb_priority = ctx.xin->flow.skb_priority;
6966 ofpacts = xin->ofpacts;
6967 ofpacts_len = xin->ofpacts_len;
6968 } else if (xin->rule) {
6969 ofpacts = xin->rule->up.ofpacts;
6970 ofpacts_len = xin->rule->up.ofpacts_len;
6975 ofpbuf_use_stub(&ctx.stack, ctx.init_stack, sizeof ctx.init_stack);
6977 if (ctx.ofproto->has_mirrors || hit_resubmit_limit) {
6978 /* Do this conditionally because the copy is expensive enough that it
6979 * shows up in profiles. */
6980 orig_flow = ctx.xin->flow;
6983 if (ctx.xin->flow.nw_frag & FLOW_NW_FRAG_ANY) {
6984 switch (ctx.ofproto->up.frag_handling) {
6985 case OFPC_FRAG_NORMAL:
6986 /* We must pretend that transport ports are unavailable. */
6987 ctx.xin->flow.tp_src = ctx.base_flow.tp_src = htons(0);
6988 ctx.xin->flow.tp_dst = ctx.base_flow.tp_dst = htons(0);
6991 case OFPC_FRAG_DROP:
6994 case OFPC_FRAG_REASM:
6997 case OFPC_FRAG_NX_MATCH:
6998 /* Nothing to do. */
7001 case OFPC_INVALID_TTL_TO_CONTROLLER:
7006 in_port = get_ofp_port(ctx.ofproto, ctx.xin->flow.in_port);
7007 special = process_special(ctx.ofproto, &ctx.xin->flow, in_port,
7010 ctx.xout->slow = special;
7012 static struct vlog_rate_limit trace_rl = VLOG_RATE_LIMIT_INIT(1, 1);
7013 struct initial_vals initial_vals;
7014 size_t sample_actions_len;
7015 uint32_t local_odp_port;
7017 initial_vals.vlan_tci = ctx.base_flow.vlan_tci;
7018 initial_vals.tunnel_ip_tos = ctx.base_flow.tunnel.ip_tos;
7020 add_sflow_action(&ctx);
7021 add_ipfix_action(&ctx);
7022 sample_actions_len = ctx.xout->odp_actions.size;
7024 if (tunnel_ecn_ok(&ctx) && (!in_port || may_receive(in_port, &ctx))) {
7025 do_xlate_actions(ofpacts, ofpacts_len, &ctx);
7027 /* We've let OFPP_NORMAL and the learning action look at the
7028 * packet, so drop it now if forwarding is disabled. */
7029 if (in_port && !stp_forward_in_state(in_port->stp_state)) {
7030 ctx.xout->odp_actions.size = sample_actions_len;
7034 if (ctx.max_resubmit_trigger && !ctx.xin->resubmit_hook) {
7035 if (!hit_resubmit_limit) {
7036 /* We didn't record the original flow. Make sure we do from
7038 hit_resubmit_limit = true;
7039 } else if (!VLOG_DROP_ERR(&trace_rl)) {
7040 struct ds ds = DS_EMPTY_INITIALIZER;
7042 ofproto_trace(ctx.ofproto, &orig_flow, ctx.xin->packet,
7043 &initial_vals, &ds);
7044 VLOG_ERR("Trace triggered by excessive resubmit "
7045 "recursion:\n%s", ds_cstr(&ds));
7050 local_odp_port = ofp_port_to_odp_port(ctx.ofproto, OFPP_LOCAL);
7051 if (!connmgr_must_output_local(ctx.ofproto->up.connmgr, &ctx.xin->flow,
7053 ctx.xout->odp_actions.data,
7054 ctx.xout->odp_actions.size)) {
7055 compose_output_action(&ctx, OFPP_LOCAL);
7057 if (ctx.ofproto->has_mirrors) {
7058 add_mirror_actions(&ctx, &orig_flow);
7060 fix_sflow_action(&ctx);
7063 ofpbuf_uninit(&ctx.stack);
7066 /* Translates the 'ofpacts_len' bytes of "struct ofpact"s starting at 'ofpacts'
7067 * into datapath actions, using 'ctx', and discards the datapath actions. */
7069 xlate_actions_for_side_effects(struct xlate_in *xin)
7071 struct xlate_out xout;
7073 xlate_actions(xin, &xout);
7074 xlate_out_uninit(&xout);
7078 xlate_report(struct xlate_ctx *ctx, const char *s)
7080 if (ctx->xin->report_hook) {
7081 ctx->xin->report_hook(ctx, s);
7085 /* OFPP_NORMAL implementation. */
7087 static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *);
7089 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
7090 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle',
7091 * the bundle on which the packet was received, returns the VLAN to which the
7094 * Both 'vid' and the return value are in the range 0...4095. */
7096 input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid)
7098 switch (in_bundle->vlan_mode) {
7099 case PORT_VLAN_ACCESS:
7100 return in_bundle->vlan;
7103 case PORT_VLAN_TRUNK:
7106 case PORT_VLAN_NATIVE_UNTAGGED:
7107 case PORT_VLAN_NATIVE_TAGGED:
7108 return vid ? vid : in_bundle->vlan;
7115 /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'.
