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 /* By default, flows in the datapath are wildcarded (megaflows). They
796 * may be disabled with the "ovs-appctl dpif/disable-megaflows" command. */
797 static bool enable_megaflows = true;
799 /* All existing ofproto_dpif instances, indexed by ->up.name. */
800 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
802 static void ofproto_dpif_unixctl_init(void);
804 static struct ofproto_dpif *
805 ofproto_dpif_cast(const struct ofproto *ofproto)
807 ovs_assert(ofproto->ofproto_class == &ofproto_dpif_class);
808 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
811 static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *,
813 static struct ofport_dpif *get_odp_port(const struct ofproto_dpif *,
815 static void ofproto_trace(struct ofproto_dpif *, const struct flow *,
816 const struct ofpbuf *,
817 const struct initial_vals *, struct ds *);
819 /* Packet processing. */
820 static void update_learning_table(struct ofproto_dpif *, const struct flow *,
821 struct flow_wildcards *, int vlan,
824 #define FLOW_MISS_MAX_BATCH 50
825 static int handle_upcalls(struct dpif_backer *, unsigned int max_batch);
827 /* Flow expiration. */
828 static int expire(struct dpif_backer *);
831 static void send_netflow_active_timeouts(struct ofproto_dpif *);
834 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
835 static size_t compose_sflow_action(const struct ofproto_dpif *,
836 struct ofpbuf *odp_actions,
837 const struct flow *, uint32_t odp_port);
838 static void compose_ipfix_action(const struct ofproto_dpif *,
839 struct ofpbuf *odp_actions,
840 const struct flow *);
841 static void add_mirror_actions(struct xlate_ctx *ctx,
842 const struct flow *flow);
843 /* Global variables. */
844 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
846 /* Initial mappings of port to bridge mappings. */
847 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
849 /* Factory functions. */
852 init(const struct shash *iface_hints)
854 struct shash_node *node;
856 /* Make a local copy, since we don't own 'iface_hints' elements. */
857 SHASH_FOR_EACH(node, iface_hints) {
858 const struct iface_hint *orig_hint = node->data;
859 struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
861 new_hint->br_name = xstrdup(orig_hint->br_name);
862 new_hint->br_type = xstrdup(orig_hint->br_type);
863 new_hint->ofp_port = orig_hint->ofp_port;
865 shash_add(&init_ofp_ports, node->name, new_hint);
870 enumerate_types(struct sset *types)
872 dp_enumerate_types(types);
876 enumerate_names(const char *type, struct sset *names)
878 struct ofproto_dpif *ofproto;
881 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
882 if (strcmp(type, ofproto->up.type)) {
885 sset_add(names, ofproto->up.name);
892 del(const char *type, const char *name)
897 error = dpif_open(name, type, &dpif);
899 error = dpif_delete(dpif);
906 port_open_type(const char *datapath_type, const char *port_type)
908 return dpif_port_open_type(datapath_type, port_type);
911 /* Type functions. */
913 static struct ofproto_dpif *
914 lookup_ofproto_dpif_by_port_name(const char *name)
916 struct ofproto_dpif *ofproto;
918 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
919 if (sset_contains(&ofproto->ports, name)) {
928 type_run(const char *type)
930 static long long int push_timer = LLONG_MIN;
931 struct dpif_backer *backer;
935 backer = shash_find_data(&all_dpif_backers, type);
937 /* This is not necessarily a problem, since backers are only
938 * created on demand. */
942 dpif_run(backer->dpif);
944 /* The most natural place to push facet statistics is when they're pulled
945 * from the datapath. However, when there are many flows in the datapath,
946 * this expensive operation can occur so frequently, that it reduces our
947 * ability to quickly set up flows. To reduce the cost, we push statistics
949 if (time_msec() > push_timer) {
950 push_timer = time_msec() + 2000;
954 /* If vswitchd started with other_config:flow_restore_wait set as "true",
955 * and the configuration has now changed to "false", enable receiving
956 * packets from the datapath. */
957 if (!backer->recv_set_enable && !ofproto_get_flow_restore_wait()) {
958 backer->recv_set_enable = true;
960 error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
962 VLOG_ERR("Failed to enable receiving packets in dpif.");
965 dpif_flow_flush(backer->dpif);
966 backer->need_revalidate = REV_RECONFIGURE;
969 if (backer->need_revalidate
970 || !tag_set_is_empty(&backer->revalidate_set)) {
971 struct tag_set revalidate_set = backer->revalidate_set;
972 bool need_revalidate = backer->need_revalidate;
973 struct ofproto_dpif *ofproto;
974 struct simap_node *node;
975 struct simap tmp_backers;
977 /* Handle tunnel garbage collection. */
978 simap_init(&tmp_backers);
979 simap_swap(&backer->tnl_backers, &tmp_backers);
981 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
982 struct ofport_dpif *iter;
984 if (backer != ofproto->backer) {
988 HMAP_FOR_EACH (iter, up.hmap_node, &ofproto->up.ports) {
991 if (!iter->tnl_port) {
995 dp_port = netdev_vport_get_dpif_port(iter->up.netdev);
996 node = simap_find(&tmp_backers, dp_port);
998 simap_put(&backer->tnl_backers, dp_port, node->data);
999 simap_delete(&tmp_backers, node);
1000 node = simap_find(&backer->tnl_backers, dp_port);
1002 node = simap_find(&backer->tnl_backers, dp_port);
1004 uint32_t odp_port = UINT32_MAX;
1006 if (!dpif_port_add(backer->dpif, iter->up.netdev,
1008 simap_put(&backer->tnl_backers, dp_port, odp_port);
1009 node = simap_find(&backer->tnl_backers, dp_port);
1014 iter->odp_port = node ? node->data : OVSP_NONE;
1015 if (tnl_port_reconfigure(&iter->up, iter->odp_port,
1017 backer->need_revalidate = REV_RECONFIGURE;
1022 SIMAP_FOR_EACH (node, &tmp_backers) {
1023 dpif_port_del(backer->dpif, node->data);
1025 simap_destroy(&tmp_backers);
1027 switch (backer->need_revalidate) {
1028 case REV_RECONFIGURE: COVERAGE_INC(rev_reconfigure); break;
1029 case REV_STP: COVERAGE_INC(rev_stp); break;
1030 case REV_PORT_TOGGLED: COVERAGE_INC(rev_port_toggled); break;
1031 case REV_FLOW_TABLE: COVERAGE_INC(rev_flow_table); break;
1032 case REV_INCONSISTENCY: COVERAGE_INC(rev_inconsistency); break;
1035 if (backer->need_revalidate) {
1036 /* Clear the drop_keys in case we should now be accepting some
1037 * formerly dropped flows. */
1038 drop_key_clear(backer);
1041 /* Clear the revalidation flags. */
1042 tag_set_init(&backer->revalidate_set);
1043 backer->need_revalidate = 0;
1045 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1046 struct facet *facet, *next;
1047 struct cls_cursor cursor;
1049 if (ofproto->backer != backer) {
1053 cls_cursor_init(&cursor, &ofproto->facets, NULL);
1054 CLS_CURSOR_FOR_EACH_SAFE (facet, next, cr, &cursor) {
1056 || tag_set_intersects(&revalidate_set, facet->xout.tags)) {
1057 facet_revalidate(facet);
1064 if (!backer->recv_set_enable) {
1065 /* Wake up before a max of 1000ms. */
1066 timer_set_duration(&backer->next_expiration, 1000);
1067 } else if (timer_expired(&backer->next_expiration)) {
1068 int delay = expire(backer);
1069 timer_set_duration(&backer->next_expiration, delay);
1072 /* Check for port changes in the dpif. */
1073 while ((error = dpif_port_poll(backer->dpif, &devname)) == 0) {
1074 struct ofproto_dpif *ofproto;
1075 struct dpif_port port;
1077 /* Don't report on the datapath's device. */
1078 if (!strcmp(devname, dpif_base_name(backer->dpif))) {
1082 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
1083 &all_ofproto_dpifs) {
1084 if (simap_contains(&ofproto->backer->tnl_backers, devname)) {
1089 ofproto = lookup_ofproto_dpif_by_port_name(devname);
1090 if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
1091 /* The port was removed. If we know the datapath,
1092 * report it through poll_set(). If we don't, it may be
1093 * notifying us of a removal we initiated, so ignore it.
1094 * If there's a pending ENOBUFS, let it stand, since
1095 * everything will be reevaluated. */
1096 if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
1097 sset_add(&ofproto->port_poll_set, devname);
1098 ofproto->port_poll_errno = 0;
1100 } else if (!ofproto) {
1101 /* The port was added, but we don't know with which
1102 * ofproto we should associate it. Delete it. */
1103 dpif_port_del(backer->dpif, port.port_no);
1105 dpif_port_destroy(&port);
1111 if (error != EAGAIN) {
1112 struct ofproto_dpif *ofproto;
1114 /* There was some sort of error, so propagate it to all
1115 * ofprotos that use this backer. */
1116 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
1117 &all_ofproto_dpifs) {
1118 if (ofproto->backer == backer) {
1119 sset_clear(&ofproto->port_poll_set);
1120 ofproto->port_poll_errno = error;
1129 dpif_backer_run_fast(struct dpif_backer *backer, int max_batch)
1133 /* If recv_set_enable is false, we should not handle upcalls. */
1134 if (!backer->recv_set_enable) {
1138 /* Handle one or more batches of upcalls, until there's nothing left to do
1139 * or until we do a fixed total amount of work.
1141 * We do work in batches because it can be much cheaper to set up a number
1142 * of flows and fire off their patches all at once. We do multiple batches
1143 * because in some cases handling a packet can cause another packet to be
1144 * queued almost immediately as part of the return flow. Both
1145 * optimizations can make major improvements on some benchmarks and
1146 * presumably for real traffic as well. */
1148 while (work < max_batch) {
1149 int retval = handle_upcalls(backer, max_batch - work);
1160 type_run_fast(const char *type)
1162 struct dpif_backer *backer;
1164 backer = shash_find_data(&all_dpif_backers, type);
1166 /* This is not necessarily a problem, since backers are only
1167 * created on demand. */
1171 return dpif_backer_run_fast(backer, FLOW_MISS_MAX_BATCH);
1177 static long long int port_rl = LLONG_MIN;
1178 static unsigned int backer_rl = 0;
1180 if (time_msec() >= port_rl) {
1181 struct ofproto_dpif *ofproto;
1182 struct ofport_dpif *ofport;
1184 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1186 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1187 port_run_fast(ofport);
1190 port_rl = time_msec() + 200;
1193 /* XXX: We have to be careful not to do too much work in this function. If
1194 * we call dpif_backer_run_fast() too often, or with too large a batch,
1195 * performance improves signifcantly, but at a cost. It's possible for the
1196 * number of flows in the datapath to increase without bound, and for poll
1197 * loops to take 10s of seconds. The correct solution to this problem,
1198 * long term, is to separate flow miss handling into it's own thread so it
1199 * isn't affected by revalidations, and expirations. Until then, this is
1200 * the best we can do. */
1201 if (++backer_rl >= 10) {
1202 struct shash_node *node;
1205 SHASH_FOR_EACH (node, &all_dpif_backers) {
1206 dpif_backer_run_fast(node->data, 1);
1212 type_wait(const char *type)
1214 struct dpif_backer *backer;
1216 backer = shash_find_data(&all_dpif_backers, type);
1218 /* This is not necessarily a problem, since backers are only
1219 * created on demand. */
1223 timer_wait(&backer->next_expiration);
1226 /* Basic life-cycle. */
1228 static int add_internal_flows(struct ofproto_dpif *);
1230 static struct ofproto *
1233 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
1234 return &ofproto->up;
1238 dealloc(struct ofproto *ofproto_)
1240 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1245 close_dpif_backer(struct dpif_backer *backer)
1247 struct shash_node *node;
1249 ovs_assert(backer->refcount > 0);
1251 if (--backer->refcount) {
1255 drop_key_clear(backer);
1256 hmap_destroy(&backer->drop_keys);
1258 simap_destroy(&backer->tnl_backers);
1259 hmap_destroy(&backer->odp_to_ofport_map);
1260 node = shash_find(&all_dpif_backers, backer->type);
1262 shash_delete(&all_dpif_backers, node);
1263 dpif_close(backer->dpif);
1268 /* Datapath port slated for removal from datapath. */
1269 struct odp_garbage {
1270 struct list list_node;
1275 open_dpif_backer(const char *type, struct dpif_backer **backerp)
1277 struct dpif_backer *backer;
1278 struct dpif_port_dump port_dump;
1279 struct dpif_port port;
1280 struct shash_node *node;
1281 struct list garbage_list;
1282 struct odp_garbage *garbage, *next;
1288 backer = shash_find_data(&all_dpif_backers, type);
1295 backer_name = xasprintf("ovs-%s", type);
1297 /* Remove any existing datapaths, since we assume we're the only
1298 * userspace controlling the datapath. */
1300 dp_enumerate_names(type, &names);
1301 SSET_FOR_EACH(name, &names) {
1302 struct dpif *old_dpif;
1304 /* Don't remove our backer if it exists. */
1305 if (!strcmp(name, backer_name)) {
1309 if (dpif_open(name, type, &old_dpif)) {
1310 VLOG_WARN("couldn't open old datapath %s to remove it", name);
1312 dpif_delete(old_dpif);
1313 dpif_close(old_dpif);
1316 sset_destroy(&names);
1318 backer = xmalloc(sizeof *backer);
1320 error = dpif_create_and_open(backer_name, type, &backer->dpif);
1323 VLOG_ERR("failed to open datapath of type %s: %s", type,
1329 backer->type = xstrdup(type);
1330 backer->refcount = 1;
1331 hmap_init(&backer->odp_to_ofport_map);
1332 hmap_init(&backer->drop_keys);
1333 timer_set_duration(&backer->next_expiration, 1000);
1334 backer->need_revalidate = 0;
1335 simap_init(&backer->tnl_backers);
1336 tag_set_init(&backer->revalidate_set);
1337 backer->recv_set_enable = !ofproto_get_flow_restore_wait();
1340 if (backer->recv_set_enable) {
1341 dpif_flow_flush(backer->dpif);
1344 /* Loop through the ports already on the datapath and remove any
1345 * that we don't need anymore. */
1346 list_init(&garbage_list);
1347 dpif_port_dump_start(&port_dump, backer->dpif);
1348 while (dpif_port_dump_next(&port_dump, &port)) {
1349 node = shash_find(&init_ofp_ports, port.name);
1350 if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
1351 garbage = xmalloc(sizeof *garbage);
1352 garbage->odp_port = port.port_no;
1353 list_push_front(&garbage_list, &garbage->list_node);
1356 dpif_port_dump_done(&port_dump);
1358 LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) {
1359 dpif_port_del(backer->dpif, garbage->odp_port);
1360 list_remove(&garbage->list_node);
1364 shash_add(&all_dpif_backers, type, backer);
1366 error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
1368 VLOG_ERR("failed to listen on datapath of type %s: %s",
1369 type, strerror(error));
1370 close_dpif_backer(backer);
1378 construct(struct ofproto *ofproto_)
1380 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1381 struct shash_node *node, *next;
1386 error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1391 max_ports = dpif_get_max_ports(ofproto->backer->dpif);
1392 ofproto_init_max_ports(ofproto_, MIN(max_ports, OFPP_MAX));
1394 ofproto->netflow = NULL;
1395 ofproto->sflow = NULL;
1396 ofproto->ipfix = NULL;
1397 ofproto->stp = NULL;
1398 hmap_init(&ofproto->bundles);
1399 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1400 for (i = 0; i < MAX_MIRRORS; i++) {
1401 ofproto->mirrors[i] = NULL;
1403 ofproto->has_bonded_bundles = false;
1405 classifier_init(&ofproto->facets);
1406 hmap_init(&ofproto->subfacets);
1407 ofproto->governor = NULL;
1408 ofproto->consistency_rl = LLONG_MIN;
1410 for (i = 0; i < N_TABLES; i++) {
1411 struct table_dpif *table = &ofproto->tables[i];
1413 table->catchall_table = NULL;
1414 table->other_table = NULL;
1415 table->basis = random_uint32();
1418 list_init(&ofproto->completions);
1420 ofproto_dpif_unixctl_init();
1422 ofproto->has_mirrors = false;
1423 ofproto->has_bundle_action = false;
1425 hmap_init(&ofproto->vlandev_map);
1426 hmap_init(&ofproto->realdev_vid_map);
1428 sset_init(&ofproto->ports);
1429 sset_init(&ofproto->ghost_ports);
1430 sset_init(&ofproto->port_poll_set);
1431 ofproto->port_poll_errno = 0;
1433 SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1434 struct iface_hint *iface_hint = node->data;
1436 if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1437 /* Check if the datapath already has this port. */
1438 if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1439 sset_add(&ofproto->ports, node->name);
1442 free(iface_hint->br_name);
1443 free(iface_hint->br_type);
1445 shash_delete(&init_ofp_ports, node);
1449 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1450 hash_string(ofproto->up.name, 0));
1451 memset(&ofproto->stats, 0, sizeof ofproto->stats);
1453 ofproto_init_tables(ofproto_, N_TABLES);
1454 error = add_internal_flows(ofproto);
1455 ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1458 ofproto->n_missed = 0;
1460 ofproto->max_n_subfacet = 0;
1461 ofproto->created = time_msec();
1462 ofproto->last_minute = ofproto->created;
1463 memset(&ofproto->hourly, 0, sizeof ofproto->hourly);
1464 memset(&ofproto->daily, 0, sizeof ofproto->daily);
1465 ofproto->subfacet_add_count = 0;
1466 ofproto->subfacet_del_count = 0;
1467 ofproto->total_subfacet_add_count = 0;
1468 ofproto->total_subfacet_del_count = 0;
1469 ofproto->total_subfacet_life_span = 0;
1470 ofproto->total_subfacet_count = 0;
1471 ofproto->n_update_stats = 0;
1477 add_internal_flow(struct ofproto_dpif *ofproto, int id,
1478 const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1480 struct ofputil_flow_mod fm;
1483 match_init_catchall(&fm.match);
1485 match_set_reg(&fm.match, 0, id);
1486 fm.new_cookie = htonll(0);
1487 fm.cookie = htonll(0);
1488 fm.cookie_mask = htonll(0);
1489 fm.table_id = TBL_INTERNAL;
1490 fm.command = OFPFC_ADD;
1491 fm.idle_timeout = 0;
1492 fm.hard_timeout = 0;
1496 fm.ofpacts = ofpacts->data;
1497 fm.ofpacts_len = ofpacts->size;
1499 error = ofproto_flow_mod(&ofproto->up, &fm);
1501 VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
1502 id, ofperr_to_string(error));
1506 *rulep = rule_dpif_lookup__(ofproto, &fm.match.flow, NULL, TBL_INTERNAL);
1507 ovs_assert(*rulep != NULL);
1513 add_internal_flows(struct ofproto_dpif *ofproto)
1515 struct ofpact_controller *controller;
1516 uint64_t ofpacts_stub[128 / 8];
1517 struct ofpbuf ofpacts;
1521 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1524 controller = ofpact_put_CONTROLLER(&ofpacts);
1525 controller->max_len = UINT16_MAX;
1526 controller->controller_id = 0;
1527 controller->reason = OFPR_NO_MATCH;
1528 ofpact_pad(&ofpacts);
1530 error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule);
1535 ofpbuf_clear(&ofpacts);
1536 error = add_internal_flow(ofproto, id++, &ofpacts,
1537 &ofproto->no_packet_in_rule);
1542 error = add_internal_flow(ofproto, id++, &ofpacts,
1543 &ofproto->drop_frags_rule);
1548 complete_operations(struct ofproto_dpif *ofproto)
1550 struct dpif_completion *c, *next;
1552 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
1553 ofoperation_complete(c->op, 0);
1554 list_remove(&c->list_node);
1560 destruct(struct ofproto *ofproto_)
1562 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1563 struct rule_dpif *rule, *next_rule;
1564 struct oftable *table;
1567 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1568 complete_operations(ofproto);
1570 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1571 struct cls_cursor cursor;
1573 cls_cursor_init(&cursor, &table->cls, NULL);
1574 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1575 ofproto_rule_destroy(&rule->up);
1579 for (i = 0; i < MAX_MIRRORS; i++) {
1580 mirror_destroy(ofproto->mirrors[i]);
1583 netflow_destroy(ofproto->netflow);
1584 dpif_sflow_destroy(ofproto->sflow);
1585 hmap_destroy(&ofproto->bundles);
1586 mac_learning_destroy(ofproto->ml);
1588 classifier_destroy(&ofproto->facets);
1589 hmap_destroy(&ofproto->subfacets);
1590 governor_destroy(ofproto->governor);
1592 hmap_destroy(&ofproto->vlandev_map);
1593 hmap_destroy(&ofproto->realdev_vid_map);
1595 sset_destroy(&ofproto->ports);
1596 sset_destroy(&ofproto->ghost_ports);
1597 sset_destroy(&ofproto->port_poll_set);
1599 close_dpif_backer(ofproto->backer);
1603 run_fast(struct ofproto *ofproto_)
1605 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1606 struct ofport_dpif *ofport;
1608 /* Do not perform any periodic activity required by 'ofproto' while
1609 * waiting for flow restore to complete. */
1610 if (ofproto_get_flow_restore_wait()) {
1614 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1615 port_run_fast(ofport);
1622 run(struct ofproto *ofproto_)
1624 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1625 struct ofport_dpif *ofport;
1626 struct ofbundle *bundle;
1630 complete_operations(ofproto);
1633 /* Do not perform any periodic activity below required by 'ofproto' while
1634 * waiting for flow restore to complete. */
1635 if (ofproto_get_flow_restore_wait()) {
1639 error = run_fast(ofproto_);
1644 if (ofproto->netflow) {
1645 if (netflow_run(ofproto->netflow)) {
1646 send_netflow_active_timeouts(ofproto);
1649 if (ofproto->sflow) {
1650 dpif_sflow_run(ofproto->sflow);
1653 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1656 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1661 mac_learning_run(ofproto->ml, &ofproto->backer->revalidate_set);
1663 /* Check the consistency of a random facet, to aid debugging. */
1664 if (time_msec() >= ofproto->consistency_rl
1665 && !classifier_is_empty(&ofproto->facets)
1666 && !ofproto->backer->need_revalidate) {
1667 struct cls_table *table;
1668 struct cls_rule *cr;
1669 struct facet *facet;
1671 ofproto->consistency_rl = time_msec() + 250;
1673 table = CONTAINER_OF(hmap_random_node(&ofproto->facets.tables),
1674 struct cls_table, hmap_node);
1675 cr = CONTAINER_OF(hmap_random_node(&table->rules), struct cls_rule,
1677 facet = CONTAINER_OF(cr, struct facet, cr);
1679 if (!tag_set_intersects(&ofproto->backer->revalidate_set,
1680 facet->xout.tags)) {
1681 if (!facet_check_consistency(facet)) {
1682 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
1687 if (ofproto->governor) {
1690 governor_run(ofproto->governor);
1692 /* If the governor has shrunk to its minimum size and the number of
1693 * subfacets has dwindled, then drop the governor entirely.
