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);
69 COVERAGE_DEFINE(subfacet_install_fail);
71 /* Maximum depth of flow table recursion (due to resubmit actions) in a
72 * flow translation. */
73 #define MAX_RESUBMIT_RECURSION 64
75 /* Number of implemented OpenFlow tables. */
76 enum { N_TABLES = 255 };
77 enum { TBL_INTERNAL = N_TABLES - 1 }; /* Used for internal hidden rules. */
78 BUILD_ASSERT_DECL(N_TABLES >= 2 && N_TABLES <= 255);
90 * - Do include packets and bytes from facets that have been deleted or
91 * whose own statistics have been folded into the rule.
93 * - Do include packets and bytes sent "by hand" that were accounted to
94 * the rule without any facet being involved (this is a rare corner
95 * case in rule_execute()).
97 * - Do not include packet or bytes that can be obtained from any facet's
98 * packet_count or byte_count member or that can be obtained from the
99 * datapath by, e.g., dpif_flow_get() for any subfacet.
101 uint64_t packet_count; /* Number of packets received. */
102 uint64_t byte_count; /* Number of bytes received. */
104 tag_type tag; /* Caches rule_calculate_tag() result. */
106 struct list facets; /* List of "struct facet"s. */
109 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
111 return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
114 static struct rule_dpif *rule_dpif_lookup(struct ofproto_dpif *,
116 struct flow_wildcards *wc);
117 static struct rule_dpif *rule_dpif_lookup__(struct ofproto_dpif *,
119 struct flow_wildcards *wc,
121 static struct rule_dpif *rule_dpif_miss_rule(struct ofproto_dpif *ofproto,
122 const struct flow *flow);
124 static void rule_get_stats(struct rule *, uint64_t *packets, uint64_t *bytes);
125 static void rule_credit_stats(struct rule_dpif *,
126 const struct dpif_flow_stats *);
127 static tag_type rule_calculate_tag(const struct flow *,
128 const struct minimask *, uint32_t basis);
129 static void rule_invalidate(const struct rule_dpif *);
131 #define MAX_MIRRORS 32
132 typedef uint32_t mirror_mask_t;
133 #define MIRROR_MASK_C(X) UINT32_C(X)
134 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
136 struct ofproto_dpif *ofproto; /* Owning ofproto. */
137 size_t idx; /* In ofproto's "mirrors" array. */
138 void *aux; /* Key supplied by ofproto's client. */
139 char *name; /* Identifier for log messages. */
141 /* Selection criteria. */
142 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
143 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
144 unsigned long *vlans; /* Bitmap of chosen VLANs, NULL selects all. */
146 /* Output (exactly one of out == NULL and out_vlan == -1 is true). */
147 struct ofbundle *out; /* Output port or NULL. */
148 int out_vlan; /* Output VLAN or -1. */
149 mirror_mask_t dup_mirrors; /* Bitmap of mirrors with the same output. */
152 int64_t packet_count; /* Number of packets sent. */
153 int64_t byte_count; /* Number of bytes sent. */
156 static void mirror_destroy(struct ofmirror *);
157 static void update_mirror_stats(struct ofproto_dpif *ofproto,
158 mirror_mask_t mirrors,
159 uint64_t packets, uint64_t bytes);
162 struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
163 struct ofproto_dpif *ofproto; /* Owning ofproto. */
164 void *aux; /* Key supplied by ofproto's client. */
165 char *name; /* Identifier for log messages. */
168 struct list ports; /* Contains "struct ofport"s. */
169 enum port_vlan_mode vlan_mode; /* VLAN mode */
170 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
171 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
172 * NULL if all VLANs are trunked. */
173 struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */
174 struct bond *bond; /* Nonnull iff more than one port. */
175 bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */
178 bool floodable; /* True if no port has OFPUTIL_PC_NO_FLOOD set. */
180 /* Port mirroring info. */
181 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
182 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
183 mirror_mask_t mirror_out; /* Mirrors that output to this bundle. */
186 static void bundle_remove(struct ofport *);
187 static void bundle_update(struct ofbundle *);
188 static void bundle_destroy(struct ofbundle *);
189 static void bundle_del_port(struct ofport_dpif *);
190 static void bundle_run(struct ofbundle *);
191 static void bundle_wait(struct ofbundle *);
192 static struct ofbundle *lookup_input_bundle(const struct ofproto_dpif *,
193 uint16_t in_port, bool warn,
194 struct ofport_dpif **in_ofportp);
196 /* A controller may use OFPP_NONE as the ingress port to indicate that
197 * it did not arrive on a "real" port. 'ofpp_none_bundle' exists for
198 * when an input bundle is needed for validation (e.g., mirroring or
199 * OFPP_NORMAL processing). It is not connected to an 'ofproto' or have
200 * any 'port' structs, so care must be taken when dealing with it. */
201 static struct ofbundle ofpp_none_bundle = {
203 .vlan_mode = PORT_VLAN_TRUNK
206 static void stp_run(struct ofproto_dpif *ofproto);
207 static void stp_wait(struct ofproto_dpif *ofproto);
208 static int set_stp_port(struct ofport *,
209 const struct ofproto_port_stp_settings *);
211 static bool ofbundle_includes_vlan(const struct ofbundle *, uint16_t vlan);
215 /* Initial values of fields of the packet that may be changed during
216 * flow processing and needed later. */
217 struct initial_vals {
218 /* This is the value of vlan_tci in the packet as actually received from
219 * dpif. This is the same as the facet's flow.vlan_tci unless the packet
220 * was received via a VLAN splinter. In that case, this value is 0
221 * (because the packet as actually received from the dpif had no 802.1Q
222 * tag) but the facet's flow.vlan_tci is set to the VLAN that the splinter
225 * This member should be removed when the VLAN splinters feature is no
229 /* If received on a tunnel, the IP TOS value of the tunnel. */
230 uint8_t tunnel_ip_tos;
234 /* Wildcards relevant in translation. Any fields that were used to
235 * calculate the action must be set for caching and kernel
236 * wildcarding to work. For example, if the flow lookup involved
237 * performing the "normal" action on IPv4 and ARP packets, 'wc'
238 * would have the 'in_port' (always set), 'dl_type' (flow match),
239 * 'vlan_tci' (normal action), and 'dl_dst' (normal action) fields
241 struct flow_wildcards wc;
243 tag_type tags; /* Tags associated with actions. */
244 enum slow_path_reason slow; /* 0 if fast path may be used. */
245 bool has_learn; /* Actions include NXAST_LEARN? */
246 bool has_normal; /* Actions output to OFPP_NORMAL? */
247 bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */
248 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
249 mirror_mask_t mirrors; /* Bitmap of associated mirrors. */
251 uint64_t odp_actions_stub[256 / 8];
252 struct ofpbuf odp_actions;
256 struct ofproto_dpif *ofproto;
258 /* Flow to which the OpenFlow actions apply. xlate_actions() will modify
259 * this flow when actions change header fields. */
262 struct initial_vals initial_vals;
264 /* The packet corresponding to 'flow', or a null pointer if we are
265 * revalidating without a packet to refer to. */
266 const struct ofpbuf *packet;
268 /* Should OFPP_NORMAL update the MAC learning table? Should "learn"
269 * actions update the flow table?
271 * We want to update these tables if we are actually processing a packet,
272 * or if we are accounting for packets that the datapath has processed, but
273 * not if we are just revalidating. */
276 /* The rule initiating translation or NULL. */
277 struct rule_dpif *rule;
279 /* The actions to translate. If 'rule' is not NULL, these may be NULL. */
280 const struct ofpact *ofpacts;
283 /* Union of the set of TCP flags seen so far in this flow. (Used only by
284 * NXAST_FIN_TIMEOUT. Set to zero to avoid updating updating rules'
288 /* If nonnull, flow translation calls this function just before executing a
289 * resubmit or OFPP_TABLE action. In addition, disables logging of traces
290 * when the recursion depth is exceeded.
292 * 'rule' is the rule being submitted into. It will be null if the
293 * resubmit or OFPP_TABLE action didn't find a matching rule.
295 * This is normally null so the client has to set it manually after
296 * calling xlate_in_init(). */
297 void (*resubmit_hook)(struct xlate_ctx *, struct rule_dpif *rule);
299 /* If nonnull, flow translation calls this function to report some
300 * significant decision, e.g. to explain why OFPP_NORMAL translation
301 * dropped a packet. */
302 void (*report_hook)(struct xlate_ctx *, const char *s);
304 /* If nonnull, flow translation credits the specified statistics to each
305 * rule reached through a resubmit or OFPP_TABLE action.
307 * This is normally null so the client has to set it manually after
308 * calling xlate_in_init(). */
309 const struct dpif_flow_stats *resubmit_stats;
312 /* Context used by xlate_actions() and its callees. */
314 struct xlate_in *xin;
315 struct xlate_out *xout;
317 struct ofproto_dpif *ofproto;
319 /* Flow at the last commit. */
320 struct flow base_flow;
322 /* Stack for the push and pop actions. Each stack element is of type
323 * "union mf_subvalue". */
324 union mf_subvalue init_stack[1024 / sizeof(union mf_subvalue)];
327 /* The rule that we are currently translating, or NULL. */
328 struct rule_dpif *rule;
330 int recurse; /* Recursion level, via xlate_table_action. */
331 bool max_resubmit_trigger; /* Recursed too deeply during translation. */
332 uint32_t orig_skb_priority; /* Priority when packet arrived. */
333 uint8_t table_id; /* OpenFlow table ID where flow was found. */
334 uint32_t sflow_n_outputs; /* Number of output ports. */
335 uint32_t sflow_odp_port; /* Output port for composing sFlow action. */
336 uint16_t user_cookie_offset;/* Used for user_action_cookie fixup. */
337 bool exit; /* No further actions should be processed. */
340 static void xlate_in_init(struct xlate_in *, struct ofproto_dpif *,
341 const struct flow *, const struct initial_vals *,
342 struct rule_dpif *, uint8_t tcp_flags,
343 const struct ofpbuf *);
345 static void xlate_out_uninit(struct xlate_out *);
347 static void xlate_actions(struct xlate_in *, struct xlate_out *);
349 static void xlate_actions_for_side_effects(struct xlate_in *);
351 static void xlate_table_action(struct xlate_ctx *, uint16_t in_port,
352 uint8_t table_id, bool may_packet_in);
354 static size_t put_userspace_action(const struct ofproto_dpif *,
355 struct ofpbuf *odp_actions,
357 const union user_action_cookie *,
360 static void compose_slow_path(const struct ofproto_dpif *, const struct flow *,
361 enum slow_path_reason,
362 uint64_t *stub, size_t stub_size,
363 const struct nlattr **actionsp,
364 size_t *actions_lenp);
366 static void xlate_report(struct xlate_ctx *ctx, const char *s);
368 static void xlate_out_copy(struct xlate_out *dst, const struct xlate_out *src);
370 /* A subfacet (see "struct subfacet" below) has three possible installation
373 * - SF_NOT_INSTALLED: Not installed in the datapath. This will only be the
374 * case just after the subfacet is created, just before the subfacet is
375 * destroyed, or if the datapath returns an error when we try to install a
378 * - SF_FAST_PATH: The subfacet's actions are installed in the datapath.
380 * - SF_SLOW_PATH: An action that sends every packet for the subfacet through
381 * ofproto_dpif is installed in the datapath.
384 SF_NOT_INSTALLED, /* No datapath flow for this subfacet. */
385 SF_FAST_PATH, /* Full actions are installed. */
386 SF_SLOW_PATH, /* Send-to-userspace action is installed. */
389 /* A dpif flow and actions associated with a facet.
391 * See also the large comment on struct facet. */
394 struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */
395 struct list list_node; /* In struct facet's 'facets' list. */
396 struct facet *facet; /* Owning facet. */
398 enum odp_key_fitness key_fitness;
402 long long int used; /* Time last used; time created if not used. */
403 long long int created; /* Time created. */
405 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
406 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
408 enum subfacet_path path; /* Installed in datapath? */
410 /* Datapath port the packet arrived on. This is needed to remove
411 * flows for ports that are no longer part of the bridge. Since the
412 * flow definition only has the OpenFlow port number and the port is
413 * no longer part of the bridge, we can't determine the datapath port
414 * number needed to delete the flow from the datapath. */
415 uint32_t odp_in_port;
418 #define SUBFACET_DESTROY_MAX_BATCH 50
420 static struct subfacet *subfacet_create(struct facet *, struct flow_miss *miss,
422 static struct subfacet *subfacet_find(struct ofproto_dpif *,
423 const struct nlattr *key, size_t key_len,
425 static void subfacet_destroy(struct subfacet *);
426 static void subfacet_destroy__(struct subfacet *);
427 static void subfacet_destroy_batch(struct ofproto_dpif *,
428 struct subfacet **, int n);
429 static void subfacet_reset_dp_stats(struct subfacet *,
430 struct dpif_flow_stats *);
431 static void subfacet_update_stats(struct subfacet *,
432 const struct dpif_flow_stats *);
433 static int subfacet_install(struct subfacet *,
434 const struct ofpbuf *odp_actions,
435 struct dpif_flow_stats *);
436 static void subfacet_uninstall(struct subfacet *);
438 /* A unique, non-overlapping instantiation of an OpenFlow flow.
440 * A facet associates a "struct flow", which represents the Open vSwitch
441 * userspace idea of an exact-match flow, with one or more subfacets.
442 * While the facet is created based on an exact-match flow, it is stored
443 * within the ofproto based on the wildcards that could be expressed
444 * based on the flow table and other configuration. (See the 'wc'
445 * description in "struct xlate_out" for more details.)
447 * Each subfacet tracks the datapath's idea of the flow equivalent to
448 * the facet. When the kernel module (or other dpif implementation) and
449 * Open vSwitch userspace agree on the definition of a flow key, there
450 * is exactly one subfacet per facet. If the dpif implementation
451 * supports more-specific flow matching than userspace, however, a facet
452 * can have more than one subfacet. Examples include the dpif
453 * implementation not supporting the same wildcards as userspace or some
454 * distinction in flow that userspace simply doesn't understand.
456 * Flow expiration works in terms of subfacets, so a facet must have at
457 * least one subfacet or it will never expire, leaking memory. */
460 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
461 struct list list_node; /* In owning rule's 'facets' list. */
462 struct rule_dpif *rule; /* Owning rule. */
465 struct list subfacets;
466 long long int used; /* Time last used; time created if not used. */
469 struct flow flow; /* Flow of the creating subfacet. */
470 struct cls_rule cr; /* In 'ofproto_dpif's facets classifier. */
474 * - Do include packets and bytes sent "by hand", e.g. with
477 * - Do include packets and bytes that were obtained from the datapath
478 * when a subfacet's statistics were reset (e.g. dpif_flow_put() with
479 * DPIF_FP_ZERO_STATS).
481 * - Do not include packets or bytes that can be obtained from the
482 * datapath for any existing subfacet.
484 uint64_t packet_count; /* Number of packets received. */
485 uint64_t byte_count; /* Number of bytes received. */
487 /* Resubmit statistics. */
488 uint64_t prev_packet_count; /* Number of packets from last stats push. */
489 uint64_t prev_byte_count; /* Number of bytes from last stats push. */
490 long long int prev_used; /* Used time from last stats push. */
493 uint64_t accounted_bytes; /* Bytes processed by facet_account(). */
494 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
495 uint8_t tcp_flags; /* TCP flags seen for this 'rule'. */
497 struct xlate_out xout;
499 /* Initial values of the packet that may be needed later. */
500 struct initial_vals initial_vals;
502 /* Storage for a single subfacet, to reduce malloc() time and space
503 * overhead. (A facet always has at least one subfacet and in the common
504 * case has exactly one subfacet. However, 'one_subfacet' may not
505 * always be valid, since it could have been removed after newer
506 * subfacets were pushed onto the 'subfacets' list.) */
507 struct subfacet one_subfacet;
509 long long int learn_rl; /* Rate limiter for facet_learn(). */
512 static struct facet *facet_create(const struct flow_miss *, struct rule_dpif *,
514 struct dpif_flow_stats *);
515 static void facet_remove(struct facet *);
516 static void facet_free(struct facet *);
518 static struct facet *facet_find(struct ofproto_dpif *, const struct flow *);
519 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
520 const struct flow *);
521 static bool facet_revalidate(struct facet *);
522 static bool facet_check_consistency(struct facet *);
524 static void facet_flush_stats(struct facet *);
526 static void facet_reset_counters(struct facet *);
527 static void facet_push_stats(struct facet *, bool may_learn);
528 static void facet_learn(struct facet *);
529 static void facet_account(struct facet *);
530 static void push_all_stats(void);
532 static bool facet_is_controller_flow(struct facet *);
535 struct hmap_node odp_port_node; /* In dpif_backer's "odp_to_ofport_map". */
539 struct ofbundle *bundle; /* Bundle that contains this port, if any. */
540 struct list bundle_node; /* In struct ofbundle's "ports" list. */
541 struct cfm *cfm; /* Connectivity Fault Management, if any. */
542 tag_type tag; /* Tag associated with this port. */
543 bool may_enable; /* May be enabled in bonds. */
544 long long int carrier_seq; /* Carrier status changes. */
545 struct tnl_port *tnl_port; /* Tunnel handle, or null. */
548 struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */
549 enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */
550 long long int stp_state_entered;
552 struct hmap priorities; /* Map of attached 'priority_to_dscp's. */
554 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
556 * This is deprecated. It is only for compatibility with broken device
557 * drivers in old versions of Linux that do not properly support VLANs when
558 * VLAN devices are not used. When broken device drivers are no longer in
559 * widespread use, we will delete these interfaces. */
560 uint16_t realdev_ofp_port;
564 /* Node in 'ofport_dpif''s 'priorities' map. Used to maintain a map from
565 * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
566 * traffic egressing the 'ofport' with that priority should be marked with. */
567 struct priority_to_dscp {
568 struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'priorities' map. */
569 uint32_t priority; /* Priority of this queue (see struct flow). */
571 uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */
574 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
576 * This is deprecated. It is only for compatibility with broken device drivers
577 * in old versions of Linux that do not properly support VLANs when VLAN
578 * devices are not used. When broken device drivers are no longer in
579 * widespread use, we will delete these interfaces. */
580 struct vlan_splinter {
581 struct hmap_node realdev_vid_node;
582 struct hmap_node vlandev_node;
583 uint16_t realdev_ofp_port;
584 uint16_t vlandev_ofp_port;
588 static uint16_t vsp_realdev_to_vlandev(const struct ofproto_dpif *,
589 uint16_t realdev_ofp_port,
591 static bool vsp_adjust_flow(const struct ofproto_dpif *, struct flow *);
592 static void vsp_remove(struct ofport_dpif *);
593 static void vsp_add(struct ofport_dpif *, uint16_t realdev_ofp_port, int vid);
595 static uint32_t ofp_port_to_odp_port(const struct ofproto_dpif *,
597 static uint16_t odp_port_to_ofp_port(const struct ofproto_dpif *,
600 static struct ofport_dpif *
601 ofport_dpif_cast(const struct ofport *ofport)
603 ovs_assert(ofport->ofproto->ofproto_class == &ofproto_dpif_class);
604 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
607 static void port_run(struct ofport_dpif *);
608 static void port_run_fast(struct ofport_dpif *);
609 static void port_wait(struct ofport_dpif *);
610 static int set_cfm(struct ofport *, const struct cfm_settings *);
611 static void ofport_clear_priorities(struct ofport_dpif *);
612 static void run_fast_rl(void);
614 struct dpif_completion {
615 struct list list_node;
616 struct ofoperation *op;
619 /* Extra information about a classifier table.
620 * Currently used just for optimized flow revalidation. */
622 /* If either of these is nonnull, then this table has a form that allows
623 * flows to be tagged to avoid revalidating most flows for the most common
624 * kinds of flow table changes. */
625 struct cls_table *catchall_table; /* Table that wildcards all fields. */
626 struct cls_table *other_table; /* Table with any other wildcard set. */
627 uint32_t basis; /* Keeps each table's tags separate. */
630 /* Reasons that we might need to revalidate every facet, and corresponding
633 * A value of 0 means that there is no need to revalidate.
635 * It would be nice to have some cleaner way to integrate with coverage
636 * counters, but with only a few reasons I guess this is good enough for
638 enum revalidate_reason {
639 REV_RECONFIGURE = 1, /* Switch configuration changed. */
640 REV_STP, /* Spanning tree protocol port status change. */
641 REV_PORT_TOGGLED, /* Port enabled or disabled by CFM, LACP, ...*/
642 REV_FLOW_TABLE, /* Flow table changed. */
643 REV_INCONSISTENCY /* Facet self-check failed. */
645 COVERAGE_DEFINE(rev_reconfigure);
646 COVERAGE_DEFINE(rev_stp);
647 COVERAGE_DEFINE(rev_port_toggled);
648 COVERAGE_DEFINE(rev_flow_table);
649 COVERAGE_DEFINE(rev_inconsistency);
651 /* Drop keys are odp flow keys which have drop flows installed in the kernel.
652 * These are datapath flows which have no associated ofproto, if they did we
653 * would use facets. */
655 struct hmap_node hmap_node;
660 /* All datapaths of a given type share a single dpif backer instance. */
665 struct timer next_expiration;
666 struct hmap odp_to_ofport_map; /* ODP port to ofport mapping. */
668 struct simap tnl_backers; /* Set of dpif ports backing tunnels. */
670 /* Facet revalidation flags applying to facets which use this backer. */
671 enum revalidate_reason need_revalidate; /* Revalidate every facet. */
672 struct tag_set revalidate_set; /* Revalidate only matching facets. */
674 struct hmap drop_keys; /* Set of dropped odp keys. */
675 bool recv_set_enable; /* Enables or disables receiving packets. */
678 /* All existing ofproto_backer instances, indexed by ofproto->up.type. */
679 static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers);
681 static void drop_key_clear(struct dpif_backer *);
682 static struct ofport_dpif *
683 odp_port_to_ofport(const struct dpif_backer *, uint32_t odp_port);
685 static void dpif_stats_update_hit_count(struct ofproto_dpif *ofproto,
687 struct avg_subfacet_rates {
688 double add_rate; /* Moving average of new flows created per minute. */
689 double del_rate; /* Moving average of flows deleted per minute. */
691 static void show_dp_rates(struct ds *ds, const char *heading,
692 const struct avg_subfacet_rates *rates);
693 static void exp_mavg(double *avg, int base, double new);
695 struct ofproto_dpif {
696 struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
698 struct dpif_backer *backer;
700 /* Special OpenFlow rules. */
701 struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */
702 struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */
703 struct rule_dpif *drop_frags_rule; /* Used in OFPC_FRAG_DROP mode. */
706 struct netflow *netflow;
707 struct dpif_sflow *sflow;
708 struct dpif_ipfix *ipfix;
709 struct hmap bundles; /* Contains "struct ofbundle"s. */
710 struct mac_learning *ml;
711 struct ofmirror *mirrors[MAX_MIRRORS];
713 bool has_bonded_bundles;
716 struct classifier facets; /* Contains 'struct facet's. */
717 struct hmap subfacets;
718 struct governor *governor;
719 long long int consistency_rl;
722 struct table_dpif tables[N_TABLES];
724 /* Support for debugging async flow mods. */
725 struct list completions;
727 bool has_bundle_action; /* True when the first bundle action appears. */
728 struct netdev_stats stats; /* To account packets generated and consumed in
733 long long int stp_last_tick;
735 /* VLAN splinters. */
736 struct hmap realdev_vid_map; /* (realdev,vid) -> vlandev. */
737 struct hmap vlandev_map; /* vlandev -> (realdev,vid). */
740 struct sset ports; /* Set of standard port names. */
741 struct sset ghost_ports; /* Ports with no datapath port. */
742 struct sset port_poll_set; /* Queued names for port_poll() reply. */
743 int port_poll_errno; /* Last errno for port_poll() reply. */
745 /* Per ofproto's dpif stats. */
749 /* Subfacet statistics.
751 * These keep track of the total number of subfacets added and deleted and
752 * flow life span. They are useful for computing the flow rates stats
753 * exposed via "ovs-appctl dpif/show". The goal is to learn about
754 * traffic patterns in ways that we can use later to improve Open vSwitch
755 * performance in new situations. */
756 long long int created; /* Time when it is created. */
757 unsigned int max_n_subfacet; /* Maximum number of flows */
759 /* The average number of subfacets... */
760 struct avg_subfacet_rates hourly; /* ...over the last hour. */
761 struct avg_subfacet_rates daily; /* ...over the last day. */
762 long long int last_minute; /* Last time 'hourly' was updated. */
764 /* Number of subfacets added or deleted since 'last_minute'. */
765 unsigned int subfacet_add_count;
766 unsigned int subfacet_del_count;
768 /* Number of subfacets added or deleted from 'created' to 'last_minute.' */
769 unsigned long long int total_subfacet_add_count;
770 unsigned long long int total_subfacet_del_count;
772 /* Sum of the number of milliseconds that each subfacet existed,
773 * over the subfacets that have been added and then later deleted. */
774 unsigned long long int total_subfacet_life_span;
776 /* Incremented by the number of currently existing subfacets, each
777 * time we pull statistics from the kernel. */
778 unsigned long long int total_subfacet_count;
780 /* Number of times we pull statistics from the kernel. */
781 unsigned long long int n_update_stats;
783 static unsigned long long int avg_subfacet_life_span(
784 const struct ofproto_dpif *);
785 static double avg_subfacet_count(const struct ofproto_dpif *ofproto);
786 static void update_moving_averages(struct ofproto_dpif *ofproto);
787 static void dpif_stats_update_hit_count(struct ofproto_dpif *ofproto,
789 static void update_max_subfacet_count(struct ofproto_dpif *ofproto);
791 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
792 * for debugging the asynchronous flow_mod implementation.) */
795 /* All existing ofproto_dpif instances, indexed by ->up.name. */
796 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
798 static void ofproto_dpif_unixctl_init(void);
800 static struct ofproto_dpif *
801 ofproto_dpif_cast(const struct ofproto *ofproto)
803 ovs_assert(ofproto->ofproto_class == &ofproto_dpif_class);
804 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
807 static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *,
809 static struct ofport_dpif *get_odp_port(const struct ofproto_dpif *,
811 static void ofproto_trace(struct ofproto_dpif *, const struct flow *,
812 const struct ofpbuf *,
813 const struct initial_vals *, struct ds *);
815 /* Packet processing. */
816 static void update_learning_table(struct ofproto_dpif *, const struct flow *,
817 struct flow_wildcards *, int vlan,
820 #define FLOW_MISS_MAX_BATCH 50
821 static int handle_upcalls(struct dpif_backer *, unsigned int max_batch);
823 /* Flow expiration. */
824 static int expire(struct dpif_backer *);
827 static void send_netflow_active_timeouts(struct ofproto_dpif *);
830 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
831 static size_t compose_sflow_action(const struct ofproto_dpif *,
832 struct ofpbuf *odp_actions,
833 const struct flow *, uint32_t odp_port);
834 static void compose_ipfix_action(const struct ofproto_dpif *,
835 struct ofpbuf *odp_actions,
836 const struct flow *);
837 static void add_mirror_actions(struct xlate_ctx *ctx,
838 const struct flow *flow);
839 /* Global variables. */
840 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
842 /* Initial mappings of port to bridge mappings. */
843 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
845 /* Factory functions. */
848 init(const struct shash *iface_hints)
850 struct shash_node *node;
852 /* Make a local copy, since we don't own 'iface_hints' elements. */
853 SHASH_FOR_EACH(node, iface_hints) {
854 const struct iface_hint *orig_hint = node->data;
855 struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
857 new_hint->br_name = xstrdup(orig_hint->br_name);
858 new_hint->br_type = xstrdup(orig_hint->br_type);
859 new_hint->ofp_port = orig_hint->ofp_port;
861 shash_add(&init_ofp_ports, node->name, new_hint);
866 enumerate_types(struct sset *types)
868 dp_enumerate_types(types);
872 enumerate_names(const char *type, struct sset *names)
874 struct ofproto_dpif *ofproto;
877 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
878 if (strcmp(type, ofproto->up.type)) {
881 sset_add(names, ofproto->up.name);
888 del(const char *type, const char *name)
893 error = dpif_open(name, type, &dpif);
895 error = dpif_delete(dpif);
902 port_open_type(const char *datapath_type, const char *port_type)
904 return dpif_port_open_type(datapath_type, port_type);
907 /* Type functions. */
909 static struct ofproto_dpif *
910 lookup_ofproto_dpif_by_port_name(const char *name)
912 struct ofproto_dpif *ofproto;
914 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
915 if (sset_contains(&ofproto->ports, name)) {
924 type_run(const char *type)
926 static long long int push_timer = LLONG_MIN;
927 struct dpif_backer *backer;
931 backer = shash_find_data(&all_dpif_backers, type);
933 /* This is not necessarily a problem, since backers are only
934 * created on demand. */
938 dpif_run(backer->dpif);
940 /* The most natural place to push facet statistics is when they're pulled
941 * from the datapath. However, when there are many flows in the datapath,
942 * this expensive operation can occur so frequently, that it reduces our
943 * ability to quickly set up flows. To reduce the cost, we push statistics
945 if (time_msec() > push_timer) {
946 push_timer = time_msec() + 2000;
950 /* If vswitchd started with other_config:flow_restore_wait set as "true",
951 * and the configuration has now changed to "false", enable receiving
952 * packets from the datapath. */
953 if (!backer->recv_set_enable && !ofproto_get_flow_restore_wait()) {
954 backer->recv_set_enable = true;
956 error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
958 VLOG_ERR("Failed to enable receiving packets in dpif.");
961 dpif_flow_flush(backer->dpif);
962 backer->need_revalidate = REV_RECONFIGURE;
965 if (backer->need_revalidate
966 || !tag_set_is_empty(&backer->revalidate_set)) {
967 struct tag_set revalidate_set = backer->revalidate_set;
968 bool need_revalidate = backer->need_revalidate;
969 struct ofproto_dpif *ofproto;
970 struct simap_node *node;
971 struct simap tmp_backers;
973 /* Handle tunnel garbage collection. */
974 simap_init(&tmp_backers);
975 simap_swap(&backer->tnl_backers, &tmp_backers);
977 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
978 struct ofport_dpif *iter;
980 if (backer != ofproto->backer) {
984 HMAP_FOR_EACH (iter, up.hmap_node, &ofproto->up.ports) {
987 if (!iter->tnl_port) {
991 dp_port = netdev_vport_get_dpif_port(iter->up.netdev);
992 node = simap_find(&tmp_backers, dp_port);
994 simap_put(&backer->tnl_backers, dp_port, node->data);
995 simap_delete(&tmp_backers, node);
996 node = simap_find(&backer->tnl_backers, dp_port);
998 node = simap_find(&backer->tnl_backers, dp_port);
1000 uint32_t odp_port = UINT32_MAX;
1002 if (!dpif_port_add(backer->dpif, iter->up.netdev,
1004 simap_put(&backer->tnl_backers, dp_port, odp_port);
1005 node = simap_find(&backer->tnl_backers, dp_port);
1010 iter->odp_port = node ? node->data : OVSP_NONE;
1011 if (tnl_port_reconfigure(&iter->up, iter->odp_port,
1013 backer->need_revalidate = REV_RECONFIGURE;
1018 SIMAP_FOR_EACH (node, &tmp_backers) {
1019 dpif_port_del(backer->dpif, node->data);
1021 simap_destroy(&tmp_backers);
1023 switch (backer->need_revalidate) {
1024 case REV_RECONFIGURE: COVERAGE_INC(rev_reconfigure); break;
1025 case REV_STP: COVERAGE_INC(rev_stp); break;
1026 case REV_PORT_TOGGLED: COVERAGE_INC(rev_port_toggled); break;
1027 case REV_FLOW_TABLE: COVERAGE_INC(rev_flow_table); break;
1028 case REV_INCONSISTENCY: COVERAGE_INC(rev_inconsistency); break;
1031 if (backer->need_revalidate) {
1032 /* Clear the drop_keys in case we should now be accepting some
1033 * formerly dropped flows. */
1034 drop_key_clear(backer);
1037 /* Clear the revalidation flags. */
1038 tag_set_init(&backer->revalidate_set);
1039 backer->need_revalidate = 0;
1041 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1042 struct facet *facet, *next;
1043 struct cls_cursor cursor;
1045 if (ofproto->backer != backer) {
1049 cls_cursor_init(&cursor, &ofproto->facets, NULL);
1050 CLS_CURSOR_FOR_EACH_SAFE (facet, next, cr, &cursor) {
1052 || tag_set_intersects(&revalidate_set, facet->xout.tags)) {
1053 facet_revalidate(facet);
1060 if (!backer->recv_set_enable) {
1061 /* Wake up before a max of 1000ms. */
1062 timer_set_duration(&backer->next_expiration, 1000);
1063 } else if (timer_expired(&backer->next_expiration)) {
1064 int delay = expire(backer);
1065 timer_set_duration(&backer->next_expiration, delay);
1068 /* Check for port changes in the dpif. */
1069 while ((error = dpif_port_poll(backer->dpif, &devname)) == 0) {
1070 struct ofproto_dpif *ofproto;
1071 struct dpif_port port;
1073 /* Don't report on the datapath's device. */
1074 if (!strcmp(devname, dpif_base_name(backer->dpif))) {
1078 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
1079 &all_ofproto_dpifs) {
1080 if (simap_contains(&ofproto->backer->tnl_backers, devname)) {
1085 ofproto = lookup_ofproto_dpif_by_port_name(devname);
1086 if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
1087 /* The port was removed. If we know the datapath,
1088 * report it through poll_set(). If we don't, it may be
1089 * notifying us of a removal we initiated, so ignore it.