7116 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
7119 * 'vid' should be the VID obtained from the 802.1Q header that was received as
7120 * part of a packet (specify 0 if there was no 802.1Q header), in the range
7123 input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn)
7125 /* Allow any VID on the OFPP_NONE port. */
7126 if (in_bundle == &ofpp_none_bundle) {
7130 switch (in_bundle->vlan_mode) {
7131 case PORT_VLAN_ACCESS:
7134 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7135 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
7136 "packet received on port %s configured as VLAN "
7137 "%"PRIu16" access port",
7138 in_bundle->ofproto->up.name, vid,
7139 in_bundle->name, in_bundle->vlan);
7145 case PORT_VLAN_NATIVE_UNTAGGED:
7146 case PORT_VLAN_NATIVE_TAGGED:
7148 /* Port must always carry its native VLAN. */
7152 case PORT_VLAN_TRUNK:
7153 if (!ofbundle_includes_vlan(in_bundle, vid)) {
7155 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7156 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet "
7157 "received on port %s not configured for trunking "
7159 in_bundle->ofproto->up.name, vid,
7160 in_bundle->name, vid);
7172 /* Given 'vlan', the VLAN that a packet belongs to, and
7173 * 'out_bundle', a bundle on which the packet is to be output, returns the VID
7174 * that should be included in the 802.1Q header. (If the return value is 0,
7175 * then the 802.1Q header should only be included in the packet if there is a
7178 * Both 'vlan' and the return value are in the range 0...4095. */
7180 output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan)
7182 switch (out_bundle->vlan_mode) {
7183 case PORT_VLAN_ACCESS:
7186 case PORT_VLAN_TRUNK:
7187 case PORT_VLAN_NATIVE_TAGGED:
7190 case PORT_VLAN_NATIVE_UNTAGGED:
7191 return vlan == out_bundle->vlan ? 0 : vlan;
7199 output_normal(struct xlate_ctx *ctx, const struct ofbundle *out_bundle,
7202 struct ofport_dpif *port;
7204 ovs_be16 tci, old_tci;
7206 vid = output_vlan_to_vid(out_bundle, vlan);
7207 if (!out_bundle->bond) {
7208 port = ofbundle_get_a_port(out_bundle);
7210 port = bond_choose_output_slave(out_bundle->bond, &ctx->xin->flow,
7211 vid, &ctx->xout->tags);
7213 /* No slaves enabled, so drop packet. */
7218 old_tci = ctx->xin->flow.vlan_tci;
7220 if (tci || out_bundle->use_priority_tags) {
7221 tci |= ctx->xin->flow.vlan_tci & htons(VLAN_PCP_MASK);
7223 tci |= htons(VLAN_CFI);
7226 ctx->xin->flow.vlan_tci = tci;
7228 compose_output_action(ctx, port->up.ofp_port);
7229 ctx->xin->flow.vlan_tci = old_tci;
7233 mirror_mask_ffs(mirror_mask_t mask)
7235 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
7240 ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
7242 return (bundle->vlan_mode != PORT_VLAN_ACCESS
7243 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
7247 ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan)
7249 return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan);
7252 /* Returns an arbitrary interface within 'bundle'. */
7253 static struct ofport_dpif *
7254 ofbundle_get_a_port(const struct ofbundle *bundle)
7256 return CONTAINER_OF(list_front(&bundle->ports),
7257 struct ofport_dpif, bundle_node);
7261 vlan_is_mirrored(const struct ofmirror *m, int vlan)
7263 return !m->vlans || bitmap_is_set(m->vlans, vlan);
7267 add_mirror_actions(struct xlate_ctx *ctx, const struct flow *orig_flow)
7269 struct ofproto_dpif *ofproto = ctx->ofproto;
7270 mirror_mask_t mirrors;
7271 struct ofbundle *in_bundle;
7274 const struct nlattr *a;
7277 in_bundle = lookup_input_bundle(ctx->ofproto, orig_flow->in_port,
7278 ctx->xin->packet != NULL, NULL);
7282 mirrors = in_bundle->src_mirrors;
7284 /* Drop frames on bundles reserved for mirroring. */
7285 if (in_bundle->mirror_out) {
7286 if (ctx->xin->packet != NULL) {
7287 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7288 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
7289 "%s, which is reserved exclusively for mirroring",
7290 ctx->ofproto->up.name, in_bundle->name);
7296 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
7297 if (!input_vid_is_valid(vid, in_bundle, ctx->xin->packet != NULL)) {
7300 vlan = input_vid_to_vlan(in_bundle, vid);
7302 /* Look at the output ports to check for destination selections. */
7304 NL_ATTR_FOR_EACH (a, left, ctx->xout->odp_actions.data,
7305 ctx->xout->odp_actions.size) {
7306 enum ovs_action_attr type = nl_attr_type(a);
7307 struct ofport_dpif *ofport;
7309 if (type != OVS_ACTION_ATTR_OUTPUT) {
7313 ofport = get_odp_port(ofproto, nl_attr_get_u32(a));
7314 if (ofport && ofport->bundle) {
7315 mirrors |= ofport->bundle->dst_mirrors;
7323 /* Restore the original packet before adding the mirror actions. */
7324 ctx->xin->flow = *orig_flow;
7329 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
7331 if (!vlan_is_mirrored(m, vlan)) {
7332 mirrors = zero_rightmost_1bit(mirrors);
7336 mirrors &= ~m->dup_mirrors;
7337 ctx->xout->mirrors |= m->dup_mirrors;
7339 output_normal(ctx, m->out, vlan);
7340 } else if (vlan != m->out_vlan
7341 && !eth_addr_is_reserved(orig_flow->dl_dst)) {
7342 struct ofbundle *bundle;
7344 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
7345 if (ofbundle_includes_vlan(bundle, m->out_vlan)
7346 && !bundle->mirror_out) {
7347 output_normal(ctx, bundle, m->out_vlan);
7355 update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors,
7356 uint64_t packets, uint64_t bytes)
7362 for (; mirrors; mirrors = zero_rightmost_1bit(mirrors)) {
7365 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
7368 /* In normal circumstances 'm' will not be NULL. However,
7369 * if mirrors are reconfigured, we can temporarily get out
7370 * of sync in facet_revalidate(). We could "correct" the
7371 * mirror list before reaching here, but doing that would
7372 * not properly account the traffic stats we've currently
7373 * accumulated for previous mirror configuration. */
7377 m->packet_count += packets;
7378 m->byte_count += bytes;
7382 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
7383 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
7384 * indicate this; newer upstream kernels use gratuitous ARP requests. */
7386 is_gratuitous_arp(const struct flow *flow)
7388 return (flow->dl_type == htons(ETH_TYPE_ARP)
7389 && eth_addr_is_broadcast(flow->dl_dst)
7390 && (flow->nw_proto == ARP_OP_REPLY
7391 || (flow->nw_proto == ARP_OP_REQUEST
7392 && flow->nw_src == flow->nw_dst)));
7396 update_learning_table(struct ofproto_dpif *ofproto,
7397 const struct flow *flow, int vlan,
7398 struct ofbundle *in_bundle)
7400 struct mac_entry *mac;
7402 /* Don't learn the OFPP_NONE port. */
7403 if (in_bundle == &ofpp_none_bundle) {
7407 if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
7411 mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan);
7412 if (is_gratuitous_arp(flow)) {
7413 /* We don't want to learn from gratuitous ARP packets that are
7414 * reflected back over bond slaves so we lock the learning table. */
7415 if (!in_bundle->bond) {
7416 mac_entry_set_grat_arp_lock(mac);
7417 } else if (mac_entry_is_grat_arp_locked(mac)) {
7422 if (mac_entry_is_new(mac) || mac->port.p != in_bundle) {
7423 /* The log messages here could actually be useful in debugging,
7424 * so keep the rate limit relatively high. */
7425 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
7426 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
7427 "on port %s in VLAN %d",
7428 ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src),
7429 in_bundle->name, vlan);
7431 mac->port.p = in_bundle;
7432 tag_set_add(&ofproto->backer->revalidate_set,
7433 mac_learning_changed(ofproto->ml, mac));
7437 static struct ofbundle *
7438 lookup_input_bundle(const struct ofproto_dpif *ofproto, uint16_t in_port,
7439 bool warn, struct ofport_dpif **in_ofportp)
7441 struct ofport_dpif *ofport;
7443 /* Find the port and bundle for the received packet. */
7444 ofport = get_ofp_port(ofproto, in_port);
7446 *in_ofportp = ofport;
7448 if (ofport && ofport->bundle) {
7449 return ofport->bundle;
7452 /* Special-case OFPP_NONE, which a controller may use as the ingress
7453 * port for traffic that it is sourcing. */
7454 if (in_port == OFPP_NONE) {
7455 return &ofpp_none_bundle;
7458 /* Odd. A few possible reasons here:
7460 * - We deleted a port but there are still a few packets queued up
7463 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
7464 * we don't know about.