1695 * For hysteresis, the number of subfacets to drop the governor is
1696 * smaller than the number needed to trigger its creation. */
1697 n_subfacets = hmap_count(&ofproto->subfacets);
1698 if (n_subfacets * 4 < ofproto->up.flow_eviction_threshold
1699 && governor_is_idle(ofproto->governor)) {
1700 governor_destroy(ofproto->governor);
1701 ofproto->governor = NULL;
1709 wait(struct ofproto *ofproto_)
1711 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1712 struct ofport_dpif *ofport;
1713 struct ofbundle *bundle;
1715 if (!clogged && !list_is_empty(&ofproto->completions)) {
1716 poll_immediate_wake();
1719 if (ofproto_get_flow_restore_wait()) {
1723 dpif_wait(ofproto->backer->dpif);
1724 dpif_recv_wait(ofproto->backer->dpif);
1725 if (ofproto->sflow) {
1726 dpif_sflow_wait(ofproto->sflow);
1728 if (!tag_set_is_empty(&ofproto->backer->revalidate_set)) {
1729 poll_immediate_wake();
1731 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1734 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1735 bundle_wait(bundle);
1737 if (ofproto->netflow) {
1738 netflow_wait(ofproto->netflow);
1740 mac_learning_wait(ofproto->ml);
1742 if (ofproto->backer->need_revalidate) {
1743 /* Shouldn't happen, but if it does just go around again. */
1744 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1745 poll_immediate_wake();
1747 if (ofproto->governor) {
1748 governor_wait(ofproto->governor);
1753 get_memory_usage(const struct ofproto *ofproto_, struct simap *usage)
1755 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1757 simap_increase(usage, "facets", classifier_count(&ofproto->facets));
1758 simap_increase(usage, "subfacets", hmap_count(&ofproto->subfacets));
1762 flush(struct ofproto *ofproto_)
1764 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1765 struct subfacet *subfacet, *next_subfacet;
1766 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
1770 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
1771 &ofproto->subfacets) {
1772 if (subfacet->path != SF_NOT_INSTALLED) {
1773 batch[n_batch++] = subfacet;
1774 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
1775 subfacet_destroy_batch(ofproto, batch, n_batch);
1779 subfacet_destroy(subfacet);
1784 subfacet_destroy_batch(ofproto, batch, n_batch);
1789 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1790 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1792 *arp_match_ip = true;
1793 *actions = (OFPUTIL_A_OUTPUT |
1794 OFPUTIL_A_SET_VLAN_VID |
1795 OFPUTIL_A_SET_VLAN_PCP |
1796 OFPUTIL_A_STRIP_VLAN |
1797 OFPUTIL_A_SET_DL_SRC |
1798 OFPUTIL_A_SET_DL_DST |
1799 OFPUTIL_A_SET_NW_SRC |
1800 OFPUTIL_A_SET_NW_DST |
1801 OFPUTIL_A_SET_NW_TOS |
1802 OFPUTIL_A_SET_TP_SRC |
1803 OFPUTIL_A_SET_TP_DST |
1808 get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots)
1810 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1811 struct dpif_dp_stats s;
1812 uint64_t n_miss, n_no_pkt_in, n_bytes, n_dropped_frags;
1815 strcpy(ots->name, "classifier");
1817 dpif_get_dp_stats(ofproto->backer->dpif, &s);
1818 rule_get_stats(&ofproto->miss_rule->up, &n_miss, &n_bytes);
1819 rule_get_stats(&ofproto->no_packet_in_rule->up, &n_no_pkt_in, &n_bytes);
1820 rule_get_stats(&ofproto->drop_frags_rule->up, &n_dropped_frags, &n_bytes);
1822 n_lookup = s.n_hit + s.n_missed - n_dropped_frags;
1823 ots->lookup_count = htonll(n_lookup);
1824 ots->matched_count = htonll(n_lookup - n_miss - n_no_pkt_in);
1827 static struct ofport *
1830 struct ofport_dpif *port = xmalloc(sizeof *port);
1835 port_dealloc(struct ofport *port_)
1837 struct ofport_dpif *port = ofport_dpif_cast(port_);
1842 port_construct(struct ofport *port_)
1844 struct ofport_dpif *port = ofport_dpif_cast(port_);
1845 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1846 const struct netdev *netdev = port->up.netdev;
1847 struct dpif_port dpif_port;
1850 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1851 port->bundle = NULL;
1853 port->tag = tag_create_random();
1854 port->may_enable = true;
1855 port->stp_port = NULL;
1856 port->stp_state = STP_DISABLED;
1857 port->tnl_port = NULL;
1858 hmap_init(&port->priorities);
1859 port->realdev_ofp_port = 0;
1860 port->vlandev_vid = 0;
1861 port->carrier_seq = netdev_get_carrier_resets(netdev);
1863 if (netdev_vport_is_patch(netdev)) {
1864 /* By bailing out here, we don't submit the port to the sFlow module
1865 * to be considered for counter polling export. This is correct
1866 * because the patch port represents an interface that sFlow considers
1867 * to be "internal" to the switch as a whole, and therefore not an
1868 * candidate for counter polling. */
1869 port->odp_port = OVSP_NONE;
1873 error = dpif_port_query_by_name(ofproto->backer->dpif,
1874 netdev_vport_get_dpif_port(netdev),
1880 port->odp_port = dpif_port.port_no;
1882 if (netdev_get_tunnel_config(netdev)) {
1883 port->tnl_port = tnl_port_add(&port->up, port->odp_port);
1885 /* Sanity-check that a mapping doesn't already exist. This
1886 * shouldn't happen for non-tunnel ports. */
1887 if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1888 VLOG_ERR("port %s already has an OpenFlow port number",
1890 dpif_port_destroy(&dpif_port);
1894 hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1895 hash_int(port->odp_port, 0));
1897 dpif_port_destroy(&dpif_port);
1899 if (ofproto->sflow) {
1900 dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1907 port_destruct(struct ofport *port_)
1909 struct ofport_dpif *port = ofport_dpif_cast(port_);
1910 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1911 const char *dp_port_name = netdev_vport_get_dpif_port(port->up.netdev);
1912 const char *devname = netdev_get_name(port->up.netdev);
1914 if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1915 /* The underlying device is still there, so delete it. This
1916 * happens when the ofproto is being destroyed, since the caller
1917 * assumes that removal of attached ports will happen as part of
1919 if (!port->tnl_port) {
1920 dpif_port_del(ofproto->backer->dpif, port->odp_port);
1922 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1925 if (port->odp_port != OVSP_NONE && !port->tnl_port) {
1926 hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1929 tnl_port_del(port->tnl_port);
1930 sset_find_and_delete(&ofproto->ports, devname);
1931 sset_find_and_delete(&ofproto->ghost_ports, devname);
1932 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1933 bundle_remove(port_);
1934 set_cfm(port_, NULL);
1935 if (ofproto->sflow) {
1936 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1939 ofport_clear_priorities(port);
1940 hmap_destroy(&port->priorities);
1944 port_modified(struct ofport *port_)
1946 struct ofport_dpif *port = ofport_dpif_cast(port_);
1948 if (port->bundle && port->bundle->bond) {
1949 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1953 cfm_set_netdev(port->cfm, port->up.netdev);
1958 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1960 struct ofport_dpif *port = ofport_dpif_cast(port_);
1961 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1962 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1964 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1965 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1966 OFPUTIL_PC_NO_PACKET_IN)) {
1967 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1969 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1970 bundle_update(port->bundle);
1976 set_sflow(struct ofproto *ofproto_,
1977 const struct ofproto_sflow_options *sflow_options)
1979 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1980 struct dpif_sflow *ds = ofproto->sflow;
1982 if (sflow_options) {
1984 struct ofport_dpif *ofport;
1986 ds = ofproto->sflow = dpif_sflow_create();
1987 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1988 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1990 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1992 dpif_sflow_set_options(ds, sflow_options);
1995 dpif_sflow_destroy(ds);
1996 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1997 ofproto->sflow = NULL;
2005 struct ofproto *ofproto_,
2006 const struct ofproto_ipfix_bridge_exporter_options *bridge_exporter_options,
2007 const struct ofproto_ipfix_flow_exporter_options *flow_exporters_options,
2008 size_t n_flow_exporters_options)
2010 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2011 struct dpif_ipfix *di = ofproto->ipfix;
2013 if (bridge_exporter_options || flow_exporters_options) {
2015 di = ofproto->ipfix = dpif_ipfix_create();
2017 dpif_ipfix_set_options(
2018 di, bridge_exporter_options, flow_exporters_options,
2019 n_flow_exporters_options);
2022 dpif_ipfix_destroy(di);
2023 ofproto->ipfix = NULL;
2030 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
2032 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2039 struct ofproto_dpif *ofproto;
2041 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2042 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2043 ofport->cfm = cfm_create(ofport->up.netdev);
2046 if (cfm_configure(ofport->cfm, s)) {
2052 cfm_destroy(ofport->cfm);
2058 get_cfm_status(const struct ofport *ofport_,
2059 struct ofproto_cfm_status *status)
2061 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2064 status->faults = cfm_get_fault(ofport->cfm);
2065 status->remote_opstate = cfm_get_opup(ofport->cfm);
2066 status->health = cfm_get_health(ofport->cfm);
2067 cfm_get_remote_mpids(ofport->cfm, &status->rmps, &status->n_rmps);
2074 /* Spanning Tree. */
2077 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
2079 struct ofproto_dpif *ofproto = ofproto_;
2080 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
2081 struct ofport_dpif *ofport;
2083 ofport = stp_port_get_aux(sp);
2085 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
2086 ofproto->up.name, port_num);
2088 struct eth_header *eth = pkt->l2;
2090 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
2091 if (eth_addr_is_zero(eth->eth_src)) {
2092 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
2093 "with unknown MAC", ofproto->up.name, port_num);
2095 send_packet(ofport, pkt);
2101 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
2103 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
2105 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2107 /* Only revalidate flows if the configuration changed. */
2108 if (!s != !ofproto->stp) {
2109 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2113 if (!ofproto->stp) {
2114 ofproto->stp = stp_create(ofproto_->name, s->system_id,
2115 send_bpdu_cb, ofproto);
2116 ofproto->stp_last_tick = time_msec();
2119 stp_set_bridge_id(ofproto->stp, s->system_id);
2120 stp_set_bridge_priority(ofproto->stp, s->priority);
2121 stp_set_hello_time(ofproto->stp, s->hello_time);
2122 stp_set_max_age(ofproto->stp, s->max_age);
2123 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
2125 struct ofport *ofport;
2127 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
2128 set_stp_port(ofport, NULL);
2131 stp_destroy(ofproto->stp);
2132 ofproto->stp = NULL;
2139 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
2141 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2145 s->bridge_id = stp_get_bridge_id(ofproto->stp);
2146 s->designated_root = stp_get_designated_root(ofproto->stp);
2147 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
2156 update_stp_port_state(struct ofport_dpif *ofport)
2158 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2159 enum stp_state state;
2161 /* Figure out new state. */
2162 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
2166 if (ofport->stp_state != state) {
2167 enum ofputil_port_state of_state;
2170 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
2171 netdev_get_name(ofport->up.netdev),
2172 stp_state_name(ofport->stp_state),
2173 stp_state_name(state));
2174 if (stp_learn_in_state(ofport->stp_state)
2175 != stp_learn_in_state(state)) {
2176 /* xxx Learning action flows should also be flushed. */
2177 mac_learning_flush(ofproto->ml,
2178 &ofproto->backer->revalidate_set);
2180 fwd_change = stp_forward_in_state(ofport->stp_state)
2181 != stp_forward_in_state(state);
2183 ofproto->backer->need_revalidate = REV_STP;
2184 ofport->stp_state = state;
2185 ofport->stp_state_entered = time_msec();
2187 if (fwd_change && ofport->bundle) {
2188 bundle_update(ofport->bundle);
2191 /* Update the STP state bits in the OpenFlow port description. */
2192 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
2193 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
2194 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
2195 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
2196 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
2198 ofproto_port_set_state(&ofport->up, of_state);
2202 /* Configures STP on 'ofport_' using the settings defined in 's'. The
2203 * caller is responsible for assigning STP port numbers and ensuring
2204 * there are no duplicates. */
2206 set_stp_port(struct ofport *ofport_,
2207 const struct ofproto_port_stp_settings *s)
2209 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2210 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2211 struct stp_port *sp = ofport->stp_port;
2213 if (!s || !s->enable) {
2215 ofport->stp_port = NULL;
2216 stp_port_disable(sp);
2217 update_stp_port_state(ofport);
2220 } else if (sp && stp_port_no(sp) != s->port_num
2221 && ofport == stp_port_get_aux(sp)) {
2222 /* The port-id changed, so disable the old one if it's not
2223 * already in use by another port. */
2224 stp_port_disable(sp);
2227 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
2228 stp_port_enable(sp);
2230 stp_port_set_aux(sp, ofport);
2231 stp_port_set_priority(sp, s->priority);
2232 stp_port_set_path_cost(sp, s->path_cost);
2234 update_stp_port_state(ofport);
2240 get_stp_port_status(struct ofport *ofport_,
2241 struct ofproto_port_stp_status *s)
2243 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2244 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2245 struct stp_port *sp = ofport->stp_port;
2247 if (!ofproto->stp || !sp) {
2253 s->port_id = stp_port_get_id(sp);
2254 s->state = stp_port_get_state(sp);
2255 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
2256 s->role = stp_port_get_role(sp);
2257 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
2263 stp_run(struct ofproto_dpif *ofproto)
2266 long long int now = time_msec();
2267 long long int elapsed = now - ofproto->stp_last_tick;
2268 struct stp_port *sp;
2271 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
2272 ofproto->stp_last_tick = now;
2274 while (stp_get_changed_port(ofproto->stp, &sp)) {
2275 struct ofport_dpif *ofport = stp_port_get_aux(sp);
2278 update_stp_port_state(ofport);
2282 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
2283 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
2289 stp_wait(struct ofproto_dpif *ofproto)
2292 poll_timer_wait(1000);
2296 /* Returns true if STP should process 'flow'. Sets fields in 'wc' that
2297 * were used to make the determination.*/
2299 stp_should_process_flow(const struct flow *flow, struct flow_wildcards *wc)
2301 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
2302 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
2306 stp_process_packet(const struct ofport_dpif *ofport,
2307 const struct ofpbuf *packet)
2309 struct ofpbuf payload = *packet;
2310 struct eth_header *eth = payload.data;
2311 struct stp_port *sp = ofport->stp_port;
2313 /* Sink packets on ports that have STP disabled when the bridge has
2315 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
2319 /* Trim off padding on payload. */
2320 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
2321 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
2324 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
2325 stp_received_bpdu(sp, payload.data, payload.size);
2329 static struct priority_to_dscp *
2330 get_priority(const struct ofport_dpif *ofport, uint32_t priority)
2332 struct priority_to_dscp *pdscp;
2335 hash = hash_int(priority, 0);
2336 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) {
2337 if (pdscp->priority == priority) {
2345 ofport_clear_priorities(struct ofport_dpif *ofport)
2347 struct priority_to_dscp *pdscp, *next;
2349 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) {
2350 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
2356 set_queues(struct ofport *ofport_,
2357 const struct ofproto_port_queue *qdscp_list,
2360 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2361 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2362 struct hmap new = HMAP_INITIALIZER(&new);
2365 for (i = 0; i < n_qdscp; i++) {
2366 struct priority_to_dscp *pdscp;
2370 dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
2371 if (dpif_queue_to_priority(ofproto->backer->dpif, qdscp_list[i].queue,
2376 pdscp = get_priority(ofport, priority);
2378 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
2380 pdscp = xmalloc(sizeof *pdscp);
2381 pdscp->priority = priority;
2383 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2386 if (pdscp->dscp != dscp) {
2388 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2391 hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0));
2394 if (!hmap_is_empty(&ofport->priorities)) {
2395 ofport_clear_priorities(ofport);
2396 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2399 hmap_swap(&new, &ofport->priorities);
2407 /* Expires all MAC learning entries associated with 'bundle' and forces its
2408 * ofproto to revalidate every flow.
2410 * Normally MAC learning entries are removed only from the ofproto associated
2411 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2412 * are removed from every ofproto. When patch ports and SLB bonds are in use
2413 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2414 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2415 * with the host from which it migrated. */
2417 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2419 struct ofproto_dpif *ofproto = bundle->ofproto;
2420 struct mac_learning *ml = ofproto->ml;
2421 struct mac_entry *mac, *next_mac;
2423 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2424 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2425 if (mac->port.p == bundle) {
2427 struct ofproto_dpif *o;
2429 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2431 struct mac_entry *e;
2433 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan,
2436 mac_learning_expire(o->ml, e);
2442 mac_learning_expire(ml, mac);
2447 static struct ofbundle *
2448 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2450 struct ofbundle *bundle;
2452 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2453 &ofproto->bundles) {
2454 if (bundle->aux == aux) {
2461 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
2462 * ones that are found to 'bundles'. */
2464 bundle_lookup_multiple(struct ofproto_dpif *ofproto,
2465 void **auxes, size_t n_auxes,
2466 struct hmapx *bundles)
2470 hmapx_init(bundles);
2471 for (i = 0; i < n_auxes; i++) {
2472 struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
2474 hmapx_add(bundles, bundle);
2480 bundle_update(struct ofbundle *bundle)
2482 struct ofport_dpif *port;
2484 bundle->floodable = true;
2485 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2486 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2487 || !stp_forward_in_state(port->stp_state)) {
2488 bundle->floodable = false;
2495 bundle_del_port(struct ofport_dpif *port)
2497 struct ofbundle *bundle = port->bundle;
2499 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2501 list_remove(&port->bundle_node);
2502 port->bundle = NULL;
2505 lacp_slave_unregister(bundle->lacp, port);
2508 bond_slave_unregister(bundle->bond, port);
2511 bundle_update(bundle);
2515 bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port,
2516 struct lacp_slave_settings *lacp)
2518 struct ofport_dpif *port;
2520 port = get_ofp_port(bundle->ofproto, ofp_port);
2525 if (port->bundle != bundle) {
2526 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2528 bundle_del_port(port);
2531 port->bundle = bundle;
2532 list_push_back(&bundle->ports, &port->bundle_node);
2533 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2534 || !stp_forward_in_state(port->stp_state)) {
2535 bundle->floodable = false;
2539 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2540 lacp_slave_register(bundle->lacp, port, lacp);
2547 bundle_destroy(struct ofbundle *bundle)
2549 struct ofproto_dpif *ofproto;
2550 struct ofport_dpif *port, *next_port;
2557 ofproto = bundle->ofproto;
2558 for (i = 0; i < MAX_MIRRORS; i++) {
2559 struct ofmirror *m = ofproto->mirrors[i];
2561 if (m->out == bundle) {
2563 } else if (hmapx_find_and_delete(&m->srcs, bundle)
2564 || hmapx_find_and_delete(&m->dsts, bundle)) {
2565 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2570 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2571 bundle_del_port(port);
2574 bundle_flush_macs(bundle, true);
2575 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2577 free(bundle->trunks);
2578 lacp_destroy(bundle->lacp);
2579 bond_destroy(bundle->bond);
2584 bundle_set(struct ofproto *ofproto_, void *aux,
2585 const struct ofproto_bundle_settings *s)
2587 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2588 bool need_flush = false;
2589 struct ofport_dpif *port;
2590 struct ofbundle *bundle;
2591 unsigned long *trunks;
2597 bundle_destroy(bundle_lookup(ofproto, aux));
2601 ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2602 ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2604 bundle = bundle_lookup(ofproto, aux);
2606 bundle = xmalloc(sizeof *bundle);
2608 bundle->ofproto = ofproto;
2609 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2610 hash_pointer(aux, 0));
2612 bundle->name = NULL;
2614 list_init(&bundle->ports);
2615 bundle->vlan_mode = PORT_VLAN_TRUNK;
2617 bundle->trunks = NULL;
2618 bundle->use_priority_tags = s->use_priority_tags;
2619 bundle->lacp = NULL;
2620 bundle->bond = NULL;
2622 bundle->floodable = true;
2624 bundle->src_mirrors = 0;
2625 bundle->dst_mirrors = 0;
2626 bundle->mirror_out = 0;
2629 if (!bundle->name || strcmp(s->name, bundle->name)) {
2631 bundle->name = xstrdup(s->name);
2636 if (!bundle->lacp) {
2637 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2638 bundle->lacp = lacp_create();
2640 lacp_configure(bundle->lacp, s->lacp);
2642 lacp_destroy(bundle->lacp);
2643 bundle->lacp = NULL;
2646 /* Update set of ports. */
2648 for (i = 0; i < s->n_slaves; i++) {
2649 if (!bundle_add_port(bundle, s->slaves[i],
2650 s->lacp ? &s->lacp_slaves[i] : NULL)) {
2654 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2655 struct ofport_dpif *next_port;
2657 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2658 for (i = 0; i < s->n_slaves; i++) {
2659 if (s->slaves[i] == port->up.ofp_port) {
2664 bundle_del_port(port);
2668 ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2670 if (list_is_empty(&bundle->ports)) {
2671 bundle_destroy(bundle);
2675 /* Set VLAN tagging mode */
2676 if (s->vlan_mode != bundle->vlan_mode
2677 || s->use_priority_tags != bundle->use_priority_tags) {
2678 bundle->vlan_mode = s->vlan_mode;
2679 bundle->use_priority_tags = s->use_priority_tags;
2684 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2685 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2687 if (vlan != bundle->vlan) {
2688 bundle->vlan = vlan;
2692 /* Get trunked VLANs. */
2693 switch (s->vlan_mode) {
2694 case PORT_VLAN_ACCESS:
2698 case PORT_VLAN_TRUNK:
2699 trunks = CONST_CAST(unsigned long *, s->trunks);
2702 case PORT_VLAN_NATIVE_UNTAGGED:
2703 case PORT_VLAN_NATIVE_TAGGED:
2704 if (vlan != 0 && (!s->trunks
2705 || !bitmap_is_set(s->trunks, vlan)
2706 || bitmap_is_set(s->trunks, 0))) {
2707 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2709 trunks = bitmap_clone(s->trunks, 4096);
2711 trunks = bitmap_allocate1(4096);
2713 bitmap_set1(trunks, vlan);
2714 bitmap_set0(trunks, 0);
2716 trunks = CONST_CAST(unsigned long *, s->trunks);
2723 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2724 free(bundle->trunks);
2725 if (trunks == s->trunks) {
2726 bundle->trunks = vlan_bitmap_clone(trunks);
2728 bundle->trunks = trunks;
2733 if (trunks != s->trunks) {
2738 if (!list_is_short(&bundle->ports)) {
2739 bundle->ofproto->has_bonded_bundles = true;
2741 if (bond_reconfigure(bundle->bond, s->bond)) {
2742 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2745 bundle->bond = bond_create(s->bond);
2746 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2749 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2750 bond_slave_register(bundle->bond, port, port->up.netdev);
2753 bond_destroy(bundle->bond);
2754 bundle->bond = NULL;
2757 /* If we changed something that would affect MAC learning, un-learn
2758 * everything on this port and force flow revalidation. */
2760 bundle_flush_macs(bundle, false);
2767 bundle_remove(struct ofport *port_)
2769 struct ofport_dpif *port = ofport_dpif_cast(port_);
2770 struct ofbundle *bundle = port->bundle;
2773 bundle_del_port(port);
2774 if (list_is_empty(&bundle->ports)) {
2775 bundle_destroy(bundle);
2776 } else if (list_is_short(&bundle->ports)) {
2777 bond_destroy(bundle->bond);
2778 bundle->bond = NULL;
2784 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2786 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2787 struct ofport_dpif *port = port_;
2788 uint8_t ea[ETH_ADDR_LEN];
2791 error = netdev_get_etheraddr(port->up.netdev, ea);
2793 struct ofpbuf packet;
2796 ofpbuf_init(&packet, 0);
2797 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2799 memcpy(packet_pdu, pdu, pdu_size);
2801 send_packet(port, &packet);
2802 ofpbuf_uninit(&packet);
2804 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2805 "%s (%s)", port->bundle->name,
2806 netdev_get_name(port->up.netdev), strerror(error));
2811 bundle_send_learning_packets(struct ofbundle *bundle)
2813 struct ofproto_dpif *ofproto = bundle->ofproto;
2814 int error, n_packets, n_errors;
2815 struct mac_entry *e;
2817 error = n_packets = n_errors = 0;