1090 * If there's a pending ENOBUFS, let it stand, since
1091 * everything will be reevaluated. */
1092 if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
1093 sset_add(&ofproto->port_poll_set, devname);
1094 ofproto->port_poll_errno = 0;
1096 } else if (!ofproto) {
1097 /* The port was added, but we don't know with which
1098 * ofproto we should associate it. Delete it. */
1099 dpif_port_del(backer->dpif, port.port_no);
1101 dpif_port_destroy(&port);
1107 if (error != EAGAIN) {
1108 struct ofproto_dpif *ofproto;
1110 /* There was some sort of error, so propagate it to all
1111 * ofprotos that use this backer. */
1112 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
1113 &all_ofproto_dpifs) {
1114 if (ofproto->backer == backer) {
1115 sset_clear(&ofproto->port_poll_set);
1116 ofproto->port_poll_errno = error;
1125 dpif_backer_run_fast(struct dpif_backer *backer, int max_batch)
1129 /* If recv_set_enable is false, we should not handle upcalls. */
1130 if (!backer->recv_set_enable) {
1134 /* Handle one or more batches of upcalls, until there's nothing left to do
1135 * or until we do a fixed total amount of work.
1137 * We do work in batches because it can be much cheaper to set up a number
1138 * of flows and fire off their patches all at once. We do multiple batches
1139 * because in some cases handling a packet can cause another packet to be
1140 * queued almost immediately as part of the return flow. Both
1141 * optimizations can make major improvements on some benchmarks and
1142 * presumably for real traffic as well. */
1144 while (work < max_batch) {
1145 int retval = handle_upcalls(backer, max_batch - work);
1156 type_run_fast(const char *type)
1158 struct dpif_backer *backer;
1160 backer = shash_find_data(&all_dpif_backers, type);
1162 /* This is not necessarily a problem, since backers are only
1163 * created on demand. */
1167 return dpif_backer_run_fast(backer, FLOW_MISS_MAX_BATCH);
1173 static long long int port_rl = LLONG_MIN;
1174 static unsigned int backer_rl = 0;
1176 if (time_msec() >= port_rl) {
1177 struct ofproto_dpif *ofproto;
1178 struct ofport_dpif *ofport;
1180 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1182 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1183 port_run_fast(ofport);
1186 port_rl = time_msec() + 200;
1189 /* XXX: We have to be careful not to do too much work in this function. If
1190 * we call dpif_backer_run_fast() too often, or with too large a batch,
1191 * performance improves signifcantly, but at a cost. It's possible for the
1192 * number of flows in the datapath to increase without bound, and for poll
1193 * loops to take 10s of seconds. The correct solution to this problem,
1194 * long term, is to separate flow miss handling into it's own thread so it
1195 * isn't affected by revalidations, and expirations. Until then, this is
1196 * the best we can do. */
1197 if (++backer_rl >= 10) {
1198 struct shash_node *node;
1201 SHASH_FOR_EACH (node, &all_dpif_backers) {
1202 dpif_backer_run_fast(node->data, 1);
1208 type_wait(const char *type)
1210 struct dpif_backer *backer;
1212 backer = shash_find_data(&all_dpif_backers, type);
1214 /* This is not necessarily a problem, since backers are only
1215 * created on demand. */
1219 timer_wait(&backer->next_expiration);
1222 /* Basic life-cycle. */
1224 static int add_internal_flows(struct ofproto_dpif *);
1226 static struct ofproto *
1229 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
1230 return &ofproto->up;
1234 dealloc(struct ofproto *ofproto_)
1236 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1241 close_dpif_backer(struct dpif_backer *backer)
1243 struct shash_node *node;
1245 ovs_assert(backer->refcount > 0);
1247 if (--backer->refcount) {
1251 drop_key_clear(backer);
1252 hmap_destroy(&backer->drop_keys);
1254 simap_destroy(&backer->tnl_backers);
1255 hmap_destroy(&backer->odp_to_ofport_map);
1256 node = shash_find(&all_dpif_backers, backer->type);
1258 shash_delete(&all_dpif_backers, node);
1259 dpif_close(backer->dpif);
1264 /* Datapath port slated for removal from datapath. */
1265 struct odp_garbage {
1266 struct list list_node;
1271 open_dpif_backer(const char *type, struct dpif_backer **backerp)
1273 struct dpif_backer *backer;
1274 struct dpif_port_dump port_dump;
1275 struct dpif_port port;
1276 struct shash_node *node;
1277 struct list garbage_list;
1278 struct odp_garbage *garbage, *next;
1284 backer = shash_find_data(&all_dpif_backers, type);
1291 backer_name = xasprintf("ovs-%s", type);
1293 /* Remove any existing datapaths, since we assume we're the only
1294 * userspace controlling the datapath. */
1296 dp_enumerate_names(type, &names);
1297 SSET_FOR_EACH(name, &names) {
1298 struct dpif *old_dpif;
1300 /* Don't remove our backer if it exists. */
1301 if (!strcmp(name, backer_name)) {
1305 if (dpif_open(name, type, &old_dpif)) {
1306 VLOG_WARN("couldn't open old datapath %s to remove it", name);
1308 dpif_delete(old_dpif);
1309 dpif_close(old_dpif);
1312 sset_destroy(&names);
1314 backer = xmalloc(sizeof *backer);
1316 error = dpif_create_and_open(backer_name, type, &backer->dpif);
1319 VLOG_ERR("failed to open datapath of type %s: %s", type,
1325 backer->type = xstrdup(type);
1326 backer->refcount = 1;
1327 hmap_init(&backer->odp_to_ofport_map);
1328 hmap_init(&backer->drop_keys);
1329 timer_set_duration(&backer->next_expiration, 1000);
1330 backer->need_revalidate = 0;
1331 simap_init(&backer->tnl_backers);
1332 tag_set_init(&backer->revalidate_set);
1333 backer->recv_set_enable = !ofproto_get_flow_restore_wait();
1336 if (backer->recv_set_enable) {
1337 dpif_flow_flush(backer->dpif);
1340 /* Loop through the ports already on the datapath and remove any
1341 * that we don't need anymore. */
1342 list_init(&garbage_list);
1343 dpif_port_dump_start(&port_dump, backer->dpif);
1344 while (dpif_port_dump_next(&port_dump, &port)) {
1345 node = shash_find(&init_ofp_ports, port.name);
1346 if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
1347 garbage = xmalloc(sizeof *garbage);
1348 garbage->odp_port = port.port_no;
1349 list_push_front(&garbage_list, &garbage->list_node);
1352 dpif_port_dump_done(&port_dump);
1354 LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) {
1355 dpif_port_del(backer->dpif, garbage->odp_port);
1356 list_remove(&garbage->list_node);
1360 shash_add(&all_dpif_backers, type, backer);
1362 error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
1364 VLOG_ERR("failed to listen on datapath of type %s: %s",
1365 type, strerror(error));
1366 close_dpif_backer(backer);
1374 construct(struct ofproto *ofproto_)
1376 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1377 struct shash_node *node, *next;
1382 error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1387 max_ports = dpif_get_max_ports(ofproto->backer->dpif);
1388 ofproto_init_max_ports(ofproto_, MIN(max_ports, OFPP_MAX));
1390 ofproto->netflow = NULL;
1391 ofproto->sflow = NULL;
1392 ofproto->ipfix = NULL;
1393 ofproto->stp = NULL;
1394 hmap_init(&ofproto->bundles);
1395 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1396 for (i = 0; i < MAX_MIRRORS; i++) {
1397 ofproto->mirrors[i] = NULL;
1399 ofproto->has_bonded_bundles = false;
1401 classifier_init(&ofproto->facets);
1402 hmap_init(&ofproto->subfacets);
1403 ofproto->governor = NULL;
1404 ofproto->consistency_rl = LLONG_MIN;
1406 for (i = 0; i < N_TABLES; i++) {
1407 struct table_dpif *table = &ofproto->tables[i];
1409 table->catchall_table = NULL;
1410 table->other_table = NULL;
1411 table->basis = random_uint32();
1414 list_init(&ofproto->completions);
1416 ofproto_dpif_unixctl_init();
1418 ofproto->has_mirrors = false;
1419 ofproto->has_bundle_action = false;
1421 hmap_init(&ofproto->vlandev_map);
1422 hmap_init(&ofproto->realdev_vid_map);
1424 sset_init(&ofproto->ports);
1425 sset_init(&ofproto->ghost_ports);
1426 sset_init(&ofproto->port_poll_set);
1427 ofproto->port_poll_errno = 0;
1429 SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1430 struct iface_hint *iface_hint = node->data;
1432 if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1433 /* Check if the datapath already has this port. */
1434 if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1435 sset_add(&ofproto->ports, node->name);
1438 free(iface_hint->br_name);
1439 free(iface_hint->br_type);
1441 shash_delete(&init_ofp_ports, node);
1445 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1446 hash_string(ofproto->up.name, 0));
1447 memset(&ofproto->stats, 0, sizeof ofproto->stats);
1449 ofproto_init_tables(ofproto_, N_TABLES);
1450 error = add_internal_flows(ofproto);
1451 ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1454 ofproto->n_missed = 0;
1456 ofproto->max_n_subfacet = 0;
1457 ofproto->created = time_msec();
1458 ofproto->last_minute = ofproto->created;
1459 memset(&ofproto->hourly, 0, sizeof ofproto->hourly);
1460 memset(&ofproto->daily, 0, sizeof ofproto->daily);
1461 ofproto->subfacet_add_count = 0;
1462 ofproto->subfacet_del_count = 0;
1463 ofproto->total_subfacet_add_count = 0;
1464 ofproto->total_subfacet_del_count = 0;
1465 ofproto->total_subfacet_life_span = 0;
1466 ofproto->total_subfacet_count = 0;
1467 ofproto->n_update_stats = 0;
1473 add_internal_flow(struct ofproto_dpif *ofproto, int id,
1474 const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1476 struct ofputil_flow_mod fm;
1479 match_init_catchall(&fm.match);
1481 match_set_reg(&fm.match, 0, id);
1482 fm.new_cookie = htonll(0);
1483 fm.cookie = htonll(0);
1484 fm.cookie_mask = htonll(0);
1485 fm.table_id = TBL_INTERNAL;
1486 fm.command = OFPFC_ADD;
1487 fm.idle_timeout = 0;
1488 fm.hard_timeout = 0;
1492 fm.ofpacts = ofpacts->data;
1493 fm.ofpacts_len = ofpacts->size;
1495 error = ofproto_flow_mod(&ofproto->up, &fm);
1497 VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
1498 id, ofperr_to_string(error));
1502 *rulep = rule_dpif_lookup__(ofproto, &fm.match.flow, NULL, TBL_INTERNAL);
1503 ovs_assert(*rulep != NULL);
1509 add_internal_flows(struct ofproto_dpif *ofproto)
1511 struct ofpact_controller *controller;
1512 uint64_t ofpacts_stub[128 / 8];
1513 struct ofpbuf ofpacts;
1517 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1520 controller = ofpact_put_CONTROLLER(&ofpacts);
1521 controller->max_len = UINT16_MAX;
1522 controller->controller_id = 0;
1523 controller->reason = OFPR_NO_MATCH;
1524 ofpact_pad(&ofpacts);
1526 error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule);
1531 ofpbuf_clear(&ofpacts);
1532 error = add_internal_flow(ofproto, id++, &ofpacts,
1533 &ofproto->no_packet_in_rule);
1538 error = add_internal_flow(ofproto, id++, &ofpacts,
1539 &ofproto->drop_frags_rule);
1544 complete_operations(struct ofproto_dpif *ofproto)
1546 struct dpif_completion *c, *next;
1548 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
1549 ofoperation_complete(c->op, 0);
1550 list_remove(&c->list_node);
1556 destruct(struct ofproto *ofproto_)
1558 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1559 struct rule_dpif *rule, *next_rule;
1560 struct oftable *table;
1563 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1564 complete_operations(ofproto);
1566 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1567 struct cls_cursor cursor;
1569 cls_cursor_init(&cursor, &table->cls, NULL);
1570 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1571 ofproto_rule_destroy(&rule->up);
1575 for (i = 0; i < MAX_MIRRORS; i++) {
1576 mirror_destroy(ofproto->mirrors[i]);
1579 netflow_destroy(ofproto->netflow);
1580 dpif_sflow_destroy(ofproto->sflow);
1581 hmap_destroy(&ofproto->bundles);
1582 mac_learning_destroy(ofproto->ml);
1584 classifier_destroy(&ofproto->facets);
1585 hmap_destroy(&ofproto->subfacets);
1586 governor_destroy(ofproto->governor);
1588 hmap_destroy(&ofproto->vlandev_map);
1589 hmap_destroy(&ofproto->realdev_vid_map);
1591 sset_destroy(&ofproto->ports);
1592 sset_destroy(&ofproto->ghost_ports);
1593 sset_destroy(&ofproto->port_poll_set);
1595 close_dpif_backer(ofproto->backer);
1599 run_fast(struct ofproto *ofproto_)
1601 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1602 struct ofport_dpif *ofport;
1604 /* Do not perform any periodic activity required by 'ofproto' while
1605 * waiting for flow restore to complete. */
1606 if (ofproto_get_flow_restore_wait()) {
1610 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1611 port_run_fast(ofport);
1618 run(struct ofproto *ofproto_)
1620 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1621 struct ofport_dpif *ofport;
1622 struct ofbundle *bundle;
1626 complete_operations(ofproto);
1629 /* Do not perform any periodic activity below required by 'ofproto' while
1630 * waiting for flow restore to complete. */
1631 if (ofproto_get_flow_restore_wait()) {
1635 error = run_fast(ofproto_);
1640 if (ofproto->netflow) {
1641 if (netflow_run(ofproto->netflow)) {
1642 send_netflow_active_timeouts(ofproto);
1645 if (ofproto->sflow) {
1646 dpif_sflow_run(ofproto->sflow);
1649 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1652 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1657 mac_learning_run(ofproto->ml, &ofproto->backer->revalidate_set);
1659 /* Check the consistency of a random facet, to aid debugging. */
1660 if (time_msec() >= ofproto->consistency_rl
1661 && !classifier_is_empty(&ofproto->facets)
1662 && !ofproto->backer->need_revalidate) {
1663 struct cls_table *table;
1664 struct cls_rule *cr;
1665 struct facet *facet;
1667 ofproto->consistency_rl = time_msec() + 250;
1669 table = CONTAINER_OF(hmap_random_node(&ofproto->facets.tables),
1670 struct cls_table, hmap_node);
1671 cr = CONTAINER_OF(hmap_random_node(&table->rules), struct cls_rule,
1673 facet = CONTAINER_OF(cr, struct facet, cr);
1675 if (!tag_set_intersects(&ofproto->backer->revalidate_set,
1676 facet->xout.tags)) {
1677 if (!facet_check_consistency(facet)) {
1678 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
1683 if (ofproto->governor) {
1686 governor_run(ofproto->governor);
1688 /* If the governor has shrunk to its minimum size and the number of
1689 * subfacets has dwindled, then drop the governor entirely.
1691 * For hysteresis, the number of subfacets to drop the governor is
1692 * smaller than the number needed to trigger its creation. */
1693 n_subfacets = hmap_count(&ofproto->subfacets);
1694 if (n_subfacets * 4 < ofproto->up.flow_eviction_threshold
1695 && governor_is_idle(ofproto->governor)) {
1696 governor_destroy(ofproto->governor);
1697 ofproto->governor = NULL;
1705 wait(struct ofproto *ofproto_)
1707 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1708 struct ofport_dpif *ofport;
1709 struct ofbundle *bundle;
1711 if (!clogged && !list_is_empty(&ofproto->completions)) {
1712 poll_immediate_wake();
1715 if (ofproto_get_flow_restore_wait()) {
1719 dpif_wait(ofproto->backer->dpif);
1720 dpif_recv_wait(ofproto->backer->dpif);
1721 if (ofproto->sflow) {
1722 dpif_sflow_wait(ofproto->sflow);
1724 if (!tag_set_is_empty(&ofproto->backer->revalidate_set)) {
1725 poll_immediate_wake();
1727 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1730 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1731 bundle_wait(bundle);
1733 if (ofproto->netflow) {
1734 netflow_wait(ofproto->netflow);
1736 mac_learning_wait(ofproto->ml);
1738 if (ofproto->backer->need_revalidate) {
1739 /* Shouldn't happen, but if it does just go around again. */
1740 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1741 poll_immediate_wake();
1743 if (ofproto->governor) {
1744 governor_wait(ofproto->governor);
1749 get_memory_usage(const struct ofproto *ofproto_, struct simap *usage)
1751 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1753 simap_increase(usage, "facets", classifier_count(&ofproto->facets));
1754 simap_increase(usage, "subfacets", hmap_count(&ofproto->subfacets));
1758 flush(struct ofproto *ofproto_)
1760 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1761 struct subfacet *subfacet, *next_subfacet;
1762 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
1766 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
1767 &ofproto->subfacets) {
1768 if (subfacet->path != SF_NOT_INSTALLED) {
1769 batch[n_batch++] = subfacet;
1770 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
1771 subfacet_destroy_batch(ofproto, batch, n_batch);
1775 subfacet_destroy(subfacet);
1780 subfacet_destroy_batch(ofproto, batch, n_batch);
1785 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1786 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1788 *arp_match_ip = true;
1789 *actions = (OFPUTIL_A_OUTPUT |
1790 OFPUTIL_A_SET_VLAN_VID |
1791 OFPUTIL_A_SET_VLAN_PCP |
1792 OFPUTIL_A_STRIP_VLAN |
1793 OFPUTIL_A_SET_DL_SRC |
1794 OFPUTIL_A_SET_DL_DST |
1795 OFPUTIL_A_SET_NW_SRC |
1796 OFPUTIL_A_SET_NW_DST |
1797 OFPUTIL_A_SET_NW_TOS |
1798 OFPUTIL_A_SET_TP_SRC |
1799 OFPUTIL_A_SET_TP_DST |
1804 get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots)
1806 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1807 struct dpif_dp_stats s;
1808 uint64_t n_miss, n_no_pkt_in, n_bytes, n_dropped_frags;
1811 strcpy(ots->name, "classifier");
1813 dpif_get_dp_stats(ofproto->backer->dpif, &s);
1814 rule_get_stats(&ofproto->miss_rule->up, &n_miss, &n_bytes);
1815 rule_get_stats(&ofproto->no_packet_in_rule->up, &n_no_pkt_in, &n_bytes);
1816 rule_get_stats(&ofproto->drop_frags_rule->up, &n_dropped_frags, &n_bytes);
1818 n_lookup = s.n_hit + s.n_missed - n_dropped_frags;
1819 ots->lookup_count = htonll(n_lookup);
1820 ots->matched_count = htonll(n_lookup - n_miss - n_no_pkt_in);
1823 static struct ofport *
1826 struct ofport_dpif *port = xmalloc(sizeof *port);
1831 port_dealloc(struct ofport *port_)
1833 struct ofport_dpif *port = ofport_dpif_cast(port_);
1838 port_construct(struct ofport *port_)
1840 struct ofport_dpif *port = ofport_dpif_cast(port_);
1841 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1842 const struct netdev *netdev = port->up.netdev;
1843 struct dpif_port dpif_port;
1846 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1847 port->bundle = NULL;
1849 port->tag = tag_create_random();
1850 port->may_enable = true;
1851 port->stp_port = NULL;
1852 port->stp_state = STP_DISABLED;
1853 port->tnl_port = NULL;
1854 hmap_init(&port->priorities);
1855 port->realdev_ofp_port = 0;
1856 port->vlandev_vid = 0;
1857 port->carrier_seq = netdev_get_carrier_resets(netdev);
1859 if (netdev_vport_is_patch(netdev)) {
1860 /* By bailing out here, we don't submit the port to the sFlow module
1861 * to be considered for counter polling export. This is correct
1862 * because the patch port represents an interface that sFlow considers
1863 * to be "internal" to the switch as a whole, and therefore not an
1864 * candidate for counter polling. */
1865 port->odp_port = OVSP_NONE;
1869 error = dpif_port_query_by_name(ofproto->backer->dpif,
1870 netdev_vport_get_dpif_port(netdev),
1876 port->odp_port = dpif_port.port_no;
1878 if (netdev_get_tunnel_config(netdev)) {
1879 port->tnl_port = tnl_port_add(&port->up, port->odp_port);
1881 /* Sanity-check that a mapping doesn't already exist. This
1882 * shouldn't happen for non-tunnel ports. */
1883 if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1884 VLOG_ERR("port %s already has an OpenFlow port number",
1886 dpif_port_destroy(&dpif_port);
1890 hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1891 hash_int(port->odp_port, 0));
1893 dpif_port_destroy(&dpif_port);
1895 if (ofproto->sflow) {
1896 dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1903 port_destruct(struct ofport *port_)
1905 struct ofport_dpif *port = ofport_dpif_cast(port_);
1906 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1907 const char *dp_port_name = netdev_vport_get_dpif_port(port->up.netdev);
1908 const char *devname = netdev_get_name(port->up.netdev);
1910 if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1911 /* The underlying device is still there, so delete it. This
1912 * happens when the ofproto is being destroyed, since the caller
1913 * assumes that removal of attached ports will happen as part of
1915 if (!port->tnl_port) {
1916 dpif_port_del(ofproto->backer->dpif, port->odp_port);
1918 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1921 if (port->odp_port != OVSP_NONE && !port->tnl_port) {
1922 hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1925 tnl_port_del(port->tnl_port);
1926 sset_find_and_delete(&ofproto->ports, devname);
1927 sset_find_and_delete(&ofproto->ghost_ports, devname);
1928 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1929 bundle_remove(port_);
1930 set_cfm(port_, NULL);
1931 if (ofproto->sflow) {
1932 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1935 ofport_clear_priorities(port);
1936 hmap_destroy(&port->priorities);
1940 port_modified(struct ofport *port_)
1942 struct ofport_dpif *port = ofport_dpif_cast(port_);
1944 if (port->bundle && port->bundle->bond) {
1945 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1949 cfm_set_netdev(port->cfm, port->up.netdev);
1954 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1956 struct ofport_dpif *port = ofport_dpif_cast(port_);
1957 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1958 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1960 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1961 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1962 OFPUTIL_PC_NO_PACKET_IN)) {
1963 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1965 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1966 bundle_update(port->bundle);
1972 set_sflow(struct ofproto *ofproto_,
1973 const struct ofproto_sflow_options *sflow_options)
1975 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1976 struct dpif_sflow *ds = ofproto->sflow;
1978 if (sflow_options) {
1980 struct ofport_dpif *ofport;
1982 ds = ofproto->sflow = dpif_sflow_create();
1983 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1984 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1986 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1988 dpif_sflow_set_options(ds, sflow_options);
1991 dpif_sflow_destroy(ds);
1992 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1993 ofproto->sflow = NULL;
2001 struct ofproto *ofproto_,
2002 const struct ofproto_ipfix_bridge_exporter_options *bridge_exporter_options,
2003 const struct ofproto_ipfix_flow_exporter_options *flow_exporters_options,
2004 size_t n_flow_exporters_options)
2006 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2007 struct dpif_ipfix *di = ofproto->ipfix;
2009 if (bridge_exporter_options || flow_exporters_options) {
2011 di = ofproto->ipfix = dpif_ipfix_create();
2013 dpif_ipfix_set_options(
2014 di, bridge_exporter_options, flow_exporters_options,
2015 n_flow_exporters_options);
2018 dpif_ipfix_destroy(di);
2019 ofproto->ipfix = NULL;
2026 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
2028 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2035 struct ofproto_dpif *ofproto;
2037 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2038 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2039 ofport->cfm = cfm_create(ofport->up.netdev);
2042 if (cfm_configure(ofport->cfm, s)) {
2048 cfm_destroy(ofport->cfm);
2054 get_cfm_status(const struct ofport *ofport_,
2055 struct ofproto_cfm_status *status)
2057 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2060 status->faults = cfm_get_fault(ofport->cfm);
2061 status->remote_opstate = cfm_get_opup(ofport->cfm);
2062 status->health = cfm_get_health(ofport->cfm);
2063 cfm_get_remote_mpids(ofport->cfm, &status->rmps, &status->n_rmps);
2070 /* Spanning Tree. */
2073 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
2075 struct ofproto_dpif *ofproto = ofproto_;
2076 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
2077 struct ofport_dpif *ofport;
2079 ofport = stp_port_get_aux(sp);
2081 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
2082 ofproto->up.name, port_num);
2084 struct eth_header *eth = pkt->l2;
2086 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
2087 if (eth_addr_is_zero(eth->eth_src)) {
2088 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
2089 "with unknown MAC", ofproto->up.name, port_num);
2091 send_packet(ofport, pkt);
2097 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
2099 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
2101 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2103 /* Only revalidate flows if the configuration changed. */
2104 if (!s != !ofproto->stp) {
2105 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2109 if (!ofproto->stp) {
2110 ofproto->stp = stp_create(ofproto_->name, s->system_id,
2111 send_bpdu_cb, ofproto);
2112 ofproto->stp_last_tick = time_msec();
2115 stp_set_bridge_id(ofproto->stp, s->system_id);
2116 stp_set_bridge_priority(ofproto->stp, s->priority);
2117 stp_set_hello_time(ofproto->stp, s->hello_time);
2118 stp_set_max_age(ofproto->stp, s->max_age);
2119 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
2121 struct ofport *ofport;
2123 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
2124 set_stp_port(ofport, NULL);
2127 stp_destroy(ofproto->stp);
2128 ofproto->stp = NULL;
2135 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
2137 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2141 s->bridge_id = stp_get_bridge_id(ofproto->stp);
2142 s->designated_root = stp_get_designated_root(ofproto->stp);
2143 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
2152 update_stp_port_state(struct ofport_dpif *ofport)
2154 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2155 enum stp_state state;
2157 /* Figure out new state. */
2158 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
2162 if (ofport->stp_state != state) {
2163 enum ofputil_port_state of_state;
2166 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
2167 netdev_get_name(ofport->up.netdev),
2168 stp_state_name(ofport->stp_state),
2169 stp_state_name(state));
2170 if (stp_learn_in_state(ofport->stp_state)
2171 != stp_learn_in_state(state)) {
2172 /* xxx Learning action flows should also be flushed. */
2173 mac_learning_flush(ofproto->ml,
2174 &ofproto->backer->revalidate_set);
2176 fwd_change = stp_forward_in_state(ofport->stp_state)
2177 != stp_forward_in_state(state);
2179 ofproto->backer->need_revalidate = REV_STP;
2180 ofport->stp_state = state;
2181 ofport->stp_state_entered = time_msec();
2183 if (fwd_change && ofport->bundle) {
2184 bundle_update(ofport->bundle);
2187 /* Update the STP state bits in the OpenFlow port description. */
2188 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
2189 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
2190 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
2191 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
2192 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
2194 ofproto_port_set_state(&ofport->up, of_state);
2198 /* Configures STP on 'ofport_' using the settings defined in 's'. The
2199 * caller is responsible for assigning STP port numbers and ensuring
2200 * there are no duplicates. */
2202 set_stp_port(struct ofport *ofport_,
2203 const struct ofproto_port_stp_settings *s)
2205 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2206 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2207 struct stp_port *sp = ofport->stp_port;
2209 if (!s || !s->enable) {
2211 ofport->stp_port = NULL;
2212 stp_port_disable(sp);
2213 update_stp_port_state(ofport);
2216 } else if (sp && stp_port_no(sp) != s->port_num
2217 && ofport == stp_port_get_aux(sp)) {
2218 /* The port-id changed, so disable the old one if it's not
2219 * already in use by another port. */
2220 stp_port_disable(sp);
2223 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
2224 stp_port_enable(sp);
2226 stp_port_set_aux(sp, ofport);
2227 stp_port_set_priority(sp, s->priority);
2228 stp_port_set_path_cost(sp, s->path_cost);
2230 update_stp_port_state(ofport);
2236 get_stp_port_status(struct ofport *ofport_,
2237 struct ofproto_port_stp_status *s)
2239 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2240 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2241 struct stp_port *sp = ofport->stp_port;
2243 if (!ofproto->stp || !sp) {
2249 s->port_id = stp_port_get_id(sp);
2250 s->state = stp_port_get_state(sp);
2251 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
2252 s->role = stp_port_get_role(sp);
2253 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
2259 stp_run(struct ofproto_dpif *ofproto)
2262 long long int now = time_msec();
2263 long long int elapsed = now - ofproto->stp_last_tick;
2264 struct stp_port *sp;
2267 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
2268 ofproto->stp_last_tick = now;
2270 while (stp_get_changed_port(ofproto->stp, &sp)) {
2271 struct ofport_dpif *ofport = stp_port_get_aux(sp);
2274 update_stp_port_state(ofport);
2278 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
2279 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
2285 stp_wait(struct ofproto_dpif *ofproto)
2288 poll_timer_wait(1000);
2292 /* Returns true if STP should process 'flow'. Sets fields in 'wc' that
2293 * were used to make the determination.*/
2295 stp_should_process_flow(const struct flow *flow, struct flow_wildcards *wc)
2297 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
2298 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
2302 stp_process_packet(const struct ofport_dpif *ofport,
2303 const struct ofpbuf *packet)
2305 struct ofpbuf payload = *packet;
2306 struct eth_header *eth = payload.data;
2307 struct stp_port *sp = ofport->stp_port;
2309 /* Sink packets on ports that have STP disabled when the bridge has
2311 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
2315 /* Trim off padding on payload. */
2316 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
2317 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
2320 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
2321 stp_received_bpdu(sp, payload.data, payload.size);
2325 static struct priority_to_dscp *
2326 get_priority(const struct ofport_dpif *ofport, uint32_t priority)
2328 struct priority_to_dscp *pdscp;
2331 hash = hash_int(priority, 0);
2332 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) {
2333 if (pdscp->priority == priority) {
2341 ofport_clear_priorities(struct ofport_dpif *ofport)
2343 struct priority_to_dscp *pdscp, *next;
2345 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) {
2346 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
2352 set_queues(struct ofport *ofport_,
2353 const struct ofproto_port_queue *qdscp_list,
2356 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2357 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2358 struct hmap new = HMAP_INITIALIZER(&new);
2361 for (i = 0; i < n_qdscp; i++) {
2362 struct priority_to_dscp *pdscp;
2366 dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
2367 if (dpif_queue_to_priority(ofproto->backer->dpif, qdscp_list[i].queue,
2372 pdscp = get_priority(ofport, priority);
2374 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
2376 pdscp = xmalloc(sizeof *pdscp);
2377 pdscp->priority = priority;
2379 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2382 if (pdscp->dscp != dscp) {
2384 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2387 hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0));
2390 if (!hmap_is_empty(&ofport->priorities)) {
2391 ofport_clear_priorities(ofport);
2392 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2395 hmap_swap(&new, &ofport->priorities);
2403 /* Expires all MAC learning entries associated with 'bundle' and forces its
2404 * ofproto to revalidate every flow.