7466 * - The ofproto client didn't configure the port as part of a bundle.
7467 * This is particularly likely to happen if a packet was received on the
7468 * port after it was created, but before the client had a chance to
7469 * configure its bundle.
7472 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7474 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
7475 "port %"PRIu16, ofproto->up.name, in_port);
7480 /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or
7481 * dropped. Returns true if they may be forwarded, false if they should be
7484 * 'in_port' must be the ofport_dpif that corresponds to flow->in_port.
7485 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
7487 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
7488 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
7489 * checked by input_vid_is_valid().
7491 * May also add tags to '*tags', although the current implementation only does
7492 * so in one special case.
7495 is_admissible(struct xlate_ctx *ctx, struct ofport_dpif *in_port,
7498 struct ofproto_dpif *ofproto = ctx->ofproto;
7499 struct flow *flow = &ctx->xin->flow;
7500 struct ofbundle *in_bundle = in_port->bundle;
7502 /* Drop frames for reserved multicast addresses
7503 * only if forward_bpdu option is absent. */
7504 if (!ofproto->up.forward_bpdu && eth_addr_is_reserved(flow->dl_dst)) {
7505 xlate_report(ctx, "packet has reserved destination MAC, dropping");
7509 if (in_bundle->bond) {
7510 struct mac_entry *mac;
7512 switch (bond_check_admissibility(in_bundle->bond, in_port,
7513 flow->dl_dst, &ctx->xout->tags)) {
7518 xlate_report(ctx, "bonding refused admissibility, dropping");
7521 case BV_DROP_IF_MOVED:
7522 mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL);
7523 if (mac && mac->port.p != in_bundle &&
7524 (!is_gratuitous_arp(flow)
7525 || mac_entry_is_grat_arp_locked(mac))) {
7526 xlate_report(ctx, "SLB bond thinks this packet looped back, "
7538 xlate_normal(struct xlate_ctx *ctx)
7540 struct ofport_dpif *in_port;
7541 struct ofbundle *in_bundle;
7542 struct mac_entry *mac;
7546 ctx->xout->has_normal = true;
7548 in_bundle = lookup_input_bundle(ctx->ofproto, ctx->xin->flow.in_port,
7549 ctx->xin->packet != NULL, &in_port);
7551 xlate_report(ctx, "no input bundle, dropping");
7555 /* Drop malformed frames. */
7556 if (ctx->xin->flow.dl_type == htons(ETH_TYPE_VLAN) &&
7557 !(ctx->xin->flow.vlan_tci & htons(VLAN_CFI))) {
7558 if (ctx->xin->packet != NULL) {
7559 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7560 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
7561 "VLAN tag received on port %s",
7562 ctx->ofproto->up.name, in_bundle->name);
7564 xlate_report(ctx, "partial VLAN tag, dropping");
7568 /* Drop frames on bundles reserved for mirroring. */
7569 if (in_bundle->mirror_out) {
7570 if (ctx->xin->packet != NULL) {
7571 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7572 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
7573 "%s, which is reserved exclusively for mirroring",
7574 ctx->ofproto->up.name, in_bundle->name);
7576 xlate_report(ctx, "input port is mirror output port, dropping");
7581 vid = vlan_tci_to_vid(ctx->xin->flow.vlan_tci);
7582 if (!input_vid_is_valid(vid, in_bundle, ctx->xin->packet != NULL)) {
7583 xlate_report(ctx, "disallowed VLAN VID for this input port, dropping");
7586 vlan = input_vid_to_vlan(in_bundle, vid);
7588 /* Check other admissibility requirements. */
7589 if (in_port && !is_admissible(ctx, in_port, vlan)) {
7593 /* Learn source MAC. */
7594 if (ctx->xin->may_learn) {
7595 update_learning_table(ctx->ofproto, &ctx->xin->flow, vlan, in_bundle);
7598 /* Determine output bundle. */
7599 mac = mac_learning_lookup(ctx->ofproto->ml, ctx->xin->flow.dl_dst, vlan,
7602 if (mac->port.p != in_bundle) {
7603 xlate_report(ctx, "forwarding to learned port");
7604 output_normal(ctx, mac->port.p, vlan);
7606 xlate_report(ctx, "learned port is input port, dropping");
7609 struct ofbundle *bundle;
7611 xlate_report(ctx, "no learned MAC for destination, flooding");
7612 HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
7613 if (bundle != in_bundle
7614 && ofbundle_includes_vlan(bundle, vlan)
7615 && bundle->floodable
7616 && !bundle->mirror_out) {
7617 output_normal(ctx, bundle, vlan);
7620 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
7624 /* Optimized flow revalidation.
7626 * It's a difficult problem, in general, to tell which facets need to have
7627 * their actions recalculated whenever the OpenFlow flow table changes. We
7628 * don't try to solve that general problem: for most kinds of OpenFlow flow
7629 * table changes, we recalculate the actions for every facet. This is
7630 * relatively expensive, but it's good enough if the OpenFlow flow table
7631 * doesn't change very often.
7633 * However, we can expect one particular kind of OpenFlow flow table change to
7634 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
7635 * of CPU on revalidating every facet whenever MAC learning modifies the flow
7636 * table, we add a special case that applies to flow tables in which every rule
7637 * has the same form (that is, the same wildcards), except that the table is
7638 * also allowed to have a single "catch-all" flow that matches all packets. We
7639 * optimize this case by tagging all of the facets that resubmit into the table
7640 * and invalidating the same tag whenever a flow changes in that table. The
7641 * end result is that we revalidate just the facets that need it (and sometimes
7642 * a few more, but not all of the facets or even all of the facets that
7643 * resubmit to the table modified by MAC learning). */
7645 /* Calculates the tag to use for 'flow' and mask 'mask' when it is inserted
7646 * into an OpenFlow table with the given 'basis'. */
7648 rule_calculate_tag(const struct flow *flow, const struct minimask *mask,
7651 if (minimask_is_catchall(mask)) {
7654 uint32_t hash = flow_hash_in_minimask(flow, mask, secret);
7655 return tag_create_deterministic(hash);
7659 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
7660 * taggability of that table.
