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_get_stats(struct rule *, uint64_t *packets, uint64_t *bytes);
122 static void rule_credit_stats(struct rule_dpif *,
123 const struct dpif_flow_stats *);
124 static void flow_push_stats(struct facet *, const struct dpif_flow_stats *);
125 static tag_type rule_calculate_tag(const struct flow *,
126 const struct minimask *, uint32_t basis);
127 static void rule_invalidate(const struct rule_dpif *);
129 #define MAX_MIRRORS 32
130 typedef uint32_t mirror_mask_t;
131 #define MIRROR_MASK_C(X) UINT32_C(X)
132 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
134 struct ofproto_dpif *ofproto; /* Owning ofproto. */
135 size_t idx; /* In ofproto's "mirrors" array. */
136 void *aux; /* Key supplied by ofproto's client. */
137 char *name; /* Identifier for log messages. */
139 /* Selection criteria. */
140 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
141 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
142 unsigned long *vlans; /* Bitmap of chosen VLANs, NULL selects all. */
144 /* Output (exactly one of out == NULL and out_vlan == -1 is true). */
145 struct ofbundle *out; /* Output port or NULL. */
146 int out_vlan; /* Output VLAN or -1. */
147 mirror_mask_t dup_mirrors; /* Bitmap of mirrors with the same output. */
150 int64_t packet_count; /* Number of packets sent. */
151 int64_t byte_count; /* Number of bytes sent. */
154 static void mirror_destroy(struct ofmirror *);
155 static void update_mirror_stats(struct ofproto_dpif *ofproto,
156 mirror_mask_t mirrors,
157 uint64_t packets, uint64_t bytes);
160 struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
161 struct ofproto_dpif *ofproto; /* Owning ofproto. */
162 void *aux; /* Key supplied by ofproto's client. */
163 char *name; /* Identifier for log messages. */
166 struct list ports; /* Contains "struct ofport"s. */
167 enum port_vlan_mode vlan_mode; /* VLAN mode */
168 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
169 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
170 * NULL if all VLANs are trunked. */
171 struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */
172 struct bond *bond; /* Nonnull iff more than one port. */
173 bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */
176 bool floodable; /* True if no port has OFPUTIL_PC_NO_FLOOD set. */
178 /* Port mirroring info. */
179 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
180 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
181 mirror_mask_t mirror_out; /* Mirrors that output to this bundle. */
184 static void bundle_remove(struct ofport *);
185 static void bundle_update(struct ofbundle *);
186 static void bundle_destroy(struct ofbundle *);
187 static void bundle_del_port(struct ofport_dpif *);
188 static void bundle_run(struct ofbundle *);
189 static void bundle_wait(struct ofbundle *);
190 static struct ofbundle *lookup_input_bundle(const struct ofproto_dpif *,
191 uint16_t in_port, bool warn,
192 struct ofport_dpif **in_ofportp);
194 /* A controller may use OFPP_NONE as the ingress port to indicate that
195 * it did not arrive on a "real" port. 'ofpp_none_bundle' exists for
196 * when an input bundle is needed for validation (e.g., mirroring or
197 * OFPP_NORMAL processing). It is not connected to an 'ofproto' or have
198 * any 'port' structs, so care must be taken when dealing with it. */
199 static struct ofbundle ofpp_none_bundle = {
201 .vlan_mode = PORT_VLAN_TRUNK
204 static void stp_run(struct ofproto_dpif *ofproto);
205 static void stp_wait(struct ofproto_dpif *ofproto);
206 static int set_stp_port(struct ofport *,
207 const struct ofproto_port_stp_settings *);
209 static bool ofbundle_includes_vlan(const struct ofbundle *, uint16_t vlan);
213 /* Initial values of fields of the packet that may be changed during
214 * flow processing and needed later. */
215 struct initial_vals {
216 /* This is the value of vlan_tci in the packet as actually received from
217 * dpif. This is the same as the facet's flow.vlan_tci unless the packet
218 * was received via a VLAN splinter. In that case, this value is 0
219 * (because the packet as actually received from the dpif had no 802.1Q
220 * tag) but the facet's flow.vlan_tci is set to the VLAN that the splinter
223 * This member should be removed when the VLAN splinters feature is no
227 /* If received on a tunnel, the IP TOS value of the tunnel. */
228 uint8_t tunnel_ip_tos;
232 tag_type tags; /* Tags associated with actions. */
233 enum slow_path_reason slow; /* 0 if fast path may be used. */
234 bool has_learn; /* Actions include NXAST_LEARN? */
235 bool has_normal; /* Actions output to OFPP_NORMAL? */
236 bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */
237 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
238 mirror_mask_t mirrors; /* Bitmap of associated mirrors. */
240 uint64_t odp_actions_stub[256 / 8];
241 struct ofpbuf odp_actions;
245 struct ofproto_dpif *ofproto;
247 /* Flow to which the OpenFlow actions apply. xlate_actions() will modify
248 * this flow when actions change header fields. */
251 struct initial_vals initial_vals;
253 /* The packet corresponding to 'flow', or a null pointer if we are
254 * revalidating without a packet to refer to. */
255 const struct ofpbuf *packet;
257 /* Should OFPP_NORMAL update the MAC learning table? Should "learn"
258 * actions update the flow table?
260 * We want to update these tables if we are actually processing a packet,
261 * or if we are accounting for packets that the datapath has processed, but
262 * not if we are just revalidating. */
265 /* The rule initiating translation or NULL. */
266 struct rule_dpif *rule;
268 /* The actions to translate. If 'rule' is not NULL, these may be NULL. */
269 const struct ofpact *ofpacts;
272 /* Union of the set of TCP flags seen so far in this flow. (Used only by
273 * NXAST_FIN_TIMEOUT. Set to zero to avoid updating updating rules'
277 /* If nonnull, flow translation calls this function just before executing a
278 * resubmit or OFPP_TABLE action. In addition, disables logging of traces
279 * when the recursion depth is exceeded.
281 * 'rule' is the rule being submitted into. It will be null if the
282 * resubmit or OFPP_TABLE action didn't find a matching rule.
284 * This is normally null so the client has to set it manually after
285 * calling xlate_in_init(). */
286 void (*resubmit_hook)(struct xlate_ctx *, struct rule_dpif *rule);
288 /* If nonnull, flow translation calls this function to report some
289 * significant decision, e.g. to explain why OFPP_NORMAL translation
290 * dropped a packet. */
291 void (*report_hook)(struct xlate_ctx *, const char *s);
293 /* If nonnull, flow translation credits the specified statistics to each
294 * rule reached through a resubmit or OFPP_TABLE action.
296 * This is normally null so the client has to set it manually after
297 * calling xlate_in_init(). */
298 const struct dpif_flow_stats *resubmit_stats;
301 /* Context used by xlate_actions() and its callees. */
303 struct xlate_in *xin;
304 struct xlate_out *xout;
306 struct ofproto_dpif *ofproto;
308 /* Flow at the last commit. */
309 struct flow base_flow;
311 /* Stack for the push and pop actions. Each stack element is of type
312 * "union mf_subvalue". */
313 union mf_subvalue init_stack[1024 / sizeof(union mf_subvalue)];
316 /* The rule that we are currently translating, or NULL. */
317 struct rule_dpif *rule;
319 int recurse; /* Recursion level, via xlate_table_action. */
320 bool max_resubmit_trigger; /* Recursed too deeply during translation. */
321 uint32_t orig_skb_priority; /* Priority when packet arrived. */
322 uint8_t table_id; /* OpenFlow table ID where flow was found. */
323 uint32_t sflow_n_outputs; /* Number of output ports. */
324 uint32_t sflow_odp_port; /* Output port for composing sFlow action. */
325 uint16_t user_cookie_offset;/* Used for user_action_cookie fixup. */
326 bool exit; /* No further actions should be processed. */
329 static void xlate_in_init(struct xlate_in *, struct ofproto_dpif *,
330 const struct flow *, const struct initial_vals *,
331 struct rule_dpif *, uint8_t tcp_flags,
332 const struct ofpbuf *);
334 static void xlate_out_uninit(struct xlate_out *);
336 static void xlate_actions(struct xlate_in *, struct xlate_out *);
338 static void xlate_actions_for_side_effects(struct xlate_in *);
340 static void xlate_table_action(struct xlate_ctx *, uint16_t in_port,
341 uint8_t table_id, bool may_packet_in);
343 static size_t put_userspace_action(const struct ofproto_dpif *,
344 struct ofpbuf *odp_actions,
346 const union user_action_cookie *,
349 static void compose_slow_path(const struct ofproto_dpif *, const struct flow *,
350 enum slow_path_reason,
351 uint64_t *stub, size_t stub_size,
352 const struct nlattr **actionsp,
353 size_t *actions_lenp);
355 static void xlate_report(struct xlate_ctx *ctx, const char *s);
357 /* A subfacet (see "struct subfacet" below) has three possible installation
360 * - SF_NOT_INSTALLED: Not installed in the datapath. This will only be the
361 * case just after the subfacet is created, just before the subfacet is
362 * destroyed, or if the datapath returns an error when we try to install a
365 * - SF_FAST_PATH: The subfacet's actions are installed in the datapath.
367 * - SF_SLOW_PATH: An action that sends every packet for the subfacet through
368 * ofproto_dpif is installed in the datapath.
371 SF_NOT_INSTALLED, /* No datapath flow for this subfacet. */
372 SF_FAST_PATH, /* Full actions are installed. */
373 SF_SLOW_PATH, /* Send-to-userspace action is installed. */
376 /* A dpif flow and actions associated with a facet.
378 * See also the large comment on struct facet. */
381 struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */
382 struct list list_node; /* In struct facet's 'facets' list. */
383 struct facet *facet; /* Owning facet. */
385 enum odp_key_fitness key_fitness;
389 long long int used; /* Time last used; time created if not used. */
390 long long int created; /* Time created. */
392 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
393 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
395 enum subfacet_path path; /* Installed in datapath? */
397 /* Datapath port the packet arrived on. This is needed to remove
398 * flows for ports that are no longer part of the bridge. Since the
399 * flow definition only has the OpenFlow port number and the port is
400 * no longer part of the bridge, we can't determine the datapath port
401 * number needed to delete the flow from the datapath. */
402 uint32_t odp_in_port;
405 #define SUBFACET_DESTROY_MAX_BATCH 50
407 static struct subfacet *subfacet_create(struct facet *, struct flow_miss *miss,
409 static struct subfacet *subfacet_find(struct ofproto_dpif *,
410 const struct nlattr *key, size_t key_len,
412 static void subfacet_destroy(struct subfacet *);
413 static void subfacet_destroy__(struct subfacet *);
414 static void subfacet_destroy_batch(struct ofproto_dpif *,
415 struct subfacet **, int n);
416 static void subfacet_reset_dp_stats(struct subfacet *,
417 struct dpif_flow_stats *);
418 static void subfacet_update_time(struct subfacet *, long long int used);
419 static void subfacet_update_stats(struct subfacet *,
420 const struct dpif_flow_stats *);
421 static int subfacet_install(struct subfacet *,
422 const struct ofpbuf *odp_actions,
423 struct dpif_flow_stats *);
424 static void subfacet_uninstall(struct subfacet *);
426 /* An exact-match instantiation of an OpenFlow flow.
428 * A facet associates a "struct flow", which represents the Open vSwitch
429 * userspace idea of an exact-match flow, with one or more subfacets. Each
430 * subfacet tracks the datapath's idea of the exact-match flow equivalent to
431 * the facet. When the kernel module (or other dpif implementation) and Open
432 * vSwitch userspace agree on the definition of a flow key, there is exactly
433 * one subfacet per facet. If the dpif implementation supports more-specific
434 * flow matching than userspace, however, a facet can have more than one
435 * subfacet, each of which corresponds to some distinction in flow that
436 * userspace simply doesn't understand.
438 * Flow expiration works in terms of subfacets, so a facet must have at least
439 * one subfacet or it will never expire, leaking memory. */
442 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
443 struct list list_node; /* In owning rule's 'facets' list. */
444 struct rule_dpif *rule; /* Owning rule. */
447 struct list subfacets;
448 long long int used; /* Time last used; time created if not used. */
455 * - Do include packets and bytes sent "by hand", e.g. with
458 * - Do include packets and bytes that were obtained from the datapath
459 * when a subfacet's statistics were reset (e.g. dpif_flow_put() with
460 * DPIF_FP_ZERO_STATS).
462 * - Do not include packets or bytes that can be obtained from the
463 * datapath for any existing subfacet.
465 uint64_t packet_count; /* Number of packets received. */
466 uint64_t byte_count; /* Number of bytes received. */
468 /* Resubmit statistics. */
469 uint64_t prev_packet_count; /* Number of packets from last stats push. */
470 uint64_t prev_byte_count; /* Number of bytes from last stats push. */
471 long long int prev_used; /* Used time from last stats push. */
474 uint64_t accounted_bytes; /* Bytes processed by facet_account(). */
475 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
476 uint8_t tcp_flags; /* TCP flags seen for this 'rule'. */
478 struct xlate_out xout;
480 /* Initial values of the packet that may be needed later. */
481 struct initial_vals initial_vals;
483 /* Storage for a single subfacet, to reduce malloc() time and space
484 * overhead. (A facet always has at least one subfacet and in the common
485 * case has exactly one subfacet. However, 'one_subfacet' may not
486 * always be valid, since it could have been removed after newer
487 * subfacets were pushed onto the 'subfacets' list.) */
488 struct subfacet one_subfacet;
490 long long int learn_rl; /* Rate limiter for facet_learn(). */
493 static struct facet *facet_create(const struct flow_miss *, uint32_t hash);
494 static void facet_remove(struct facet *);
495 static void facet_free(struct facet *);
497 static struct facet *facet_find(struct ofproto_dpif *,
498 const struct flow *, uint32_t hash);
499 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
500 const struct flow *, uint32_t hash);
501 static bool facet_revalidate(struct facet *);
502 static bool facet_check_consistency(struct facet *);
504 static void facet_flush_stats(struct facet *);
506 static void facet_update_time(struct facet *, long long int used);
507 static void facet_reset_counters(struct facet *);
508 static void facet_push_stats(struct facet *);
509 static void facet_learn(struct facet *);
510 static void facet_account(struct facet *);
511 static void push_all_stats(void);
513 static bool facet_is_controller_flow(struct facet *);
516 struct hmap_node odp_port_node; /* In dpif_backer's "odp_to_ofport_map". */
520 struct ofbundle *bundle; /* Bundle that contains this port, if any. */
521 struct list bundle_node; /* In struct ofbundle's "ports" list. */
522 struct cfm *cfm; /* Connectivity Fault Management, if any. */
523 tag_type tag; /* Tag associated with this port. */
524 bool may_enable; /* May be enabled in bonds. */
525 long long int carrier_seq; /* Carrier status changes. */
526 struct tnl_port *tnl_port; /* Tunnel handle, or null. */
529 struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */
530 enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */
531 long long int stp_state_entered;
533 struct hmap priorities; /* Map of attached 'priority_to_dscp's. */
535 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
537 * This is deprecated. It is only for compatibility with broken device
538 * drivers in old versions of Linux that do not properly support VLANs when
539 * VLAN devices are not used. When broken device drivers are no longer in
540 * widespread use, we will delete these interfaces. */
541 uint16_t realdev_ofp_port;
545 /* Node in 'ofport_dpif''s 'priorities' map. Used to maintain a map from
546 * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
547 * traffic egressing the 'ofport' with that priority should be marked with. */
548 struct priority_to_dscp {
549 struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'priorities' map. */
550 uint32_t priority; /* Priority of this queue (see struct flow). */
552 uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */
555 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
557 * This is deprecated. It is only for compatibility with broken device drivers
558 * in old versions of Linux that do not properly support VLANs when VLAN
559 * devices are not used. When broken device drivers are no longer in
560 * widespread use, we will delete these interfaces. */
561 struct vlan_splinter {
562 struct hmap_node realdev_vid_node;
563 struct hmap_node vlandev_node;
564 uint16_t realdev_ofp_port;
565 uint16_t vlandev_ofp_port;
569 static uint16_t vsp_realdev_to_vlandev(const struct ofproto_dpif *,
570 uint16_t realdev_ofp_port,
572 static bool vsp_adjust_flow(const struct ofproto_dpif *, struct flow *);
573 static void vsp_remove(struct ofport_dpif *);
574 static void vsp_add(struct ofport_dpif *, uint16_t realdev_ofp_port, int vid);
576 static uint32_t ofp_port_to_odp_port(const struct ofproto_dpif *,
578 static uint16_t odp_port_to_ofp_port(const struct ofproto_dpif *,
581 static struct ofport_dpif *
582 ofport_dpif_cast(const struct ofport *ofport)
584 ovs_assert(ofport->ofproto->ofproto_class == &ofproto_dpif_class);
585 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
588 static void port_run(struct ofport_dpif *);
589 static void port_run_fast(struct ofport_dpif *);
590 static void port_wait(struct ofport_dpif *);
591 static int set_cfm(struct ofport *, const struct cfm_settings *);
592 static void ofport_clear_priorities(struct ofport_dpif *);
593 static void run_fast_rl(void);
595 struct dpif_completion {
596 struct list list_node;
597 struct ofoperation *op;
600 /* Extra information about a classifier table.
601 * Currently used just for optimized flow revalidation. */
603 /* If either of these is nonnull, then this table has a form that allows
604 * flows to be tagged to avoid revalidating most flows for the most common
605 * kinds of flow table changes. */
606 struct cls_table *catchall_table; /* Table that wildcards all fields. */
607 struct cls_table *other_table; /* Table with any other wildcard set. */
608 uint32_t basis; /* Keeps each table's tags separate. */
611 /* Reasons that we might need to revalidate every facet, and corresponding
614 * A value of 0 means that there is no need to revalidate.
616 * It would be nice to have some cleaner way to integrate with coverage
617 * counters, but with only a few reasons I guess this is good enough for
619 enum revalidate_reason {
620 REV_RECONFIGURE = 1, /* Switch configuration changed. */
621 REV_STP, /* Spanning tree protocol port status change. */
622 REV_PORT_TOGGLED, /* Port enabled or disabled by CFM, LACP, ...*/
623 REV_FLOW_TABLE, /* Flow table changed. */
624 REV_INCONSISTENCY /* Facet self-check failed. */
626 COVERAGE_DEFINE(rev_reconfigure);
627 COVERAGE_DEFINE(rev_stp);
628 COVERAGE_DEFINE(rev_port_toggled);
629 COVERAGE_DEFINE(rev_flow_table);
630 COVERAGE_DEFINE(rev_inconsistency);
632 /* Drop keys are odp flow keys which have drop flows installed in the kernel.
633 * These are datapath flows which have no associated ofproto, if they did we
634 * would use facets. */
636 struct hmap_node hmap_node;
641 /* All datapaths of a given type share a single dpif backer instance. */
646 struct timer next_expiration;
647 struct hmap odp_to_ofport_map; /* ODP port to ofport mapping. */
649 struct simap tnl_backers; /* Set of dpif ports backing tunnels. */
651 /* Facet revalidation flags applying to facets which use this backer. */
652 enum revalidate_reason need_revalidate; /* Revalidate every facet. */
653 struct tag_set revalidate_set; /* Revalidate only matching facets. */
655 struct hmap drop_keys; /* Set of dropped odp keys. */
656 bool recv_set_enable; /* Enables or disables receiving packets. */
659 /* All existing ofproto_backer instances, indexed by ofproto->up.type. */
660 static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers);
662 static void drop_key_clear(struct dpif_backer *);
663 static struct ofport_dpif *
664 odp_port_to_ofport(const struct dpif_backer *, uint32_t odp_port);
666 static void dpif_stats_update_hit_count(struct ofproto_dpif *ofproto,
668 struct avg_subfacet_rates {
669 double add_rate; /* Moving average of new flows created per minute. */
670 double del_rate; /* Moving average of flows deleted per minute. */
672 static void show_dp_rates(struct ds *ds, const char *heading,
673 const struct avg_subfacet_rates *rates);
674 static void exp_mavg(double *avg, int base, double new);
676 struct ofproto_dpif {
677 struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
679 struct dpif_backer *backer;
681 /* Special OpenFlow rules. */
682 struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */
683 struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */
686 struct netflow *netflow;
687 struct dpif_sflow *sflow;
688 struct dpif_ipfix *ipfix;
689 struct hmap bundles; /* Contains "struct ofbundle"s. */
690 struct mac_learning *ml;
691 struct ofmirror *mirrors[MAX_MIRRORS];
693 bool has_bonded_bundles;
697 struct hmap subfacets;
698 struct governor *governor;
699 long long int consistency_rl;
702 struct table_dpif tables[N_TABLES];
704 /* Support for debugging async flow mods. */
705 struct list completions;
707 bool has_bundle_action; /* True when the first bundle action appears. */
708 struct netdev_stats stats; /* To account packets generated and consumed in
713 long long int stp_last_tick;
715 /* VLAN splinters. */
716 struct hmap realdev_vid_map; /* (realdev,vid) -> vlandev. */
717 struct hmap vlandev_map; /* vlandev -> (realdev,vid). */
720 struct sset ports; /* Set of standard port names. */
721 struct sset ghost_ports; /* Ports with no datapath port. */
722 struct sset port_poll_set; /* Queued names for port_poll() reply. */
723 int port_poll_errno; /* Last errno for port_poll() reply. */
725 /* Per ofproto's dpif stats. */
729 /* Subfacet statistics.
731 * These keep track of the total number of subfacets added and deleted and
732 * flow life span. They are useful for computing the flow rates stats
733 * exposed via "ovs-appctl dpif/show". The goal is to learn about
734 * traffic patterns in ways that we can use later to improve Open vSwitch
735 * performance in new situations. */
736 long long int created; /* Time when it is created. */
737 unsigned int max_n_subfacet; /* Maximum number of flows */
739 /* The average number of subfacets... */
740 struct avg_subfacet_rates hourly; /* ...over the last hour. */
741 struct avg_subfacet_rates daily; /* ...over the last day. */
742 long long int last_minute; /* Last time 'hourly' was updated. */
744 /* Number of subfacets added or deleted since 'last_minute'. */
745 unsigned int subfacet_add_count;
746 unsigned int subfacet_del_count;
748 /* Number of subfacets added or deleted from 'created' to 'last_minute.' */
749 unsigned long long int total_subfacet_add_count;
750 unsigned long long int total_subfacet_del_count;
752 /* Sum of the number of milliseconds that each subfacet existed,
753 * over the subfacets that have been added and then later deleted. */
754 unsigned long long int total_subfacet_life_span;
756 /* Incremented by the number of currently existing subfacets, each
757 * time we pull statistics from the kernel. */
758 unsigned long long int total_subfacet_count;
760 /* Number of times we pull statistics from the kernel. */
761 unsigned long long int n_update_stats;
763 static unsigned long long int avg_subfacet_life_span(
764 const struct ofproto_dpif *);
765 static double avg_subfacet_count(const struct ofproto_dpif *ofproto);
766 static void update_moving_averages(struct ofproto_dpif *ofproto);
767 static void dpif_stats_update_hit_count(struct ofproto_dpif *ofproto,
769 static void update_max_subfacet_count(struct ofproto_dpif *ofproto);
771 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
772 * for debugging the asynchronous flow_mod implementation.) */
775 /* All existing ofproto_dpif instances, indexed by ->up.name. */
776 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
778 static void ofproto_dpif_unixctl_init(void);
780 static struct ofproto_dpif *
781 ofproto_dpif_cast(const struct ofproto *ofproto)
783 ovs_assert(ofproto->ofproto_class == &ofproto_dpif_class);
784 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
787 static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *,
789 static struct ofport_dpif *get_odp_port(const struct ofproto_dpif *,
791 static void ofproto_trace(struct ofproto_dpif *, const struct flow *,
792 const struct ofpbuf *,
793 const struct initial_vals *, struct ds *);
795 /* Packet processing. */
796 static void update_learning_table(struct ofproto_dpif *,
797 const struct flow *, int vlan,
800 #define FLOW_MISS_MAX_BATCH 50
801 static int handle_upcalls(struct dpif_backer *, unsigned int max_batch);
803 /* Flow expiration. */
804 static int expire(struct dpif_backer *);
807 static void send_netflow_active_timeouts(struct ofproto_dpif *);
810 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
811 static size_t compose_sflow_action(const struct ofproto_dpif *,
812 struct ofpbuf *odp_actions,
813 const struct flow *, uint32_t odp_port);
814 static void compose_ipfix_action(const struct ofproto_dpif *,
815 struct ofpbuf *odp_actions,
816 const struct flow *);
817 static void add_mirror_actions(struct xlate_ctx *ctx,
818 const struct flow *flow);
819 /* Global variables. */
820 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
822 /* Initial mappings of port to bridge mappings. */
823 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
825 /* Factory functions. */
828 init(const struct shash *iface_hints)
830 struct shash_node *node;
832 /* Make a local copy, since we don't own 'iface_hints' elements. */
833 SHASH_FOR_EACH(node, iface_hints) {
834 const struct iface_hint *orig_hint = node->data;
835 struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
837 new_hint->br_name = xstrdup(orig_hint->br_name);
838 new_hint->br_type = xstrdup(orig_hint->br_type);
839 new_hint->ofp_port = orig_hint->ofp_port;
841 shash_add(&init_ofp_ports, node->name, new_hint);
846 enumerate_types(struct sset *types)
848 dp_enumerate_types(types);
852 enumerate_names(const char *type, struct sset *names)
854 struct ofproto_dpif *ofproto;
857 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
858 if (strcmp(type, ofproto->up.type)) {
861 sset_add(names, ofproto->up.name);
868 del(const char *type, const char *name)
873 error = dpif_open(name, type, &dpif);
875 error = dpif_delete(dpif);
882 port_open_type(const char *datapath_type, const char *port_type)
884 return dpif_port_open_type(datapath_type, port_type);
887 /* Type functions. */
889 static struct ofproto_dpif *
890 lookup_ofproto_dpif_by_port_name(const char *name)
892 struct ofproto_dpif *ofproto;
894 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
895 if (sset_contains(&ofproto->ports, name)) {
904 type_run(const char *type)
906 static long long int push_timer = LLONG_MIN;
907 struct dpif_backer *backer;
911 backer = shash_find_data(&all_dpif_backers, type);
913 /* This is not necessarily a problem, since backers are only
914 * created on demand. */
918 dpif_run(backer->dpif);
920 /* The most natural place to push facet statistics is when they're pulled
921 * from the datapath. However, when there are many flows in the datapath,
922 * this expensive operation can occur so frequently, that it reduces our
923 * ability to quickly set up flows. To reduce the cost, we push statistics
925 if (time_msec() > push_timer) {
926 push_timer = time_msec() + 2000;
930 /* If vswitchd started with other_config:flow_restore_wait set as "true",
931 * and the configuration has now changed to "false", enable receiving
932 * packets from the datapath. */
933 if (!backer->recv_set_enable && !ofproto_get_flow_restore_wait()) {
934 backer->recv_set_enable = true;
936 error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
938 VLOG_ERR("Failed to enable receiving packets in dpif.");
941 dpif_flow_flush(backer->dpif);
942 backer->need_revalidate = REV_RECONFIGURE;
945 if (backer->need_revalidate
946 || !tag_set_is_empty(&backer->revalidate_set)) {
947 struct tag_set revalidate_set = backer->revalidate_set;
948 bool need_revalidate = backer->need_revalidate;
949 struct ofproto_dpif *ofproto;
950 struct simap_node *node;
951 struct simap tmp_backers;
953 /* Handle tunnel garbage collection. */
954 simap_init(&tmp_backers);
955 simap_swap(&backer->tnl_backers, &tmp_backers);
957 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
958 struct ofport_dpif *iter;
960 if (backer != ofproto->backer) {
964 HMAP_FOR_EACH (iter, up.hmap_node, &ofproto->up.ports) {
967 if (!iter->tnl_port) {
971 dp_port = netdev_vport_get_dpif_port(iter->up.netdev);
972 node = simap_find(&tmp_backers, dp_port);
974 simap_put(&backer->tnl_backers, dp_port, node->data);
975 simap_delete(&tmp_backers, node);
976 node = simap_find(&backer->tnl_backers, dp_port);
978 node = simap_find(&backer->tnl_backers, dp_port);
980 uint32_t odp_port = UINT32_MAX;
982 if (!dpif_port_add(backer->dpif, iter->up.netdev,
984 simap_put(&backer->tnl_backers, dp_port, odp_port);
985 node = simap_find(&backer->tnl_backers, dp_port);
990 iter->odp_port = node ? node->data : OVSP_NONE;
991 if (tnl_port_reconfigure(&iter->up, iter->odp_port,
993 backer->need_revalidate = REV_RECONFIGURE;
998 SIMAP_FOR_EACH (node, &tmp_backers) {
999 dpif_port_del(backer->dpif, node->data);
1001 simap_destroy(&tmp_backers);
1003 switch (backer->need_revalidate) {
1004 case REV_RECONFIGURE: COVERAGE_INC(rev_reconfigure); break;
1005 case REV_STP: COVERAGE_INC(rev_stp); break;
1006 case REV_PORT_TOGGLED: COVERAGE_INC(rev_port_toggled); break;
1007 case REV_FLOW_TABLE: COVERAGE_INC(rev_flow_table); break;
1008 case REV_INCONSISTENCY: COVERAGE_INC(rev_inconsistency); break;
1011 if (backer->need_revalidate) {
1012 /* Clear the drop_keys in case we should now be accepting some
1013 * formerly dropped flows. */
1014 drop_key_clear(backer);
1017 /* Clear the revalidation flags. */
1018 tag_set_init(&backer->revalidate_set);
1019 backer->need_revalidate = 0;
1021 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1022 struct facet *facet, *next;
1024 if (ofproto->backer != backer) {
1028 HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &ofproto->facets) {
1030 || tag_set_intersects(&revalidate_set, facet->xout.tags)) {
1031 facet_revalidate(facet);
1038 if (!backer->recv_set_enable) {
1039 /* Wake up before a max of 1000ms. */
1040 timer_set_duration(&backer->next_expiration, 1000);
1041 } else if (timer_expired(&backer->next_expiration)) {
1042 int delay = expire(backer);
1043 timer_set_duration(&backer->next_expiration, delay);
1046 /* Check for port changes in the dpif. */
1047 while ((error = dpif_port_poll(backer->dpif, &devname)) == 0) {
1048 struct ofproto_dpif *ofproto;
1049 struct dpif_port port;
1051 /* Don't report on the datapath's device. */
1052 if (!strcmp(devname, dpif_base_name(backer->dpif))) {
1056 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
1057 &all_ofproto_dpifs) {
1058 if (simap_contains(&ofproto->backer->tnl_backers, devname)) {
1063 ofproto = lookup_ofproto_dpif_by_port_name(devname);
1064 if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
1065 /* The port was removed. If we know the datapath,
1066 * report it through poll_set(). If we don't, it may be
1067 * notifying us of a removal we initiated, so ignore it.