2406 * Normally MAC learning entries are removed only from the ofproto associated
2407 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2408 * are removed from every ofproto. When patch ports and SLB bonds are in use
2409 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2410 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2411 * with the host from which it migrated. */
2413 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2415 struct ofproto_dpif *ofproto = bundle->ofproto;
2416 struct mac_learning *ml = ofproto->ml;
2417 struct mac_entry *mac, *next_mac;
2419 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2420 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2421 if (mac->port.p == bundle) {
2423 struct ofproto_dpif *o;
2425 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2427 struct mac_entry *e;
2429 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan,
2432 mac_learning_expire(o->ml, e);
2438 mac_learning_expire(ml, mac);
2443 static struct ofbundle *
2444 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2446 struct ofbundle *bundle;
2448 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2449 &ofproto->bundles) {
2450 if (bundle->aux == aux) {
2457 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
2458 * ones that are found to 'bundles'. */
2460 bundle_lookup_multiple(struct ofproto_dpif *ofproto,
2461 void **auxes, size_t n_auxes,
2462 struct hmapx *bundles)
2466 hmapx_init(bundles);
2467 for (i = 0; i < n_auxes; i++) {
2468 struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
2470 hmapx_add(bundles, bundle);
2476 bundle_update(struct ofbundle *bundle)
2478 struct ofport_dpif *port;
2480 bundle->floodable = true;
2481 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2482 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2483 || !stp_forward_in_state(port->stp_state)) {
2484 bundle->floodable = false;
2491 bundle_del_port(struct ofport_dpif *port)
2493 struct ofbundle *bundle = port->bundle;
2495 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2497 list_remove(&port->bundle_node);
2498 port->bundle = NULL;
2501 lacp_slave_unregister(bundle->lacp, port);
2504 bond_slave_unregister(bundle->bond, port);
2507 bundle_update(bundle);
2511 bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port,
2512 struct lacp_slave_settings *lacp)
2514 struct ofport_dpif *port;
2516 port = get_ofp_port(bundle->ofproto, ofp_port);
2521 if (port->bundle != bundle) {
2522 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2524 bundle_del_port(port);
2527 port->bundle = bundle;
2528 list_push_back(&bundle->ports, &port->bundle_node);
2529 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2530 || !stp_forward_in_state(port->stp_state)) {
2531 bundle->floodable = false;
2535 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2536 lacp_slave_register(bundle->lacp, port, lacp);
2543 bundle_destroy(struct ofbundle *bundle)
2545 struct ofproto_dpif *ofproto;
2546 struct ofport_dpif *port, *next_port;
2553 ofproto = bundle->ofproto;
2554 for (i = 0; i < MAX_MIRRORS; i++) {
2555 struct ofmirror *m = ofproto->mirrors[i];
2557 if (m->out == bundle) {
2559 } else if (hmapx_find_and_delete(&m->srcs, bundle)
2560 || hmapx_find_and_delete(&m->dsts, bundle)) {
2561 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2566 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2567 bundle_del_port(port);
2570 bundle_flush_macs(bundle, true);
2571 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2573 free(bundle->trunks);
2574 lacp_destroy(bundle->lacp);
2575 bond_destroy(bundle->bond);
2580 bundle_set(struct ofproto *ofproto_, void *aux,
2581 const struct ofproto_bundle_settings *s)
2583 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2584 bool need_flush = false;
2585 struct ofport_dpif *port;
2586 struct ofbundle *bundle;
2587 unsigned long *trunks;
2593 bundle_destroy(bundle_lookup(ofproto, aux));
2597 ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2598 ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2600 bundle = bundle_lookup(ofproto, aux);
2602 bundle = xmalloc(sizeof *bundle);
2604 bundle->ofproto = ofproto;
2605 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2606 hash_pointer(aux, 0));
2608 bundle->name = NULL;
2610 list_init(&bundle->ports);
2611 bundle->vlan_mode = PORT_VLAN_TRUNK;
2613 bundle->trunks = NULL;
2614 bundle->use_priority_tags = s->use_priority_tags;
2615 bundle->lacp = NULL;
2616 bundle->bond = NULL;
2618 bundle->floodable = true;
2620 bundle->src_mirrors = 0;
2621 bundle->dst_mirrors = 0;
2622 bundle->mirror_out = 0;
2625 if (!bundle->name || strcmp(s->name, bundle->name)) {
2627 bundle->name = xstrdup(s->name);
2632 if (!bundle->lacp) {
2633 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2634 bundle->lacp = lacp_create();
2636 lacp_configure(bundle->lacp, s->lacp);
2638 lacp_destroy(bundle->lacp);
2639 bundle->lacp = NULL;
2642 /* Update set of ports. */
2644 for (i = 0; i < s->n_slaves; i++) {
2645 if (!bundle_add_port(bundle, s->slaves[i],
2646 s->lacp ? &s->lacp_slaves[i] : NULL)) {
2650 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2651 struct ofport_dpif *next_port;
2653 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2654 for (i = 0; i < s->n_slaves; i++) {
2655 if (s->slaves[i] == port->up.ofp_port) {
2660 bundle_del_port(port);
2664 ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2666 if (list_is_empty(&bundle->ports)) {
2667 bundle_destroy(bundle);
2671 /* Set VLAN tagging mode */
2672 if (s->vlan_mode != bundle->vlan_mode
2673 || s->use_priority_tags != bundle->use_priority_tags) {
2674 bundle->vlan_mode = s->vlan_mode;
2675 bundle->use_priority_tags = s->use_priority_tags;
2680 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2681 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2683 if (vlan != bundle->vlan) {
2684 bundle->vlan = vlan;
2688 /* Get trunked VLANs. */
2689 switch (s->vlan_mode) {
2690 case PORT_VLAN_ACCESS:
2694 case PORT_VLAN_TRUNK:
2695 trunks = CONST_CAST(unsigned long *, s->trunks);
2698 case PORT_VLAN_NATIVE_UNTAGGED:
2699 case PORT_VLAN_NATIVE_TAGGED:
2700 if (vlan != 0 && (!s->trunks
2701 || !bitmap_is_set(s->trunks, vlan)
2702 || bitmap_is_set(s->trunks, 0))) {
2703 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2705 trunks = bitmap_clone(s->trunks, 4096);
2707 trunks = bitmap_allocate1(4096);
2709 bitmap_set1(trunks, vlan);
2710 bitmap_set0(trunks, 0);
2712 trunks = CONST_CAST(unsigned long *, s->trunks);
2719 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2720 free(bundle->trunks);
2721 if (trunks == s->trunks) {
2722 bundle->trunks = vlan_bitmap_clone(trunks);
2724 bundle->trunks = trunks;
2729 if (trunks != s->trunks) {
2734 if (!list_is_short(&bundle->ports)) {
2735 bundle->ofproto->has_bonded_bundles = true;
2737 if (bond_reconfigure(bundle->bond, s->bond)) {
2738 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2741 bundle->bond = bond_create(s->bond);
2742 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2745 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2746 bond_slave_register(bundle->bond, port, port->up.netdev);
2749 bond_destroy(bundle->bond);
2750 bundle->bond = NULL;
2753 /* If we changed something that would affect MAC learning, un-learn
2754 * everything on this port and force flow revalidation. */
2756 bundle_flush_macs(bundle, false);
2763 bundle_remove(struct ofport *port_)
2765 struct ofport_dpif *port = ofport_dpif_cast(port_);
2766 struct ofbundle *bundle = port->bundle;
2769 bundle_del_port(port);
2770 if (list_is_empty(&bundle->ports)) {
2771 bundle_destroy(bundle);
2772 } else if (list_is_short(&bundle->ports)) {
2773 bond_destroy(bundle->bond);
2774 bundle->bond = NULL;
2780 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2782 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2783 struct ofport_dpif *port = port_;
2784 uint8_t ea[ETH_ADDR_LEN];
2787 error = netdev_get_etheraddr(port->up.netdev, ea);
2789 struct ofpbuf packet;
2792 ofpbuf_init(&packet, 0);
2793 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2795 memcpy(packet_pdu, pdu, pdu_size);
2797 send_packet(port, &packet);
2798 ofpbuf_uninit(&packet);
2800 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2801 "%s (%s)", port->bundle->name,
2802 netdev_get_name(port->up.netdev), strerror(error));
2807 bundle_send_learning_packets(struct ofbundle *bundle)
2809 struct ofproto_dpif *ofproto = bundle->ofproto;
2810 int error, n_packets, n_errors;
2811 struct mac_entry *e;
2813 error = n_packets = n_errors = 0;
2814 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2815 if (e->port.p != bundle) {
2816 struct ofpbuf *learning_packet;
2817 struct ofport_dpif *port;
2821 /* The assignment to "port" is unnecessary but makes "grep"ing for
2822 * struct ofport_dpif more effective. */
2823 learning_packet = bond_compose_learning_packet(bundle->bond,
2827 ret = send_packet(port, learning_packet);
2828 ofpbuf_delete(learning_packet);
2838 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2839 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2840 "packets, last error was: %s",
2841 bundle->name, n_errors, n_packets, strerror(error));
2843 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2844 bundle->name, n_packets);
2849 bundle_run(struct ofbundle *bundle)
2852 lacp_run(bundle->lacp, send_pdu_cb);
2855 struct ofport_dpif *port;
2857 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2858 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2861 bond_run(bundle->bond, &bundle->ofproto->backer->revalidate_set,
2862 lacp_status(bundle->lacp));
2863 if (bond_should_send_learning_packets(bundle->bond)) {
2864 bundle_send_learning_packets(bundle);
2870 bundle_wait(struct ofbundle *bundle)
2873 lacp_wait(bundle->lacp);
2876 bond_wait(bundle->bond);
2883 mirror_scan(struct ofproto_dpif *ofproto)
2887 for (idx = 0; idx < MAX_MIRRORS; idx++) {
2888 if (!ofproto->mirrors[idx]) {
2895 static struct ofmirror *
2896 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
2900 for (i = 0; i < MAX_MIRRORS; i++) {
2901 struct ofmirror *mirror = ofproto->mirrors[i];
2902 if (mirror && mirror->aux == aux) {
2910 /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */
2912 mirror_update_dups(struct ofproto_dpif *ofproto)
2916 for (i = 0; i < MAX_MIRRORS; i++) {
2917 struct ofmirror *m = ofproto->mirrors[i];
2920 m->dup_mirrors = MIRROR_MASK_C(1) << i;
2924 for (i = 0; i < MAX_MIRRORS; i++) {
2925 struct ofmirror *m1 = ofproto->mirrors[i];
2932 for (j = i + 1; j < MAX_MIRRORS; j++) {
2933 struct ofmirror *m2 = ofproto->mirrors[j];
2935 if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) {
2936 m1->dup_mirrors |= MIRROR_MASK_C(1) << j;
2937 m2->dup_mirrors |= m1->dup_mirrors;
2944 mirror_set(struct ofproto *ofproto_, void *aux,
2945 const struct ofproto_mirror_settings *s)
2947 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2948 mirror_mask_t mirror_bit;
2949 struct ofbundle *bundle;
2950 struct ofmirror *mirror;
2951 struct ofbundle *out;
2952 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
2953 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
2956 mirror = mirror_lookup(ofproto, aux);
2958 mirror_destroy(mirror);
2964 idx = mirror_scan(ofproto);
2966 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
2968 ofproto->up.name, MAX_MIRRORS, s->name);
2972 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
2973 mirror->ofproto = ofproto;
2976 mirror->out_vlan = -1;
2977 mirror->name = NULL;
2980 if (!mirror->name || strcmp(s->name, mirror->name)) {
2982 mirror->name = xstrdup(s->name);
2985 /* Get the new configuration. */
2986 if (s->out_bundle) {
2987 out = bundle_lookup(ofproto, s->out_bundle);
2989 mirror_destroy(mirror);
2995 out_vlan = s->out_vlan;
2997 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
2998 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
3000 /* If the configuration has not changed, do nothing. */
3001 if (hmapx_equals(&srcs, &mirror->srcs)
3002 && hmapx_equals(&dsts, &mirror->dsts)
3003 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
3004 && mirror->out == out
3005 && mirror->out_vlan == out_vlan)
3007 hmapx_destroy(&srcs);
3008 hmapx_destroy(&dsts);
3012 hmapx_swap(&srcs, &mirror->srcs);
3013 hmapx_destroy(&srcs);
3015 hmapx_swap(&dsts, &mirror->dsts);
3016 hmapx_destroy(&dsts);
3018 free(mirror->vlans);
3019 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
3022 mirror->out_vlan = out_vlan;
3024 /* Update bundles. */
3025 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
3026 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
3027 if (hmapx_contains(&mirror->srcs, bundle)) {
3028 bundle->src_mirrors |= mirror_bit;
3030 bundle->src_mirrors &= ~mirror_bit;
3033 if (hmapx_contains(&mirror->dsts, bundle)) {
3034 bundle->dst_mirrors |= mirror_bit;
3036 bundle->dst_mirrors &= ~mirror_bit;
3039 if (mirror->out == bundle) {
3040 bundle->mirror_out |= mirror_bit;
3042 bundle->mirror_out &= ~mirror_bit;
3046 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3047 ofproto->has_mirrors = true;
3048 mac_learning_flush(ofproto->ml,
3049 &ofproto->backer->revalidate_set);
3050 mirror_update_dups(ofproto);
3056 mirror_destroy(struct ofmirror *mirror)
3058 struct ofproto_dpif *ofproto;
3059 mirror_mask_t mirror_bit;
3060 struct ofbundle *bundle;
3067 ofproto = mirror->ofproto;
3068 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3069 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
3071 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
3072 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
3073 bundle->src_mirrors &= ~mirror_bit;
3074 bundle->dst_mirrors &= ~mirror_bit;
3075 bundle->mirror_out &= ~mirror_bit;
3078 hmapx_destroy(&mirror->srcs);
3079 hmapx_destroy(&mirror->dsts);
3080 free(mirror->vlans);
3082 ofproto->mirrors[mirror->idx] = NULL;
3086 mirror_update_dups(ofproto);
3088 ofproto->has_mirrors = false;
3089 for (i = 0; i < MAX_MIRRORS; i++) {
3090 if (ofproto->mirrors[i]) {
3091 ofproto->has_mirrors = true;
3098 mirror_get_stats(struct ofproto *ofproto_, void *aux,
3099 uint64_t *packets, uint64_t *bytes)
3101 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3102 struct ofmirror *mirror = mirror_lookup(ofproto, aux);
3105 *packets = *bytes = UINT64_MAX;
3111 *packets = mirror->packet_count;
3112 *bytes = mirror->byte_count;
3118 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
3120 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3121 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
3122 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
3128 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
3130 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3131 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
3132 return bundle && bundle->mirror_out != 0;
3136 forward_bpdu_changed(struct ofproto *ofproto_)
3138 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3139 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3143 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
3146 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3147 mac_learning_set_idle_time(ofproto->ml, idle_time);
3148 mac_learning_set_max_entries(ofproto->ml, max_entries);
3153 static struct ofport_dpif *
3154 get_ofp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
3156 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
3157 return ofport ? ofport_dpif_cast(ofport) : NULL;
3160 static struct ofport_dpif *
3161 get_odp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
3163 struct ofport_dpif *port = odp_port_to_ofport(ofproto->backer, odp_port);
3164 return port && &ofproto->up == port->up.ofproto ? port : NULL;
3168 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
3169 struct ofproto_port *ofproto_port,
3170 struct dpif_port *dpif_port)
3172 ofproto_port->name = dpif_port->name;
3173 ofproto_port->type = dpif_port->type;
3174 ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
3177 static struct ofport_dpif *
3178 ofport_get_peer(const struct ofport_dpif *ofport_dpif)
3180 const struct ofproto_dpif *ofproto;
3183 peer = netdev_vport_patch_peer(ofport_dpif->up.netdev);
3188 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
3189 struct ofport *ofport;
3191 ofport = shash_find_data(&ofproto->up.port_by_name, peer);
3192 if (ofport && ofport->ofproto->ofproto_class == &ofproto_dpif_class) {
3193 return ofport_dpif_cast(ofport);
3200 port_run_fast(struct ofport_dpif *ofport)
3202 if (ofport->cfm && cfm_should_send_ccm(ofport->cfm)) {
3203 struct ofpbuf packet;
3205 ofpbuf_init(&packet, 0);
3206 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
3207 send_packet(ofport, &packet);
3208 ofpbuf_uninit(&packet);
3213 port_run(struct ofport_dpif *ofport)
3215 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
3216 bool carrier_changed = carrier_seq != ofport->carrier_seq;
3217 bool enable = netdev_get_carrier(ofport->up.netdev);
3219 ofport->carrier_seq = carrier_seq;
3221 port_run_fast(ofport);
3223 if (ofport->tnl_port
3224 && tnl_port_reconfigure(&ofport->up, ofport->odp_port,
3225 &ofport->tnl_port)) {
3226 ofproto_dpif_cast(ofport->up.ofproto)->backer->need_revalidate = true;
3230 int cfm_opup = cfm_get_opup(ofport->cfm);
3232 cfm_run(ofport->cfm);
3233 enable = enable && !cfm_get_fault(ofport->cfm);
3235 if (cfm_opup >= 0) {
3236 enable = enable && cfm_opup;
3240 if (ofport->bundle) {
3241 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
3242 if (carrier_changed) {
3243 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
3247 if (ofport->may_enable != enable) {
3248 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3250 if (ofproto->has_bundle_action) {
3251 ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
3255 ofport->may_enable = enable;
3259 port_wait(struct ofport_dpif *ofport)
3262 cfm_wait(ofport->cfm);
3267 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
3268 struct ofproto_port *ofproto_port)
3270 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3271 struct dpif_port dpif_port;
3274 if (sset_contains(&ofproto->ghost_ports, devname)) {
3275 const char *type = netdev_get_type_from_name(devname);
3277 /* We may be called before ofproto->up.port_by_name is populated with
3278 * the appropriate ofport. For this reason, we must get the name and
3279 * type from the netdev layer directly. */
3281 const struct ofport *ofport;
3283 ofport = shash_find_data(&ofproto->up.port_by_name, devname);
3284 ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
3285 ofproto_port->name = xstrdup(devname);
3286 ofproto_port->type = xstrdup(type);
3292 if (!sset_contains(&ofproto->ports, devname)) {
3295 error = dpif_port_query_by_name(ofproto->backer->dpif,
3296 devname, &dpif_port);
3298 ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
3304 port_add(struct ofproto *ofproto_, struct netdev *netdev)
3306 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3307 const char *dp_port_name = netdev_vport_get_dpif_port(netdev);
3308 const char *devname = netdev_get_name(netdev);
3310 if (netdev_vport_is_patch(netdev)) {
3311 sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
3315 if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
3316 uint32_t port_no = UINT32_MAX;
3319 error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
3323 if (netdev_get_tunnel_config(netdev)) {
3324 simap_put(&ofproto->backer->tnl_backers, dp_port_name, port_no);
3328 if (netdev_get_tunnel_config(netdev)) {
3329 sset_add(&ofproto->ghost_ports, devname);
3331 sset_add(&ofproto->ports, devname);
3337 port_del(struct ofproto *ofproto_, uint16_t ofp_port)
3339 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3340 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
3347 sset_find_and_delete(&ofproto->ghost_ports,
3348 netdev_get_name(ofport->up.netdev));
3349 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3350 if (!ofport->tnl_port) {
3351 error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
3353 /* The caller is going to close ofport->up.netdev. If this is a
3354 * bonded port, then the bond is using that netdev, so remove it
3355 * from the bond. The client will need to reconfigure everything
3356 * after deleting ports, so then the slave will get re-added. */
3357 bundle_remove(&ofport->up);
3364 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
3366 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3371 error = netdev_get_stats(ofport->up.netdev, stats);
3373 if (!error && ofport_->ofp_port == OFPP_LOCAL) {
3374 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3376 /* ofproto->stats.tx_packets represents packets that we created
3377 * internally and sent to some port (e.g. packets sent with
3378 * send_packet()). Account for them as if they had come from
3379 * OFPP_LOCAL and got forwarded. */
3381 if (stats->rx_packets != UINT64_MAX) {
3382 stats->rx_packets += ofproto->stats.tx_packets;
3385 if (stats->rx_bytes != UINT64_MAX) {
3386 stats->rx_bytes += ofproto->stats.tx_bytes;
3389 /* ofproto->stats.rx_packets represents packets that were received on
3390 * some port and we processed internally and dropped (e.g. STP).
3391 * Account for them as if they had been forwarded to OFPP_LOCAL. */
3393 if (stats->tx_packets != UINT64_MAX) {
3394 stats->tx_packets += ofproto->stats.rx_packets;
3397 if (stats->tx_bytes != UINT64_MAX) {
3398 stats->tx_bytes += ofproto->stats.rx_bytes;
3405 /* Account packets for LOCAL port. */
3407 ofproto_update_local_port_stats(const struct ofproto *ofproto_,
3408 size_t tx_size, size_t rx_size)
3410 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3413 ofproto->stats.rx_packets++;
3414 ofproto->stats.rx_bytes += rx_size;
3417 ofproto->stats.tx_packets++;
3418 ofproto->stats.tx_bytes += tx_size;
3422 struct port_dump_state {
3427 struct ofproto_port port;
3432 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
3434 *statep = xzalloc(sizeof(struct port_dump_state));
3439 port_dump_next(const struct ofproto *ofproto_, void *state_,
3440 struct ofproto_port *port)
3442 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3443 struct port_dump_state *state = state_;
3444 const struct sset *sset;
3445 struct sset_node *node;
3447 if (state->has_port) {
3448 ofproto_port_destroy(&state->port);
3449 state->has_port = false;
3451 sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
3452 while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
3455 error = port_query_by_name(ofproto_, node->name, &state->port);
3457 *port = state->port;
3458 state->has_port = true;
3460 } else if (error != ENODEV) {
3465 if (!state->ghost) {
3466 state->ghost = true;
3469 return port_dump_next(ofproto_, state_, port);
3476 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3478 struct port_dump_state *state = state_;
3480 if (state->has_port) {
3481 ofproto_port_destroy(&state->port);
3488 port_poll(const struct ofproto *ofproto_, char **devnamep)
3490 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3492 if (ofproto->port_poll_errno) {
3493 int error = ofproto->port_poll_errno;
3494 ofproto->port_poll_errno = 0;
3498 if (sset_is_empty(&ofproto->port_poll_set)) {
3502 *devnamep = sset_pop(&ofproto->port_poll_set);
3507 port_poll_wait(const struct ofproto *ofproto_)
3509 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3510 dpif_port_poll_wait(ofproto->backer->dpif);
3514 port_is_lacp_current(const struct ofport *ofport_)
3516 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3517 return (ofport->bundle && ofport->bundle->lacp
3518 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3522 /* Upcall handling. */
3524 /* Flow miss batching.
3526 * Some dpifs implement operations faster when you hand them off in a batch.
3527 * To allow batching, "struct flow_miss" queues the dpif-related work needed
3528 * for a given flow. Each "struct flow_miss" corresponds to sending one or
3529 * more packets, plus possibly installing the flow in the dpif.
3531 * So far we only batch the operations that affect flow setup time the most.
3532 * It's possible to batch more than that, but the benefit might be minimal. */
3534 struct hmap_node hmap_node;
3535 struct ofproto_dpif *ofproto;
3537 enum odp_key_fitness key_fitness;
3538 const struct nlattr *key;
3540 struct initial_vals initial_vals;
3541 struct list packets;
3542 enum dpif_upcall_type upcall_type;
3543 uint32_t odp_in_port;
3546 struct flow_miss_op {
3547 struct dpif_op dpif_op;
3549 uint64_t slow_stub[128 / 8]; /* Buffer for compose_slow_path() */
3550 struct xlate_out xout;
3551 bool xout_garbage; /* 'xout' needs to be uninitialized? */
3553 struct ofpbuf mask; /* Flow mask for "put" ops. */
3554 struct odputil_keybuf maskbuf;
3556 /* If this is a "put" op, then a pointer to the subfacet that should
3557 * be marked as uninstalled if the operation fails. */
3558 struct subfacet *subfacet;
3561 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
3562 * OpenFlow controller as necessary according to their individual
3563 * configurations. */
3565 send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet,
3566 const struct flow *flow)
3568 struct ofputil_packet_in pin;
3570 pin.packet = packet->data;
3571 pin.packet_len = packet->size;
3572 pin.reason = OFPR_NO_MATCH;
3573 pin.controller_id = 0;
3578 pin.send_len = 0; /* not used for flow table misses */
3580 flow_get_metadata(flow, &pin.fmd);
3582 connmgr_send_packet_in(ofproto->up.connmgr, &pin);
3585 static enum slow_path_reason
3586 process_special(struct xlate_ctx *ctx, const struct flow *flow,
3587 const struct ofport_dpif *ofport, const struct ofpbuf *packet)
3589 struct ofproto_dpif *ofproto = ctx->ofproto;
3590 struct flow_wildcards *wc = &ctx->xout->wc;
3594 } else if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow, wc)) {
3596 cfm_process_heartbeat(ofport->cfm, packet);
3599 } else if (ofport->bundle && ofport->bundle->lacp
3600 && flow->dl_type == htons(ETH_TYPE_LACP)) {
3602 lacp_process_packet(ofport->bundle->lacp, ofport, packet);
3605 } else if (ofproto->stp && stp_should_process_flow(flow, wc)) {
3607 stp_process_packet(ofport, packet);
3615 static struct flow_miss *
3616 flow_miss_find(struct hmap *todo, const struct ofproto_dpif *ofproto,
3617 const struct flow *flow, uint32_t hash)
3619 struct flow_miss *miss;
3621 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
3622 if (miss->ofproto == ofproto && flow_equal(&miss->flow, flow)) {
3630 /* Partially Initializes 'op' as an "execute" operation for 'miss' and
3631 * 'packet'. The caller must initialize op->actions and op->actions_len. If
3632 * 'miss' is associated with a subfacet the caller must also initialize the
3633 * returned op->subfacet, and if anything needs to be freed after processing
3634 * the op, the caller must initialize op->garbage also. */
3636 init_flow_miss_execute_op(struct flow_miss *miss, struct ofpbuf *packet,
3637 struct flow_miss_op *op)
3639 if (miss->flow.vlan_tci != miss->initial_vals.vlan_tci) {
3640 /* This packet was received on a VLAN splinter port. We
3641 * added a VLAN to the packet to make the packet resemble
3642 * the flow, but the actions were composed assuming that
3643 * the packet contained no VLAN. So, we must remove the
3644 * VLAN header from the packet before trying to execute the
3646 eth_pop_vlan(packet);
3649 op->subfacet = NULL;
3650 op->xout_garbage = false;
3651 op->dpif_op.type = DPIF_OP_EXECUTE;
3652 op->dpif_op.u.execute.key = miss->key;
3653 op->dpif_op.u.execute.key_len = miss->key_len;
3654 op->dpif_op.u.execute.packet = packet;
3655 ofpbuf_use_stack(&op->mask, &op->maskbuf, sizeof op->maskbuf);
3658 /* Helper for handle_flow_miss_without_facet() and
3659 * handle_flow_miss_with_facet(). */
3661 handle_flow_miss_common(struct rule_dpif *rule,
3662 struct ofpbuf *packet, const struct flow *flow)
3664 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3666 if (rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
3668 * Extra-special case for fail-open mode.
3670 * We are in fail-open mode and the packet matched the fail-open
3671 * rule, but we are connected to a controller too. We should send
3672 * the packet up to the controller in the hope that it will try to
3673 * set up a flow and thereby allow us to exit fail-open.
3675 * See the top-level comment in fail-open.c for more information.
3677 send_packet_in_miss(ofproto, packet, flow);
3681 /* Figures out whether a flow that missed in 'ofproto', whose details are in
3682 * 'miss' masked by 'wc', is likely to be worth tracking in detail in userspace
3683 * and (usually) installing a datapath flow. The answer is usually "yes" (a
3684 * return value of true). However, for short flows the cost of bookkeeping is
3685 * much higher than the benefits, so when the datapath holds a large number of
3686 * flows we impose some heuristics to decide which flows are likely to be worth
3689 flow_miss_should_make_facet(struct flow_miss *miss, struct flow_wildcards *wc)
3691 struct ofproto_dpif *ofproto = miss->ofproto;
3694 if (!ofproto->governor) {
3697 n_subfacets = hmap_count(&ofproto->subfacets);
3698 if (n_subfacets * 2 <= ofproto->up.flow_eviction_threshold) {
3702 ofproto->governor = governor_create(ofproto->up.name);
3705 hash = flow_hash_in_wildcards(&miss->flow, wc, 0);
3706 return governor_should_install_flow(ofproto->governor, hash,
3707 list_size(&miss->packets));
3710 /* Handles 'miss' without creating a facet or subfacet or creating any datapath
3711 * flow. 'miss->flow' must have matched 'rule' and been xlated into 'xout'.
3712 * May add an "execute" operation to 'ops' and increment '*n_ops'. */
3714 handle_flow_miss_without_facet(struct rule_dpif *rule, struct xlate_out *xout,
3715 struct flow_miss *miss,
3716 struct flow_miss_op *ops, size_t *n_ops)
3718 struct ofpbuf *packet;
3720 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3722 COVERAGE_INC(facet_suppress);
3724 handle_flow_miss_common(rule, packet, &miss->flow);
3727 struct xlate_in xin;
3729 xlate_in_init(&xin, miss->ofproto, &miss->flow,
3730 &miss->initial_vals, rule, 0, packet);
3731 xlate_actions_for_side_effects(&xin);
3734 if (xout->odp_actions.size) {
3735 struct flow_miss_op *op = &ops[*n_ops];
3736 struct dpif_execute *execute = &op->dpif_op.u.execute;
3738 init_flow_miss_execute_op(miss, packet, op);
3739 xlate_out_copy(&op->xout, xout);
3740 execute->actions = op->xout.odp_actions.data;
3741 execute->actions_len = op->xout.odp_actions.size;
3742 op->xout_garbage = true;
3749 /* Handles 'miss', which matches 'facet'. May add any required datapath
3750 * operations to 'ops', incrementing '*n_ops' for each new op.
3752 * All of the packets in 'miss' are considered to have arrived at time 'now'.
3753 * This is really important only for new facets: if we just called time_msec()
3754 * here, then the new subfacet or its packets could look (occasionally) as
3755 * though it was used some time after the facet was used. That can make a
3756 * one-packet flow look like it has a nonzero duration, which looks odd in
3757 * e.g. NetFlow statistics.