7662 * This function must be called after *each* change to a flow table. If you
7663 * skip calling it on some changes then the pointer comparisons at the end can
7664 * be invalid if you get unlucky. For example, if a flow removal causes a
7665 * cls_table to be destroyed and then a flow insertion causes a cls_table with
7666 * different wildcards to be created with the same address, then this function
7667 * will incorrectly skip revalidation. */
7669 table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
7671 struct table_dpif *table = &ofproto->tables[table_id];
7672 const struct oftable *oftable = &ofproto->up.tables[table_id];
7673 struct cls_table *catchall, *other;
7674 struct cls_table *t;
7676 catchall = other = NULL;
7678 switch (hmap_count(&oftable->cls.tables)) {
7680 /* We could tag this OpenFlow table but it would make the logic a
7681 * little harder and it's a corner case that doesn't seem worth it
7687 HMAP_FOR_EACH (t, hmap_node, &oftable->cls.tables) {
7688 if (cls_table_is_catchall(t)) {
7690 } else if (!other) {
7693 /* Indicate that we can't tag this by setting both tables to
7694 * NULL. (We know that 'catchall' is already NULL.) */
7701 /* Can't tag this table. */
7705 if (table->catchall_table != catchall || table->other_table != other) {
7706 table->catchall_table = catchall;
7707 table->other_table = other;
7708 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
7712 /* Given 'rule' that has changed in some way (either it is a rule being
7713 * inserted, a rule being deleted, or a rule whose actions are being
7714 * modified), marks facets for revalidation to ensure that packets will be
7715 * forwarded correctly according to the new state of the flow table.
7717 * This function must be called after *each* change to a flow table. See
7718 * the comment on table_update_taggable() for more information. */
7720 rule_invalidate(const struct rule_dpif *rule)
7722 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
7724 table_update_taggable(ofproto, rule->up.table_id);
7726 if (!ofproto->backer->need_revalidate) {
7727 struct table_dpif *table = &ofproto->tables[rule->up.table_id];
7729 if (table->other_table && rule->tag) {
7730 tag_set_add(&ofproto->backer->revalidate_set, rule->tag);
7732 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
7738 set_frag_handling(struct ofproto *ofproto_,
7739 enum ofp_config_flags frag_handling)
7741 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7742 if (frag_handling != OFPC_FRAG_REASM) {
7743 ofproto->backer->need_revalidate = REV_RECONFIGURE;
7751 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
7752 const struct flow *flow,
7753 const struct ofpact *ofpacts, size_t ofpacts_len)
7755 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7756 struct initial_vals initial_vals;
7757 struct odputil_keybuf keybuf;
7758 struct dpif_flow_stats stats;
7759 struct xlate_out xout;
7760 struct xlate_in xin;
7764 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
7765 odp_flow_key_from_flow(&key, flow,
7766 ofp_port_to_odp_port(ofproto, flow->in_port));
7768 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
7770 initial_vals.vlan_tci = flow->vlan_tci;
7771 initial_vals.tunnel_ip_tos = 0;
7772 xlate_in_init(&xin, ofproto, flow, &initial_vals, NULL, stats.tcp_flags,
7774 xin.resubmit_stats = &stats;
7775 xin.ofpacts_len = ofpacts_len;
7776 xin.ofpacts = ofpacts;
7778 xlate_actions(&xin, &xout);
7779 dpif_execute(ofproto->backer->dpif, key.data, key.size,
7780 xout.odp_actions.data, xout.odp_actions.size, packet);
7781 xlate_out_uninit(&xout);
7789 set_netflow(struct ofproto *ofproto_,
7790 const struct netflow_options *netflow_options)
7792 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7794 if (netflow_options) {
7795 if (!ofproto->netflow) {
7796 ofproto->netflow = netflow_create();
7798 return netflow_set_options(ofproto->netflow, netflow_options);
7800 netflow_destroy(ofproto->netflow);
7801 ofproto->netflow = NULL;
7807 get_netflow_ids(const struct ofproto *ofproto_,
7808 uint8_t *engine_type, uint8_t *engine_id)
7810 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7812 dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
7816 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
7818 if (!facet_is_controller_flow(facet) &&
7819 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
7820 struct subfacet *subfacet;
7821 struct ofexpired expired;
7823 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
7824 if (subfacet->path == SF_FAST_PATH) {
7825 struct dpif_flow_stats stats;
7827 subfacet_install(subfacet, &facet->xout.odp_actions, &stats);
7828 subfacet_update_stats(subfacet, &stats);
7832 expired.flow = facet->flow;
7833 expired.packet_count = facet->packet_count;
7834 expired.byte_count = facet->byte_count;
7835 expired.used = facet->used;
7836 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
7841 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
7843 struct facet *facet;
7845 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
7846 send_active_timeout(ofproto, facet);
7850 static struct ofproto_dpif *
7851 ofproto_dpif_lookup(const char *name)
7853 struct ofproto_dpif *ofproto;
7855 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
7856 hash_string(name, 0), &all_ofproto_dpifs) {
7857 if (!strcmp(ofproto->up.name, name)) {
7865 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
7866 const char *argv[], void *aux OVS_UNUSED)
7868 struct ofproto_dpif *ofproto;
7871 ofproto = ofproto_dpif_lookup(argv[1]);
7873 unixctl_command_reply_error(conn, "no such bridge");
7876 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
7878 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
7879 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
7883 unixctl_command_reply(conn, "table successfully flushed");
7887 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
7888 const char *argv[], void *aux OVS_UNUSED)
7890 struct ds ds = DS_EMPTY_INITIALIZER;
7891 const struct ofproto_dpif *ofproto;
7892 const struct mac_entry *e;