2818 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2819 if (e->port.p != bundle) {
2820 struct ofpbuf *learning_packet;
2821 struct ofport_dpif *port;
2825 /* The assignment to "port" is unnecessary but makes "grep"ing for
2826 * struct ofport_dpif more effective. */
2827 learning_packet = bond_compose_learning_packet(bundle->bond,
2831 ret = send_packet(port, learning_packet);
2832 ofpbuf_delete(learning_packet);
2842 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2843 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2844 "packets, last error was: %s",
2845 bundle->name, n_errors, n_packets, strerror(error));
2847 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2848 bundle->name, n_packets);
2853 bundle_run(struct ofbundle *bundle)
2856 lacp_run(bundle->lacp, send_pdu_cb);
2859 struct ofport_dpif *port;
2861 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2862 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2865 bond_run(bundle->bond, &bundle->ofproto->backer->revalidate_set,
2866 lacp_status(bundle->lacp));
2867 if (bond_should_send_learning_packets(bundle->bond)) {
2868 bundle_send_learning_packets(bundle);
2874 bundle_wait(struct ofbundle *bundle)
2877 lacp_wait(bundle->lacp);
2880 bond_wait(bundle->bond);
2887 mirror_scan(struct ofproto_dpif *ofproto)
2891 for (idx = 0; idx < MAX_MIRRORS; idx++) {
2892 if (!ofproto->mirrors[idx]) {
2899 static struct ofmirror *
2900 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
2904 for (i = 0; i < MAX_MIRRORS; i++) {
2905 struct ofmirror *mirror = ofproto->mirrors[i];
2906 if (mirror && mirror->aux == aux) {
2914 /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */
2916 mirror_update_dups(struct ofproto_dpif *ofproto)
2920 for (i = 0; i < MAX_MIRRORS; i++) {
2921 struct ofmirror *m = ofproto->mirrors[i];
2924 m->dup_mirrors = MIRROR_MASK_C(1) << i;
2928 for (i = 0; i < MAX_MIRRORS; i++) {
2929 struct ofmirror *m1 = ofproto->mirrors[i];
2936 for (j = i + 1; j < MAX_MIRRORS; j++) {
2937 struct ofmirror *m2 = ofproto->mirrors[j];
2939 if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) {
2940 m1->dup_mirrors |= MIRROR_MASK_C(1) << j;
2941 m2->dup_mirrors |= m1->dup_mirrors;
2948 mirror_set(struct ofproto *ofproto_, void *aux,
2949 const struct ofproto_mirror_settings *s)
2951 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2952 mirror_mask_t mirror_bit;
2953 struct ofbundle *bundle;
2954 struct ofmirror *mirror;
2955 struct ofbundle *out;
2956 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
2957 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
2960 mirror = mirror_lookup(ofproto, aux);
2962 mirror_destroy(mirror);
2968 idx = mirror_scan(ofproto);
2970 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
2972 ofproto->up.name, MAX_MIRRORS, s->name);
2976 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
2977 mirror->ofproto = ofproto;
2980 mirror->out_vlan = -1;
2981 mirror->name = NULL;
2984 if (!mirror->name || strcmp(s->name, mirror->name)) {
2986 mirror->name = xstrdup(s->name);
2989 /* Get the new configuration. */
2990 if (s->out_bundle) {
2991 out = bundle_lookup(ofproto, s->out_bundle);
2993 mirror_destroy(mirror);
2999 out_vlan = s->out_vlan;
3001 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
3002 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
3004 /* If the configuration has not changed, do nothing. */
3005 if (hmapx_equals(&srcs, &mirror->srcs)
3006 && hmapx_equals(&dsts, &mirror->dsts)
3007 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
3008 && mirror->out == out
3009 && mirror->out_vlan == out_vlan)
3011 hmapx_destroy(&srcs);
3012 hmapx_destroy(&dsts);
3016 hmapx_swap(&srcs, &mirror->srcs);
3017 hmapx_destroy(&srcs);
3019 hmapx_swap(&dsts, &mirror->dsts);
3020 hmapx_destroy(&dsts);
3022 free(mirror->vlans);
3023 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
3026 mirror->out_vlan = out_vlan;
3028 /* Update bundles. */
3029 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
3030 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
3031 if (hmapx_contains(&mirror->srcs, bundle)) {
3032 bundle->src_mirrors |= mirror_bit;
3034 bundle->src_mirrors &= ~mirror_bit;
3037 if (hmapx_contains(&mirror->dsts, bundle)) {
3038 bundle->dst_mirrors |= mirror_bit;
3040 bundle->dst_mirrors &= ~mirror_bit;
3043 if (mirror->out == bundle) {
3044 bundle->mirror_out |= mirror_bit;
3046 bundle->mirror_out &= ~mirror_bit;
3050 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3051 ofproto->has_mirrors = true;
3052 mac_learning_flush(ofproto->ml,
3053 &ofproto->backer->revalidate_set);
3054 mirror_update_dups(ofproto);
3060 mirror_destroy(struct ofmirror *mirror)
3062 struct ofproto_dpif *ofproto;
3063 mirror_mask_t mirror_bit;
3064 struct ofbundle *bundle;
3071 ofproto = mirror->ofproto;
3072 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3073 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
3075 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
3076 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
3077 bundle->src_mirrors &= ~mirror_bit;
3078 bundle->dst_mirrors &= ~mirror_bit;
3079 bundle->mirror_out &= ~mirror_bit;
3082 hmapx_destroy(&mirror->srcs);
3083 hmapx_destroy(&mirror->dsts);
3084 free(mirror->vlans);
3086 ofproto->mirrors[mirror->idx] = NULL;
3090 mirror_update_dups(ofproto);
3092 ofproto->has_mirrors = false;
3093 for (i = 0; i < MAX_MIRRORS; i++) {
3094 if (ofproto->mirrors[i]) {
3095 ofproto->has_mirrors = true;
3102 mirror_get_stats(struct ofproto *ofproto_, void *aux,
3103 uint64_t *packets, uint64_t *bytes)
3105 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3106 struct ofmirror *mirror = mirror_lookup(ofproto, aux);
3109 *packets = *bytes = UINT64_MAX;
3115 *packets = mirror->packet_count;
3116 *bytes = mirror->byte_count;
3122 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
3124 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3125 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
3126 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
3132 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
3134 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3135 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
3136 return bundle && bundle->mirror_out != 0;
3140 forward_bpdu_changed(struct ofproto *ofproto_)
3142 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3143 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3147 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
3150 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3151 mac_learning_set_idle_time(ofproto->ml, idle_time);
3152 mac_learning_set_max_entries(ofproto->ml, max_entries);
3157 static struct ofport_dpif *
3158 get_ofp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
3160 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
3161 return ofport ? ofport_dpif_cast(ofport) : NULL;
3164 static struct ofport_dpif *
3165 get_odp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
3167 struct ofport_dpif *port = odp_port_to_ofport(ofproto->backer, odp_port);
3168 return port && &ofproto->up == port->up.ofproto ? port : NULL;
3172 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
3173 struct ofproto_port *ofproto_port,
3174 struct dpif_port *dpif_port)
3176 ofproto_port->name = dpif_port->name;
3177 ofproto_port->type = dpif_port->type;
3178 ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
3181 static struct ofport_dpif *
3182 ofport_get_peer(const struct ofport_dpif *ofport_dpif)
3184 const struct ofproto_dpif *ofproto;
3187 peer = netdev_vport_patch_peer(ofport_dpif->up.netdev);
3192 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
3193 struct ofport *ofport;
3195 ofport = shash_find_data(&ofproto->up.port_by_name, peer);
3196 if (ofport && ofport->ofproto->ofproto_class == &ofproto_dpif_class) {
3197 return ofport_dpif_cast(ofport);
3204 port_run_fast(struct ofport_dpif *ofport)
3206 if (ofport->cfm && cfm_should_send_ccm(ofport->cfm)) {
3207 struct ofpbuf packet;
3209 ofpbuf_init(&packet, 0);
3210 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
3211 send_packet(ofport, &packet);
3212 ofpbuf_uninit(&packet);
3217 port_run(struct ofport_dpif *ofport)
3219 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
3220 bool carrier_changed = carrier_seq != ofport->carrier_seq;
3221 bool enable = netdev_get_carrier(ofport->up.netdev);
3223 ofport->carrier_seq = carrier_seq;
3225 port_run_fast(ofport);
3227 if (ofport->tnl_port
3228 && tnl_port_reconfigure(&ofport->up, ofport->odp_port,
3229 &ofport->tnl_port)) {
3230 ofproto_dpif_cast(ofport->up.ofproto)->backer->need_revalidate = true;
3234 int cfm_opup = cfm_get_opup(ofport->cfm);
3236 cfm_run(ofport->cfm);
3237 enable = enable && !cfm_get_fault(ofport->cfm);
3239 if (cfm_opup >= 0) {
3240 enable = enable && cfm_opup;
3244 if (ofport->bundle) {
3245 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
3246 if (carrier_changed) {
3247 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
3251 if (ofport->may_enable != enable) {
3252 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3254 if (ofproto->has_bundle_action) {
3255 ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
3259 ofport->may_enable = enable;
3263 port_wait(struct ofport_dpif *ofport)
3266 cfm_wait(ofport->cfm);
3271 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
3272 struct ofproto_port *ofproto_port)
3274 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3275 struct dpif_port dpif_port;
3278 if (sset_contains(&ofproto->ghost_ports, devname)) {
3279 const char *type = netdev_get_type_from_name(devname);
3281 /* We may be called before ofproto->up.port_by_name is populated with
3282 * the appropriate ofport. For this reason, we must get the name and
3283 * type from the netdev layer directly. */
3285 const struct ofport *ofport;
3287 ofport = shash_find_data(&ofproto->up.port_by_name, devname);
3288 ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
3289 ofproto_port->name = xstrdup(devname);
3290 ofproto_port->type = xstrdup(type);
3296 if (!sset_contains(&ofproto->ports, devname)) {
3299 error = dpif_port_query_by_name(ofproto->backer->dpif,
3300 devname, &dpif_port);
3302 ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
3308 port_add(struct ofproto *ofproto_, struct netdev *netdev)
3310 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3311 const char *dp_port_name = netdev_vport_get_dpif_port(netdev);
3312 const char *devname = netdev_get_name(netdev);
3314 if (netdev_vport_is_patch(netdev)) {
3315 sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
3319 if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
3320 uint32_t port_no = UINT32_MAX;
3323 error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
3327 if (netdev_get_tunnel_config(netdev)) {
3328 simap_put(&ofproto->backer->tnl_backers, dp_port_name, port_no);
3332 if (netdev_get_tunnel_config(netdev)) {
3333 sset_add(&ofproto->ghost_ports, devname);
3335 sset_add(&ofproto->ports, devname);
3341 port_del(struct ofproto *ofproto_, uint16_t ofp_port)
3343 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3344 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
3351 sset_find_and_delete(&ofproto->ghost_ports,
3352 netdev_get_name(ofport->up.netdev));
3353 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3354 if (!ofport->tnl_port) {
3355 error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
3357 /* The caller is going to close ofport->up.netdev. If this is a
3358 * bonded port, then the bond is using that netdev, so remove it
3359 * from the bond. The client will need to reconfigure everything
3360 * after deleting ports, so then the slave will get re-added. */
3361 bundle_remove(&ofport->up);
3368 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
3370 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3375 error = netdev_get_stats(ofport->up.netdev, stats);
3377 if (!error && ofport_->ofp_port == OFPP_LOCAL) {
3378 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3380 /* ofproto->stats.tx_packets represents packets that we created
3381 * internally and sent to some port (e.g. packets sent with
3382 * send_packet()). Account for them as if they had come from
3383 * OFPP_LOCAL and got forwarded. */
3385 if (stats->rx_packets != UINT64_MAX) {
3386 stats->rx_packets += ofproto->stats.tx_packets;
3389 if (stats->rx_bytes != UINT64_MAX) {
3390 stats->rx_bytes += ofproto->stats.tx_bytes;
3393 /* ofproto->stats.rx_packets represents packets that were received on
3394 * some port and we processed internally and dropped (e.g. STP).
3395 * Account for them as if they had been forwarded to OFPP_LOCAL. */
3397 if (stats->tx_packets != UINT64_MAX) {
3398 stats->tx_packets += ofproto->stats.rx_packets;
3401 if (stats->tx_bytes != UINT64_MAX) {
3402 stats->tx_bytes += ofproto->stats.rx_bytes;
3409 /* Account packets for LOCAL port. */
3411 ofproto_update_local_port_stats(const struct ofproto *ofproto_,
3412 size_t tx_size, size_t rx_size)
3414 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3417 ofproto->stats.rx_packets++;
3418 ofproto->stats.rx_bytes += rx_size;
3421 ofproto->stats.tx_packets++;
3422 ofproto->stats.tx_bytes += tx_size;
3426 struct port_dump_state {
3431 struct ofproto_port port;
3436 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
3438 *statep = xzalloc(sizeof(struct port_dump_state));
3443 port_dump_next(const struct ofproto *ofproto_, void *state_,
3444 struct ofproto_port *port)
3446 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3447 struct port_dump_state *state = state_;
3448 const struct sset *sset;
3449 struct sset_node *node;
3451 if (state->has_port) {
3452 ofproto_port_destroy(&state->port);
3453 state->has_port = false;
3455 sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
3456 while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
3459 error = port_query_by_name(ofproto_, node->name, &state->port);
3461 *port = state->port;
3462 state->has_port = true;
3464 } else if (error != ENODEV) {
3469 if (!state->ghost) {
3470 state->ghost = true;
3473 return port_dump_next(ofproto_, state_, port);
3480 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3482 struct port_dump_state *state = state_;
3484 if (state->has_port) {
3485 ofproto_port_destroy(&state->port);
3492 port_poll(const struct ofproto *ofproto_, char **devnamep)
3494 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3496 if (ofproto->port_poll_errno) {
3497 int error = ofproto->port_poll_errno;
3498 ofproto->port_poll_errno = 0;
3502 if (sset_is_empty(&ofproto->port_poll_set)) {
3506 *devnamep = sset_pop(&ofproto->port_poll_set);
3511 port_poll_wait(const struct ofproto *ofproto_)
3513 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3514 dpif_port_poll_wait(ofproto->backer->dpif);
3518 port_is_lacp_current(const struct ofport *ofport_)
3520 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3521 return (ofport->bundle && ofport->bundle->lacp
3522 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3526 /* Upcall handling. */
3528 /* Flow miss batching.
3530 * Some dpifs implement operations faster when you hand them off in a batch.
3531 * To allow batching, "struct flow_miss" queues the dpif-related work needed
3532 * for a given flow. Each "struct flow_miss" corresponds to sending one or
3533 * more packets, plus possibly installing the flow in the dpif.
3535 * So far we only batch the operations that affect flow setup time the most.
3536 * It's possible to batch more than that, but the benefit might be minimal. */
3538 struct hmap_node hmap_node;
3539 struct ofproto_dpif *ofproto;
3541 enum odp_key_fitness key_fitness;
3542 const struct nlattr *key;
3544 struct initial_vals initial_vals;
3545 struct list packets;
3546 enum dpif_upcall_type upcall_type;
3547 uint32_t odp_in_port;
3550 struct flow_miss_op {
3551 struct dpif_op dpif_op;
3553 uint64_t slow_stub[128 / 8]; /* Buffer for compose_slow_path() */
3554 struct xlate_out xout;
3555 bool xout_garbage; /* 'xout' needs to be uninitialized? */
3557 struct ofpbuf mask; /* Flow mask for "put" ops. */
3558 struct odputil_keybuf maskbuf;
3560 /* If this is a "put" op, then a pointer to the subfacet that should
3561 * be marked as uninstalled if the operation fails. */
3562 struct subfacet *subfacet;
3565 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
3566 * OpenFlow controller as necessary according to their individual
3567 * configurations. */
3569 send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet,
3570 const struct flow *flow)
3572 struct ofputil_packet_in pin;
3574 pin.packet = packet->data;
3575 pin.packet_len = packet->size;
3576 pin.reason = OFPR_NO_MATCH;
3577 pin.controller_id = 0;
3582 pin.send_len = 0; /* not used for flow table misses */
3584 flow_get_metadata(flow, &pin.fmd);
3586 connmgr_send_packet_in(ofproto->up.connmgr, &pin);
3589 static enum slow_path_reason
3590 process_special(struct xlate_ctx *ctx, const struct flow *flow,
3591 const struct ofport_dpif *ofport, const struct ofpbuf *packet)
3593 struct ofproto_dpif *ofproto = ctx->ofproto;
3594 struct flow_wildcards *wc = &ctx->xout->wc;
3598 } else if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow, wc)) {
3600 cfm_process_heartbeat(ofport->cfm, packet);
3603 } else if (ofport->bundle && ofport->bundle->lacp
3604 && flow->dl_type == htons(ETH_TYPE_LACP)) {
3606 lacp_process_packet(ofport->bundle->lacp, ofport, packet);
3609 } else if (ofproto->stp && stp_should_process_flow(flow, wc)) {
3611 stp_process_packet(ofport, packet);
3619 static struct flow_miss *
3620 flow_miss_find(struct hmap *todo, const struct ofproto_dpif *ofproto,
3621 const struct flow *flow, uint32_t hash)
3623 struct flow_miss *miss;
3625 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
3626 if (miss->ofproto == ofproto && flow_equal(&miss->flow, flow)) {
3634 /* Partially Initializes 'op' as an "execute" operation for 'miss' and
3635 * 'packet'. The caller must initialize op->actions and op->actions_len. If
3636 * 'miss' is associated with a subfacet the caller must also initialize the
3637 * returned op->subfacet, and if anything needs to be freed after processing
3638 * the op, the caller must initialize op->garbage also. */
3640 init_flow_miss_execute_op(struct flow_miss *miss, struct ofpbuf *packet,
3641 struct flow_miss_op *op)
3643 if (miss->flow.vlan_tci != miss->initial_vals.vlan_tci) {
3644 /* This packet was received on a VLAN splinter port. We
3645 * added a VLAN to the packet to make the packet resemble
3646 * the flow, but the actions were composed assuming that
3647 * the packet contained no VLAN. So, we must remove the
3648 * VLAN header from the packet before trying to execute the
3650 eth_pop_vlan(packet);
3653 op->subfacet = NULL;
3654 op->xout_garbage = false;
3655 op->dpif_op.type = DPIF_OP_EXECUTE;
3656 op->dpif_op.u.execute.key = miss->key;
3657 op->dpif_op.u.execute.key_len = miss->key_len;
3658 op->dpif_op.u.execute.packet = packet;
3659 ofpbuf_use_stack(&op->mask, &op->maskbuf, sizeof op->maskbuf);
3662 /* Helper for handle_flow_miss_without_facet() and
3663 * handle_flow_miss_with_facet(). */
3665 handle_flow_miss_common(struct rule_dpif *rule,
3666 struct ofpbuf *packet, const struct flow *flow)
3668 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3670 if (rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
3672 * Extra-special case for fail-open mode.
3674 * We are in fail-open mode and the packet matched the fail-open
3675 * rule, but we are connected to a controller too. We should send
3676 * the packet up to the controller in the hope that it will try to
3677 * set up a flow and thereby allow us to exit fail-open.
3679 * See the top-level comment in fail-open.c for more information.
3681 send_packet_in_miss(ofproto, packet, flow);
3685 /* Figures out whether a flow that missed in 'ofproto', whose details are in
3686 * 'miss' masked by 'wc', is likely to be worth tracking in detail in userspace
3687 * and (usually) installing a datapath flow. The answer is usually "yes" (a
3688 * return value of true). However, for short flows the cost of bookkeeping is
3689 * much higher than the benefits, so when the datapath holds a large number of
3690 * flows we impose some heuristics to decide which flows are likely to be worth
3693 flow_miss_should_make_facet(struct flow_miss *miss, struct flow_wildcards *wc)
3695 struct ofproto_dpif *ofproto = miss->ofproto;
3698 if (!ofproto->governor) {
3701 n_subfacets = hmap_count(&ofproto->subfacets);
3702 if (n_subfacets * 2 <= ofproto->up.flow_eviction_threshold) {
3706 ofproto->governor = governor_create(ofproto->up.name);
3709 hash = flow_hash_in_wildcards(&miss->flow, wc, 0);
3710 return governor_should_install_flow(ofproto->governor, hash,
3711 list_size(&miss->packets));
3714 /* Handles 'miss' without creating a facet or subfacet or creating any datapath
3715 * flow. 'miss->flow' must have matched 'rule' and been xlated into 'xout'.
3716 * May add an "execute" operation to 'ops' and increment '*n_ops'. */
3718 handle_flow_miss_without_facet(struct rule_dpif *rule, struct xlate_out *xout,
3719 struct flow_miss *miss,
3720 struct flow_miss_op *ops, size_t *n_ops)
3722 struct ofpbuf *packet;
3724 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3726 COVERAGE_INC(facet_suppress);
3728 handle_flow_miss_common(rule, packet, &miss->flow);
3731 struct xlate_in xin;
3733 xlate_in_init(&xin, miss->ofproto, &miss->flow,
3734 &miss->initial_vals, rule, 0, packet);
3735 xlate_actions_for_side_effects(&xin);
3738 if (xout->odp_actions.size) {
3739 struct flow_miss_op *op = &ops[*n_ops];
3740 struct dpif_execute *execute = &op->dpif_op.u.execute;
3742 init_flow_miss_execute_op(miss, packet, op);
3743 xlate_out_copy(&op->xout, xout);
3744 execute->actions = op->xout.odp_actions.data;
3745 execute->actions_len = op->xout.odp_actions.size;
3746 op->xout_garbage = true;
3753 /* Handles 'miss', which matches 'facet'. May add any required datapath
3754 * operations to 'ops', incrementing '*n_ops' for each new op.
3756 * All of the packets in 'miss' are considered to have arrived at time 'now'.
3757 * This is really important only for new facets: if we just called time_msec()
3758 * here, then the new subfacet or its packets could look (occasionally) as
3759 * though it was used some time after the facet was used. That can make a
3760 * one-packet flow look like it has a nonzero duration, which looks odd in
3761 * e.g. NetFlow statistics.
3763 * If non-null, 'stats' will be folded into 'facet'. */
3765 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
3766 long long int now, struct dpif_flow_stats *stats,
3767 struct flow_miss_op *ops, size_t *n_ops)
3769 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3770 enum subfacet_path want_path;
3771 struct subfacet *subfacet;
3772 struct ofpbuf *packet;
3774 subfacet = subfacet_create(facet, miss, now);
3775 want_path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
3777 subfacet_update_stats(subfacet, stats);
3780 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3781 struct flow_miss_op *op = &ops[*n_ops];
3783 handle_flow_miss_common(facet->rule, packet, &miss->flow);
3785 if (want_path != SF_FAST_PATH) {
3786 struct xlate_in xin;
3788 xlate_in_init(&xin, ofproto, &miss->flow, &miss->initial_vals,
3789 facet->rule, 0, packet);
3790 xlate_actions_for_side_effects(&xin);
3793 if (facet->xout.odp_actions.size) {
3794 struct dpif_execute *execute = &op->dpif_op.u.execute;
3796 init_flow_miss_execute_op(miss, packet, op);
3797 execute->actions = facet->xout.odp_actions.data,
3798 execute->actions_len = facet->xout.odp_actions.size;
3803 if (miss->upcall_type == DPIF_UC_MISS || subfacet->path != want_path) {
3804 struct flow_miss_op *op = &ops[(*n_ops)++];
3805 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
3807 subfacet->path = want_path;
3809 ofpbuf_use_stack(&op->mask, &op->maskbuf, sizeof op->maskbuf);
3810 if (enable_megaflows) {
3811 odp_flow_key_from_mask(&op->mask, &facet->xout.wc.masks,
3812 &miss->flow, UINT32_MAX);
3815 op->xout_garbage = false;
3816 op->dpif_op.type = DPIF_OP_FLOW_PUT;
3817 op->subfacet = subfacet;
3818 put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
3819 put->key = miss->key;
3820 put->key_len = miss->key_len;
3821 put->mask = op->mask.data;
3822 put->mask_len = op->mask.size;
3824 if (want_path == SF_FAST_PATH) {
3825 put->actions = facet->xout.odp_actions.data;
3826 put->actions_len = facet->xout.odp_actions.size;
3828 compose_slow_path(ofproto, &miss->flow, facet->xout.slow,
3829 op->slow_stub, sizeof op->slow_stub,
3830 &put->actions, &put->actions_len);
3836 /* Handles flow miss 'miss'. May add any required datapath operations
3837 * to 'ops', incrementing '*n_ops' for each new op. */
3839 handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops,
3842 struct ofproto_dpif *ofproto = miss->ofproto;
3843 struct dpif_flow_stats stats__;
3844 struct dpif_flow_stats *stats = &stats__;
3845 struct ofpbuf *packet;
3846 struct facet *facet;
3850 memset(stats, 0, sizeof *stats);
3852 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3853 stats->tcp_flags |= packet_get_tcp_flags(packet, &miss->flow);
3854 stats->n_bytes += packet->size;
3858 facet = facet_lookup_valid(ofproto, &miss->flow);
3860 struct flow_wildcards wc;
3861 struct rule_dpif *rule;
3862 struct xlate_out xout;
3863 struct xlate_in xin;
3865 flow_wildcards_init_catchall(&wc);
3866 rule = rule_dpif_lookup(ofproto, &miss->flow, &wc);
3867 rule_credit_stats(rule, stats);
3869 xlate_in_init(&xin, ofproto, &miss->flow, &miss->initial_vals, rule,
3870 stats->tcp_flags, NULL);
3871 xin.resubmit_stats = stats;
3872 xin.may_learn = true;
3873 xlate_actions(&xin, &xout);
3874 flow_wildcards_or(&xout.wc, &xout.wc, &wc);
3876 /* There does not exist a bijection between 'struct flow' and datapath
3877 * flow keys with fitness ODP_FIT_TO_LITTLE. This breaks a fundamental
3878 * assumption used throughout the facet and subfacet handling code.