3759 * If non-null, 'stats' will be folded into 'facet'. */
3761 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
3762 long long int now, struct dpif_flow_stats *stats,
3763 struct flow_miss_op *ops, size_t *n_ops)
3765 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3766 enum subfacet_path want_path;
3767 struct subfacet *subfacet;
3768 struct ofpbuf *packet;
3770 subfacet = subfacet_create(facet, miss, now);
3771 want_path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
3773 subfacet_update_stats(subfacet, stats);
3776 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3777 struct flow_miss_op *op = &ops[*n_ops];
3779 handle_flow_miss_common(facet->rule, packet, &miss->flow);
3781 if (want_path != SF_FAST_PATH) {
3782 struct xlate_in xin;
3784 xlate_in_init(&xin, ofproto, &miss->flow, &miss->initial_vals,
3785 facet->rule, 0, packet);
3786 xlate_actions_for_side_effects(&xin);
3789 if (facet->xout.odp_actions.size) {
3790 struct dpif_execute *execute = &op->dpif_op.u.execute;
3792 init_flow_miss_execute_op(miss, packet, op);
3793 execute->actions = facet->xout.odp_actions.data,
3794 execute->actions_len = facet->xout.odp_actions.size;
3799 if (miss->upcall_type == DPIF_UC_MISS || subfacet->path != want_path) {
3800 struct flow_miss_op *op = &ops[(*n_ops)++];
3801 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
3803 subfacet->path = want_path;
3805 ofpbuf_use_stack(&op->mask, &op->maskbuf, sizeof op->maskbuf);
3806 odp_flow_key_from_mask(&op->mask, &facet->xout.wc.masks,
3807 &miss->flow, UINT32_MAX);
3809 op->xout_garbage = false;
3810 op->dpif_op.type = DPIF_OP_FLOW_PUT;
3811 op->subfacet = subfacet;
3812 put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
3813 put->key = miss->key;
3814 put->key_len = miss->key_len;
3815 put->mask = op->mask.data;
3816 put->mask_len = op->mask.size;
3818 if (want_path == SF_FAST_PATH) {
3819 put->actions = facet->xout.odp_actions.data;
3820 put->actions_len = facet->xout.odp_actions.size;
3822 compose_slow_path(ofproto, &miss->flow, facet->xout.slow,
3823 op->slow_stub, sizeof op->slow_stub,
3824 &put->actions, &put->actions_len);
3830 /* Handles flow miss 'miss'. May add any required datapath operations
3831 * to 'ops', incrementing '*n_ops' for each new op. */
3833 handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops,
3836 struct ofproto_dpif *ofproto = miss->ofproto;
3837 struct dpif_flow_stats stats__;
3838 struct dpif_flow_stats *stats = &stats__;
3839 struct ofpbuf *packet;
3840 struct facet *facet;
3844 memset(stats, 0, sizeof *stats);
3846 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3847 stats->tcp_flags |= packet_get_tcp_flags(packet, &miss->flow);
3848 stats->n_bytes += packet->size;
3852 facet = facet_lookup_valid(ofproto, &miss->flow);
3854 struct flow_wildcards wc;
3855 struct rule_dpif *rule;
3856 struct xlate_out xout;
3857 struct xlate_in xin;
3859 flow_wildcards_init_catchall(&wc);
3860 rule = rule_dpif_lookup(ofproto, &miss->flow, &wc);
3861 rule_credit_stats(rule, stats);
3863 xlate_in_init(&xin, ofproto, &miss->flow, &miss->initial_vals, rule,
3864 stats->tcp_flags, NULL);
3865 xin.resubmit_stats = stats;
3866 xin.may_learn = true;
3867 xlate_actions(&xin, &xout);
3868 flow_wildcards_or(&xout.wc, &xout.wc, &wc);
3870 /* There does not exist a bijection between 'struct flow' and datapath
3871 * flow keys with fitness ODP_FIT_TO_LITTLE. This breaks a fundamental
3872 * assumption used throughout the facet and subfacet handling code.
3873 * Since we have to handle these misses in userspace anyway, we simply
3874 * skip facet creation, avoiding the problem altogether. */
3875 if (miss->key_fitness == ODP_FIT_TOO_LITTLE
3876 || !flow_miss_should_make_facet(miss, &xout.wc)) {
3877 handle_flow_miss_without_facet(rule, &xout, miss, ops, n_ops);
3881 facet = facet_create(miss, rule, &xout, stats);
3884 handle_flow_miss_with_facet(miss, facet, now, stats, ops, n_ops);
3887 static struct drop_key *
3888 drop_key_lookup(const struct dpif_backer *backer, const struct nlattr *key,
3891 struct drop_key *drop_key;
3893 HMAP_FOR_EACH_WITH_HASH (drop_key, hmap_node, hash_bytes(key, key_len, 0),
3894 &backer->drop_keys) {
3895 if (drop_key->key_len == key_len
3896 && !memcmp(drop_key->key, key, key_len)) {
3904 drop_key_clear(struct dpif_backer *backer)
3906 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3907 struct drop_key *drop_key, *next;
3909 HMAP_FOR_EACH_SAFE (drop_key, next, hmap_node, &backer->drop_keys) {
3912 error = dpif_flow_del(backer->dpif, drop_key->key, drop_key->key_len,
3914 if (error && !VLOG_DROP_WARN(&rl)) {
3915 struct ds ds = DS_EMPTY_INITIALIZER;
3916 odp_flow_key_format(drop_key->key, drop_key->key_len, &ds);
3917 VLOG_WARN("Failed to delete drop key (%s) (%s)", strerror(error),
3922 hmap_remove(&backer->drop_keys, &drop_key->hmap_node);
3923 free(drop_key->key);
3928 /* Given a datpath, packet, and flow metadata ('backer', 'packet', and 'key'
3929 * respectively), populates 'flow' with the result of odp_flow_key_to_flow().
3930 * Optionally, if nonnull, populates 'fitnessp' with the fitness of 'flow' as
3931 * returned by odp_flow_key_to_flow(). Also, optionally populates 'ofproto'
3932 * with the ofproto_dpif, and 'odp_in_port' with the datapath in_port, that
3933 * 'packet' ingressed.
3935 * If 'ofproto' is nonnull, requires 'flow''s in_port to exist. Otherwise sets
3936 * 'flow''s in_port to OFPP_NONE.
3938 * This function does post-processing on data returned from
3939 * odp_flow_key_to_flow() to help make VLAN splinters transparent to the rest
3940 * of the upcall processing logic. In particular, if the extracted in_port is
3941 * a VLAN splinter port, it replaces flow->in_port by the "real" port, sets
3942 * flow->vlan_tci correctly for the VLAN of the VLAN splinter port, and pushes
3943 * a VLAN header onto 'packet' (if it is nonnull).
3945 * Optionally, if 'initial_vals' is nonnull, sets 'initial_vals->vlan_tci'
3946 * to the VLAN TCI with which the packet was really received, that is, the
3947 * actual VLAN TCI extracted by odp_flow_key_to_flow(). (This differs from
3948 * the value returned in flow->vlan_tci only for packets received on
3949 * VLAN splinters.) Also, if received on an IP tunnel, sets
3950 * 'initial_vals->tunnel_ip_tos' to the tunnel's IP TOS.
3952 * Similarly, this function also includes some logic to help with tunnels. It
3953 * may modify 'flow' as necessary to make the tunneling implementation
3954 * transparent to the upcall processing logic.
3956 * Returns 0 if successful, ENODEV if the parsed flow has no associated ofport,
3957 * or some other positive errno if there are other problems. */
3959 ofproto_receive(const struct dpif_backer *backer, struct ofpbuf *packet,
3960 const struct nlattr *key, size_t key_len,
3961 struct flow *flow, enum odp_key_fitness *fitnessp,
3962 struct ofproto_dpif **ofproto, uint32_t *odp_in_port,
3963 struct initial_vals *initial_vals)
3965 const struct ofport_dpif *port;
3966 enum odp_key_fitness fitness;
3969 fitness = odp_flow_key_to_flow(key, key_len, flow);
3970 if (fitness == ODP_FIT_ERROR) {
3976 initial_vals->vlan_tci = flow->vlan_tci;
3977 initial_vals->tunnel_ip_tos = flow->tunnel.ip_tos;
3981 *odp_in_port = flow->in_port;
3984 if (tnl_port_should_receive(flow)) {
3985 const struct ofport *ofport = tnl_port_receive(flow);
3987 flow->in_port = OFPP_NONE;
3990 port = ofport_dpif_cast(ofport);
3992 /* We can't reproduce 'key' from 'flow'. */
3993 fitness = fitness == ODP_FIT_PERFECT ? ODP_FIT_TOO_MUCH : fitness;
3995 /* XXX: Since the tunnel module is not scoped per backer, it's
3996 * theoretically possible that we'll receive an ofport belonging to an
3997 * entirely different datapath. In practice, this can't happen because
3998 * no platforms has two separate datapaths which each support
4000 ovs_assert(ofproto_dpif_cast(port->up.ofproto)->backer == backer);
4002 port = odp_port_to_ofport(backer, flow->in_port);
4004 flow->in_port = OFPP_NONE;
4008 flow->in_port = port->up.ofp_port;
4009 if (vsp_adjust_flow(ofproto_dpif_cast(port->up.ofproto), flow)) {
4011 /* Make the packet resemble the flow, so that it gets sent to
4012 * an OpenFlow controller properly, so that it looks correct
4013 * for sFlow, and so that flow_extract() will get the correct
4014 * vlan_tci if it is called on 'packet'.
4016 * The allocated space inside 'packet' probably also contains
4017 * 'key', that is, both 'packet' and 'key' are probably part of
4018 * a struct dpif_upcall (see the large comment on that
4019 * structure definition), so pushing data on 'packet' is in
4020 * general not a good idea since it could overwrite 'key' or
4021 * free it as a side effect. However, it's OK in this special
4022 * case because we know that 'packet' is inside a Netlink
4023 * attribute: pushing 4 bytes will just overwrite the 4-byte
4024 * "struct nlattr", which is fine since we don't need that
4025 * header anymore. */
4026 eth_push_vlan(packet, flow->vlan_tci);
4028 /* We can't reproduce 'key' from 'flow'. */
4029 fitness = fitness == ODP_FIT_PERFECT ? ODP_FIT_TOO_MUCH : fitness;
4035 *ofproto = ofproto_dpif_cast(port->up.ofproto);
4040 *fitnessp = fitness;
4046 handle_miss_upcalls(struct dpif_backer *backer, struct dpif_upcall *upcalls,
4049 struct dpif_upcall *upcall;
4050 struct flow_miss *miss;
4051 struct flow_miss misses[FLOW_MISS_MAX_BATCH];
4052 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
4053 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
4063 /* Construct the to-do list.
4065 * This just amounts to extracting the flow from each packet and sticking
4066 * the packets that have the same flow in the same "flow_miss" structure so
4067 * that we can process them together. */
4070 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
4071 struct flow_miss *miss = &misses[n_misses];
4072 struct flow_miss *existing_miss;
4073 struct ofproto_dpif *ofproto;
4074 uint32_t odp_in_port;
4079 error = ofproto_receive(backer, upcall->packet, upcall->key,
4080 upcall->key_len, &flow, &miss->key_fitness,
4081 &ofproto, &odp_in_port, &miss->initial_vals);
4082 if (error == ENODEV) {
4083 struct drop_key *drop_key;
4085 /* Received packet on port for which we couldn't associate
4086 * an ofproto. This can happen if a port is removed while
4087 * traffic is being received. Print a rate-limited message
4088 * in case it happens frequently. Install a drop flow so
4089 * that future packets of the flow are inexpensively dropped
4091 VLOG_INFO_RL(&rl, "received packet on unassociated port %"PRIu32,
4094 drop_key = drop_key_lookup(backer, upcall->key, upcall->key_len);
4096 drop_key = xmalloc(sizeof *drop_key);
4097 drop_key->key = xmemdup(upcall->key, upcall->key_len);
4098 drop_key->key_len = upcall->key_len;
4100 hmap_insert(&backer->drop_keys, &drop_key->hmap_node,
4101 hash_bytes(drop_key->key, drop_key->key_len, 0));
4102 dpif_flow_put(backer->dpif, DPIF_FP_CREATE | DPIF_FP_MODIFY,
4103 drop_key->key, drop_key->key_len,
4104 NULL, 0, NULL, 0, NULL);
4112 ofproto->n_missed++;
4113 flow_extract(upcall->packet, flow.skb_priority, flow.skb_mark,
4114 &flow.tunnel, flow.in_port, &miss->flow);
4116 /* Add other packets to a to-do list. */
4117 hash = flow_hash(&miss->flow, 0);
4118 existing_miss = flow_miss_find(&todo, ofproto, &miss->flow, hash);
4119 if (!existing_miss) {
4120 hmap_insert(&todo, &miss->hmap_node, hash);
4121 miss->ofproto = ofproto;
4122 miss->key = upcall->key;
4123 miss->key_len = upcall->key_len;
4124 miss->upcall_type = upcall->type;
4125 miss->odp_in_port = odp_in_port;
4126 list_init(&miss->packets);
4130 miss = existing_miss;
4132 list_push_back(&miss->packets, &upcall->packet->list_node);
4135 /* Process each element in the to-do list, constructing the set of
4136 * operations to batch. */
4138 HMAP_FOR_EACH (miss, hmap_node, &todo) {
4139 handle_flow_miss(miss, flow_miss_ops, &n_ops);
4141 ovs_assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
4143 /* Execute batch. */
4144 for (i = 0; i < n_ops; i++) {
4145 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
4147 dpif_operate(backer->dpif, dpif_ops, n_ops);
4149 for (i = 0; i < n_ops; i++) {
4150 if (dpif_ops[i]->error != 0
4151 && flow_miss_ops[i].dpif_op.type == DPIF_OP_FLOW_PUT
4152 && flow_miss_ops[i].subfacet) {
4153 struct subfacet *subfacet = flow_miss_ops[i].subfacet;
4155 COVERAGE_INC(subfacet_install_fail);
4157 subfacet->path = SF_NOT_INSTALLED;
4161 if (flow_miss_ops[i].xout_garbage) {
4162 xlate_out_uninit(&flow_miss_ops[i].xout);
4165 hmap_destroy(&todo);
4168 static enum { SFLOW_UPCALL, MISS_UPCALL, BAD_UPCALL, FLOW_SAMPLE_UPCALL,
4170 classify_upcall(const struct dpif_upcall *upcall)
4172 size_t userdata_len;
4173 union user_action_cookie cookie;
4175 /* First look at the upcall type. */
4176 switch (upcall->type) {
4177 case DPIF_UC_ACTION:
4183 case DPIF_N_UC_TYPES:
4185 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
4189 /* "action" upcalls need a closer look. */
4190 if (!upcall->userdata) {
4191 VLOG_WARN_RL(&rl, "action upcall missing cookie");
4194 userdata_len = nl_attr_get_size(upcall->userdata);
4195 if (userdata_len < sizeof cookie.type
4196 || userdata_len > sizeof cookie) {
4197 VLOG_WARN_RL(&rl, "action upcall cookie has unexpected size %zu",
4201 memset(&cookie, 0, sizeof cookie);
4202 memcpy(&cookie, nl_attr_get(upcall->userdata), userdata_len);
4203 if (userdata_len == sizeof cookie.sflow
4204 && cookie.type == USER_ACTION_COOKIE_SFLOW) {
4205 return SFLOW_UPCALL;
4206 } else if (userdata_len == sizeof cookie.slow_path
4207 && cookie.type == USER_ACTION_COOKIE_SLOW_PATH) {
4209 } else if (userdata_len == sizeof cookie.flow_sample
4210 && cookie.type == USER_ACTION_COOKIE_FLOW_SAMPLE) {
4211 return FLOW_SAMPLE_UPCALL;
4212 } else if (userdata_len == sizeof cookie.ipfix
4213 && cookie.type == USER_ACTION_COOKIE_IPFIX) {
4214 return IPFIX_UPCALL;
4216 VLOG_WARN_RL(&rl, "invalid user cookie of type %"PRIu16
4217 " and size %zu", cookie.type, userdata_len);
4223 handle_sflow_upcall(struct dpif_backer *backer,
4224 const struct dpif_upcall *upcall)
4226 struct ofproto_dpif *ofproto;
4227 union user_action_cookie cookie;
4229 uint32_t odp_in_port;
4231 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
4232 &flow, NULL, &ofproto, &odp_in_port, NULL)
4233 || !ofproto->sflow) {
4237 memset(&cookie, 0, sizeof cookie);
4238 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.sflow);
4239 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
4240 odp_in_port, &cookie);
4244 handle_flow_sample_upcall(struct dpif_backer *backer,
4245 const struct dpif_upcall *upcall)
4247 struct ofproto_dpif *ofproto;
4248 union user_action_cookie cookie;
4251 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
4252 &flow, NULL, &ofproto, NULL, NULL)
4253 || !ofproto->ipfix) {
4257 memset(&cookie, 0, sizeof cookie);
4258 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.flow_sample);
4260 /* The flow reflects exactly the contents of the packet. Sample
4261 * the packet using it. */
4262 dpif_ipfix_flow_sample(ofproto->ipfix, upcall->packet, &flow,
4263 cookie.flow_sample.collector_set_id,
4264 cookie.flow_sample.probability,
4265 cookie.flow_sample.obs_domain_id,
4266 cookie.flow_sample.obs_point_id);
4270 handle_ipfix_upcall(struct dpif_backer *backer,
4271 const struct dpif_upcall *upcall)
4273 struct ofproto_dpif *ofproto;
4276 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
4277 &flow, NULL, &ofproto, NULL, NULL)
4278 || !ofproto->ipfix) {
4282 /* The flow reflects exactly the contents of the packet. Sample
4283 * the packet using it. */
4284 dpif_ipfix_bridge_sample(ofproto->ipfix, upcall->packet, &flow);
4288 handle_upcalls(struct dpif_backer *backer, unsigned int max_batch)
4290 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
4291 struct ofpbuf miss_bufs[FLOW_MISS_MAX_BATCH];
4292 uint64_t miss_buf_stubs[FLOW_MISS_MAX_BATCH][4096 / 8];
4297 ovs_assert(max_batch <= FLOW_MISS_MAX_BATCH);
4300 for (n_processed = 0; n_processed < max_batch; n_processed++) {
4301 struct dpif_upcall *upcall = &misses[n_misses];
4302 struct ofpbuf *buf = &miss_bufs[n_misses];
4305 ofpbuf_use_stub(buf, miss_buf_stubs[n_misses],
4306 sizeof miss_buf_stubs[n_misses]);
4307 error = dpif_recv(backer->dpif, upcall, buf);
4313 switch (classify_upcall(upcall)) {
4315 /* Handle it later. */
4320 handle_sflow_upcall(backer, upcall);
4324 case FLOW_SAMPLE_UPCALL:
4325 handle_flow_sample_upcall(backer, upcall);
4330 handle_ipfix_upcall(backer, upcall);
4340 /* Handle deferred MISS_UPCALL processing. */
4341 handle_miss_upcalls(backer, misses, n_misses);
4342 for (i = 0; i < n_misses; i++) {
4343 ofpbuf_uninit(&miss_bufs[i]);
4349 /* Flow expiration. */
4351 static int subfacet_max_idle(const struct ofproto_dpif *);
4352 static void update_stats(struct dpif_backer *);
4353 static void rule_expire(struct rule_dpif *);
4354 static void expire_subfacets(struct ofproto_dpif *, int dp_max_idle);
4356 /* This function is called periodically by run(). Its job is to collect
4357 * updates for the flows that have been installed into the datapath, most
4358 * importantly when they last were used, and then use that information to
4359 * expire flows that have not been used recently.
4361 * Returns the number of milliseconds after which it should be called again. */
4363 expire(struct dpif_backer *backer)
4365 struct ofproto_dpif *ofproto;
4366 int max_idle = INT32_MAX;
4368 /* Periodically clear out the drop keys in an effort to keep them
4369 * relatively few. */
4370 drop_key_clear(backer);
4372 /* Update stats for each flow in the backer. */
4373 update_stats(backer);
4375 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4376 struct rule *rule, *next_rule;
4379 if (ofproto->backer != backer) {
4383 /* Keep track of the max number of flows per ofproto_dpif. */
4384 update_max_subfacet_count(ofproto);
4386 /* Expire subfacets that have been idle too long. */
4387 dp_max_idle = subfacet_max_idle(ofproto);
4388 expire_subfacets(ofproto, dp_max_idle);
4390 max_idle = MIN(max_idle, dp_max_idle);
4392 /* Expire OpenFlow flows whose idle_timeout or hard_timeout
4394 LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
4395 &ofproto->up.expirable) {
4396 rule_expire(rule_dpif_cast(rule));
4399 /* All outstanding data in existing flows has been accounted, so it's a
4400 * good time to do bond rebalancing. */
4401 if (ofproto->has_bonded_bundles) {
4402 struct ofbundle *bundle;
4404 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
4406 bond_rebalance(bundle->bond, &backer->revalidate_set);
4412 return MIN(max_idle, 1000);
4415 /* Updates flow table statistics given that the datapath just reported 'stats'
4416 * as 'subfacet''s statistics. */
4418 update_subfacet_stats(struct subfacet *subfacet,
4419 const struct dpif_flow_stats *stats)
4421 struct facet *facet = subfacet->facet;
4422 struct dpif_flow_stats diff;
4424 diff.tcp_flags = stats->tcp_flags;
4425 diff.used = stats->used;
4427 if (stats->n_packets >= subfacet->dp_packet_count) {
4428 diff.n_packets = stats->n_packets - subfacet->dp_packet_count;
4430 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
4434 if (stats->n_bytes >= subfacet->dp_byte_count) {
4435 diff.n_bytes = stats->n_bytes - subfacet->dp_byte_count;
4437 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
4441 subfacet->dp_packet_count = stats->n_packets;
4442 subfacet->dp_byte_count = stats->n_bytes;
4443 subfacet_update_stats(subfacet, &diff);
4445 if (facet->accounted_bytes < facet->byte_count) {
4447 facet_account(facet);
4448 facet->accounted_bytes = facet->byte_count;
4452 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
4453 * about, or a flow that shouldn't be installed but was anyway. Delete it. */
4455 delete_unexpected_flow(struct ofproto_dpif *ofproto,
4456 const struct nlattr *key, size_t key_len)
4458 if (!VLOG_DROP_WARN(&rl)) {
4462 odp_flow_key_format(key, key_len, &s);
4463 VLOG_WARN("unexpected flow on %s: %s", ofproto->up.name, ds_cstr(&s));
4467 COVERAGE_INC(facet_unexpected);
4468 dpif_flow_del(ofproto->backer->dpif, key, key_len, NULL);
4471 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
4473 * This function also pushes statistics updates to rules which each facet
4474 * resubmits into. Generally these statistics will be accurate. However, if a
4475 * facet changes the rule it resubmits into at some time in between
4476 * update_stats() runs, it is possible that statistics accrued to the
4477 * old rule will be incorrectly attributed to the new rule. This could be
4478 * avoided by calling update_stats() whenever rules are created or
4479 * deleted. However, the performance impact of making so many calls to the
4480 * datapath do not justify the benefit of having perfectly accurate statistics.
4482 * In addition, this function maintains per ofproto flow hit counts. The patch
4483 * port is not treated specially. e.g. A packet ingress from br0 patched into
4484 * br1 will increase the hit count of br0 by 1, however, does not affect
4485 * the hit or miss counts of br1.
4488 update_stats(struct dpif_backer *backer)
4490 const struct dpif_flow_stats *stats;
4491 struct dpif_flow_dump dump;
4492 struct ofproto_dpif *ofproto;
4493 const struct nlattr *key, *mask;
4494 size_t key_len, mask_len;
4496 dpif_flow_dump_start(&dump, backer->dpif);
4497 while (dpif_flow_dump_next(&dump, &key, &key_len,
4498 &mask, &mask_len, NULL, NULL, &stats)) {
4500 struct subfacet *subfacet;
4503 if (ofproto_receive(backer, NULL, key, key_len, &flow, NULL, &ofproto,
4508 ofproto->total_subfacet_count += hmap_count(&ofproto->subfacets);
4509 ofproto->n_update_stats++;
4511 key_hash = odp_flow_key_hash(key, key_len);
4512 subfacet = subfacet_find(ofproto, key, key_len, key_hash);
4513 switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
4515 /* Update ofproto_dpif's hit count. */
4516 if (stats->n_packets > subfacet->dp_packet_count) {
4517 uint64_t delta = stats->n_packets - subfacet->dp_packet_count;
4518 dpif_stats_update_hit_count(ofproto, delta);
4521 update_subfacet_stats(subfacet, stats);
4525 /* Stats are updated per-packet. */
4528 case SF_NOT_INSTALLED:
4530 delete_unexpected_flow(ofproto, key, key_len);
4535 dpif_flow_dump_done(&dump);
4537 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4538 update_moving_averages(ofproto);
4543 /* Calculates and returns the number of milliseconds of idle time after which
4544 * subfacets should expire from the datapath. When a subfacet expires, we fold
4545 * its statistics into its facet, and when a facet's last subfacet expires, we
4546 * fold its statistic into its rule. */
4548 subfacet_max_idle(const struct ofproto_dpif *ofproto)
4551 * Idle time histogram.
4553 * Most of the time a switch has a relatively small number of subfacets.
4554 * When this is the case we might as well keep statistics for all of them
4555 * in userspace and to cache them in the kernel datapath for performance as
4558 * As the number of subfacets increases, the memory required to maintain
4559 * statistics about them in userspace and in the kernel becomes
4560 * significant. However, with a large number of subfacets it is likely
4561 * that only a few of them are "heavy hitters" that consume a large amount
4562 * of bandwidth. At this point, only heavy hitters are worth caching in
4563 * the kernel and maintaining in userspaces; other subfacets we can
4566 * The technique used to compute the idle time is to build a histogram with
4567 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
4568 * that is installed in the kernel gets dropped in the appropriate bucket.
4569 * After the histogram has been built, we compute the cutoff so that only
4570 * the most-recently-used 1% of subfacets (but at least
4571 * ofproto->up.flow_eviction_threshold flows) are kept cached. At least
4572 * the most-recently-used bucket of subfacets is kept, so actually an
4573 * arbitrary number of subfacets can be kept in any given expiration run
4574 * (though the next run will delete most of those unless they receive
4577 * This requires a second pass through the subfacets, in addition to the
4578 * pass made by update_stats(), because the former function never looks at
4579 * uninstallable subfacets.
4581 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
4582 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
4583 int buckets[N_BUCKETS] = { 0 };
4584 int total, subtotal, bucket;
4585 struct subfacet *subfacet;
4589 total = hmap_count(&ofproto->subfacets);
4590 if (total <= ofproto->up.flow_eviction_threshold) {
4591 return N_BUCKETS * BUCKET_WIDTH;
4594 /* Build histogram. */
4596 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
4597 long long int idle = now - subfacet->used;
4598 int bucket = (idle <= 0 ? 0
4599 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
4600 : (unsigned int) idle / BUCKET_WIDTH);
4604 /* Find the first bucket whose flows should be expired. */
4605 subtotal = bucket = 0;
4607 subtotal += buckets[bucket++];
4608 } while (bucket < N_BUCKETS &&
4609 subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100));
4611 if (VLOG_IS_DBG_ENABLED()) {
4615 ds_put_cstr(&s, "keep");
4616 for (i = 0; i < N_BUCKETS; i++) {
4618 ds_put_cstr(&s, ", drop");
4621 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
4624 VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s));
4628 return bucket * BUCKET_WIDTH;
4632 expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle)
4634 /* Cutoff time for most flows. */
4635 long long int normal_cutoff = time_msec() - dp_max_idle;
4637 /* We really want to keep flows for special protocols around, so use a more
4638 * conservative cutoff. */
4639 long long int special_cutoff = time_msec() - 10000;
4641 struct subfacet *subfacet, *next_subfacet;
4642 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
4646 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
4647 &ofproto->subfacets) {
4648 long long int cutoff;
4650 cutoff = (subfacet->facet->xout.slow & (SLOW_CFM | SLOW_LACP
4654 if (subfacet->used < cutoff) {
4655 if (subfacet->path != SF_NOT_INSTALLED) {
4656 batch[n_batch++] = subfacet;
4657 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
4658 subfacet_destroy_batch(ofproto, batch, n_batch);
4662 subfacet_destroy(subfacet);
4668 subfacet_destroy_batch(ofproto, batch, n_batch);
4672 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
4673 * then delete it entirely. */
4675 rule_expire(struct rule_dpif *rule)
4677 struct facet *facet, *next_facet;
4681 if (rule->up.pending) {
4682 /* We'll have to expire it later. */
4686 /* Has 'rule' expired? */
4688 if (rule->up.hard_timeout
4689 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
4690 reason = OFPRR_HARD_TIMEOUT;
4691 } else if (rule->up.idle_timeout
4692 && now > rule->up.used + rule->up.idle_timeout * 1000) {
4693 reason = OFPRR_IDLE_TIMEOUT;
4698 COVERAGE_INC(ofproto_dpif_expired);
4700 /* Update stats. (This is a no-op if the rule expired due to an idle
4701 * timeout, because that only happens when the rule has no facets left.) */
4702 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4703 facet_remove(facet);
4706 /* Get rid of the rule. */
4707 ofproto_rule_expire(&rule->up, reason);
4712 /* Creates and returns a new facet based on 'miss'.
4714 * The caller must already have determined that no facet with an identical
4715 * 'miss->flow' exists in 'miss->ofproto'.
4717 * 'rule' and 'xout' must have been created based on 'miss'.
4719 * 'facet'' statistics are initialized based on 'stats'.
4721 * The facet will initially have no subfacets. The caller should create (at
4722 * least) one subfacet with subfacet_create(). */
4723 static struct facet *
4724 facet_create(const struct flow_miss *miss, struct rule_dpif *rule,
4725 struct xlate_out *xout, struct dpif_flow_stats *stats)
4727 struct ofproto_dpif *ofproto = miss->ofproto;
4728 struct facet *facet;
4731 facet = xzalloc(sizeof *facet);
4732 facet->packet_count = facet->prev_packet_count = stats->n_packets;
4733 facet->byte_count = facet->prev_byte_count = stats->n_bytes;
4734 facet->tcp_flags = stats->tcp_flags;
4735 facet->used = stats->used;
4736 facet->flow = miss->flow;
4737 facet->initial_vals = miss->initial_vals;
4738 facet->learn_rl = time_msec() + 500;
4741 list_push_back(&facet->rule->facets, &facet->list_node);
4742 list_init(&facet->subfacets);
4743 netflow_flow_init(&facet->nf_flow);
4744 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
4746 xlate_out_copy(&facet->xout, xout);
4748 match_init(&match, &facet->flow, &facet->xout.wc);
4749 cls_rule_init(&facet->cr, &match, OFP_DEFAULT_PRIORITY);
4750 classifier_insert(&ofproto->facets, &facet->cr);
4752 facet->nf_flow.output_iface = facet->xout.nf_output_iface;
4758 facet_free(struct facet *facet)
4761 xlate_out_uninit(&facet->xout);
4766 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
4767 * 'packet', which arrived on 'in_port'. */
4769 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
4770 const struct nlattr *odp_actions, size_t actions_len,
4771 struct ofpbuf *packet)
4773 struct odputil_keybuf keybuf;
4777 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4778 odp_flow_key_from_flow(&key, flow,
4779 ofp_port_to_odp_port(ofproto, flow->in_port));
4781 error = dpif_execute(ofproto->backer->dpif, key.data, key.size,
4782 odp_actions, actions_len, packet);
4786 /* Remove 'facet' from its ofproto and free up the associated memory:
4788 * - If 'facet' was installed in the datapath, uninstalls it and updates its
4789 * rule's statistics, via subfacet_uninstall().
4791 * - Removes 'facet' from its rule and from ofproto->facets.
4794 facet_remove(struct facet *facet)
4796 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4797 struct subfacet *subfacet, *next_subfacet;
4799 ovs_assert(!list_is_empty(&facet->subfacets));
4801 /* First uninstall all of the subfacets to get final statistics. */
4802 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4803 subfacet_uninstall(subfacet);
4806 /* Flush the final stats to the rule.
4808 * This might require us to have at least one subfacet around so that we
4809 * can use its actions for accounting in facet_account(), which is why we
4810 * have uninstalled but not yet destroyed the subfacets. */
4811 facet_flush_stats(facet);
4813 /* Now we're really all done so destroy everything. */
4814 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
4815 &facet->subfacets) {
4816 subfacet_destroy__(subfacet);
4818 classifier_remove(&ofproto->facets, &facet->cr);
4819 cls_rule_destroy(&facet->cr);
4820 list_remove(&facet->list_node);
4824 /* Feed information from 'facet' back into the learning table to keep it in
4825 * sync with what is actually flowing through the datapath. */
4827 facet_learn(struct facet *facet)
4829 long long int now = time_msec();
4831 if (!facet->xout.has_fin_timeout && now < facet->learn_rl) {
4835 facet->learn_rl = now + 500;
4837 if (!facet->xout.has_learn
4838 && !facet->xout.has_normal
4839 && (!facet->xout.has_fin_timeout
4840 || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
4844 facet_push_stats(facet, true);
4848 facet_account(struct facet *facet)
4850 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4851 const struct nlattr *a;
4856 if (!facet->xout.has_normal || !ofproto->has_bonded_bundles) {
4859 n_bytes = facet->byte_count - facet->accounted_bytes;
4861 /* This loop feeds byte counters to bond_account() for rebalancing to use
4862 * as a basis. We also need to track the actual VLAN on which the packet
4863 * is going to be sent to ensure that it matches the one passed to
4864 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
4867 * We use the actions from an arbitrary subfacet because they should all
4868 * be equally valid for our purpose. */
4869 vlan_tci = facet->flow.vlan_tci;
4870 NL_ATTR_FOR_EACH_UNSAFE (a, left, facet->xout.odp_actions.data,
4871 facet->xout.odp_actions.size) {
4872 const struct ovs_action_push_vlan *vlan;
4873 struct ofport_dpif *port;
4875 switch (nl_attr_type(a)) {
4876 case OVS_ACTION_ATTR_OUTPUT:
4877 port = get_odp_port(ofproto, nl_attr_get_u32(a));
4878 if (port && port->bundle && port->bundle->bond) {
4879 bond_account(port->bundle->bond, &facet->flow,
4880 vlan_tci_to_vid(vlan_tci), n_bytes);
4884 case OVS_ACTION_ATTR_POP_VLAN:
4885 vlan_tci = htons(0);
4888 case OVS_ACTION_ATTR_PUSH_VLAN:
4889 vlan = nl_attr_get(a);
4890 vlan_tci = vlan->vlan_tci;
4896 /* Returns true if the only action for 'facet' is to send to the controller.