1068 * If there's a pending ENOBUFS, let it stand, since
1069 * everything will be reevaluated. */
1070 if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
1071 sset_add(&ofproto->port_poll_set, devname);
1072 ofproto->port_poll_errno = 0;
1074 } else if (!ofproto) {
1075 /* The port was added, but we don't know with which
1076 * ofproto we should associate it. Delete it. */
1077 dpif_port_del(backer->dpif, port.port_no);
1079 dpif_port_destroy(&port);
1085 if (error != EAGAIN) {
1086 struct ofproto_dpif *ofproto;
1088 /* There was some sort of error, so propagate it to all
1089 * ofprotos that use this backer. */
1090 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
1091 &all_ofproto_dpifs) {
1092 if (ofproto->backer == backer) {
1093 sset_clear(&ofproto->port_poll_set);
1094 ofproto->port_poll_errno = error;
1103 dpif_backer_run_fast(struct dpif_backer *backer, int max_batch)
1107 /* If recv_set_enable is false, we should not handle upcalls. */
1108 if (!backer->recv_set_enable) {
1112 /* Handle one or more batches of upcalls, until there's nothing left to do
1113 * or until we do a fixed total amount of work.
1115 * We do work in batches because it can be much cheaper to set up a number
1116 * of flows and fire off their patches all at once. We do multiple batches
1117 * because in some cases handling a packet can cause another packet to be
1118 * queued almost immediately as part of the return flow. Both
1119 * optimizations can make major improvements on some benchmarks and
1120 * presumably for real traffic as well. */
1122 while (work < max_batch) {
1123 int retval = handle_upcalls(backer, max_batch - work);
1134 type_run_fast(const char *type)
1136 struct dpif_backer *backer;
1138 backer = shash_find_data(&all_dpif_backers, type);
1140 /* This is not necessarily a problem, since backers are only
1141 * created on demand. */
1145 return dpif_backer_run_fast(backer, FLOW_MISS_MAX_BATCH);
1151 static long long int port_rl = LLONG_MIN;
1152 static unsigned int backer_rl = 0;
1154 if (time_msec() >= port_rl) {
1155 struct ofproto_dpif *ofproto;
1156 struct ofport_dpif *ofport;
1158 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1160 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1161 port_run_fast(ofport);
1164 port_rl = time_msec() + 200;
1167 /* XXX: We have to be careful not to do too much work in this function. If
1168 * we call dpif_backer_run_fast() too often, or with too large a batch,
1169 * performance improves signifcantly, but at a cost. It's possible for the
1170 * number of flows in the datapath to increase without bound, and for poll
1171 * loops to take 10s of seconds. The correct solution to this problem,
1172 * long term, is to separate flow miss handling into it's own thread so it
1173 * isn't affected by revalidations, and expirations. Until then, this is
1174 * the best we can do. */
1175 if (++backer_rl >= 10) {
1176 struct shash_node *node;
1179 SHASH_FOR_EACH (node, &all_dpif_backers) {
1180 dpif_backer_run_fast(node->data, 1);
1186 type_wait(const char *type)
1188 struct dpif_backer *backer;
1190 backer = shash_find_data(&all_dpif_backers, type);
1192 /* This is not necessarily a problem, since backers are only
1193 * created on demand. */
1197 timer_wait(&backer->next_expiration);
1200 /* Basic life-cycle. */
1202 static int add_internal_flows(struct ofproto_dpif *);
1204 static struct ofproto *
1207 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
1208 return &ofproto->up;
1212 dealloc(struct ofproto *ofproto_)
1214 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1219 close_dpif_backer(struct dpif_backer *backer)
1221 struct shash_node *node;
1223 ovs_assert(backer->refcount > 0);
1225 if (--backer->refcount) {
1229 drop_key_clear(backer);
1230 hmap_destroy(&backer->drop_keys);
1232 simap_destroy(&backer->tnl_backers);
1233 hmap_destroy(&backer->odp_to_ofport_map);
1234 node = shash_find(&all_dpif_backers, backer->type);
1236 shash_delete(&all_dpif_backers, node);
1237 dpif_close(backer->dpif);
1242 /* Datapath port slated for removal from datapath. */
1243 struct odp_garbage {
1244 struct list list_node;
1249 open_dpif_backer(const char *type, struct dpif_backer **backerp)
1251 struct dpif_backer *backer;
1252 struct dpif_port_dump port_dump;
1253 struct dpif_port port;
1254 struct shash_node *node;
1255 struct list garbage_list;
1256 struct odp_garbage *garbage, *next;
1262 backer = shash_find_data(&all_dpif_backers, type);
1269 backer_name = xasprintf("ovs-%s", type);
1271 /* Remove any existing datapaths, since we assume we're the only
1272 * userspace controlling the datapath. */
1274 dp_enumerate_names(type, &names);
1275 SSET_FOR_EACH(name, &names) {
1276 struct dpif *old_dpif;
1278 /* Don't remove our backer if it exists. */
1279 if (!strcmp(name, backer_name)) {
1283 if (dpif_open(name, type, &old_dpif)) {
1284 VLOG_WARN("couldn't open old datapath %s to remove it", name);
1286 dpif_delete(old_dpif);
1287 dpif_close(old_dpif);
1290 sset_destroy(&names);
1292 backer = xmalloc(sizeof *backer);
1294 error = dpif_create_and_open(backer_name, type, &backer->dpif);
1297 VLOG_ERR("failed to open datapath of type %s: %s", type,
1303 backer->type = xstrdup(type);
1304 backer->refcount = 1;
1305 hmap_init(&backer->odp_to_ofport_map);
1306 hmap_init(&backer->drop_keys);
1307 timer_set_duration(&backer->next_expiration, 1000);
1308 backer->need_revalidate = 0;
1309 simap_init(&backer->tnl_backers);
1310 tag_set_init(&backer->revalidate_set);
1311 backer->recv_set_enable = !ofproto_get_flow_restore_wait();
1314 if (backer->recv_set_enable) {
1315 dpif_flow_flush(backer->dpif);
1318 /* Loop through the ports already on the datapath and remove any
1319 * that we don't need anymore. */
1320 list_init(&garbage_list);
1321 dpif_port_dump_start(&port_dump, backer->dpif);
1322 while (dpif_port_dump_next(&port_dump, &port)) {
1323 node = shash_find(&init_ofp_ports, port.name);
1324 if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
1325 garbage = xmalloc(sizeof *garbage);
1326 garbage->odp_port = port.port_no;
1327 list_push_front(&garbage_list, &garbage->list_node);
1330 dpif_port_dump_done(&port_dump);
1332 LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) {
1333 dpif_port_del(backer->dpif, garbage->odp_port);
1334 list_remove(&garbage->list_node);
1338 shash_add(&all_dpif_backers, type, backer);
1340 error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
1342 VLOG_ERR("failed to listen on datapath of type %s: %s",
1343 type, strerror(error));
1344 close_dpif_backer(backer);
1352 construct(struct ofproto *ofproto_)
1354 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1355 struct shash_node *node, *next;
1360 error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1365 max_ports = dpif_get_max_ports(ofproto->backer->dpif);
1366 ofproto_init_max_ports(ofproto_, MIN(max_ports, OFPP_MAX));
1368 ofproto->netflow = NULL;
1369 ofproto->sflow = NULL;
1370 ofproto->ipfix = NULL;
1371 ofproto->stp = NULL;
1372 hmap_init(&ofproto->bundles);
1373 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1374 for (i = 0; i < MAX_MIRRORS; i++) {
1375 ofproto->mirrors[i] = NULL;
1377 ofproto->has_bonded_bundles = false;
1379 hmap_init(&ofproto->facets);
1380 hmap_init(&ofproto->subfacets);
1381 ofproto->governor = NULL;
1382 ofproto->consistency_rl = LLONG_MIN;
1384 for (i = 0; i < N_TABLES; i++) {
1385 struct table_dpif *table = &ofproto->tables[i];
1387 table->catchall_table = NULL;
1388 table->other_table = NULL;
1389 table->basis = random_uint32();
1392 list_init(&ofproto->completions);
1394 ofproto_dpif_unixctl_init();
1396 ofproto->has_mirrors = false;
1397 ofproto->has_bundle_action = false;
1399 hmap_init(&ofproto->vlandev_map);
1400 hmap_init(&ofproto->realdev_vid_map);
1402 sset_init(&ofproto->ports);
1403 sset_init(&ofproto->ghost_ports);
1404 sset_init(&ofproto->port_poll_set);
1405 ofproto->port_poll_errno = 0;
1407 SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1408 struct iface_hint *iface_hint = node->data;
1410 if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1411 /* Check if the datapath already has this port. */
1412 if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1413 sset_add(&ofproto->ports, node->name);
1416 free(iface_hint->br_name);
1417 free(iface_hint->br_type);
1419 shash_delete(&init_ofp_ports, node);
1423 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1424 hash_string(ofproto->up.name, 0));
1425 memset(&ofproto->stats, 0, sizeof ofproto->stats);
1427 ofproto_init_tables(ofproto_, N_TABLES);
1428 error = add_internal_flows(ofproto);
1429 ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1432 ofproto->n_missed = 0;
1434 ofproto->max_n_subfacet = 0;
1435 ofproto->created = time_msec();
1436 ofproto->last_minute = ofproto->created;
1437 memset(&ofproto->hourly, 0, sizeof ofproto->hourly);
1438 memset(&ofproto->daily, 0, sizeof ofproto->daily);
1439 ofproto->subfacet_add_count = 0;
1440 ofproto->subfacet_del_count = 0;
1441 ofproto->total_subfacet_add_count = 0;
1442 ofproto->total_subfacet_del_count = 0;
1443 ofproto->total_subfacet_life_span = 0;
1444 ofproto->total_subfacet_count = 0;
1445 ofproto->n_update_stats = 0;
1451 add_internal_flow(struct ofproto_dpif *ofproto, int id,
1452 const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1454 struct ofputil_flow_mod fm;
1457 match_init_catchall(&fm.match);
1459 match_set_reg(&fm.match, 0, id);
1460 fm.new_cookie = htonll(0);
1461 fm.cookie = htonll(0);
1462 fm.cookie_mask = htonll(0);
1463 fm.table_id = TBL_INTERNAL;
1464 fm.command = OFPFC_ADD;
1465 fm.idle_timeout = 0;
1466 fm.hard_timeout = 0;
1470 fm.ofpacts = ofpacts->data;
1471 fm.ofpacts_len = ofpacts->size;
1473 error = ofproto_flow_mod(&ofproto->up, &fm);
1475 VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
1476 id, ofperr_to_string(error));
1480 *rulep = rule_dpif_lookup__(ofproto, &fm.match.flow, TBL_INTERNAL);
1481 ovs_assert(*rulep != NULL);
1487 add_internal_flows(struct ofproto_dpif *ofproto)
1489 struct ofpact_controller *controller;
1490 uint64_t ofpacts_stub[128 / 8];
1491 struct ofpbuf ofpacts;
1495 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1498 controller = ofpact_put_CONTROLLER(&ofpacts);
1499 controller->max_len = UINT16_MAX;
1500 controller->controller_id = 0;
1501 controller->reason = OFPR_NO_MATCH;
1502 ofpact_pad(&ofpacts);
1504 error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule);
1509 ofpbuf_clear(&ofpacts);
1510 error = add_internal_flow(ofproto, id++, &ofpacts,
1511 &ofproto->no_packet_in_rule);
1516 complete_operations(struct ofproto_dpif *ofproto)
1518 struct dpif_completion *c, *next;
1520 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
1521 ofoperation_complete(c->op, 0);
1522 list_remove(&c->list_node);
1528 destruct(struct ofproto *ofproto_)
1530 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1531 struct rule_dpif *rule, *next_rule;
1532 struct oftable *table;
1535 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1536 complete_operations(ofproto);
1538 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1539 struct cls_cursor cursor;
1541 cls_cursor_init(&cursor, &table->cls, NULL);
1542 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1543 ofproto_rule_destroy(&rule->up);
1547 for (i = 0; i < MAX_MIRRORS; i++) {
1548 mirror_destroy(ofproto->mirrors[i]);
1551 netflow_destroy(ofproto->netflow);
1552 dpif_sflow_destroy(ofproto->sflow);
1553 hmap_destroy(&ofproto->bundles);
1554 mac_learning_destroy(ofproto->ml);
1556 hmap_destroy(&ofproto->facets);
1557 hmap_destroy(&ofproto->subfacets);
1558 governor_destroy(ofproto->governor);
1560 hmap_destroy(&ofproto->vlandev_map);
1561 hmap_destroy(&ofproto->realdev_vid_map);
1563 sset_destroy(&ofproto->ports);
1564 sset_destroy(&ofproto->ghost_ports);
1565 sset_destroy(&ofproto->port_poll_set);
1567 close_dpif_backer(ofproto->backer);
1571 run_fast(struct ofproto *ofproto_)
1573 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1574 struct ofport_dpif *ofport;
1576 /* Do not perform any periodic activity required by 'ofproto' while
1577 * waiting for flow restore to complete. */
1578 if (ofproto_get_flow_restore_wait()) {
1582 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1583 port_run_fast(ofport);
1590 run(struct ofproto *ofproto_)
1592 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1593 struct ofport_dpif *ofport;
1594 struct ofbundle *bundle;
1598 complete_operations(ofproto);
1601 /* Do not perform any periodic activity below required by 'ofproto' while
1602 * waiting for flow restore to complete. */
1603 if (ofproto_get_flow_restore_wait()) {
1607 error = run_fast(ofproto_);
1612 if (ofproto->netflow) {
1613 if (netflow_run(ofproto->netflow)) {
1614 send_netflow_active_timeouts(ofproto);
1617 if (ofproto->sflow) {
1618 dpif_sflow_run(ofproto->sflow);
1621 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1624 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1629 mac_learning_run(ofproto->ml, &ofproto->backer->revalidate_set);
1631 /* Check the consistency of a random facet, to aid debugging. */
1632 if (time_msec() >= ofproto->consistency_rl
1633 && !hmap_is_empty(&ofproto->facets)
1634 && !ofproto->backer->need_revalidate) {
1635 struct facet *facet;
1637 ofproto->consistency_rl = time_msec() + 250;
1639 facet = CONTAINER_OF(hmap_random_node(&ofproto->facets),
1640 struct facet, hmap_node);
1641 if (!tag_set_intersects(&ofproto->backer->revalidate_set,
1642 facet->xout.tags)) {
1643 if (!facet_check_consistency(facet)) {
1644 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
1649 if (ofproto->governor) {
1652 governor_run(ofproto->governor);
1654 /* If the governor has shrunk to its minimum size and the number of
1655 * subfacets has dwindled, then drop the governor entirely.
1657 * For hysteresis, the number of subfacets to drop the governor is
1658 * smaller than the number needed to trigger its creation. */
1659 n_subfacets = hmap_count(&ofproto->subfacets);
1660 if (n_subfacets * 4 < ofproto->up.flow_eviction_threshold
1661 && governor_is_idle(ofproto->governor)) {
1662 governor_destroy(ofproto->governor);
1663 ofproto->governor = NULL;
1671 wait(struct ofproto *ofproto_)
1673 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1674 struct ofport_dpif *ofport;
1675 struct ofbundle *bundle;
1677 if (!clogged && !list_is_empty(&ofproto->completions)) {
1678 poll_immediate_wake();
1681 if (ofproto_get_flow_restore_wait()) {
1685 dpif_wait(ofproto->backer->dpif);
1686 dpif_recv_wait(ofproto->backer->dpif);
1687 if (ofproto->sflow) {
1688 dpif_sflow_wait(ofproto->sflow);
1690 if (!tag_set_is_empty(&ofproto->backer->revalidate_set)) {
1691 poll_immediate_wake();
1693 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1696 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1697 bundle_wait(bundle);
1699 if (ofproto->netflow) {
1700 netflow_wait(ofproto->netflow);
1702 mac_learning_wait(ofproto->ml);
1704 if (ofproto->backer->need_revalidate) {
1705 /* Shouldn't happen, but if it does just go around again. */
1706 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1707 poll_immediate_wake();
1709 if (ofproto->governor) {
1710 governor_wait(ofproto->governor);
1715 get_memory_usage(const struct ofproto *ofproto_, struct simap *usage)
1717 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1719 simap_increase(usage, "facets", hmap_count(&ofproto->facets));
1720 simap_increase(usage, "subfacets", hmap_count(&ofproto->subfacets));
1724 flush(struct ofproto *ofproto_)
1726 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1727 struct subfacet *subfacet, *next_subfacet;
1728 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
1732 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
1733 &ofproto->subfacets) {
1734 if (subfacet->path != SF_NOT_INSTALLED) {
1735 batch[n_batch++] = subfacet;
1736 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
1737 subfacet_destroy_batch(ofproto, batch, n_batch);
1741 subfacet_destroy(subfacet);
1746 subfacet_destroy_batch(ofproto, batch, n_batch);
1751 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1752 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1754 *arp_match_ip = true;
1755 *actions = (OFPUTIL_A_OUTPUT |
1756 OFPUTIL_A_SET_VLAN_VID |
1757 OFPUTIL_A_SET_VLAN_PCP |
1758 OFPUTIL_A_STRIP_VLAN |
1759 OFPUTIL_A_SET_DL_SRC |
1760 OFPUTIL_A_SET_DL_DST |
1761 OFPUTIL_A_SET_NW_SRC |
1762 OFPUTIL_A_SET_NW_DST |
1763 OFPUTIL_A_SET_NW_TOS |
1764 OFPUTIL_A_SET_TP_SRC |
1765 OFPUTIL_A_SET_TP_DST |
1770 get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots)
1772 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1773 struct dpif_dp_stats s;
1774 uint64_t n_miss, n_no_pkt_in, n_bytes;
1777 strcpy(ots->name, "classifier");
1779 dpif_get_dp_stats(ofproto->backer->dpif, &s);
1780 rule_get_stats(&ofproto->miss_rule->up, &n_miss, &n_bytes);
1781 rule_get_stats(&ofproto->no_packet_in_rule->up, &n_no_pkt_in, &n_bytes);
1783 n_lookup = s.n_hit + s.n_missed;
1784 ots->lookup_count = htonll(n_lookup);
1785 ots->matched_count = htonll(n_lookup - n_miss - n_no_pkt_in);
1788 static struct ofport *
1791 struct ofport_dpif *port = xmalloc(sizeof *port);
1796 port_dealloc(struct ofport *port_)
1798 struct ofport_dpif *port = ofport_dpif_cast(port_);
1803 port_construct(struct ofport *port_)
1805 struct ofport_dpif *port = ofport_dpif_cast(port_);
1806 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1807 const struct netdev *netdev = port->up.netdev;
1808 struct dpif_port dpif_port;
1811 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1812 port->bundle = NULL;
1814 port->tag = tag_create_random();
1815 port->may_enable = true;
1816 port->stp_port = NULL;
1817 port->stp_state = STP_DISABLED;
1818 port->tnl_port = NULL;
1819 hmap_init(&port->priorities);
1820 port->realdev_ofp_port = 0;
1821 port->vlandev_vid = 0;
1822 port->carrier_seq = netdev_get_carrier_resets(netdev);
1824 if (netdev_vport_is_patch(netdev)) {
1825 /* By bailing out here, we don't submit the port to the sFlow module
1826 * to be considered for counter polling export. This is correct
1827 * because the patch port represents an interface that sFlow considers
1828 * to be "internal" to the switch as a whole, and therefore not an
1829 * candidate for counter polling. */
1830 port->odp_port = OVSP_NONE;
1834 error = dpif_port_query_by_name(ofproto->backer->dpif,
1835 netdev_vport_get_dpif_port(netdev),
1841 port->odp_port = dpif_port.port_no;
1843 if (netdev_get_tunnel_config(netdev)) {
1844 port->tnl_port = tnl_port_add(&port->up, port->odp_port);
1846 /* Sanity-check that a mapping doesn't already exist. This
1847 * shouldn't happen for non-tunnel ports. */
1848 if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1849 VLOG_ERR("port %s already has an OpenFlow port number",
1851 dpif_port_destroy(&dpif_port);
1855 hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1856 hash_int(port->odp_port, 0));
1858 dpif_port_destroy(&dpif_port);
1860 if (ofproto->sflow) {
1861 dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1868 port_destruct(struct ofport *port_)
1870 struct ofport_dpif *port = ofport_dpif_cast(port_);
1871 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1872 const char *dp_port_name = netdev_vport_get_dpif_port(port->up.netdev);
1873 const char *devname = netdev_get_name(port->up.netdev);
1875 if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1876 /* The underlying device is still there, so delete it. This
1877 * happens when the ofproto is being destroyed, since the caller
1878 * assumes that removal of attached ports will happen as part of
1880 if (!port->tnl_port) {
1881 dpif_port_del(ofproto->backer->dpif, port->odp_port);
1883 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1886 if (port->odp_port != OVSP_NONE && !port->tnl_port) {
1887 hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1890 tnl_port_del(port->tnl_port);
1891 sset_find_and_delete(&ofproto->ports, devname);
1892 sset_find_and_delete(&ofproto->ghost_ports, devname);
1893 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1894 bundle_remove(port_);
1895 set_cfm(port_, NULL);
1896 if (ofproto->sflow) {
1897 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1900 ofport_clear_priorities(port);
1901 hmap_destroy(&port->priorities);
1905 port_modified(struct ofport *port_)
1907 struct ofport_dpif *port = ofport_dpif_cast(port_);
1909 if (port->bundle && port->bundle->bond) {
1910 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1914 cfm_set_netdev(port->cfm, port->up.netdev);
1919 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1921 struct ofport_dpif *port = ofport_dpif_cast(port_);
1922 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1923 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1925 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1926 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1927 OFPUTIL_PC_NO_PACKET_IN)) {
1928 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1930 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1931 bundle_update(port->bundle);
1937 set_sflow(struct ofproto *ofproto_,
1938 const struct ofproto_sflow_options *sflow_options)
1940 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1941 struct dpif_sflow *ds = ofproto->sflow;
1943 if (sflow_options) {
1945 struct ofport_dpif *ofport;
1947 ds = ofproto->sflow = dpif_sflow_create();
1948 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1949 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1951 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1953 dpif_sflow_set_options(ds, sflow_options);
1956 dpif_sflow_destroy(ds);
1957 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1958 ofproto->sflow = NULL;
1966 struct ofproto *ofproto_,
1967 const struct ofproto_ipfix_bridge_exporter_options *bridge_exporter_options,
1968 const struct ofproto_ipfix_flow_exporter_options *flow_exporters_options,
1969 size_t n_flow_exporters_options)
1971 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1972 struct dpif_ipfix *di = ofproto->ipfix;
1974 if (bridge_exporter_options || flow_exporters_options) {
1976 di = ofproto->ipfix = dpif_ipfix_create();
1978 dpif_ipfix_set_options(
1979 di, bridge_exporter_options, flow_exporters_options,
1980 n_flow_exporters_options);
1983 dpif_ipfix_destroy(di);
1984 ofproto->ipfix = NULL;
1991 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1993 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2000 struct ofproto_dpif *ofproto;
2002 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2003 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2004 ofport->cfm = cfm_create(ofport->up.netdev);
2007 if (cfm_configure(ofport->cfm, s)) {
2013 cfm_destroy(ofport->cfm);
2019 get_cfm_status(const struct ofport *ofport_,
2020 struct ofproto_cfm_status *status)
2022 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2025 status->faults = cfm_get_fault(ofport->cfm);
2026 status->remote_opstate = cfm_get_opup(ofport->cfm);
2027 status->health = cfm_get_health(ofport->cfm);
2028 cfm_get_remote_mpids(ofport->cfm, &status->rmps, &status->n_rmps);
2035 /* Spanning Tree. */
2038 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
2040 struct ofproto_dpif *ofproto = ofproto_;
2041 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
2042 struct ofport_dpif *ofport;
2044 ofport = stp_port_get_aux(sp);
2046 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
2047 ofproto->up.name, port_num);
2049 struct eth_header *eth = pkt->l2;
2051 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
2052 if (eth_addr_is_zero(eth->eth_src)) {
2053 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
2054 "with unknown MAC", ofproto->up.name, port_num);
2056 send_packet(ofport, pkt);
2062 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
2064 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
2066 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2068 /* Only revalidate flows if the configuration changed. */
2069 if (!s != !ofproto->stp) {
2070 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2074 if (!ofproto->stp) {
2075 ofproto->stp = stp_create(ofproto_->name, s->system_id,
2076 send_bpdu_cb, ofproto);
2077 ofproto->stp_last_tick = time_msec();
2080 stp_set_bridge_id(ofproto->stp, s->system_id);
2081 stp_set_bridge_priority(ofproto->stp, s->priority);
2082 stp_set_hello_time(ofproto->stp, s->hello_time);
2083 stp_set_max_age(ofproto->stp, s->max_age);
2084 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
2086 struct ofport *ofport;
2088 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
2089 set_stp_port(ofport, NULL);
2092 stp_destroy(ofproto->stp);
2093 ofproto->stp = NULL;
2100 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
2102 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2106 s->bridge_id = stp_get_bridge_id(ofproto->stp);
2107 s->designated_root = stp_get_designated_root(ofproto->stp);
2108 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
2117 update_stp_port_state(struct ofport_dpif *ofport)
2119 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2120 enum stp_state state;
2122 /* Figure out new state. */
2123 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
2127 if (ofport->stp_state != state) {
2128 enum ofputil_port_state of_state;
2131 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
2132 netdev_get_name(ofport->up.netdev),
2133 stp_state_name(ofport->stp_state),
2134 stp_state_name(state));
2135 if (stp_learn_in_state(ofport->stp_state)
2136 != stp_learn_in_state(state)) {
2137 /* xxx Learning action flows should also be flushed. */
2138 mac_learning_flush(ofproto->ml,
2139 &ofproto->backer->revalidate_set);
2141 fwd_change = stp_forward_in_state(ofport->stp_state)
2142 != stp_forward_in_state(state);
2144 ofproto->backer->need_revalidate = REV_STP;
2145 ofport->stp_state = state;
2146 ofport->stp_state_entered = time_msec();
2148 if (fwd_change && ofport->bundle) {
2149 bundle_update(ofport->bundle);
2152 /* Update the STP state bits in the OpenFlow port description. */
2153 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
2154 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
2155 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
2156 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
2157 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
2159 ofproto_port_set_state(&ofport->up, of_state);
2163 /* Configures STP on 'ofport_' using the settings defined in 's'. The
2164 * caller is responsible for assigning STP port numbers and ensuring
2165 * there are no duplicates. */
2167 set_stp_port(struct ofport *ofport_,
2168 const struct ofproto_port_stp_settings *s)
2170 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2171 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2172 struct stp_port *sp = ofport->stp_port;
2174 if (!s || !s->enable) {
2176 ofport->stp_port = NULL;
2177 stp_port_disable(sp);
2178 update_stp_port_state(ofport);
2181 } else if (sp && stp_port_no(sp) != s->port_num
2182 && ofport == stp_port_get_aux(sp)) {
2183 /* The port-id changed, so disable the old one if it's not
2184 * already in use by another port. */
2185 stp_port_disable(sp);
2188 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
2189 stp_port_enable(sp);
2191 stp_port_set_aux(sp, ofport);
2192 stp_port_set_priority(sp, s->priority);
2193 stp_port_set_path_cost(sp, s->path_cost);
2195 update_stp_port_state(ofport);
2201 get_stp_port_status(struct ofport *ofport_,
2202 struct ofproto_port_stp_status *s)
2204 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2205 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2206 struct stp_port *sp = ofport->stp_port;
2208 if (!ofproto->stp || !sp) {
2214 s->port_id = stp_port_get_id(sp);
2215 s->state = stp_port_get_state(sp);
2216 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
2217 s->role = stp_port_get_role(sp);
2218 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
2224 stp_run(struct ofproto_dpif *ofproto)
2227 long long int now = time_msec();
2228 long long int elapsed = now - ofproto->stp_last_tick;
2229 struct stp_port *sp;
2232 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
2233 ofproto->stp_last_tick = now;
2235 while (stp_get_changed_port(ofproto->stp, &sp)) {
2236 struct ofport_dpif *ofport = stp_port_get_aux(sp);
2239 update_stp_port_state(ofport);
2243 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
2244 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
2250 stp_wait(struct ofproto_dpif *ofproto)
2253 poll_timer_wait(1000);
2257 /* Returns true if STP should process 'flow'. */
2259 stp_should_process_flow(const struct flow *flow)
2261 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
2265 stp_process_packet(const struct ofport_dpif *ofport,
2266 const struct ofpbuf *packet)
2268 struct ofpbuf payload = *packet;
2269 struct eth_header *eth = payload.data;
2270 struct stp_port *sp = ofport->stp_port;
2272 /* Sink packets on ports that have STP disabled when the bridge has
2274 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
2278 /* Trim off padding on payload. */
2279 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
2280 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
2283 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
2284 stp_received_bpdu(sp, payload.data, payload.size);
2288 static struct priority_to_dscp *
2289 get_priority(const struct ofport_dpif *ofport, uint32_t priority)
2291 struct priority_to_dscp *pdscp;
2294 hash = hash_int(priority, 0);
2295 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) {
2296 if (pdscp->priority == priority) {
2304 ofport_clear_priorities(struct ofport_dpif *ofport)
2306 struct priority_to_dscp *pdscp, *next;
2308 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) {
2309 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
2315 set_queues(struct ofport *ofport_,
2316 const struct ofproto_port_queue *qdscp_list,
2319 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2320 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2321 struct hmap new = HMAP_INITIALIZER(&new);
2324 for (i = 0; i < n_qdscp; i++) {
2325 struct priority_to_dscp *pdscp;
2329 dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
2330 if (dpif_queue_to_priority(ofproto->backer->dpif, qdscp_list[i].queue,
2335 pdscp = get_priority(ofport, priority);
2337 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
2339 pdscp = xmalloc(sizeof *pdscp);
2340 pdscp->priority = priority;
2342 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2345 if (pdscp->dscp != dscp) {
2347 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2350 hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0));
2353 if (!hmap_is_empty(&ofport->priorities)) {
2354 ofport_clear_priorities(ofport);
2355 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2358 hmap_swap(&new, &ofport->priorities);
2366 /* Expires all MAC learning entries associated with 'bundle' and forces its
2367 * ofproto to revalidate every flow.