3879 * Since we have to handle these misses in userspace anyway, we simply
3880 * skip facet creation, avoiding the problem altogether. */
3881 if (miss->key_fitness == ODP_FIT_TOO_LITTLE
3882 || !flow_miss_should_make_facet(miss, &xout.wc)) {
3883 handle_flow_miss_without_facet(rule, &xout, miss, ops, n_ops);
3887 facet = facet_create(miss, rule, &xout, stats);
3890 handle_flow_miss_with_facet(miss, facet, now, stats, ops, n_ops);
3893 static struct drop_key *
3894 drop_key_lookup(const struct dpif_backer *backer, const struct nlattr *key,
3897 struct drop_key *drop_key;
3899 HMAP_FOR_EACH_WITH_HASH (drop_key, hmap_node, hash_bytes(key, key_len, 0),
3900 &backer->drop_keys) {
3901 if (drop_key->key_len == key_len
3902 && !memcmp(drop_key->key, key, key_len)) {
3910 drop_key_clear(struct dpif_backer *backer)
3912 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3913 struct drop_key *drop_key, *next;
3915 HMAP_FOR_EACH_SAFE (drop_key, next, hmap_node, &backer->drop_keys) {
3918 error = dpif_flow_del(backer->dpif, drop_key->key, drop_key->key_len,
3920 if (error && !VLOG_DROP_WARN(&rl)) {
3921 struct ds ds = DS_EMPTY_INITIALIZER;
3922 odp_flow_key_format(drop_key->key, drop_key->key_len, &ds);
3923 VLOG_WARN("Failed to delete drop key (%s) (%s)", strerror(error),
3928 hmap_remove(&backer->drop_keys, &drop_key->hmap_node);
3929 free(drop_key->key);
3934 /* Given a datpath, packet, and flow metadata ('backer', 'packet', and 'key'
3935 * respectively), populates 'flow' with the result of odp_flow_key_to_flow().
3936 * Optionally, if nonnull, populates 'fitnessp' with the fitness of 'flow' as
3937 * returned by odp_flow_key_to_flow(). Also, optionally populates 'ofproto'
3938 * with the ofproto_dpif, and 'odp_in_port' with the datapath in_port, that
3939 * 'packet' ingressed.
3941 * If 'ofproto' is nonnull, requires 'flow''s in_port to exist. Otherwise sets
3942 * 'flow''s in_port to OFPP_NONE.
3944 * This function does post-processing on data returned from
3945 * odp_flow_key_to_flow() to help make VLAN splinters transparent to the rest
3946 * of the upcall processing logic. In particular, if the extracted in_port is
3947 * a VLAN splinter port, it replaces flow->in_port by the "real" port, sets
3948 * flow->vlan_tci correctly for the VLAN of the VLAN splinter port, and pushes
3949 * a VLAN header onto 'packet' (if it is nonnull).
3951 * Optionally, if 'initial_vals' is nonnull, sets 'initial_vals->vlan_tci'
3952 * to the VLAN TCI with which the packet was really received, that is, the
3953 * actual VLAN TCI extracted by odp_flow_key_to_flow(). (This differs from
3954 * the value returned in flow->vlan_tci only for packets received on
3955 * VLAN splinters.) Also, if received on an IP tunnel, sets
3956 * 'initial_vals->tunnel_ip_tos' to the tunnel's IP TOS.
3958 * Similarly, this function also includes some logic to help with tunnels. It
3959 * may modify 'flow' as necessary to make the tunneling implementation
3960 * transparent to the upcall processing logic.
3962 * Returns 0 if successful, ENODEV if the parsed flow has no associated ofport,
3963 * or some other positive errno if there are other problems. */
3965 ofproto_receive(const struct dpif_backer *backer, struct ofpbuf *packet,
3966 const struct nlattr *key, size_t key_len,
3967 struct flow *flow, enum odp_key_fitness *fitnessp,
3968 struct ofproto_dpif **ofproto, uint32_t *odp_in_port,
3969 struct initial_vals *initial_vals)
3971 const struct ofport_dpif *port;
3972 enum odp_key_fitness fitness;
3975 fitness = odp_flow_key_to_flow(key, key_len, flow);
3976 if (fitness == ODP_FIT_ERROR) {
3982 initial_vals->vlan_tci = flow->vlan_tci;
3983 initial_vals->tunnel_ip_tos = flow->tunnel.ip_tos;
3987 *odp_in_port = flow->in_port;
3990 if (tnl_port_should_receive(flow)) {
3991 const struct ofport *ofport = tnl_port_receive(flow);
3993 flow->in_port = OFPP_NONE;
3996 port = ofport_dpif_cast(ofport);
3998 /* We can't reproduce 'key' from 'flow'. */
3999 fitness = fitness == ODP_FIT_PERFECT ? ODP_FIT_TOO_MUCH : fitness;
4001 /* XXX: Since the tunnel module is not scoped per backer, it's
4002 * theoretically possible that we'll receive an ofport belonging to an
4003 * entirely different datapath. In practice, this can't happen because
4004 * no platforms has two separate datapaths which each support
4006 ovs_assert(ofproto_dpif_cast(port->up.ofproto)->backer == backer);
4008 port = odp_port_to_ofport(backer, flow->in_port);
4010 flow->in_port = OFPP_NONE;
4014 flow->in_port = port->up.ofp_port;
4015 if (vsp_adjust_flow(ofproto_dpif_cast(port->up.ofproto), flow)) {
4017 /* Make the packet resemble the flow, so that it gets sent to
4018 * an OpenFlow controller properly, so that it looks correct
4019 * for sFlow, and so that flow_extract() will get the correct
4020 * vlan_tci if it is called on 'packet'.
4022 * The allocated space inside 'packet' probably also contains
4023 * 'key', that is, both 'packet' and 'key' are probably part of
4024 * a struct dpif_upcall (see the large comment on that
4025 * structure definition), so pushing data on 'packet' is in
4026 * general not a good idea since it could overwrite 'key' or
4027 * free it as a side effect. However, it's OK in this special
4028 * case because we know that 'packet' is inside a Netlink
4029 * attribute: pushing 4 bytes will just overwrite the 4-byte
4030 * "struct nlattr", which is fine since we don't need that
4031 * header anymore. */
4032 eth_push_vlan(packet, flow->vlan_tci);
4034 /* We can't reproduce 'key' from 'flow'. */
4035 fitness = fitness == ODP_FIT_PERFECT ? ODP_FIT_TOO_MUCH : fitness;
4041 *ofproto = ofproto_dpif_cast(port->up.ofproto);
4046 *fitnessp = fitness;
4052 handle_miss_upcalls(struct dpif_backer *backer, struct dpif_upcall *upcalls,
4055 struct dpif_upcall *upcall;
4056 struct flow_miss *miss;
4057 struct flow_miss misses[FLOW_MISS_MAX_BATCH];
4058 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
4059 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
4069 /* Construct the to-do list.
4071 * This just amounts to extracting the flow from each packet and sticking
4072 * the packets that have the same flow in the same "flow_miss" structure so
4073 * that we can process them together. */
4076 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
4077 struct flow_miss *miss = &misses[n_misses];
4078 struct flow_miss *existing_miss;
4079 struct ofproto_dpif *ofproto;
4080 uint32_t odp_in_port;
4085 error = ofproto_receive(backer, upcall->packet, upcall->key,
4086 upcall->key_len, &flow, &miss->key_fitness,
4087 &ofproto, &odp_in_port, &miss->initial_vals);
4088 if (error == ENODEV) {
4089 struct drop_key *drop_key;
4091 /* Received packet on port for which we couldn't associate
4092 * an ofproto. This can happen if a port is removed while
4093 * traffic is being received. Print a rate-limited message
4094 * in case it happens frequently. Install a drop flow so
4095 * that future packets of the flow are inexpensively dropped
4097 VLOG_INFO_RL(&rl, "received packet on unassociated port %"PRIu32,
4100 drop_key = drop_key_lookup(backer, upcall->key, upcall->key_len);
4102 drop_key = xmalloc(sizeof *drop_key);
4103 drop_key->key = xmemdup(upcall->key, upcall->key_len);
4104 drop_key->key_len = upcall->key_len;
4106 hmap_insert(&backer->drop_keys, &drop_key->hmap_node,
4107 hash_bytes(drop_key->key, drop_key->key_len, 0));
4108 dpif_flow_put(backer->dpif, DPIF_FP_CREATE | DPIF_FP_MODIFY,
4109 drop_key->key, drop_key->key_len,
4110 NULL, 0, NULL, 0, NULL);
4118 ofproto->n_missed++;
4119 flow_extract(upcall->packet, flow.skb_priority, flow.skb_mark,
4120 &flow.tunnel, flow.in_port, &miss->flow);
4122 /* Add other packets to a to-do list. */
4123 hash = flow_hash(&miss->flow, 0);
4124 existing_miss = flow_miss_find(&todo, ofproto, &miss->flow, hash);
4125 if (!existing_miss) {
4126 hmap_insert(&todo, &miss->hmap_node, hash);
4127 miss->ofproto = ofproto;
4128 miss->key = upcall->key;
4129 miss->key_len = upcall->key_len;
4130 miss->upcall_type = upcall->type;
4131 miss->odp_in_port = odp_in_port;
4132 list_init(&miss->packets);
4136 miss = existing_miss;
4138 list_push_back(&miss->packets, &upcall->packet->list_node);
4141 /* Process each element in the to-do list, constructing the set of
4142 * operations to batch. */
4144 HMAP_FOR_EACH (miss, hmap_node, &todo) {
4145 handle_flow_miss(miss, flow_miss_ops, &n_ops);
4147 ovs_assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
4149 /* Execute batch. */
4150 for (i = 0; i < n_ops; i++) {
4151 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
4153 dpif_operate(backer->dpif, dpif_ops, n_ops);
4155 for (i = 0; i < n_ops; i++) {
4156 if (dpif_ops[i]->error != 0
4157 && flow_miss_ops[i].dpif_op.type == DPIF_OP_FLOW_PUT
4158 && flow_miss_ops[i].subfacet) {
4159 struct subfacet *subfacet = flow_miss_ops[i].subfacet;
4161 COVERAGE_INC(subfacet_install_fail);
4163 subfacet->path = SF_NOT_INSTALLED;
4167 if (flow_miss_ops[i].xout_garbage) {
4168 xlate_out_uninit(&flow_miss_ops[i].xout);
4171 hmap_destroy(&todo);
4174 static enum { SFLOW_UPCALL, MISS_UPCALL, BAD_UPCALL, FLOW_SAMPLE_UPCALL,
4176 classify_upcall(const struct dpif_upcall *upcall)
4178 size_t userdata_len;
4179 union user_action_cookie cookie;
4181 /* First look at the upcall type. */
4182 switch (upcall->type) {
4183 case DPIF_UC_ACTION:
4189 case DPIF_N_UC_TYPES:
4191 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
4195 /* "action" upcalls need a closer look. */
4196 if (!upcall->userdata) {
4197 VLOG_WARN_RL(&rl, "action upcall missing cookie");
4200 userdata_len = nl_attr_get_size(upcall->userdata);
4201 if (userdata_len < sizeof cookie.type
4202 || userdata_len > sizeof cookie) {
4203 VLOG_WARN_RL(&rl, "action upcall cookie has unexpected size %zu",
4207 memset(&cookie, 0, sizeof cookie);
4208 memcpy(&cookie, nl_attr_get(upcall->userdata), userdata_len);
4209 if (userdata_len == sizeof cookie.sflow
4210 && cookie.type == USER_ACTION_COOKIE_SFLOW) {
4211 return SFLOW_UPCALL;
4212 } else if (userdata_len == sizeof cookie.slow_path
4213 && cookie.type == USER_ACTION_COOKIE_SLOW_PATH) {
4215 } else if (userdata_len == sizeof cookie.flow_sample
4216 && cookie.type == USER_ACTION_COOKIE_FLOW_SAMPLE) {
4217 return FLOW_SAMPLE_UPCALL;
4218 } else if (userdata_len == sizeof cookie.ipfix
4219 && cookie.type == USER_ACTION_COOKIE_IPFIX) {
4220 return IPFIX_UPCALL;
4222 VLOG_WARN_RL(&rl, "invalid user cookie of type %"PRIu16
4223 " and size %zu", cookie.type, userdata_len);
4229 handle_sflow_upcall(struct dpif_backer *backer,
4230 const struct dpif_upcall *upcall)
4232 struct ofproto_dpif *ofproto;
4233 union user_action_cookie cookie;
4235 uint32_t odp_in_port;
4237 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
4238 &flow, NULL, &ofproto, &odp_in_port, NULL)
4239 || !ofproto->sflow) {
4243 memset(&cookie, 0, sizeof cookie);
4244 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.sflow);
4245 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
4246 odp_in_port, &cookie);
4250 handle_flow_sample_upcall(struct dpif_backer *backer,
4251 const struct dpif_upcall *upcall)
4253 struct ofproto_dpif *ofproto;
4254 union user_action_cookie cookie;
4257 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
4258 &flow, NULL, &ofproto, NULL, NULL)
4259 || !ofproto->ipfix) {
4263 memset(&cookie, 0, sizeof cookie);
4264 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.flow_sample);
4266 /* The flow reflects exactly the contents of the packet. Sample
4267 * the packet using it. */
4268 dpif_ipfix_flow_sample(ofproto->ipfix, upcall->packet, &flow,
4269 cookie.flow_sample.collector_set_id,
4270 cookie.flow_sample.probability,
4271 cookie.flow_sample.obs_domain_id,
4272 cookie.flow_sample.obs_point_id);
4276 handle_ipfix_upcall(struct dpif_backer *backer,
4277 const struct dpif_upcall *upcall)
4279 struct ofproto_dpif *ofproto;
4282 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
4283 &flow, NULL, &ofproto, NULL, NULL)
4284 || !ofproto->ipfix) {
4288 /* The flow reflects exactly the contents of the packet. Sample
4289 * the packet using it. */
4290 dpif_ipfix_bridge_sample(ofproto->ipfix, upcall->packet, &flow);
4294 handle_upcalls(struct dpif_backer *backer, unsigned int max_batch)
4296 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
4297 struct ofpbuf miss_bufs[FLOW_MISS_MAX_BATCH];
4298 uint64_t miss_buf_stubs[FLOW_MISS_MAX_BATCH][4096 / 8];
4303 ovs_assert(max_batch <= FLOW_MISS_MAX_BATCH);
4306 for (n_processed = 0; n_processed < max_batch; n_processed++) {
4307 struct dpif_upcall *upcall = &misses[n_misses];
4308 struct ofpbuf *buf = &miss_bufs[n_misses];
4311 ofpbuf_use_stub(buf, miss_buf_stubs[n_misses],
4312 sizeof miss_buf_stubs[n_misses]);
4313 error = dpif_recv(backer->dpif, upcall, buf);
4319 switch (classify_upcall(upcall)) {
4321 /* Handle it later. */
4326 handle_sflow_upcall(backer, upcall);
4330 case FLOW_SAMPLE_UPCALL:
4331 handle_flow_sample_upcall(backer, upcall);
4336 handle_ipfix_upcall(backer, upcall);
4346 /* Handle deferred MISS_UPCALL processing. */
4347 handle_miss_upcalls(backer, misses, n_misses);
4348 for (i = 0; i < n_misses; i++) {
4349 ofpbuf_uninit(&miss_bufs[i]);
4355 /* Flow expiration. */
4357 static int subfacet_max_idle(const struct ofproto_dpif *);
4358 static void update_stats(struct dpif_backer *);
4359 static void rule_expire(struct rule_dpif *);
4360 static void expire_subfacets(struct ofproto_dpif *, int dp_max_idle);
4362 /* This function is called periodically by run(). Its job is to collect
4363 * updates for the flows that have been installed into the datapath, most
4364 * importantly when they last were used, and then use that information to
4365 * expire flows that have not been used recently.
4367 * Returns the number of milliseconds after which it should be called again. */
4369 expire(struct dpif_backer *backer)
4371 struct ofproto_dpif *ofproto;
4372 int max_idle = INT32_MAX;
4374 /* Periodically clear out the drop keys in an effort to keep them
4375 * relatively few. */
4376 drop_key_clear(backer);
4378 /* Update stats for each flow in the backer. */
4379 update_stats(backer);
4381 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4382 struct rule *rule, *next_rule;
4385 if (ofproto->backer != backer) {
4389 /* Keep track of the max number of flows per ofproto_dpif. */
4390 update_max_subfacet_count(ofproto);
4392 /* Expire subfacets that have been idle too long. */
4393 dp_max_idle = subfacet_max_idle(ofproto);
4394 expire_subfacets(ofproto, dp_max_idle);
4396 max_idle = MIN(max_idle, dp_max_idle);
4398 /* Expire OpenFlow flows whose idle_timeout or hard_timeout
4400 LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
4401 &ofproto->up.expirable) {
4402 rule_expire(rule_dpif_cast(rule));
4405 /* All outstanding data in existing flows has been accounted, so it's a
4406 * good time to do bond rebalancing. */
4407 if (ofproto->has_bonded_bundles) {
4408 struct ofbundle *bundle;
4410 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
4412 bond_rebalance(bundle->bond, &backer->revalidate_set);
4418 return MIN(max_idle, 1000);
4421 /* Updates flow table statistics given that the datapath just reported 'stats'
4422 * as 'subfacet''s statistics. */
4424 update_subfacet_stats(struct subfacet *subfacet,
4425 const struct dpif_flow_stats *stats)
4427 struct facet *facet = subfacet->facet;
4428 struct dpif_flow_stats diff;
4430 diff.tcp_flags = stats->tcp_flags;
4431 diff.used = stats->used;
4433 if (stats->n_packets >= subfacet->dp_packet_count) {
4434 diff.n_packets = stats->n_packets - subfacet->dp_packet_count;
4436 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
4440 if (stats->n_bytes >= subfacet->dp_byte_count) {
4441 diff.n_bytes = stats->n_bytes - subfacet->dp_byte_count;
4443 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
4447 subfacet->dp_packet_count = stats->n_packets;
4448 subfacet->dp_byte_count = stats->n_bytes;
4449 subfacet_update_stats(subfacet, &diff);
4451 if (facet->accounted_bytes < facet->byte_count) {
4453 facet_account(facet);
4454 facet->accounted_bytes = facet->byte_count;
4458 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
4459 * about, or a flow that shouldn't be installed but was anyway. Delete it. */
4461 delete_unexpected_flow(struct ofproto_dpif *ofproto,
4462 const struct nlattr *key, size_t key_len)
4464 if (!VLOG_DROP_WARN(&rl)) {
4468 odp_flow_key_format(key, key_len, &s);
4469 VLOG_WARN("unexpected flow on %s: %s", ofproto->up.name, ds_cstr(&s));
4473 COVERAGE_INC(facet_unexpected);
4474 dpif_flow_del(ofproto->backer->dpif, key, key_len, NULL);
4477 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
4479 * This function also pushes statistics updates to rules which each facet
4480 * resubmits into. Generally these statistics will be accurate. However, if a
4481 * facet changes the rule it resubmits into at some time in between
4482 * update_stats() runs, it is possible that statistics accrued to the
4483 * old rule will be incorrectly attributed to the new rule. This could be
4484 * avoided by calling update_stats() whenever rules are created or
4485 * deleted. However, the performance impact of making so many calls to the
4486 * datapath do not justify the benefit of having perfectly accurate statistics.
4488 * In addition, this function maintains per ofproto flow hit counts. The patch
4489 * port is not treated specially. e.g. A packet ingress from br0 patched into
4490 * br1 will increase the hit count of br0 by 1, however, does not affect
4491 * the hit or miss counts of br1.
4494 update_stats(struct dpif_backer *backer)
4496 const struct dpif_flow_stats *stats;
4497 struct dpif_flow_dump dump;
4498 struct ofproto_dpif *ofproto;
4499 const struct nlattr *key, *mask;
4500 size_t key_len, mask_len;
4502 dpif_flow_dump_start(&dump, backer->dpif);
4503 while (dpif_flow_dump_next(&dump, &key, &key_len,
4504 &mask, &mask_len, NULL, NULL, &stats)) {
4506 struct subfacet *subfacet;
4509 if (ofproto_receive(backer, NULL, key, key_len, &flow, NULL, &ofproto,
4514 ofproto->total_subfacet_count += hmap_count(&ofproto->subfacets);
4515 ofproto->n_update_stats++;
4517 key_hash = odp_flow_key_hash(key, key_len);
4518 subfacet = subfacet_find(ofproto, key, key_len, key_hash);
4519 switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
4521 /* Update ofproto_dpif's hit count. */
4522 if (stats->n_packets > subfacet->dp_packet_count) {
4523 uint64_t delta = stats->n_packets - subfacet->dp_packet_count;
4524 dpif_stats_update_hit_count(ofproto, delta);
4527 update_subfacet_stats(subfacet, stats);
4531 /* Stats are updated per-packet. */
4534 case SF_NOT_INSTALLED:
4536 delete_unexpected_flow(ofproto, key, key_len);
4541 dpif_flow_dump_done(&dump);
4543 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4544 update_moving_averages(ofproto);
4549 /* Calculates and returns the number of milliseconds of idle time after which
4550 * subfacets should expire from the datapath. When a subfacet expires, we fold
4551 * its statistics into its facet, and when a facet's last subfacet expires, we
4552 * fold its statistic into its rule. */
4554 subfacet_max_idle(const struct ofproto_dpif *ofproto)
4557 * Idle time histogram.
4559 * Most of the time a switch has a relatively small number of subfacets.
4560 * When this is the case we might as well keep statistics for all of them
4561 * in userspace and to cache them in the kernel datapath for performance as
4564 * As the number of subfacets increases, the memory required to maintain
4565 * statistics about them in userspace and in the kernel becomes
4566 * significant. However, with a large number of subfacets it is likely
4567 * that only a few of them are "heavy hitters" that consume a large amount
4568 * of bandwidth. At this point, only heavy hitters are worth caching in
4569 * the kernel and maintaining in userspaces; other subfacets we can
4572 * The technique used to compute the idle time is to build a histogram with
4573 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
4574 * that is installed in the kernel gets dropped in the appropriate bucket.
4575 * After the histogram has been built, we compute the cutoff so that only
4576 * the most-recently-used 1% of subfacets (but at least
4577 * ofproto->up.flow_eviction_threshold flows) are kept cached. At least
4578 * the most-recently-used bucket of subfacets is kept, so actually an
4579 * arbitrary number of subfacets can be kept in any given expiration run
4580 * (though the next run will delete most of those unless they receive
4583 * This requires a second pass through the subfacets, in addition to the
4584 * pass made by update_stats(), because the former function never looks at
4585 * uninstallable subfacets.