4897 * (We don't report NetFlow expiration messages for such facets because they
4898 * are just part of the control logic for the network, not real traffic). */
4900 facet_is_controller_flow(struct facet *facet)
4903 const struct rule *rule = &facet->rule->up;
4904 const struct ofpact *ofpacts = rule->ofpacts;
4905 size_t ofpacts_len = rule->ofpacts_len;
4907 if (ofpacts_len > 0 &&
4908 ofpacts->type == OFPACT_CONTROLLER &&
4909 ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len)) {
4916 /* Folds all of 'facet''s statistics into its rule. Also updates the
4917 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
4918 * 'facet''s statistics in the datapath should have been zeroed and folded into
4919 * its packet and byte counts before this function is called. */
4921 facet_flush_stats(struct facet *facet)
4923 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4924 struct subfacet *subfacet;
4926 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4927 ovs_assert(!subfacet->dp_byte_count);
4928 ovs_assert(!subfacet->dp_packet_count);
4931 facet_push_stats(facet, false);
4932 if (facet->accounted_bytes < facet->byte_count) {
4933 facet_account(facet);
4934 facet->accounted_bytes = facet->byte_count;
4937 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
4938 struct ofexpired expired;
4939 expired.flow = facet->flow;
4940 expired.packet_count = facet->packet_count;
4941 expired.byte_count = facet->byte_count;
4942 expired.used = facet->used;
4943 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4946 /* Reset counters to prevent double counting if 'facet' ever gets
4948 facet_reset_counters(facet);
4950 netflow_flow_clear(&facet->nf_flow);
4951 facet->tcp_flags = 0;
4954 /* Searches 'ofproto''s table of facets for one which would be responsible for
4955 * 'flow'. Returns it if found, otherwise a null pointer.
4957 * The returned facet might need revalidation; use facet_lookup_valid()
4958 * instead if that is important. */
4959 static struct facet *
4960 facet_find(struct ofproto_dpif *ofproto, const struct flow *flow)
4962 struct cls_rule *cr = classifier_lookup(&ofproto->facets, flow, NULL);
4963 return cr ? CONTAINER_OF(cr, struct facet, cr) : NULL;
4966 /* Searches 'ofproto''s table of facets for one capable that covers
4967 * 'flow'. Returns it if found, otherwise a null pointer.
4969 * The returned facet is guaranteed to be valid. */
4970 static struct facet *
4971 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow)
4973 struct facet *facet;
4975 facet = facet_find(ofproto, flow);
4977 && (ofproto->backer->need_revalidate
4978 || tag_set_intersects(&ofproto->backer->revalidate_set,
4980 && !facet_revalidate(facet)) {
4988 facet_check_consistency(struct facet *facet)
4990 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
4992 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4994 struct xlate_out xout;
4995 struct xlate_in xin;
4997 struct rule_dpif *rule;
5000 /* Check the rule for consistency. */
5001 rule = rule_dpif_lookup(ofproto, &facet->flow, NULL);
5002 if (rule != facet->rule) {
5003 if (!VLOG_DROP_WARN(&rl)) {
5004 struct ds s = DS_EMPTY_INITIALIZER;
5006 flow_format(&s, &facet->flow);
5007 ds_put_format(&s, ": facet associated with wrong rule (was "
5008 "table=%"PRIu8",", facet->rule->up.table_id);
5009 cls_rule_format(&facet->rule->up.cr, &s);
5010 ds_put_format(&s, ") (should have been table=%"PRIu8",",
5012 cls_rule_format(&rule->up.cr, &s);
5013 ds_put_cstr(&s, ")\n");
5020 /* Check the datapath actions for consistency. */
5021 xlate_in_init(&xin, ofproto, &facet->flow, &facet->initial_vals, rule,
5023 xlate_actions(&xin, &xout);
5025 ok = ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)
5026 && facet->xout.slow == xout.slow;
5027 if (!ok && !VLOG_DROP_WARN(&rl)) {
5028 struct ds s = DS_EMPTY_INITIALIZER;
5030 flow_format(&s, &facet->flow);
5031 ds_put_cstr(&s, ": inconsistency in facet");
5033 if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
5034 ds_put_cstr(&s, " (actions were: ");
5035 format_odp_actions(&s, facet->xout.odp_actions.data,
5036 facet->xout.odp_actions.size);
5037 ds_put_cstr(&s, ") (correct actions: ");
5038 format_odp_actions(&s, xout.odp_actions.data,
5039 xout.odp_actions.size);
5040 ds_put_cstr(&s, ")");
5043 if (facet->xout.slow != xout.slow) {
5044 ds_put_format(&s, " slow path incorrect. should be %d", xout.slow);
5049 xlate_out_uninit(&xout);
5054 /* Re-searches the classifier for 'facet':
5056 * - If the rule found is different from 'facet''s current rule, moves
5057 * 'facet' to the new rule and recompiles its actions.
5059 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
5060 * where it is and recompiles its actions anyway.
5062 * - If any of 'facet''s subfacets correspond to a new flow according to
5063 * ofproto_receive(), 'facet' is removed.
5065 * Returns true if 'facet' is still valid. False if 'facet' was removed. */
5067 facet_revalidate(struct facet *facet)
5069 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5070 struct rule_dpif *new_rule;
5071 struct subfacet *subfacet;
5072 struct flow_wildcards wc;
5073 struct xlate_out xout;
5074 struct xlate_in xin;
5076 COVERAGE_INC(facet_revalidate);
5078 /* Check that child subfacets still correspond to this facet. Tunnel
5079 * configuration changes could cause a subfacet's OpenFlow in_port to
5081 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
5082 struct ofproto_dpif *recv_ofproto;
5083 struct flow recv_flow;
5086 error = ofproto_receive(ofproto->backer, NULL, subfacet->key,
5087 subfacet->key_len, &recv_flow, NULL,
5088 &recv_ofproto, NULL, NULL);
5090 || recv_ofproto != ofproto
5091 || facet != facet_find(ofproto, &recv_flow)) {
5092 facet_remove(facet);
5097 flow_wildcards_init_catchall(&wc);
5098 new_rule = rule_dpif_lookup(ofproto, &facet->flow, &wc);
5100 /* Calculate new datapath actions.
5102 * We do not modify any 'facet' state yet, because we might need to, e.g.,
5103 * emit a NetFlow expiration and, if so, we need to have the old state
5104 * around to properly compose it. */
5105 xlate_in_init(&xin, ofproto, &facet->flow, &facet->initial_vals, new_rule,
5107 xlate_actions(&xin, &xout);
5108 flow_wildcards_or(&xout.wc, &xout.wc, &wc);
5110 /* A facet's slow path reason should only change under dramatic
5111 * circumstances. Rather than try to update everything, it's simpler to
5112 * remove the facet and start over.
5114 * More importantly, if a facet's wildcards change, it will be relatively
5115 * difficult to figure out if its subfacets still belong to it, and if not
5116 * which facet they may belong to. Again, to avoid the complexity, we
5117 * simply give up instead. */
5118 if (facet->xout.slow != xout.slow
5119 || memcmp(&facet->xout.wc, &xout.wc, sizeof xout.wc)) {
5120 facet_remove(facet);
5121 xlate_out_uninit(&xout);
5125 if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
5126 LIST_FOR_EACH(subfacet, list_node, &facet->subfacets) {
5127 if (subfacet->path == SF_FAST_PATH) {
5128 struct dpif_flow_stats stats;
5130 subfacet_install(subfacet, &xout.odp_actions, &stats);
5131 subfacet_update_stats(subfacet, &stats);
5135 facet_flush_stats(facet);
5137 ofpbuf_clear(&facet->xout.odp_actions);
5138 ofpbuf_put(&facet->xout.odp_actions, xout.odp_actions.data,
5139 xout.odp_actions.size);
5142 /* Update 'facet' now that we've taken care of all the old state. */
5143 facet->xout.tags = xout.tags;
5144 facet->xout.slow = xout.slow;
5145 facet->xout.has_learn = xout.has_learn;
5146 facet->xout.has_normal = xout.has_normal;
5147 facet->xout.has_fin_timeout = xout.has_fin_timeout;
5148 facet->xout.nf_output_iface = xout.nf_output_iface;
5149 facet->xout.mirrors = xout.mirrors;
5150 facet->nf_flow.output_iface = facet->xout.nf_output_iface;
5152 if (facet->rule != new_rule) {
5153 COVERAGE_INC(facet_changed_rule);
5154 list_remove(&facet->list_node);
5155 list_push_back(&new_rule->facets, &facet->list_node);
5156 facet->rule = new_rule;
5157 facet->used = new_rule->up.created;
5158 facet->prev_used = facet->used;
5161 xlate_out_uninit(&xout);
5166 facet_reset_counters(struct facet *facet)
5168 facet->packet_count = 0;
5169 facet->byte_count = 0;
5170 facet->prev_packet_count = 0;
5171 facet->prev_byte_count = 0;
5172 facet->accounted_bytes = 0;
5176 facet_push_stats(struct facet *facet, bool may_learn)
5178 struct dpif_flow_stats stats;
5180 ovs_assert(facet->packet_count >= facet->prev_packet_count);
5181 ovs_assert(facet->byte_count >= facet->prev_byte_count);
5182 ovs_assert(facet->used >= facet->prev_used);
5184 stats.n_packets = facet->packet_count - facet->prev_packet_count;
5185 stats.n_bytes = facet->byte_count - facet->prev_byte_count;
5186 stats.used = facet->used;
5187 stats.tcp_flags = facet->tcp_flags;
5189 if (may_learn || stats.n_packets || facet->used > facet->prev_used) {
5190 struct ofproto_dpif *ofproto =
5191 ofproto_dpif_cast(facet->rule->up.ofproto);
5193 struct ofport_dpif *in_port;
5194 struct xlate_in xin;
5196 facet->prev_packet_count = facet->packet_count;
5197 facet->prev_byte_count = facet->byte_count;
5198 facet->prev_used = facet->used;
5200 in_port = get_ofp_port(ofproto, facet->flow.in_port);
5201 if (in_port && in_port->tnl_port) {
5202 netdev_vport_inc_rx(in_port->up.netdev, &stats);
5205 rule_credit_stats(facet->rule, &stats);
5206 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow,
5208 netflow_flow_update_flags(&facet->nf_flow, facet->tcp_flags);
5209 update_mirror_stats(ofproto, facet->xout.mirrors, stats.n_packets,
5212 xlate_in_init(&xin, ofproto, &facet->flow, &facet->initial_vals,
5213 facet->rule, stats.tcp_flags, NULL);
5214 xin.resubmit_stats = &stats;
5215 xin.may_learn = may_learn;
5216 xlate_actions_for_side_effects(&xin);
5221 push_all_stats__(bool run_fast)
5223 static long long int rl = LLONG_MIN;
5224 struct ofproto_dpif *ofproto;
5226 if (time_msec() < rl) {
5230 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5231 struct cls_cursor cursor;
5232 struct facet *facet;
5234 cls_cursor_init(&cursor, &ofproto->facets, NULL);
5235 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5236 facet_push_stats(facet, false);
5243 rl = time_msec() + 100;
5247 push_all_stats(void)
5249 push_all_stats__(true);
5253 rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats)
5255 rule->packet_count += stats->n_packets;
5256 rule->byte_count += stats->n_bytes;
5257 ofproto_rule_update_used(&rule->up, stats->used);
5262 static struct subfacet *
5263 subfacet_find(struct ofproto_dpif *ofproto,
5264 const struct nlattr *key, size_t key_len, uint32_t key_hash)
5266 struct subfacet *subfacet;
5268 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
5269 &ofproto->subfacets) {
5270 if (subfacet->key_len == key_len
5271 && !memcmp(key, subfacet->key, key_len)) {
5279 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
5280 * 'key_fitness', 'key', and 'key_len' members in 'miss'. Returns the
5281 * existing subfacet if there is one, otherwise creates and returns a
5283 static struct subfacet *
5284 subfacet_create(struct facet *facet, struct flow_miss *miss,
5287 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5288 enum odp_key_fitness key_fitness = miss->key_fitness;
5289 const struct nlattr *key = miss->key;
5290 size_t key_len = miss->key_len;
5292 struct subfacet *subfacet;
5294 key_hash = odp_flow_key_hash(key, key_len);
5296 if (list_is_empty(&facet->subfacets)) {
5297 subfacet = &facet->one_subfacet;
5299 subfacet = subfacet_find(ofproto, key, key_len, key_hash);
5301 if (subfacet->facet == facet) {
5305 /* This shouldn't happen. */
5306 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
5307 subfacet_destroy(subfacet);
5310 subfacet = xmalloc(sizeof *subfacet);
5313 hmap_insert(&ofproto->subfacets, &subfacet->hmap_node, key_hash);
5314 list_push_back(&facet->subfacets, &subfacet->list_node);
5315 subfacet->facet = facet;
5316 subfacet->key_fitness = key_fitness;
5317 subfacet->key = xmemdup(key, key_len);
5318 subfacet->key_len = key_len;
5319 subfacet->used = now;
5320 subfacet->created = now;
5321 subfacet->dp_packet_count = 0;
5322 subfacet->dp_byte_count = 0;
5323 subfacet->path = SF_NOT_INSTALLED;
5324 subfacet->odp_in_port = miss->odp_in_port;
5326 ofproto->subfacet_add_count++;
5330 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
5331 * its facet within 'ofproto', and frees it. */
5333 subfacet_destroy__(struct subfacet *subfacet)
5335 struct facet *facet = subfacet->facet;
5336 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5338 /* Update ofproto stats before uninstall the subfacet. */
5339 ofproto->subfacet_del_count++;
5340 ofproto->total_subfacet_life_span += (time_msec() - subfacet->created);
5342 subfacet_uninstall(subfacet);
5343 hmap_remove(&ofproto->subfacets, &subfacet->hmap_node);
5344 list_remove(&subfacet->list_node);
5345 free(subfacet->key);
5346 if (subfacet != &facet->one_subfacet) {
5351 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
5352 * last remaining subfacet in its facet destroys the facet too. */
5354 subfacet_destroy(struct subfacet *subfacet)
5356 struct facet *facet = subfacet->facet;
5358 if (list_is_singleton(&facet->subfacets)) {
5359 /* facet_remove() needs at least one subfacet (it will remove it). */
5360 facet_remove(facet);
5362 subfacet_destroy__(subfacet);
5367 subfacet_destroy_batch(struct ofproto_dpif *ofproto,
5368 struct subfacet **subfacets, int n)
5370 struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH];
5371 struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH];
5372 struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH];
5375 for (i = 0; i < n; i++) {
5376 ops[i].type = DPIF_OP_FLOW_DEL;
5377 ops[i].u.flow_del.key = subfacets[i]->key;
5378 ops[i].u.flow_del.key_len = subfacets[i]->key_len;
5379 ops[i].u.flow_del.stats = &stats[i];
5383 dpif_operate(ofproto->backer->dpif, opsp, n);
5384 for (i = 0; i < n; i++) {
5385 subfacet_reset_dp_stats(subfacets[i], &stats[i]);
5386 subfacets[i]->path = SF_NOT_INSTALLED;
5387 subfacet_destroy(subfacets[i]);
5392 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
5393 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
5394 * in the datapath will be zeroed and 'stats' will be updated with traffic new
5395 * since 'subfacet' was last updated.
5397 * Returns 0 if successful, otherwise a positive errno value. */
5399 subfacet_install(struct subfacet *subfacet, const struct ofpbuf *odp_actions,
5400 struct dpif_flow_stats *stats)
5402 struct facet *facet = subfacet->facet;
5403 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5404 enum subfacet_path path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
5405 const struct nlattr *actions = odp_actions->data;
5406 size_t actions_len = odp_actions->size;
5407 struct odputil_keybuf maskbuf;
5410 uint64_t slow_path_stub[128 / 8];
5411 enum dpif_flow_put_flags flags;
5414 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
5416 flags |= DPIF_FP_ZERO_STATS;
5419 if (path == SF_SLOW_PATH) {
5420 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
5421 slow_path_stub, sizeof slow_path_stub,
5422 &actions, &actions_len);
5425 ofpbuf_use_stack(&mask, &maskbuf, sizeof maskbuf);
5426 odp_flow_key_from_mask(&mask, &facet->xout.wc.masks,
5427 &facet->flow, UINT32_MAX);
5429 ret = dpif_flow_put(ofproto->backer->dpif, flags, subfacet->key,
5430 subfacet->key_len, mask.data, mask.size,
5431 actions, actions_len, stats);
5434 subfacet_reset_dp_stats(subfacet, stats);
5438 COVERAGE_INC(subfacet_install_fail);
5440 subfacet->path = path;
5445 /* If 'subfacet' is installed in the datapath, uninstalls it. */
5447 subfacet_uninstall(struct subfacet *subfacet)
5449 if (subfacet->path != SF_NOT_INSTALLED) {
5450 struct rule_dpif *rule = subfacet->facet->rule;
5451 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5452 struct dpif_flow_stats stats;
5455 error = dpif_flow_del(ofproto->backer->dpif, subfacet->key,
5456 subfacet->key_len, &stats);
5457 subfacet_reset_dp_stats(subfacet, &stats);
5459 subfacet_update_stats(subfacet, &stats);
5461 subfacet->path = SF_NOT_INSTALLED;
5463 ovs_assert(subfacet->dp_packet_count == 0);
5464 ovs_assert(subfacet->dp_byte_count == 0);
5468 /* Resets 'subfacet''s datapath statistics counters. This should be called
5469 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
5470 * non-null, it should contain the statistics returned by dpif when 'subfacet'
5471 * was reset in the datapath. 'stats' will be modified to include only
5472 * statistics new since 'subfacet' was last updated. */
5474 subfacet_reset_dp_stats(struct subfacet *subfacet,
5475 struct dpif_flow_stats *stats)
5478 && subfacet->dp_packet_count <= stats->n_packets
5479 && subfacet->dp_byte_count <= stats->n_bytes) {
5480 stats->n_packets -= subfacet->dp_packet_count;
5481 stats->n_bytes -= subfacet->dp_byte_count;
5484 subfacet->dp_packet_count = 0;
5485 subfacet->dp_byte_count = 0;
5488 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
5490 * Because of the meaning of a subfacet's counters, it only makes sense to do
5491 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
5492 * represents a packet that was sent by hand or if it represents statistics
5493 * that have been cleared out of the datapath. */
5495 subfacet_update_stats(struct subfacet *subfacet,
5496 const struct dpif_flow_stats *stats)
5498 if (stats->n_packets || stats->used > subfacet->used) {
5499 struct facet *facet = subfacet->facet;
5501 subfacet->used = MAX(subfacet->used, stats->used);
5502 facet->used = MAX(facet->used, stats->used);
5503 facet->packet_count += stats->n_packets;
5504 facet->byte_count += stats->n_bytes;
5505 facet->tcp_flags |= stats->tcp_flags;
5511 /* Lookup 'flow' in 'ofproto''s classifier. If 'wc' is non-null, sets
5512 * the fields that were relevant as part of the lookup. */
5513 static struct rule_dpif *
5514 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow,
5515 struct flow_wildcards *wc)
5517 struct rule_dpif *rule;
5519 rule = rule_dpif_lookup__(ofproto, flow, wc, 0);
5524 return rule_dpif_miss_rule(ofproto, flow);
5527 static struct rule_dpif *
5528 rule_dpif_lookup__(struct ofproto_dpif *ofproto, const struct flow *flow,
5529 struct flow_wildcards *wc, uint8_t table_id)
5531 struct cls_rule *cls_rule;
5532 struct classifier *cls;
5535 if (table_id >= N_TABLES) {
5540 memset(&wc->masks.dl_type, 0xff, sizeof wc->masks.dl_type);
5541 wc->masks.nw_frag |= FLOW_NW_FRAG_MASK;
5544 cls = &ofproto->up.tables[table_id].cls;
5545 frag = (flow->nw_frag & FLOW_NW_FRAG_ANY) != 0;
5546 if (frag && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
5547 /* We must pretend that transport ports are unavailable. */
5548 struct flow ofpc_normal_flow = *flow;
5549 ofpc_normal_flow.tp_src = htons(0);
5550 ofpc_normal_flow.tp_dst = htons(0);
5551 cls_rule = classifier_lookup(cls, &ofpc_normal_flow, wc);
5552 } else if (frag && ofproto->up.frag_handling == OFPC_FRAG_DROP) {
5553 cls_rule = &ofproto->drop_frags_rule->up.cr;
5555 flow_wildcards_init_exact(wc);
5558 cls_rule = classifier_lookup(cls, flow, wc);
5560 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
5563 static struct rule_dpif *
5564 rule_dpif_miss_rule(struct ofproto_dpif *ofproto, const struct flow *flow)
5566 struct ofport_dpif *port;
5568 port = get_ofp_port(ofproto, flow->in_port);
5570 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, flow->in_port);
5571 return ofproto->miss_rule;
5574 if (port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN) {
5575 return ofproto->no_packet_in_rule;
5577 return ofproto->miss_rule;
5581 complete_operation(struct rule_dpif *rule)
5583 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5585 rule_invalidate(rule);
5587 struct dpif_completion *c = xmalloc(sizeof *c);
5588 c->op = rule->up.pending;
5589 list_push_back(&ofproto->completions, &c->list_node);
5591 ofoperation_complete(rule->up.pending, 0);
5595 static struct rule *
5598 struct rule_dpif *rule = xmalloc(sizeof *rule);
5603 rule_dealloc(struct rule *rule_)
5605 struct rule_dpif *rule = rule_dpif_cast(rule_);
5610 rule_construct(struct rule *rule_)
5612 struct rule_dpif *rule = rule_dpif_cast(rule_);
5613 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5614 struct rule_dpif *victim;
5617 rule->packet_count = 0;
5618 rule->byte_count = 0;
5620 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
5621 if (victim && !list_is_empty(&victim->facets)) {
5622 struct facet *facet;
5624 rule->facets = victim->facets;
5625 list_moved(&rule->facets);
5626 LIST_FOR_EACH (facet, list_node, &rule->facets) {
5627 /* XXX: We're only clearing our local counters here. It's possible
5628 * that quite a few packets are unaccounted for in the datapath
5629 * statistics. These will be accounted to the new rule instead of
5630 * cleared as required. This could be fixed by clearing out the
5631 * datapath statistics for this facet, but currently it doesn't
5633 facet_reset_counters(facet);
5637 /* Must avoid list_moved() in this case. */
5638 list_init(&rule->facets);
5641 table_id = rule->up.table_id;
5643 rule->tag = victim->tag;
5644 } else if (table_id == 0) {
5649 miniflow_expand(&rule->up.cr.match.flow, &flow);
5650 rule->tag = rule_calculate_tag(&flow, &rule->up.cr.match.mask,
5651 ofproto->tables[table_id].basis);
5654 complete_operation(rule);
5659 rule_destruct(struct rule *rule_)
5661 struct rule_dpif *rule = rule_dpif_cast(rule_);
5662 struct facet *facet, *next_facet;
5664 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
5665 facet_revalidate(facet);
5668 complete_operation(rule);
5672 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
5674 struct rule_dpif *rule = rule_dpif_cast(rule_);
5676 /* push_all_stats() can handle flow misses which, when using the learn
5677 * action, can cause rules to be added and deleted. This can corrupt our
5678 * caller's datastructures which assume that rule_get_stats() doesn't have
5679 * an impact on the flow table. To be safe, we disable miss handling. */
5680 push_all_stats__(false);
5682 /* Start from historical data for 'rule' itself that are no longer tracked
5683 * in facets. This counts, for example, facets that have expired. */
5684 *packets = rule->packet_count;
5685 *bytes = rule->byte_count;
5689 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
5690 struct ofpbuf *packet)
5692 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5693 struct initial_vals initial_vals;
5694 struct dpif_flow_stats stats;
5695 struct xlate_out xout;
5696 struct xlate_in xin;
5698 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
5699 rule_credit_stats(rule, &stats);
5701 initial_vals.vlan_tci = flow->vlan_tci;
5702 initial_vals.tunnel_ip_tos = flow->tunnel.ip_tos;
5703 xlate_in_init(&xin, ofproto, flow, &initial_vals, rule, stats.tcp_flags,
5705 xin.resubmit_stats = &stats;
5706 xlate_actions(&xin, &xout);
5708 execute_odp_actions(ofproto, flow, xout.odp_actions.data,
5709 xout.odp_actions.size, packet);
5711 xlate_out_uninit(&xout);
5715 rule_execute(struct rule *rule, const struct flow *flow,
5716 struct ofpbuf *packet)
5718 rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
5719 ofpbuf_delete(packet);
5724 rule_modify_actions(struct rule *rule_)
5726 struct rule_dpif *rule = rule_dpif_cast(rule_);
5728 complete_operation(rule);
5731 /* Sends 'packet' out 'ofport'.
5732 * May modify 'packet'.
5733 * Returns 0 if successful, otherwise a positive errno value. */
5735 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
5737 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
5738 uint64_t odp_actions_stub[1024 / 8];
5739 struct ofpbuf key, odp_actions;
5740 struct odputil_keybuf keybuf;
5745 flow_extract(packet, 0, 0, NULL, OFPP_LOCAL, &flow);
5746 if (netdev_vport_is_patch(ofport->up.netdev)) {
5747 struct ofproto_dpif *peer_ofproto;
5748 struct dpif_flow_stats stats;
5749 struct ofport_dpif *peer;
5750 struct rule_dpif *rule;
5752 peer = ofport_get_peer(ofport);
5757 dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
5758 netdev_vport_inc_tx(ofport->up.netdev, &stats);
5759 netdev_vport_inc_rx(peer->up.netdev, &stats);
5761 flow.in_port = peer->up.ofp_port;
5762 peer_ofproto = ofproto_dpif_cast(peer->up.ofproto);
5763 rule = rule_dpif_lookup(peer_ofproto, &flow, NULL);
5764 rule_dpif_execute(rule, &flow, packet);
5769 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5771 if (ofport->tnl_port) {
5772 struct flow_wildcards wc;
5773 struct dpif_flow_stats stats;
5775 odp_port = tnl_port_send(ofport->tnl_port, &flow, &wc);
5776 if (odp_port == OVSP_NONE) {
5780 dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
5781 netdev_vport_inc_tx(ofport->up.netdev, &stats);
5782 odp_put_tunnel_action(&flow.tunnel, &odp_actions);
5783 odp_put_skb_mark_action(flow.skb_mark, &odp_actions);
5785 odp_port = vsp_realdev_to_vlandev(ofproto, ofport->odp_port,
5787 if (odp_port != ofport->odp_port) {
5788 eth_pop_vlan(packet);
5789 flow.vlan_tci = htons(0);
5793 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5794 odp_flow_key_from_flow(&key, &flow,
5795 ofp_port_to_odp_port(ofproto, flow.in_port));
5797 compose_sflow_action(ofproto, &odp_actions, &flow, odp_port);
5798 compose_ipfix_action(ofproto, &odp_actions, &flow);
5800 nl_msg_put_u32(&odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port);
5801 error = dpif_execute(ofproto->backer->dpif,
5803 odp_actions.data, odp_actions.size,
5805 ofpbuf_uninit(&odp_actions);
5808 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)",
5809 ofproto->up.name, odp_port, strerror(error));
5811 ofproto_update_local_port_stats(ofport->up.ofproto, packet->size, 0);
5815 /* OpenFlow to datapath action translation. */
5817 static bool may_receive(const struct ofport_dpif *, struct xlate_ctx *);
5818 static void do_xlate_actions(const struct ofpact *, size_t ofpacts_len,
5819 struct xlate_ctx *);
5820 static void xlate_normal(struct xlate_ctx *);
5822 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
5823 * The action will state 'slow' as the reason that the action is in the slow
5824 * path. (This is purely informational: it allows a human viewing "ovs-dpctl
5825 * dump-flows" output to see why a flow is in the slow path.)
5827 * The 'stub_size' bytes in 'stub' will be used to store the action.
5828 * 'stub_size' must be large enough for the action.
5830 * The action and its size will be stored in '*actionsp' and '*actions_lenp',
5833 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
5834 enum slow_path_reason slow,
5835 uint64_t *stub, size_t stub_size,
5836 const struct nlattr **actionsp, size_t *actions_lenp)
5838 union user_action_cookie cookie;
5841 cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
5842 cookie.slow_path.unused = 0;
5843 cookie.slow_path.reason = slow;
5845 ofpbuf_use_stack(&buf, stub, stub_size);
5846 if (slow & (SLOW_CFM | SLOW_LACP | SLOW_STP)) {
5847 uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif, UINT32_MAX);
5848 odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
5850 put_userspace_action(ofproto, &buf, flow, &cookie,
5851 sizeof cookie.slow_path);
5853 *actionsp = buf.data;
5854 *actions_lenp = buf.size;
5858 put_userspace_action(const struct ofproto_dpif *ofproto,
5859 struct ofpbuf *odp_actions,
5860 const struct flow *flow,
5861 const union user_action_cookie *cookie,
5862 const size_t cookie_size)
5866 pid = dpif_port_get_pid(ofproto->backer->dpif,
5867 ofp_port_to_odp_port(ofproto, flow->in_port));
5869 return odp_put_userspace_action(pid, cookie, cookie_size, odp_actions);
5872 /* Compose SAMPLE action for sFlow or IPFIX. The given probability is
5873 * the number of packets out of UINT32_MAX to sample. The given
5874 * cookie is passed back in the callback for each sampled packet.