2369 * Normally MAC learning entries are removed only from the ofproto associated
2370 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2371 * are removed from every ofproto. When patch ports and SLB bonds are in use
2372 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2373 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2374 * with the host from which it migrated. */
2376 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2378 struct ofproto_dpif *ofproto = bundle->ofproto;
2379 struct mac_learning *ml = ofproto->ml;
2380 struct mac_entry *mac, *next_mac;
2382 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2383 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2384 if (mac->port.p == bundle) {
2386 struct ofproto_dpif *o;
2388 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2390 struct mac_entry *e;
2392 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan,
2395 mac_learning_expire(o->ml, e);
2401 mac_learning_expire(ml, mac);
2406 static struct ofbundle *
2407 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2409 struct ofbundle *bundle;
2411 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2412 &ofproto->bundles) {
2413 if (bundle->aux == aux) {
2420 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
2421 * ones that are found to 'bundles'. */
2423 bundle_lookup_multiple(struct ofproto_dpif *ofproto,
2424 void **auxes, size_t n_auxes,
2425 struct hmapx *bundles)
2429 hmapx_init(bundles);
2430 for (i = 0; i < n_auxes; i++) {
2431 struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
2433 hmapx_add(bundles, bundle);
2439 bundle_update(struct ofbundle *bundle)
2441 struct ofport_dpif *port;
2443 bundle->floodable = true;
2444 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2445 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2446 || !stp_forward_in_state(port->stp_state)) {
2447 bundle->floodable = false;
2454 bundle_del_port(struct ofport_dpif *port)
2456 struct ofbundle *bundle = port->bundle;
2458 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2460 list_remove(&port->bundle_node);
2461 port->bundle = NULL;
2464 lacp_slave_unregister(bundle->lacp, port);
2467 bond_slave_unregister(bundle->bond, port);
2470 bundle_update(bundle);
2474 bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port,
2475 struct lacp_slave_settings *lacp)
2477 struct ofport_dpif *port;
2479 port = get_ofp_port(bundle->ofproto, ofp_port);
2484 if (port->bundle != bundle) {
2485 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2487 bundle_del_port(port);
2490 port->bundle = bundle;
2491 list_push_back(&bundle->ports, &port->bundle_node);
2492 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2493 || !stp_forward_in_state(port->stp_state)) {
2494 bundle->floodable = false;
2498 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2499 lacp_slave_register(bundle->lacp, port, lacp);
2506 bundle_destroy(struct ofbundle *bundle)
2508 struct ofproto_dpif *ofproto;
2509 struct ofport_dpif *port, *next_port;
2516 ofproto = bundle->ofproto;
2517 for (i = 0; i < MAX_MIRRORS; i++) {
2518 struct ofmirror *m = ofproto->mirrors[i];
2520 if (m->out == bundle) {
2522 } else if (hmapx_find_and_delete(&m->srcs, bundle)
2523 || hmapx_find_and_delete(&m->dsts, bundle)) {
2524 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2529 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2530 bundle_del_port(port);
2533 bundle_flush_macs(bundle, true);
2534 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2536 free(bundle->trunks);
2537 lacp_destroy(bundle->lacp);
2538 bond_destroy(bundle->bond);
2543 bundle_set(struct ofproto *ofproto_, void *aux,
2544 const struct ofproto_bundle_settings *s)
2546 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2547 bool need_flush = false;
2548 struct ofport_dpif *port;
2549 struct ofbundle *bundle;
2550 unsigned long *trunks;
2556 bundle_destroy(bundle_lookup(ofproto, aux));
2560 ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2561 ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2563 bundle = bundle_lookup(ofproto, aux);
2565 bundle = xmalloc(sizeof *bundle);
2567 bundle->ofproto = ofproto;
2568 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2569 hash_pointer(aux, 0));
2571 bundle->name = NULL;
2573 list_init(&bundle->ports);
2574 bundle->vlan_mode = PORT_VLAN_TRUNK;
2576 bundle->trunks = NULL;
2577 bundle->use_priority_tags = s->use_priority_tags;
2578 bundle->lacp = NULL;
2579 bundle->bond = NULL;
2581 bundle->floodable = true;
2583 bundle->src_mirrors = 0;
2584 bundle->dst_mirrors = 0;
2585 bundle->mirror_out = 0;
2588 if (!bundle->name || strcmp(s->name, bundle->name)) {
2590 bundle->name = xstrdup(s->name);
2595 if (!bundle->lacp) {
2596 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2597 bundle->lacp = lacp_create();
2599 lacp_configure(bundle->lacp, s->lacp);
2601 lacp_destroy(bundle->lacp);
2602 bundle->lacp = NULL;
2605 /* Update set of ports. */
2607 for (i = 0; i < s->n_slaves; i++) {
2608 if (!bundle_add_port(bundle, s->slaves[i],
2609 s->lacp ? &s->lacp_slaves[i] : NULL)) {
2613 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2614 struct ofport_dpif *next_port;
2616 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2617 for (i = 0; i < s->n_slaves; i++) {
2618 if (s->slaves[i] == port->up.ofp_port) {
2623 bundle_del_port(port);
2627 ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2629 if (list_is_empty(&bundle->ports)) {
2630 bundle_destroy(bundle);
2634 /* Set VLAN tagging mode */
2635 if (s->vlan_mode != bundle->vlan_mode
2636 || s->use_priority_tags != bundle->use_priority_tags) {
2637 bundle->vlan_mode = s->vlan_mode;
2638 bundle->use_priority_tags = s->use_priority_tags;
2643 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2644 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2646 if (vlan != bundle->vlan) {
2647 bundle->vlan = vlan;
2651 /* Get trunked VLANs. */
2652 switch (s->vlan_mode) {
2653 case PORT_VLAN_ACCESS:
2657 case PORT_VLAN_TRUNK:
2658 trunks = CONST_CAST(unsigned long *, s->trunks);
2661 case PORT_VLAN_NATIVE_UNTAGGED:
2662 case PORT_VLAN_NATIVE_TAGGED:
2663 if (vlan != 0 && (!s->trunks
2664 || !bitmap_is_set(s->trunks, vlan)
2665 || bitmap_is_set(s->trunks, 0))) {
2666 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2668 trunks = bitmap_clone(s->trunks, 4096);
2670 trunks = bitmap_allocate1(4096);
2672 bitmap_set1(trunks, vlan);
2673 bitmap_set0(trunks, 0);
2675 trunks = CONST_CAST(unsigned long *, s->trunks);
2682 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2683 free(bundle->trunks);
2684 if (trunks == s->trunks) {
2685 bundle->trunks = vlan_bitmap_clone(trunks);
2687 bundle->trunks = trunks;
2692 if (trunks != s->trunks) {
2697 if (!list_is_short(&bundle->ports)) {
2698 bundle->ofproto->has_bonded_bundles = true;
2700 if (bond_reconfigure(bundle->bond, s->bond)) {
2701 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2704 bundle->bond = bond_create(s->bond);
2705 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2708 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2709 bond_slave_register(bundle->bond, port, port->up.netdev);
2712 bond_destroy(bundle->bond);
2713 bundle->bond = NULL;
2716 /* If we changed something that would affect MAC learning, un-learn
2717 * everything on this port and force flow revalidation. */
2719 bundle_flush_macs(bundle, false);
2726 bundle_remove(struct ofport *port_)
2728 struct ofport_dpif *port = ofport_dpif_cast(port_);
2729 struct ofbundle *bundle = port->bundle;
2732 bundle_del_port(port);
2733 if (list_is_empty(&bundle->ports)) {
2734 bundle_destroy(bundle);
2735 } else if (list_is_short(&bundle->ports)) {
2736 bond_destroy(bundle->bond);
2737 bundle->bond = NULL;
2743 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2745 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2746 struct ofport_dpif *port = port_;
2747 uint8_t ea[ETH_ADDR_LEN];
2750 error = netdev_get_etheraddr(port->up.netdev, ea);
2752 struct ofpbuf packet;
2755 ofpbuf_init(&packet, 0);
2756 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2758 memcpy(packet_pdu, pdu, pdu_size);
2760 send_packet(port, &packet);
2761 ofpbuf_uninit(&packet);
2763 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2764 "%s (%s)", port->bundle->name,
2765 netdev_get_name(port->up.netdev), strerror(error));
2770 bundle_send_learning_packets(struct ofbundle *bundle)
2772 struct ofproto_dpif *ofproto = bundle->ofproto;
2773 int error, n_packets, n_errors;
2774 struct mac_entry *e;
2776 error = n_packets = n_errors = 0;
2777 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2778 if (e->port.p != bundle) {
2779 struct ofpbuf *learning_packet;
2780 struct ofport_dpif *port;
2784 /* The assignment to "port" is unnecessary but makes "grep"ing for
2785 * struct ofport_dpif more effective. */
2786 learning_packet = bond_compose_learning_packet(bundle->bond,
2790 ret = send_packet(port, learning_packet);
2791 ofpbuf_delete(learning_packet);
2801 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2802 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2803 "packets, last error was: %s",
2804 bundle->name, n_errors, n_packets, strerror(error));
2806 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2807 bundle->name, n_packets);
2812 bundle_run(struct ofbundle *bundle)
2815 lacp_run(bundle->lacp, send_pdu_cb);
2818 struct ofport_dpif *port;
2820 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2821 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2824 bond_run(bundle->bond, &bundle->ofproto->backer->revalidate_set,
2825 lacp_status(bundle->lacp));
2826 if (bond_should_send_learning_packets(bundle->bond)) {
2827 bundle_send_learning_packets(bundle);
2833 bundle_wait(struct ofbundle *bundle)
2836 lacp_wait(bundle->lacp);
2839 bond_wait(bundle->bond);
2846 mirror_scan(struct ofproto_dpif *ofproto)
2850 for (idx = 0; idx < MAX_MIRRORS; idx++) {
2851 if (!ofproto->mirrors[idx]) {
2858 static struct ofmirror *
2859 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
2863 for (i = 0; i < MAX_MIRRORS; i++) {
2864 struct ofmirror *mirror = ofproto->mirrors[i];
2865 if (mirror && mirror->aux == aux) {
2873 /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */
2875 mirror_update_dups(struct ofproto_dpif *ofproto)
2879 for (i = 0; i < MAX_MIRRORS; i++) {
2880 struct ofmirror *m = ofproto->mirrors[i];
2883 m->dup_mirrors = MIRROR_MASK_C(1) << i;
2887 for (i = 0; i < MAX_MIRRORS; i++) {
2888 struct ofmirror *m1 = ofproto->mirrors[i];
2895 for (j = i + 1; j < MAX_MIRRORS; j++) {
2896 struct ofmirror *m2 = ofproto->mirrors[j];
2898 if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) {
2899 m1->dup_mirrors |= MIRROR_MASK_C(1) << j;
2900 m2->dup_mirrors |= m1->dup_mirrors;
2907 mirror_set(struct ofproto *ofproto_, void *aux,
2908 const struct ofproto_mirror_settings *s)
2910 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2911 mirror_mask_t mirror_bit;
2912 struct ofbundle *bundle;
2913 struct ofmirror *mirror;
2914 struct ofbundle *out;
2915 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
2916 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
2919 mirror = mirror_lookup(ofproto, aux);
2921 mirror_destroy(mirror);
2927 idx = mirror_scan(ofproto);
2929 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
2931 ofproto->up.name, MAX_MIRRORS, s->name);
2935 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
2936 mirror->ofproto = ofproto;
2939 mirror->out_vlan = -1;
2940 mirror->name = NULL;
2943 if (!mirror->name || strcmp(s->name, mirror->name)) {
2945 mirror->name = xstrdup(s->name);
2948 /* Get the new configuration. */
2949 if (s->out_bundle) {
2950 out = bundle_lookup(ofproto, s->out_bundle);
2952 mirror_destroy(mirror);
2958 out_vlan = s->out_vlan;
2960 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
2961 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
2963 /* If the configuration has not changed, do nothing. */
2964 if (hmapx_equals(&srcs, &mirror->srcs)
2965 && hmapx_equals(&dsts, &mirror->dsts)
2966 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
2967 && mirror->out == out
2968 && mirror->out_vlan == out_vlan)
2970 hmapx_destroy(&srcs);
2971 hmapx_destroy(&dsts);
2975 hmapx_swap(&srcs, &mirror->srcs);
2976 hmapx_destroy(&srcs);
2978 hmapx_swap(&dsts, &mirror->dsts);
2979 hmapx_destroy(&dsts);
2981 free(mirror->vlans);
2982 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
2985 mirror->out_vlan = out_vlan;
2987 /* Update bundles. */
2988 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2989 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
2990 if (hmapx_contains(&mirror->srcs, bundle)) {
2991 bundle->src_mirrors |= mirror_bit;
2993 bundle->src_mirrors &= ~mirror_bit;
2996 if (hmapx_contains(&mirror->dsts, bundle)) {
2997 bundle->dst_mirrors |= mirror_bit;
2999 bundle->dst_mirrors &= ~mirror_bit;
3002 if (mirror->out == bundle) {
3003 bundle->mirror_out |= mirror_bit;
3005 bundle->mirror_out &= ~mirror_bit;
3009 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3010 ofproto->has_mirrors = true;
3011 mac_learning_flush(ofproto->ml,
3012 &ofproto->backer->revalidate_set);
3013 mirror_update_dups(ofproto);
3019 mirror_destroy(struct ofmirror *mirror)
3021 struct ofproto_dpif *ofproto;
3022 mirror_mask_t mirror_bit;
3023 struct ofbundle *bundle;
3030 ofproto = mirror->ofproto;
3031 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3032 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
3034 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
3035 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
3036 bundle->src_mirrors &= ~mirror_bit;
3037 bundle->dst_mirrors &= ~mirror_bit;
3038 bundle->mirror_out &= ~mirror_bit;
3041 hmapx_destroy(&mirror->srcs);
3042 hmapx_destroy(&mirror->dsts);
3043 free(mirror->vlans);
3045 ofproto->mirrors[mirror->idx] = NULL;
3049 mirror_update_dups(ofproto);
3051 ofproto->has_mirrors = false;
3052 for (i = 0; i < MAX_MIRRORS; i++) {
3053 if (ofproto->mirrors[i]) {
3054 ofproto->has_mirrors = true;
3061 mirror_get_stats(struct ofproto *ofproto_, void *aux,
3062 uint64_t *packets, uint64_t *bytes)
3064 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3065 struct ofmirror *mirror = mirror_lookup(ofproto, aux);
3068 *packets = *bytes = UINT64_MAX;
3074 *packets = mirror->packet_count;
3075 *bytes = mirror->byte_count;
3081 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
3083 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3084 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
3085 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
3091 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
3093 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3094 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
3095 return bundle && bundle->mirror_out != 0;
3099 forward_bpdu_changed(struct ofproto *ofproto_)
3101 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3102 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3106 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
3109 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3110 mac_learning_set_idle_time(ofproto->ml, idle_time);
3111 mac_learning_set_max_entries(ofproto->ml, max_entries);
3116 static struct ofport_dpif *
3117 get_ofp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
3119 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
3120 return ofport ? ofport_dpif_cast(ofport) : NULL;
3123 static struct ofport_dpif *
3124 get_odp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
3126 struct ofport_dpif *port = odp_port_to_ofport(ofproto->backer, odp_port);
3127 return port && &ofproto->up == port->up.ofproto ? port : NULL;
3131 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
3132 struct ofproto_port *ofproto_port,
3133 struct dpif_port *dpif_port)
3135 ofproto_port->name = dpif_port->name;
3136 ofproto_port->type = dpif_port->type;
3137 ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
3140 static struct ofport_dpif *
3141 ofport_get_peer(const struct ofport_dpif *ofport_dpif)
3143 const struct ofproto_dpif *ofproto;
3146 peer = netdev_vport_patch_peer(ofport_dpif->up.netdev);
3151 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
3152 struct ofport *ofport;
3154 ofport = shash_find_data(&ofproto->up.port_by_name, peer);
3155 if (ofport && ofport->ofproto->ofproto_class == &ofproto_dpif_class) {
3156 return ofport_dpif_cast(ofport);
3163 port_run_fast(struct ofport_dpif *ofport)
3165 if (ofport->cfm && cfm_should_send_ccm(ofport->cfm)) {
3166 struct ofpbuf packet;
3168 ofpbuf_init(&packet, 0);
3169 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
3170 send_packet(ofport, &packet);
3171 ofpbuf_uninit(&packet);
3176 port_run(struct ofport_dpif *ofport)
3178 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
3179 bool carrier_changed = carrier_seq != ofport->carrier_seq;
3180 bool enable = netdev_get_carrier(ofport->up.netdev);
3182 ofport->carrier_seq = carrier_seq;
3184 port_run_fast(ofport);
3186 if (ofport->tnl_port
3187 && tnl_port_reconfigure(&ofport->up, ofport->odp_port,
3188 &ofport->tnl_port)) {
3189 ofproto_dpif_cast(ofport->up.ofproto)->backer->need_revalidate = true;
3193 int cfm_opup = cfm_get_opup(ofport->cfm);
3195 cfm_run(ofport->cfm);
3196 enable = enable && !cfm_get_fault(ofport->cfm);
3198 if (cfm_opup >= 0) {
3199 enable = enable && cfm_opup;
3203 if (ofport->bundle) {
3204 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
3205 if (carrier_changed) {
3206 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
3210 if (ofport->may_enable != enable) {
3211 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3213 if (ofproto->has_bundle_action) {
3214 ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
3218 ofport->may_enable = enable;
3222 port_wait(struct ofport_dpif *ofport)
3225 cfm_wait(ofport->cfm);
3230 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
3231 struct ofproto_port *ofproto_port)
3233 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3234 struct dpif_port dpif_port;
3237 if (sset_contains(&ofproto->ghost_ports, devname)) {
3238 const char *type = netdev_get_type_from_name(devname);
3240 /* We may be called before ofproto->up.port_by_name is populated with
3241 * the appropriate ofport. For this reason, we must get the name and
3242 * type from the netdev layer directly. */
3244 const struct ofport *ofport;
3246 ofport = shash_find_data(&ofproto->up.port_by_name, devname);
3247 ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
3248 ofproto_port->name = xstrdup(devname);
3249 ofproto_port->type = xstrdup(type);
3255 if (!sset_contains(&ofproto->ports, devname)) {
3258 error = dpif_port_query_by_name(ofproto->backer->dpif,
3259 devname, &dpif_port);
3261 ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
3267 port_add(struct ofproto *ofproto_, struct netdev *netdev)
3269 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3270 const char *dp_port_name = netdev_vport_get_dpif_port(netdev);
3271 const char *devname = netdev_get_name(netdev);
3273 if (netdev_vport_is_patch(netdev)) {
3274 sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
3278 if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
3279 uint32_t port_no = UINT32_MAX;
3282 error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
3286 if (netdev_get_tunnel_config(netdev)) {
3287 simap_put(&ofproto->backer->tnl_backers, dp_port_name, port_no);
3291 if (netdev_get_tunnel_config(netdev)) {
3292 sset_add(&ofproto->ghost_ports, devname);
3294 sset_add(&ofproto->ports, devname);
3300 port_del(struct ofproto *ofproto_, uint16_t ofp_port)
3302 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3303 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
3310 sset_find_and_delete(&ofproto->ghost_ports,
3311 netdev_get_name(ofport->up.netdev));
3312 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3313 if (!ofport->tnl_port) {
3314 error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
3316 /* The caller is going to close ofport->up.netdev. If this is a
3317 * bonded port, then the bond is using that netdev, so remove it
3318 * from the bond. The client will need to reconfigure everything
3319 * after deleting ports, so then the slave will get re-added. */
3320 bundle_remove(&ofport->up);
3327 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
3329 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3334 error = netdev_get_stats(ofport->up.netdev, stats);
3336 if (!error && ofport_->ofp_port == OFPP_LOCAL) {
3337 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3339 /* ofproto->stats.tx_packets represents packets that we created
3340 * internally and sent to some port (e.g. packets sent with
3341 * send_packet()). Account for them as if they had come from
3342 * OFPP_LOCAL and got forwarded. */
3344 if (stats->rx_packets != UINT64_MAX) {
3345 stats->rx_packets += ofproto->stats.tx_packets;
3348 if (stats->rx_bytes != UINT64_MAX) {
3349 stats->rx_bytes += ofproto->stats.tx_bytes;
3352 /* ofproto->stats.rx_packets represents packets that were received on
3353 * some port and we processed internally and dropped (e.g. STP).
3354 * Account for them as if they had been forwarded to OFPP_LOCAL. */
3356 if (stats->tx_packets != UINT64_MAX) {
3357 stats->tx_packets += ofproto->stats.rx_packets;
3360 if (stats->tx_bytes != UINT64_MAX) {
3361 stats->tx_bytes += ofproto->stats.rx_bytes;
3368 /* Account packets for LOCAL port. */
3370 ofproto_update_local_port_stats(const struct ofproto *ofproto_,
3371 size_t tx_size, size_t rx_size)
3373 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3376 ofproto->stats.rx_packets++;
3377 ofproto->stats.rx_bytes += rx_size;
3380 ofproto->stats.tx_packets++;
3381 ofproto->stats.tx_bytes += tx_size;
3385 struct port_dump_state {
3390 struct ofproto_port port;
3395 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
3397 *statep = xzalloc(sizeof(struct port_dump_state));
3402 port_dump_next(const struct ofproto *ofproto_, void *state_,
3403 struct ofproto_port *port)
3405 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3406 struct port_dump_state *state = state_;
3407 const struct sset *sset;
3408 struct sset_node *node;
3410 if (state->has_port) {
3411 ofproto_port_destroy(&state->port);
3412 state->has_port = false;
3414 sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
3415 while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
3418 error = port_query_by_name(ofproto_, node->name, &state->port);
3420 *port = state->port;
3421 state->has_port = true;
3423 } else if (error != ENODEV) {
3428 if (!state->ghost) {
3429 state->ghost = true;
3432 return port_dump_next(ofproto_, state_, port);
3439 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3441 struct port_dump_state *state = state_;
3443 if (state->has_port) {
3444 ofproto_port_destroy(&state->port);
3451 port_poll(const struct ofproto *ofproto_, char **devnamep)
3453 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3455 if (ofproto->port_poll_errno) {
3456 int error = ofproto->port_poll_errno;
3457 ofproto->port_poll_errno = 0;
3461 if (sset_is_empty(&ofproto->port_poll_set)) {
3465 *devnamep = sset_pop(&ofproto->port_poll_set);
3470 port_poll_wait(const struct ofproto *ofproto_)
3472 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3473 dpif_port_poll_wait(ofproto->backer->dpif);
3477 port_is_lacp_current(const struct ofport *ofport_)
3479 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3480 return (ofport->bundle && ofport->bundle->lacp
3481 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3485 /* Upcall handling. */
3487 /* Flow miss batching.
3489 * Some dpifs implement operations faster when you hand them off in a batch.
3490 * To allow batching, "struct flow_miss" queues the dpif-related work needed
3491 * for a given flow. Each "struct flow_miss" corresponds to sending one or
3492 * more packets, plus possibly installing the flow in the dpif.
3494 * So far we only batch the operations that affect flow setup time the most.
3495 * It's possible to batch more than that, but the benefit might be minimal. */
3497 struct hmap_node hmap_node;
3498 struct ofproto_dpif *ofproto;
3500 enum odp_key_fitness key_fitness;
3501 const struct nlattr *key;
3503 struct initial_vals initial_vals;
3504 struct list packets;
3505 enum dpif_upcall_type upcall_type;
3506 uint32_t odp_in_port;
3509 struct flow_miss_op {
3510 struct dpif_op dpif_op;
3512 uint64_t slow_stub[128 / 8]; /* Buffer for compose_slow_path() */
3513 struct xlate_out xout;
3514 bool xout_garbage; /* 'xout' needs to be uninitialized? */
3517 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
3518 * OpenFlow controller as necessary according to their individual
3519 * configurations. */
3521 send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet,
3522 const struct flow *flow)
3524 struct ofputil_packet_in pin;
3526 pin.packet = packet->data;
3527 pin.packet_len = packet->size;
3528 pin.reason = OFPR_NO_MATCH;
3529 pin.controller_id = 0;
3534 pin.send_len = 0; /* not used for flow table misses */
3536 flow_get_metadata(flow, &pin.fmd);
3538 connmgr_send_packet_in(ofproto->up.connmgr, &pin);
3541 static enum slow_path_reason
3542 process_special(struct ofproto_dpif *ofproto, const struct flow *flow,
3543 const struct ofport_dpif *ofport, const struct ofpbuf *packet)
3547 } else if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) {
3549 cfm_process_heartbeat(ofport->cfm, packet);
3552 } else if (ofport->bundle && ofport->bundle->lacp
3553 && flow->dl_type == htons(ETH_TYPE_LACP)) {
3555 lacp_process_packet(ofport->bundle->lacp, ofport, packet);
3558 } else if (ofproto->stp && stp_should_process_flow(flow)) {
3560 stp_process_packet(ofport, packet);
3568 static struct flow_miss *
3569 flow_miss_find(struct hmap *todo, const struct ofproto_dpif *ofproto,
3570 const struct flow *flow, uint32_t hash)
3572 struct flow_miss *miss;
3574 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
3575 if (miss->ofproto == ofproto && flow_equal(&miss->flow, flow)) {
3583 /* Partially Initializes 'op' as an "execute" operation for 'miss' and
3584 * 'packet'. The caller must initialize op->actions and op->actions_len. If
3585 * 'miss' is associated with a subfacet the caller must also initialize the
3586 * returned op->subfacet, and if anything needs to be freed after processing
3587 * the op, the caller must initialize op->garbage also. */
3589 init_flow_miss_execute_op(struct flow_miss *miss, struct ofpbuf *packet,
3590 struct flow_miss_op *op)
3592 if (miss->flow.vlan_tci != miss->initial_vals.vlan_tci) {
3593 /* This packet was received on a VLAN splinter port. We
3594 * added a VLAN to the packet to make the packet resemble
3595 * the flow, but the actions were composed assuming that
3596 * the packet contained no VLAN. So, we must remove the
3597 * VLAN header from the packet before trying to execute the
3599 eth_pop_vlan(packet);
3602 op->xout_garbage = false;
3603 op->dpif_op.type = DPIF_OP_EXECUTE;
3604 op->dpif_op.u.execute.key = miss->key;
3605 op->dpif_op.u.execute.key_len = miss->key_len;
3606 op->dpif_op.u.execute.packet = packet;
3609 /* Helper for handle_flow_miss_without_facet() and
3610 * handle_flow_miss_with_facet(). */
3612 handle_flow_miss_common(struct rule_dpif *rule,
3613 struct ofpbuf *packet, const struct flow *flow)
3615 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3617 if (rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
3619 * Extra-special case for fail-open mode.
3621 * We are in fail-open mode and the packet matched the fail-open
3622 * rule, but we are connected to a controller too. We should send
3623 * the packet up to the controller in the hope that it will try to
3624 * set up a flow and thereby allow us to exit fail-open.
3626 * See the top-level comment in fail-open.c for more information.
3628 send_packet_in_miss(ofproto, packet, flow);
3632 /* Figures out whether a flow that missed in 'ofproto', whose details are in
3633 * 'miss', is likely to be worth tracking in detail in userspace and (usually)
3634 * installing a datapath flow. The answer is usually "yes" (a return value of
3635 * true). However, for short flows the cost of bookkeeping is much higher than
3636 * the benefits, so when the datapath holds a large number of flows we impose
3637 * some heuristics to decide which flows are likely to be worth tracking. */
3639 flow_miss_should_make_facet(struct ofproto_dpif *ofproto,
3640 struct flow_miss *miss, uint32_t hash)
3642 if (!ofproto->governor) {
3645 n_subfacets = hmap_count(&ofproto->subfacets);
3646 if (n_subfacets * 2 <= ofproto->up.flow_eviction_threshold) {
3650 ofproto->governor = governor_create(ofproto->up.name);
3653 return governor_should_install_flow(ofproto->governor, hash,
3654 list_size(&miss->packets));
3657 /* Handles 'miss', which matches 'rule', without creating a facet or subfacet
3658 * or creating any datapath flow. May add an "execute" operation to 'ops' and
3659 * increment '*n_ops'. */
3661 handle_flow_miss_without_facet(struct flow_miss *miss,
3662 struct flow_miss_op *ops, size_t *n_ops)
3664 struct rule_dpif *rule = rule_dpif_lookup(miss->ofproto, &miss->flow);
3665 long long int now = time_msec();
3666 struct ofpbuf *packet;
3667 struct xlate_in xin;
3669 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3670 struct flow_miss_op *op = &ops[*n_ops];
3671 struct dpif_flow_stats stats;
3673 COVERAGE_INC(facet_suppress);
3675 handle_flow_miss_common(rule, packet, &miss->flow);
3677 dpif_flow_stats_extract(&miss->flow, packet, now, &stats);
3678 rule_credit_stats(rule, &stats);
3680 xlate_in_init(&xin, miss->ofproto, &miss->flow, &miss->initial_vals,
3681 rule, stats.tcp_flags, packet);
3682 xin.resubmit_stats = &stats;
3683 xlate_actions(&xin, &op->xout);
3685 if (op->xout.odp_actions.size) {
3686 struct dpif_execute *execute = &op->dpif_op.u.execute;
3688 init_flow_miss_execute_op(miss, packet, op);
3689 execute->actions = op->xout.odp_actions.data;
3690 execute->actions_len = op->xout.odp_actions.size;
3691 op->xout_garbage = true;
3695 xlate_out_uninit(&op->xout);
3700 /* Handles 'miss', which matches 'facet'. May add any required datapath
3701 * operations to 'ops', incrementing '*n_ops' for each new op.
3703 * All of the packets in 'miss' are considered to have arrived at time 'now'.