4587 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
4588 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
4589 int buckets[N_BUCKETS] = { 0 };
4590 int total, subtotal, bucket;
4591 struct subfacet *subfacet;
4595 total = hmap_count(&ofproto->subfacets);
4596 if (total <= ofproto->up.flow_eviction_threshold) {
4597 return N_BUCKETS * BUCKET_WIDTH;
4600 /* Build histogram. */
4602 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
4603 long long int idle = now - subfacet->used;
4604 int bucket = (idle <= 0 ? 0
4605 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
4606 : (unsigned int) idle / BUCKET_WIDTH);
4610 /* Find the first bucket whose flows should be expired. */
4611 subtotal = bucket = 0;
4613 subtotal += buckets[bucket++];
4614 } while (bucket < N_BUCKETS &&
4615 subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100));
4617 if (VLOG_IS_DBG_ENABLED()) {
4621 ds_put_cstr(&s, "keep");
4622 for (i = 0; i < N_BUCKETS; i++) {
4624 ds_put_cstr(&s, ", drop");
4627 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
4630 VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s));
4634 return bucket * BUCKET_WIDTH;
4638 expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle)
4640 /* Cutoff time for most flows. */
4641 long long int normal_cutoff = time_msec() - dp_max_idle;
4643 /* We really want to keep flows for special protocols around, so use a more
4644 * conservative cutoff. */
4645 long long int special_cutoff = time_msec() - 10000;
4647 struct subfacet *subfacet, *next_subfacet;
4648 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
4652 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
4653 &ofproto->subfacets) {
4654 long long int cutoff;
4656 cutoff = (subfacet->facet->xout.slow & (SLOW_CFM | SLOW_LACP
4660 if (subfacet->used < cutoff) {
4661 if (subfacet->path != SF_NOT_INSTALLED) {
4662 batch[n_batch++] = subfacet;
4663 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
4664 subfacet_destroy_batch(ofproto, batch, n_batch);
4668 subfacet_destroy(subfacet);
4674 subfacet_destroy_batch(ofproto, batch, n_batch);
4678 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
4679 * then delete it entirely. */
4681 rule_expire(struct rule_dpif *rule)
4683 struct facet *facet, *next_facet;
4687 if (rule->up.pending) {
4688 /* We'll have to expire it later. */
4692 /* Has 'rule' expired? */
4694 if (rule->up.hard_timeout
4695 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
4696 reason = OFPRR_HARD_TIMEOUT;
4697 } else if (rule->up.idle_timeout
4698 && now > rule->up.used + rule->up.idle_timeout * 1000) {
4699 reason = OFPRR_IDLE_TIMEOUT;
4704 COVERAGE_INC(ofproto_dpif_expired);
4706 /* Update stats. (This is a no-op if the rule expired due to an idle
4707 * timeout, because that only happens when the rule has no facets left.) */
4708 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4709 facet_remove(facet);
4712 /* Get rid of the rule. */
4713 ofproto_rule_expire(&rule->up, reason);
4718 /* Creates and returns a new facet based on 'miss'.
4720 * The caller must already have determined that no facet with an identical
4721 * 'miss->flow' exists in 'miss->ofproto'.
4723 * 'rule' and 'xout' must have been created based on 'miss'.
4725 * 'facet'' statistics are initialized based on 'stats'.
4727 * The facet will initially have no subfacets. The caller should create (at
4728 * least) one subfacet with subfacet_create(). */
4729 static struct facet *
4730 facet_create(const struct flow_miss *miss, struct rule_dpif *rule,
4731 struct xlate_out *xout, struct dpif_flow_stats *stats)
4733 struct ofproto_dpif *ofproto = miss->ofproto;
4734 struct facet *facet;
4737 facet = xzalloc(sizeof *facet);
4738 facet->packet_count = facet->prev_packet_count = stats->n_packets;
4739 facet->byte_count = facet->prev_byte_count = stats->n_bytes;
4740 facet->tcp_flags = stats->tcp_flags;
4741 facet->used = stats->used;
4742 facet->flow = miss->flow;
4743 facet->initial_vals = miss->initial_vals;
4744 facet->learn_rl = time_msec() + 500;
4747 list_push_back(&facet->rule->facets, &facet->list_node);
4748 list_init(&facet->subfacets);
4749 netflow_flow_init(&facet->nf_flow);
4750 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
4752 xlate_out_copy(&facet->xout, xout);
4754 match_init(&match, &facet->flow, &facet->xout.wc);
4755 cls_rule_init(&facet->cr, &match, OFP_DEFAULT_PRIORITY);
4756 classifier_insert(&ofproto->facets, &facet->cr);
4758 facet->nf_flow.output_iface = facet->xout.nf_output_iface;
4764 facet_free(struct facet *facet)
4767 xlate_out_uninit(&facet->xout);
4772 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
4773 * 'packet', which arrived on 'in_port'. */
4775 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
4776 const struct nlattr *odp_actions, size_t actions_len,
4777 struct ofpbuf *packet)
4779 struct odputil_keybuf keybuf;
4783 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4784 odp_flow_key_from_flow(&key, flow,
4785 ofp_port_to_odp_port(ofproto, flow->in_port));
4787 error = dpif_execute(ofproto->backer->dpif, key.data, key.size,
4788 odp_actions, actions_len, packet);
4792 /* Remove 'facet' from its ofproto and free up the associated memory:
4794 * - If 'facet' was installed in the datapath, uninstalls it and updates its
4795 * rule's statistics, via subfacet_uninstall().
4797 * - Removes 'facet' from its rule and from ofproto->facets.
4800 facet_remove(struct facet *facet)
4802 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4803 struct subfacet *subfacet, *next_subfacet;
4805 ovs_assert(!list_is_empty(&facet->subfacets));
4807 /* First uninstall all of the subfacets to get final statistics. */
4808 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4809 subfacet_uninstall(subfacet);
4812 /* Flush the final stats to the rule.
4814 * This might require us to have at least one subfacet around so that we
4815 * can use its actions for accounting in facet_account(), which is why we
4816 * have uninstalled but not yet destroyed the subfacets. */
4817 facet_flush_stats(facet);
4819 /* Now we're really all done so destroy everything. */
4820 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
4821 &facet->subfacets) {
4822 subfacet_destroy__(subfacet);
4824 classifier_remove(&ofproto->facets, &facet->cr);
4825 cls_rule_destroy(&facet->cr);
4826 list_remove(&facet->list_node);
4830 /* Feed information from 'facet' back into the learning table to keep it in
4831 * sync with what is actually flowing through the datapath. */
4833 facet_learn(struct facet *facet)
4835 long long int now = time_msec();
4837 if (!facet->xout.has_fin_timeout && now < facet->learn_rl) {
4841 facet->learn_rl = now + 500;
4843 if (!facet->xout.has_learn
4844 && !facet->xout.has_normal
4845 && (!facet->xout.has_fin_timeout
4846 || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
4850 facet_push_stats(facet, true);
4854 facet_account(struct facet *facet)
4856 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4857 const struct nlattr *a;
4862 if (!facet->xout.has_normal || !ofproto->has_bonded_bundles) {
4865 n_bytes = facet->byte_count - facet->accounted_bytes;
4867 /* This loop feeds byte counters to bond_account() for rebalancing to use
4868 * as a basis. We also need to track the actual VLAN on which the packet
4869 * is going to be sent to ensure that it matches the one passed to
4870 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
4873 * We use the actions from an arbitrary subfacet because they should all
4874 * be equally valid for our purpose. */
4875 vlan_tci = facet->flow.vlan_tci;
4876 NL_ATTR_FOR_EACH_UNSAFE (a, left, facet->xout.odp_actions.data,
4877 facet->xout.odp_actions.size) {
4878 const struct ovs_action_push_vlan *vlan;
4879 struct ofport_dpif *port;
4881 switch (nl_attr_type(a)) {
4882 case OVS_ACTION_ATTR_OUTPUT:
4883 port = get_odp_port(ofproto, nl_attr_get_u32(a));
4884 if (port && port->bundle && port->bundle->bond) {
4885 bond_account(port->bundle->bond, &facet->flow,
4886 vlan_tci_to_vid(vlan_tci), n_bytes);
4890 case OVS_ACTION_ATTR_POP_VLAN:
4891 vlan_tci = htons(0);
4894 case OVS_ACTION_ATTR_PUSH_VLAN:
4895 vlan = nl_attr_get(a);
4896 vlan_tci = vlan->vlan_tci;
4902 /* Returns true if the only action for 'facet' is to send to the controller.
4903 * (We don't report NetFlow expiration messages for such facets because they
4904 * are just part of the control logic for the network, not real traffic). */
4906 facet_is_controller_flow(struct facet *facet)
4909 const struct rule *rule = &facet->rule->up;
4910 const struct ofpact *ofpacts = rule->ofpacts;
4911 size_t ofpacts_len = rule->ofpacts_len;
4913 if (ofpacts_len > 0 &&
4914 ofpacts->type == OFPACT_CONTROLLER &&
4915 ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len)) {
4922 /* Folds all of 'facet''s statistics into its rule. Also updates the
4923 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
4924 * 'facet''s statistics in the datapath should have been zeroed and folded into
4925 * its packet and byte counts before this function is called. */
4927 facet_flush_stats(struct facet *facet)
4929 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4930 struct subfacet *subfacet;
4932 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4933 ovs_assert(!subfacet->dp_byte_count);
4934 ovs_assert(!subfacet->dp_packet_count);
4937 facet_push_stats(facet, false);
4938 if (facet->accounted_bytes < facet->byte_count) {
4939 facet_account(facet);
4940 facet->accounted_bytes = facet->byte_count;
4943 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
4944 struct ofexpired expired;
4945 expired.flow = facet->flow;
4946 expired.packet_count = facet->packet_count;
4947 expired.byte_count = facet->byte_count;
4948 expired.used = facet->used;
4949 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4952 /* Reset counters to prevent double counting if 'facet' ever gets
4954 facet_reset_counters(facet);
4956 netflow_flow_clear(&facet->nf_flow);
4957 facet->tcp_flags = 0;
4960 /* Searches 'ofproto''s table of facets for one which would be responsible for
4961 * 'flow'. Returns it if found, otherwise a null pointer.
4963 * The returned facet might need revalidation; use facet_lookup_valid()
4964 * instead if that is important. */
4965 static struct facet *
4966 facet_find(struct ofproto_dpif *ofproto, const struct flow *flow)
4968 struct cls_rule *cr = classifier_lookup(&ofproto->facets, flow, NULL);
4969 return cr ? CONTAINER_OF(cr, struct facet, cr) : NULL;
4972 /* Searches 'ofproto''s table of facets for one capable that covers
4973 * 'flow'. Returns it if found, otherwise a null pointer.
4975 * The returned facet is guaranteed to be valid. */
4976 static struct facet *
4977 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow)
4979 struct facet *facet;
4981 facet = facet_find(ofproto, flow);
4983 && (ofproto->backer->need_revalidate
4984 || tag_set_intersects(&ofproto->backer->revalidate_set,
4986 && !facet_revalidate(facet)) {
4994 facet_check_consistency(struct facet *facet)
4996 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
4998 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5000 struct xlate_out xout;
5001 struct xlate_in xin;
5003 struct rule_dpif *rule;
5006 /* Check the rule for consistency. */
5007 rule = rule_dpif_lookup(ofproto, &facet->flow, NULL);
5008 if (rule != facet->rule) {
5009 if (!VLOG_DROP_WARN(&rl)) {
5010 struct ds s = DS_EMPTY_INITIALIZER;
5012 flow_format(&s, &facet->flow);
5013 ds_put_format(&s, ": facet associated with wrong rule (was "
5014 "table=%"PRIu8",", facet->rule->up.table_id);
5015 cls_rule_format(&facet->rule->up.cr, &s);
5016 ds_put_format(&s, ") (should have been table=%"PRIu8",",
5018 cls_rule_format(&rule->up.cr, &s);
5019 ds_put_cstr(&s, ")\n");
5026 /* Check the datapath actions for consistency. */
5027 xlate_in_init(&xin, ofproto, &facet->flow, &facet->initial_vals, rule,
5029 xlate_actions(&xin, &xout);
5031 ok = ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)
5032 && facet->xout.slow == xout.slow;
5033 if (!ok && !VLOG_DROP_WARN(&rl)) {
5034 struct ds s = DS_EMPTY_INITIALIZER;
5036 flow_format(&s, &facet->flow);
5037 ds_put_cstr(&s, ": inconsistency in facet");
5039 if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
5040 ds_put_cstr(&s, " (actions were: ");
5041 format_odp_actions(&s, facet->xout.odp_actions.data,
5042 facet->xout.odp_actions.size);
5043 ds_put_cstr(&s, ") (correct actions: ");
5044 format_odp_actions(&s, xout.odp_actions.data,
5045 xout.odp_actions.size);
5046 ds_put_cstr(&s, ")");
5049 if (facet->xout.slow != xout.slow) {
5050 ds_put_format(&s, " slow path incorrect. should be %d", xout.slow);
5055 xlate_out_uninit(&xout);
5060 /* Re-searches the classifier for 'facet':
5062 * - If the rule found is different from 'facet''s current rule, moves
5063 * 'facet' to the new rule and recompiles its actions.
5065 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
5066 * where it is and recompiles its actions anyway.
5068 * - If any of 'facet''s subfacets correspond to a new flow according to
5069 * ofproto_receive(), 'facet' is removed.
5071 * Returns true if 'facet' is still valid. False if 'facet' was removed. */
5073 facet_revalidate(struct facet *facet)
5075 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5076 struct rule_dpif *new_rule;
5077 struct subfacet *subfacet;
5078 struct flow_wildcards wc;
5079 struct xlate_out xout;
5080 struct xlate_in xin;
5082 COVERAGE_INC(facet_revalidate);
5084 /* Check that child subfacets still correspond to this facet. Tunnel
5085 * configuration changes could cause a subfacet's OpenFlow in_port to
5087 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
5088 struct ofproto_dpif *recv_ofproto;
5089 struct flow recv_flow;
5092 error = ofproto_receive(ofproto->backer, NULL, subfacet->key,
5093 subfacet->key_len, &recv_flow, NULL,
5094 &recv_ofproto, NULL, NULL);
5096 || recv_ofproto != ofproto
5097 || facet != facet_find(ofproto, &recv_flow)) {
5098 facet_remove(facet);
5103 flow_wildcards_init_catchall(&wc);
5104 new_rule = rule_dpif_lookup(ofproto, &facet->flow, &wc);
5106 /* Calculate new datapath actions.
5108 * We do not modify any 'facet' state yet, because we might need to, e.g.,
5109 * emit a NetFlow expiration and, if so, we need to have the old state
5110 * around to properly compose it. */
5111 xlate_in_init(&xin, ofproto, &facet->flow, &facet->initial_vals, new_rule,
5113 xlate_actions(&xin, &xout);
5114 flow_wildcards_or(&xout.wc, &xout.wc, &wc);
5116 /* A facet's slow path reason should only change under dramatic
5117 * circumstances. Rather than try to update everything, it's simpler to
5118 * remove the facet and start over.
5120 * More importantly, if a facet's wildcards change, it will be relatively
5121 * difficult to figure out if its subfacets still belong to it, and if not
5122 * which facet they may belong to. Again, to avoid the complexity, we
5123 * simply give up instead. */
5124 if (facet->xout.slow != xout.slow
5125 || memcmp(&facet->xout.wc, &xout.wc, sizeof xout.wc)) {
5126 facet_remove(facet);
5127 xlate_out_uninit(&xout);
5131 if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
5132 LIST_FOR_EACH(subfacet, list_node, &facet->subfacets) {
5133 if (subfacet->path == SF_FAST_PATH) {
5134 struct dpif_flow_stats stats;
5136 subfacet_install(subfacet, &xout.odp_actions, &stats);
5137 subfacet_update_stats(subfacet, &stats);
5141 facet_flush_stats(facet);
5143 ofpbuf_clear(&facet->xout.odp_actions);
5144 ofpbuf_put(&facet->xout.odp_actions, xout.odp_actions.data,
5145 xout.odp_actions.size);
5148 /* Update 'facet' now that we've taken care of all the old state. */
5149 facet->xout.tags = xout.tags;
5150 facet->xout.slow = xout.slow;
5151 facet->xout.has_learn = xout.has_learn;
5152 facet->xout.has_normal = xout.has_normal;
5153 facet->xout.has_fin_timeout = xout.has_fin_timeout;
5154 facet->xout.nf_output_iface = xout.nf_output_iface;
5155 facet->xout.mirrors = xout.mirrors;
5156 facet->nf_flow.output_iface = facet->xout.nf_output_iface;
5158 if (facet->rule != new_rule) {
5159 COVERAGE_INC(facet_changed_rule);
5160 list_remove(&facet->list_node);
5161 list_push_back(&new_rule->facets, &facet->list_node);
5162 facet->rule = new_rule;
5163 facet->used = new_rule->up.created;
5164 facet->prev_used = facet->used;
5167 xlate_out_uninit(&xout);
5172 facet_reset_counters(struct facet *facet)
5174 facet->packet_count = 0;
5175 facet->byte_count = 0;
5176 facet->prev_packet_count = 0;
5177 facet->prev_byte_count = 0;
5178 facet->accounted_bytes = 0;
5182 facet_push_stats(struct facet *facet, bool may_learn)
5184 struct dpif_flow_stats stats;
5186 ovs_assert(facet->packet_count >= facet->prev_packet_count);
5187 ovs_assert(facet->byte_count >= facet->prev_byte_count);
5188 ovs_assert(facet->used >= facet->prev_used);
5190 stats.n_packets = facet->packet_count - facet->prev_packet_count;
5191 stats.n_bytes = facet->byte_count - facet->prev_byte_count;
5192 stats.used = facet->used;
5193 stats.tcp_flags = facet->tcp_flags;
5195 if (may_learn || stats.n_packets || facet->used > facet->prev_used) {
5196 struct ofproto_dpif *ofproto =
5197 ofproto_dpif_cast(facet->rule->up.ofproto);
5199 struct ofport_dpif *in_port;
5200 struct xlate_in xin;
5202 facet->prev_packet_count = facet->packet_count;
5203 facet->prev_byte_count = facet->byte_count;
5204 facet->prev_used = facet->used;
5206 in_port = get_ofp_port(ofproto, facet->flow.in_port);
5207 if (in_port && in_port->tnl_port) {
5208 netdev_vport_inc_rx(in_port->up.netdev, &stats);
5211 rule_credit_stats(facet->rule, &stats);
5212 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow,
5214 netflow_flow_update_flags(&facet->nf_flow, facet->tcp_flags);
5215 update_mirror_stats(ofproto, facet->xout.mirrors, stats.n_packets,
5218 xlate_in_init(&xin, ofproto, &facet->flow, &facet->initial_vals,
5219 facet->rule, stats.tcp_flags, NULL);
5220 xin.resubmit_stats = &stats;
5221 xin.may_learn = may_learn;
5222 xlate_actions_for_side_effects(&xin);
5227 push_all_stats__(bool run_fast)
5229 static long long int rl = LLONG_MIN;
5230 struct ofproto_dpif *ofproto;
5232 if (time_msec() < rl) {
5236 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5237 struct cls_cursor cursor;
5238 struct facet *facet;
5240 cls_cursor_init(&cursor, &ofproto->facets, NULL);
5241 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5242 facet_push_stats(facet, false);
5249 rl = time_msec() + 100;
5253 push_all_stats(void)
5255 push_all_stats__(true);
5259 rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats)
5261 rule->packet_count += stats->n_packets;
5262 rule->byte_count += stats->n_bytes;
5263 ofproto_rule_update_used(&rule->up, stats->used);
5268 static struct subfacet *
5269 subfacet_find(struct ofproto_dpif *ofproto,
5270 const struct nlattr *key, size_t key_len, uint32_t key_hash)
5272 struct subfacet *subfacet;
5274 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
5275 &ofproto->subfacets) {
5276 if (subfacet->key_len == key_len
5277 && !memcmp(key, subfacet->key, key_len)) {
5285 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
5286 * 'key_fitness', 'key', and 'key_len' members in 'miss'. Returns the
5287 * existing subfacet if there is one, otherwise creates and returns a
5289 static struct subfacet *
5290 subfacet_create(struct facet *facet, struct flow_miss *miss,
5293 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5294 enum odp_key_fitness key_fitness = miss->key_fitness;
5295 const struct nlattr *key = miss->key;
5296 size_t key_len = miss->key_len;
5298 struct subfacet *subfacet;
5300 key_hash = odp_flow_key_hash(key, key_len);
5302 if (list_is_empty(&facet->subfacets)) {
5303 subfacet = &facet->one_subfacet;
5305 subfacet = subfacet_find(ofproto, key, key_len, key_hash);
5307 if (subfacet->facet == facet) {
5311 /* This shouldn't happen. */
5312 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
5313 subfacet_destroy(subfacet);
5316 subfacet = xmalloc(sizeof *subfacet);
5319 hmap_insert(&ofproto->subfacets, &subfacet->hmap_node, key_hash);
5320 list_push_back(&facet->subfacets, &subfacet->list_node);
5321 subfacet->facet = facet;
5322 subfacet->key_fitness = key_fitness;
5323 subfacet->key = xmemdup(key, key_len);
5324 subfacet->key_len = key_len;
5325 subfacet->used = now;
5326 subfacet->created = now;
5327 subfacet->dp_packet_count = 0;
5328 subfacet->dp_byte_count = 0;
5329 subfacet->path = SF_NOT_INSTALLED;
5330 subfacet->odp_in_port = miss->odp_in_port;
5332 ofproto->subfacet_add_count++;
5336 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
5337 * its facet within 'ofproto', and frees it. */
5339 subfacet_destroy__(struct subfacet *subfacet)
5341 struct facet *facet = subfacet->facet;
5342 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5344 /* Update ofproto stats before uninstall the subfacet. */
5345 ofproto->subfacet_del_count++;
5346 ofproto->total_subfacet_life_span += (time_msec() - subfacet->created);
5348 subfacet_uninstall(subfacet);
5349 hmap_remove(&ofproto->subfacets, &subfacet->hmap_node);
5350 list_remove(&subfacet->list_node);
5351 free(subfacet->key);
5352 if (subfacet != &facet->one_subfacet) {
5357 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
5358 * last remaining subfacet in its facet destroys the facet too. */
5360 subfacet_destroy(struct subfacet *subfacet)
5362 struct facet *facet = subfacet->facet;
5364 if (list_is_singleton(&facet->subfacets)) {
5365 /* facet_remove() needs at least one subfacet (it will remove it). */
5366 facet_remove(facet);
5368 subfacet_destroy__(subfacet);
5373 subfacet_destroy_batch(struct ofproto_dpif *ofproto,
5374 struct subfacet **subfacets, int n)
5376 struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH];
5377 struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH];
5378 struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH];
5381 for (i = 0; i < n; i++) {
5382 ops[i].type = DPIF_OP_FLOW_DEL;
5383 ops[i].u.flow_del.key = subfacets[i]->key;
5384 ops[i].u.flow_del.key_len = subfacets[i]->key_len;
5385 ops[i].u.flow_del.stats = &stats[i];
5389 dpif_operate(ofproto->backer->dpif, opsp, n);
5390 for (i = 0; i < n; i++) {
5391 subfacet_reset_dp_stats(subfacets[i], &stats[i]);
5392 subfacets[i]->path = SF_NOT_INSTALLED;
5393 subfacet_destroy(subfacets[i]);
5398 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
5399 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
5400 * in the datapath will be zeroed and 'stats' will be updated with traffic new
5401 * since 'subfacet' was last updated.