7894 ofproto = ofproto_dpif_lookup(argv[1]);
7896 unixctl_command_reply_error(conn, "no such bridge");
7900 ds_put_cstr(&ds, " port VLAN MAC Age\n");
7901 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
7902 struct ofbundle *bundle = e->port.p;
7903 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
7904 ofbundle_get_a_port(bundle)->odp_port,
7905 e->vlan, ETH_ADDR_ARGS(e->mac),
7906 mac_entry_age(ofproto->ml, e));
7908 unixctl_command_reply(conn, ds_cstr(&ds));
7913 struct xlate_out xout;
7914 struct xlate_in xin;
7920 trace_format_rule(struct ds *result, uint8_t table_id, int level,
7921 const struct rule_dpif *rule)
7923 ds_put_char_multiple(result, '\t', level);
7925 ds_put_cstr(result, "No match\n");
7929 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
7930 table_id, ntohll(rule->up.flow_cookie));
7931 cls_rule_format(&rule->up.cr, result);
7932 ds_put_char(result, '\n');
7934 ds_put_char_multiple(result, '\t', level);
7935 ds_put_cstr(result, "OpenFlow ");
7936 ofpacts_format(rule->up.ofpacts, rule->up.ofpacts_len, result);
7937 ds_put_char(result, '\n');
7941 trace_format_flow(struct ds *result, int level, const char *title,
7942 struct trace_ctx *trace)
7944 ds_put_char_multiple(result, '\t', level);
7945 ds_put_format(result, "%s: ", title);
7946 if (flow_equal(&trace->xin.flow, &trace->flow)) {
7947 ds_put_cstr(result, "unchanged");
7949 flow_format(result, &trace->xin.flow);
7950 trace->flow = trace->xin.flow;
7952 ds_put_char(result, '\n');
7956 trace_format_regs(struct ds *result, int level, const char *title,
7957 struct trace_ctx *trace)
7961 ds_put_char_multiple(result, '\t', level);
7962 ds_put_format(result, "%s:", title);
7963 for (i = 0; i < FLOW_N_REGS; i++) {
7964 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
7966 ds_put_char(result, '\n');
7970 trace_format_odp(struct ds *result, int level, const char *title,
7971 struct trace_ctx *trace)
7973 struct ofpbuf *odp_actions = &trace->xout.odp_actions;
7975 ds_put_char_multiple(result, '\t', level);
7976 ds_put_format(result, "%s: ", title);
7977 format_odp_actions(result, odp_actions->data, odp_actions->size);
7978 ds_put_char(result, '\n');
7982 trace_resubmit(struct xlate_ctx *ctx, struct rule_dpif *rule)
7984 struct trace_ctx *trace = CONTAINER_OF(ctx->xin, struct trace_ctx, xin);
7985 struct ds *result = trace->result;
7987 ds_put_char(result, '\n');
7988 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
7989 trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace);
7990 trace_format_odp(result, ctx->recurse + 1, "Resubmitted odp", trace);
7991 trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule);
7995 trace_report(struct xlate_ctx *ctx, const char *s)
7997 struct trace_ctx *trace = CONTAINER_OF(ctx->xin, struct trace_ctx, xin);
7998 struct ds *result = trace->result;
8000 ds_put_char_multiple(result, '\t', ctx->recurse);
8001 ds_put_cstr(result, s);
8002 ds_put_char(result, '\n');
8006 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
8007 void *aux OVS_UNUSED)
8009 const char *dpname = argv[1];
8010 struct ofproto_dpif *ofproto;
8011 struct ofpbuf odp_key;
8012 struct ofpbuf *packet;
8013 struct initial_vals initial_vals;
8019 ofpbuf_init(&odp_key, 0);
8022 ofproto = ofproto_dpif_lookup(dpname);
8024 unixctl_command_reply_error(conn, "Unknown ofproto (use ofproto/list "
8028 if (argc == 3 || (argc == 4 && !strcmp(argv[3], "-generate"))) {
8029 /* ofproto/trace dpname flow [-generate] */
8030 const char *flow_s = argv[2];
8031 const char *generate_s = argv[3];
8033 /* Allow 'flow_s' to be either a datapath flow or an OpenFlow-like
8034 * flow. We guess which type it is based on whether 'flow_s' contains
8035 * an '(', since a datapath flow always contains '(') but an
8036 * OpenFlow-like flow should not (in fact it's allowed but I believe
8037 * that's not documented anywhere).
8039 * An alternative would be to try to parse 'flow_s' both ways, but then
8040 * it would be tricky giving a sensible error message. After all, do
8041 * you just say "syntax error" or do you present both error messages?
8042 * Both choices seem lousy. */
8043 if (strchr(flow_s, '(')) {
8046 /* Convert string to datapath key. */
8047 ofpbuf_init(&odp_key, 0);
8048 error = odp_flow_key_from_string(flow_s, NULL, &odp_key);
8050 unixctl_command_reply_error(conn, "Bad flow syntax");
8054 /* The user might have specified the wrong ofproto but within the
8055 * same backer. That's OK, ofproto_receive() can find the right
8057 if (ofproto_receive(ofproto->backer, NULL, odp_key.data,
8058 odp_key.size, &flow, NULL, &ofproto, NULL,
8060 unixctl_command_reply_error(conn, "Invalid flow");
8063 ds_put_format(&result, "Bridge: %s\n", ofproto->up.name);
8067 error_s = parse_ofp_exact_flow(&flow, argv[2]);
8069 unixctl_command_reply_error(conn, error_s);
8074 initial_vals.vlan_tci = flow.vlan_tci;
8075 initial_vals.tunnel_ip_tos = flow.tunnel.ip_tos;
8078 /* Generate a packet, if requested. */
8080 packet = ofpbuf_new(0);
8081 flow_compose(packet, &flow);
8083 } else if (argc == 7) {
8084 /* ofproto/trace dpname priority tun_id in_port mark packet */
8085 const char *priority_s = argv[2];
8086 const char *tun_id_s = argv[3];
8087 const char *in_port_s = argv[4];
8088 const char *mark_s = argv[5];
8089 const char *packet_s = argv[6];
8090 uint32_t in_port = atoi(in_port_s);
8091 ovs_be64 tun_id = htonll(strtoull(tun_id_s, NULL, 0));
8092 uint32_t priority = atoi(priority_s);
8093 uint32_t mark = atoi(mark_s);
8096 msg = eth_from_hex(packet_s, &packet);
8098 unixctl_command_reply_error(conn, msg);
8102 ds_put_cstr(&result, "Packet: ");
8103 s = ofp_packet_to_string(packet->data, packet->size);
8104 ds_put_cstr(&result, s);
8107 flow_extract(packet, priority, mark, NULL, in_port, &flow);
8108 flow.tunnel.tun_id = tun_id;
8109 initial_vals.vlan_tci = flow.vlan_tci;
8110 initial_vals.tunnel_ip_tos = flow.tunnel.ip_tos;
8112 unixctl_command_reply_error(conn, "Bad command syntax");
8116 ofproto_trace(ofproto, &flow, packet, &initial_vals, &result);