3704 * This is really important only for new facets: if we just called time_msec()
3705 * here, then the new subfacet or its packets could look (occasionally) as
3706 * though it was used some time after the facet was used. That can make a
3707 * one-packet flow look like it has a nonzero duration, which looks odd in
3708 * e.g. NetFlow statistics. */
3710 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
3712 struct flow_miss_op *ops, size_t *n_ops)
3714 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3715 enum subfacet_path want_path;
3716 struct subfacet *subfacet;
3717 struct ofpbuf *packet;
3719 subfacet = subfacet_create(facet, miss, now);
3720 want_path = subfacet->facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
3722 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3723 struct flow_miss_op *op = &ops[*n_ops];
3724 struct dpif_flow_stats stats;
3726 handle_flow_miss_common(facet->rule, packet, &miss->flow);
3728 if (want_path != SF_FAST_PATH) {
3729 struct xlate_in xin;
3731 xlate_in_init(&xin, ofproto, &facet->flow, &facet->initial_vals,
3732 facet->rule, 0, packet);
3733 xlate_actions_for_side_effects(&xin);
3736 dpif_flow_stats_extract(&facet->flow, packet, now, &stats);
3737 subfacet_update_stats(subfacet, &stats);
3739 if (facet->xout.odp_actions.size) {
3740 struct dpif_execute *execute = &op->dpif_op.u.execute;
3742 init_flow_miss_execute_op(miss, packet, op);
3743 execute->actions = facet->xout.odp_actions.data,
3744 execute->actions_len = facet->xout.odp_actions.size;
3749 if (miss->upcall_type == DPIF_UC_MISS || subfacet->path != want_path) {
3750 struct flow_miss_op *op = &ops[(*n_ops)++];
3751 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
3753 subfacet->path = want_path;
3755 op->xout_garbage = false;
3756 op->dpif_op.type = DPIF_OP_FLOW_PUT;
3757 put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
3758 put->key = miss->key;
3759 put->key_len = miss->key_len;
3760 if (want_path == SF_FAST_PATH) {
3761 put->actions = facet->xout.odp_actions.data;
3762 put->actions_len = facet->xout.odp_actions.size;
3764 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
3765 op->slow_stub, sizeof op->slow_stub,
3766 &put->actions, &put->actions_len);
3772 /* Handles flow miss 'miss'. May add any required datapath operations
3773 * to 'ops', incrementing '*n_ops' for each new op. */
3775 handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops,
3778 struct ofproto_dpif *ofproto = miss->ofproto;
3779 struct facet *facet;
3783 /* The caller must ensure that miss->hmap_node.hash contains
3784 * flow_hash(miss->flow, 0). */
3785 hash = miss->hmap_node.hash;
3787 facet = facet_lookup_valid(ofproto, &miss->flow, hash);
3789 /* There does not exist a bijection between 'struct flow' and datapath
3790 * flow keys with fitness ODP_FIT_TO_LITTLE. This breaks a fundamental
3791 * assumption used throughout the facet and subfacet handling code.
3792 * Since we have to handle these misses in userspace anyway, we simply
3793 * skip facet creation, avoiding the problem alltogether. */
3794 if (miss->key_fitness == ODP_FIT_TOO_LITTLE
3795 || !flow_miss_should_make_facet(ofproto, miss, hash)) {
3796 handle_flow_miss_without_facet(miss, ops, n_ops);
3800 facet = facet_create(miss, hash);
3805 handle_flow_miss_with_facet(miss, facet, now, ops, n_ops);
3808 static struct drop_key *
3809 drop_key_lookup(const struct dpif_backer *backer, const struct nlattr *key,
3812 struct drop_key *drop_key;
3814 HMAP_FOR_EACH_WITH_HASH (drop_key, hmap_node, hash_bytes(key, key_len, 0),
3815 &backer->drop_keys) {
3816 if (drop_key->key_len == key_len
3817 && !memcmp(drop_key->key, key, key_len)) {
3825 drop_key_clear(struct dpif_backer *backer)
3827 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3828 struct drop_key *drop_key, *next;
3830 HMAP_FOR_EACH_SAFE (drop_key, next, hmap_node, &backer->drop_keys) {
3833 error = dpif_flow_del(backer->dpif, drop_key->key, drop_key->key_len,
3835 if (error && !VLOG_DROP_WARN(&rl)) {
3836 struct ds ds = DS_EMPTY_INITIALIZER;
3837 odp_flow_key_format(drop_key->key, drop_key->key_len, &ds);
3838 VLOG_WARN("Failed to delete drop key (%s) (%s)", strerror(error),
3843 hmap_remove(&backer->drop_keys, &drop_key->hmap_node);
3844 free(drop_key->key);
3849 /* Given a datpath, packet, and flow metadata ('backer', 'packet', and 'key'
3850 * respectively), populates 'flow' with the result of odp_flow_key_to_flow().
3851 * Optionally, if nonnull, populates 'fitnessp' with the fitness of 'flow' as
3852 * returned by odp_flow_key_to_flow(). Also, optionally populates 'ofproto'
3853 * with the ofproto_dpif, and 'odp_in_port' with the datapath in_port, that
3854 * 'packet' ingressed.
3856 * If 'ofproto' is nonnull, requires 'flow''s in_port to exist. Otherwise sets
3857 * 'flow''s in_port to OFPP_NONE.
3859 * This function does post-processing on data returned from
3860 * odp_flow_key_to_flow() to help make VLAN splinters transparent to the rest
3861 * of the upcall processing logic. In particular, if the extracted in_port is
3862 * a VLAN splinter port, it replaces flow->in_port by the "real" port, sets
3863 * flow->vlan_tci correctly for the VLAN of the VLAN splinter port, and pushes
3864 * a VLAN header onto 'packet' (if it is nonnull).
3866 * Optionally, if 'initial_vals' is nonnull, sets 'initial_vals->vlan_tci'
3867 * to the VLAN TCI with which the packet was really received, that is, the
3868 * actual VLAN TCI extracted by odp_flow_key_to_flow(). (This differs from
3869 * the value returned in flow->vlan_tci only for packets received on
3870 * VLAN splinters.) Also, if received on an IP tunnel, sets
3871 * 'initial_vals->tunnel_ip_tos' to the tunnel's IP TOS.
3873 * Similarly, this function also includes some logic to help with tunnels. It
3874 * may modify 'flow' as necessary to make the tunneling implementation
3875 * transparent to the upcall processing logic.
3877 * Returns 0 if successful, ENODEV if the parsed flow has no associated ofport,
3878 * or some other positive errno if there are other problems. */
3880 ofproto_receive(const struct dpif_backer *backer, struct ofpbuf *packet,
3881 const struct nlattr *key, size_t key_len,
3882 struct flow *flow, enum odp_key_fitness *fitnessp,
3883 struct ofproto_dpif **ofproto, uint32_t *odp_in_port,
3884 struct initial_vals *initial_vals)
3886 const struct ofport_dpif *port;
3887 enum odp_key_fitness fitness;
3890 fitness = odp_flow_key_to_flow(key, key_len, flow);
3891 if (fitness == ODP_FIT_ERROR) {
3897 initial_vals->vlan_tci = flow->vlan_tci;
3898 initial_vals->tunnel_ip_tos = flow->tunnel.ip_tos;
3902 *odp_in_port = flow->in_port;
3905 if (tnl_port_should_receive(flow)) {
3906 const struct ofport *ofport = tnl_port_receive(flow);
3908 flow->in_port = OFPP_NONE;
3911 port = ofport_dpif_cast(ofport);
3913 /* We can't reproduce 'key' from 'flow'. */
3914 fitness = fitness == ODP_FIT_PERFECT ? ODP_FIT_TOO_MUCH : fitness;
3916 /* XXX: Since the tunnel module is not scoped per backer, it's
3917 * theoretically possible that we'll receive an ofport belonging to an
3918 * entirely different datapath. In practice, this can't happen because
3919 * no platforms has two separate datapaths which each support
3921 ovs_assert(ofproto_dpif_cast(port->up.ofproto)->backer == backer);
3923 port = odp_port_to_ofport(backer, flow->in_port);
3925 flow->in_port = OFPP_NONE;
3929 flow->in_port = port->up.ofp_port;
3930 if (vsp_adjust_flow(ofproto_dpif_cast(port->up.ofproto), flow)) {
3932 /* Make the packet resemble the flow, so that it gets sent to
3933 * an OpenFlow controller properly, so that it looks correct
3934 * for sFlow, and so that flow_extract() will get the correct
3935 * vlan_tci if it is called on 'packet'.
3937 * The allocated space inside 'packet' probably also contains
3938 * 'key', that is, both 'packet' and 'key' are probably part of
3939 * a struct dpif_upcall (see the large comment on that
3940 * structure definition), so pushing data on 'packet' is in
3941 * general not a good idea since it could overwrite 'key' or
3942 * free it as a side effect. However, it's OK in this special
3943 * case because we know that 'packet' is inside a Netlink
3944 * attribute: pushing 4 bytes will just overwrite the 4-byte
3945 * "struct nlattr", which is fine since we don't need that
3946 * header anymore. */
3947 eth_push_vlan(packet, flow->vlan_tci);
3949 /* We can't reproduce 'key' from 'flow'. */
3950 fitness = fitness == ODP_FIT_PERFECT ? ODP_FIT_TOO_MUCH : fitness;
3956 *ofproto = ofproto_dpif_cast(port->up.ofproto);
3961 *fitnessp = fitness;
3967 handle_miss_upcalls(struct dpif_backer *backer, struct dpif_upcall *upcalls,
3970 struct dpif_upcall *upcall;
3971 struct flow_miss *miss;
3972 struct flow_miss misses[FLOW_MISS_MAX_BATCH];
3973 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
3974 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
3984 /* Construct the to-do list.
3986 * This just amounts to extracting the flow from each packet and sticking
3987 * the packets that have the same flow in the same "flow_miss" structure so
3988 * that we can process them together. */
3991 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
3992 struct flow_miss *miss = &misses[n_misses];
3993 struct flow_miss *existing_miss;
3994 struct ofproto_dpif *ofproto;
3995 uint32_t odp_in_port;
4000 error = ofproto_receive(backer, upcall->packet, upcall->key,
4001 upcall->key_len, &flow, &miss->key_fitness,
4002 &ofproto, &odp_in_port, &miss->initial_vals);
4003 if (error == ENODEV) {
4004 struct drop_key *drop_key;
4006 /* Received packet on port for which we couldn't associate
4007 * an ofproto. This can happen if a port is removed while
4008 * traffic is being received. Print a rate-limited message
4009 * in case it happens frequently. Install a drop flow so
4010 * that future packets of the flow are inexpensively dropped
4012 VLOG_INFO_RL(&rl, "received packet on unassociated port %"PRIu32,
4015 drop_key = drop_key_lookup(backer, upcall->key, upcall->key_len);
4017 drop_key = xmalloc(sizeof *drop_key);
4018 drop_key->key = xmemdup(upcall->key, upcall->key_len);
4019 drop_key->key_len = upcall->key_len;
4021 hmap_insert(&backer->drop_keys, &drop_key->hmap_node,
4022 hash_bytes(drop_key->key, drop_key->key_len, 0));
4023 dpif_flow_put(backer->dpif, DPIF_FP_CREATE | DPIF_FP_MODIFY,
4024 drop_key->key, drop_key->key_len, NULL, 0, NULL);
4032 ofproto->n_missed++;
4033 flow_extract(upcall->packet, flow.skb_priority, flow.skb_mark,
4034 &flow.tunnel, flow.in_port, &miss->flow);
4036 /* Add other packets to a to-do list. */
4037 hash = flow_hash(&miss->flow, 0);
4038 existing_miss = flow_miss_find(&todo, ofproto, &miss->flow, hash);
4039 if (!existing_miss) {
4040 hmap_insert(&todo, &miss->hmap_node, hash);
4041 miss->ofproto = ofproto;
4042 miss->key = upcall->key;
4043 miss->key_len = upcall->key_len;
4044 miss->upcall_type = upcall->type;
4045 miss->odp_in_port = odp_in_port;
4046 list_init(&miss->packets);
4050 miss = existing_miss;
4052 list_push_back(&miss->packets, &upcall->packet->list_node);
4055 /* Process each element in the to-do list, constructing the set of
4056 * operations to batch. */
4058 HMAP_FOR_EACH (miss, hmap_node, &todo) {
4059 handle_flow_miss(miss, flow_miss_ops, &n_ops);
4061 ovs_assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
4063 /* Execute batch. */
4064 for (i = 0; i < n_ops; i++) {
4065 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
4067 dpif_operate(backer->dpif, dpif_ops, n_ops);
4070 for (i = 0; i < n_ops; i++) {
4071 if (flow_miss_ops[i].xout_garbage) {
4072 xlate_out_uninit(&flow_miss_ops[i].xout);
4075 hmap_destroy(&todo);
4078 static enum { SFLOW_UPCALL, MISS_UPCALL, BAD_UPCALL, FLOW_SAMPLE_UPCALL,
4080 classify_upcall(const struct dpif_upcall *upcall)
4082 size_t userdata_len;
4083 union user_action_cookie cookie;
4085 /* First look at the upcall type. */
4086 switch (upcall->type) {
4087 case DPIF_UC_ACTION:
4093 case DPIF_N_UC_TYPES:
4095 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
4099 /* "action" upcalls need a closer look. */
4100 if (!upcall->userdata) {
4101 VLOG_WARN_RL(&rl, "action upcall missing cookie");
4104 userdata_len = nl_attr_get_size(upcall->userdata);
4105 if (userdata_len < sizeof cookie.type
4106 || userdata_len > sizeof cookie) {
4107 VLOG_WARN_RL(&rl, "action upcall cookie has unexpected size %zu",
4111 memset(&cookie, 0, sizeof cookie);
4112 memcpy(&cookie, nl_attr_get(upcall->userdata), userdata_len);
4113 if (userdata_len == sizeof cookie.sflow
4114 && cookie.type == USER_ACTION_COOKIE_SFLOW) {
4115 return SFLOW_UPCALL;
4116 } else if (userdata_len == sizeof cookie.slow_path
4117 && cookie.type == USER_ACTION_COOKIE_SLOW_PATH) {
4119 } else if (userdata_len == sizeof cookie.flow_sample
4120 && cookie.type == USER_ACTION_COOKIE_FLOW_SAMPLE) {
4121 return FLOW_SAMPLE_UPCALL;
4122 } else if (userdata_len == sizeof cookie.ipfix
4123 && cookie.type == USER_ACTION_COOKIE_IPFIX) {
4124 return IPFIX_UPCALL;
4126 VLOG_WARN_RL(&rl, "invalid user cookie of type %"PRIu16
4127 " and size %zu", cookie.type, userdata_len);
4133 handle_sflow_upcall(struct dpif_backer *backer,
4134 const struct dpif_upcall *upcall)
4136 struct ofproto_dpif *ofproto;
4137 union user_action_cookie cookie;
4139 uint32_t odp_in_port;
4141 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
4142 &flow, NULL, &ofproto, &odp_in_port, NULL)
4143 || !ofproto->sflow) {
4147 memset(&cookie, 0, sizeof cookie);
4148 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.sflow);
4149 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
4150 odp_in_port, &cookie);
4154 handle_flow_sample_upcall(struct dpif_backer *backer,
4155 const struct dpif_upcall *upcall)
4157 struct ofproto_dpif *ofproto;
4158 union user_action_cookie cookie;
4161 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
4162 &flow, NULL, &ofproto, NULL, NULL)
4163 || !ofproto->ipfix) {
4167 memset(&cookie, 0, sizeof cookie);
4168 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.flow_sample);
4170 /* The flow reflects exactly the contents of the packet. Sample
4171 * the packet using it. */
4172 dpif_ipfix_flow_sample(ofproto->ipfix, upcall->packet, &flow,
4173 cookie.flow_sample.collector_set_id,
4174 cookie.flow_sample.probability,
4175 cookie.flow_sample.obs_domain_id,
4176 cookie.flow_sample.obs_point_id);
4180 handle_ipfix_upcall(struct dpif_backer *backer,
4181 const struct dpif_upcall *upcall)
4183 struct ofproto_dpif *ofproto;
4186 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
4187 &flow, NULL, &ofproto, NULL, NULL)
4188 || !ofproto->ipfix) {
4192 /* The flow reflects exactly the contents of the packet. Sample
4193 * the packet using it. */
4194 dpif_ipfix_bridge_sample(ofproto->ipfix, upcall->packet, &flow);
4198 handle_upcalls(struct dpif_backer *backer, unsigned int max_batch)
4200 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
4201 struct ofpbuf miss_bufs[FLOW_MISS_MAX_BATCH];
4202 uint64_t miss_buf_stubs[FLOW_MISS_MAX_BATCH][4096 / 8];
4207 ovs_assert(max_batch <= FLOW_MISS_MAX_BATCH);
4210 for (n_processed = 0; n_processed < max_batch; n_processed++) {
4211 struct dpif_upcall *upcall = &misses[n_misses];
4212 struct ofpbuf *buf = &miss_bufs[n_misses];
4215 ofpbuf_use_stub(buf, miss_buf_stubs[n_misses],
4216 sizeof miss_buf_stubs[n_misses]);
4217 error = dpif_recv(backer->dpif, upcall, buf);
4223 switch (classify_upcall(upcall)) {
4225 /* Handle it later. */
4230 handle_sflow_upcall(backer, upcall);
4234 case FLOW_SAMPLE_UPCALL:
4235 handle_flow_sample_upcall(backer, upcall);
4240 handle_ipfix_upcall(backer, upcall);
4250 /* Handle deferred MISS_UPCALL processing. */
4251 handle_miss_upcalls(backer, misses, n_misses);
4252 for (i = 0; i < n_misses; i++) {
4253 ofpbuf_uninit(&miss_bufs[i]);
4259 /* Flow expiration. */
4261 static int subfacet_max_idle(const struct ofproto_dpif *);
4262 static void update_stats(struct dpif_backer *);
4263 static void rule_expire(struct rule_dpif *);
4264 static void expire_subfacets(struct ofproto_dpif *, int dp_max_idle);
4266 /* This function is called periodically by run(). Its job is to collect
4267 * updates for the flows that have been installed into the datapath, most
4268 * importantly when they last were used, and then use that information to
4269 * expire flows that have not been used recently.
4271 * Returns the number of milliseconds after which it should be called again. */
4273 expire(struct dpif_backer *backer)
4275 struct ofproto_dpif *ofproto;
4276 int max_idle = INT32_MAX;
4278 /* Periodically clear out the drop keys in an effort to keep them
4279 * relatively few. */
4280 drop_key_clear(backer);
4282 /* Update stats for each flow in the backer. */
4283 update_stats(backer);
4285 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4286 struct rule *rule, *next_rule;
4289 if (ofproto->backer != backer) {
4293 /* Keep track of the max number of flows per ofproto_dpif. */
4294 update_max_subfacet_count(ofproto);
4296 /* Expire subfacets that have been idle too long. */
4297 dp_max_idle = subfacet_max_idle(ofproto);
4298 expire_subfacets(ofproto, dp_max_idle);
4300 max_idle = MIN(max_idle, dp_max_idle);
4302 /* Expire OpenFlow flows whose idle_timeout or hard_timeout
4304 LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
4305 &ofproto->up.expirable) {
4306 rule_expire(rule_dpif_cast(rule));
4309 /* All outstanding data in existing flows has been accounted, so it's a
4310 * good time to do bond rebalancing. */
4311 if (ofproto->has_bonded_bundles) {
4312 struct ofbundle *bundle;
4314 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
4316 bond_rebalance(bundle->bond, &backer->revalidate_set);
4322 return MIN(max_idle, 1000);
4325 /* Updates flow table statistics given that the datapath just reported 'stats'
4326 * as 'subfacet''s statistics. */
4328 update_subfacet_stats(struct subfacet *subfacet,
4329 const struct dpif_flow_stats *stats)
4331 struct facet *facet = subfacet->facet;
4333 if (stats->n_packets >= subfacet->dp_packet_count) {
4334 uint64_t extra = stats->n_packets - subfacet->dp_packet_count;
4335 facet->packet_count += extra;
4337 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
4340 if (stats->n_bytes >= subfacet->dp_byte_count) {
4341 facet->byte_count += stats->n_bytes - subfacet->dp_byte_count;
4343 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
4346 subfacet->dp_packet_count = stats->n_packets;
4347 subfacet->dp_byte_count = stats->n_bytes;
4349 facet->tcp_flags |= stats->tcp_flags;
4351 subfacet_update_time(subfacet, stats->used);
4352 if (facet->accounted_bytes < facet->byte_count) {
4354 facet_account(facet);
4355 facet->accounted_bytes = facet->byte_count;
4359 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
4360 * about, or a flow that shouldn't be installed but was anyway. Delete it. */
4362 delete_unexpected_flow(struct ofproto_dpif *ofproto,
4363 const struct nlattr *key, size_t key_len)
4365 if (!VLOG_DROP_WARN(&rl)) {
4369 odp_flow_key_format(key, key_len, &s);
4370 VLOG_WARN("unexpected flow on %s: %s", ofproto->up.name, ds_cstr(&s));
4374 COVERAGE_INC(facet_unexpected);
4375 dpif_flow_del(ofproto->backer->dpif, key, key_len, NULL);
4378 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
4380 * This function also pushes statistics updates to rules which each facet
4381 * resubmits into. Generally these statistics will be accurate. However, if a
4382 * facet changes the rule it resubmits into at some time in between
4383 * update_stats() runs, it is possible that statistics accrued to the
4384 * old rule will be incorrectly attributed to the new rule. This could be
4385 * avoided by calling update_stats() whenever rules are created or
4386 * deleted. However, the performance impact of making so many calls to the
4387 * datapath do not justify the benefit of having perfectly accurate statistics.
4389 * In addition, this function maintains per ofproto flow hit counts. The patch
4390 * port is not treated specially. e.g. A packet ingress from br0 patched into
4391 * br1 will increase the hit count of br0 by 1, however, does not affect
4392 * the hit or miss counts of br1.
4395 update_stats(struct dpif_backer *backer)
4397 const struct dpif_flow_stats *stats;
4398 struct dpif_flow_dump dump;
4399 const struct nlattr *key;
4400 struct ofproto_dpif *ofproto;
4403 dpif_flow_dump_start(&dump, backer->dpif);
4404 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
4406 struct subfacet *subfacet;
4407 struct ofport_dpif *ofport;
4410 if (ofproto_receive(backer, NULL, key, key_len, &flow, NULL, &ofproto,
4415 ofproto->total_subfacet_count += hmap_count(&ofproto->subfacets);
4416 ofproto->n_update_stats++;
4418 ofport = get_ofp_port(ofproto, flow.in_port);
4419 if (ofport && ofport->tnl_port) {
4420 netdev_vport_inc_rx(ofport->up.netdev, stats);
4423 key_hash = odp_flow_key_hash(key, key_len);
4424 subfacet = subfacet_find(ofproto, key, key_len, key_hash);
4425 switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
4427 /* Update ofproto_dpif's hit count. */
4428 if (stats->n_packets > subfacet->dp_packet_count) {
4429 uint64_t delta = stats->n_packets - subfacet->dp_packet_count;
4430 dpif_stats_update_hit_count(ofproto, delta);
4433 update_subfacet_stats(subfacet, stats);
4437 /* Stats are updated per-packet. */
4440 case SF_NOT_INSTALLED:
4442 delete_unexpected_flow(ofproto, key, key_len);
4447 dpif_flow_dump_done(&dump);
4449 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4450 update_moving_averages(ofproto);
4455 /* Calculates and returns the number of milliseconds of idle time after which
4456 * subfacets should expire from the datapath. When a subfacet expires, we fold
4457 * its statistics into its facet, and when a facet's last subfacet expires, we
4458 * fold its statistic into its rule. */
4460 subfacet_max_idle(const struct ofproto_dpif *ofproto)
4463 * Idle time histogram.
4465 * Most of the time a switch has a relatively small number of subfacets.
4466 * When this is the case we might as well keep statistics for all of them
4467 * in userspace and to cache them in the kernel datapath for performance as
4470 * As the number of subfacets increases, the memory required to maintain
4471 * statistics about them in userspace and in the kernel becomes
4472 * significant. However, with a large number of subfacets it is likely
4473 * that only a few of them are "heavy hitters" that consume a large amount
4474 * of bandwidth. At this point, only heavy hitters are worth caching in
4475 * the kernel and maintaining in userspaces; other subfacets we can
4478 * The technique used to compute the idle time is to build a histogram with
4479 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
4480 * that is installed in the kernel gets dropped in the appropriate bucket.
4481 * After the histogram has been built, we compute the cutoff so that only
4482 * the most-recently-used 1% of subfacets (but at least
4483 * ofproto->up.flow_eviction_threshold flows) are kept cached. At least
4484 * the most-recently-used bucket of subfacets is kept, so actually an
4485 * arbitrary number of subfacets can be kept in any given expiration run
4486 * (though the next run will delete most of those unless they receive
4489 * This requires a second pass through the subfacets, in addition to the
4490 * pass made by update_stats(), because the former function never looks at
4491 * uninstallable subfacets.
4493 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
4494 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
4495 int buckets[N_BUCKETS] = { 0 };
4496 int total, subtotal, bucket;
4497 struct subfacet *subfacet;
4501 total = hmap_count(&ofproto->subfacets);
4502 if (total <= ofproto->up.flow_eviction_threshold) {
4503 return N_BUCKETS * BUCKET_WIDTH;
4506 /* Build histogram. */
4508 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
4509 long long int idle = now - subfacet->used;
4510 int bucket = (idle <= 0 ? 0
4511 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
4512 : (unsigned int) idle / BUCKET_WIDTH);
4516 /* Find the first bucket whose flows should be expired. */
4517 subtotal = bucket = 0;
4519 subtotal += buckets[bucket++];
4520 } while (bucket < N_BUCKETS &&
4521 subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100));
4523 if (VLOG_IS_DBG_ENABLED()) {
4527 ds_put_cstr(&s, "keep");
4528 for (i = 0; i < N_BUCKETS; i++) {
4530 ds_put_cstr(&s, ", drop");
4533 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
4536 VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s));
4540 return bucket * BUCKET_WIDTH;
4544 expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle)
4546 /* Cutoff time for most flows. */
4547 long long int normal_cutoff = time_msec() - dp_max_idle;
4549 /* We really want to keep flows for special protocols around, so use a more
4550 * conservative cutoff. */
4551 long long int special_cutoff = time_msec() - 10000;
4553 struct subfacet *subfacet, *next_subfacet;
4554 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
4558 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
4559 &ofproto->subfacets) {
4560 long long int cutoff;
4562 cutoff = (subfacet->facet->xout.slow & (SLOW_CFM | SLOW_LACP
4566 if (subfacet->used < cutoff) {
4567 if (subfacet->path != SF_NOT_INSTALLED) {
4568 batch[n_batch++] = subfacet;
4569 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
4570 subfacet_destroy_batch(ofproto, batch, n_batch);
4574 subfacet_destroy(subfacet);
4580 subfacet_destroy_batch(ofproto, batch, n_batch);
4584 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
4585 * then delete it entirely. */
4587 rule_expire(struct rule_dpif *rule)
4589 struct facet *facet, *next_facet;
4593 if (rule->up.pending) {
4594 /* We'll have to expire it later. */
4598 /* Has 'rule' expired? */
4600 if (rule->up.hard_timeout
4601 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
4602 reason = OFPRR_HARD_TIMEOUT;
4603 } else if (rule->up.idle_timeout
4604 && now > rule->up.used + rule->up.idle_timeout * 1000) {
4605 reason = OFPRR_IDLE_TIMEOUT;
4610 COVERAGE_INC(ofproto_dpif_expired);
4612 /* Update stats. (This is a no-op if the rule expired due to an idle
4613 * timeout, because that only happens when the rule has no facets left.) */
4614 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4615 facet_remove(facet);
4618 /* Get rid of the rule. */
4619 ofproto_rule_expire(&rule->up, reason);
4624 /* Creates and returns a new facet based on 'miss'.
4626 * The caller must already have determined that no facet with an identical
4627 * 'miss->flow' exists in 'miss->ofproto'.
4629 * 'hash' must be the return value of flow_hash(miss->flow, 0).
4631 * The facet will initially have no subfacets. The caller should create (at
4632 * least) one subfacet with subfacet_create(). */
4633 static struct facet *
4634 facet_create(const struct flow_miss *miss, uint32_t hash)
4636 struct ofproto_dpif *ofproto = miss->ofproto;
4637 struct xlate_in xin;
4638 struct facet *facet;
4640 facet = xzalloc(sizeof *facet);
4641 facet->used = time_msec();
4642 facet->flow = miss->flow;
4643 facet->initial_vals = miss->initial_vals;
4644 facet->rule = rule_dpif_lookup(ofproto, &facet->flow);
4645 facet->learn_rl = time_msec() + 500;
4647 hmap_insert(&ofproto->facets, &facet->hmap_node, hash);
4648 list_push_back(&facet->rule->facets, &facet->list_node);
4649 list_init(&facet->subfacets);
4650 netflow_flow_init(&facet->nf_flow);
4651 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
4653 xlate_in_init(&xin, ofproto, &facet->flow, &facet->initial_vals,
4654 facet->rule, 0, NULL);
4655 xin.may_learn = true;
4656 xlate_actions(&xin, &facet->xout);
4657 facet->nf_flow.output_iface = facet->xout.nf_output_iface;
4663 facet_free(struct facet *facet)
4666 xlate_out_uninit(&facet->xout);
4671 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
4672 * 'packet', which arrived on 'in_port'. */
4674 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
4675 const struct nlattr *odp_actions, size_t actions_len,
4676 struct ofpbuf *packet)
4678 struct odputil_keybuf keybuf;
4682 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4683 odp_flow_key_from_flow(&key, flow,
4684 ofp_port_to_odp_port(ofproto, flow->in_port));
4686 error = dpif_execute(ofproto->backer->dpif, key.data, key.size,
4687 odp_actions, actions_len, packet);
4691 /* Remove 'facet' from 'ofproto' and free up the associated memory:
4693 * - If 'facet' was installed in the datapath, uninstalls it and updates its
4694 * rule's statistics, via subfacet_uninstall().
4696 * - Removes 'facet' from its rule and from ofproto->facets.
4699 facet_remove(struct facet *facet)
4701 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4702 struct subfacet *subfacet, *next_subfacet;
4704 ovs_assert(!list_is_empty(&facet->subfacets));
4706 /* First uninstall all of the subfacets to get final statistics. */
4707 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4708 subfacet_uninstall(subfacet);
4711 /* Flush the final stats to the rule.