5403 * Returns 0 if successful, otherwise a positive errno value. */
5405 subfacet_install(struct subfacet *subfacet, const struct ofpbuf *odp_actions,
5406 struct dpif_flow_stats *stats)
5408 struct facet *facet = subfacet->facet;
5409 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5410 enum subfacet_path path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
5411 const struct nlattr *actions = odp_actions->data;
5412 size_t actions_len = odp_actions->size;
5413 struct odputil_keybuf maskbuf;
5416 uint64_t slow_path_stub[128 / 8];
5417 enum dpif_flow_put_flags flags;
5420 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
5422 flags |= DPIF_FP_ZERO_STATS;
5425 if (path == SF_SLOW_PATH) {
5426 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
5427 slow_path_stub, sizeof slow_path_stub,
5428 &actions, &actions_len);
5431 ofpbuf_use_stack(&mask, &maskbuf, sizeof maskbuf);
5432 if (enable_megaflows) {
5433 odp_flow_key_from_mask(&mask, &facet->xout.wc.masks,
5434 &facet->flow, UINT32_MAX);
5437 ret = dpif_flow_put(ofproto->backer->dpif, flags, subfacet->key,
5438 subfacet->key_len, mask.data, mask.size,
5439 actions, actions_len, stats);
5442 subfacet_reset_dp_stats(subfacet, stats);
5446 COVERAGE_INC(subfacet_install_fail);
5448 subfacet->path = path;
5453 /* If 'subfacet' is installed in the datapath, uninstalls it. */
5455 subfacet_uninstall(struct subfacet *subfacet)
5457 if (subfacet->path != SF_NOT_INSTALLED) {
5458 struct rule_dpif *rule = subfacet->facet->rule;
5459 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5460 struct dpif_flow_stats stats;
5463 error = dpif_flow_del(ofproto->backer->dpif, subfacet->key,
5464 subfacet->key_len, &stats);
5465 subfacet_reset_dp_stats(subfacet, &stats);
5467 subfacet_update_stats(subfacet, &stats);
5469 subfacet->path = SF_NOT_INSTALLED;
5471 ovs_assert(subfacet->dp_packet_count == 0);
5472 ovs_assert(subfacet->dp_byte_count == 0);
5476 /* Resets 'subfacet''s datapath statistics counters. This should be called
5477 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
5478 * non-null, it should contain the statistics returned by dpif when 'subfacet'
5479 * was reset in the datapath. 'stats' will be modified to include only
5480 * statistics new since 'subfacet' was last updated. */
5482 subfacet_reset_dp_stats(struct subfacet *subfacet,
5483 struct dpif_flow_stats *stats)
5486 && subfacet->dp_packet_count <= stats->n_packets
5487 && subfacet->dp_byte_count <= stats->n_bytes) {
5488 stats->n_packets -= subfacet->dp_packet_count;
5489 stats->n_bytes -= subfacet->dp_byte_count;
5492 subfacet->dp_packet_count = 0;
5493 subfacet->dp_byte_count = 0;
5496 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
5498 * Because of the meaning of a subfacet's counters, it only makes sense to do
5499 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
5500 * represents a packet that was sent by hand or if it represents statistics
5501 * that have been cleared out of the datapath. */
5503 subfacet_update_stats(struct subfacet *subfacet,
5504 const struct dpif_flow_stats *stats)
5506 if (stats->n_packets || stats->used > subfacet->used) {
5507 struct facet *facet = subfacet->facet;
5509 subfacet->used = MAX(subfacet->used, stats->used);
5510 facet->used = MAX(facet->used, stats->used);
5511 facet->packet_count += stats->n_packets;
5512 facet->byte_count += stats->n_bytes;
5513 facet->tcp_flags |= stats->tcp_flags;
5519 /* Lookup 'flow' in 'ofproto''s classifier. If 'wc' is non-null, sets
5520 * the fields that were relevant as part of the lookup. */
5521 static struct rule_dpif *
5522 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow,
5523 struct flow_wildcards *wc)
5525 struct rule_dpif *rule;
5527 rule = rule_dpif_lookup__(ofproto, flow, wc, 0);
5532 return rule_dpif_miss_rule(ofproto, flow);
5535 static struct rule_dpif *
5536 rule_dpif_lookup__(struct ofproto_dpif *ofproto, const struct flow *flow,
5537 struct flow_wildcards *wc, uint8_t table_id)
5539 struct cls_rule *cls_rule;
5540 struct classifier *cls;
5543 if (table_id >= N_TABLES) {
5548 memset(&wc->masks.dl_type, 0xff, sizeof wc->masks.dl_type);
5549 wc->masks.nw_frag |= FLOW_NW_FRAG_MASK;
5552 cls = &ofproto->up.tables[table_id].cls;
5553 frag = (flow->nw_frag & FLOW_NW_FRAG_ANY) != 0;
5554 if (frag && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
5555 /* We must pretend that transport ports are unavailable. */
5556 struct flow ofpc_normal_flow = *flow;
5557 ofpc_normal_flow.tp_src = htons(0);
5558 ofpc_normal_flow.tp_dst = htons(0);
5559 cls_rule = classifier_lookup(cls, &ofpc_normal_flow, wc);
5560 } else if (frag && ofproto->up.frag_handling == OFPC_FRAG_DROP) {
5561 cls_rule = &ofproto->drop_frags_rule->up.cr;
5563 flow_wildcards_init_exact(wc);
5566 cls_rule = classifier_lookup(cls, flow, wc);
5568 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
5571 static struct rule_dpif *
5572 rule_dpif_miss_rule(struct ofproto_dpif *ofproto, const struct flow *flow)
5574 struct ofport_dpif *port;
5576 port = get_ofp_port(ofproto, flow->in_port);
5578 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, flow->in_port);
5579 return ofproto->miss_rule;
5582 if (port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN) {
5583 return ofproto->no_packet_in_rule;
5585 return ofproto->miss_rule;
5589 complete_operation(struct rule_dpif *rule)
5591 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5593 rule_invalidate(rule);
5595 struct dpif_completion *c = xmalloc(sizeof *c);
5596 c->op = rule->up.pending;
5597 list_push_back(&ofproto->completions, &c->list_node);
5599 ofoperation_complete(rule->up.pending, 0);
5603 static struct rule *
5606 struct rule_dpif *rule = xmalloc(sizeof *rule);
5611 rule_dealloc(struct rule *rule_)
5613 struct rule_dpif *rule = rule_dpif_cast(rule_);
5618 rule_construct(struct rule *rule_)
5620 struct rule_dpif *rule = rule_dpif_cast(rule_);
5621 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5622 struct rule_dpif *victim;
5625 rule->packet_count = 0;
5626 rule->byte_count = 0;
5628 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
5629 if (victim && !list_is_empty(&victim->facets)) {
5630 struct facet *facet;
5632 rule->facets = victim->facets;
5633 list_moved(&rule->facets);
5634 LIST_FOR_EACH (facet, list_node, &rule->facets) {
5635 /* XXX: We're only clearing our local counters here. It's possible
5636 * that quite a few packets are unaccounted for in the datapath
5637 * statistics. These will be accounted to the new rule instead of
5638 * cleared as required. This could be fixed by clearing out the
5639 * datapath statistics for this facet, but currently it doesn't
5641 facet_reset_counters(facet);
5645 /* Must avoid list_moved() in this case. */
5646 list_init(&rule->facets);
5649 table_id = rule->up.table_id;
5651 rule->tag = victim->tag;
5652 } else if (table_id == 0) {
5657 miniflow_expand(&rule->up.cr.match.flow, &flow);
5658 rule->tag = rule_calculate_tag(&flow, &rule->up.cr.match.mask,
5659 ofproto->tables[table_id].basis);
5662 complete_operation(rule);
5667 rule_destruct(struct rule *rule_)
5669 struct rule_dpif *rule = rule_dpif_cast(rule_);
5670 struct facet *facet, *next_facet;
5672 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
5673 facet_revalidate(facet);
5676 complete_operation(rule);
5680 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
5682 struct rule_dpif *rule = rule_dpif_cast(rule_);
5684 /* push_all_stats() can handle flow misses which, when using the learn
5685 * action, can cause rules to be added and deleted. This can corrupt our
5686 * caller's datastructures which assume that rule_get_stats() doesn't have
5687 * an impact on the flow table. To be safe, we disable miss handling. */
5688 push_all_stats__(false);
5690 /* Start from historical data for 'rule' itself that are no longer tracked
5691 * in facets. This counts, for example, facets that have expired. */
5692 *packets = rule->packet_count;
5693 *bytes = rule->byte_count;
5697 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
5698 struct ofpbuf *packet)
5700 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5701 struct initial_vals initial_vals;
5702 struct dpif_flow_stats stats;
5703 struct xlate_out xout;
5704 struct xlate_in xin;
5706 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
5707 rule_credit_stats(rule, &stats);
5709 initial_vals.vlan_tci = flow->vlan_tci;
5710 initial_vals.tunnel_ip_tos = flow->tunnel.ip_tos;
5711 xlate_in_init(&xin, ofproto, flow, &initial_vals, rule, stats.tcp_flags,
5713 xin.resubmit_stats = &stats;
5714 xlate_actions(&xin, &xout);
5716 execute_odp_actions(ofproto, flow, xout.odp_actions.data,
5717 xout.odp_actions.size, packet);
5719 xlate_out_uninit(&xout);
5723 rule_execute(struct rule *rule, const struct flow *flow,
5724 struct ofpbuf *packet)
5726 rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
5727 ofpbuf_delete(packet);
5732 rule_modify_actions(struct rule *rule_)
5734 struct rule_dpif *rule = rule_dpif_cast(rule_);
5736 complete_operation(rule);
5739 /* Sends 'packet' out 'ofport'.
5740 * May modify 'packet'.
5741 * Returns 0 if successful, otherwise a positive errno value. */
5743 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
5745 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
5746 uint64_t odp_actions_stub[1024 / 8];
5747 struct ofpbuf key, odp_actions;
5748 struct odputil_keybuf keybuf;
5753 flow_extract(packet, 0, 0, NULL, OFPP_LOCAL, &flow);
5754 if (netdev_vport_is_patch(ofport->up.netdev)) {
5755 struct ofproto_dpif *peer_ofproto;
5756 struct dpif_flow_stats stats;
5757 struct ofport_dpif *peer;
5758 struct rule_dpif *rule;
5760 peer = ofport_get_peer(ofport);
5765 dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
5766 netdev_vport_inc_tx(ofport->up.netdev, &stats);
5767 netdev_vport_inc_rx(peer->up.netdev, &stats);
5769 flow.in_port = peer->up.ofp_port;
5770 peer_ofproto = ofproto_dpif_cast(peer->up.ofproto);
5771 rule = rule_dpif_lookup(peer_ofproto, &flow, NULL);
5772 rule_dpif_execute(rule, &flow, packet);
5777 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5779 if (ofport->tnl_port) {
5780 struct flow_wildcards wc;
5781 struct dpif_flow_stats stats;
5783 odp_port = tnl_port_send(ofport->tnl_port, &flow, &wc);
5784 if (odp_port == OVSP_NONE) {
5788 dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
5789 netdev_vport_inc_tx(ofport->up.netdev, &stats);
5790 odp_put_tunnel_action(&flow.tunnel, &odp_actions);
5791 odp_put_skb_mark_action(flow.skb_mark, &odp_actions);
5793 odp_port = vsp_realdev_to_vlandev(ofproto, ofport->odp_port,
5795 if (odp_port != ofport->odp_port) {
5796 eth_pop_vlan(packet);
5797 flow.vlan_tci = htons(0);
5801 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5802 odp_flow_key_from_flow(&key, &flow,
5803 ofp_port_to_odp_port(ofproto, flow.in_port));
5805 compose_sflow_action(ofproto, &odp_actions, &flow, odp_port);
5806 compose_ipfix_action(ofproto, &odp_actions, &flow);
5808 nl_msg_put_u32(&odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port);
5809 error = dpif_execute(ofproto->backer->dpif,
5811 odp_actions.data, odp_actions.size,
5813 ofpbuf_uninit(&odp_actions);
5816 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)",
5817 ofproto->up.name, odp_port, strerror(error));
5819 ofproto_update_local_port_stats(ofport->up.ofproto, packet->size, 0);
5823 /* OpenFlow to datapath action translation. */
5825 static bool may_receive(const struct ofport_dpif *, struct xlate_ctx *);
5826 static void do_xlate_actions(const struct ofpact *, size_t ofpacts_len,
5827 struct xlate_ctx *);
5828 static void xlate_normal(struct xlate_ctx *);
5830 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
5831 * The action will state 'slow' as the reason that the action is in the slow
5832 * path. (This is purely informational: it allows a human viewing "ovs-dpctl
5833 * dump-flows" output to see why a flow is in the slow path.)
5835 * The 'stub_size' bytes in 'stub' will be used to store the action.
5836 * 'stub_size' must be large enough for the action.
5838 * The action and its size will be stored in '*actionsp' and '*actions_lenp',
5841 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
5842 enum slow_path_reason slow,
5843 uint64_t *stub, size_t stub_size,
5844 const struct nlattr **actionsp, size_t *actions_lenp)
5846 union user_action_cookie cookie;
5849 cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
5850 cookie.slow_path.unused = 0;
5851 cookie.slow_path.reason = slow;
5853 ofpbuf_use_stack(&buf, stub, stub_size);
5854 if (slow & (SLOW_CFM | SLOW_LACP | SLOW_STP)) {
5855 uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif, UINT32_MAX);
5856 odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
5858 put_userspace_action(ofproto, &buf, flow, &cookie,
5859 sizeof cookie.slow_path);
5861 *actionsp = buf.data;
5862 *actions_lenp = buf.size;
5866 put_userspace_action(const struct ofproto_dpif *ofproto,
5867 struct ofpbuf *odp_actions,
5868 const struct flow *flow,
5869 const union user_action_cookie *cookie,
5870 const size_t cookie_size)
5874 pid = dpif_port_get_pid(ofproto->backer->dpif,
5875 ofp_port_to_odp_port(ofproto, flow->in_port));
5877 return odp_put_userspace_action(pid, cookie, cookie_size, odp_actions);
5880 /* Compose SAMPLE action for sFlow or IPFIX. The given probability is
5881 * the number of packets out of UINT32_MAX to sample. The given
5882 * cookie is passed back in the callback for each sampled packet.
5885 compose_sample_action(const struct ofproto_dpif *ofproto,
5886 struct ofpbuf *odp_actions,
5887 const struct flow *flow,
5888 const uint32_t probability,
5889 const union user_action_cookie *cookie,
5890 const size_t cookie_size)
5892 size_t sample_offset, actions_offset;
5895 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
5897 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
5899 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
5900 cookie_offset = put_userspace_action(ofproto, odp_actions, flow, cookie,
5903 nl_msg_end_nested(odp_actions, actions_offset);
5904 nl_msg_end_nested(odp_actions, sample_offset);
5905 return cookie_offset;
5909 compose_sflow_cookie(const struct ofproto_dpif *ofproto,
5910 ovs_be16 vlan_tci, uint32_t odp_port,
5911 unsigned int n_outputs, union user_action_cookie *cookie)
5915 cookie->type = USER_ACTION_COOKIE_SFLOW;
5916 cookie->sflow.vlan_tci = vlan_tci;
5918 /* See http://www.sflow.org/sflow_version_5.txt (search for "Input/output
5919 * port information") for the interpretation of cookie->output. */
5920 switch (n_outputs) {
5922 /* 0x40000000 | 256 means "packet dropped for unknown reason". */
5923 cookie->sflow.output = 0x40000000 | 256;
5927 ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port);
5929 cookie->sflow.output = ifindex;
5934 /* 0x80000000 means "multiple output ports. */
5935 cookie->sflow.output = 0x80000000 | n_outputs;
5940 /* Compose SAMPLE action for sFlow bridge sampling. */
5942 compose_sflow_action(const struct ofproto_dpif *ofproto,
5943 struct ofpbuf *odp_actions,
5944 const struct flow *flow,
5947 uint32_t probability;
5948 union user_action_cookie cookie;
5950 if (!ofproto->sflow || flow->in_port == OFPP_NONE) {
5954 probability = dpif_sflow_get_probability(ofproto->sflow);
5955 compose_sflow_cookie(ofproto, htons(0), odp_port,
5956 odp_port == OVSP_NONE ? 0 : 1, &cookie);
5958 return compose_sample_action(ofproto, odp_actions, flow, probability,
5959 &cookie, sizeof cookie.sflow);
5963 compose_flow_sample_cookie(uint16_t probability, uint32_t collector_set_id,
5964 uint32_t obs_domain_id, uint32_t obs_point_id,
5965 union user_action_cookie *cookie)
5967 cookie->type = USER_ACTION_COOKIE_FLOW_SAMPLE;
5968 cookie->flow_sample.probability = probability;
5969 cookie->flow_sample.collector_set_id = collector_set_id;
5970 cookie->flow_sample.obs_domain_id = obs_domain_id;
5971 cookie->flow_sample.obs_point_id = obs_point_id;
5975 compose_ipfix_cookie(union user_action_cookie *cookie)
5977 cookie->type = USER_ACTION_COOKIE_IPFIX;
5980 /* Compose SAMPLE action for IPFIX bridge sampling. */
5982 compose_ipfix_action(const struct ofproto_dpif *ofproto,
5983 struct ofpbuf *odp_actions,
5984 const struct flow *flow)
5986 uint32_t probability;
5987 union user_action_cookie cookie;
5989 if (!ofproto->ipfix || flow->in_port == OFPP_NONE) {
5993 probability = dpif_ipfix_get_bridge_exporter_probability(ofproto->ipfix);
5994 compose_ipfix_cookie(&cookie);
5996 compose_sample_action(ofproto, odp_actions, flow, probability,
5997 &cookie, sizeof cookie.ipfix);
6000 /* SAMPLE action for sFlow must be first action in any given list of
6001 * actions. At this point we do not have all information required to
6002 * build it. So try to build sample action as complete as possible. */
6004 add_sflow_action(struct xlate_ctx *ctx)
6006 ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto,
6007 &ctx->xout->odp_actions,
6008 &ctx->xin->flow, OVSP_NONE);
6009 ctx->sflow_odp_port = 0;
6010 ctx->sflow_n_outputs = 0;
6013 /* SAMPLE action for IPFIX must be 1st or 2nd action in any given list
6014 * of actions, eventually after the SAMPLE action for sFlow. */
6016 add_ipfix_action(struct xlate_ctx *ctx)
6018 compose_ipfix_action(ctx->ofproto, &ctx->xout->odp_actions,
6022 /* Fix SAMPLE action according to data collected while composing ODP actions.
6023 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
6024 * USERSPACE action's user-cookie which is required for sflow. */
6026 fix_sflow_action(struct xlate_ctx *ctx)
6028 const struct flow *base = &ctx->base_flow;
6029 union user_action_cookie *cookie;
6031 if (!ctx->user_cookie_offset) {
6035 cookie = ofpbuf_at(&ctx->xout->odp_actions, ctx->user_cookie_offset,
6036 sizeof cookie->sflow);
6037 ovs_assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
6039 compose_sflow_cookie(ctx->ofproto, base->vlan_tci,
6040 ctx->sflow_odp_port, ctx->sflow_n_outputs, cookie);
6044 compose_output_action__(struct xlate_ctx *ctx, uint16_t ofp_port,
6047 const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
6048 struct flow_wildcards *wc = &ctx->xout->wc;
6049 ovs_be16 flow_vlan_tci;
6050 uint32_t flow_skb_mark;
6051 uint8_t flow_nw_tos;
6052 struct priority_to_dscp *pdscp;
6053 uint32_t out_port, odp_port;
6055 /* If 'struct flow' gets additional metadata, we'll need to zero it out
6056 * before traversing a patch port. */
6057 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 20);
6060 xlate_report(ctx, "Nonexistent output port");
6062 } else if (ofport->up.pp.config & OFPUTIL_PC_NO_FWD) {
6063 xlate_report(ctx, "OFPPC_NO_FWD set, skipping output");
6065 } else if (check_stp && !stp_forward_in_state(ofport->stp_state)) {
6066 xlate_report(ctx, "STP not in forwarding state, skipping output");
6070 if (netdev_vport_is_patch(ofport->up.netdev)) {
6071 struct ofport_dpif *peer = ofport_get_peer(ofport);
6072 struct flow old_flow = ctx->xin->flow;
6073 const struct ofproto_dpif *peer_ofproto;
6074 enum slow_path_reason special;
6075 struct ofport_dpif *in_port;
6078 xlate_report(ctx, "Nonexistent patch port peer");
6082 peer_ofproto = ofproto_dpif_cast(peer->up.ofproto);
6083 if (peer_ofproto->backer != ctx->ofproto->backer) {
6084 xlate_report(ctx, "Patch port peer on a different datapath");
6088 ctx->ofproto = ofproto_dpif_cast(peer->up.ofproto);
6089 ctx->xin->flow.in_port = peer->up.ofp_port;
6090 ctx->xin->flow.metadata = htonll(0);
6091 memset(&ctx->xin->flow.tunnel, 0, sizeof ctx->xin->flow.tunnel);
6092 memset(ctx->xin->flow.regs, 0, sizeof ctx->xin->flow.regs);
6094 in_port = get_ofp_port(ctx->ofproto, ctx->xin->flow.in_port);
6095 special = process_special(ctx, &ctx->xin->flow, in_port,
6098 ctx->xout->slow = special;
6099 } else if (!in_port || may_receive(in_port, ctx)) {
6100 if (!in_port || stp_forward_in_state(in_port->stp_state)) {
6101 xlate_table_action(ctx, ctx->xin->flow.in_port, 0, true);
6103 /* Forwarding is disabled by STP. Let OFPP_NORMAL and the
6104 * learning action look at the packet, then drop it. */
6105 struct flow old_base_flow = ctx->base_flow;
6106 size_t old_size = ctx->xout->odp_actions.size;
6107 xlate_table_action(ctx, ctx->xin->flow.in_port, 0, true);
6108 ctx->base_flow = old_base_flow;
6109 ctx->xout->odp_actions.size = old_size;
6113 ctx->xin->flow = old_flow;
6114 ctx->ofproto = ofproto_dpif_cast(ofport->up.ofproto);
6116 if (ctx->xin->resubmit_stats) {
6117 netdev_vport_inc_tx(ofport->up.netdev, ctx->xin->resubmit_stats);
6118 netdev_vport_inc_rx(peer->up.netdev, ctx->xin->resubmit_stats);
6124 flow_vlan_tci = ctx->xin->flow.vlan_tci;
6125 flow_skb_mark = ctx->xin->flow.skb_mark;
6126 flow_nw_tos = ctx->xin->flow.nw_tos;
6128 pdscp = get_priority(ofport, ctx->xin->flow.skb_priority);
6130 wc->masks.nw_tos |= IP_ECN_MASK;
6131 ctx->xin->flow.nw_tos &= ~IP_DSCP_MASK;
6132 ctx->xin->flow.nw_tos |= pdscp->dscp;
6135 if (ofport->tnl_port) {
6136 /* Save tunnel metadata so that changes made due to
6137 * the Logical (tunnel) Port are not visible for any further
6138 * matches, while explicit set actions on tunnel metadata are.