8117 unixctl_command_reply(conn, ds_cstr(&result));
8120 ds_destroy(&result);
8121 ofpbuf_delete(packet);
8122 ofpbuf_uninit(&odp_key);
8126 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
8127 const struct ofpbuf *packet,
8128 const struct initial_vals *initial_vals, struct ds *ds)
8130 struct rule_dpif *rule;
8132 ds_put_cstr(ds, "Flow: ");
8133 flow_format(ds, flow);
8134 ds_put_char(ds, '\n');
8136 rule = rule_dpif_lookup(ofproto, flow);
8138 trace_format_rule(ds, 0, 0, rule);
8139 if (rule == ofproto->miss_rule) {
8140 ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
8141 } else if (rule == ofproto->no_packet_in_rule) {
8142 ds_put_cstr(ds, "\nNo match, packets dropped because "
8143 "OFPPC_NO_PACKET_IN is set on in_port.\n");
8147 uint64_t odp_actions_stub[1024 / 8];
8148 struct ofpbuf odp_actions;
8150 struct trace_ctx trace;
8153 tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
8156 ofpbuf_use_stub(&odp_actions,
8157 odp_actions_stub, sizeof odp_actions_stub);
8158 xlate_in_init(&trace.xin, ofproto, flow, initial_vals, rule, tcp_flags,
8160 trace.xin.resubmit_hook = trace_resubmit;
8161 trace.xin.report_hook = trace_report;
8162 xlate_actions(&trace.xin, &trace.xout);
8164 ds_put_char(ds, '\n');
8165 trace_format_flow(ds, 0, "Final flow", &trace);
8166 ds_put_cstr(ds, "Datapath actions: ");
8167 format_odp_actions(ds, trace.xout.odp_actions.data,
8168 trace.xout.odp_actions.size);
8170 if (trace.xout.slow) {
8171 ds_put_cstr(ds, "\nThis flow is handled by the userspace "
8172 "slow path because it:");
8173 switch (trace.xout.slow) {
8175 ds_put_cstr(ds, "\n\t- Consists of CFM packets.");
8178 ds_put_cstr(ds, "\n\t- Consists of LACP packets.");
8181 ds_put_cstr(ds, "\n\t- Consists of STP packets.");
8183 case SLOW_CONTROLLER:
8184 ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages "
8185 "to the OpenFlow controller.");
8192 xlate_out_uninit(&trace.xout);
8197 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
8198 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
8201 unixctl_command_reply(conn, NULL);
8205 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
8206 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
8209 unixctl_command_reply(conn, NULL);
8212 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
8213 * 'reply' describing the results. */
8215 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
8217 struct facet *facet;
8221 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
8222 if (!facet_check_consistency(facet)) {
8227 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
8231 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
8232 ofproto->up.name, errors);
8234 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
8239 ofproto_dpif_self_check(struct unixctl_conn *conn,
8240 int argc, const char *argv[], void *aux OVS_UNUSED)
8242 struct ds reply = DS_EMPTY_INITIALIZER;
8243 struct ofproto_dpif *ofproto;
8246 ofproto = ofproto_dpif_lookup(argv[1]);
8248 unixctl_command_reply_error(conn, "Unknown ofproto (use "
8249 "ofproto/list for help)");
8252 ofproto_dpif_self_check__(ofproto, &reply);
8254 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
8255 ofproto_dpif_self_check__(ofproto, &reply);
8259 unixctl_command_reply(conn, ds_cstr(&reply));
8263 /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list
8264 * of the 'ofproto_shash' nodes. It is the responsibility of the caller
8265 * to destroy 'ofproto_shash' and free the returned value. */
8266 static const struct shash_node **
8267 get_ofprotos(struct shash *ofproto_shash)
8269 const struct ofproto_dpif *ofproto;
8271 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
8272 char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
8273 shash_add_nocopy(ofproto_shash, name, ofproto);
8276 return shash_sort(ofproto_shash);
8280 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
8281 const char *argv[] OVS_UNUSED,
8282 void *aux OVS_UNUSED)
8284 struct ds ds = DS_EMPTY_INITIALIZER;
8285 struct shash ofproto_shash;
8286 const struct shash_node **sorted_ofprotos;
8289 shash_init(&ofproto_shash);
8290 sorted_ofprotos = get_ofprotos(&ofproto_shash);
8291 for (i = 0; i < shash_count(&ofproto_shash); i++) {
8292 const struct shash_node *node = sorted_ofprotos[i];
8293 ds_put_format(&ds, "%s\n", node->name);
8296 shash_destroy(&ofproto_shash);
8297 free(sorted_ofprotos);
8299 unixctl_command_reply(conn, ds_cstr(&ds));
8304 show_dp_format(const struct ofproto_dpif *ofproto, struct ds *ds)
8306 const struct shash_node **ports;
8308 struct avg_subfacet_rates lifetime;
8309 unsigned long long int minutes;
8310 const int min_ms = 60 * 1000; /* milliseconds in one minute. */
8312 minutes = (time_msec() - ofproto->created) / min_ms;
8315 lifetime.add_rate = (double)ofproto->total_subfacet_add_count
8317 lifetime.del_rate = (double)ofproto->total_subfacet_del_count
8320 lifetime.add_rate = 0.0;
8321 lifetime.del_rate = 0.0;
8324 ds_put_format(ds, "%s (%s):\n", ofproto->up.name,
8325 dpif_name(ofproto->backer->dpif));
8327 "\tlookups: hit:%"PRIu64" missed:%"PRIu64"\n",
8328 ofproto->n_hit, ofproto->n_missed);
8329 ds_put_format(ds, "\tflows: cur: %zu, avg: %5.3f, max: %d,"
8330 " life span: %llu(ms)\n",
8331 hmap_count(&ofproto->subfacets),
8332 avg_subfacet_count(ofproto),
8333 ofproto->max_n_subfacet,
8334 avg_subfacet_life_span(ofproto));
8335 if (minutes >= 60) {
8336 show_dp_rates(ds, "\t\thourly avg:", &ofproto->hourly);
8338 if (minutes >= 60 * 24) {
8339 show_dp_rates(ds, "\t\tdaily avg:", &ofproto->daily);
8341 show_dp_rates(ds, "\t\toverall avg:", &lifetime);
8343 ports = shash_sort(&ofproto->up.port_by_name);
8344 for (i = 0; i < shash_count(&ofproto->up.port_by_name); i++) {
8345 const struct shash_node *node = ports[i];
8346 struct ofport *ofport = node->data;
8347 const char *name = netdev_get_name(ofport->netdev);
8348 const char *type = netdev_get_type(ofport->netdev);
8351 ds_put_format(ds, "\t%s %u/", name, ofport->ofp_port);
8353 odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