4713 * This might require us to have at least one subfacet around so that we
4714 * can use its actions for accounting in facet_account(), which is why we
4715 * have uninstalled but not yet destroyed the subfacets. */
4716 facet_flush_stats(facet);
4718 /* Now we're really all done so destroy everything. */
4719 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
4720 &facet->subfacets) {
4721 subfacet_destroy__(subfacet);
4723 hmap_remove(&ofproto->facets, &facet->hmap_node);
4724 list_remove(&facet->list_node);
4728 /* Feed information from 'facet' back into the learning table to keep it in
4729 * sync with what is actually flowing through the datapath. */
4731 facet_learn(struct facet *facet)
4733 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4734 long long int now = time_msec();
4735 struct xlate_in xin;
4737 if (!facet->xout.has_fin_timeout && now < facet->learn_rl) {
4741 facet->learn_rl = now + 500;
4743 if (!facet->xout.has_learn
4744 && !facet->xout.has_normal
4745 && (!facet->xout.has_fin_timeout
4746 || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
4750 xlate_in_init(&xin, ofproto, &facet->flow, &facet->initial_vals,
4751 facet->rule, facet->tcp_flags, NULL);
4752 xin.may_learn = true;
4753 xlate_actions_for_side_effects(&xin);
4757 facet_account(struct facet *facet)
4759 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4760 const struct nlattr *a;
4765 if (!facet->xout.has_normal || !ofproto->has_bonded_bundles) {
4768 n_bytes = facet->byte_count - facet->accounted_bytes;
4770 /* This loop feeds byte counters to bond_account() for rebalancing to use
4771 * as a basis. We also need to track the actual VLAN on which the packet
4772 * is going to be sent to ensure that it matches the one passed to
4773 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
4776 * We use the actions from an arbitrary subfacet because they should all
4777 * be equally valid for our purpose. */
4778 vlan_tci = facet->flow.vlan_tci;
4779 NL_ATTR_FOR_EACH_UNSAFE (a, left, facet->xout.odp_actions.data,
4780 facet->xout.odp_actions.size) {
4781 const struct ovs_action_push_vlan *vlan;
4782 struct ofport_dpif *port;
4784 switch (nl_attr_type(a)) {
4785 case OVS_ACTION_ATTR_OUTPUT:
4786 port = get_odp_port(ofproto, nl_attr_get_u32(a));
4787 if (port && port->bundle && port->bundle->bond) {
4788 bond_account(port->bundle->bond, &facet->flow,
4789 vlan_tci_to_vid(vlan_tci), n_bytes);
4793 case OVS_ACTION_ATTR_POP_VLAN:
4794 vlan_tci = htons(0);
4797 case OVS_ACTION_ATTR_PUSH_VLAN:
4798 vlan = nl_attr_get(a);
4799 vlan_tci = vlan->vlan_tci;
4805 /* Returns true if the only action for 'facet' is to send to the controller.
4806 * (We don't report NetFlow expiration messages for such facets because they
4807 * are just part of the control logic for the network, not real traffic). */
4809 facet_is_controller_flow(struct facet *facet)
4812 const struct rule *rule = &facet->rule->up;
4813 const struct ofpact *ofpacts = rule->ofpacts;
4814 size_t ofpacts_len = rule->ofpacts_len;
4816 if (ofpacts_len > 0 &&
4817 ofpacts->type == OFPACT_CONTROLLER &&
4818 ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len)) {
4825 /* Folds all of 'facet''s statistics into its rule. Also updates the
4826 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
4827 * 'facet''s statistics in the datapath should have been zeroed and folded into
4828 * its packet and byte counts before this function is called. */
4830 facet_flush_stats(struct facet *facet)
4832 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4833 struct subfacet *subfacet;
4835 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4836 ovs_assert(!subfacet->dp_byte_count);
4837 ovs_assert(!subfacet->dp_packet_count);
4840 facet_push_stats(facet);
4841 if (facet->accounted_bytes < facet->byte_count) {
4842 facet_account(facet);
4843 facet->accounted_bytes = facet->byte_count;
4846 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
4847 struct ofexpired expired;
4848 expired.flow = facet->flow;
4849 expired.packet_count = facet->packet_count;
4850 expired.byte_count = facet->byte_count;
4851 expired.used = facet->used;
4852 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4855 facet->rule->packet_count += facet->packet_count;
4856 facet->rule->byte_count += facet->byte_count;
4858 /* Reset counters to prevent double counting if 'facet' ever gets
4860 facet_reset_counters(facet);
4862 netflow_flow_clear(&facet->nf_flow);
4863 facet->tcp_flags = 0;
4866 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
4867 * Returns it if found, otherwise a null pointer.
4869 * 'hash' must be the return value of flow_hash(flow, 0).
4871 * The returned facet might need revalidation; use facet_lookup_valid()
4872 * instead if that is important. */
4873 static struct facet *
4874 facet_find(struct ofproto_dpif *ofproto,
4875 const struct flow *flow, uint32_t hash)
4877 struct facet *facet;
4879 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, hash, &ofproto->facets) {
4880 if (flow_equal(flow, &facet->flow)) {
4888 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
4889 * Returns it if found, otherwise a null pointer.
4891 * 'hash' must be the return value of flow_hash(flow, 0).
4893 * The returned facet is guaranteed to be valid. */
4894 static struct facet *
4895 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow,
4898 struct facet *facet;
4900 facet = facet_find(ofproto, flow, hash);
4902 && (ofproto->backer->need_revalidate
4903 || tag_set_intersects(&ofproto->backer->revalidate_set,
4905 && !facet_revalidate(facet)) {
4913 facet_check_consistency(struct facet *facet)
4915 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
4917 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4919 struct xlate_out xout;
4920 struct xlate_in xin;
4922 struct rule_dpif *rule;
4925 /* Check the rule for consistency. */
4926 rule = rule_dpif_lookup(ofproto, &facet->flow);
4927 if (rule != facet->rule) {
4928 if (!VLOG_DROP_WARN(&rl)) {
4929 struct ds s = DS_EMPTY_INITIALIZER;
4931 flow_format(&s, &facet->flow);
4932 ds_put_format(&s, ": facet associated with wrong rule (was "
4933 "table=%"PRIu8",", facet->rule->up.table_id);
4934 cls_rule_format(&facet->rule->up.cr, &s);
4935 ds_put_format(&s, ") (should have been table=%"PRIu8",",
4937 cls_rule_format(&rule->up.cr, &s);
4938 ds_put_cstr(&s, ")\n");
4945 /* Check the datapath actions for consistency. */
4946 xlate_in_init(&xin, ofproto, &facet->flow, &facet->initial_vals, rule,
4948 xlate_actions(&xin, &xout);
4950 ok = ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)
4951 && facet->xout.slow == xout.slow;
4952 if (!ok && !VLOG_DROP_WARN(&rl)) {
4953 struct ds s = DS_EMPTY_INITIALIZER;
4955 flow_format(&s, &facet->flow);
4956 ds_put_cstr(&s, ": inconsistency in facet");
4958 if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4959 ds_put_cstr(&s, " (actions were: ");
4960 format_odp_actions(&s, facet->xout.odp_actions.data,
4961 facet->xout.odp_actions.size);
4962 ds_put_cstr(&s, ") (correct actions: ");
4963 format_odp_actions(&s, xout.odp_actions.data,
4964 xout.odp_actions.size);
4965 ds_put_cstr(&s, ")");
4968 if (facet->xout.slow != xout.slow) {
4969 ds_put_format(&s, " slow path incorrect. should be %d", xout.slow);
4974 xlate_out_uninit(&xout);
4979 /* Re-searches the classifier for 'facet':
4981 * - If the rule found is different from 'facet''s current rule, moves
4982 * 'facet' to the new rule and recompiles its actions.
4984 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
4985 * where it is and recompiles its actions anyway.
4987 * - If any of 'facet''s subfacets correspond to a new flow according to
4988 * ofproto_receive(), 'facet' is removed.
4990 * Returns true if 'facet' is still valid. False if 'facet' was removed. */
4992 facet_revalidate(struct facet *facet)
4994 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4995 struct rule_dpif *new_rule;
4996 struct subfacet *subfacet;
4997 struct xlate_out xout;
4998 struct xlate_in xin;
5000 COVERAGE_INC(facet_revalidate);
5002 /* Check that child subfacets still correspond to this facet. Tunnel
5003 * configuration changes could cause a subfacet's OpenFlow in_port to
5005 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
5006 struct ofproto_dpif *recv_ofproto;
5007 struct flow recv_flow;
5010 error = ofproto_receive(ofproto->backer, NULL, subfacet->key,
5011 subfacet->key_len, &recv_flow, NULL,
5012 &recv_ofproto, NULL, NULL);
5014 || recv_ofproto != ofproto
5015 || memcmp(&recv_flow, &facet->flow, sizeof recv_flow)) {
5016 facet_remove(facet);
5021 new_rule = rule_dpif_lookup(ofproto, &facet->flow);
5023 /* Calculate new datapath actions.
5025 * We do not modify any 'facet' state yet, because we might need to, e.g.,
5026 * emit a NetFlow expiration and, if so, we need to have the old state
5027 * around to properly compose it. */
5028 xlate_in_init(&xin, ofproto, &facet->flow, &facet->initial_vals, new_rule,
5030 xlate_actions(&xin, &xout);
5032 /* A facet's slow path reason should only change under dramatic
5033 * circumstances. Rather than try to update everything, it's simpler to
5034 * remove the facet and start over. */
5035 if (facet->xout.slow != xout.slow) {
5036 facet_remove(facet);
5037 xlate_out_uninit(&xout);
5041 if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
5042 LIST_FOR_EACH(subfacet, list_node, &facet->subfacets) {
5043 if (subfacet->path == SF_FAST_PATH) {
5044 struct dpif_flow_stats stats;
5046 subfacet_install(subfacet, &xout.odp_actions, &stats);
5047 subfacet_update_stats(subfacet, &stats);
5051 facet_flush_stats(facet);
5053 ofpbuf_clear(&facet->xout.odp_actions);
5054 ofpbuf_put(&facet->xout.odp_actions, xout.odp_actions.data,
5055 xout.odp_actions.size);
5058 /* Update 'facet' now that we've taken care of all the old state. */
5059 facet->xout.tags = xout.tags;
5060 facet->xout.slow = xout.slow;
5061 facet->xout.has_learn = xout.has_learn;
5062 facet->xout.has_normal = xout.has_normal;
5063 facet->xout.has_fin_timeout = xout.has_fin_timeout;
5064 facet->xout.nf_output_iface = xout.nf_output_iface;
5065 facet->xout.mirrors = xout.mirrors;
5066 facet->nf_flow.output_iface = facet->xout.nf_output_iface;
5068 if (facet->rule != new_rule) {
5069 COVERAGE_INC(facet_changed_rule);
5070 list_remove(&facet->list_node);
5071 list_push_back(&new_rule->facets, &facet->list_node);
5072 facet->rule = new_rule;
5073 facet->used = new_rule->up.created;
5074 facet->prev_used = facet->used;
5077 xlate_out_uninit(&xout);
5081 /* Updates 'facet''s used time. Caller is responsible for calling
5082 * facet_push_stats() to update the flows which 'facet' resubmits into. */
5084 facet_update_time(struct facet *facet, long long int used)
5086 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5087 if (used > facet->used) {
5089 ofproto_rule_update_used(&facet->rule->up, used);
5090 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
5095 facet_reset_counters(struct facet *facet)
5097 facet->packet_count = 0;
5098 facet->byte_count = 0;
5099 facet->prev_packet_count = 0;
5100 facet->prev_byte_count = 0;
5101 facet->accounted_bytes = 0;
5105 facet_push_stats(struct facet *facet)
5107 struct dpif_flow_stats stats;
5109 ovs_assert(facet->packet_count >= facet->prev_packet_count);
5110 ovs_assert(facet->byte_count >= facet->prev_byte_count);
5111 ovs_assert(facet->used >= facet->prev_used);
5113 stats.n_packets = facet->packet_count - facet->prev_packet_count;
5114 stats.n_bytes = facet->byte_count - facet->prev_byte_count;
5115 stats.used = facet->used;
5116 stats.tcp_flags = 0;
5118 if (stats.n_packets || stats.n_bytes || facet->used > facet->prev_used) {
5119 facet->prev_packet_count = facet->packet_count;
5120 facet->prev_byte_count = facet->byte_count;
5121 facet->prev_used = facet->used;
5123 flow_push_stats(facet, &stats);
5125 update_mirror_stats(ofproto_dpif_cast(facet->rule->up.ofproto),
5126 facet->xout.mirrors, stats.n_packets,
5132 push_all_stats__(bool run_fast)
5134 static long long int rl = LLONG_MIN;
5135 struct ofproto_dpif *ofproto;
5137 if (time_msec() < rl) {
5141 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5142 struct facet *facet;
5144 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
5145 facet_push_stats(facet);
5152 rl = time_msec() + 100;
5156 push_all_stats(void)
5158 push_all_stats__(true);
5162 rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats)
5164 rule->packet_count += stats->n_packets;
5165 rule->byte_count += stats->n_bytes;
5166 ofproto_rule_update_used(&rule->up, stats->used);
5169 /* Pushes flow statistics to the rules which 'facet->flow' resubmits
5170 * into given 'facet->rule''s actions and mirrors. */
5172 flow_push_stats(struct facet *facet, const struct dpif_flow_stats *stats)
5174 struct rule_dpif *rule = facet->rule;
5175 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5176 struct xlate_in xin;
5178 ofproto_rule_update_used(&rule->up, stats->used);
5180 xlate_in_init(&xin, ofproto, &facet->flow, &facet->initial_vals, rule,
5182 xin.resubmit_stats = stats;
5183 xlate_actions_for_side_effects(&xin);
5188 static struct subfacet *
5189 subfacet_find(struct ofproto_dpif *ofproto,
5190 const struct nlattr *key, size_t key_len, uint32_t key_hash)
5192 struct subfacet *subfacet;
5194 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
5195 &ofproto->subfacets) {
5196 if (subfacet->key_len == key_len
5197 && !memcmp(key, subfacet->key, key_len)) {
5205 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
5206 * 'key_fitness', 'key', and 'key_len' members in 'miss'. Returns the
5207 * existing subfacet if there is one, otherwise creates and returns a
5209 static struct subfacet *
5210 subfacet_create(struct facet *facet, struct flow_miss *miss,
5213 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5214 enum odp_key_fitness key_fitness = miss->key_fitness;
5215 const struct nlattr *key = miss->key;
5216 size_t key_len = miss->key_len;
5218 struct subfacet *subfacet;
5220 key_hash = odp_flow_key_hash(key, key_len);
5222 if (list_is_empty(&facet->subfacets)) {
5223 subfacet = &facet->one_subfacet;
5225 subfacet = subfacet_find(ofproto, key, key_len, key_hash);
5227 if (subfacet->facet == facet) {
5231 /* This shouldn't happen. */
5232 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
5233 subfacet_destroy(subfacet);
5236 subfacet = xmalloc(sizeof *subfacet);
5239 hmap_insert(&ofproto->subfacets, &subfacet->hmap_node, key_hash);
5240 list_push_back(&facet->subfacets, &subfacet->list_node);
5241 subfacet->facet = facet;
5242 subfacet->key_fitness = key_fitness;
5243 subfacet->key = xmemdup(key, key_len);
5244 subfacet->key_len = key_len;
5245 subfacet->used = now;
5246 subfacet->created = now;
5247 subfacet->dp_packet_count = 0;
5248 subfacet->dp_byte_count = 0;
5249 subfacet->path = SF_NOT_INSTALLED;
5250 subfacet->odp_in_port = miss->odp_in_port;
5252 ofproto->subfacet_add_count++;
5256 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
5257 * its facet within 'ofproto', and frees it. */
5259 subfacet_destroy__(struct subfacet *subfacet)
5261 struct facet *facet = subfacet->facet;
5262 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5264 /* Update ofproto stats before uninstall the subfacet. */
5265 ofproto->subfacet_del_count++;
5266 ofproto->total_subfacet_life_span += (time_msec() - subfacet->created);
5268 subfacet_uninstall(subfacet);
5269 hmap_remove(&ofproto->subfacets, &subfacet->hmap_node);
5270 list_remove(&subfacet->list_node);
5271 free(subfacet->key);
5272 if (subfacet != &facet->one_subfacet) {
5277 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
5278 * last remaining subfacet in its facet destroys the facet too. */
5280 subfacet_destroy(struct subfacet *subfacet)
5282 struct facet *facet = subfacet->facet;
5284 if (list_is_singleton(&facet->subfacets)) {
5285 /* facet_remove() needs at least one subfacet (it will remove it). */
5286 facet_remove(facet);
5288 subfacet_destroy__(subfacet);
5293 subfacet_destroy_batch(struct ofproto_dpif *ofproto,
5294 struct subfacet **subfacets, int n)
5296 struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH];
5297 struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH];
5298 struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH];
5301 for (i = 0; i < n; i++) {
5302 ops[i].type = DPIF_OP_FLOW_DEL;
5303 ops[i].u.flow_del.key = subfacets[i]->key;
5304 ops[i].u.flow_del.key_len = subfacets[i]->key_len;
5305 ops[i].u.flow_del.stats = &stats[i];
5309 dpif_operate(ofproto->backer->dpif, opsp, n);
5310 for (i = 0; i < n; i++) {
5311 subfacet_reset_dp_stats(subfacets[i], &stats[i]);
5312 subfacets[i]->path = SF_NOT_INSTALLED;
5313 subfacet_destroy(subfacets[i]);
5318 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
5319 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
5320 * in the datapath will be zeroed and 'stats' will be updated with traffic new
5321 * since 'subfacet' was last updated.
5323 * Returns 0 if successful, otherwise a positive errno value. */
5325 subfacet_install(struct subfacet *subfacet, const struct ofpbuf *odp_actions,
5326 struct dpif_flow_stats *stats)
5328 struct facet *facet = subfacet->facet;
5329 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5330 enum subfacet_path path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
5331 const struct nlattr *actions = odp_actions->data;
5332 size_t actions_len = odp_actions->size;
5334 uint64_t slow_path_stub[128 / 8];
5335 enum dpif_flow_put_flags flags;
5338 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
5340 flags |= DPIF_FP_ZERO_STATS;
5343 if (path == SF_SLOW_PATH) {
5344 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
5345 slow_path_stub, sizeof slow_path_stub,
5346 &actions, &actions_len);
5349 ret = dpif_flow_put(ofproto->backer->dpif, flags, subfacet->key,
5350 subfacet->key_len, actions, actions_len, stats);
5353 subfacet_reset_dp_stats(subfacet, stats);
5357 subfacet->path = path;
5362 /* If 'subfacet' is installed in the datapath, uninstalls it. */
5364 subfacet_uninstall(struct subfacet *subfacet)
5366 if (subfacet->path != SF_NOT_INSTALLED) {
5367 struct rule_dpif *rule = subfacet->facet->rule;
5368 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5369 struct dpif_flow_stats stats;
5372 error = dpif_flow_del(ofproto->backer->dpif, subfacet->key,
5373 subfacet->key_len, &stats);
5374 subfacet_reset_dp_stats(subfacet, &stats);
5376 subfacet_update_stats(subfacet, &stats);
5378 subfacet->path = SF_NOT_INSTALLED;
5380 ovs_assert(subfacet->dp_packet_count == 0);
5381 ovs_assert(subfacet->dp_byte_count == 0);
5385 /* Resets 'subfacet''s datapath statistics counters. This should be called
5386 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
5387 * non-null, it should contain the statistics returned by dpif when 'subfacet'
5388 * was reset in the datapath. 'stats' will be modified to include only
5389 * statistics new since 'subfacet' was last updated. */
5391 subfacet_reset_dp_stats(struct subfacet *subfacet,
5392 struct dpif_flow_stats *stats)
5395 && subfacet->dp_packet_count <= stats->n_packets
5396 && subfacet->dp_byte_count <= stats->n_bytes) {
5397 stats->n_packets -= subfacet->dp_packet_count;
5398 stats->n_bytes -= subfacet->dp_byte_count;
5401 subfacet->dp_packet_count = 0;
5402 subfacet->dp_byte_count = 0;
5405 /* Updates 'subfacet''s used time. The caller is responsible for calling
5406 * facet_push_stats() to update the flows which 'subfacet' resubmits into. */
5408 subfacet_update_time(struct subfacet *subfacet, long long int used)
5410 if (used > subfacet->used) {
5411 subfacet->used = used;
5412 facet_update_time(subfacet->facet, used);
5416 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
5418 * Because of the meaning of a subfacet's counters, it only makes sense to do
5419 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
5420 * represents a packet that was sent by hand or if it represents statistics
5421 * that have been cleared out of the datapath. */
5423 subfacet_update_stats(struct subfacet *subfacet,
5424 const struct dpif_flow_stats *stats)
5426 if (stats->n_packets || stats->used > subfacet->used) {
5427 struct facet *facet = subfacet->facet;
5429 subfacet_update_time(subfacet, stats->used);
5430 facet->packet_count += stats->n_packets;
5431 facet->byte_count += stats->n_bytes;
5432 facet->tcp_flags |= stats->tcp_flags;
5433 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
5439 static struct rule_dpif *
5440 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow)
5442 struct rule_dpif *rule;
5444 rule = rule_dpif_lookup__(ofproto, flow, 0);
5449 return rule_dpif_miss_rule(ofproto, flow);
5452 static struct rule_dpif *
5453 rule_dpif_lookup__(struct ofproto_dpif *ofproto, const struct flow *flow,
5456 struct cls_rule *cls_rule;
5457 struct classifier *cls;
5459 if (table_id >= N_TABLES) {
5463 cls = &ofproto->up.tables[table_id].cls;
5464 if (flow->nw_frag & FLOW_NW_FRAG_ANY
5465 && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
5466 /* For OFPC_NORMAL frag_handling, we must pretend that transport ports
5467 * are unavailable. */
5468 struct flow ofpc_normal_flow = *flow;
5469 ofpc_normal_flow.tp_src = htons(0);
5470 ofpc_normal_flow.tp_dst = htons(0);
5471 cls_rule = classifier_lookup(cls, &ofpc_normal_flow);
5473 cls_rule = classifier_lookup(cls, flow);
5475 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
5478 static struct rule_dpif *
5479 rule_dpif_miss_rule(struct ofproto_dpif *ofproto, const struct flow *flow)
5481 struct ofport_dpif *port;
5483 port = get_ofp_port(ofproto, flow->in_port);
5485 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, flow->in_port);
5486 return ofproto->miss_rule;
5489 if (port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN) {
5490 return ofproto->no_packet_in_rule;
5492 return ofproto->miss_rule;
5496 complete_operation(struct rule_dpif *rule)
5498 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5500 rule_invalidate(rule);
5502 struct dpif_completion *c = xmalloc(sizeof *c);
5503 c->op = rule->up.pending;
5504 list_push_back(&ofproto->completions, &c->list_node);
5506 ofoperation_complete(rule->up.pending, 0);
5510 static struct rule *
5513 struct rule_dpif *rule = xmalloc(sizeof *rule);
5518 rule_dealloc(struct rule *rule_)
5520 struct rule_dpif *rule = rule_dpif_cast(rule_);
5525 rule_construct(struct rule *rule_)
5527 struct rule_dpif *rule = rule_dpif_cast(rule_);
5528 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5529 struct rule_dpif *victim;
5532 rule->packet_count = 0;
5533 rule->byte_count = 0;
5535 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
5536 if (victim && !list_is_empty(&victim->facets)) {
5537 struct facet *facet;
5539 rule->facets = victim->facets;
5540 list_moved(&rule->facets);
5541 LIST_FOR_EACH (facet, list_node, &rule->facets) {
5542 /* XXX: We're only clearing our local counters here. It's possible
5543 * that quite a few packets are unaccounted for in the datapath
5544 * statistics. These will be accounted to the new rule instead of
5545 * cleared as required. This could be fixed by clearing out the
5546 * datapath statistics for this facet, but currently it doesn't
5548 facet_reset_counters(facet);
5552 /* Must avoid list_moved() in this case. */
5553 list_init(&rule->facets);
5556 table_id = rule->up.table_id;
5558 rule->tag = victim->tag;
5559 } else if (table_id == 0) {
5564 miniflow_expand(&rule->up.cr.match.flow, &flow);
5565 rule->tag = rule_calculate_tag(&flow, &rule->up.cr.match.mask,
5566 ofproto->tables[table_id].basis);
5569 complete_operation(rule);
5574 rule_destruct(struct rule *rule_)
5576 struct rule_dpif *rule = rule_dpif_cast(rule_);
5577 struct facet *facet, *next_facet;
5579 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
5580 facet_revalidate(facet);
5583 complete_operation(rule);
5587 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
5589 struct rule_dpif *rule = rule_dpif_cast(rule_);
5590 struct facet *facet;
5592 /* push_all_stats() can handle flow misses which, when using the learn
5593 * action, can cause rules to be added and deleted. This can corrupt our
5594 * caller's datastructures which assume that rule_get_stats() doesn't have
5595 * an impact on the flow table. To be safe, we disable miss handling. */
5596 push_all_stats__(false);
5598 /* Start from historical data for 'rule' itself that are no longer tracked
5599 * in facets. This counts, for example, facets that have expired. */
5600 *packets = rule->packet_count;
5601 *bytes = rule->byte_count;
5603 /* Add any statistics that are tracked by facets. This includes
5604 * statistical data recently updated by ofproto_update_stats() as well as
5605 * stats for packets that were executed "by hand" via dpif_execute(). */
5606 LIST_FOR_EACH (facet, list_node, &rule->facets) {
5607 *packets += facet->packet_count;
5608 *bytes += facet->byte_count;
5613 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
5614 struct ofpbuf *packet)
5616 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5617 struct initial_vals initial_vals;
5618 struct dpif_flow_stats stats;
5619 struct xlate_out xout;
5620 struct xlate_in xin;
5622 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
5623 rule_credit_stats(rule, &stats);
5625 initial_vals.vlan_tci = flow->vlan_tci;
5626 initial_vals.tunnel_ip_tos = flow->tunnel.ip_tos;
5627 xlate_in_init(&xin, ofproto, flow, &initial_vals, rule, stats.tcp_flags,
5629 xin.resubmit_stats = &stats;
5630 xlate_actions(&xin, &xout);
5632 execute_odp_actions(ofproto, flow, xout.odp_actions.data,
5633 xout.odp_actions.size, packet);
5635 xlate_out_uninit(&xout);
5639 rule_execute(struct rule *rule, const struct flow *flow,
5640 struct ofpbuf *packet)
5642 rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
5643 ofpbuf_delete(packet);
5648 rule_modify_actions(struct rule *rule_)
5650 struct rule_dpif *rule = rule_dpif_cast(rule_);
5652 complete_operation(rule);
5655 /* Sends 'packet' out 'ofport'.
5656 * May modify 'packet'.
5657 * Returns 0 if successful, otherwise a positive errno value. */
5659 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
5661 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
5662 uint64_t odp_actions_stub[1024 / 8];
5663 struct ofpbuf key, odp_actions;
5664 struct odputil_keybuf keybuf;
5669 flow_extract(packet, 0, 0, NULL, OFPP_LOCAL, &flow);
5670 if (netdev_vport_is_patch(ofport->up.netdev)) {
5671 struct ofproto_dpif *peer_ofproto;
5672 struct dpif_flow_stats stats;
5673 struct ofport_dpif *peer;
5674 struct rule_dpif *rule;
5676 peer = ofport_get_peer(ofport);
5681 dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
5682 netdev_vport_inc_tx(ofport->up.netdev, &stats);
5683 netdev_vport_inc_rx(peer->up.netdev, &stats);
5685 flow.in_port = peer->up.ofp_port;
5686 peer_ofproto = ofproto_dpif_cast(peer->up.ofproto);
5687 rule = rule_dpif_lookup(peer_ofproto, &flow);
5688 rule_dpif_execute(rule, &flow, packet);
5693 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5695 if (ofport->tnl_port) {
5696 struct dpif_flow_stats stats;
5698 odp_port = tnl_port_send(ofport->tnl_port, &flow);
5699 if (odp_port == OVSP_NONE) {
5703 dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
5704 netdev_vport_inc_tx(ofport->up.netdev, &stats);
5705 odp_put_tunnel_action(&flow.tunnel, &odp_actions);
5706 odp_put_skb_mark_action(flow.skb_mark, &odp_actions);
5708 odp_port = vsp_realdev_to_vlandev(ofproto, ofport->odp_port,
5710 if (odp_port != ofport->odp_port) {
5711 eth_pop_vlan(packet);
5712 flow.vlan_tci = htons(0);
5716 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5717 odp_flow_key_from_flow(&key, &flow,
5718 ofp_port_to_odp_port(ofproto, flow.in_port));
5720 compose_sflow_action(ofproto, &odp_actions, &flow, odp_port);
5721 compose_ipfix_action(ofproto, &odp_actions, &flow);
5723 nl_msg_put_u32(&odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port);
5724 error = dpif_execute(ofproto->backer->dpif,
5726 odp_actions.data, odp_actions.size,
5728 ofpbuf_uninit(&odp_actions);
5731 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)",
5732 ofproto->up.name, odp_port, strerror(error));
5734 ofproto_update_local_port_stats(ofport->up.ofproto, packet->size, 0);
5738 /* OpenFlow to datapath action translation. */
5740 static bool may_receive(const struct ofport_dpif *, struct xlate_ctx *);
5741 static void do_xlate_actions(const struct ofpact *, size_t ofpacts_len,
5742 struct xlate_ctx *);
5743 static void xlate_normal(struct xlate_ctx *);
5745 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
5746 * The action will state 'slow' as the reason that the action is in the slow
5747 * path. (This is purely informational: it allows a human viewing "ovs-dpctl
5748 * dump-flows" output to see why a flow is in the slow path.)
5750 * The 'stub_size' bytes in 'stub' will be used to store the action.
5751 * 'stub_size' must be large enough for the action.
5753 * The action and its size will be stored in '*actionsp' and '*actions_lenp',
5756 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
5757 enum slow_path_reason slow,
5758 uint64_t *stub, size_t stub_size,
5759 const struct nlattr **actionsp, size_t *actions_lenp)
5761 union user_action_cookie cookie;
5764 cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
5765 cookie.slow_path.unused = 0;
5766 cookie.slow_path.reason = slow;
5768 ofpbuf_use_stack(&buf, stub, stub_size);
5769 if (slow & (SLOW_CFM | SLOW_LACP | SLOW_STP)) {
5770 uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif, UINT32_MAX);
5771 odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
5773 put_userspace_action(ofproto, &buf, flow, &cookie,
5774 sizeof cookie.slow_path);
5776 *actionsp = buf.data;
5777 *actions_lenp = buf.size;
5781 put_userspace_action(const struct ofproto_dpif *ofproto,
5782 struct ofpbuf *odp_actions,
5783 const struct flow *flow,
5784 const union user_action_cookie *cookie,
5785 const size_t cookie_size)
5789 pid = dpif_port_get_pid(ofproto->backer->dpif,
5790 ofp_port_to_odp_port(ofproto, flow->in_port));
5792 return odp_put_userspace_action(pid, cookie, cookie_size, odp_actions);
5795 /* Compose SAMPLE action for sFlow or IPFIX. The given probability is
5796 * the number of packets out of UINT32_MAX to sample. The given
5797 * cookie is passed back in the callback for each sampled packet.