6140 struct flow_tnl flow_tnl = ctx->xin->flow.tunnel;
6141 odp_port = tnl_port_send(ofport->tnl_port, &ctx->xin->flow,
6143 if (odp_port == OVSP_NONE) {
6144 xlate_report(ctx, "Tunneling decided against output");
6145 goto out; /* restore flow_nw_tos */
6148 if (ctx->xin->resubmit_stats) {
6149 netdev_vport_inc_tx(ofport->up.netdev, ctx->xin->resubmit_stats);
6151 out_port = odp_port;
6152 commit_odp_tunnel_action(&ctx->xin->flow, &ctx->base_flow,
6153 &ctx->xout->odp_actions);
6154 ctx->xin->flow.tunnel = flow_tnl; /* Restore tunnel metadata */
6156 uint16_t vlandev_port;
6158 odp_port = ofport->odp_port;
6159 if (!hmap_is_empty(&ctx->ofproto->realdev_vid_map)) {
6160 wc->masks.vlan_tci |= htons(VLAN_VID_MASK | VLAN_CFI);
6162 vlandev_port = vsp_realdev_to_vlandev(ctx->ofproto, ofp_port,
6163 ctx->xin->flow.vlan_tci);
6164 if (vlandev_port == ofp_port) {
6165 out_port = odp_port;
6167 out_port = ofp_port_to_odp_port(ctx->ofproto, vlandev_port);
6168 ctx->xin->flow.vlan_tci = htons(0);
6170 ctx->xin->flow.skb_mark &= ~IPSEC_MARK;
6172 commit_odp_actions(&ctx->xin->flow, &ctx->base_flow,
6173 &ctx->xout->odp_actions, &ctx->xout->wc);
6174 nl_msg_put_u32(&ctx->xout->odp_actions, OVS_ACTION_ATTR_OUTPUT, out_port);
6176 ctx->sflow_odp_port = odp_port;
6177 ctx->sflow_n_outputs++;
6178 ctx->xout->nf_output_iface = ofp_port;
6181 ctx->xin->flow.vlan_tci = flow_vlan_tci;
6182 ctx->xin->flow.skb_mark = flow_skb_mark;
6184 ctx->xin->flow.nw_tos = flow_nw_tos;
6188 compose_output_action(struct xlate_ctx *ctx, uint16_t ofp_port)
6190 compose_output_action__(ctx, ofp_port, true);
6194 tag_the_flow(struct xlate_ctx *ctx, struct rule_dpif *rule)
6196 struct ofproto_dpif *ofproto = ctx->ofproto;
6197 uint8_t table_id = ctx->table_id;
6199 if (table_id > 0 && table_id < N_TABLES) {
6200 struct table_dpif *table = &ofproto->tables[table_id];
6201 if (table->other_table) {
6202 ctx->xout->tags |= (rule && rule->tag
6204 : rule_calculate_tag(&ctx->xin->flow,
6205 &table->other_table->mask,
6211 /* Common rule processing in one place to avoid duplicating code. */
6212 static struct rule_dpif *
6213 ctx_rule_hooks(struct xlate_ctx *ctx, struct rule_dpif *rule,
6216 if (ctx->xin->resubmit_hook) {
6217 ctx->xin->resubmit_hook(ctx, rule);
6219 if (rule == NULL && may_packet_in) {
6221 * check if table configuration flags
6222 * OFPTC_TABLE_MISS_CONTROLLER, default.
6223 * OFPTC_TABLE_MISS_CONTINUE,
6224 * OFPTC_TABLE_MISS_DROP
6225 * When OF1.0, OFPTC_TABLE_MISS_CONTINUE is used. What to do?
6227 rule = rule_dpif_miss_rule(ctx->ofproto, &ctx->xin->flow);
6229 if (rule && ctx->xin->resubmit_stats) {
6230 rule_credit_stats(rule, ctx->xin->resubmit_stats);
6236 xlate_table_action(struct xlate_ctx *ctx,
6237 uint16_t in_port, uint8_t table_id, bool may_packet_in)
6239 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
6240 struct rule_dpif *rule;
6241 uint16_t old_in_port = ctx->xin->flow.in_port;
6242 uint8_t old_table_id = ctx->table_id;
6244 ctx->table_id = table_id;
6246 /* Look up a flow with 'in_port' as the input port. */
6247 ctx->xin->flow.in_port = in_port;
6248 rule = rule_dpif_lookup__(ctx->ofproto, &ctx->xin->flow,
6249 &ctx->xout->wc, table_id);
6251 tag_the_flow(ctx, rule);
6253 /* Restore the original input port. Otherwise OFPP_NORMAL and
6254 * OFPP_IN_PORT will have surprising behavior. */
6255 ctx->xin->flow.in_port = old_in_port;
6257 rule = ctx_rule_hooks(ctx, rule, may_packet_in);
6260 struct rule_dpif *old_rule = ctx->rule;
6264 do_xlate_actions(rule->up.ofpacts, rule->up.ofpacts_len, ctx);
6265 ctx->rule = old_rule;
6269 ctx->table_id = old_table_id;
6271 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
6273 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
6274 MAX_RESUBMIT_RECURSION);
6275 ctx->max_resubmit_trigger = true;
6280 xlate_ofpact_resubmit(struct xlate_ctx *ctx,
6281 const struct ofpact_resubmit *resubmit)
6286 in_port = resubmit->in_port;
6287 if (in_port == OFPP_IN_PORT) {
6288 in_port = ctx->xin->flow.in_port;
6291 table_id = resubmit->table_id;
6292 if (table_id == 255) {
6293 table_id = ctx->table_id;
6296 xlate_table_action(ctx, in_port, table_id, false);
6300 flood_packets(struct xlate_ctx *ctx, bool all)
6302 struct ofport_dpif *ofport;
6304 HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) {
6305 uint16_t ofp_port = ofport->up.ofp_port;
6307 if (ofp_port == ctx->xin->flow.in_port) {
6312 compose_output_action__(ctx, ofp_port, false);
6313 } else if (!(ofport->up.pp.config & OFPUTIL_PC_NO_FLOOD)) {
6314 compose_output_action(ctx, ofp_port);
6318 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
6322 execute_controller_action(struct xlate_ctx *ctx, int len,
6323 enum ofp_packet_in_reason reason,
6324 uint16_t controller_id)
6326 struct ofputil_packet_in pin;
6327 struct ofpbuf *packet;
6329 ovs_assert(!ctx->xout->slow || ctx->xout->slow == SLOW_CONTROLLER);
6330 ctx->xout->slow = SLOW_CONTROLLER;
6331 if (!ctx->xin->packet) {
6335 packet = ofpbuf_clone(ctx->xin->packet);
6337 if (packet->l2 && packet->l3) {
6338 struct eth_header *eh;
6339 uint16_t mpls_depth;
6341 eth_pop_vlan(packet);
6344 memcpy(eh->eth_src, ctx->xin->flow.dl_src, sizeof eh->eth_src);
6345 memcpy(eh->eth_dst, ctx->xin->flow.dl_dst, sizeof eh->eth_dst);
6347 if (ctx->xin->flow.vlan_tci & htons(VLAN_CFI)) {
6348 eth_push_vlan(packet, ctx->xin->flow.vlan_tci);
6351 mpls_depth = eth_mpls_depth(packet);
6353 if (mpls_depth < ctx->xin->flow.mpls_depth) {
6354 push_mpls(packet, ctx->xin->flow.dl_type, ctx->xin->flow.mpls_lse);
6355 } else if (mpls_depth > ctx->xin->flow.mpls_depth) {
6356 pop_mpls(packet, ctx->xin->flow.dl_type);
6357 } else if (mpls_depth) {
6358 set_mpls_lse(packet, ctx->xin->flow.mpls_lse);
6362 if (ctx->xin->flow.dl_type == htons(ETH_TYPE_IP)) {
6363 packet_set_ipv4(packet, ctx->xin->flow.nw_src,
6364 ctx->xin->flow.nw_dst, ctx->xin->flow.nw_tos,
6365 ctx->xin->flow.nw_ttl);
6369 if (ctx->xin->flow.nw_proto == IPPROTO_TCP) {
6370 packet_set_tcp_port(packet, ctx->xin->flow.tp_src,
6371 ctx->xin->flow.tp_dst);
6372 } else if (ctx->xin->flow.nw_proto == IPPROTO_UDP) {
6373 packet_set_udp_port(packet, ctx->xin->flow.tp_src,
6374 ctx->xin->flow.tp_dst);
6380 pin.packet = packet->data;
6381 pin.packet_len = packet->size;
6382 pin.reason = reason;
6383 pin.controller_id = controller_id;
6384 pin.table_id = ctx->table_id;
6385 pin.cookie = ctx->rule ? ctx->rule->up.flow_cookie : 0;
6388 flow_get_metadata(&ctx->xin->flow, &pin.fmd);
6390 connmgr_send_packet_in(ctx->ofproto->up.connmgr, &pin);
6391 ofpbuf_delete(packet);
6395 execute_mpls_push_action(struct xlate_ctx *ctx, ovs_be16 eth_type)
6397 struct flow_wildcards *wc = &ctx->xout->wc;
6398 ovs_assert(eth_type_mpls(eth_type));
6400 memset(&wc->masks.mpls_lse, 0xff, sizeof wc->masks.mpls_lse);
6401 memset(&wc->masks.mpls_depth, 0xff, sizeof wc->masks.mpls_depth);
6403 if (ctx->base_flow.mpls_depth) {
6404 ctx->xin->flow.mpls_lse &= ~htonl(MPLS_BOS_MASK);
6405 ctx->xin->flow.mpls_depth++;
6410 if (ctx->xin->flow.dl_type == htons(ETH_TYPE_IPV6)) {
6411 label = htonl(0x2); /* IPV6 Explicit Null. */
6413 label = htonl(0x0); /* IPV4 Explicit Null. */
6415 wc->masks.nw_tos |= IP_DSCP_MASK;
6416 wc->masks.nw_ttl = 0xff;
6417 tc = (ctx->xin->flow.nw_tos & IP_DSCP_MASK) >> 2;
6418 ttl = ctx->xin->flow.nw_ttl ? ctx->xin->flow.nw_ttl : 0x40;
6419 ctx->xin->flow.mpls_lse = set_mpls_lse_values(ttl, tc, 1, label);
6420 ctx->xin->flow.mpls_depth = 1;
6422 ctx->xin->flow.dl_type = eth_type;
6426 execute_mpls_pop_action(struct xlate_ctx *ctx, ovs_be16 eth_type)
6428 struct flow_wildcards *wc = &ctx->xout->wc;
6430 ovs_assert(eth_type_mpls(ctx->xin->flow.dl_type));
6431 ovs_assert(!eth_type_mpls(eth_type));
6433 memset(&wc->masks.mpls_lse, 0xff, sizeof wc->masks.mpls_lse);
6434 memset(&wc->masks.mpls_depth, 0xff, sizeof wc->masks.mpls_depth);
6436 if (ctx->xin->flow.mpls_depth) {
6437 ctx->xin->flow.mpls_depth--;
6438 ctx->xin->flow.mpls_lse = htonl(0);
6439 if (!ctx->xin->flow.mpls_depth) {
6440 ctx->xin->flow.dl_type = eth_type;
6446 compose_dec_ttl(struct xlate_ctx *ctx, struct ofpact_cnt_ids *ids)
6448 if (ctx->xin->flow.dl_type != htons(ETH_TYPE_IP) &&
6449 ctx->xin->flow.dl_type != htons(ETH_TYPE_IPV6)) {
6453 ctx->xout->wc.masks.nw_ttl = 0xff;
6454 if (ctx->xin->flow.nw_ttl > 1) {
6455 ctx->xin->flow.nw_ttl--;
6460 for (i = 0; i < ids->n_controllers; i++) {
6461 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL,
6465 /* Stop processing for current table. */
6471 execute_set_mpls_ttl_action(struct xlate_ctx *ctx, uint8_t ttl)
6473 if (!eth_type_mpls(ctx->xin->flow.dl_type)) {
6477 set_mpls_lse_ttl(&ctx->xin->flow.mpls_lse, ttl);
6482 execute_dec_mpls_ttl_action(struct xlate_ctx *ctx)
6484 uint8_t ttl = mpls_lse_to_ttl(ctx->xin->flow.mpls_lse);
6485 struct flow_wildcards *wc = &ctx->xout->wc;
6487 memset(&wc->masks.mpls_lse, 0xff, sizeof wc->masks.mpls_lse);
6489 if (!eth_type_mpls(ctx->xin->flow.dl_type)) {
6495 set_mpls_lse_ttl(&ctx->xin->flow.mpls_lse, ttl);
6498 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL, 0);
6500 /* Stop processing for current table. */
6506 xlate_output_action(struct xlate_ctx *ctx,
6507 uint16_t port, uint16_t max_len, bool may_packet_in)
6509 uint16_t prev_nf_output_iface = ctx->xout->nf_output_iface;
6511 ctx->xout->nf_output_iface = NF_OUT_DROP;
6515 compose_output_action(ctx, ctx->xin->flow.in_port);
6518 xlate_table_action(ctx, ctx->xin->flow.in_port, 0, may_packet_in);
6524 flood_packets(ctx, false);
6527 flood_packets(ctx, true);
6529 case OFPP_CONTROLLER:
6530 execute_controller_action(ctx, max_len, OFPR_ACTION, 0);
6536 if (port != ctx->xin->flow.in_port) {
6537 compose_output_action(ctx, port);
6539 xlate_report(ctx, "skipping output to input port");
6544 if (prev_nf_output_iface == NF_OUT_FLOOD) {
6545 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
6546 } else if (ctx->xout->nf_output_iface == NF_OUT_DROP) {
6547 ctx->xout->nf_output_iface = prev_nf_output_iface;
6548 } else if (prev_nf_output_iface != NF_OUT_DROP &&
6549 ctx->xout->nf_output_iface != NF_OUT_FLOOD) {
6550 ctx->xout->nf_output_iface = NF_OUT_MULTI;
6555 xlate_output_reg_action(struct xlate_ctx *ctx,
6556 const struct ofpact_output_reg *or)
6558 uint64_t port = mf_get_subfield(&or->src, &ctx->xin->flow);
6559 if (port <= UINT16_MAX) {
6560 union mf_subvalue value;
6562 memset(&value, 0xff, sizeof value);
6563 mf_write_subfield_flow(&or->src, &value, &ctx->xout->wc.masks);
6564 xlate_output_action(ctx, port, or->max_len, false);
6569 xlate_enqueue_action(struct xlate_ctx *ctx,
6570 const struct ofpact_enqueue *enqueue)
6572 uint16_t ofp_port = enqueue->port;
6573 uint32_t queue_id = enqueue->queue;
6574 uint32_t flow_priority, priority;
6577 /* Translate queue to priority. */
6578 error = dpif_queue_to_priority(ctx->ofproto->backer->dpif,
6579 queue_id, &priority);
6581 /* Fall back to ordinary output action. */
6582 xlate_output_action(ctx, enqueue->port, 0, false);
6586 /* Check output port. */
6587 if (ofp_port == OFPP_IN_PORT) {
6588 ofp_port = ctx->xin->flow.in_port;
6589 } else if (ofp_port == ctx->xin->flow.in_port) {
6593 /* Add datapath actions. */
6594 flow_priority = ctx->xin->flow.skb_priority;
6595 ctx->xin->flow.skb_priority = priority;
6596 compose_output_action(ctx, ofp_port);
6597 ctx->xin->flow.skb_priority = flow_priority;
6599 /* Update NetFlow output port. */
6600 if (ctx->xout->nf_output_iface == NF_OUT_DROP) {
6601 ctx->xout->nf_output_iface = ofp_port;
6602 } else if (ctx->xout->nf_output_iface != NF_OUT_FLOOD) {
6603 ctx->xout->nf_output_iface = NF_OUT_MULTI;
6608 xlate_set_queue_action(struct xlate_ctx *ctx, uint32_t queue_id)
6610 uint32_t skb_priority;
6612 if (!dpif_queue_to_priority(ctx->ofproto->backer->dpif,
6613 queue_id, &skb_priority)) {
6614 ctx->xin->flow.skb_priority = skb_priority;
6616 /* Couldn't translate queue to a priority. Nothing to do. A warning
6617 * has already been logged. */
6622 slave_enabled_cb(uint16_t ofp_port, void *ofproto_)
6624 struct ofproto_dpif *ofproto = ofproto_;
6625 struct ofport_dpif *port;
6635 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
6638 port = get_ofp_port(ofproto, ofp_port);
6639 return port ? port->may_enable : false;
6644 xlate_bundle_action(struct xlate_ctx *ctx,
6645 const struct ofpact_bundle *bundle)
6649 port = bundle_execute(bundle, &ctx->xin->flow, &ctx->xout->wc,
6650 slave_enabled_cb, ctx->ofproto);
6651 if (bundle->dst.field) {
6652 nxm_reg_load(&bundle->dst, port, &ctx->xin->flow);
6654 xlate_output_action(ctx, port, 0, false);
6659 xlate_learn_action(struct xlate_ctx *ctx,
6660 const struct ofpact_learn *learn)
6662 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
6663 struct ofputil_flow_mod fm;
6664 uint64_t ofpacts_stub[1024 / 8];
6665 struct ofpbuf ofpacts;
6668 ctx->xout->has_learn = true;
6670 learn_mask(learn, &ctx->xout->wc);
6672 if (!ctx->xin->may_learn) {
6676 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
6677 learn_execute(learn, &ctx->xin->flow, &fm, &ofpacts);
6679 error = ofproto_flow_mod(&ctx->ofproto->up, &fm);
6680 if (error && !VLOG_DROP_WARN(&rl)) {
6681 VLOG_WARN("learning action failed to modify flow table (%s)",
6682 ofperr_get_name(error));
6685 ofpbuf_uninit(&ofpacts);
6688 /* Reduces '*timeout' to no more than 'max'. A value of zero in either case
6689 * means "infinite". */
6691 reduce_timeout(uint16_t max, uint16_t *timeout)
6693 if (max && (!*timeout || *timeout > max)) {
6699 xlate_fin_timeout(struct xlate_ctx *ctx,
6700 const struct ofpact_fin_timeout *oft)
6702 if (ctx->xin->tcp_flags & (TCP_FIN | TCP_RST) && ctx->rule) {
6703 struct rule_dpif *rule = ctx->rule;
6705 reduce_timeout(oft->fin_idle_timeout, &rule->up.idle_timeout);
6706 reduce_timeout(oft->fin_hard_timeout, &rule->up.hard_timeout);
6711 xlate_sample_action(struct xlate_ctx *ctx,
6712 const struct ofpact_sample *os)
6714 union user_action_cookie cookie;
6715 /* Scale the probability from 16-bit to 32-bit while representing
6716 * the same percentage. */
6717 uint32_t probability = (os->probability << 16) | os->probability;
6719 commit_odp_actions(&ctx->xin->flow, &ctx->base_flow,
6720 &ctx->xout->odp_actions, &ctx->xout->wc);
6722 compose_flow_sample_cookie(os->probability, os->collector_set_id,
6723 os->obs_domain_id, os->obs_point_id, &cookie);
6724 compose_sample_action(ctx->ofproto, &ctx->xout->odp_actions, &ctx->xin->flow,
6725 probability, &cookie, sizeof cookie.flow_sample);
6729 may_receive(const struct ofport_dpif *port, struct xlate_ctx *ctx)
6731 if (port->up.pp.config & (eth_addr_equals(ctx->xin->flow.dl_dst,
6733 ? OFPUTIL_PC_NO_RECV_STP
6734 : OFPUTIL_PC_NO_RECV)) {
6738 /* Only drop packets here if both forwarding and learning are
6739 * disabled. If just learning is enabled, we need to have
6740 * OFPP_NORMAL and the learning action have a look at the packet
6741 * before we can drop it. */
6742 if (!stp_forward_in_state(port->stp_state)
6743 && !stp_learn_in_state(port->stp_state)) {
6751 tunnel_ecn_ok(struct xlate_ctx *ctx)
6753 if (is_ip_any(&ctx->base_flow)
6754 && (ctx->base_flow.tunnel.ip_tos & IP_ECN_MASK) == IP_ECN_CE) {
6755 if ((ctx->base_flow.nw_tos & IP_ECN_MASK) == IP_ECN_NOT_ECT) {
6756 VLOG_WARN_RL(&rl, "dropping tunnel packet marked ECN CE"
6757 " but is not ECN capable");
6760 /* Set the ECN CE value in the tunneled packet. */
6761 ctx->xin->flow.nw_tos |= IP_ECN_CE;
6769 do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len,
6770 struct xlate_ctx *ctx)
6772 bool was_evictable = true;
6773 const struct ofpact *a;
6776 /* Don't let the rule we're working on get evicted underneath us. */
6777 was_evictable = ctx->rule->up.evictable;
6778 ctx->rule->up.evictable = false;
6781 do_xlate_actions_again:
6782 OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) {
6783 struct ofpact_controller *controller;
6784 const struct ofpact_metadata *metadata;
6792 xlate_output_action(ctx, ofpact_get_OUTPUT(a)->port,
6793 ofpact_get_OUTPUT(a)->max_len, true);
6796 case OFPACT_CONTROLLER:
6797 controller = ofpact_get_CONTROLLER(a);
6798 execute_controller_action(ctx, controller->max_len,
6800 controller->controller_id);
6803 case OFPACT_ENQUEUE:
6804 xlate_enqueue_action(ctx, ofpact_get_ENQUEUE(a));
6807 case OFPACT_SET_VLAN_VID:
6808 ctx->xin->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
6809 ctx->xin->flow.vlan_tci |=
6810 (htons(ofpact_get_SET_VLAN_VID(a)->vlan_vid)
6814 case OFPACT_SET_VLAN_PCP:
6815 ctx->xin->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
6816 ctx->xin->flow.vlan_tci |=
6817 htons((ofpact_get_SET_VLAN_PCP(a)->vlan_pcp << VLAN_PCP_SHIFT)
6821 case OFPACT_STRIP_VLAN:
6822 ctx->xin->flow.vlan_tci = htons(0);
6825 case OFPACT_PUSH_VLAN:
6826 /* XXX 802.1AD(QinQ) */
6827 ctx->xin->flow.vlan_tci = htons(VLAN_CFI);
6830 case OFPACT_SET_ETH_SRC:
6831 memcpy(ctx->xin->flow.dl_src, ofpact_get_SET_ETH_SRC(a)->mac,
6835 case OFPACT_SET_ETH_DST:
6836 memcpy(ctx->xin->flow.dl_dst, ofpact_get_SET_ETH_DST(a)->mac,
6840 case OFPACT_SET_IPV4_SRC:
6841 if (ctx->xin->flow.dl_type == htons(ETH_TYPE_IP)) {
6842 ctx->xin->flow.nw_src = ofpact_get_SET_IPV4_SRC(a)->ipv4;
6846 case OFPACT_SET_IPV4_DST:
6847 if (ctx->xin->flow.dl_type == htons(ETH_TYPE_IP)) {
6848 ctx->xin->flow.nw_dst = ofpact_get_SET_IPV4_DST(a)->ipv4;
6852 case OFPACT_SET_IPV4_DSCP:
6853 /* OpenFlow 1.0 only supports IPv4. */
6854 if (ctx->xin->flow.dl_type == htons(ETH_TYPE_IP)) {
6855 ctx->xin->flow.nw_tos &= ~IP_DSCP_MASK;
6856 ctx->xin->flow.nw_tos |= ofpact_get_SET_IPV4_DSCP(a)->dscp;
6860 case OFPACT_SET_L4_SRC_PORT:
6861 memset(&ctx->xout->wc.masks.nw_proto, 0xff,
6862 sizeof ctx->xout->wc.masks.nw_proto);
6863 if (is_ip_any(&ctx->xin->flow)) {
6864 ctx->xin->flow.tp_src =
6865 htons(ofpact_get_SET_L4_SRC_PORT(a)->port);
6869 case OFPACT_SET_L4_DST_PORT:
6870 memset(&ctx->xout->wc.masks.nw_proto, 0xff,
6871 sizeof ctx->xout->wc.masks.nw_proto);
6872 if (is_ip_any(&ctx->xin->flow)) {
6873 ctx->xin->flow.tp_dst =
6874 htons(ofpact_get_SET_L4_DST_PORT(a)->port);
6878 case OFPACT_RESUBMIT:
6879 xlate_ofpact_resubmit(ctx, ofpact_get_RESUBMIT(a));
6882 case OFPACT_SET_TUNNEL:
6883 ctx->xin->flow.tunnel.tun_id =
6884 htonll(ofpact_get_SET_TUNNEL(a)->tun_id);
6887 case OFPACT_SET_QUEUE:
6888 xlate_set_queue_action(ctx, ofpact_get_SET_QUEUE(a)->queue_id);
6891 case OFPACT_POP_QUEUE:
6892 ctx->xin->flow.skb_priority = ctx->orig_skb_priority;
6895 case OFPACT_REG_MOVE:
6896 nxm_execute_reg_move(ofpact_get_REG_MOVE(a), &ctx->xin->flow,
6900 case OFPACT_REG_LOAD:
6901 nxm_execute_reg_load(ofpact_get_REG_LOAD(a), &ctx->xin->flow);
6904 case OFPACT_STACK_PUSH:
6905 nxm_execute_stack_push(ofpact_get_STACK_PUSH(a), &ctx->xin->flow,
6906 &ctx->xout->wc, &ctx->stack);
6909 case OFPACT_STACK_POP:
6910 nxm_execute_stack_pop(ofpact_get_STACK_POP(a), &ctx->xin->flow,
6914 case OFPACT_PUSH_MPLS:
6915 execute_mpls_push_action(ctx, ofpact_get_PUSH_MPLS(a)->ethertype);
6918 case OFPACT_POP_MPLS:
6919 execute_mpls_pop_action(ctx, ofpact_get_POP_MPLS(a)->ethertype);
6922 case OFPACT_SET_MPLS_TTL:
6923 if (execute_set_mpls_ttl_action(ctx,
6924 ofpact_get_SET_MPLS_TTL(a)->ttl)) {
6929 case OFPACT_DEC_MPLS_TTL:
6930 if (execute_dec_mpls_ttl_action(ctx)) {
6935 case OFPACT_DEC_TTL:
6936 if (compose_dec_ttl(ctx, ofpact_get_DEC_TTL(a))) {
6942 /* Nothing to do. */
6945 case OFPACT_MULTIPATH:
6946 multipath_execute(ofpact_get_MULTIPATH(a), &ctx->xin->flow,
6951 ctx->ofproto->has_bundle_action = true;
6952 xlate_bundle_action(ctx, ofpact_get_BUNDLE(a));
6955 case OFPACT_OUTPUT_REG:
6956 xlate_output_reg_action(ctx, ofpact_get_OUTPUT_REG(a));
6960 xlate_learn_action(ctx, ofpact_get_LEARN(a));
6967 case OFPACT_FIN_TIMEOUT:
6968 memset(&ctx->xout->wc.masks.nw_proto, 0xff,
6969 sizeof ctx->xout->wc.masks.nw_proto);
6970 ctx->xout->has_fin_timeout = true;
6971 xlate_fin_timeout(ctx, ofpact_get_FIN_TIMEOUT(a));
6974 case OFPACT_CLEAR_ACTIONS:
6976 * Nothing to do because writa-actions is not supported for now.