5877 compose_sample_action(const struct ofproto_dpif *ofproto,
5878 struct ofpbuf *odp_actions,
5879 const struct flow *flow,
5880 const uint32_t probability,
5881 const union user_action_cookie *cookie,
5882 const size_t cookie_size)
5884 size_t sample_offset, actions_offset;
5887 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
5889 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
5891 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
5892 cookie_offset = put_userspace_action(ofproto, odp_actions, flow, cookie,
5895 nl_msg_end_nested(odp_actions, actions_offset);
5896 nl_msg_end_nested(odp_actions, sample_offset);
5897 return cookie_offset;
5901 compose_sflow_cookie(const struct ofproto_dpif *ofproto,
5902 ovs_be16 vlan_tci, uint32_t odp_port,
5903 unsigned int n_outputs, union user_action_cookie *cookie)
5907 cookie->type = USER_ACTION_COOKIE_SFLOW;
5908 cookie->sflow.vlan_tci = vlan_tci;
5910 /* See http://www.sflow.org/sflow_version_5.txt (search for "Input/output
5911 * port information") for the interpretation of cookie->output. */
5912 switch (n_outputs) {
5914 /* 0x40000000 | 256 means "packet dropped for unknown reason". */
5915 cookie->sflow.output = 0x40000000 | 256;
5919 ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port);
5921 cookie->sflow.output = ifindex;
5926 /* 0x80000000 means "multiple output ports. */
5927 cookie->sflow.output = 0x80000000 | n_outputs;
5932 /* Compose SAMPLE action for sFlow bridge sampling. */
5934 compose_sflow_action(const struct ofproto_dpif *ofproto,
5935 struct ofpbuf *odp_actions,
5936 const struct flow *flow,
5939 uint32_t probability;
5940 union user_action_cookie cookie;
5942 if (!ofproto->sflow || flow->in_port == OFPP_NONE) {
5946 probability = dpif_sflow_get_probability(ofproto->sflow);
5947 compose_sflow_cookie(ofproto, htons(0), odp_port,
5948 odp_port == OVSP_NONE ? 0 : 1, &cookie);
5950 return compose_sample_action(ofproto, odp_actions, flow, probability,
5951 &cookie, sizeof cookie.sflow);
5955 compose_flow_sample_cookie(uint16_t probability, uint32_t collector_set_id,
5956 uint32_t obs_domain_id, uint32_t obs_point_id,
5957 union user_action_cookie *cookie)
5959 cookie->type = USER_ACTION_COOKIE_FLOW_SAMPLE;
5960 cookie->flow_sample.probability = probability;
5961 cookie->flow_sample.collector_set_id = collector_set_id;
5962 cookie->flow_sample.obs_domain_id = obs_domain_id;
5963 cookie->flow_sample.obs_point_id = obs_point_id;
5967 compose_ipfix_cookie(union user_action_cookie *cookie)
5969 cookie->type = USER_ACTION_COOKIE_IPFIX;
5972 /* Compose SAMPLE action for IPFIX bridge sampling. */
5974 compose_ipfix_action(const struct ofproto_dpif *ofproto,
5975 struct ofpbuf *odp_actions,
5976 const struct flow *flow)
5978 uint32_t probability;
5979 union user_action_cookie cookie;
5981 if (!ofproto->ipfix || flow->in_port == OFPP_NONE) {
5985 probability = dpif_ipfix_get_bridge_exporter_probability(ofproto->ipfix);
5986 compose_ipfix_cookie(&cookie);
5988 compose_sample_action(ofproto, odp_actions, flow, probability,
5989 &cookie, sizeof cookie.ipfix);
5992 /* SAMPLE action for sFlow must be first action in any given list of
5993 * actions. At this point we do not have all information required to
5994 * build it. So try to build sample action as complete as possible. */
5996 add_sflow_action(struct xlate_ctx *ctx)
5998 ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto,
5999 &ctx->xout->odp_actions,
6000 &ctx->xin->flow, OVSP_NONE);
6001 ctx->sflow_odp_port = 0;
6002 ctx->sflow_n_outputs = 0;
6005 /* SAMPLE action for IPFIX must be 1st or 2nd action in any given list
6006 * of actions, eventually after the SAMPLE action for sFlow. */
6008 add_ipfix_action(struct xlate_ctx *ctx)
6010 compose_ipfix_action(ctx->ofproto, &ctx->xout->odp_actions,
6014 /* Fix SAMPLE action according to data collected while composing ODP actions.
6015 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
6016 * USERSPACE action's user-cookie which is required for sflow. */
6018 fix_sflow_action(struct xlate_ctx *ctx)
6020 const struct flow *base = &ctx->base_flow;
6021 union user_action_cookie *cookie;
6023 if (!ctx->user_cookie_offset) {
6027 cookie = ofpbuf_at(&ctx->xout->odp_actions, ctx->user_cookie_offset,
6028 sizeof cookie->sflow);
6029 ovs_assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
6031 compose_sflow_cookie(ctx->ofproto, base->vlan_tci,
6032 ctx->sflow_odp_port, ctx->sflow_n_outputs, cookie);
6036 compose_output_action__(struct xlate_ctx *ctx, uint16_t ofp_port,
6039 const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
6040 struct flow_wildcards *wc = &ctx->xout->wc;
6041 ovs_be16 flow_vlan_tci;
6042 uint32_t flow_skb_mark;
6043 uint8_t flow_nw_tos;
6044 struct priority_to_dscp *pdscp;
6045 uint32_t out_port, odp_port;
6047 /* If 'struct flow' gets additional metadata, we'll need to zero it out
6048 * before traversing a patch port. */
6049 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 20);
6052 xlate_report(ctx, "Nonexistent output port");
6054 } else if (ofport->up.pp.config & OFPUTIL_PC_NO_FWD) {
6055 xlate_report(ctx, "OFPPC_NO_FWD set, skipping output");
6057 } else if (check_stp && !stp_forward_in_state(ofport->stp_state)) {
6058 xlate_report(ctx, "STP not in forwarding state, skipping output");
6062 if (netdev_vport_is_patch(ofport->up.netdev)) {
6063 struct ofport_dpif *peer = ofport_get_peer(ofport);
6064 struct flow old_flow = ctx->xin->flow;
6065 const struct ofproto_dpif *peer_ofproto;
6066 enum slow_path_reason special;
6067 struct ofport_dpif *in_port;
6070 xlate_report(ctx, "Nonexistent patch port peer");
6074 peer_ofproto = ofproto_dpif_cast(peer->up.ofproto);
6075 if (peer_ofproto->backer != ctx->ofproto->backer) {
6076 xlate_report(ctx, "Patch port peer on a different datapath");
6080 ctx->ofproto = ofproto_dpif_cast(peer->up.ofproto);
6081 ctx->xin->flow.in_port = peer->up.ofp_port;
6082 ctx->xin->flow.metadata = htonll(0);
6083 memset(&ctx->xin->flow.tunnel, 0, sizeof ctx->xin->flow.tunnel);
6084 memset(ctx->xin->flow.regs, 0, sizeof ctx->xin->flow.regs);
6086 in_port = get_ofp_port(ctx->ofproto, ctx->xin->flow.in_port);
6087 special = process_special(ctx, &ctx->xin->flow, in_port,
6090 ctx->xout->slow = special;
6091 } else if (!in_port || may_receive(in_port, ctx)) {
6092 if (!in_port || stp_forward_in_state(in_port->stp_state)) {
6093 xlate_table_action(ctx, ctx->xin->flow.in_port, 0, true);
6095 /* Forwarding is disabled by STP. Let OFPP_NORMAL and the
6096 * learning action look at the packet, then drop it. */
6097 struct flow old_base_flow = ctx->base_flow;
6098 size_t old_size = ctx->xout->odp_actions.size;
6099 xlate_table_action(ctx, ctx->xin->flow.in_port, 0, true);
6100 ctx->base_flow = old_base_flow;
6101 ctx->xout->odp_actions.size = old_size;
6105 ctx->xin->flow = old_flow;
6106 ctx->ofproto = ofproto_dpif_cast(ofport->up.ofproto);
6108 if (ctx->xin->resubmit_stats) {
6109 netdev_vport_inc_tx(ofport->up.netdev, ctx->xin->resubmit_stats);
6110 netdev_vport_inc_rx(peer->up.netdev, ctx->xin->resubmit_stats);
6116 flow_vlan_tci = ctx->xin->flow.vlan_tci;
6117 flow_skb_mark = ctx->xin->flow.skb_mark;
6118 flow_nw_tos = ctx->xin->flow.nw_tos;
6120 pdscp = get_priority(ofport, ctx->xin->flow.skb_priority);
6122 wc->masks.nw_tos |= IP_ECN_MASK;
6123 ctx->xin->flow.nw_tos &= ~IP_DSCP_MASK;
6124 ctx->xin->flow.nw_tos |= pdscp->dscp;
6127 if (ofport->tnl_port) {
6128 /* Save tunnel metadata so that changes made due to
6129 * the Logical (tunnel) Port are not visible for any further
6130 * matches, while explicit set actions on tunnel metadata are.
6132 struct flow_tnl flow_tnl = ctx->xin->flow.tunnel;
6133 odp_port = tnl_port_send(ofport->tnl_port, &ctx->xin->flow,
6135 if (odp_port == OVSP_NONE) {
6136 xlate_report(ctx, "Tunneling decided against output");
6137 goto out; /* restore flow_nw_tos */
6140 if (ctx->xin->resubmit_stats) {
6141 netdev_vport_inc_tx(ofport->up.netdev, ctx->xin->resubmit_stats);
6143 out_port = odp_port;
6144 commit_odp_tunnel_action(&ctx->xin->flow, &ctx->base_flow,
6145 &ctx->xout->odp_actions);
6146 ctx->xin->flow.tunnel = flow_tnl; /* Restore tunnel metadata */
6148 uint16_t vlandev_port;
6150 odp_port = ofport->odp_port;
6151 if (!hmap_is_empty(&ctx->ofproto->realdev_vid_map)) {
6152 wc->masks.vlan_tci |= htons(VLAN_VID_MASK | VLAN_CFI);
6154 vlandev_port = vsp_realdev_to_vlandev(ctx->ofproto, ofp_port,
6155 ctx->xin->flow.vlan_tci);
6156 if (vlandev_port == ofp_port) {
6157 out_port = odp_port;
6159 out_port = ofp_port_to_odp_port(ctx->ofproto, vlandev_port);
6160 ctx->xin->flow.vlan_tci = htons(0);
6162 ctx->xin->flow.skb_mark &= ~IPSEC_MARK;
6164 commit_odp_actions(&ctx->xin->flow, &ctx->base_flow,
6165 &ctx->xout->odp_actions, &ctx->xout->wc);
6166 nl_msg_put_u32(&ctx->xout->odp_actions, OVS_ACTION_ATTR_OUTPUT, out_port);
6168 ctx->sflow_odp_port = odp_port;
6169 ctx->sflow_n_outputs++;
6170 ctx->xout->nf_output_iface = ofp_port;
6173 ctx->xin->flow.vlan_tci = flow_vlan_tci;
6174 ctx->xin->flow.skb_mark = flow_skb_mark;
6176 ctx->xin->flow.nw_tos = flow_nw_tos;
6180 compose_output_action(struct xlate_ctx *ctx, uint16_t ofp_port)
6182 compose_output_action__(ctx, ofp_port, true);
6186 tag_the_flow(struct xlate_ctx *ctx, struct rule_dpif *rule)
6188 struct ofproto_dpif *ofproto = ctx->ofproto;
6189 uint8_t table_id = ctx->table_id;
6191 if (table_id > 0 && table_id < N_TABLES) {
6192 struct table_dpif *table = &ofproto->tables[table_id];
6193 if (table->other_table) {
6194 ctx->xout->tags |= (rule && rule->tag
6196 : rule_calculate_tag(&ctx->xin->flow,
6197 &table->other_table->mask,
6203 /* Common rule processing in one place to avoid duplicating code. */
6204 static struct rule_dpif *
6205 ctx_rule_hooks(struct xlate_ctx *ctx, struct rule_dpif *rule,
6208 if (ctx->xin->resubmit_hook) {
6209 ctx->xin->resubmit_hook(ctx, rule);
6211 if (rule == NULL && may_packet_in) {
6213 * check if table configuration flags
6214 * OFPTC_TABLE_MISS_CONTROLLER, default.
6215 * OFPTC_TABLE_MISS_CONTINUE,
6216 * OFPTC_TABLE_MISS_DROP
6217 * When OF1.0, OFPTC_TABLE_MISS_CONTINUE is used. What to do?
6219 rule = rule_dpif_miss_rule(ctx->ofproto, &ctx->xin->flow);
6221 if (rule && ctx->xin->resubmit_stats) {
6222 rule_credit_stats(rule, ctx->xin->resubmit_stats);
6228 xlate_table_action(struct xlate_ctx *ctx,
6229 uint16_t in_port, uint8_t table_id, bool may_packet_in)
6231 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
6232 struct rule_dpif *rule;
6233 uint16_t old_in_port = ctx->xin->flow.in_port;
6234 uint8_t old_table_id = ctx->table_id;
6236 ctx->table_id = table_id;
6238 /* Look up a flow with 'in_port' as the input port. */
6239 ctx->xin->flow.in_port = in_port;
6240 rule = rule_dpif_lookup__(ctx->ofproto, &ctx->xin->flow,
6241 &ctx->xout->wc, table_id);
6243 tag_the_flow(ctx, rule);
6245 /* Restore the original input port. Otherwise OFPP_NORMAL and
6246 * OFPP_IN_PORT will have surprising behavior. */
6247 ctx->xin->flow.in_port = old_in_port;
6249 rule = ctx_rule_hooks(ctx, rule, may_packet_in);
6252 struct rule_dpif *old_rule = ctx->rule;
6256 do_xlate_actions(rule->up.ofpacts, rule->up.ofpacts_len, ctx);
6257 ctx->rule = old_rule;
6261 ctx->table_id = old_table_id;
6263 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
6265 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
6266 MAX_RESUBMIT_RECURSION);
6267 ctx->max_resubmit_trigger = true;
6272 xlate_ofpact_resubmit(struct xlate_ctx *ctx,
6273 const struct ofpact_resubmit *resubmit)
6278 in_port = resubmit->in_port;
6279 if (in_port == OFPP_IN_PORT) {
6280 in_port = ctx->xin->flow.in_port;
6283 table_id = resubmit->table_id;
6284 if (table_id == 255) {
6285 table_id = ctx->table_id;
6288 xlate_table_action(ctx, in_port, table_id, false);
6292 flood_packets(struct xlate_ctx *ctx, bool all)
6294 struct ofport_dpif *ofport;
6296 HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) {
6297 uint16_t ofp_port = ofport->up.ofp_port;
6299 if (ofp_port == ctx->xin->flow.in_port) {
6304 compose_output_action__(ctx, ofp_port, false);
6305 } else if (!(ofport->up.pp.config & OFPUTIL_PC_NO_FLOOD)) {
6306 compose_output_action(ctx, ofp_port);
6310 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
6314 execute_controller_action(struct xlate_ctx *ctx, int len,
6315 enum ofp_packet_in_reason reason,
6316 uint16_t controller_id)
6318 struct ofputil_packet_in pin;
6319 struct ofpbuf *packet;
6321 ovs_assert(!ctx->xout->slow || ctx->xout->slow == SLOW_CONTROLLER);
6322 ctx->xout->slow = SLOW_CONTROLLER;
6323 if (!ctx->xin->packet) {
6327 packet = ofpbuf_clone(ctx->xin->packet);
6329 if (packet->l2 && packet->l3) {
6330 struct eth_header *eh;
6331 uint16_t mpls_depth;
6333 eth_pop_vlan(packet);
6336 memcpy(eh->eth_src, ctx->xin->flow.dl_src, sizeof eh->eth_src);
6337 memcpy(eh->eth_dst, ctx->xin->flow.dl_dst, sizeof eh->eth_dst);
6339 if (ctx->xin->flow.vlan_tci & htons(VLAN_CFI)) {
6340 eth_push_vlan(packet, ctx->xin->flow.vlan_tci);
6343 mpls_depth = eth_mpls_depth(packet);
6345 if (mpls_depth < ctx->xin->flow.mpls_depth) {
6346 push_mpls(packet, ctx->xin->flow.dl_type, ctx->xin->flow.mpls_lse);
6347 } else if (mpls_depth > ctx->xin->flow.mpls_depth) {
6348 pop_mpls(packet, ctx->xin->flow.dl_type);
6349 } else if (mpls_depth) {
6350 set_mpls_lse(packet, ctx->xin->flow.mpls_lse);
6354 if (ctx->xin->flow.dl_type == htons(ETH_TYPE_IP)) {
6355 packet_set_ipv4(packet, ctx->xin->flow.nw_src,
6356 ctx->xin->flow.nw_dst, ctx->xin->flow.nw_tos,
6357 ctx->xin->flow.nw_ttl);
6361 if (ctx->xin->flow.nw_proto == IPPROTO_TCP) {
6362 packet_set_tcp_port(packet, ctx->xin->flow.tp_src,
6363 ctx->xin->flow.tp_dst);
6364 } else if (ctx->xin->flow.nw_proto == IPPROTO_UDP) {
6365 packet_set_udp_port(packet, ctx->xin->flow.tp_src,
6366 ctx->xin->flow.tp_dst);
6372 pin.packet = packet->data;
6373 pin.packet_len = packet->size;
6374 pin.reason = reason;
6375 pin.controller_id = controller_id;
6376 pin.table_id = ctx->table_id;
6377 pin.cookie = ctx->rule ? ctx->rule->up.flow_cookie : 0;
6380 flow_get_metadata(&ctx->xin->flow, &pin.fmd);
6382 connmgr_send_packet_in(ctx->ofproto->up.connmgr, &pin);
6383 ofpbuf_delete(packet);
6387 execute_mpls_push_action(struct xlate_ctx *ctx, ovs_be16 eth_type)
6389 struct flow_wildcards *wc = &ctx->xout->wc;
6390 ovs_assert(eth_type_mpls(eth_type));
6392 memset(&wc->masks.mpls_lse, 0xff, sizeof wc->masks.mpls_lse);
6393 memset(&wc->masks.mpls_depth, 0xff, sizeof wc->masks.mpls_depth);
6395 if (ctx->base_flow.mpls_depth) {
6396 ctx->xin->flow.mpls_lse &= ~htonl(MPLS_BOS_MASK);
6397 ctx->xin->flow.mpls_depth++;
6402 if (ctx->xin->flow.dl_type == htons(ETH_TYPE_IPV6)) {
6403 label = htonl(0x2); /* IPV6 Explicit Null. */
6405 label = htonl(0x0); /* IPV4 Explicit Null. */
6407 wc->masks.nw_tos |= IP_DSCP_MASK;
6408 wc->masks.nw_ttl = 0xff;
6409 tc = (ctx->xin->flow.nw_tos & IP_DSCP_MASK) >> 2;
6410 ttl = ctx->xin->flow.nw_ttl ? ctx->xin->flow.nw_ttl : 0x40;
6411 ctx->xin->flow.mpls_lse = set_mpls_lse_values(ttl, tc, 1, label);
6412 ctx->xin->flow.mpls_depth = 1;
6414 ctx->xin->flow.dl_type = eth_type;
6418 execute_mpls_pop_action(struct xlate_ctx *ctx, ovs_be16 eth_type)
6420 struct flow_wildcards *wc = &ctx->xout->wc;
6422 ovs_assert(eth_type_mpls(ctx->xin->flow.dl_type));
6423 ovs_assert(!eth_type_mpls(eth_type));
6425 memset(&wc->masks.mpls_lse, 0xff, sizeof wc->masks.mpls_lse);
6426 memset(&wc->masks.mpls_depth, 0xff, sizeof wc->masks.mpls_depth);
6428 if (ctx->xin->flow.mpls_depth) {
6429 ctx->xin->flow.mpls_depth--;
6430 ctx->xin->flow.mpls_lse = htonl(0);
6431 if (!ctx->xin->flow.mpls_depth) {
6432 ctx->xin->flow.dl_type = eth_type;
6438 compose_dec_ttl(struct xlate_ctx *ctx, struct ofpact_cnt_ids *ids)
6440 if (ctx->xin->flow.dl_type != htons(ETH_TYPE_IP) &&
6441 ctx->xin->flow.dl_type != htons(ETH_TYPE_IPV6)) {
6445 ctx->xout->wc.masks.nw_ttl = 0xff;
6446 if (ctx->xin->flow.nw_ttl > 1) {
6447 ctx->xin->flow.nw_ttl--;
6452 for (i = 0; i < ids->n_controllers; i++) {
6453 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL,
6457 /* Stop processing for current table. */
6463 execute_set_mpls_ttl_action(struct xlate_ctx *ctx, uint8_t ttl)
6465 if (!eth_type_mpls(ctx->xin->flow.dl_type)) {
6469 set_mpls_lse_ttl(&ctx->xin->flow.mpls_lse, ttl);
6474 execute_dec_mpls_ttl_action(struct xlate_ctx *ctx)
6476 uint8_t ttl = mpls_lse_to_ttl(ctx->xin->flow.mpls_lse);
6477 struct flow_wildcards *wc = &ctx->xout->wc;
6479 memset(&wc->masks.mpls_lse, 0xff, sizeof wc->masks.mpls_lse);
6481 if (!eth_type_mpls(ctx->xin->flow.dl_type)) {
6487 set_mpls_lse_ttl(&ctx->xin->flow.mpls_lse, ttl);
6490 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL, 0);
6492 /* Stop processing for current table. */
6498 xlate_output_action(struct xlate_ctx *ctx,
6499 uint16_t port, uint16_t max_len, bool may_packet_in)
6501 uint16_t prev_nf_output_iface = ctx->xout->nf_output_iface;
6503 ctx->xout->nf_output_iface = NF_OUT_DROP;
6507 compose_output_action(ctx, ctx->xin->flow.in_port);
6510 xlate_table_action(ctx, ctx->xin->flow.in_port, 0, may_packet_in);
6516 flood_packets(ctx, false);
6519 flood_packets(ctx, true);
6521 case OFPP_CONTROLLER:
6522 execute_controller_action(ctx, max_len, OFPR_ACTION, 0);
6528 if (port != ctx->xin->flow.in_port) {
6529 compose_output_action(ctx, port);
6531 xlate_report(ctx, "skipping output to input port");
6536 if (prev_nf_output_iface == NF_OUT_FLOOD) {
6537 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
6538 } else if (ctx->xout->nf_output_iface == NF_OUT_DROP) {
6539 ctx->xout->nf_output_iface = prev_nf_output_iface;
6540 } else if (prev_nf_output_iface != NF_OUT_DROP &&
6541 ctx->xout->nf_output_iface != NF_OUT_FLOOD) {
6542 ctx->xout->nf_output_iface = NF_OUT_MULTI;
6547 xlate_output_reg_action(struct xlate_ctx *ctx,
6548 const struct ofpact_output_reg *or)
6550 uint64_t port = mf_get_subfield(&or->src, &ctx->xin->flow);
6551 if (port <= UINT16_MAX) {
6552 union mf_subvalue value;
6554 memset(&value, 0xff, sizeof value);
6555 mf_write_subfield_flow(&or->src, &value, &ctx->xout->wc.masks);
6556 xlate_output_action(ctx, port, or->max_len, false);
6561 xlate_enqueue_action(struct xlate_ctx *ctx,
6562 const struct ofpact_enqueue *enqueue)
6564 uint16_t ofp_port = enqueue->port;
6565 uint32_t queue_id = enqueue->queue;
6566 uint32_t flow_priority, priority;
6569 /* Translate queue to priority. */
6570 error = dpif_queue_to_priority(ctx->ofproto->backer->dpif,
6571 queue_id, &priority);
6573 /* Fall back to ordinary output action. */
6574 xlate_output_action(ctx, enqueue->port, 0, false);
6578 /* Check output port. */
6579 if (ofp_port == OFPP_IN_PORT) {
6580 ofp_port = ctx->xin->flow.in_port;
6581 } else if (ofp_port == ctx->xin->flow.in_port) {
6585 /* Add datapath actions. */
6586 flow_priority = ctx->xin->flow.skb_priority;
6587 ctx->xin->flow.skb_priority = priority;
6588 compose_output_action(ctx, ofp_port);
6589 ctx->xin->flow.skb_priority = flow_priority;
6591 /* Update NetFlow output port. */
6592 if (ctx->xout->nf_output_iface == NF_OUT_DROP) {
6593 ctx->xout->nf_output_iface = ofp_port;
6594 } else if (ctx->xout->nf_output_iface != NF_OUT_FLOOD) {
6595 ctx->xout->nf_output_iface = NF_OUT_MULTI;
6600 xlate_set_queue_action(struct xlate_ctx *ctx, uint32_t queue_id)
6602 uint32_t skb_priority;
6604 if (!dpif_queue_to_priority(ctx->ofproto->backer->dpif,
6605 queue_id, &skb_priority)) {
6606 ctx->xin->flow.skb_priority = skb_priority;
6608 /* Couldn't translate queue to a priority. Nothing to do. A warning
6609 * has already been logged. */
6614 slave_enabled_cb(uint16_t ofp_port, void *ofproto_)
6616 struct ofproto_dpif *ofproto = ofproto_;
6617 struct ofport_dpif *port;
6627 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
6630 port = get_ofp_port(ofproto, ofp_port);
6631 return port ? port->may_enable : false;
6636 xlate_bundle_action(struct xlate_ctx *ctx,
6637 const struct ofpact_bundle *bundle)
6641 port = bundle_execute(bundle, &ctx->xin->flow, &ctx->xout->wc,
6642 slave_enabled_cb, ctx->ofproto);
6643 if (bundle->dst.field) {
6644 nxm_reg_load(&bundle->dst, port, &ctx->xin->flow);
6646 xlate_output_action(ctx, port, 0, false);
6651 xlate_learn_action(struct xlate_ctx *ctx,
6652 const struct ofpact_learn *learn)
6654 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
6655 struct ofputil_flow_mod fm;
6656 uint64_t ofpacts_stub[1024 / 8];
6657 struct ofpbuf ofpacts;
6660 ctx->xout->has_learn = true;
6662 learn_mask(learn, &ctx->xout->wc);
6664 if (!ctx->xin->may_learn) {
6668 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
6669 learn_execute(learn, &ctx->xin->flow, &fm, &ofpacts);
6671 error = ofproto_flow_mod(&ctx->ofproto->up, &fm);
6672 if (error && !VLOG_DROP_WARN(&rl)) {
6673 VLOG_WARN("learning action failed to modify flow table (%s)",
6674 ofperr_get_name(error));
6677 ofpbuf_uninit(&ofpacts);
6680 /* Reduces '*timeout' to no more than 'max'. A value of zero in either case
6681 * means "infinite". */
6683 reduce_timeout(uint16_t max, uint16_t *timeout)
6685 if (max && (!*timeout || *timeout > max)) {
6691 xlate_fin_timeout(struct xlate_ctx *ctx,
6692 const struct ofpact_fin_timeout *oft)
6694 if (ctx->xin->tcp_flags & (TCP_FIN | TCP_RST) && ctx->rule) {
6695 struct rule_dpif *rule = ctx->rule;
6697 reduce_timeout(oft->fin_idle_timeout, &rule->up.idle_timeout);
6698 reduce_timeout(oft->fin_hard_timeout, &rule->up.hard_timeout);
6703 xlate_sample_action(struct xlate_ctx *ctx,
6704 const struct ofpact_sample *os)
6706 union user_action_cookie cookie;
6707 /* Scale the probability from 16-bit to 32-bit while representing
6708 * the same percentage. */
6709 uint32_t probability = (os->probability << 16) | os->probability;
6711 commit_odp_actions(&ctx->xin->flow, &ctx->base_flow,
6712 &ctx->xout->odp_actions, &ctx->xout->wc);
6714 compose_flow_sample_cookie(os->probability, os->collector_set_id,
6715 os->obs_domain_id, os->obs_point_id, &cookie);
6716 compose_sample_action(ctx->ofproto, &ctx->xout->odp_actions, &ctx->xin->flow,
6717 probability, &cookie, sizeof cookie.flow_sample);
6721 may_receive(const struct ofport_dpif *port, struct xlate_ctx *ctx)
6723 if (port->up.pp.config & (eth_addr_equals(ctx->xin->flow.dl_dst,
6725 ? OFPUTIL_PC_NO_RECV_STP
6726 : OFPUTIL_PC_NO_RECV)) {
6730 /* Only drop packets here if both forwarding and learning are
6731 * disabled. If just learning is enabled, we need to have
6732 * OFPP_NORMAL and the learning action have a look at the packet
6733 * before we can drop it. */
6734 if (!stp_forward_in_state(port->stp_state)
6735 && !stp_learn_in_state(port->stp_state)) {
6743 tunnel_ecn_ok(struct xlate_ctx *ctx)
6745 if (is_ip_any(&ctx->base_flow)
6746 && (ctx->base_flow.tunnel.ip_tos & IP_ECN_MASK) == IP_ECN_CE) {
6747 if ((ctx->base_flow.nw_tos & IP_ECN_MASK) == IP_ECN_NOT_ECT) {
6748 VLOG_WARN_RL(&rl, "dropping tunnel packet marked ECN CE"
6749 " but is not ECN capable");
6752 /* Set the ECN CE value in the tunneled packet. */
6753 ctx->xin->flow.nw_tos |= IP_ECN_CE;
6761 do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len,
6762 struct xlate_ctx *ctx)
6764 bool was_evictable = true;
6765 const struct ofpact *a;
6768 /* Don't let the rule we're working on get evicted underneath us. */
6769 was_evictable = ctx->rule->up.evictable;
6770 ctx->rule->up.evictable = false;
6773 do_xlate_actions_again:
6774 OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) {
6775 struct ofpact_controller *controller;
6776 const struct ofpact_metadata *metadata;
6784 xlate_output_action(ctx, ofpact_get_OUTPUT(a)->port,
6785 ofpact_get_OUTPUT(a)->max_len, true);
6788 case OFPACT_CONTROLLER:
6789 controller = ofpact_get_CONTROLLER(a);
6790 execute_controller_action(ctx, controller->max_len,
6792 controller->controller_id);
6795 case OFPACT_ENQUEUE:
6796 xlate_enqueue_action(ctx, ofpact_get_ENQUEUE(a));
6799 case OFPACT_SET_VLAN_VID:
6800 ctx->xin->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
6801 ctx->xin->flow.vlan_tci |=
6802 (htons(ofpact_get_SET_VLAN_VID(a)->vlan_vid)
6806 case OFPACT_SET_VLAN_PCP:
6807 ctx->xin->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
6808 ctx->xin->flow.vlan_tci |=
6809 htons((ofpact_get_SET_VLAN_PCP(a)->vlan_pcp << VLAN_PCP_SHIFT)
6813 case OFPACT_STRIP_VLAN:
6814 ctx->xin->flow.vlan_tci = htons(0);
6817 case OFPACT_PUSH_VLAN:
6818 /* XXX 802.1AD(QinQ) */
6819 ctx->xin->flow.vlan_tci = htons(VLAN_CFI);
6822 case OFPACT_SET_ETH_SRC:
6823 memcpy(ctx->xin->flow.dl_src, ofpact_get_SET_ETH_SRC(a)->mac,
6827 case OFPACT_SET_ETH_DST:
6828 memcpy(ctx->xin->flow.dl_dst, ofpact_get_SET_ETH_DST(a)->mac,
6832 case OFPACT_SET_IPV4_SRC:
6833 if (ctx->xin->flow.dl_type == htons(ETH_TYPE_IP)) {
6834 ctx->xin->flow.nw_src = ofpact_get_SET_IPV4_SRC(a)->ipv4;
6838 case OFPACT_SET_IPV4_DST:
6839 if (ctx->xin->flow.dl_type == htons(ETH_TYPE_IP)) {
6840 ctx->xin->flow.nw_dst = ofpact_get_SET_IPV4_DST(a)->ipv4;
6844 case OFPACT_SET_IPV4_DSCP:
6845 /* OpenFlow 1.0 only supports IPv4. */
6846 if (ctx->xin->flow.dl_type == htons(ETH_TYPE_IP)) {
6847 ctx->xin->flow.nw_tos &= ~IP_DSCP_MASK;
6848 ctx->xin->flow.nw_tos |= ofpact_get_SET_IPV4_DSCP(a)->dscp;
6852 case OFPACT_SET_L4_SRC_PORT:
6853 memset(&ctx->xout->wc.masks.nw_proto, 0xff,
6854 sizeof ctx->xout->wc.masks.nw_proto);
6855 if (is_ip_any(&ctx->xin->flow)) {
6856 ctx->xin->flow.tp_src =
6857 htons(ofpact_get_SET_L4_SRC_PORT(a)->port);
6861 case OFPACT_SET_L4_DST_PORT:
6862 memset(&ctx->xout->wc.masks.nw_proto, 0xff,
6863 sizeof ctx->xout->wc.masks.nw_proto);
6864 if (is_ip_any(&ctx->xin->flow)) {
6865 ctx->xin->flow.tp_dst =
6866 htons(ofpact_get_SET_L4_DST_PORT(a)->port);
6870 case OFPACT_RESUBMIT:
6871 xlate_ofpact_resubmit(ctx, ofpact_get_RESUBMIT(a));
6874 case OFPACT_SET_TUNNEL:
6875 ctx->xin->flow.tunnel.tun_id =
6876 htonll(ofpact_get_SET_TUNNEL(a)->tun_id);
6879 case OFPACT_SET_QUEUE:
6880 xlate_set_queue_action(ctx, ofpact_get_SET_QUEUE(a)->queue_id);
6883 case OFPACT_POP_QUEUE:
6884 ctx->xin->flow.skb_priority = ctx->orig_skb_priority;
6887 case OFPACT_REG_MOVE:
6888 nxm_execute_reg_move(ofpact_get_REG_MOVE(a), &ctx->xin->flow,
6892 case OFPACT_REG_LOAD:
6893 nxm_execute_reg_load(ofpact_get_REG_LOAD(a), &ctx->xin->flow);
6896 case OFPACT_STACK_PUSH:
6897 nxm_execute_stack_push(ofpact_get_STACK_PUSH(a), &ctx->xin->flow,
6898 &ctx->xout->wc, &ctx->stack);
6901 case OFPACT_STACK_POP:
6902 nxm_execute_stack_pop(ofpact_get_STACK_POP(a), &ctx->xin->flow,
6906 case OFPACT_PUSH_MPLS:
6907 execute_mpls_push_action(ctx, ofpact_get_PUSH_MPLS(a)->ethertype);
6910 case OFPACT_POP_MPLS:
6911 execute_mpls_pop_action(ctx, ofpact_get_POP_MPLS(a)->ethertype);
6914 case OFPACT_SET_MPLS_TTL:
6915 if (execute_set_mpls_ttl_action(ctx,
6916 ofpact_get_SET_MPLS_TTL(a)->ttl)) {
6921 case OFPACT_DEC_MPLS_TTL:
6922 if (execute_dec_mpls_ttl_action(ctx)) {
6927 case OFPACT_DEC_TTL:
6928 if (compose_dec_ttl(ctx, ofpact_get_DEC_TTL(a))) {
6934 /* Nothing to do. */
6937 case OFPACT_MULTIPATH:
6938 multipath_execute(ofpact_get_MULTIPATH(a), &ctx->xin->flow,
6943 ctx->ofproto->has_bundle_action = true;
6944 xlate_bundle_action(ctx, ofpact_get_BUNDLE(a));
6947 case OFPACT_OUTPUT_REG:
6948 xlate_output_reg_action(ctx, ofpact_get_OUTPUT_REG(a));
6952 xlate_learn_action(ctx, ofpact_get_LEARN(a));
6959 case OFPACT_FIN_TIMEOUT:
6960 memset(&ctx->xout->wc.masks.nw_proto, 0xff,
6961 sizeof ctx->xout->wc.masks.nw_proto);
6962 ctx->xout->has_fin_timeout = true;
6963 xlate_fin_timeout(ctx, ofpact_get_FIN_TIMEOUT(a));
6966 case OFPACT_CLEAR_ACTIONS:
6968 * Nothing to do because writa-actions is not supported for now.