5800 compose_sample_action(const struct ofproto_dpif *ofproto,
5801 struct ofpbuf *odp_actions,
5802 const struct flow *flow,
5803 const uint32_t probability,
5804 const union user_action_cookie *cookie,
5805 const size_t cookie_size)
5807 size_t sample_offset, actions_offset;
5810 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
5812 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
5814 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
5815 cookie_offset = put_userspace_action(ofproto, odp_actions, flow, cookie,
5818 nl_msg_end_nested(odp_actions, actions_offset);
5819 nl_msg_end_nested(odp_actions, sample_offset);
5820 return cookie_offset;
5824 compose_sflow_cookie(const struct ofproto_dpif *ofproto,
5825 ovs_be16 vlan_tci, uint32_t odp_port,
5826 unsigned int n_outputs, union user_action_cookie *cookie)
5830 cookie->type = USER_ACTION_COOKIE_SFLOW;
5831 cookie->sflow.vlan_tci = vlan_tci;
5833 /* See http://www.sflow.org/sflow_version_5.txt (search for "Input/output
5834 * port information") for the interpretation of cookie->output. */
5835 switch (n_outputs) {
5837 /* 0x40000000 | 256 means "packet dropped for unknown reason". */
5838 cookie->sflow.output = 0x40000000 | 256;
5842 ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port);
5844 cookie->sflow.output = ifindex;
5849 /* 0x80000000 means "multiple output ports. */
5850 cookie->sflow.output = 0x80000000 | n_outputs;
5855 /* Compose SAMPLE action for sFlow bridge sampling. */
5857 compose_sflow_action(const struct ofproto_dpif *ofproto,
5858 struct ofpbuf *odp_actions,
5859 const struct flow *flow,
5862 uint32_t probability;
5863 union user_action_cookie cookie;
5865 if (!ofproto->sflow || flow->in_port == OFPP_NONE) {
5869 probability = dpif_sflow_get_probability(ofproto->sflow);
5870 compose_sflow_cookie(ofproto, htons(0), odp_port,
5871 odp_port == OVSP_NONE ? 0 : 1, &cookie);
5873 return compose_sample_action(ofproto, odp_actions, flow, probability,
5874 &cookie, sizeof cookie.sflow);
5878 compose_flow_sample_cookie(uint16_t probability, uint32_t collector_set_id,
5879 uint32_t obs_domain_id, uint32_t obs_point_id,
5880 union user_action_cookie *cookie)
5882 cookie->type = USER_ACTION_COOKIE_FLOW_SAMPLE;
5883 cookie->flow_sample.probability = probability;
5884 cookie->flow_sample.collector_set_id = collector_set_id;
5885 cookie->flow_sample.obs_domain_id = obs_domain_id;
5886 cookie->flow_sample.obs_point_id = obs_point_id;
5890 compose_ipfix_cookie(union user_action_cookie *cookie)
5892 cookie->type = USER_ACTION_COOKIE_IPFIX;
5895 /* Compose SAMPLE action for IPFIX bridge sampling. */
5897 compose_ipfix_action(const struct ofproto_dpif *ofproto,
5898 struct ofpbuf *odp_actions,
5899 const struct flow *flow)
5901 uint32_t probability;
5902 union user_action_cookie cookie;
5904 if (!ofproto->ipfix || flow->in_port == OFPP_NONE) {
5908 probability = dpif_ipfix_get_bridge_exporter_probability(ofproto->ipfix);
5909 compose_ipfix_cookie(&cookie);
5911 compose_sample_action(ofproto, odp_actions, flow, probability,
5912 &cookie, sizeof cookie.ipfix);
5915 /* SAMPLE action for sFlow must be first action in any given list of
5916 * actions. At this point we do not have all information required to
5917 * build it. So try to build sample action as complete as possible. */
5919 add_sflow_action(struct xlate_ctx *ctx)
5921 ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto,
5922 &ctx->xout->odp_actions,
5923 &ctx->xin->flow, OVSP_NONE);
5924 ctx->sflow_odp_port = 0;
5925 ctx->sflow_n_outputs = 0;
5928 /* SAMPLE action for IPFIX must be 1st or 2nd action in any given list
5929 * of actions, eventually after the SAMPLE action for sFlow. */
5931 add_ipfix_action(struct xlate_ctx *ctx)
5933 compose_ipfix_action(ctx->ofproto, &ctx->xout->odp_actions,
5937 /* Fix SAMPLE action according to data collected while composing ODP actions.
5938 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
5939 * USERSPACE action's user-cookie which is required for sflow. */
5941 fix_sflow_action(struct xlate_ctx *ctx)
5943 const struct flow *base = &ctx->base_flow;
5944 union user_action_cookie *cookie;
5946 if (!ctx->user_cookie_offset) {
5950 cookie = ofpbuf_at(&ctx->xout->odp_actions, ctx->user_cookie_offset,
5951 sizeof cookie->sflow);
5952 ovs_assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
5954 compose_sflow_cookie(ctx->ofproto, base->vlan_tci,
5955 ctx->sflow_odp_port, ctx->sflow_n_outputs, cookie);
5959 compose_output_action__(struct xlate_ctx *ctx, uint16_t ofp_port,
5962 const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
5963 ovs_be16 flow_vlan_tci;
5964 uint32_t flow_skb_mark;
5965 uint8_t flow_nw_tos;
5966 struct priority_to_dscp *pdscp;
5967 uint32_t out_port, odp_port;
5969 /* If 'struct flow' gets additional metadata, we'll need to zero it out
5970 * before traversing a patch port. */
5971 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 20);
5974 xlate_report(ctx, "Nonexistent output port");
5976 } else if (ofport->up.pp.config & OFPUTIL_PC_NO_FWD) {
5977 xlate_report(ctx, "OFPPC_NO_FWD set, skipping output");
5979 } else if (check_stp && !stp_forward_in_state(ofport->stp_state)) {
5980 xlate_report(ctx, "STP not in forwarding state, skipping output");
5984 if (netdev_vport_is_patch(ofport->up.netdev)) {
5985 struct ofport_dpif *peer = ofport_get_peer(ofport);
5986 struct flow old_flow = ctx->xin->flow;
5987 const struct ofproto_dpif *peer_ofproto;
5988 enum slow_path_reason special;
5989 struct ofport_dpif *in_port;
5992 xlate_report(ctx, "Nonexistent patch port peer");
5996 peer_ofproto = ofproto_dpif_cast(peer->up.ofproto);
5997 if (peer_ofproto->backer != ctx->ofproto->backer) {
5998 xlate_report(ctx, "Patch port peer on a different datapath");
6002 ctx->ofproto = ofproto_dpif_cast(peer->up.ofproto);
6003 ctx->xin->flow.in_port = peer->up.ofp_port;
6004 ctx->xin->flow.metadata = htonll(0);
6005 memset(&ctx->xin->flow.tunnel, 0, sizeof ctx->xin->flow.tunnel);
6006 memset(ctx->xin->flow.regs, 0, sizeof ctx->xin->flow.regs);
6008 in_port = get_ofp_port(ctx->ofproto, ctx->xin->flow.in_port);
6009 special = process_special(ctx->ofproto, &ctx->xin->flow, in_port,
6012 ctx->xout->slow = special;
6013 } else if (!in_port || may_receive(in_port, ctx)) {
6014 if (!in_port || stp_forward_in_state(in_port->stp_state)) {
6015 xlate_table_action(ctx, ctx->xin->flow.in_port, 0, true);
6017 /* Forwarding is disabled by STP. Let OFPP_NORMAL and the
6018 * learning action look at the packet, then drop it. */
6019 struct flow old_base_flow = ctx->base_flow;
6020 size_t old_size = ctx->xout->odp_actions.size;
6021 xlate_table_action(ctx, ctx->xin->flow.in_port, 0, true);
6022 ctx->base_flow = old_base_flow;
6023 ctx->xout->odp_actions.size = old_size;
6027 ctx->xin->flow = old_flow;
6028 ctx->ofproto = ofproto_dpif_cast(ofport->up.ofproto);
6030 if (ctx->xin->resubmit_stats) {
6031 netdev_vport_inc_tx(ofport->up.netdev, ctx->xin->resubmit_stats);
6032 netdev_vport_inc_rx(peer->up.netdev, ctx->xin->resubmit_stats);
6038 flow_vlan_tci = ctx->xin->flow.vlan_tci;
6039 flow_skb_mark = ctx->xin->flow.skb_mark;
6040 flow_nw_tos = ctx->xin->flow.nw_tos;
6042 pdscp = get_priority(ofport, ctx->xin->flow.skb_priority);
6044 ctx->xin->flow.nw_tos &= ~IP_DSCP_MASK;
6045 ctx->xin->flow.nw_tos |= pdscp->dscp;
6048 if (ofport->tnl_port) {
6049 /* Save tunnel metadata so that changes made due to
6050 * the Logical (tunnel) Port are not visible for any further
6051 * matches, while explicit set actions on tunnel metadata are.
6053 struct flow_tnl flow_tnl = ctx->xin->flow.tunnel;
6054 odp_port = tnl_port_send(ofport->tnl_port, &ctx->xin->flow);
6055 if (odp_port == OVSP_NONE) {
6056 xlate_report(ctx, "Tunneling decided against output");
6057 goto out; /* restore flow_nw_tos */
6060 if (ctx->xin->resubmit_stats) {
6061 netdev_vport_inc_tx(ofport->up.netdev, ctx->xin->resubmit_stats);
6063 out_port = odp_port;
6064 commit_odp_tunnel_action(&ctx->xin->flow, &ctx->base_flow,
6065 &ctx->xout->odp_actions);
6066 ctx->xin->flow.tunnel = flow_tnl; /* Restore tunnel metadata */
6068 uint16_t vlandev_port;
6069 odp_port = ofport->odp_port;
6070 vlandev_port = vsp_realdev_to_vlandev(ctx->ofproto, ofp_port,
6071 ctx->xin->flow.vlan_tci);
6072 if (vlandev_port == ofp_port) {
6073 out_port = odp_port;
6075 out_port = ofp_port_to_odp_port(ctx->ofproto, vlandev_port);
6076 ctx->xin->flow.vlan_tci = htons(0);
6078 ctx->xin->flow.skb_mark &= ~IPSEC_MARK;
6080 commit_odp_actions(&ctx->xin->flow, &ctx->base_flow,
6081 &ctx->xout->odp_actions);
6082 nl_msg_put_u32(&ctx->xout->odp_actions, OVS_ACTION_ATTR_OUTPUT, out_port);
6084 ctx->sflow_odp_port = odp_port;
6085 ctx->sflow_n_outputs++;
6086 ctx->xout->nf_output_iface = ofp_port;
6089 ctx->xin->flow.vlan_tci = flow_vlan_tci;
6090 ctx->xin->flow.skb_mark = flow_skb_mark;
6092 ctx->xin->flow.nw_tos = flow_nw_tos;
6096 compose_output_action(struct xlate_ctx *ctx, uint16_t ofp_port)
6098 compose_output_action__(ctx, ofp_port, true);
6102 tag_the_flow(struct xlate_ctx *ctx, struct rule_dpif *rule)
6104 struct ofproto_dpif *ofproto = ctx->ofproto;
6105 uint8_t table_id = ctx->table_id;
6107 if (table_id > 0 && table_id < N_TABLES) {
6108 struct table_dpif *table = &ofproto->tables[table_id];
6109 if (table->other_table) {
6110 ctx->xout->tags |= (rule && rule->tag
6112 : rule_calculate_tag(&ctx->xin->flow,
6113 &table->other_table->mask,
6119 /* Common rule processing in one place to avoid duplicating code. */
6120 static struct rule_dpif *
6121 ctx_rule_hooks(struct xlate_ctx *ctx, struct rule_dpif *rule,
6124 if (ctx->xin->resubmit_hook) {
6125 ctx->xin->resubmit_hook(ctx, rule);
6127 if (rule == NULL && may_packet_in) {
6129 * check if table configuration flags
6130 * OFPTC_TABLE_MISS_CONTROLLER, default.
6131 * OFPTC_TABLE_MISS_CONTINUE,
6132 * OFPTC_TABLE_MISS_DROP
6133 * When OF1.0, OFPTC_TABLE_MISS_CONTINUE is used. What to do?
6135 rule = rule_dpif_miss_rule(ctx->ofproto, &ctx->xin->flow);
6137 if (rule && ctx->xin->resubmit_stats) {
6138 rule_credit_stats(rule, ctx->xin->resubmit_stats);
6144 xlate_table_action(struct xlate_ctx *ctx,
6145 uint16_t in_port, uint8_t table_id, bool may_packet_in)
6147 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
6148 struct rule_dpif *rule;
6149 uint16_t old_in_port = ctx->xin->flow.in_port;
6150 uint8_t old_table_id = ctx->table_id;
6152 ctx->table_id = table_id;
6154 /* Look up a flow with 'in_port' as the input port. */
6155 ctx->xin->flow.in_port = in_port;
6156 rule = rule_dpif_lookup__(ctx->ofproto, &ctx->xin->flow, table_id);
6158 tag_the_flow(ctx, rule);
6160 /* Restore the original input port. Otherwise OFPP_NORMAL and
6161 * OFPP_IN_PORT will have surprising behavior. */
6162 ctx->xin->flow.in_port = old_in_port;
6164 rule = ctx_rule_hooks(ctx, rule, may_packet_in);
6167 struct rule_dpif *old_rule = ctx->rule;
6171 do_xlate_actions(rule->up.ofpacts, rule->up.ofpacts_len, ctx);
6172 ctx->rule = old_rule;
6176 ctx->table_id = old_table_id;
6178 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
6180 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
6181 MAX_RESUBMIT_RECURSION);
6182 ctx->max_resubmit_trigger = true;
6187 xlate_ofpact_resubmit(struct xlate_ctx *ctx,
6188 const struct ofpact_resubmit *resubmit)
6193 in_port = resubmit->in_port;
6194 if (in_port == OFPP_IN_PORT) {
6195 in_port = ctx->xin->flow.in_port;
6198 table_id = resubmit->table_id;
6199 if (table_id == 255) {
6200 table_id = ctx->table_id;
6203 xlate_table_action(ctx, in_port, table_id, false);
6207 flood_packets(struct xlate_ctx *ctx, bool all)
6209 struct ofport_dpif *ofport;
6211 HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) {
6212 uint16_t ofp_port = ofport->up.ofp_port;
6214 if (ofp_port == ctx->xin->flow.in_port) {
6219 compose_output_action__(ctx, ofp_port, false);
6220 } else if (!(ofport->up.pp.config & OFPUTIL_PC_NO_FLOOD)) {
6221 compose_output_action(ctx, ofp_port);
6225 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
6229 execute_controller_action(struct xlate_ctx *ctx, int len,
6230 enum ofp_packet_in_reason reason,
6231 uint16_t controller_id)
6233 struct ofputil_packet_in pin;
6234 struct ofpbuf *packet;
6236 ovs_assert(!ctx->xout->slow || ctx->xout->slow == SLOW_CONTROLLER);
6237 ctx->xout->slow = SLOW_CONTROLLER;
6238 if (!ctx->xin->packet) {
6242 packet = ofpbuf_clone(ctx->xin->packet);
6244 if (packet->l2 && packet->l3) {
6245 struct eth_header *eh;
6246 uint16_t mpls_depth;
6248 eth_pop_vlan(packet);
6251 memcpy(eh->eth_src, ctx->xin->flow.dl_src, sizeof eh->eth_src);
6252 memcpy(eh->eth_dst, ctx->xin->flow.dl_dst, sizeof eh->eth_dst);
6254 if (ctx->xin->flow.vlan_tci & htons(VLAN_CFI)) {
6255 eth_push_vlan(packet, ctx->xin->flow.vlan_tci);
6258 mpls_depth = eth_mpls_depth(packet);
6260 if (mpls_depth < ctx->xin->flow.mpls_depth) {
6261 push_mpls(packet, ctx->xin->flow.dl_type, ctx->xin->flow.mpls_lse);
6262 } else if (mpls_depth > ctx->xin->flow.mpls_depth) {
6263 pop_mpls(packet, ctx->xin->flow.dl_type);
6264 } else if (mpls_depth) {
6265 set_mpls_lse(packet, ctx->xin->flow.mpls_lse);
6269 if (ctx->xin->flow.dl_type == htons(ETH_TYPE_IP)) {
6270 packet_set_ipv4(packet, ctx->xin->flow.nw_src,
6271 ctx->xin->flow.nw_dst, ctx->xin->flow.nw_tos,
6272 ctx->xin->flow.nw_ttl);
6276 if (ctx->xin->flow.nw_proto == IPPROTO_TCP) {
6277 packet_set_tcp_port(packet, ctx->xin->flow.tp_src,
6278 ctx->xin->flow.tp_dst);
6279 } else if (ctx->xin->flow.nw_proto == IPPROTO_UDP) {
6280 packet_set_udp_port(packet, ctx->xin->flow.tp_src,
6281 ctx->xin->flow.tp_dst);
6287 pin.packet = packet->data;
6288 pin.packet_len = packet->size;
6289 pin.reason = reason;
6290 pin.controller_id = controller_id;
6291 pin.table_id = ctx->table_id;
6292 pin.cookie = ctx->rule ? ctx->rule->up.flow_cookie : 0;
6295 flow_get_metadata(&ctx->xin->flow, &pin.fmd);
6297 connmgr_send_packet_in(ctx->ofproto->up.connmgr, &pin);
6298 ofpbuf_delete(packet);
6302 execute_mpls_push_action(struct xlate_ctx *ctx, ovs_be16 eth_type)
6304 ovs_assert(eth_type_mpls(eth_type));
6306 if (ctx->base_flow.mpls_depth) {
6307 ctx->xin->flow.mpls_lse &= ~htonl(MPLS_BOS_MASK);
6308 ctx->xin->flow.mpls_depth++;
6313 if (ctx->xin->flow.dl_type == htons(ETH_TYPE_IPV6)) {
6314 label = htonl(0x2); /* IPV6 Explicit Null. */
6316 label = htonl(0x0); /* IPV4 Explicit Null. */
6318 tc = (ctx->xin->flow.nw_tos & IP_DSCP_MASK) >> 2;
6319 ttl = ctx->xin->flow.nw_ttl ? ctx->xin->flow.nw_ttl : 0x40;
6320 ctx->xin->flow.mpls_lse = set_mpls_lse_values(ttl, tc, 1, label);
6321 ctx->xin->flow.mpls_depth = 1;
6323 ctx->xin->flow.dl_type = eth_type;
6327 execute_mpls_pop_action(struct xlate_ctx *ctx, ovs_be16 eth_type)
6329 ovs_assert(eth_type_mpls(ctx->xin->flow.dl_type));
6330 ovs_assert(!eth_type_mpls(eth_type));
6332 if (ctx->xin->flow.mpls_depth) {
6333 ctx->xin->flow.mpls_depth--;
6334 ctx->xin->flow.mpls_lse = htonl(0);
6335 if (!ctx->xin->flow.mpls_depth) {
6336 ctx->xin->flow.dl_type = eth_type;
6342 compose_dec_ttl(struct xlate_ctx *ctx, struct ofpact_cnt_ids *ids)
6344 if (ctx->xin->flow.dl_type != htons(ETH_TYPE_IP) &&
6345 ctx->xin->flow.dl_type != htons(ETH_TYPE_IPV6)) {
6349 if (ctx->xin->flow.nw_ttl > 1) {
6350 ctx->xin->flow.nw_ttl--;
6355 for (i = 0; i < ids->n_controllers; i++) {
6356 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL,
6360 /* Stop processing for current table. */
6366 execute_set_mpls_ttl_action(struct xlate_ctx *ctx, uint8_t ttl)
6368 if (!eth_type_mpls(ctx->xin->flow.dl_type)) {
6372 set_mpls_lse_ttl(&ctx->xin->flow.mpls_lse, ttl);
6377 execute_dec_mpls_ttl_action(struct xlate_ctx *ctx)
6379 uint8_t ttl = mpls_lse_to_ttl(ctx->xin->flow.mpls_lse);
6381 if (!eth_type_mpls(ctx->xin->flow.dl_type)) {
6387 set_mpls_lse_ttl(&ctx->xin->flow.mpls_lse, ttl);
6390 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL, 0);
6392 /* Stop processing for current table. */
6398 xlate_output_action(struct xlate_ctx *ctx,
6399 uint16_t port, uint16_t max_len, bool may_packet_in)
6401 uint16_t prev_nf_output_iface = ctx->xout->nf_output_iface;
6403 ctx->xout->nf_output_iface = NF_OUT_DROP;
6407 compose_output_action(ctx, ctx->xin->flow.in_port);
6410 xlate_table_action(ctx, ctx->xin->flow.in_port, 0, may_packet_in);
6416 flood_packets(ctx, false);
6419 flood_packets(ctx, true);
6421 case OFPP_CONTROLLER:
6422 execute_controller_action(ctx, max_len, OFPR_ACTION, 0);
6428 if (port != ctx->xin->flow.in_port) {
6429 compose_output_action(ctx, port);
6431 xlate_report(ctx, "skipping output to input port");
6436 if (prev_nf_output_iface == NF_OUT_FLOOD) {
6437 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
6438 } else if (ctx->xout->nf_output_iface == NF_OUT_DROP) {
6439 ctx->xout->nf_output_iface = prev_nf_output_iface;
6440 } else if (prev_nf_output_iface != NF_OUT_DROP &&
6441 ctx->xout->nf_output_iface != NF_OUT_FLOOD) {
6442 ctx->xout->nf_output_iface = NF_OUT_MULTI;
6447 xlate_output_reg_action(struct xlate_ctx *ctx,
6448 const struct ofpact_output_reg *or)
6450 uint64_t port = mf_get_subfield(&or->src, &ctx->xin->flow);
6451 if (port <= UINT16_MAX) {
6452 xlate_output_action(ctx, port, or->max_len, false);
6457 xlate_enqueue_action(struct xlate_ctx *ctx,
6458 const struct ofpact_enqueue *enqueue)
6460 uint16_t ofp_port = enqueue->port;
6461 uint32_t queue_id = enqueue->queue;
6462 uint32_t flow_priority, priority;
6465 /* Translate queue to priority. */
6466 error = dpif_queue_to_priority(ctx->ofproto->backer->dpif,
6467 queue_id, &priority);
6469 /* Fall back to ordinary output action. */
6470 xlate_output_action(ctx, enqueue->port, 0, false);
6474 /* Check output port. */
6475 if (ofp_port == OFPP_IN_PORT) {
6476 ofp_port = ctx->xin->flow.in_port;
6477 } else if (ofp_port == ctx->xin->flow.in_port) {
6481 /* Add datapath actions. */
6482 flow_priority = ctx->xin->flow.skb_priority;
6483 ctx->xin->flow.skb_priority = priority;
6484 compose_output_action(ctx, ofp_port);
6485 ctx->xin->flow.skb_priority = flow_priority;
6487 /* Update NetFlow output port. */
6488 if (ctx->xout->nf_output_iface == NF_OUT_DROP) {
6489 ctx->xout->nf_output_iface = ofp_port;
6490 } else if (ctx->xout->nf_output_iface != NF_OUT_FLOOD) {
6491 ctx->xout->nf_output_iface = NF_OUT_MULTI;
6496 xlate_set_queue_action(struct xlate_ctx *ctx, uint32_t queue_id)
6498 uint32_t skb_priority;
6500 if (!dpif_queue_to_priority(ctx->ofproto->backer->dpif,
6501 queue_id, &skb_priority)) {
6502 ctx->xin->flow.skb_priority = skb_priority;
6504 /* Couldn't translate queue to a priority. Nothing to do. A warning
6505 * has already been logged. */
6510 slave_enabled_cb(uint16_t ofp_port, void *ofproto_)
6512 struct ofproto_dpif *ofproto = ofproto_;
6513 struct ofport_dpif *port;
6523 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
6526 port = get_ofp_port(ofproto, ofp_port);
6527 return port ? port->may_enable : false;
6532 xlate_bundle_action(struct xlate_ctx *ctx,
6533 const struct ofpact_bundle *bundle)
6537 port = bundle_execute(bundle, &ctx->xin->flow, slave_enabled_cb,
6539 if (bundle->dst.field) {
6540 nxm_reg_load(&bundle->dst, port, &ctx->xin->flow);
6542 xlate_output_action(ctx, port, 0, false);
6547 xlate_learn_action(struct xlate_ctx *ctx,
6548 const struct ofpact_learn *learn)
6550 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
6551 struct ofputil_flow_mod fm;
6552 uint64_t ofpacts_stub[1024 / 8];
6553 struct ofpbuf ofpacts;
6556 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
6557 learn_execute(learn, &ctx->xin->flow, &fm, &ofpacts);
6559 error = ofproto_flow_mod(&ctx->ofproto->up, &fm);
6560 if (error && !VLOG_DROP_WARN(&rl)) {
6561 VLOG_WARN("learning action failed to modify flow table (%s)",
6562 ofperr_get_name(error));
6565 ofpbuf_uninit(&ofpacts);
6568 /* Reduces '*timeout' to no more than 'max'. A value of zero in either case
6569 * means "infinite". */
6571 reduce_timeout(uint16_t max, uint16_t *timeout)
6573 if (max && (!*timeout || *timeout > max)) {
6579 xlate_fin_timeout(struct xlate_ctx *ctx,
6580 const struct ofpact_fin_timeout *oft)
6582 if (ctx->xin->tcp_flags & (TCP_FIN | TCP_RST) && ctx->rule) {
6583 struct rule_dpif *rule = ctx->rule;
6585 reduce_timeout(oft->fin_idle_timeout, &rule->up.idle_timeout);
6586 reduce_timeout(oft->fin_hard_timeout, &rule->up.hard_timeout);
6591 xlate_sample_action(struct xlate_ctx *ctx,
6592 const struct ofpact_sample *os)
6594 union user_action_cookie cookie;
6595 /* Scale the probability from 16-bit to 32-bit while representing
6596 * the same percentage. */
6597 uint32_t probability = (os->probability << 16) | os->probability;
6599 commit_odp_actions(&ctx->xin->flow, &ctx->base_flow,
6600 &ctx->xout->odp_actions);
6602 compose_flow_sample_cookie(os->probability, os->collector_set_id,
6603 os->obs_domain_id, os->obs_point_id, &cookie);
6604 compose_sample_action(ctx->ofproto, &ctx->xout->odp_actions, &ctx->xin->flow,
6605 probability, &cookie, sizeof cookie.flow_sample);
6609 may_receive(const struct ofport_dpif *port, struct xlate_ctx *ctx)
6611 if (port->up.pp.config & (eth_addr_equals(ctx->xin->flow.dl_dst,
6613 ? OFPUTIL_PC_NO_RECV_STP
6614 : OFPUTIL_PC_NO_RECV)) {
6618 /* Only drop packets here if both forwarding and learning are
6619 * disabled. If just learning is enabled, we need to have
6620 * OFPP_NORMAL and the learning action have a look at the packet
6621 * before we can drop it. */
6622 if (!stp_forward_in_state(port->stp_state)
6623 && !stp_learn_in_state(port->stp_state)) {
6631 tunnel_ecn_ok(struct xlate_ctx *ctx)
6633 if (is_ip_any(&ctx->base_flow)
6634 && (ctx->base_flow.tunnel.ip_tos & IP_ECN_MASK) == IP_ECN_CE) {
6635 if ((ctx->base_flow.nw_tos & IP_ECN_MASK) == IP_ECN_NOT_ECT) {
6636 VLOG_WARN_RL(&rl, "dropping tunnel packet marked ECN CE"
6637 " but is not ECN capable");
6640 /* Set the ECN CE value in the tunneled packet. */
6641 ctx->xin->flow.nw_tos |= IP_ECN_CE;
6649 do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len,
6650 struct xlate_ctx *ctx)
6652 bool was_evictable = true;
6653 const struct ofpact *a;
6656 /* Don't let the rule we're working on get evicted underneath us. */
6657 was_evictable = ctx->rule->up.evictable;
6658 ctx->rule->up.evictable = false;
6661 do_xlate_actions_again:
6662 OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) {
6663 struct ofpact_controller *controller;
6664 const struct ofpact_metadata *metadata;
6672 xlate_output_action(ctx, ofpact_get_OUTPUT(a)->port,
6673 ofpact_get_OUTPUT(a)->max_len, true);
6676 case OFPACT_CONTROLLER:
6677 controller = ofpact_get_CONTROLLER(a);
6678 execute_controller_action(ctx, controller->max_len,
6680 controller->controller_id);
6683 case OFPACT_ENQUEUE:
6684 xlate_enqueue_action(ctx, ofpact_get_ENQUEUE(a));
6687 case OFPACT_SET_VLAN_VID:
6688 ctx->xin->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
6689 ctx->xin->flow.vlan_tci |=
6690 (htons(ofpact_get_SET_VLAN_VID(a)->vlan_vid)
6694 case OFPACT_SET_VLAN_PCP:
6695 ctx->xin->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
6696 ctx->xin->flow.vlan_tci |=
6697 htons((ofpact_get_SET_VLAN_PCP(a)->vlan_pcp << VLAN_PCP_SHIFT)
6701 case OFPACT_STRIP_VLAN:
6702 ctx->xin->flow.vlan_tci = htons(0);
6705 case OFPACT_PUSH_VLAN:
6706 /* XXX 802.1AD(QinQ) */
6707 ctx->xin->flow.vlan_tci = htons(VLAN_CFI);
6710 case OFPACT_SET_ETH_SRC:
6711 memcpy(ctx->xin->flow.dl_src, ofpact_get_SET_ETH_SRC(a)->mac,
6715 case OFPACT_SET_ETH_DST:
6716 memcpy(ctx->xin->flow.dl_dst, ofpact_get_SET_ETH_DST(a)->mac,
6720 case OFPACT_SET_IPV4_SRC:
6721 if (ctx->xin->flow.dl_type == htons(ETH_TYPE_IP)) {
6722 ctx->xin->flow.nw_src = ofpact_get_SET_IPV4_SRC(a)->ipv4;
6726 case OFPACT_SET_IPV4_DST:
6727 if (ctx->xin->flow.dl_type == htons(ETH_TYPE_IP)) {
6728 ctx->xin->flow.nw_dst = ofpact_get_SET_IPV4_DST(a)->ipv4;
6732 case OFPACT_SET_IPV4_DSCP:
6733 /* OpenFlow 1.0 only supports IPv4. */
6734 if (ctx->xin->flow.dl_type == htons(ETH_TYPE_IP)) {
6735 ctx->xin->flow.nw_tos &= ~IP_DSCP_MASK;
6736 ctx->xin->flow.nw_tos |= ofpact_get_SET_IPV4_DSCP(a)->dscp;
6740 case OFPACT_SET_L4_SRC_PORT:
6741 if (is_ip_any(&ctx->xin->flow)) {
6742 ctx->xin->flow.tp_src =
6743 htons(ofpact_get_SET_L4_SRC_PORT(a)->port);
6747 case OFPACT_SET_L4_DST_PORT:
6748 if (is_ip_any(&ctx->xin->flow)) {
6749 ctx->xin->flow.tp_dst =
6750 htons(ofpact_get_SET_L4_DST_PORT(a)->port);
6754 case OFPACT_RESUBMIT:
6755 xlate_ofpact_resubmit(ctx, ofpact_get_RESUBMIT(a));
6758 case OFPACT_SET_TUNNEL:
6759 ctx->xin->flow.tunnel.tun_id =
6760 htonll(ofpact_get_SET_TUNNEL(a)->tun_id);
6763 case OFPACT_SET_QUEUE:
6764 xlate_set_queue_action(ctx, ofpact_get_SET_QUEUE(a)->queue_id);
6767 case OFPACT_POP_QUEUE:
6768 ctx->xin->flow.skb_priority = ctx->orig_skb_priority;
6771 case OFPACT_REG_MOVE:
6772 nxm_execute_reg_move(ofpact_get_REG_MOVE(a), &ctx->xin->flow);
6775 case OFPACT_REG_LOAD:
6776 nxm_execute_reg_load(ofpact_get_REG_LOAD(a), &ctx->xin->flow);
6779 case OFPACT_STACK_PUSH:
6780 nxm_execute_stack_push(ofpact_get_STACK_PUSH(a), &ctx->xin->flow,
6784 case OFPACT_STACK_POP:
6785 nxm_execute_stack_pop(ofpact_get_STACK_POP(a), &ctx->xin->flow,
6789 case OFPACT_PUSH_MPLS:
6790 execute_mpls_push_action(ctx, ofpact_get_PUSH_MPLS(a)->ethertype);
6793 case OFPACT_POP_MPLS:
6794 execute_mpls_pop_action(ctx, ofpact_get_POP_MPLS(a)->ethertype);
6797 case OFPACT_SET_MPLS_TTL:
6798 if (execute_set_mpls_ttl_action(ctx,
6799 ofpact_get_SET_MPLS_TTL(a)->ttl)) {
6804 case OFPACT_DEC_MPLS_TTL:
6805 if (execute_dec_mpls_ttl_action(ctx)) {
6810 case OFPACT_DEC_TTL:
6811 if (compose_dec_ttl(ctx, ofpact_get_DEC_TTL(a))) {
6817 /* Nothing to do. */
6820 case OFPACT_MULTIPATH:
6821 multipath_execute(ofpact_get_MULTIPATH(a), &ctx->xin->flow);
6825 ctx->ofproto->has_bundle_action = true;
6826 xlate_bundle_action(ctx, ofpact_get_BUNDLE(a));
6829 case OFPACT_OUTPUT_REG:
6830 xlate_output_reg_action(ctx, ofpact_get_OUTPUT_REG(a));
6834 ctx->xout->has_learn = true;
6835 if (ctx->xin->may_learn) {
6836 xlate_learn_action(ctx, ofpact_get_LEARN(a));
6844 case OFPACT_FIN_TIMEOUT:
6845 ctx->xout->has_fin_timeout = true;
6846 xlate_fin_timeout(ctx, ofpact_get_FIN_TIMEOUT(a));
6849 case OFPACT_CLEAR_ACTIONS:
6851 * Nothing to do because writa-actions is not supported for now.