8354 if (odp_port != OVSP_NONE) {
8355 ds_put_format(ds, "%"PRIu32":", odp_port);
8357 ds_put_cstr(ds, "none:");
8360 if (strcmp(type, "system")) {
8361 struct netdev *netdev;
8364 ds_put_format(ds, " (%s", type);
8366 error = netdev_open(name, type, &netdev);
8371 error = netdev_get_config(netdev, &config);
8373 const struct smap_node **nodes;
8376 nodes = smap_sort(&config);
8377 for (i = 0; i < smap_count(&config); i++) {
8378 const struct smap_node *node = nodes[i];
8379 ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
8380 node->key, node->value);
8384 smap_destroy(&config);
8386 netdev_close(netdev);
8388 ds_put_char(ds, ')');
8390 ds_put_char(ds, '\n');
8396 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc,
8397 const char *argv[], void *aux OVS_UNUSED)
8399 struct ds ds = DS_EMPTY_INITIALIZER;
8400 const struct ofproto_dpif *ofproto;
8404 for (i = 1; i < argc; i++) {
8405 ofproto = ofproto_dpif_lookup(argv[i]);
8407 ds_put_format(&ds, "Unknown bridge %s (use dpif/dump-dps "
8408 "for help)", argv[i]);
8409 unixctl_command_reply_error(conn, ds_cstr(&ds));
8412 show_dp_format(ofproto, &ds);
8415 struct shash ofproto_shash;
8416 const struct shash_node **sorted_ofprotos;
8419 shash_init(&ofproto_shash);
8420 sorted_ofprotos = get_ofprotos(&ofproto_shash);
8421 for (i = 0; i < shash_count(&ofproto_shash); i++) {
8422 const struct shash_node *node = sorted_ofprotos[i];
8423 show_dp_format(node->data, &ds);
8426 shash_destroy(&ofproto_shash);
8427 free(sorted_ofprotos);
8430 unixctl_command_reply(conn, ds_cstr(&ds));
8435 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
8436 int argc OVS_UNUSED, const char *argv[],
8437 void *aux OVS_UNUSED)
8439 struct ds ds = DS_EMPTY_INITIALIZER;
8440 const struct ofproto_dpif *ofproto;
8441 struct subfacet *subfacet;
8443 ofproto = ofproto_dpif_lookup(argv[1]);
8445 unixctl_command_reply_error(conn, "no such bridge");
8449 update_stats(ofproto->backer);
8451 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
8452 struct facet *facet = subfacet->facet;
8454 odp_flow_key_format(subfacet->key, subfacet->key_len, &ds);
8456 ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:",
8457 subfacet->dp_packet_count, subfacet->dp_byte_count);
8458 if (subfacet->used) {
8459 ds_put_format(&ds, "%.3fs",
8460 (time_msec() - subfacet->used) / 1000.0);
8462 ds_put_format(&ds, "never");
8464 if (subfacet->facet->tcp_flags) {
8465 ds_put_cstr(&ds, ", flags:");
8466 packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags);
8469 ds_put_cstr(&ds, ", actions:");
8470 if (facet->xout.slow) {
8471 uint64_t slow_path_stub[128 / 8];
8472 const struct nlattr *actions;
8475 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
8476 slow_path_stub, sizeof slow_path_stub,
8477 &actions, &actions_len);
8478 format_odp_actions(&ds, actions, actions_len);
8480 format_odp_actions(&ds, facet->xout.odp_actions.data,
8481 facet->xout.odp_actions.size);
8483 ds_put_char(&ds, '\n');
8486 unixctl_command_reply(conn, ds_cstr(&ds));
8491 ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn,
8492 int argc OVS_UNUSED, const char *argv[],
8493 void *aux OVS_UNUSED)
8495 struct ds ds = DS_EMPTY_INITIALIZER;
8496 struct ofproto_dpif *ofproto;
8498 ofproto = ofproto_dpif_lookup(argv[1]);
8500 unixctl_command_reply_error(conn, "no such bridge");
8504 flush(&ofproto->up);
8506 unixctl_command_reply(conn, ds_cstr(&ds));
8511 ofproto_dpif_unixctl_init(void)
8513 static bool registered;
8519 unixctl_command_register(
8521 "bridge {priority tun_id in_port mark packet | odp_flow [-generate]}",
8522 2, 6, ofproto_unixctl_trace, NULL);
8523 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
8524 ofproto_unixctl_fdb_flush, NULL);
8525 unixctl_command_register("fdb/show", "bridge", 1, 1,
8526 ofproto_unixctl_fdb_show, NULL);
8527 unixctl_command_register("ofproto/clog", "", 0, 0,
8528 ofproto_dpif_clog, NULL);
8529 unixctl_command_register("ofproto/unclog", "", 0, 0,
8530 ofproto_dpif_unclog, NULL);
8531 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
8532 ofproto_dpif_self_check, NULL);
8533 unixctl_command_register("dpif/dump-dps", "", 0, 0,
8534 ofproto_unixctl_dpif_dump_dps, NULL);
8535 unixctl_command_register("dpif/show", "[bridge]", 0, INT_MAX,
8536 ofproto_unixctl_dpif_show, NULL);
8537 unixctl_command_register("dpif/dump-flows", "bridge", 1, 1,
8538 ofproto_unixctl_dpif_dump_flows, NULL);
8539 unixctl_command_register("dpif/del-flows", "bridge", 1, 1,
8540 ofproto_unixctl_dpif_del_flows, NULL);
8543 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
8545 * This is deprecated. It is only for compatibility with broken device drivers
8546 * in old versions of Linux that do not properly support VLANs when VLAN
8547 * devices are not used. When broken device drivers are no longer in
8548 * widespread use, we will delete these interfaces. */
8551 set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid)
8553 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
8554 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
8556 if (realdev_ofp_port == ofport->realdev_ofp_port
8557 && vid == ofport->vlandev_vid) {
8561 ofproto->backer->need_revalidate = REV_RECONFIGURE;
8563 if (ofport->realdev_ofp_port) {
8566 if (realdev_ofp_port && ofport->bundle) {
8567 /* vlandevs are enslaved to their realdevs, so they are not allowed to
8568 * themselves be part of a bundle. */
8569 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
8572 ofport->realdev_ofp_port = realdev_ofp_port;
8573 ofport->vlandev_vid = vid;
8575 if (realdev_ofp_port) {
8576 vsp_add(ofport, realdev_ofp_port, vid);
8583 hash_realdev_vid(uint16_t realdev_ofp_port, int vid)
8585 return hash_2words(realdev_ofp_port, vid);
8588 /* Returns the OFP port number of the Linux VLAN device that corresponds to
8589 * 'vlan_tci' on the network device with port number 'realdev_ofp_port' in
8590 * 'struct ofport_dpif'. For example, given 'realdev_ofp_port' of eth0 and
8591 * 'vlan_tci' 9, it would return the port number of eth0.9.