6969 * When writa-actions is supported, clear-actions also must
6970 * be supported at the same time.
6974 case OFPACT_WRITE_METADATA:
6975 metadata = ofpact_get_WRITE_METADATA(a);
6976 ctx->xin->flow.metadata &= ~metadata->mask;
6977 ctx->xin->flow.metadata |= metadata->metadata & metadata->mask;
6980 case OFPACT_GOTO_TABLE: {
6981 /* It is assumed that goto-table is the last action. */
6982 struct ofpact_goto_table *ogt = ofpact_get_GOTO_TABLE(a);
6983 struct rule_dpif *rule;
6985 ovs_assert(ctx->table_id < ogt->table_id);
6987 ctx->table_id = ogt->table_id;
6989 /* Look up a flow from the new table. */
6990 rule = rule_dpif_lookup__(ctx->ofproto, &ctx->xin->flow,
6991 &ctx->xout->wc, ctx->table_id);
6993 tag_the_flow(ctx, rule);
6995 rule = ctx_rule_hooks(ctx, rule, true);
6999 ctx->rule->up.evictable = was_evictable;
7002 was_evictable = rule->up.evictable;
7003 rule->up.evictable = false;
7005 /* Tail recursion removal. */
7006 ofpacts = rule->up.ofpacts;
7007 ofpacts_len = rule->up.ofpacts_len;
7008 goto do_xlate_actions_again;
7014 xlate_sample_action(ctx, ofpact_get_SAMPLE(a));
7021 ctx->rule->up.evictable = was_evictable;
7026 xlate_in_init(struct xlate_in *xin, struct ofproto_dpif *ofproto,
7027 const struct flow *flow,
7028 const struct initial_vals *initial_vals,
7029 struct rule_dpif *rule, uint8_t tcp_flags,
7030 const struct ofpbuf *packet)
7032 xin->ofproto = ofproto;
7034 xin->packet = packet;
7035 xin->may_learn = packet != NULL;
7037 xin->ofpacts = NULL;
7038 xin->ofpacts_len = 0;
7039 xin->tcp_flags = tcp_flags;
7040 xin->resubmit_hook = NULL;
7041 xin->report_hook = NULL;
7042 xin->resubmit_stats = NULL;
7045 xin->initial_vals = *initial_vals;
7047 xin->initial_vals.vlan_tci = xin->flow.vlan_tci;
7048 xin->initial_vals.tunnel_ip_tos = xin->flow.tunnel.ip_tos;
7053 xlate_out_uninit(struct xlate_out *xout)
7056 ofpbuf_uninit(&xout->odp_actions);
7060 /* Translates the 'ofpacts_len' bytes of "struct ofpacts" starting at 'ofpacts'
7061 * into datapath actions in 'odp_actions', using 'ctx'. */
7063 xlate_actions(struct xlate_in *xin, struct xlate_out *xout)
7065 /* Normally false. Set to true if we ever hit MAX_RESUBMIT_RECURSION, so
7066 * that in the future we always keep a copy of the original flow for
7067 * tracing purposes. */
7068 static bool hit_resubmit_limit;
7069 struct flow_wildcards *wc = &xout->wc;
7071 enum slow_path_reason special;
7072 const struct ofpact *ofpacts;
7073 struct ofport_dpif *in_port;
7074 struct flow orig_flow;
7075 struct xlate_ctx ctx;
7078 COVERAGE_INC(ofproto_dpif_xlate);
7080 /* Flow initialization rules:
7081 * - 'base_flow' must match the kernel's view of the packet at the
7082 * time that action processing starts. 'flow' represents any
7083 * transformations we wish to make through actions.
7084 * - By default 'base_flow' and 'flow' are the same since the input
7085 * packet matches the output before any actions are applied.
7086 * - When using VLAN splinters, 'base_flow''s VLAN is set to the value
7087 * of the received packet as seen by the kernel. If we later output
7088 * to another device without any modifications this will cause us to
7089 * insert a new tag since the original one was stripped off by the
7091 * - Tunnel 'flow' is largely cleared when transitioning between
7092 * the input and output stages since it does not make sense to output
7093 * a packet with the exact headers that it was received with (i.e.
7094 * the destination IP is us). The one exception is the tun_id, which
7095 * is preserved to allow use in later resubmit lookups and loads into
7097 * - Tunnel 'base_flow' is completely cleared since that is what the
7098 * kernel does. If we wish to maintain the original values an action
7099 * needs to be generated. */
7104 ctx.ofproto = xin->ofproto;
7105 ctx.rule = xin->rule;
7107 ctx.base_flow = ctx.xin->flow;
7108 memset(&ctx.base_flow.tunnel, 0, sizeof ctx.base_flow.tunnel);
7109 ctx.base_flow.vlan_tci = xin->initial_vals.vlan_tci;
7110 ctx.base_flow.tunnel.ip_tos = xin->initial_vals.tunnel_ip_tos;
7112 flow_wildcards_init_catchall(&ctx.xout->wc);
7113 memset(&wc->masks.in_port, 0xff, sizeof wc->masks.in_port);
7114 memset(&wc->masks.skb_priority, 0xff, sizeof wc->masks.skb_priority);
7115 memset(&wc->masks.dl_type, 0xff, sizeof wc->masks.dl_type);
7116 wc->masks.nw_frag |= FLOW_NW_FRAG_MASK;
7118 if (tnl_port_should_receive(&ctx.xin->flow)) {
7119 memset(&wc->masks.tunnel, 0xff, sizeof wc->masks.tunnel);
7122 /* Disable most wildcarding for NetFlow. */
7123 if (xin->ofproto->netflow) {
7124 netflow_mask_wc(wc);
7129 ctx.xout->has_learn = false;
7130 ctx.xout->has_normal = false;
7131 ctx.xout->has_fin_timeout = false;
7132 ctx.xout->nf_output_iface = NF_OUT_DROP;
7133 ctx.xout->mirrors = 0;
7135 ofpbuf_use_stub(&ctx.xout->odp_actions, ctx.xout->odp_actions_stub,
7136 sizeof ctx.xout->odp_actions_stub);
7137 ofpbuf_reserve(&ctx.xout->odp_actions, NL_A_U32_SIZE);
7140 ctx.max_resubmit_trigger = false;
7141 ctx.orig_skb_priority = ctx.xin->flow.skb_priority;
7146 ofpacts = xin->ofpacts;
7147 ofpacts_len = xin->ofpacts_len;
7148 } else if (xin->rule) {
7149 ofpacts = xin->rule->up.ofpacts;
7150 ofpacts_len = xin->rule->up.ofpacts_len;
7155 ofpbuf_use_stub(&ctx.stack, ctx.init_stack, sizeof ctx.init_stack);
7157 if (ctx.ofproto->has_mirrors || hit_resubmit_limit) {
7158 /* Do this conditionally because the copy is expensive enough that it
7159 * shows up in profiles. */
7160 orig_flow = ctx.xin->flow;
7163 if (ctx.xin->flow.nw_frag & FLOW_NW_FRAG_ANY) {
7164 switch (ctx.ofproto->up.frag_handling) {
7165 case OFPC_FRAG_NORMAL:
7166 /* We must pretend that transport ports are unavailable. */
7167 ctx.xin->flow.tp_src = ctx.base_flow.tp_src = htons(0);
7168 ctx.xin->flow.tp_dst = ctx.base_flow.tp_dst = htons(0);
7171 case OFPC_FRAG_DROP:
7174 case OFPC_FRAG_REASM:
7177 case OFPC_FRAG_NX_MATCH:
7178 /* Nothing to do. */
7181 case OFPC_INVALID_TTL_TO_CONTROLLER:
7186 in_port = get_ofp_port(ctx.ofproto, ctx.xin->flow.in_port);
7187 special = process_special(&ctx, &ctx.xin->flow, in_port,
7190 ctx.xout->slow = special;
7192 static struct vlog_rate_limit trace_rl = VLOG_RATE_LIMIT_INIT(1, 1);
7193 struct initial_vals initial_vals;
7194 size_t sample_actions_len;
7195 uint32_t local_odp_port;
7197 initial_vals.vlan_tci = ctx.base_flow.vlan_tci;
7198 initial_vals.tunnel_ip_tos = ctx.base_flow.tunnel.ip_tos;
7200 add_sflow_action(&ctx);
7201 add_ipfix_action(&ctx);
7202 sample_actions_len = ctx.xout->odp_actions.size;
7204 if (tunnel_ecn_ok(&ctx) && (!in_port || may_receive(in_port, &ctx))) {
7205 do_xlate_actions(ofpacts, ofpacts_len, &ctx);
7207 /* We've let OFPP_NORMAL and the learning action look at the
7208 * packet, so drop it now if forwarding is disabled. */
7209 if (in_port && !stp_forward_in_state(in_port->stp_state)) {
7210 ctx.xout->odp_actions.size = sample_actions_len;
7214 if (ctx.max_resubmit_trigger && !ctx.xin->resubmit_hook) {
7215 if (!hit_resubmit_limit) {
7216 /* We didn't record the original flow. Make sure we do from
7218 hit_resubmit_limit = true;
7219 } else if (!VLOG_DROP_ERR(&trace_rl)) {
7220 struct ds ds = DS_EMPTY_INITIALIZER;
7222 ofproto_trace(ctx.ofproto, &orig_flow, ctx.xin->packet,
7223 &initial_vals, &ds);
7224 VLOG_ERR("Trace triggered by excessive resubmit "
7225 "recursion:\n%s", ds_cstr(&ds));
7230 local_odp_port = ofp_port_to_odp_port(ctx.ofproto, OFPP_LOCAL);
7231 if (!connmgr_must_output_local(ctx.ofproto->up.connmgr, &ctx.xin->flow,
7233 ctx.xout->odp_actions.data,
7234 ctx.xout->odp_actions.size)) {
7235 compose_output_action(&ctx, OFPP_LOCAL);
7237 if (ctx.ofproto->has_mirrors) {
7238 add_mirror_actions(&ctx, &orig_flow);
7240 fix_sflow_action(&ctx);
7243 ofpbuf_uninit(&ctx.stack);
7245 /* Clear the metadata and register wildcard masks, because we won't
7246 * use non-header fields as part of the cache. */
7247 memset(&ctx.xout->wc.masks.metadata, 0,
7248 sizeof ctx.xout->wc.masks.metadata);
7249 memset(&ctx.xout->wc.masks.regs, 0, sizeof ctx.xout->wc.masks.regs);
7252 /* Translates the 'ofpacts_len' bytes of "struct ofpact"s starting at 'ofpacts'
7253 * into datapath actions, using 'ctx', and discards the datapath actions. */
7255 xlate_actions_for_side_effects(struct xlate_in *xin)
7257 struct xlate_out xout;
7259 xlate_actions(xin, &xout);
7260 xlate_out_uninit(&xout);
7264 xlate_report(struct xlate_ctx *ctx, const char *s)
7266 if (ctx->xin->report_hook) {
7267 ctx->xin->report_hook(ctx, s);
7272 xlate_out_copy(struct xlate_out *dst, const struct xlate_out *src)
7275 dst->tags = src->tags;
7276 dst->slow = src->slow;
7277 dst->has_learn = src->has_learn;
7278 dst->has_normal = src->has_normal;
7279 dst->has_fin_timeout = src->has_fin_timeout;
7280 dst->nf_output_iface = src->nf_output_iface;
7281 dst->mirrors = src->mirrors;
7283 ofpbuf_use_stub(&dst->odp_actions, dst->odp_actions_stub,
7284 sizeof dst->odp_actions_stub);
7285 ofpbuf_put(&dst->odp_actions, src->odp_actions.data,
7286 src->odp_actions.size);
7289 /* OFPP_NORMAL implementation. */
7291 static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *);
7293 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
7294 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle',
7295 * the bundle on which the packet was received, returns the VLAN to which the
7298 * Both 'vid' and the return value are in the range 0...4095. */
7300 input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid)
7302 switch (in_bundle->vlan_mode) {
7303 case PORT_VLAN_ACCESS:
7304 return in_bundle->vlan;
7307 case PORT_VLAN_TRUNK:
7310 case PORT_VLAN_NATIVE_UNTAGGED:
7311 case PORT_VLAN_NATIVE_TAGGED:
7312 return vid ? vid : in_bundle->vlan;
7319 /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'.
7320 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
7323 * 'vid' should be the VID obtained from the 802.1Q header that was received as
7324 * part of a packet (specify 0 if there was no 802.1Q header), in the range
7327 input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn)
7329 /* Allow any VID on the OFPP_NONE port. */
7330 if (in_bundle == &ofpp_none_bundle) {
7334 switch (in_bundle->vlan_mode) {
7335 case PORT_VLAN_ACCESS:
7338 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7339 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
7340 "packet received on port %s configured as VLAN "
7341 "%"PRIu16" access port",
7342 in_bundle->ofproto->up.name, vid,
7343 in_bundle->name, in_bundle->vlan);
7349 case PORT_VLAN_NATIVE_UNTAGGED:
7350 case PORT_VLAN_NATIVE_TAGGED:
7352 /* Port must always carry its native VLAN. */
7356 case PORT_VLAN_TRUNK:
7357 if (!ofbundle_includes_vlan(in_bundle, vid)) {
7359 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7360 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet "
7361 "received on port %s not configured for trunking "
7363 in_bundle->ofproto->up.name, vid,
7364 in_bundle->name, vid);
7376 /* Given 'vlan', the VLAN that a packet belongs to, and
7377 * 'out_bundle', a bundle on which the packet is to be output, returns the VID
7378 * that should be included in the 802.1Q header. (If the return value is 0,
7379 * then the 802.1Q header should only be included in the packet if there is a
7382 * Both 'vlan' and the return value are in the range 0...4095. */
7384 output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan)
7386 switch (out_bundle->vlan_mode) {
7387 case PORT_VLAN_ACCESS:
7390 case PORT_VLAN_TRUNK:
7391 case PORT_VLAN_NATIVE_TAGGED:
7394 case PORT_VLAN_NATIVE_UNTAGGED:
7395 return vlan == out_bundle->vlan ? 0 : vlan;
7403 output_normal(struct xlate_ctx *ctx, const struct ofbundle *out_bundle,
7406 struct ofport_dpif *port;
7408 ovs_be16 tci, old_tci;
7410 vid = output_vlan_to_vid(out_bundle, vlan);
7411 if (!out_bundle->bond) {
7412 port = ofbundle_get_a_port(out_bundle);
7414 port = bond_choose_output_slave(out_bundle->bond, &ctx->xin->flow,
7415 &ctx->xout->wc, vid, &ctx->xout->tags);
7417 /* No slaves enabled, so drop packet. */
7422 old_tci = ctx->xin->flow.vlan_tci;
7424 if (tci || out_bundle->use_priority_tags) {
7425 tci |= ctx->xin->flow.vlan_tci & htons(VLAN_PCP_MASK);
7427 tci |= htons(VLAN_CFI);
7430 ctx->xin->flow.vlan_tci = tci;
7432 compose_output_action(ctx, port->up.ofp_port);
7433 ctx->xin->flow.vlan_tci = old_tci;
7437 mirror_mask_ffs(mirror_mask_t mask)
7439 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
7444 ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
7446 return (bundle->vlan_mode != PORT_VLAN_ACCESS
7447 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
7451 ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan)
7453 return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan);
7456 /* Returns an arbitrary interface within 'bundle'. */
7457 static struct ofport_dpif *
7458 ofbundle_get_a_port(const struct ofbundle *bundle)
7460 return CONTAINER_OF(list_front(&bundle->ports),
7461 struct ofport_dpif, bundle_node);
7465 vlan_is_mirrored(const struct ofmirror *m, int vlan)
7467 return !m->vlans || bitmap_is_set(m->vlans, vlan);
7471 add_mirror_actions(struct xlate_ctx *ctx, const struct flow *orig_flow)
7473 struct ofproto_dpif *ofproto = ctx->ofproto;
7474 mirror_mask_t mirrors;
7475 struct ofbundle *in_bundle;
7478 const struct nlattr *a;
7481 in_bundle = lookup_input_bundle(ctx->ofproto, orig_flow->in_port,
7482 ctx->xin->packet != NULL, NULL);
7486 mirrors = in_bundle->src_mirrors;
7488 /* Drop frames on bundles reserved for mirroring. */
7489 if (in_bundle->mirror_out) {
7490 if (ctx->xin->packet != NULL) {
7491 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7492 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
7493 "%s, which is reserved exclusively for mirroring",
7494 ctx->ofproto->up.name, in_bundle->name);
7500 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
7501 if (!input_vid_is_valid(vid, in_bundle, ctx->xin->packet != NULL)) {
7504 vlan = input_vid_to_vlan(in_bundle, vid);
7506 /* Look at the output ports to check for destination selections. */
7508 NL_ATTR_FOR_EACH (a, left, ctx->xout->odp_actions.data,
7509 ctx->xout->odp_actions.size) {
7510 enum ovs_action_attr type = nl_attr_type(a);
7511 struct ofport_dpif *ofport;
7513 if (type != OVS_ACTION_ATTR_OUTPUT) {
7517 ofport = get_odp_port(ofproto, nl_attr_get_u32(a));
7518 if (ofport && ofport->bundle) {
7519 mirrors |= ofport->bundle->dst_mirrors;
7527 /* Restore the original packet before adding the mirror actions. */
7528 ctx->xin->flow = *orig_flow;
7533 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
7536 ctx->xout->wc.masks.vlan_tci |= htons(VLAN_CFI | VLAN_VID_MASK);
7539 if (!vlan_is_mirrored(m, vlan)) {
7540 mirrors = zero_rightmost_1bit(mirrors);
7544 mirrors &= ~m->dup_mirrors;
7545 ctx->xout->mirrors |= m->dup_mirrors;
7547 output_normal(ctx, m->out, vlan);
7548 } else if (vlan != m->out_vlan
7549 && !eth_addr_is_reserved(orig_flow->dl_dst)) {
7550 struct ofbundle *bundle;
7552 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
7553 if (ofbundle_includes_vlan(bundle, m->out_vlan)
7554 && !bundle->mirror_out) {
7555 output_normal(ctx, bundle, m->out_vlan);
7563 update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors,
7564 uint64_t packets, uint64_t bytes)
7570 for (; mirrors; mirrors = zero_rightmost_1bit(mirrors)) {
7573 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
7576 /* In normal circumstances 'm' will not be NULL. However,
7577 * if mirrors are reconfigured, we can temporarily get out
7578 * of sync in facet_revalidate(). We could "correct" the
7579 * mirror list before reaching here, but doing that would
7580 * not properly account the traffic stats we've currently
7581 * accumulated for previous mirror configuration. */
7585 m->packet_count += packets;
7586 m->byte_count += bytes;
7590 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
7591 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
7592 * indicate this; newer upstream kernels use gratuitous ARP requests. */
7594 is_gratuitous_arp(const struct flow *flow, struct flow_wildcards *wc)
7596 if (flow->dl_type != htons(ETH_TYPE_ARP)) {
7600 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
7601 if (!eth_addr_is_broadcast(flow->dl_dst)) {
7605 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
7606 if (flow->nw_proto == ARP_OP_REPLY) {
7608 } else if (flow->nw_proto == ARP_OP_REQUEST) {
7609 memset(&wc->masks.nw_src, 0xff, sizeof wc->masks.nw_src);
7610 memset(&wc->masks.nw_dst, 0xff, sizeof wc->masks.nw_dst);
7612 return flow->nw_src == flow->nw_dst;
7619 update_learning_table(struct ofproto_dpif *ofproto,
7620 const struct flow *flow, struct flow_wildcards *wc,
7621 int vlan, struct ofbundle *in_bundle)
7623 struct mac_entry *mac;
7625 /* Don't learn the OFPP_NONE port. */
7626 if (in_bundle == &ofpp_none_bundle) {
7630 if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
7634 mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan);
7635 if (is_gratuitous_arp(flow, wc)) {
7636 /* We don't want to learn from gratuitous ARP packets that are
7637 * reflected back over bond slaves so we lock the learning table. */
7638 if (!in_bundle->bond) {
7639 mac_entry_set_grat_arp_lock(mac);
7640 } else if (mac_entry_is_grat_arp_locked(mac)) {
7645 if (mac_entry_is_new(mac) || mac->port.p != in_bundle) {
7646 /* The log messages here could actually be useful in debugging,
7647 * so keep the rate limit relatively high. */
7648 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
7649 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
7650 "on port %s in VLAN %d",
7651 ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src),
7652 in_bundle->name, vlan);
7654 mac->port.p = in_bundle;
7655 tag_set_add(&ofproto->backer->revalidate_set,
7656 mac_learning_changed(ofproto->ml, mac));
7660 static struct ofbundle *
7661 lookup_input_bundle(const struct ofproto_dpif *ofproto, uint16_t in_port,
7662 bool warn, struct ofport_dpif **in_ofportp)
7664 struct ofport_dpif *ofport;
7666 /* Find the port and bundle for the received packet. */
7667 ofport = get_ofp_port(ofproto, in_port);
7669 *in_ofportp = ofport;
7671 if (ofport && ofport->bundle) {
7672 return ofport->bundle;
7675 /* Special-case OFPP_NONE, which a controller may use as the ingress
7676 * port for traffic that it is sourcing. */
7677 if (in_port == OFPP_NONE) {
7678 return &ofpp_none_bundle;
7681 /* Odd. A few possible reasons here:
7683 * - We deleted a port but there are still a few packets queued up
7686 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
7687 * we don't know about.
7689 * - The ofproto client didn't configure the port as part of a bundle.
7690 * This is particularly likely to happen if a packet was received on the
7691 * port after it was created, but before the client had a chance to
7692 * configure its bundle.
7695 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7697 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
7698 "port %"PRIu16, ofproto->up.name, in_port);
7703 /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or
7704 * dropped. Returns true if they may be forwarded, false if they should be
7707 * 'in_port' must be the ofport_dpif that corresponds to flow->in_port.
7708 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
7710 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
7711 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
7712 * checked by input_vid_is_valid().
7714 * May also add tags to '*tags', although the current implementation only does
7715 * so in one special case.
7718 is_admissible(struct xlate_ctx *ctx, struct ofport_dpif *in_port,
7721 struct ofproto_dpif *ofproto = ctx->ofproto;
7722 struct flow *flow = &ctx->xin->flow;
7723 struct ofbundle *in_bundle = in_port->bundle;
7725 /* Drop frames for reserved multicast addresses
7726 * only if forward_bpdu option is absent. */
7727 if (!ofproto->up.forward_bpdu && eth_addr_is_reserved(flow->dl_dst)) {
7728 xlate_report(ctx, "packet has reserved destination MAC, dropping");
7732 if (in_bundle->bond) {
7733 struct mac_entry *mac;
7735 switch (bond_check_admissibility(in_bundle->bond, in_port,
7736 flow->dl_dst, &ctx->xout->tags)) {
7741 xlate_report(ctx, "bonding refused admissibility, dropping");
7744 case BV_DROP_IF_MOVED:
7745 mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL);
7746 if (mac && mac->port.p != in_bundle &&
7747 (!is_gratuitous_arp(flow, &ctx->xout->wc)
7748 || mac_entry_is_grat_arp_locked(mac))) {
7749 xlate_report(ctx, "SLB bond thinks this packet looped back, "
7761 xlate_normal(struct xlate_ctx *ctx)
7763 struct flow_wildcards *wc = &ctx->xout->wc;
7764 struct ofport_dpif *in_port;
7765 struct ofbundle *in_bundle;
7766 struct mac_entry *mac;
7770 ctx->xout->has_normal = true;
7772 /* Check the dl_type, since we may check for gratuituous ARP. */
7773 memset(&wc->masks.dl_src, 0xff, sizeof wc->masks.dl_src);
7774 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
7775 wc->masks.vlan_tci |= htons(VLAN_VID_MASK | VLAN_CFI);
7777 in_bundle = lookup_input_bundle(ctx->ofproto, ctx->xin->flow.in_port,
7778 ctx->xin->packet != NULL, &in_port);
7780 xlate_report(ctx, "no input bundle, dropping");
7784 /* Drop malformed frames. */
7785 if (ctx->xin->flow.dl_type == htons(ETH_TYPE_VLAN) &&
7786 !(ctx->xin->flow.vlan_tci & htons(VLAN_CFI))) {
7787 if (ctx->xin->packet != NULL) {
7788 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7789 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
7790 "VLAN tag received on port %s",
7791 ctx->ofproto->up.name, in_bundle->name);
7793 xlate_report(ctx, "partial VLAN tag, dropping");
7797 /* Drop frames on bundles reserved for mirroring. */
7798 if (in_bundle->mirror_out) {
7799 if (ctx->xin->packet != NULL) {
7800 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7801 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
7802 "%s, which is reserved exclusively for mirroring",
7803 ctx->ofproto->up.name, in_bundle->name);
7805 xlate_report(ctx, "input port is mirror output port, dropping");
7810 vid = vlan_tci_to_vid(ctx->xin->flow.vlan_tci);
7811 if (!input_vid_is_valid(vid, in_bundle, ctx->xin->packet != NULL)) {
7812 xlate_report(ctx, "disallowed VLAN VID for this input port, dropping");
7815 vlan = input_vid_to_vlan(in_bundle, vid);
7817 /* Check other admissibility requirements. */
7818 if (in_port && !is_admissible(ctx, in_port, vlan)) {
7822 /* Learn source MAC. */
7823 if (ctx->xin->may_learn) {
7824 update_learning_table(ctx->ofproto, &ctx->xin->flow, wc,
7828 /* Determine output bundle. */
7829 mac = mac_learning_lookup(ctx->ofproto->ml, ctx->xin->flow.dl_dst, vlan,
7832 if (mac->port.p != in_bundle) {
7833 xlate_report(ctx, "forwarding to learned port");
7834 output_normal(ctx, mac->port.p, vlan);
7836 xlate_report(ctx, "learned port is input port, dropping");
7839 struct ofbundle *bundle;
7841 xlate_report(ctx, "no learned MAC for destination, flooding");
7842 HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
7843 if (bundle != in_bundle
7844 && ofbundle_includes_vlan(bundle, vlan)
7845 && bundle->floodable
7846 && !bundle->mirror_out) {
7847 output_normal(ctx, bundle, vlan);
7850 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
7854 /* Optimized flow revalidation.
7856 * It's a difficult problem, in general, to tell which facets need to have
7857 * their actions recalculated whenever the OpenFlow flow table changes. We
7858 * don't try to solve that general problem: for most kinds of OpenFlow flow
7859 * table changes, we recalculate the actions for every facet. This is
7860 * relatively expensive, but it's good enough if the OpenFlow flow table
7861 * doesn't change very often.
7863 * However, we can expect one particular kind of OpenFlow flow table change to
7864 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
7865 * of CPU on revalidating every facet whenever MAC learning modifies the flow
7866 * table, we add a special case that applies to flow tables in which every rule
7867 * has the same form (that is, the same wildcards), except that the table is
7868 * also allowed to have a single "catch-all" flow that matches all packets. We
7869 * optimize this case by tagging all of the facets that resubmit into the table
7870 * and invalidating the same tag whenever a flow changes in that table. The
7871 * end result is that we revalidate just the facets that need it (and sometimes
7872 * a few more, but not all of the facets or even all of the facets that
7873 * resubmit to the table modified by MAC learning). */
7875 /* Calculates the tag to use for 'flow' and mask 'mask' when it is inserted
7876 * into an OpenFlow table with the given 'basis'. */
7878 rule_calculate_tag(const struct flow *flow, const struct minimask *mask,
7881 if (minimask_is_catchall(mask)) {
7884 uint32_t hash = flow_hash_in_minimask(flow, mask, secret);
7885 return tag_create_deterministic(hash);
7889 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
7890 * taggability of that table.
7892 * This function must be called after *each* change to a flow table. If you
7893 * skip calling it on some changes then the pointer comparisons at the end can
7894 * be invalid if you get unlucky. For example, if a flow removal causes a
7895 * cls_table to be destroyed and then a flow insertion causes a cls_table with
7896 * different wildcards to be created with the same address, then this function
7897 * will incorrectly skip revalidation. */
7899 table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
7901 struct table_dpif *table = &ofproto->tables[table_id];
7902 const struct oftable *oftable = &ofproto->up.tables[table_id];
7903 struct cls_table *catchall, *other;
7904 struct cls_table *t;
7906 catchall = other = NULL;
7908 switch (hmap_count(&oftable->cls.tables)) {
7910 /* We could tag this OpenFlow table but it would make the logic a
7911 * little harder and it's a corner case that doesn't seem worth it
7917 HMAP_FOR_EACH (t, hmap_node, &oftable->cls.tables) {
7918 if (cls_table_is_catchall(t)) {
7920 } else if (!other) {
7923 /* Indicate that we can't tag this by setting both tables to
7924 * NULL. (We know that 'catchall' is already NULL.) */
7931 /* Can't tag this table. */
7935 if (table->catchall_table != catchall || table->other_table != other) {
7936 table->catchall_table = catchall;
7937 table->other_table = other;
7938 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
7942 /* Given 'rule' that has changed in some way (either it is a rule being
7943 * inserted, a rule being deleted, or a rule whose actions are being
7944 * modified), marks facets for revalidation to ensure that packets will be
7945 * forwarded correctly according to the new state of the flow table.