6852 * When writa-actions is supported, clear-actions also must
6853 * be supported at the same time.
6857 case OFPACT_WRITE_METADATA:
6858 metadata = ofpact_get_WRITE_METADATA(a);
6859 ctx->xin->flow.metadata &= ~metadata->mask;
6860 ctx->xin->flow.metadata |= metadata->metadata & metadata->mask;
6863 case OFPACT_GOTO_TABLE: {
6864 /* It is assumed that goto-table is the last action. */
6865 struct ofpact_goto_table *ogt = ofpact_get_GOTO_TABLE(a);
6866 struct rule_dpif *rule;
6868 ovs_assert(ctx->table_id < ogt->table_id);
6870 ctx->table_id = ogt->table_id;
6872 /* Look up a flow from the new table. */
6873 rule = rule_dpif_lookup__(ctx->ofproto, &ctx->xin->flow, ctx->table_id);
6875 tag_the_flow(ctx, rule);
6877 rule = ctx_rule_hooks(ctx, rule, true);
6881 ctx->rule->up.evictable = was_evictable;
6884 was_evictable = rule->up.evictable;
6885 rule->up.evictable = false;
6887 /* Tail recursion removal. */
6888 ofpacts = rule->up.ofpacts;
6889 ofpacts_len = rule->up.ofpacts_len;
6890 goto do_xlate_actions_again;
6896 xlate_sample_action(ctx, ofpact_get_SAMPLE(a));
6903 ctx->rule->up.evictable = was_evictable;
6908 xlate_in_init(struct xlate_in *xin, struct ofproto_dpif *ofproto,
6909 const struct flow *flow,
6910 const struct initial_vals *initial_vals,
6911 struct rule_dpif *rule, uint8_t tcp_flags,
6912 const struct ofpbuf *packet)
6914 xin->ofproto = ofproto;
6916 xin->packet = packet;
6917 xin->may_learn = packet != NULL;
6919 xin->ofpacts = NULL;
6920 xin->ofpacts_len = 0;
6921 xin->tcp_flags = tcp_flags;
6922 xin->resubmit_hook = NULL;
6923 xin->report_hook = NULL;
6924 xin->resubmit_stats = NULL;
6927 xin->initial_vals = *initial_vals;
6929 xin->initial_vals.vlan_tci = xin->flow.vlan_tci;
6930 xin->initial_vals.tunnel_ip_tos = xin->flow.tunnel.ip_tos;
6935 xlate_out_uninit(struct xlate_out *xout)
6938 ofpbuf_uninit(&xout->odp_actions);
6942 /* Translates the 'ofpacts_len' bytes of "struct ofpacts" starting at 'ofpacts'
6943 * into datapath actions in 'odp_actions', using 'ctx'. */
6945 xlate_actions(struct xlate_in *xin, struct xlate_out *xout)
6947 /* Normally false. Set to true if we ever hit MAX_RESUBMIT_RECURSION, so
6948 * that in the future we always keep a copy of the original flow for
6949 * tracing purposes. */
6950 static bool hit_resubmit_limit;
6952 enum slow_path_reason special;
6953 const struct ofpact *ofpacts;
6954 struct ofport_dpif *in_port;
6955 struct flow orig_flow;
6956 struct xlate_ctx ctx;
6959 COVERAGE_INC(ofproto_dpif_xlate);
6961 /* Flow initialization rules:
6962 * - 'base_flow' must match the kernel's view of the packet at the
6963 * time that action processing starts. 'flow' represents any
6964 * transformations we wish to make through actions.
6965 * - By default 'base_flow' and 'flow' are the same since the input
6966 * packet matches the output before any actions are applied.
6967 * - When using VLAN splinters, 'base_flow''s VLAN is set to the value
6968 * of the received packet as seen by the kernel. If we later output
6969 * to another device without any modifications this will cause us to
6970 * insert a new tag since the original one was stripped off by the
6972 * - Tunnel 'flow' is largely cleared when transitioning between
6973 * the input and output stages since it does not make sense to output
6974 * a packet with the exact headers that it was received with (i.e.
6975 * the destination IP is us). The one exception is the tun_id, which
6976 * is preserved to allow use in later resubmit lookups and loads into
6978 * - Tunnel 'base_flow' is completely cleared since that is what the
6979 * kernel does. If we wish to maintain the original values an action
6980 * needs to be generated. */
6985 ctx.ofproto = xin->ofproto;
6986 ctx.rule = xin->rule;
6988 ctx.base_flow = ctx.xin->flow;
6989 memset(&ctx.base_flow.tunnel, 0, sizeof ctx.base_flow.tunnel);
6990 ctx.base_flow.vlan_tci = xin->initial_vals.vlan_tci;
6991 ctx.base_flow.tunnel.ip_tos = xin->initial_vals.tunnel_ip_tos;
6995 ctx.xout->has_learn = false;
6996 ctx.xout->has_normal = false;
6997 ctx.xout->has_fin_timeout = false;
6998 ctx.xout->nf_output_iface = NF_OUT_DROP;
6999 ctx.xout->mirrors = 0;
7001 ofpbuf_use_stub(&ctx.xout->odp_actions, ctx.xout->odp_actions_stub,
7002 sizeof ctx.xout->odp_actions_stub);
7003 ofpbuf_reserve(&ctx.xout->odp_actions, NL_A_U32_SIZE);
7006 ctx.max_resubmit_trigger = false;
7007 ctx.orig_skb_priority = ctx.xin->flow.skb_priority;
7012 ofpacts = xin->ofpacts;
7013 ofpacts_len = xin->ofpacts_len;
7014 } else if (xin->rule) {
7015 ofpacts = xin->rule->up.ofpacts;
7016 ofpacts_len = xin->rule->up.ofpacts_len;
7021 ofpbuf_use_stub(&ctx.stack, ctx.init_stack, sizeof ctx.init_stack);
7023 if (ctx.ofproto->has_mirrors || hit_resubmit_limit) {
7024 /* Do this conditionally because the copy is expensive enough that it
7025 * shows up in profiles. */
7026 orig_flow = ctx.xin->flow;
7029 if (ctx.xin->flow.nw_frag & FLOW_NW_FRAG_ANY) {
7030 switch (ctx.ofproto->up.frag_handling) {
7031 case OFPC_FRAG_NORMAL:
7032 /* We must pretend that transport ports are unavailable. */
7033 ctx.xin->flow.tp_src = ctx.base_flow.tp_src = htons(0);
7034 ctx.xin->flow.tp_dst = ctx.base_flow.tp_dst = htons(0);
7037 case OFPC_FRAG_DROP:
7040 case OFPC_FRAG_REASM:
7043 case OFPC_FRAG_NX_MATCH:
7044 /* Nothing to do. */
7047 case OFPC_INVALID_TTL_TO_CONTROLLER:
7052 in_port = get_ofp_port(ctx.ofproto, ctx.xin->flow.in_port);
7053 special = process_special(ctx.ofproto, &ctx.xin->flow, in_port,
7056 ctx.xout->slow = special;
7058 static struct vlog_rate_limit trace_rl = VLOG_RATE_LIMIT_INIT(1, 1);
7059 struct initial_vals initial_vals;
7060 size_t sample_actions_len;
7061 uint32_t local_odp_port;
7063 initial_vals.vlan_tci = ctx.base_flow.vlan_tci;
7064 initial_vals.tunnel_ip_tos = ctx.base_flow.tunnel.ip_tos;
7066 add_sflow_action(&ctx);
7067 add_ipfix_action(&ctx);
7068 sample_actions_len = ctx.xout->odp_actions.size;
7070 if (tunnel_ecn_ok(&ctx) && (!in_port || may_receive(in_port, &ctx))) {
7071 do_xlate_actions(ofpacts, ofpacts_len, &ctx);
7073 /* We've let OFPP_NORMAL and the learning action look at the
7074 * packet, so drop it now if forwarding is disabled. */
7075 if (in_port && !stp_forward_in_state(in_port->stp_state)) {
7076 ctx.xout->odp_actions.size = sample_actions_len;
7080 if (ctx.max_resubmit_trigger && !ctx.xin->resubmit_hook) {
7081 if (!hit_resubmit_limit) {
7082 /* We didn't record the original flow. Make sure we do from
7084 hit_resubmit_limit = true;
7085 } else if (!VLOG_DROP_ERR(&trace_rl)) {
7086 struct ds ds = DS_EMPTY_INITIALIZER;
7088 ofproto_trace(ctx.ofproto, &orig_flow, ctx.xin->packet,
7089 &initial_vals, &ds);
7090 VLOG_ERR("Trace triggered by excessive resubmit "
7091 "recursion:\n%s", ds_cstr(&ds));
7096 local_odp_port = ofp_port_to_odp_port(ctx.ofproto, OFPP_LOCAL);
7097 if (!connmgr_must_output_local(ctx.ofproto->up.connmgr, &ctx.xin->flow,
7099 ctx.xout->odp_actions.data,
7100 ctx.xout->odp_actions.size)) {
7101 compose_output_action(&ctx, OFPP_LOCAL);
7103 if (ctx.ofproto->has_mirrors) {
7104 add_mirror_actions(&ctx, &orig_flow);
7106 fix_sflow_action(&ctx);
7109 ofpbuf_uninit(&ctx.stack);
7112 /* Translates the 'ofpacts_len' bytes of "struct ofpact"s starting at 'ofpacts'
7113 * into datapath actions, using 'ctx', and discards the datapath actions. */
7115 xlate_actions_for_side_effects(struct xlate_in *xin)
7117 struct xlate_out xout;
7119 xlate_actions(xin, &xout);
7120 xlate_out_uninit(&xout);
7124 xlate_report(struct xlate_ctx *ctx, const char *s)
7126 if (ctx->xin->report_hook) {
7127 ctx->xin->report_hook(ctx, s);
7131 /* OFPP_NORMAL implementation. */
7133 static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *);
7135 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
7136 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle',
7137 * the bundle on which the packet was received, returns the VLAN to which the
7140 * Both 'vid' and the return value are in the range 0...4095. */
7142 input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid)
7144 switch (in_bundle->vlan_mode) {
7145 case PORT_VLAN_ACCESS:
7146 return in_bundle->vlan;
7149 case PORT_VLAN_TRUNK:
7152 case PORT_VLAN_NATIVE_UNTAGGED:
7153 case PORT_VLAN_NATIVE_TAGGED:
7154 return vid ? vid : in_bundle->vlan;
7161 /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'.
7162 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
7165 * 'vid' should be the VID obtained from the 802.1Q header that was received as
7166 * part of a packet (specify 0 if there was no 802.1Q header), in the range
7169 input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn)
7171 /* Allow any VID on the OFPP_NONE port. */
7172 if (in_bundle == &ofpp_none_bundle) {
7176 switch (in_bundle->vlan_mode) {
7177 case PORT_VLAN_ACCESS:
7180 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7181 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
7182 "packet received on port %s configured as VLAN "
7183 "%"PRIu16" access port",
7184 in_bundle->ofproto->up.name, vid,
7185 in_bundle->name, in_bundle->vlan);
7191 case PORT_VLAN_NATIVE_UNTAGGED:
7192 case PORT_VLAN_NATIVE_TAGGED:
7194 /* Port must always carry its native VLAN. */
7198 case PORT_VLAN_TRUNK:
7199 if (!ofbundle_includes_vlan(in_bundle, vid)) {
7201 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7202 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet "
7203 "received on port %s not configured for trunking "
7205 in_bundle->ofproto->up.name, vid,
7206 in_bundle->name, vid);
7218 /* Given 'vlan', the VLAN that a packet belongs to, and
7219 * 'out_bundle', a bundle on which the packet is to be output, returns the VID
7220 * that should be included in the 802.1Q header. (If the return value is 0,
7221 * then the 802.1Q header should only be included in the packet if there is a
7224 * Both 'vlan' and the return value are in the range 0...4095. */
7226 output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan)
7228 switch (out_bundle->vlan_mode) {
7229 case PORT_VLAN_ACCESS:
7232 case PORT_VLAN_TRUNK:
7233 case PORT_VLAN_NATIVE_TAGGED:
7236 case PORT_VLAN_NATIVE_UNTAGGED:
7237 return vlan == out_bundle->vlan ? 0 : vlan;
7245 output_normal(struct xlate_ctx *ctx, const struct ofbundle *out_bundle,
7248 struct ofport_dpif *port;
7250 ovs_be16 tci, old_tci;
7252 vid = output_vlan_to_vid(out_bundle, vlan);
7253 if (!out_bundle->bond) {
7254 port = ofbundle_get_a_port(out_bundle);
7256 port = bond_choose_output_slave(out_bundle->bond, &ctx->xin->flow,
7257 vid, &ctx->xout->tags);
7259 /* No slaves enabled, so drop packet. */
7264 old_tci = ctx->xin->flow.vlan_tci;
7266 if (tci || out_bundle->use_priority_tags) {
7267 tci |= ctx->xin->flow.vlan_tci & htons(VLAN_PCP_MASK);
7269 tci |= htons(VLAN_CFI);
7272 ctx->xin->flow.vlan_tci = tci;
7274 compose_output_action(ctx, port->up.ofp_port);
7275 ctx->xin->flow.vlan_tci = old_tci;
7279 mirror_mask_ffs(mirror_mask_t mask)
7281 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
7286 ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
7288 return (bundle->vlan_mode != PORT_VLAN_ACCESS
7289 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
7293 ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan)
7295 return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan);
7298 /* Returns an arbitrary interface within 'bundle'. */
7299 static struct ofport_dpif *
7300 ofbundle_get_a_port(const struct ofbundle *bundle)
7302 return CONTAINER_OF(list_front(&bundle->ports),
7303 struct ofport_dpif, bundle_node);
7307 vlan_is_mirrored(const struct ofmirror *m, int vlan)
7309 return !m->vlans || bitmap_is_set(m->vlans, vlan);
7313 add_mirror_actions(struct xlate_ctx *ctx, const struct flow *orig_flow)
7315 struct ofproto_dpif *ofproto = ctx->ofproto;
7316 mirror_mask_t mirrors;
7317 struct ofbundle *in_bundle;
7320 const struct nlattr *a;
7323 in_bundle = lookup_input_bundle(ctx->ofproto, orig_flow->in_port,
7324 ctx->xin->packet != NULL, NULL);
7328 mirrors = in_bundle->src_mirrors;
7330 /* Drop frames on bundles reserved for mirroring. */
7331 if (in_bundle->mirror_out) {
7332 if (ctx->xin->packet != NULL) {
7333 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7334 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
7335 "%s, which is reserved exclusively for mirroring",
7336 ctx->ofproto->up.name, in_bundle->name);
7342 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
7343 if (!input_vid_is_valid(vid, in_bundle, ctx->xin->packet != NULL)) {
7346 vlan = input_vid_to_vlan(in_bundle, vid);
7348 /* Look at the output ports to check for destination selections. */
7350 NL_ATTR_FOR_EACH (a, left, ctx->xout->odp_actions.data,
7351 ctx->xout->odp_actions.size) {
7352 enum ovs_action_attr type = nl_attr_type(a);
7353 struct ofport_dpif *ofport;
7355 if (type != OVS_ACTION_ATTR_OUTPUT) {
7359 ofport = get_odp_port(ofproto, nl_attr_get_u32(a));
7360 if (ofport && ofport->bundle) {
7361 mirrors |= ofport->bundle->dst_mirrors;
7369 /* Restore the original packet before adding the mirror actions. */
7370 ctx->xin->flow = *orig_flow;
7375 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
7377 if (!vlan_is_mirrored(m, vlan)) {
7378 mirrors = zero_rightmost_1bit(mirrors);
7382 mirrors &= ~m->dup_mirrors;
7383 ctx->xout->mirrors |= m->dup_mirrors;
7385 output_normal(ctx, m->out, vlan);
7386 } else if (vlan != m->out_vlan
7387 && !eth_addr_is_reserved(orig_flow->dl_dst)) {
7388 struct ofbundle *bundle;
7390 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
7391 if (ofbundle_includes_vlan(bundle, m->out_vlan)
7392 && !bundle->mirror_out) {
7393 output_normal(ctx, bundle, m->out_vlan);
7401 update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors,
7402 uint64_t packets, uint64_t bytes)
7408 for (; mirrors; mirrors = zero_rightmost_1bit(mirrors)) {
7411 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
7414 /* In normal circumstances 'm' will not be NULL. However,
7415 * if mirrors are reconfigured, we can temporarily get out
7416 * of sync in facet_revalidate(). We could "correct" the
7417 * mirror list before reaching here, but doing that would
7418 * not properly account the traffic stats we've currently
7419 * accumulated for previous mirror configuration. */
7423 m->packet_count += packets;
7424 m->byte_count += bytes;
7428 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
7429 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
7430 * indicate this; newer upstream kernels use gratuitous ARP requests. */
7432 is_gratuitous_arp(const struct flow *flow)
7434 return (flow->dl_type == htons(ETH_TYPE_ARP)
7435 && eth_addr_is_broadcast(flow->dl_dst)
7436 && (flow->nw_proto == ARP_OP_REPLY
7437 || (flow->nw_proto == ARP_OP_REQUEST
7438 && flow->nw_src == flow->nw_dst)));
7442 update_learning_table(struct ofproto_dpif *ofproto,
7443 const struct flow *flow, int vlan,
7444 struct ofbundle *in_bundle)
7446 struct mac_entry *mac;
7448 /* Don't learn the OFPP_NONE port. */
7449 if (in_bundle == &ofpp_none_bundle) {
7453 if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
7457 mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan);
7458 if (is_gratuitous_arp(flow)) {
7459 /* We don't want to learn from gratuitous ARP packets that are
7460 * reflected back over bond slaves so we lock the learning table. */
7461 if (!in_bundle->bond) {
7462 mac_entry_set_grat_arp_lock(mac);
7463 } else if (mac_entry_is_grat_arp_locked(mac)) {
7468 if (mac_entry_is_new(mac) || mac->port.p != in_bundle) {
7469 /* The log messages here could actually be useful in debugging,
7470 * so keep the rate limit relatively high. */
7471 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
7472 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
7473 "on port %s in VLAN %d",
7474 ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src),
7475 in_bundle->name, vlan);
7477 mac->port.p = in_bundle;
7478 tag_set_add(&ofproto->backer->revalidate_set,
7479 mac_learning_changed(ofproto->ml, mac));
7483 static struct ofbundle *
7484 lookup_input_bundle(const struct ofproto_dpif *ofproto, uint16_t in_port,
7485 bool warn, struct ofport_dpif **in_ofportp)
7487 struct ofport_dpif *ofport;
7489 /* Find the port and bundle for the received packet. */
7490 ofport = get_ofp_port(ofproto, in_port);
7492 *in_ofportp = ofport;
7494 if (ofport && ofport->bundle) {
7495 return ofport->bundle;
7498 /* Special-case OFPP_NONE, which a controller may use as the ingress
7499 * port for traffic that it is sourcing. */
7500 if (in_port == OFPP_NONE) {
7501 return &ofpp_none_bundle;
7504 /* Odd. A few possible reasons here:
7506 * - We deleted a port but there are still a few packets queued up
7509 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
7510 * we don't know about.
7512 * - The ofproto client didn't configure the port as part of a bundle.
7513 * This is particularly likely to happen if a packet was received on the
7514 * port after it was created, but before the client had a chance to
7515 * configure its bundle.
7518 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7520 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
7521 "port %"PRIu16, ofproto->up.name, in_port);
7526 /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or
7527 * dropped. Returns true if they may be forwarded, false if they should be
7530 * 'in_port' must be the ofport_dpif that corresponds to flow->in_port.
7531 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
7533 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
7534 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
7535 * checked by input_vid_is_valid().
7537 * May also add tags to '*tags', although the current implementation only does
7538 * so in one special case.
7541 is_admissible(struct xlate_ctx *ctx, struct ofport_dpif *in_port,
7544 struct ofproto_dpif *ofproto = ctx->ofproto;
7545 struct flow *flow = &ctx->xin->flow;
7546 struct ofbundle *in_bundle = in_port->bundle;
7548 /* Drop frames for reserved multicast addresses
7549 * only if forward_bpdu option is absent. */
7550 if (!ofproto->up.forward_bpdu && eth_addr_is_reserved(flow->dl_dst)) {
7551 xlate_report(ctx, "packet has reserved destination MAC, dropping");
7555 if (in_bundle->bond) {
7556 struct mac_entry *mac;
7558 switch (bond_check_admissibility(in_bundle->bond, in_port,
7559 flow->dl_dst, &ctx->xout->tags)) {
7564 xlate_report(ctx, "bonding refused admissibility, dropping");
7567 case BV_DROP_IF_MOVED:
7568 mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL);
7569 if (mac && mac->port.p != in_bundle &&
7570 (!is_gratuitous_arp(flow)
7571 || mac_entry_is_grat_arp_locked(mac))) {
7572 xlate_report(ctx, "SLB bond thinks this packet looped back, "
7584 xlate_normal(struct xlate_ctx *ctx)
7586 struct ofport_dpif *in_port;
7587 struct ofbundle *in_bundle;
7588 struct mac_entry *mac;
7592 ctx->xout->has_normal = true;
7594 in_bundle = lookup_input_bundle(ctx->ofproto, ctx->xin->flow.in_port,
7595 ctx->xin->packet != NULL, &in_port);
7597 xlate_report(ctx, "no input bundle, dropping");
7601 /* Drop malformed frames. */
7602 if (ctx->xin->flow.dl_type == htons(ETH_TYPE_VLAN) &&
7603 !(ctx->xin->flow.vlan_tci & htons(VLAN_CFI))) {
7604 if (ctx->xin->packet != NULL) {
7605 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7606 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
7607 "VLAN tag received on port %s",
7608 ctx->ofproto->up.name, in_bundle->name);
7610 xlate_report(ctx, "partial VLAN tag, dropping");
7614 /* Drop frames on bundles reserved for mirroring. */
7615 if (in_bundle->mirror_out) {
7616 if (ctx->xin->packet != NULL) {
7617 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7618 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
7619 "%s, which is reserved exclusively for mirroring",
7620 ctx->ofproto->up.name, in_bundle->name);
7622 xlate_report(ctx, "input port is mirror output port, dropping");
7627 vid = vlan_tci_to_vid(ctx->xin->flow.vlan_tci);
7628 if (!input_vid_is_valid(vid, in_bundle, ctx->xin->packet != NULL)) {
7629 xlate_report(ctx, "disallowed VLAN VID for this input port, dropping");
7632 vlan = input_vid_to_vlan(in_bundle, vid);
7634 /* Check other admissibility requirements. */
7635 if (in_port && !is_admissible(ctx, in_port, vlan)) {
7639 /* Learn source MAC. */
7640 if (ctx->xin->may_learn) {
7641 update_learning_table(ctx->ofproto, &ctx->xin->flow, vlan, in_bundle);
7644 /* Determine output bundle. */
7645 mac = mac_learning_lookup(ctx->ofproto->ml, ctx->xin->flow.dl_dst, vlan,
7648 if (mac->port.p != in_bundle) {
7649 xlate_report(ctx, "forwarding to learned port");
7650 output_normal(ctx, mac->port.p, vlan);
7652 xlate_report(ctx, "learned port is input port, dropping");
7655 struct ofbundle *bundle;
7657 xlate_report(ctx, "no learned MAC for destination, flooding");
7658 HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
7659 if (bundle != in_bundle
7660 && ofbundle_includes_vlan(bundle, vlan)
7661 && bundle->floodable
7662 && !bundle->mirror_out) {
7663 output_normal(ctx, bundle, vlan);
7666 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
7670 /* Optimized flow revalidation.
7672 * It's a difficult problem, in general, to tell which facets need to have
7673 * their actions recalculated whenever the OpenFlow flow table changes. We
7674 * don't try to solve that general problem: for most kinds of OpenFlow flow
7675 * table changes, we recalculate the actions for every facet. This is
7676 * relatively expensive, but it's good enough if the OpenFlow flow table
7677 * doesn't change very often.
7679 * However, we can expect one particular kind of OpenFlow flow table change to
7680 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
7681 * of CPU on revalidating every facet whenever MAC learning modifies the flow
7682 * table, we add a special case that applies to flow tables in which every rule
7683 * has the same form (that is, the same wildcards), except that the table is
7684 * also allowed to have a single "catch-all" flow that matches all packets. We
7685 * optimize this case by tagging all of the facets that resubmit into the table
7686 * and invalidating the same tag whenever a flow changes in that table. The
7687 * end result is that we revalidate just the facets that need it (and sometimes
7688 * a few more, but not all of the facets or even all of the facets that
7689 * resubmit to the table modified by MAC learning). */
7691 /* Calculates the tag to use for 'flow' and mask 'mask' when it is inserted
7692 * into an OpenFlow table with the given 'basis'. */
7694 rule_calculate_tag(const struct flow *flow, const struct minimask *mask,
7697 if (minimask_is_catchall(mask)) {
7700 uint32_t hash = flow_hash_in_minimask(flow, mask, secret);
7701 return tag_create_deterministic(hash);
7705 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
7706 * taggability of that table.
7708 * This function must be called after *each* change to a flow table. If you
7709 * skip calling it on some changes then the pointer comparisons at the end can
7710 * be invalid if you get unlucky. For example, if a flow removal causes a
7711 * cls_table to be destroyed and then a flow insertion causes a cls_table with
7712 * different wildcards to be created with the same address, then this function
7713 * will incorrectly skip revalidation. */
7715 table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
7717 struct table_dpif *table = &ofproto->tables[table_id];
7718 const struct oftable *oftable = &ofproto->up.tables[table_id];
7719 struct cls_table *catchall, *other;
7720 struct cls_table *t;
7722 catchall = other = NULL;
7724 switch (hmap_count(&oftable->cls.tables)) {
7726 /* We could tag this OpenFlow table but it would make the logic a
7727 * little harder and it's a corner case that doesn't seem worth it
7733 HMAP_FOR_EACH (t, hmap_node, &oftable->cls.tables) {
7734 if (cls_table_is_catchall(t)) {
7736 } else if (!other) {
7739 /* Indicate that we can't tag this by setting both tables to
7740 * NULL. (We know that 'catchall' is already NULL.) */
7747 /* Can't tag this table. */
7751 if (table->catchall_table != catchall || table->other_table != other) {
7752 table->catchall_table = catchall;
7753 table->other_table = other;
7754 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
7758 /* Given 'rule' that has changed in some way (either it is a rule being
7759 * inserted, a rule being deleted, or a rule whose actions are being
7760 * modified), marks facets for revalidation to ensure that packets will be
7761 * forwarded correctly according to the new state of the flow table.