8593 * Unless VLAN splinters are enabled for port 'realdev_ofp_port', this
8594 * function just returns its 'realdev_ofp_port' argument. */
8596 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
8597 uint16_t realdev_ofp_port, ovs_be16 vlan_tci)
8599 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
8600 int vid = vlan_tci_to_vid(vlan_tci);
8601 const struct vlan_splinter *vsp;
8603 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
8604 hash_realdev_vid(realdev_ofp_port, vid),
8605 &ofproto->realdev_vid_map) {
8606 if (vsp->realdev_ofp_port == realdev_ofp_port
8607 && vsp->vid == vid) {
8608 return vsp->vlandev_ofp_port;
8612 return realdev_ofp_port;
8615 static struct vlan_splinter *
8616 vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port)
8618 struct vlan_splinter *vsp;
8620 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, hash_int(vlandev_ofp_port, 0),
8621 &ofproto->vlandev_map) {
8622 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
8630 /* Returns the OpenFlow port number of the "real" device underlying the Linux
8631 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
8632 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
8633 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
8634 * eth0 and store 9 in '*vid'.
8636 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
8637 * VLAN device. Unless VLAN splinters are enabled, this is what this function
8640 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
8641 uint16_t vlandev_ofp_port, int *vid)
8643 if (!hmap_is_empty(&ofproto->vlandev_map)) {
8644 const struct vlan_splinter *vsp;
8646 vsp = vlandev_find(ofproto, vlandev_ofp_port);
8651 return vsp->realdev_ofp_port;
8657 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
8658 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
8659 * 'flow->in_port' to the "real" device backing the VLAN device, sets
8660 * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is
8661 * always the case unless VLAN splinters are enabled), returns false without
8662 * making any changes. */
8664 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
8669 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port, &vid);
8674 /* Cause the flow to be processed as if it came in on the real device with
8675 * the VLAN device's VLAN ID. */
8676 flow->in_port = realdev;
8677 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
8682 vsp_remove(struct ofport_dpif *port)
8684 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
8685 struct vlan_splinter *vsp;
8687 vsp = vlandev_find(ofproto, port->up.ofp_port);
8689 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
8690 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
8693 port->realdev_ofp_port = 0;
8695 VLOG_ERR("missing vlan device record");
8700 vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid)
8702 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
8704 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
8705 && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid))
8706 == realdev_ofp_port)) {
8707 struct vlan_splinter *vsp;
8709 vsp = xmalloc(sizeof *vsp);
8710 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
8711 hash_int(port->up.ofp_port, 0));
8712 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
8713 hash_realdev_vid(realdev_ofp_port, vid));
8714 vsp->realdev_ofp_port = realdev_ofp_port;
8715 vsp->vlandev_ofp_port = port->up.ofp_port;
8718 port->realdev_ofp_port = realdev_ofp_port;
8720 VLOG_ERR("duplicate vlan device record");
8725 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
8727 const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
8728 return ofport ? ofport->odp_port : OVSP_NONE;
8731 static struct ofport_dpif *
8732 odp_port_to_ofport(const struct dpif_backer *backer, uint32_t odp_port)
8734 struct ofport_dpif *port;
8736 HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node,
8737 hash_int(odp_port, 0),
8738 &backer->odp_to_ofport_map) {
8739 if (port->odp_port == odp_port) {
8748 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
8750 struct ofport_dpif *port;
8752 port = odp_port_to_ofport(ofproto->backer, odp_port);
8753 if (port && &ofproto->up == port->up.ofproto) {
8754 return port->up.ofp_port;
8759 static unsigned long long int
8760 avg_subfacet_life_span(const struct ofproto_dpif *ofproto)
8762 unsigned long long int dc;
8763 unsigned long long int avg;
8765 dc = ofproto->total_subfacet_del_count + ofproto->subfacet_del_count;
8766 avg = dc ? ofproto->total_subfacet_life_span / dc : 0;
8772 avg_subfacet_count(const struct ofproto_dpif *ofproto)
8776 if (ofproto->n_update_stats) {
8777 avg_c = (double)ofproto->total_subfacet_count
8778 / ofproto->n_update_stats;
8785 show_dp_rates(struct ds *ds, const char *heading,
8786 const struct avg_subfacet_rates *rates)
8788 ds_put_format(ds, "%s add rate: %5.3f/min, del rate: %5.3f/min\n",
8789 heading, rates->add_rate, rates->del_rate);
8793 update_max_subfacet_count(struct ofproto_dpif *ofproto)
8795 ofproto->max_n_subfacet = MAX(ofproto->max_n_subfacet,
8796 hmap_count(&ofproto->subfacets));
8799 /* Compute exponentially weighted moving average, adding 'new' as the newest,
8800 * most heavily weighted element. 'base' designates the rate of decay: after
8801 * 'base' further updates, 'new''s weight in the EWMA decays to about 1/e
8804 exp_mavg(double *avg, int base, double new)
8806 *avg = (*avg * (base - 1) + new) / base;
8810 update_moving_averages(struct ofproto_dpif *ofproto)
8812 const int min_ms = 60 * 1000; /* milliseconds in one minute. */
8814 /* Update hourly averages on the minute boundaries. */
8815 if (time_msec() - ofproto->last_minute >= min_ms) {
8816 exp_mavg(&ofproto->hourly.add_rate, 60, ofproto->subfacet_add_count);
8817 exp_mavg(&ofproto->hourly.del_rate, 60, ofproto->subfacet_del_count);
8819 /* Update daily averages on the hour boundaries. */
8820 if ((ofproto->last_minute - ofproto->created) / min_ms % 60 == 59) {
8821 exp_mavg(&ofproto->daily.add_rate, 24, ofproto->hourly.add_rate);
8822 exp_mavg(&ofproto->daily.del_rate, 24, ofproto->hourly.del_rate);
8825 ofproto->total_subfacet_add_count += ofproto->subfacet_add_count;
8826 ofproto->total_subfacet_del_count += ofproto->subfacet_del_count;
8827 ofproto->subfacet_add_count = 0;
8828 ofproto->subfacet_del_count = 0;
8829 ofproto->last_minute += min_ms;
8834 dpif_stats_update_hit_count(struct ofproto_dpif *ofproto, uint64_t delta)
8836 ofproto->n_hit += delta;
8839 const struct ofproto_class ofproto_dpif_class = {
8874 port_is_lacp_current,
8875 NULL, /* rule_choose_table */
8882 rule_modify_actions,
8894 get_stp_port_status,
8901 is_mirror_output_bundle,
8902 forward_bpdu_changed,
8903 set_mac_table_config,