7947 * This function must be called after *each* change to a flow table. See
7948 * the comment on table_update_taggable() for more information. */
7950 rule_invalidate(const struct rule_dpif *rule)
7952 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
7954 table_update_taggable(ofproto, rule->up.table_id);
7956 if (!ofproto->backer->need_revalidate) {
7957 struct table_dpif *table = &ofproto->tables[rule->up.table_id];
7959 if (table->other_table && rule->tag) {
7960 tag_set_add(&ofproto->backer->revalidate_set, rule->tag);
7962 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
7968 set_frag_handling(struct ofproto *ofproto_,
7969 enum ofp_config_flags frag_handling)
7971 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7972 if (frag_handling != OFPC_FRAG_REASM) {
7973 ofproto->backer->need_revalidate = REV_RECONFIGURE;
7981 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
7982 const struct flow *flow,
7983 const struct ofpact *ofpacts, size_t ofpacts_len)
7985 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7986 struct initial_vals initial_vals;
7987 struct odputil_keybuf keybuf;
7988 struct dpif_flow_stats stats;
7989 struct xlate_out xout;
7990 struct xlate_in xin;
7994 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
7995 odp_flow_key_from_flow(&key, flow,
7996 ofp_port_to_odp_port(ofproto, flow->in_port));
7998 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
8000 initial_vals.vlan_tci = flow->vlan_tci;
8001 initial_vals.tunnel_ip_tos = 0;
8002 xlate_in_init(&xin, ofproto, flow, &initial_vals, NULL, stats.tcp_flags,
8004 xin.resubmit_stats = &stats;
8005 xin.ofpacts_len = ofpacts_len;
8006 xin.ofpacts = ofpacts;
8008 xlate_actions(&xin, &xout);
8009 dpif_execute(ofproto->backer->dpif, key.data, key.size,
8010 xout.odp_actions.data, xout.odp_actions.size, packet);
8011 xlate_out_uninit(&xout);
8019 set_netflow(struct ofproto *ofproto_,
8020 const struct netflow_options *netflow_options)
8022 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
8024 if (netflow_options) {
8025 if (!ofproto->netflow) {
8026 ofproto->netflow = netflow_create();
8028 return netflow_set_options(ofproto->netflow, netflow_options);
8030 netflow_destroy(ofproto->netflow);
8031 ofproto->netflow = NULL;
8037 get_netflow_ids(const struct ofproto *ofproto_,
8038 uint8_t *engine_type, uint8_t *engine_id)
8040 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
8042 dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
8046 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
8048 if (!facet_is_controller_flow(facet) &&
8049 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
8050 struct subfacet *subfacet;
8051 struct ofexpired expired;
8053 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
8054 if (subfacet->path == SF_FAST_PATH) {
8055 struct dpif_flow_stats stats;
8057 subfacet_install(subfacet, &facet->xout.odp_actions,
8059 subfacet_update_stats(subfacet, &stats);
8063 expired.flow = facet->flow;
8064 expired.packet_count = facet->packet_count;
8065 expired.byte_count = facet->byte_count;
8066 expired.used = facet->used;
8067 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
8072 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
8074 struct cls_cursor cursor;
8075 struct facet *facet;
8077 cls_cursor_init(&cursor, &ofproto->facets, NULL);
8078 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
8079 send_active_timeout(ofproto, facet);
8083 static struct ofproto_dpif *
8084 ofproto_dpif_lookup(const char *name)
8086 struct ofproto_dpif *ofproto;
8088 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
8089 hash_string(name, 0), &all_ofproto_dpifs) {
8090 if (!strcmp(ofproto->up.name, name)) {
8098 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
8099 const char *argv[], void *aux OVS_UNUSED)
8101 struct ofproto_dpif *ofproto;
8104 ofproto = ofproto_dpif_lookup(argv[1]);
8106 unixctl_command_reply_error(conn, "no such bridge");
8109 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
8111 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
8112 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
8116 unixctl_command_reply(conn, "table successfully flushed");
8120 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
8121 const char *argv[], void *aux OVS_UNUSED)
8123 struct ds ds = DS_EMPTY_INITIALIZER;
8124 const struct ofproto_dpif *ofproto;
8125 const struct mac_entry *e;
8127 ofproto = ofproto_dpif_lookup(argv[1]);
8129 unixctl_command_reply_error(conn, "no such bridge");
8133 ds_put_cstr(&ds, " port VLAN MAC Age\n");
8134 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
8135 struct ofbundle *bundle = e->port.p;
8136 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
8137 ofbundle_get_a_port(bundle)->odp_port,
8138 e->vlan, ETH_ADDR_ARGS(e->mac),
8139 mac_entry_age(ofproto->ml, e));
8141 unixctl_command_reply(conn, ds_cstr(&ds));
8146 struct xlate_out xout;
8147 struct xlate_in xin;
8153 trace_format_rule(struct ds *result, uint8_t table_id, int level,
8154 const struct rule_dpif *rule)
8156 ds_put_char_multiple(result, '\t', level);
8158 ds_put_cstr(result, "No match\n");
8162 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
8163 table_id, ntohll(rule->up.flow_cookie));
8164 cls_rule_format(&rule->up.cr, result);
8165 ds_put_char(result, '\n');
8167 ds_put_char_multiple(result, '\t', level);
8168 ds_put_cstr(result, "OpenFlow ");
8169 ofpacts_format(rule->up.ofpacts, rule->up.ofpacts_len, result);
8170 ds_put_char(result, '\n');
8174 trace_format_flow(struct ds *result, int level, const char *title,
8175 struct trace_ctx *trace)
8177 ds_put_char_multiple(result, '\t', level);
8178 ds_put_format(result, "%s: ", title);
8179 if (flow_equal(&trace->xin.flow, &trace->flow)) {
8180 ds_put_cstr(result, "unchanged");
8182 flow_format(result, &trace->xin.flow);
8183 trace->flow = trace->xin.flow;
8185 ds_put_char(result, '\n');
8189 trace_format_regs(struct ds *result, int level, const char *title,
8190 struct trace_ctx *trace)
8194 ds_put_char_multiple(result, '\t', level);
8195 ds_put_format(result, "%s:", title);
8196 for (i = 0; i < FLOW_N_REGS; i++) {
8197 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
8199 ds_put_char(result, '\n');
8203 trace_format_odp(struct ds *result, int level, const char *title,
8204 struct trace_ctx *trace)
8206 struct ofpbuf *odp_actions = &trace->xout.odp_actions;
8208 ds_put_char_multiple(result, '\t', level);
8209 ds_put_format(result, "%s: ", title);
8210 format_odp_actions(result, odp_actions->data, odp_actions->size);
8211 ds_put_char(result, '\n');
8215 trace_resubmit(struct xlate_ctx *ctx, struct rule_dpif *rule)
8217 struct trace_ctx *trace = CONTAINER_OF(ctx->xin, struct trace_ctx, xin);
8218 struct ds *result = trace->result;
8220 ds_put_char(result, '\n');
8221 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
8222 trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace);
8223 trace_format_odp(result, ctx->recurse + 1, "Resubmitted odp", trace);
8224 trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule);
8228 trace_report(struct xlate_ctx *ctx, const char *s)
8230 struct trace_ctx *trace = CONTAINER_OF(ctx->xin, struct trace_ctx, xin);
8231 struct ds *result = trace->result;
8233 ds_put_char_multiple(result, '\t', ctx->recurse);
8234 ds_put_cstr(result, s);
8235 ds_put_char(result, '\n');
8239 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
8240 void *aux OVS_UNUSED)
8242 const char *dpname = argv[1];
8243 struct ofproto_dpif *ofproto;
8244 struct ofpbuf odp_key;
8245 struct ofpbuf *packet;
8246 struct initial_vals initial_vals;
8252 ofpbuf_init(&odp_key, 0);
8255 ofproto = ofproto_dpif_lookup(dpname);
8257 unixctl_command_reply_error(conn, "Unknown ofproto (use ofproto/list "
8261 if (argc == 3 || (argc == 4 && !strcmp(argv[3], "-generate"))) {
8262 /* ofproto/trace dpname flow [-generate] */
8263 const char *flow_s = argv[2];
8264 const char *generate_s = argv[3];
8266 /* Allow 'flow_s' to be either a datapath flow or an OpenFlow-like
8267 * flow. We guess which type it is based on whether 'flow_s' contains
8268 * an '(', since a datapath flow always contains '(') but an
8269 * OpenFlow-like flow should not (in fact it's allowed but I believe
8270 * that's not documented anywhere).
8272 * An alternative would be to try to parse 'flow_s' both ways, but then
8273 * it would be tricky giving a sensible error message. After all, do
8274 * you just say "syntax error" or do you present both error messages?
8275 * Both choices seem lousy. */
8276 if (strchr(flow_s, '(')) {
8279 /* Convert string to datapath key. */
8280 ofpbuf_init(&odp_key, 0);
8281 error = odp_flow_from_string(flow_s, NULL, &odp_key, NULL);
8283 unixctl_command_reply_error(conn, "Bad flow syntax");
8287 /* The user might have specified the wrong ofproto but within the
8288 * same backer. That's OK, ofproto_receive() can find the right
8290 if (ofproto_receive(ofproto->backer, NULL, odp_key.data,
8291 odp_key.size, &flow, NULL, &ofproto, NULL,
8293 unixctl_command_reply_error(conn, "Invalid flow");
8296 ds_put_format(&result, "Bridge: %s\n", ofproto->up.name);
8300 error_s = parse_ofp_exact_flow(&flow, argv[2]);
8302 unixctl_command_reply_error(conn, error_s);
8307 initial_vals.vlan_tci = flow.vlan_tci;
8308 initial_vals.tunnel_ip_tos = flow.tunnel.ip_tos;
8311 /* Generate a packet, if requested. */
8313 packet = ofpbuf_new(0);
8314 flow_compose(packet, &flow);
8316 } else if (argc == 7) {
8317 /* ofproto/trace dpname priority tun_id in_port mark packet */
8318 const char *priority_s = argv[2];
8319 const char *tun_id_s = argv[3];
8320 const char *in_port_s = argv[4];
8321 const char *mark_s = argv[5];
8322 const char *packet_s = argv[6];
8323 uint32_t in_port = atoi(in_port_s);
8324 ovs_be64 tun_id = htonll(strtoull(tun_id_s, NULL, 0));
8325 uint32_t priority = atoi(priority_s);
8326 uint32_t mark = atoi(mark_s);
8329 msg = eth_from_hex(packet_s, &packet);
8331 unixctl_command_reply_error(conn, msg);
8335 ds_put_cstr(&result, "Packet: ");
8336 s = ofp_packet_to_string(packet->data, packet->size);
8337 ds_put_cstr(&result, s);
8340 flow_extract(packet, priority, mark, NULL, in_port, &flow);
8341 flow.tunnel.tun_id = tun_id;
8342 initial_vals.vlan_tci = flow.vlan_tci;
8343 initial_vals.tunnel_ip_tos = flow.tunnel.ip_tos;
8345 unixctl_command_reply_error(conn, "Bad command syntax");
8349 ofproto_trace(ofproto, &flow, packet, &initial_vals, &result);
8350 unixctl_command_reply(conn, ds_cstr(&result));
8353 ds_destroy(&result);
8354 ofpbuf_delete(packet);
8355 ofpbuf_uninit(&odp_key);
8359 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
8360 const struct ofpbuf *packet,
8361 const struct initial_vals *initial_vals, struct ds *ds)
8363 struct rule_dpif *rule;
8365 ds_put_cstr(ds, "Flow: ");
8366 flow_format(ds, flow);
8367 ds_put_char(ds, '\n');
8369 rule = rule_dpif_lookup(ofproto, flow, NULL);
8371 trace_format_rule(ds, 0, 0, rule);
8372 if (rule == ofproto->miss_rule) {
8373 ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
8374 } else if (rule == ofproto->no_packet_in_rule) {
8375 ds_put_cstr(ds, "\nNo match, packets dropped because "
8376 "OFPPC_NO_PACKET_IN is set on in_port.\n");
8377 } else if (rule == ofproto->drop_frags_rule) {
8378 ds_put_cstr(ds, "\nPackets dropped because they are IP fragments "
8379 "and the fragment handling mode is \"drop\".\n");
8383 uint64_t odp_actions_stub[1024 / 8];
8384 struct ofpbuf odp_actions;
8385 struct trace_ctx trace;
8389 tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
8392 ofpbuf_use_stub(&odp_actions,
8393 odp_actions_stub, sizeof odp_actions_stub);
8394 xlate_in_init(&trace.xin, ofproto, flow, initial_vals, rule, tcp_flags,
8396 trace.xin.resubmit_hook = trace_resubmit;
8397 trace.xin.report_hook = trace_report;
8399 xlate_actions(&trace.xin, &trace.xout);
8401 ds_put_char(ds, '\n');
8402 trace_format_flow(ds, 0, "Final flow", &trace);
8404 match_init(&match, flow, &trace.xout.wc);
8405 ds_put_cstr(ds, "Relevant fields: ");
8406 match_format(&match, ds, OFP_DEFAULT_PRIORITY);
8407 ds_put_char(ds, '\n');
8409 ds_put_cstr(ds, "Datapath actions: ");
8410 format_odp_actions(ds, trace.xout.odp_actions.data,
8411 trace.xout.odp_actions.size);
8413 if (trace.xout.slow) {
8414 ds_put_cstr(ds, "\nThis flow is handled by the userspace "
8415 "slow path because it:");
8416 switch (trace.xout.slow) {
8418 ds_put_cstr(ds, "\n\t- Consists of CFM packets.");
8421 ds_put_cstr(ds, "\n\t- Consists of LACP packets.");
8424 ds_put_cstr(ds, "\n\t- Consists of STP packets.");
8426 case SLOW_CONTROLLER:
8427 ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages "
8428 "to the OpenFlow controller.");
8435 xlate_out_uninit(&trace.xout);
8440 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
8441 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
8444 unixctl_command_reply(conn, NULL);
8448 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
8449 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
8452 unixctl_command_reply(conn, NULL);
8455 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
8456 * 'reply' describing the results. */
8458 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
8460 struct cls_cursor cursor;
8461 struct facet *facet;
8465 cls_cursor_init(&cursor, &ofproto->facets, NULL);
8466 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
8467 if (!facet_check_consistency(facet)) {
8472 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
8476 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
8477 ofproto->up.name, errors);
8479 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
8484 ofproto_dpif_self_check(struct unixctl_conn *conn,
8485 int argc, const char *argv[], void *aux OVS_UNUSED)
8487 struct ds reply = DS_EMPTY_INITIALIZER;
8488 struct ofproto_dpif *ofproto;
8491 ofproto = ofproto_dpif_lookup(argv[1]);
8493 unixctl_command_reply_error(conn, "Unknown ofproto (use "
8494 "ofproto/list for help)");
8497 ofproto_dpif_self_check__(ofproto, &reply);
8499 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
8500 ofproto_dpif_self_check__(ofproto, &reply);
8504 unixctl_command_reply(conn, ds_cstr(&reply));
8508 /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list
8509 * of the 'ofproto_shash' nodes. It is the responsibility of the caller
8510 * to destroy 'ofproto_shash' and free the returned value. */
8511 static const struct shash_node **
8512 get_ofprotos(struct shash *ofproto_shash)
8514 const struct ofproto_dpif *ofproto;
8516 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
8517 char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
8518 shash_add_nocopy(ofproto_shash, name, ofproto);
8521 return shash_sort(ofproto_shash);
8525 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
8526 const char *argv[] OVS_UNUSED,
8527 void *aux OVS_UNUSED)
8529 struct ds ds = DS_EMPTY_INITIALIZER;
8530 struct shash ofproto_shash;
8531 const struct shash_node **sorted_ofprotos;
8534 shash_init(&ofproto_shash);
8535 sorted_ofprotos = get_ofprotos(&ofproto_shash);
8536 for (i = 0; i < shash_count(&ofproto_shash); i++) {
8537 const struct shash_node *node = sorted_ofprotos[i];
8538 ds_put_format(&ds, "%s\n", node->name);
8541 shash_destroy(&ofproto_shash);
8542 free(sorted_ofprotos);
8544 unixctl_command_reply(conn, ds_cstr(&ds));
8549 show_dp_format(const struct ofproto_dpif *ofproto, struct ds *ds)
8551 const struct shash_node **ports;
8553 struct avg_subfacet_rates lifetime;
8554 unsigned long long int minutes;
8555 const int min_ms = 60 * 1000; /* milliseconds in one minute. */
8557 minutes = (time_msec() - ofproto->created) / min_ms;
8560 lifetime.add_rate = (double)ofproto->total_subfacet_add_count
8562 lifetime.del_rate = (double)ofproto->total_subfacet_del_count
8565 lifetime.add_rate = 0.0;
8566 lifetime.del_rate = 0.0;
8569 ds_put_format(ds, "%s (%s):\n", ofproto->up.name,
8570 dpif_name(ofproto->backer->dpif));
8572 "\tlookups: hit:%"PRIu64" missed:%"PRIu64"\n",
8573 ofproto->n_hit, ofproto->n_missed);
8574 ds_put_format(ds, "\tflows: cur: %zu, avg: %5.3f, max: %d,"
8575 " life span: %llu(ms)\n",
8576 hmap_count(&ofproto->subfacets),
8577 avg_subfacet_count(ofproto),
8578 ofproto->max_n_subfacet,
8579 avg_subfacet_life_span(ofproto));
8580 if (minutes >= 60) {
8581 show_dp_rates(ds, "\t\thourly avg:", &ofproto->hourly);
8583 if (minutes >= 60 * 24) {
8584 show_dp_rates(ds, "\t\tdaily avg:", &ofproto->daily);
8586 show_dp_rates(ds, "\t\toverall avg:", &lifetime);
8588 ports = shash_sort(&ofproto->up.port_by_name);
8589 for (i = 0; i < shash_count(&ofproto->up.port_by_name); i++) {
8590 const struct shash_node *node = ports[i];
8591 struct ofport *ofport = node->data;
8592 const char *name = netdev_get_name(ofport->netdev);
8593 const char *type = netdev_get_type(ofport->netdev);
8596 ds_put_format(ds, "\t%s %u/", name, ofport->ofp_port);
8598 odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
8599 if (odp_port != OVSP_NONE) {
8600 ds_put_format(ds, "%"PRIu32":", odp_port);
8602 ds_put_cstr(ds, "none:");
8605 if (strcmp(type, "system")) {
8606 struct netdev *netdev;
8609 ds_put_format(ds, " (%s", type);
8611 error = netdev_open(name, type, &netdev);
8616 error = netdev_get_config(netdev, &config);
8618 const struct smap_node **nodes;
8621 nodes = smap_sort(&config);
8622 for (i = 0; i < smap_count(&config); i++) {
8623 const struct smap_node *node = nodes[i];
8624 ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
8625 node->key, node->value);
8629 smap_destroy(&config);
8631 netdev_close(netdev);
8633 ds_put_char(ds, ')');
8635 ds_put_char(ds, '\n');
8641 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc,
8642 const char *argv[], void *aux OVS_UNUSED)
8644 struct ds ds = DS_EMPTY_INITIALIZER;
8645 const struct ofproto_dpif *ofproto;
8649 for (i = 1; i < argc; i++) {
8650 ofproto = ofproto_dpif_lookup(argv[i]);
8652 ds_put_format(&ds, "Unknown bridge %s (use dpif/dump-dps "
8653 "for help)", argv[i]);
8654 unixctl_command_reply_error(conn, ds_cstr(&ds));
8657 show_dp_format(ofproto, &ds);
8660 struct shash ofproto_shash;
8661 const struct shash_node **sorted_ofprotos;
8664 shash_init(&ofproto_shash);
8665 sorted_ofprotos = get_ofprotos(&ofproto_shash);
8666 for (i = 0; i < shash_count(&ofproto_shash); i++) {
8667 const struct shash_node *node = sorted_ofprotos[i];
8668 show_dp_format(node->data, &ds);
8671 shash_destroy(&ofproto_shash);
8672 free(sorted_ofprotos);
8675 unixctl_command_reply(conn, ds_cstr(&ds));
8679 /* Dump the megaflow (facet) cache. This is useful to check the
8680 * correctness of flow wildcarding, since the same mechanism is used for
8681 * both xlate caching and kernel wildcarding.
8683 * It's important to note that in the output the flow description uses
8684 * OpenFlow (OFP) ports, but the actions use datapath (ODP) ports.
8686 * This command is only needed for advanced debugging, so it's not
8687 * documented in the man page. */
8689 ofproto_unixctl_dpif_dump_megaflows(struct unixctl_conn *conn,
8690 int argc OVS_UNUSED, const char *argv[],
8691 void *aux OVS_UNUSED)
8693 struct ds ds = DS_EMPTY_INITIALIZER;
8694 const struct ofproto_dpif *ofproto;
8695 long long int now = time_msec();
8696 struct cls_cursor cursor;
8697 struct facet *facet;
8699 ofproto = ofproto_dpif_lookup(argv[1]);
8701 unixctl_command_reply_error(conn, "no such bridge");
8705 cls_cursor_init(&cursor, &ofproto->facets, NULL);
8706 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
8707 cls_rule_format(&facet->cr, &ds);
8708 ds_put_cstr(&ds, ", ");
8709 ds_put_format(&ds, "n_subfacets:%zu, ", list_size(&facet->subfacets));
8710 ds_put_format(&ds, "used:%.3fs, ", (now - facet->used) / 1000.0);
8711 ds_put_cstr(&ds, "Datapath actions: ");
8712 if (facet->xout.slow) {
8713 uint64_t slow_path_stub[128 / 8];
8714 const struct nlattr *actions;
8717 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
8718 slow_path_stub, sizeof slow_path_stub,
8719 &actions, &actions_len);
8720 format_odp_actions(&ds, actions, actions_len);
8722 format_odp_actions(&ds, facet->xout.odp_actions.data,
8723 facet->xout.odp_actions.size);
8725 ds_put_cstr(&ds, "\n");
8728 ds_chomp(&ds, '\n');
8729 unixctl_command_reply(conn, ds_cstr(&ds));
8734 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
8735 int argc OVS_UNUSED, const char *argv[],
8736 void *aux OVS_UNUSED)
8738 struct ds ds = DS_EMPTY_INITIALIZER;
8739 const struct ofproto_dpif *ofproto;
8740 struct subfacet *subfacet;
8742 ofproto = ofproto_dpif_lookup(argv[1]);
8744 unixctl_command_reply_error(conn, "no such bridge");
8748 update_stats(ofproto->backer);
8750 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
8751 struct facet *facet = subfacet->facet;
8753 odp_flow_key_format(subfacet->key, subfacet->key_len, &ds);
8755 ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:",
8756 subfacet->dp_packet_count, subfacet->dp_byte_count);
8757 if (subfacet->used) {
8758 ds_put_format(&ds, "%.3fs",
8759 (time_msec() - subfacet->used) / 1000.0);
8761 ds_put_format(&ds, "never");
8763 if (subfacet->facet->tcp_flags) {
8764 ds_put_cstr(&ds, ", flags:");
8765 packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags);
8768 ds_put_cstr(&ds, ", actions:");
8769 if (facet->xout.slow) {
8770 uint64_t slow_path_stub[128 / 8];
8771 const struct nlattr *actions;
8774 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
8775 slow_path_stub, sizeof slow_path_stub,
8776 &actions, &actions_len);
8777 format_odp_actions(&ds, actions, actions_len);
8779 format_odp_actions(&ds, facet->xout.odp_actions.data,
8780 facet->xout.odp_actions.size);
8782 ds_put_char(&ds, '\n');
8785 unixctl_command_reply(conn, ds_cstr(&ds));
8790 ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn,
8791 int argc OVS_UNUSED, const char *argv[],
8792 void *aux OVS_UNUSED)
8794 struct ds ds = DS_EMPTY_INITIALIZER;
8795 struct ofproto_dpif *ofproto;
8797 ofproto = ofproto_dpif_lookup(argv[1]);
8799 unixctl_command_reply_error(conn, "no such bridge");
8803 flush(&ofproto->up);
8805 unixctl_command_reply(conn, ds_cstr(&ds));
8810 ofproto_dpif_unixctl_init(void)
8812 static bool registered;
8818 unixctl_command_register(
8820 "bridge {priority tun_id in_port mark packet | odp_flow [-generate]}",
8821 2, 6, ofproto_unixctl_trace, NULL);
8822 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
8823 ofproto_unixctl_fdb_flush, NULL);
8824 unixctl_command_register("fdb/show", "bridge", 1, 1,
8825 ofproto_unixctl_fdb_show, NULL);
8826 unixctl_command_register("ofproto/clog", "", 0, 0,
8827 ofproto_dpif_clog, NULL);
8828 unixctl_command_register("ofproto/unclog", "", 0, 0,
8829 ofproto_dpif_unclog, NULL);
8830 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
8831 ofproto_dpif_self_check, NULL);
8832 unixctl_command_register("dpif/dump-dps", "", 0, 0,
8833 ofproto_unixctl_dpif_dump_dps, NULL);
8834 unixctl_command_register("dpif/show", "[bridge]", 0, INT_MAX,
8835 ofproto_unixctl_dpif_show, NULL);
8836 unixctl_command_register("dpif/dump-flows", "bridge", 1, 1,
8837 ofproto_unixctl_dpif_dump_flows, NULL);
8838 unixctl_command_register("dpif/del-flows", "bridge", 1, 1,
8839 ofproto_unixctl_dpif_del_flows, NULL);
8840 unixctl_command_register("dpif/dump-megaflows", "bridge", 1, 1,
8841 ofproto_unixctl_dpif_dump_megaflows, NULL);
8844 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
8846 * This is deprecated. It is only for compatibility with broken device drivers
8847 * in old versions of Linux that do not properly support VLANs when VLAN
8848 * devices are not used. When broken device drivers are no longer in
8849 * widespread use, we will delete these interfaces. */
8852 set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid)
8854 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
8855 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
8857 if (realdev_ofp_port == ofport->realdev_ofp_port
8858 && vid == ofport->vlandev_vid) {
8862 ofproto->backer->need_revalidate = REV_RECONFIGURE;
8864 if (ofport->realdev_ofp_port) {
8867 if (realdev_ofp_port && ofport->bundle) {
8868 /* vlandevs are enslaved to their realdevs, so they are not allowed to
8869 * themselves be part of a bundle. */
8870 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
8873 ofport->realdev_ofp_port = realdev_ofp_port;
8874 ofport->vlandev_vid = vid;
8876 if (realdev_ofp_port) {
8877 vsp_add(ofport, realdev_ofp_port, vid);
8884 hash_realdev_vid(uint16_t realdev_ofp_port, int vid)
8886 return hash_2words(realdev_ofp_port, vid);
8889 /* Returns the OFP port number of the Linux VLAN device that corresponds to
8890 * 'vlan_tci' on the network device with port number 'realdev_ofp_port' in
8891 * 'struct ofport_dpif'. For example, given 'realdev_ofp_port' of eth0 and
8892 * 'vlan_tci' 9, it would return the port number of eth0.9.
8894 * Unless VLAN splinters are enabled for port 'realdev_ofp_port', this
8895 * function just returns its 'realdev_ofp_port' argument. */
8897 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
8898 uint16_t realdev_ofp_port, ovs_be16 vlan_tci)
8900 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
8901 int vid = vlan_tci_to_vid(vlan_tci);
8902 const struct vlan_splinter *vsp;
8904 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
8905 hash_realdev_vid(realdev_ofp_port, vid),
8906 &ofproto->realdev_vid_map) {
8907 if (vsp->realdev_ofp_port == realdev_ofp_port
8908 && vsp->vid == vid) {
8909 return vsp->vlandev_ofp_port;
8913 return realdev_ofp_port;
8916 static struct vlan_splinter *
8917 vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port)
8919 struct vlan_splinter *vsp;
8921 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, hash_int(vlandev_ofp_port, 0),
8922 &ofproto->vlandev_map) {
8923 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
8931 /* Returns the OpenFlow port number of the "real" device underlying the Linux
8932 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
8933 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
8934 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
8935 * eth0 and store 9 in '*vid'.
8937 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
8938 * VLAN device. Unless VLAN splinters are enabled, this is what this function
8941 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
8942 uint16_t vlandev_ofp_port, int *vid)
8944 if (!hmap_is_empty(&ofproto->vlandev_map)) {
8945 const struct vlan_splinter *vsp;
8947 vsp = vlandev_find(ofproto, vlandev_ofp_port);
8952 return vsp->realdev_ofp_port;
8958 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
8959 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
8960 * 'flow->in_port' to the "real" device backing the VLAN device, sets
8961 * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is
8962 * always the case unless VLAN splinters are enabled), returns false without
8963 * making any changes. */
8965 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
8970 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port, &vid);
8975 /* Cause the flow to be processed as if it came in on the real device with
8976 * the VLAN device's VLAN ID. */
8977 flow->in_port = realdev;
8978 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
8983 vsp_remove(struct ofport_dpif *port)
8985 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
8986 struct vlan_splinter *vsp;
8988 vsp = vlandev_find(ofproto, port->up.ofp_port);
8990 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
8991 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
8994 port->realdev_ofp_port = 0;
8996 VLOG_ERR("missing vlan device record");
9001 vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid)
9003 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
9005 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
9006 && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid))
9007 == realdev_ofp_port)) {
9008 struct vlan_splinter *vsp;
9010 vsp = xmalloc(sizeof *vsp);
9011 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
9012 hash_int(port->up.ofp_port, 0));
9013 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
9014 hash_realdev_vid(realdev_ofp_port, vid));
9015 vsp->realdev_ofp_port = realdev_ofp_port;
9016 vsp->vlandev_ofp_port = port->up.ofp_port;
9019 port->realdev_ofp_port = realdev_ofp_port;
9021 VLOG_ERR("duplicate vlan device record");
9026 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
9028 const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
9029 return ofport ? ofport->odp_port : OVSP_NONE;
9032 static struct ofport_dpif *
9033 odp_port_to_ofport(const struct dpif_backer *backer, uint32_t odp_port)
9035 struct ofport_dpif *port;
9037 HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node,
9038 hash_int(odp_port, 0),
9039 &backer->odp_to_ofport_map) {
9040 if (port->odp_port == odp_port) {
9049 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
9051 struct ofport_dpif *port;
9053 port = odp_port_to_ofport(ofproto->backer, odp_port);
9054 if (port && &ofproto->up == port->up.ofproto) {
9055 return port->up.ofp_port;
9060 static unsigned long long int
9061 avg_subfacet_life_span(const struct ofproto_dpif *ofproto)
9063 unsigned long long int dc;
9064 unsigned long long int avg;
9066 dc = ofproto->total_subfacet_del_count + ofproto->subfacet_del_count;
9067 avg = dc ? ofproto->total_subfacet_life_span / dc : 0;
9073 avg_subfacet_count(const struct ofproto_dpif *ofproto)
9077 if (ofproto->n_update_stats) {
9078 avg_c = (double)ofproto->total_subfacet_count
9079 / ofproto->n_update_stats;
9086 show_dp_rates(struct ds *ds, const char *heading,
9087 const struct avg_subfacet_rates *rates)
9089 ds_put_format(ds, "%s add rate: %5.3f/min, del rate: %5.3f/min\n",
9090 heading, rates->add_rate, rates->del_rate);
9094 update_max_subfacet_count(struct ofproto_dpif *ofproto)
9096 ofproto->max_n_subfacet = MAX(ofproto->max_n_subfacet,
9097 hmap_count(&ofproto->subfacets));
9100 /* Compute exponentially weighted moving average, adding 'new' as the newest,
9101 * most heavily weighted element. 'base' designates the rate of decay: after
9102 * 'base' further updates, 'new''s weight in the EWMA decays to about 1/e
9105 exp_mavg(double *avg, int base, double new)
9107 *avg = (*avg * (base - 1) + new) / base;
9111 update_moving_averages(struct ofproto_dpif *ofproto)
9113 const int min_ms = 60 * 1000; /* milliseconds in one minute. */
9115 /* Update hourly averages on the minute boundaries. */
9116 if (time_msec() - ofproto->last_minute >= min_ms) {
9117 exp_mavg(&ofproto->hourly.add_rate, 60, ofproto->subfacet_add_count);
9118 exp_mavg(&ofproto->hourly.del_rate, 60, ofproto->subfacet_del_count);
9120 /* Update daily averages on the hour boundaries. */
9121 if ((ofproto->last_minute - ofproto->created) / min_ms % 60 == 59) {
9122 exp_mavg(&ofproto->daily.add_rate, 24, ofproto->hourly.add_rate);
9123 exp_mavg(&ofproto->daily.del_rate, 24, ofproto->hourly.del_rate);
9126 ofproto->total_subfacet_add_count += ofproto->subfacet_add_count;
9127 ofproto->total_subfacet_del_count += ofproto->subfacet_del_count;
9128 ofproto->subfacet_add_count = 0;
9129 ofproto->subfacet_del_count = 0;
9130 ofproto->last_minute += min_ms;
9135 dpif_stats_update_hit_count(struct ofproto_dpif *ofproto, uint64_t delta)
9137 ofproto->n_hit += delta;
9140 const struct ofproto_class ofproto_dpif_class = {
9175 port_is_lacp_current,
9176 NULL, /* rule_choose_table */
9183 rule_modify_actions,
9195 get_stp_port_status,
9202 is_mirror_output_bundle,
9203 forward_bpdu_changed,
9204 set_mac_table_config,