7763 * This function must be called after *each* change to a flow table. See
7764 * the comment on table_update_taggable() for more information. */
7766 rule_invalidate(const struct rule_dpif *rule)
7768 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
7770 table_update_taggable(ofproto, rule->up.table_id);
7772 if (!ofproto->backer->need_revalidate) {
7773 struct table_dpif *table = &ofproto->tables[rule->up.table_id];
7775 if (table->other_table && rule->tag) {
7776 tag_set_add(&ofproto->backer->revalidate_set, rule->tag);
7778 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
7784 set_frag_handling(struct ofproto *ofproto_,
7785 enum ofp_config_flags frag_handling)
7787 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7788 if (frag_handling != OFPC_FRAG_REASM) {
7789 ofproto->backer->need_revalidate = REV_RECONFIGURE;
7797 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
7798 const struct flow *flow,
7799 const struct ofpact *ofpacts, size_t ofpacts_len)
7801 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7802 struct initial_vals initial_vals;
7803 struct odputil_keybuf keybuf;
7804 struct dpif_flow_stats stats;
7805 struct xlate_out xout;
7806 struct xlate_in xin;
7810 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
7811 odp_flow_key_from_flow(&key, flow,
7812 ofp_port_to_odp_port(ofproto, flow->in_port));
7814 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
7816 initial_vals.vlan_tci = flow->vlan_tci;
7817 initial_vals.tunnel_ip_tos = 0;
7818 xlate_in_init(&xin, ofproto, flow, &initial_vals, NULL, stats.tcp_flags,
7820 xin.resubmit_stats = &stats;
7821 xin.ofpacts_len = ofpacts_len;
7822 xin.ofpacts = ofpacts;
7824 xlate_actions(&xin, &xout);
7825 dpif_execute(ofproto->backer->dpif, key.data, key.size,
7826 xout.odp_actions.data, xout.odp_actions.size, packet);
7827 xlate_out_uninit(&xout);
7835 set_netflow(struct ofproto *ofproto_,
7836 const struct netflow_options *netflow_options)
7838 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7840 if (netflow_options) {
7841 if (!ofproto->netflow) {
7842 ofproto->netflow = netflow_create();
7844 return netflow_set_options(ofproto->netflow, netflow_options);
7846 netflow_destroy(ofproto->netflow);
7847 ofproto->netflow = NULL;
7853 get_netflow_ids(const struct ofproto *ofproto_,
7854 uint8_t *engine_type, uint8_t *engine_id)
7856 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7858 dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
7862 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
7864 if (!facet_is_controller_flow(facet) &&
7865 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
7866 struct subfacet *subfacet;
7867 struct ofexpired expired;
7869 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
7870 if (subfacet->path == SF_FAST_PATH) {
7871 struct dpif_flow_stats stats;
7873 subfacet_install(subfacet, &facet->xout.odp_actions, &stats);
7874 subfacet_update_stats(subfacet, &stats);
7878 expired.flow = facet->flow;
7879 expired.packet_count = facet->packet_count;
7880 expired.byte_count = facet->byte_count;
7881 expired.used = facet->used;
7882 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
7887 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
7889 struct facet *facet;
7891 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
7892 send_active_timeout(ofproto, facet);
7896 static struct ofproto_dpif *
7897 ofproto_dpif_lookup(const char *name)
7899 struct ofproto_dpif *ofproto;
7901 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
7902 hash_string(name, 0), &all_ofproto_dpifs) {
7903 if (!strcmp(ofproto->up.name, name)) {
7911 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
7912 const char *argv[], void *aux OVS_UNUSED)
7914 struct ofproto_dpif *ofproto;
7917 ofproto = ofproto_dpif_lookup(argv[1]);
7919 unixctl_command_reply_error(conn, "no such bridge");
7922 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
7924 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
7925 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
7929 unixctl_command_reply(conn, "table successfully flushed");
7933 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
7934 const char *argv[], void *aux OVS_UNUSED)
7936 struct ds ds = DS_EMPTY_INITIALIZER;
7937 const struct ofproto_dpif *ofproto;
7938 const struct mac_entry *e;
7940 ofproto = ofproto_dpif_lookup(argv[1]);
7942 unixctl_command_reply_error(conn, "no such bridge");
7946 ds_put_cstr(&ds, " port VLAN MAC Age\n");
7947 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
7948 struct ofbundle *bundle = e->port.p;
7949 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
7950 ofbundle_get_a_port(bundle)->odp_port,
7951 e->vlan, ETH_ADDR_ARGS(e->mac),
7952 mac_entry_age(ofproto->ml, e));
7954 unixctl_command_reply(conn, ds_cstr(&ds));
7959 struct xlate_out xout;
7960 struct xlate_in xin;
7966 trace_format_rule(struct ds *result, uint8_t table_id, int level,
7967 const struct rule_dpif *rule)
7969 ds_put_char_multiple(result, '\t', level);
7971 ds_put_cstr(result, "No match\n");
7975 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
7976 table_id, ntohll(rule->up.flow_cookie));
7977 cls_rule_format(&rule->up.cr, result);
7978 ds_put_char(result, '\n');
7980 ds_put_char_multiple(result, '\t', level);
7981 ds_put_cstr(result, "OpenFlow ");
7982 ofpacts_format(rule->up.ofpacts, rule->up.ofpacts_len, result);
7983 ds_put_char(result, '\n');
7987 trace_format_flow(struct ds *result, int level, const char *title,
7988 struct trace_ctx *trace)
7990 ds_put_char_multiple(result, '\t', level);
7991 ds_put_format(result, "%s: ", title);
7992 if (flow_equal(&trace->xin.flow, &trace->flow)) {
7993 ds_put_cstr(result, "unchanged");
7995 flow_format(result, &trace->xin.flow);
7996 trace->flow = trace->xin.flow;
7998 ds_put_char(result, '\n');
8002 trace_format_regs(struct ds *result, int level, const char *title,
8003 struct trace_ctx *trace)
8007 ds_put_char_multiple(result, '\t', level);
8008 ds_put_format(result, "%s:", title);
8009 for (i = 0; i < FLOW_N_REGS; i++) {
8010 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
8012 ds_put_char(result, '\n');
8016 trace_format_odp(struct ds *result, int level, const char *title,
8017 struct trace_ctx *trace)
8019 struct ofpbuf *odp_actions = &trace->xout.odp_actions;
8021 ds_put_char_multiple(result, '\t', level);
8022 ds_put_format(result, "%s: ", title);
8023 format_odp_actions(result, odp_actions->data, odp_actions->size);
8024 ds_put_char(result, '\n');
8028 trace_resubmit(struct xlate_ctx *ctx, struct rule_dpif *rule)
8030 struct trace_ctx *trace = CONTAINER_OF(ctx->xin, struct trace_ctx, xin);
8031 struct ds *result = trace->result;
8033 ds_put_char(result, '\n');
8034 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
8035 trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace);
8036 trace_format_odp(result, ctx->recurse + 1, "Resubmitted odp", trace);
8037 trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule);
8041 trace_report(struct xlate_ctx *ctx, const char *s)
8043 struct trace_ctx *trace = CONTAINER_OF(ctx->xin, struct trace_ctx, xin);
8044 struct ds *result = trace->result;
8046 ds_put_char_multiple(result, '\t', ctx->recurse);
8047 ds_put_cstr(result, s);
8048 ds_put_char(result, '\n');
8052 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
8053 void *aux OVS_UNUSED)
8055 const char *dpname = argv[1];
8056 struct ofproto_dpif *ofproto;
8057 struct ofpbuf odp_key;
8058 struct ofpbuf *packet;
8059 struct initial_vals initial_vals;
8065 ofpbuf_init(&odp_key, 0);
8068 ofproto = ofproto_dpif_lookup(dpname);
8070 unixctl_command_reply_error(conn, "Unknown ofproto (use ofproto/list "
8074 if (argc == 3 || (argc == 4 && !strcmp(argv[3], "-generate"))) {
8075 /* ofproto/trace dpname flow [-generate] */
8076 const char *flow_s = argv[2];
8077 const char *generate_s = argv[3];
8079 /* Allow 'flow_s' to be either a datapath flow or an OpenFlow-like
8080 * flow. We guess which type it is based on whether 'flow_s' contains
8081 * an '(', since a datapath flow always contains '(') but an
8082 * OpenFlow-like flow should not (in fact it's allowed but I believe
8083 * that's not documented anywhere).
8085 * An alternative would be to try to parse 'flow_s' both ways, but then
8086 * it would be tricky giving a sensible error message. After all, do
8087 * you just say "syntax error" or do you present both error messages?
8088 * Both choices seem lousy. */
8089 if (strchr(flow_s, '(')) {
8092 /* Convert string to datapath key. */
8093 ofpbuf_init(&odp_key, 0);
8094 error = odp_flow_key_from_string(flow_s, NULL, &odp_key);
8096 unixctl_command_reply_error(conn, "Bad flow syntax");
8100 /* The user might have specified the wrong ofproto but within the
8101 * same backer. That's OK, ofproto_receive() can find the right
8103 if (ofproto_receive(ofproto->backer, NULL, odp_key.data,
8104 odp_key.size, &flow, NULL, &ofproto, NULL,
8106 unixctl_command_reply_error(conn, "Invalid flow");
8109 ds_put_format(&result, "Bridge: %s\n", ofproto->up.name);
8113 error_s = parse_ofp_exact_flow(&flow, argv[2]);
8115 unixctl_command_reply_error(conn, error_s);
8120 initial_vals.vlan_tci = flow.vlan_tci;
8121 initial_vals.tunnel_ip_tos = flow.tunnel.ip_tos;
8124 /* Generate a packet, if requested. */
8126 packet = ofpbuf_new(0);
8127 flow_compose(packet, &flow);
8129 } else if (argc == 7) {
8130 /* ofproto/trace dpname priority tun_id in_port mark packet */
8131 const char *priority_s = argv[2];
8132 const char *tun_id_s = argv[3];
8133 const char *in_port_s = argv[4];
8134 const char *mark_s = argv[5];
8135 const char *packet_s = argv[6];
8136 uint32_t in_port = atoi(in_port_s);
8137 ovs_be64 tun_id = htonll(strtoull(tun_id_s, NULL, 0));
8138 uint32_t priority = atoi(priority_s);
8139 uint32_t mark = atoi(mark_s);
8142 msg = eth_from_hex(packet_s, &packet);
8144 unixctl_command_reply_error(conn, msg);
8148 ds_put_cstr(&result, "Packet: ");
8149 s = ofp_packet_to_string(packet->data, packet->size);
8150 ds_put_cstr(&result, s);
8153 flow_extract(packet, priority, mark, NULL, in_port, &flow);
8154 flow.tunnel.tun_id = tun_id;
8155 initial_vals.vlan_tci = flow.vlan_tci;
8156 initial_vals.tunnel_ip_tos = flow.tunnel.ip_tos;
8158 unixctl_command_reply_error(conn, "Bad command syntax");
8162 ofproto_trace(ofproto, &flow, packet, &initial_vals, &result);
8163 unixctl_command_reply(conn, ds_cstr(&result));
8166 ds_destroy(&result);
8167 ofpbuf_delete(packet);
8168 ofpbuf_uninit(&odp_key);
8172 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
8173 const struct ofpbuf *packet,
8174 const struct initial_vals *initial_vals, struct ds *ds)
8176 struct rule_dpif *rule;
8178 ds_put_cstr(ds, "Flow: ");
8179 flow_format(ds, flow);
8180 ds_put_char(ds, '\n');
8182 rule = rule_dpif_lookup(ofproto, flow);
8184 trace_format_rule(ds, 0, 0, rule);
8185 if (rule == ofproto->miss_rule) {
8186 ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
8187 } else if (rule == ofproto->no_packet_in_rule) {
8188 ds_put_cstr(ds, "\nNo match, packets dropped because "
8189 "OFPPC_NO_PACKET_IN is set on in_port.\n");
8193 uint64_t odp_actions_stub[1024 / 8];
8194 struct ofpbuf odp_actions;
8196 struct trace_ctx trace;
8199 tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
8202 ofpbuf_use_stub(&odp_actions,
8203 odp_actions_stub, sizeof odp_actions_stub);
8204 xlate_in_init(&trace.xin, ofproto, flow, initial_vals, rule, tcp_flags,
8206 trace.xin.resubmit_hook = trace_resubmit;
8207 trace.xin.report_hook = trace_report;
8208 xlate_actions(&trace.xin, &trace.xout);
8210 ds_put_char(ds, '\n');
8211 trace_format_flow(ds, 0, "Final flow", &trace);
8212 ds_put_cstr(ds, "Datapath actions: ");
8213 format_odp_actions(ds, trace.xout.odp_actions.data,
8214 trace.xout.odp_actions.size);
8216 if (trace.xout.slow) {
8217 ds_put_cstr(ds, "\nThis flow is handled by the userspace "
8218 "slow path because it:");
8219 switch (trace.xout.slow) {
8221 ds_put_cstr(ds, "\n\t- Consists of CFM packets.");
8224 ds_put_cstr(ds, "\n\t- Consists of LACP packets.");
8227 ds_put_cstr(ds, "\n\t- Consists of STP packets.");
8229 case SLOW_CONTROLLER:
8230 ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages "
8231 "to the OpenFlow controller.");
8238 xlate_out_uninit(&trace.xout);
8243 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
8244 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
8247 unixctl_command_reply(conn, NULL);
8251 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
8252 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
8255 unixctl_command_reply(conn, NULL);
8258 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
8259 * 'reply' describing the results. */
8261 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
8263 struct facet *facet;
8267 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
8268 if (!facet_check_consistency(facet)) {
8273 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
8277 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
8278 ofproto->up.name, errors);
8280 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
8285 ofproto_dpif_self_check(struct unixctl_conn *conn,
8286 int argc, const char *argv[], void *aux OVS_UNUSED)
8288 struct ds reply = DS_EMPTY_INITIALIZER;
8289 struct ofproto_dpif *ofproto;
8292 ofproto = ofproto_dpif_lookup(argv[1]);
8294 unixctl_command_reply_error(conn, "Unknown ofproto (use "
8295 "ofproto/list for help)");
8298 ofproto_dpif_self_check__(ofproto, &reply);
8300 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
8301 ofproto_dpif_self_check__(ofproto, &reply);
8305 unixctl_command_reply(conn, ds_cstr(&reply));
8309 /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list
8310 * of the 'ofproto_shash' nodes. It is the responsibility of the caller
8311 * to destroy 'ofproto_shash' and free the returned value. */
8312 static const struct shash_node **
8313 get_ofprotos(struct shash *ofproto_shash)
8315 const struct ofproto_dpif *ofproto;
8317 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
8318 char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
8319 shash_add_nocopy(ofproto_shash, name, ofproto);
8322 return shash_sort(ofproto_shash);
8326 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
8327 const char *argv[] OVS_UNUSED,
8328 void *aux OVS_UNUSED)
8330 struct ds ds = DS_EMPTY_INITIALIZER;
8331 struct shash ofproto_shash;
8332 const struct shash_node **sorted_ofprotos;
8335 shash_init(&ofproto_shash);
8336 sorted_ofprotos = get_ofprotos(&ofproto_shash);
8337 for (i = 0; i < shash_count(&ofproto_shash); i++) {
8338 const struct shash_node *node = sorted_ofprotos[i];
8339 ds_put_format(&ds, "%s\n", node->name);
8342 shash_destroy(&ofproto_shash);
8343 free(sorted_ofprotos);
8345 unixctl_command_reply(conn, ds_cstr(&ds));
8350 show_dp_format(const struct ofproto_dpif *ofproto, struct ds *ds)
8352 const struct shash_node **ports;
8354 struct avg_subfacet_rates lifetime;
8355 unsigned long long int minutes;
8356 const int min_ms = 60 * 1000; /* milliseconds in one minute. */
8358 minutes = (time_msec() - ofproto->created) / min_ms;
8361 lifetime.add_rate = (double)ofproto->total_subfacet_add_count
8363 lifetime.del_rate = (double)ofproto->total_subfacet_del_count
8366 lifetime.add_rate = 0.0;
8367 lifetime.del_rate = 0.0;
8370 ds_put_format(ds, "%s (%s):\n", ofproto->up.name,
8371 dpif_name(ofproto->backer->dpif));
8373 "\tlookups: hit:%"PRIu64" missed:%"PRIu64"\n",
8374 ofproto->n_hit, ofproto->n_missed);
8375 ds_put_format(ds, "\tflows: cur: %zu, avg: %5.3f, max: %d,"
8376 " life span: %llu(ms)\n",
8377 hmap_count(&ofproto->subfacets),
8378 avg_subfacet_count(ofproto),
8379 ofproto->max_n_subfacet,
8380 avg_subfacet_life_span(ofproto));
8381 if (minutes >= 60) {
8382 show_dp_rates(ds, "\t\thourly avg:", &ofproto->hourly);
8384 if (minutes >= 60 * 24) {
8385 show_dp_rates(ds, "\t\tdaily avg:", &ofproto->daily);
8387 show_dp_rates(ds, "\t\toverall avg:", &lifetime);
8389 ports = shash_sort(&ofproto->up.port_by_name);
8390 for (i = 0; i < shash_count(&ofproto->up.port_by_name); i++) {
8391 const struct shash_node *node = ports[i];
8392 struct ofport *ofport = node->data;
8393 const char *name = netdev_get_name(ofport->netdev);
8394 const char *type = netdev_get_type(ofport->netdev);
8397 ds_put_format(ds, "\t%s %u/", name, ofport->ofp_port);
8399 odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
8400 if (odp_port != OVSP_NONE) {
8401 ds_put_format(ds, "%"PRIu32":", odp_port);
8403 ds_put_cstr(ds, "none:");
8406 if (strcmp(type, "system")) {
8407 struct netdev *netdev;
8410 ds_put_format(ds, " (%s", type);
8412 error = netdev_open(name, type, &netdev);
8417 error = netdev_get_config(netdev, &config);
8419 const struct smap_node **nodes;
8422 nodes = smap_sort(&config);
8423 for (i = 0; i < smap_count(&config); i++) {
8424 const struct smap_node *node = nodes[i];
8425 ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
8426 node->key, node->value);
8430 smap_destroy(&config);
8432 netdev_close(netdev);
8434 ds_put_char(ds, ')');
8436 ds_put_char(ds, '\n');
8442 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc,
8443 const char *argv[], void *aux OVS_UNUSED)
8445 struct ds ds = DS_EMPTY_INITIALIZER;
8446 const struct ofproto_dpif *ofproto;
8450 for (i = 1; i < argc; i++) {
8451 ofproto = ofproto_dpif_lookup(argv[i]);
8453 ds_put_format(&ds, "Unknown bridge %s (use dpif/dump-dps "
8454 "for help)", argv[i]);
8455 unixctl_command_reply_error(conn, ds_cstr(&ds));
8458 show_dp_format(ofproto, &ds);
8461 struct shash ofproto_shash;
8462 const struct shash_node **sorted_ofprotos;
8465 shash_init(&ofproto_shash);
8466 sorted_ofprotos = get_ofprotos(&ofproto_shash);
8467 for (i = 0; i < shash_count(&ofproto_shash); i++) {
8468 const struct shash_node *node = sorted_ofprotos[i];
8469 show_dp_format(node->data, &ds);
8472 shash_destroy(&ofproto_shash);
8473 free(sorted_ofprotos);
8476 unixctl_command_reply(conn, ds_cstr(&ds));
8481 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
8482 int argc OVS_UNUSED, const char *argv[],
8483 void *aux OVS_UNUSED)
8485 struct ds ds = DS_EMPTY_INITIALIZER;
8486 const struct ofproto_dpif *ofproto;
8487 struct subfacet *subfacet;
8489 ofproto = ofproto_dpif_lookup(argv[1]);
8491 unixctl_command_reply_error(conn, "no such bridge");
8495 update_stats(ofproto->backer);
8497 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
8498 struct facet *facet = subfacet->facet;
8500 odp_flow_key_format(subfacet->key, subfacet->key_len, &ds);
8502 ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:",
8503 subfacet->dp_packet_count, subfacet->dp_byte_count);
8504 if (subfacet->used) {
8505 ds_put_format(&ds, "%.3fs",
8506 (time_msec() - subfacet->used) / 1000.0);
8508 ds_put_format(&ds, "never");
8510 if (subfacet->facet->tcp_flags) {
8511 ds_put_cstr(&ds, ", flags:");
8512 packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags);
8515 ds_put_cstr(&ds, ", actions:");
8516 if (facet->xout.slow) {
8517 uint64_t slow_path_stub[128 / 8];
8518 const struct nlattr *actions;
8521 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
8522 slow_path_stub, sizeof slow_path_stub,
8523 &actions, &actions_len);
8524 format_odp_actions(&ds, actions, actions_len);
8526 format_odp_actions(&ds, facet->xout.odp_actions.data,
8527 facet->xout.odp_actions.size);
8529 ds_put_char(&ds, '\n');
8532 unixctl_command_reply(conn, ds_cstr(&ds));
8537 ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn,
8538 int argc OVS_UNUSED, const char *argv[],
8539 void *aux OVS_UNUSED)
8541 struct ds ds = DS_EMPTY_INITIALIZER;
8542 struct ofproto_dpif *ofproto;
8544 ofproto = ofproto_dpif_lookup(argv[1]);
8546 unixctl_command_reply_error(conn, "no such bridge");
8550 flush(&ofproto->up);
8552 unixctl_command_reply(conn, ds_cstr(&ds));
8557 ofproto_dpif_unixctl_init(void)
8559 static bool registered;
8565 unixctl_command_register(
8567 "bridge {priority tun_id in_port mark packet | odp_flow [-generate]}",
8568 2, 6, ofproto_unixctl_trace, NULL);
8569 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
8570 ofproto_unixctl_fdb_flush, NULL);
8571 unixctl_command_register("fdb/show", "bridge", 1, 1,
8572 ofproto_unixctl_fdb_show, NULL);
8573 unixctl_command_register("ofproto/clog", "", 0, 0,
8574 ofproto_dpif_clog, NULL);
8575 unixctl_command_register("ofproto/unclog", "", 0, 0,
8576 ofproto_dpif_unclog, NULL);
8577 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
8578 ofproto_dpif_self_check, NULL);
8579 unixctl_command_register("dpif/dump-dps", "", 0, 0,
8580 ofproto_unixctl_dpif_dump_dps, NULL);
8581 unixctl_command_register("dpif/show", "[bridge]", 0, INT_MAX,
8582 ofproto_unixctl_dpif_show, NULL);
8583 unixctl_command_register("dpif/dump-flows", "bridge", 1, 1,
8584 ofproto_unixctl_dpif_dump_flows, NULL);
8585 unixctl_command_register("dpif/del-flows", "bridge", 1, 1,
8586 ofproto_unixctl_dpif_del_flows, NULL);
8589 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
8591 * This is deprecated. It is only for compatibility with broken device drivers
8592 * in old versions of Linux that do not properly support VLANs when VLAN
8593 * devices are not used. When broken device drivers are no longer in
8594 * widespread use, we will delete these interfaces. */
8597 set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid)
8599 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
8600 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
8602 if (realdev_ofp_port == ofport->realdev_ofp_port
8603 && vid == ofport->vlandev_vid) {
8607 ofproto->backer->need_revalidate = REV_RECONFIGURE;
8609 if (ofport->realdev_ofp_port) {
8612 if (realdev_ofp_port && ofport->bundle) {
8613 /* vlandevs are enslaved to their realdevs, so they are not allowed to
8614 * themselves be part of a bundle. */
8615 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
8618 ofport->realdev_ofp_port = realdev_ofp_port;
8619 ofport->vlandev_vid = vid;
8621 if (realdev_ofp_port) {
8622 vsp_add(ofport, realdev_ofp_port, vid);
8629 hash_realdev_vid(uint16_t realdev_ofp_port, int vid)
8631 return hash_2words(realdev_ofp_port, vid);
8634 /* Returns the OFP port number of the Linux VLAN device that corresponds to
8635 * 'vlan_tci' on the network device with port number 'realdev_ofp_port' in
8636 * 'struct ofport_dpif'. For example, given 'realdev_ofp_port' of eth0 and
8637 * 'vlan_tci' 9, it would return the port number of eth0.9.
8639 * Unless VLAN splinters are enabled for port 'realdev_ofp_port', this
8640 * function just returns its 'realdev_ofp_port' argument. */
8642 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
8643 uint16_t realdev_ofp_port, ovs_be16 vlan_tci)
8645 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
8646 int vid = vlan_tci_to_vid(vlan_tci);
8647 const struct vlan_splinter *vsp;
8649 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
8650 hash_realdev_vid(realdev_ofp_port, vid),
8651 &ofproto->realdev_vid_map) {
8652 if (vsp->realdev_ofp_port == realdev_ofp_port
8653 && vsp->vid == vid) {
8654 return vsp->vlandev_ofp_port;
8658 return realdev_ofp_port;
8661 static struct vlan_splinter *
8662 vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port)
8664 struct vlan_splinter *vsp;
8666 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, hash_int(vlandev_ofp_port, 0),
8667 &ofproto->vlandev_map) {
8668 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
8676 /* Returns the OpenFlow port number of the "real" device underlying the Linux
8677 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
8678 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
8679 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
8680 * eth0 and store 9 in '*vid'.
8682 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
8683 * VLAN device. Unless VLAN splinters are enabled, this is what this function
8686 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
8687 uint16_t vlandev_ofp_port, int *vid)
8689 if (!hmap_is_empty(&ofproto->vlandev_map)) {
8690 const struct vlan_splinter *vsp;
8692 vsp = vlandev_find(ofproto, vlandev_ofp_port);
8697 return vsp->realdev_ofp_port;
8703 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
8704 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
8705 * 'flow->in_port' to the "real" device backing the VLAN device, sets
8706 * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is
8707 * always the case unless VLAN splinters are enabled), returns false without
8708 * making any changes. */
8710 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
8715 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port, &vid);
8720 /* Cause the flow to be processed as if it came in on the real device with
8721 * the VLAN device's VLAN ID. */
8722 flow->in_port = realdev;
8723 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
8728 vsp_remove(struct ofport_dpif *port)
8730 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
8731 struct vlan_splinter *vsp;
8733 vsp = vlandev_find(ofproto, port->up.ofp_port);
8735 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
8736 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
8739 port->realdev_ofp_port = 0;
8741 VLOG_ERR("missing vlan device record");
8746 vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid)
8748 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
8750 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
8751 && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid))
8752 == realdev_ofp_port)) {
8753 struct vlan_splinter *vsp;
8755 vsp = xmalloc(sizeof *vsp);
8756 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
8757 hash_int(port->up.ofp_port, 0));
8758 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
8759 hash_realdev_vid(realdev_ofp_port, vid));
8760 vsp->realdev_ofp_port = realdev_ofp_port;
8761 vsp->vlandev_ofp_port = port->up.ofp_port;
8764 port->realdev_ofp_port = realdev_ofp_port;
8766 VLOG_ERR("duplicate vlan device record");
8771 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
8773 const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
8774 return ofport ? ofport->odp_port : OVSP_NONE;
8777 static struct ofport_dpif *
8778 odp_port_to_ofport(const struct dpif_backer *backer, uint32_t odp_port)
8780 struct ofport_dpif *port;
8782 HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node,
8783 hash_int(odp_port, 0),
8784 &backer->odp_to_ofport_map) {
8785 if (port->odp_port == odp_port) {
8794 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
8796 struct ofport_dpif *port;
8798 port = odp_port_to_ofport(ofproto->backer, odp_port);
8799 if (port && &ofproto->up == port->up.ofproto) {
8800 return port->up.ofp_port;
8805 static unsigned long long int
8806 avg_subfacet_life_span(const struct ofproto_dpif *ofproto)
8808 unsigned long long int dc;
8809 unsigned long long int avg;
8811 dc = ofproto->total_subfacet_del_count + ofproto->subfacet_del_count;
8812 avg = dc ? ofproto->total_subfacet_life_span / dc : 0;
8818 avg_subfacet_count(const struct ofproto_dpif *ofproto)
8822 if (ofproto->n_update_stats) {
8823 avg_c = (double)ofproto->total_subfacet_count
8824 / ofproto->n_update_stats;
8831 show_dp_rates(struct ds *ds, const char *heading,
8832 const struct avg_subfacet_rates *rates)
8834 ds_put_format(ds, "%s add rate: %5.3f/min, del rate: %5.3f/min\n",
8835 heading, rates->add_rate, rates->del_rate);
8839 update_max_subfacet_count(struct ofproto_dpif *ofproto)
8841 ofproto->max_n_subfacet = MAX(ofproto->max_n_subfacet,
8842 hmap_count(&ofproto->subfacets));
8845 /* Compute exponentially weighted moving average, adding 'new' as the newest,
8846 * most heavily weighted element. 'base' designates the rate of decay: after
8847 * 'base' further updates, 'new''s weight in the EWMA decays to about 1/e
8850 exp_mavg(double *avg, int base, double new)
8852 *avg = (*avg * (base - 1) + new) / base;
8856 update_moving_averages(struct ofproto_dpif *ofproto)
8858 const int min_ms = 60 * 1000; /* milliseconds in one minute. */
8860 /* Update hourly averages on the minute boundaries. */
8861 if (time_msec() - ofproto->last_minute >= min_ms) {
8862 exp_mavg(&ofproto->hourly.add_rate, 60, ofproto->subfacet_add_count);
8863 exp_mavg(&ofproto->hourly.del_rate, 60, ofproto->subfacet_del_count);
8865 /* Update daily averages on the hour boundaries. */
8866 if ((ofproto->last_minute - ofproto->created) / min_ms % 60 == 59) {
8867 exp_mavg(&ofproto->daily.add_rate, 24, ofproto->hourly.add_rate);
8868 exp_mavg(&ofproto->daily.del_rate, 24, ofproto->hourly.del_rate);
8871 ofproto->total_subfacet_add_count += ofproto->subfacet_add_count;
8872 ofproto->total_subfacet_del_count += ofproto->subfacet_del_count;
8873 ofproto->subfacet_add_count = 0;
8874 ofproto->subfacet_del_count = 0;
8875 ofproto->last_minute += min_ms;
8880 dpif_stats_update_hit_count(struct ofproto_dpif *ofproto, uint64_t delta)
8882 ofproto->n_hit += delta;
8885 const struct ofproto_class ofproto_dpif_class = {
8920 port_is_lacp_current,
8921 NULL, /* rule_choose_table */
8928 rule_modify_actions,
8940 get_stp_port_status,
8947 is_mirror_output_bundle,
8948 forward_bpdu_changed,
8949 set_mac_table_config,