6977 * When writa-actions is supported, clear-actions also must
6978 * be supported at the same time.
6982 case OFPACT_WRITE_METADATA:
6983 metadata = ofpact_get_WRITE_METADATA(a);
6984 ctx->xin->flow.metadata &= ~metadata->mask;
6985 ctx->xin->flow.metadata |= metadata->metadata & metadata->mask;
6988 case OFPACT_GOTO_TABLE: {
6989 /* It is assumed that goto-table is the last action. */
6990 struct ofpact_goto_table *ogt = ofpact_get_GOTO_TABLE(a);
6991 struct rule_dpif *rule;
6993 ovs_assert(ctx->table_id < ogt->table_id);
6995 ctx->table_id = ogt->table_id;
6997 /* Look up a flow from the new table. */
6998 rule = rule_dpif_lookup__(ctx->ofproto, &ctx->xin->flow,
6999 &ctx->xout->wc, ctx->table_id);
7001 tag_the_flow(ctx, rule);
7003 rule = ctx_rule_hooks(ctx, rule, true);
7007 ctx->rule->up.evictable = was_evictable;
7010 was_evictable = rule->up.evictable;
7011 rule->up.evictable = false;
7013 /* Tail recursion removal. */
7014 ofpacts = rule->up.ofpacts;
7015 ofpacts_len = rule->up.ofpacts_len;
7016 goto do_xlate_actions_again;
7022 xlate_sample_action(ctx, ofpact_get_SAMPLE(a));
7029 ctx->rule->up.evictable = was_evictable;
7034 xlate_in_init(struct xlate_in *xin, struct ofproto_dpif *ofproto,
7035 const struct flow *flow,
7036 const struct initial_vals *initial_vals,
7037 struct rule_dpif *rule, uint8_t tcp_flags,
7038 const struct ofpbuf *packet)
7040 xin->ofproto = ofproto;
7042 xin->packet = packet;
7043 xin->may_learn = packet != NULL;
7045 xin->ofpacts = NULL;
7046 xin->ofpacts_len = 0;
7047 xin->tcp_flags = tcp_flags;
7048 xin->resubmit_hook = NULL;
7049 xin->report_hook = NULL;
7050 xin->resubmit_stats = NULL;
7053 xin->initial_vals = *initial_vals;
7055 xin->initial_vals.vlan_tci = xin->flow.vlan_tci;
7056 xin->initial_vals.tunnel_ip_tos = xin->flow.tunnel.ip_tos;
7061 xlate_out_uninit(struct xlate_out *xout)
7064 ofpbuf_uninit(&xout->odp_actions);
7068 /* Translates the 'ofpacts_len' bytes of "struct ofpacts" starting at 'ofpacts'
7069 * into datapath actions in 'odp_actions', using 'ctx'. */
7071 xlate_actions(struct xlate_in *xin, struct xlate_out *xout)
7073 /* Normally false. Set to true if we ever hit MAX_RESUBMIT_RECURSION, so
7074 * that in the future we always keep a copy of the original flow for
7075 * tracing purposes. */
7076 static bool hit_resubmit_limit;
7077 struct flow_wildcards *wc = &xout->wc;
7079 enum slow_path_reason special;
7080 const struct ofpact *ofpacts;
7081 struct ofport_dpif *in_port;
7082 struct flow orig_flow;
7083 struct xlate_ctx ctx;
7086 COVERAGE_INC(ofproto_dpif_xlate);
7088 /* Flow initialization rules:
7089 * - 'base_flow' must match the kernel's view of the packet at the
7090 * time that action processing starts. 'flow' represents any
7091 * transformations we wish to make through actions.
7092 * - By default 'base_flow' and 'flow' are the same since the input
7093 * packet matches the output before any actions are applied.
7094 * - When using VLAN splinters, 'base_flow''s VLAN is set to the value
7095 * of the received packet as seen by the kernel. If we later output
7096 * to another device without any modifications this will cause us to
7097 * insert a new tag since the original one was stripped off by the
7099 * - Tunnel 'flow' is largely cleared when transitioning between
7100 * the input and output stages since it does not make sense to output
7101 * a packet with the exact headers that it was received with (i.e.
7102 * the destination IP is us). The one exception is the tun_id, which
7103 * is preserved to allow use in later resubmit lookups and loads into
7105 * - Tunnel 'base_flow' is completely cleared since that is what the
7106 * kernel does. If we wish to maintain the original values an action
7107 * needs to be generated. */
7112 ctx.ofproto = xin->ofproto;
7113 ctx.rule = xin->rule;
7115 ctx.base_flow = ctx.xin->flow;
7116 memset(&ctx.base_flow.tunnel, 0, sizeof ctx.base_flow.tunnel);
7117 ctx.base_flow.vlan_tci = xin->initial_vals.vlan_tci;
7118 ctx.base_flow.tunnel.ip_tos = xin->initial_vals.tunnel_ip_tos;
7120 flow_wildcards_init_catchall(&ctx.xout->wc);
7121 memset(&wc->masks.in_port, 0xff, sizeof wc->masks.in_port);
7122 memset(&wc->masks.skb_priority, 0xff, sizeof wc->masks.skb_priority);
7123 memset(&wc->masks.dl_type, 0xff, sizeof wc->masks.dl_type);
7124 wc->masks.nw_frag |= FLOW_NW_FRAG_MASK;
7126 if (tnl_port_should_receive(&ctx.xin->flow)) {
7127 memset(&wc->masks.tunnel, 0xff, sizeof wc->masks.tunnel);
7130 /* Disable most wildcarding for NetFlow. */
7131 if (xin->ofproto->netflow) {
7132 netflow_mask_wc(&ctx.xin->flow, wc);
7137 ctx.xout->has_learn = false;
7138 ctx.xout->has_normal = false;
7139 ctx.xout->has_fin_timeout = false;
7140 ctx.xout->nf_output_iface = NF_OUT_DROP;
7141 ctx.xout->mirrors = 0;
7143 ofpbuf_use_stub(&ctx.xout->odp_actions, ctx.xout->odp_actions_stub,
7144 sizeof ctx.xout->odp_actions_stub);
7145 ofpbuf_reserve(&ctx.xout->odp_actions, NL_A_U32_SIZE);
7148 ctx.max_resubmit_trigger = false;
7149 ctx.orig_skb_priority = ctx.xin->flow.skb_priority;
7154 ofpacts = xin->ofpacts;
7155 ofpacts_len = xin->ofpacts_len;
7156 } else if (xin->rule) {
7157 ofpacts = xin->rule->up.ofpacts;
7158 ofpacts_len = xin->rule->up.ofpacts_len;
7163 ofpbuf_use_stub(&ctx.stack, ctx.init_stack, sizeof ctx.init_stack);
7165 if (ctx.ofproto->has_mirrors || hit_resubmit_limit) {
7166 /* Do this conditionally because the copy is expensive enough that it
7167 * shows up in profiles. */
7168 orig_flow = ctx.xin->flow;
7171 if (ctx.xin->flow.nw_frag & FLOW_NW_FRAG_ANY) {
7172 switch (ctx.ofproto->up.frag_handling) {
7173 case OFPC_FRAG_NORMAL:
7174 /* We must pretend that transport ports are unavailable. */
7175 ctx.xin->flow.tp_src = ctx.base_flow.tp_src = htons(0);
7176 ctx.xin->flow.tp_dst = ctx.base_flow.tp_dst = htons(0);
7179 case OFPC_FRAG_DROP:
7182 case OFPC_FRAG_REASM:
7185 case OFPC_FRAG_NX_MATCH:
7186 /* Nothing to do. */
7189 case OFPC_INVALID_TTL_TO_CONTROLLER:
7194 in_port = get_ofp_port(ctx.ofproto, ctx.xin->flow.in_port);
7195 special = process_special(&ctx, &ctx.xin->flow, in_port,
7198 ctx.xout->slow = special;
7200 static struct vlog_rate_limit trace_rl = VLOG_RATE_LIMIT_INIT(1, 1);
7201 struct initial_vals initial_vals;
7202 size_t sample_actions_len;
7203 uint32_t local_odp_port;
7205 initial_vals.vlan_tci = ctx.base_flow.vlan_tci;
7206 initial_vals.tunnel_ip_tos = ctx.base_flow.tunnel.ip_tos;
7208 add_sflow_action(&ctx);
7209 add_ipfix_action(&ctx);
7210 sample_actions_len = ctx.xout->odp_actions.size;
7212 if (tunnel_ecn_ok(&ctx) && (!in_port || may_receive(in_port, &ctx))) {
7213 do_xlate_actions(ofpacts, ofpacts_len, &ctx);
7215 /* We've let OFPP_NORMAL and the learning action look at the
7216 * packet, so drop it now if forwarding is disabled. */
7217 if (in_port && !stp_forward_in_state(in_port->stp_state)) {
7218 ctx.xout->odp_actions.size = sample_actions_len;
7222 if (ctx.max_resubmit_trigger && !ctx.xin->resubmit_hook) {
7223 if (!hit_resubmit_limit) {
7224 /* We didn't record the original flow. Make sure we do from
7226 hit_resubmit_limit = true;
7227 } else if (!VLOG_DROP_ERR(&trace_rl)) {
7228 struct ds ds = DS_EMPTY_INITIALIZER;
7230 ofproto_trace(ctx.ofproto, &orig_flow, ctx.xin->packet,
7231 &initial_vals, &ds);
7232 VLOG_ERR("Trace triggered by excessive resubmit "
7233 "recursion:\n%s", ds_cstr(&ds));
7238 local_odp_port = ofp_port_to_odp_port(ctx.ofproto, OFPP_LOCAL);
7239 if (!connmgr_must_output_local(ctx.ofproto->up.connmgr, &ctx.xin->flow,
7241 ctx.xout->odp_actions.data,
7242 ctx.xout->odp_actions.size)) {
7243 compose_output_action(&ctx, OFPP_LOCAL);
7245 if (ctx.ofproto->has_mirrors) {
7246 add_mirror_actions(&ctx, &orig_flow);
7248 fix_sflow_action(&ctx);
7251 ofpbuf_uninit(&ctx.stack);
7253 /* Clear the metadata and register wildcard masks, because we won't
7254 * use non-header fields as part of the cache. */
7255 memset(&ctx.xout->wc.masks.metadata, 0,
7256 sizeof ctx.xout->wc.masks.metadata);
7257 memset(&ctx.xout->wc.masks.regs, 0, sizeof ctx.xout->wc.masks.regs);
7260 /* Translates the 'ofpacts_len' bytes of "struct ofpact"s starting at 'ofpacts'
7261 * into datapath actions, using 'ctx', and discards the datapath actions. */
7263 xlate_actions_for_side_effects(struct xlate_in *xin)
7265 struct xlate_out xout;
7267 xlate_actions(xin, &xout);
7268 xlate_out_uninit(&xout);
7272 xlate_report(struct xlate_ctx *ctx, const char *s)
7274 if (ctx->xin->report_hook) {
7275 ctx->xin->report_hook(ctx, s);
7280 xlate_out_copy(struct xlate_out *dst, const struct xlate_out *src)
7283 dst->tags = src->tags;
7284 dst->slow = src->slow;
7285 dst->has_learn = src->has_learn;
7286 dst->has_normal = src->has_normal;
7287 dst->has_fin_timeout = src->has_fin_timeout;
7288 dst->nf_output_iface = src->nf_output_iface;
7289 dst->mirrors = src->mirrors;
7291 ofpbuf_use_stub(&dst->odp_actions, dst->odp_actions_stub,
7292 sizeof dst->odp_actions_stub);
7293 ofpbuf_put(&dst->odp_actions, src->odp_actions.data,
7294 src->odp_actions.size);
7297 /* OFPP_NORMAL implementation. */
7299 static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *);
7301 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
7302 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle',
7303 * the bundle on which the packet was received, returns the VLAN to which the
7306 * Both 'vid' and the return value are in the range 0...4095. */
7308 input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid)
7310 switch (in_bundle->vlan_mode) {
7311 case PORT_VLAN_ACCESS:
7312 return in_bundle->vlan;
7315 case PORT_VLAN_TRUNK:
7318 case PORT_VLAN_NATIVE_UNTAGGED:
7319 case PORT_VLAN_NATIVE_TAGGED:
7320 return vid ? vid : in_bundle->vlan;
7327 /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'.
7328 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
7331 * 'vid' should be the VID obtained from the 802.1Q header that was received as
7332 * part of a packet (specify 0 if there was no 802.1Q header), in the range
7335 input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn)
7337 /* Allow any VID on the OFPP_NONE port. */
7338 if (in_bundle == &ofpp_none_bundle) {
7342 switch (in_bundle->vlan_mode) {
7343 case PORT_VLAN_ACCESS:
7346 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7347 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
7348 "packet received on port %s configured as VLAN "
7349 "%"PRIu16" access port",
7350 in_bundle->ofproto->up.name, vid,
7351 in_bundle->name, in_bundle->vlan);
7357 case PORT_VLAN_NATIVE_UNTAGGED:
7358 case PORT_VLAN_NATIVE_TAGGED:
7360 /* Port must always carry its native VLAN. */
7364 case PORT_VLAN_TRUNK:
7365 if (!ofbundle_includes_vlan(in_bundle, vid)) {
7367 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7368 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet "
7369 "received on port %s not configured for trunking "
7371 in_bundle->ofproto->up.name, vid,
7372 in_bundle->name, vid);
7384 /* Given 'vlan', the VLAN that a packet belongs to, and
7385 * 'out_bundle', a bundle on which the packet is to be output, returns the VID
7386 * that should be included in the 802.1Q header. (If the return value is 0,
7387 * then the 802.1Q header should only be included in the packet if there is a
7390 * Both 'vlan' and the return value are in the range 0...4095. */
7392 output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan)
7394 switch (out_bundle->vlan_mode) {
7395 case PORT_VLAN_ACCESS:
7398 case PORT_VLAN_TRUNK:
7399 case PORT_VLAN_NATIVE_TAGGED:
7402 case PORT_VLAN_NATIVE_UNTAGGED:
7403 return vlan == out_bundle->vlan ? 0 : vlan;
7411 output_normal(struct xlate_ctx *ctx, const struct ofbundle *out_bundle,
7414 struct ofport_dpif *port;
7416 ovs_be16 tci, old_tci;
7418 vid = output_vlan_to_vid(out_bundle, vlan);
7419 if (!out_bundle->bond) {
7420 port = ofbundle_get_a_port(out_bundle);
7422 port = bond_choose_output_slave(out_bundle->bond, &ctx->xin->flow,
7423 &ctx->xout->wc, vid, &ctx->xout->tags);
7425 /* No slaves enabled, so drop packet. */
7430 old_tci = ctx->xin->flow.vlan_tci;
7432 if (tci || out_bundle->use_priority_tags) {
7433 tci |= ctx->xin->flow.vlan_tci & htons(VLAN_PCP_MASK);
7435 tci |= htons(VLAN_CFI);
7438 ctx->xin->flow.vlan_tci = tci;
7440 compose_output_action(ctx, port->up.ofp_port);
7441 ctx->xin->flow.vlan_tci = old_tci;
7445 mirror_mask_ffs(mirror_mask_t mask)
7447 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
7452 ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
7454 return (bundle->vlan_mode != PORT_VLAN_ACCESS
7455 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
7459 ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan)
7461 return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan);
7464 /* Returns an arbitrary interface within 'bundle'. */
7465 static struct ofport_dpif *
7466 ofbundle_get_a_port(const struct ofbundle *bundle)
7468 return CONTAINER_OF(list_front(&bundle->ports),
7469 struct ofport_dpif, bundle_node);
7473 vlan_is_mirrored(const struct ofmirror *m, int vlan)
7475 return !m->vlans || bitmap_is_set(m->vlans, vlan);
7479 add_mirror_actions(struct xlate_ctx *ctx, const struct flow *orig_flow)
7481 struct ofproto_dpif *ofproto = ctx->ofproto;
7482 mirror_mask_t mirrors;
7483 struct ofbundle *in_bundle;
7486 const struct nlattr *a;
7489 in_bundle = lookup_input_bundle(ctx->ofproto, orig_flow->in_port,
7490 ctx->xin->packet != NULL, NULL);
7494 mirrors = in_bundle->src_mirrors;
7496 /* Drop frames on bundles reserved for mirroring. */
7497 if (in_bundle->mirror_out) {
7498 if (ctx->xin->packet != NULL) {
7499 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7500 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
7501 "%s, which is reserved exclusively for mirroring",
7502 ctx->ofproto->up.name, in_bundle->name);
7508 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
7509 if (!input_vid_is_valid(vid, in_bundle, ctx->xin->packet != NULL)) {
7512 vlan = input_vid_to_vlan(in_bundle, vid);
7514 /* Look at the output ports to check for destination selections. */
7516 NL_ATTR_FOR_EACH (a, left, ctx->xout->odp_actions.data,
7517 ctx->xout->odp_actions.size) {
7518 enum ovs_action_attr type = nl_attr_type(a);
7519 struct ofport_dpif *ofport;
7521 if (type != OVS_ACTION_ATTR_OUTPUT) {
7525 ofport = get_odp_port(ofproto, nl_attr_get_u32(a));
7526 if (ofport && ofport->bundle) {
7527 mirrors |= ofport->bundle->dst_mirrors;
7535 /* Restore the original packet before adding the mirror actions. */
7536 ctx->xin->flow = *orig_flow;
7541 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
7544 ctx->xout->wc.masks.vlan_tci |= htons(VLAN_CFI | VLAN_VID_MASK);
7547 if (!vlan_is_mirrored(m, vlan)) {
7548 mirrors = zero_rightmost_1bit(mirrors);
7552 mirrors &= ~m->dup_mirrors;
7553 ctx->xout->mirrors |= m->dup_mirrors;
7555 output_normal(ctx, m->out, vlan);
7556 } else if (vlan != m->out_vlan
7557 && !eth_addr_is_reserved(orig_flow->dl_dst)) {
7558 struct ofbundle *bundle;
7560 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
7561 if (ofbundle_includes_vlan(bundle, m->out_vlan)
7562 && !bundle->mirror_out) {
7563 output_normal(ctx, bundle, m->out_vlan);
7571 update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors,
7572 uint64_t packets, uint64_t bytes)
7578 for (; mirrors; mirrors = zero_rightmost_1bit(mirrors)) {
7581 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
7584 /* In normal circumstances 'm' will not be NULL. However,
7585 * if mirrors are reconfigured, we can temporarily get out
7586 * of sync in facet_revalidate(). We could "correct" the
7587 * mirror list before reaching here, but doing that would
7588 * not properly account the traffic stats we've currently
7589 * accumulated for previous mirror configuration. */
7593 m->packet_count += packets;
7594 m->byte_count += bytes;
7598 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
7599 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
7600 * indicate this; newer upstream kernels use gratuitous ARP requests. */
7602 is_gratuitous_arp(const struct flow *flow, struct flow_wildcards *wc)
7604 if (flow->dl_type != htons(ETH_TYPE_ARP)) {
7608 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
7609 if (!eth_addr_is_broadcast(flow->dl_dst)) {
7613 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
7614 if (flow->nw_proto == ARP_OP_REPLY) {
7616 } else if (flow->nw_proto == ARP_OP_REQUEST) {
7617 memset(&wc->masks.nw_src, 0xff, sizeof wc->masks.nw_src);
7618 memset(&wc->masks.nw_dst, 0xff, sizeof wc->masks.nw_dst);
7620 return flow->nw_src == flow->nw_dst;
7627 update_learning_table(struct ofproto_dpif *ofproto,
7628 const struct flow *flow, struct flow_wildcards *wc,
7629 int vlan, struct ofbundle *in_bundle)
7631 struct mac_entry *mac;
7633 /* Don't learn the OFPP_NONE port. */
7634 if (in_bundle == &ofpp_none_bundle) {
7638 if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
7642 mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan);
7643 if (is_gratuitous_arp(flow, wc)) {
7644 /* We don't want to learn from gratuitous ARP packets that are
7645 * reflected back over bond slaves so we lock the learning table. */
7646 if (!in_bundle->bond) {
7647 mac_entry_set_grat_arp_lock(mac);
7648 } else if (mac_entry_is_grat_arp_locked(mac)) {
7653 if (mac_entry_is_new(mac) || mac->port.p != in_bundle) {
7654 /* The log messages here could actually be useful in debugging,
7655 * so keep the rate limit relatively high. */
7656 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
7657 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
7658 "on port %s in VLAN %d",
7659 ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src),
7660 in_bundle->name, vlan);
7662 mac->port.p = in_bundle;
7663 tag_set_add(&ofproto->backer->revalidate_set,
7664 mac_learning_changed(ofproto->ml, mac));
7668 static struct ofbundle *
7669 lookup_input_bundle(const struct ofproto_dpif *ofproto, uint16_t in_port,
7670 bool warn, struct ofport_dpif **in_ofportp)
7672 struct ofport_dpif *ofport;
7674 /* Find the port and bundle for the received packet. */
7675 ofport = get_ofp_port(ofproto, in_port);
7677 *in_ofportp = ofport;
7679 if (ofport && ofport->bundle) {
7680 return ofport->bundle;
7683 /* Special-case OFPP_NONE, which a controller may use as the ingress
7684 * port for traffic that it is sourcing. */
7685 if (in_port == OFPP_NONE) {
7686 return &ofpp_none_bundle;
7689 /* Odd. A few possible reasons here:
7691 * - We deleted a port but there are still a few packets queued up
7694 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
7695 * we don't know about.
7697 * - The ofproto client didn't configure the port as part of a bundle.
7698 * This is particularly likely to happen if a packet was received on the
7699 * port after it was created, but before the client had a chance to
7700 * configure its bundle.
7703 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7705 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
7706 "port %"PRIu16, ofproto->up.name, in_port);
7711 /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or
7712 * dropped. Returns true if they may be forwarded, false if they should be
7715 * 'in_port' must be the ofport_dpif that corresponds to flow->in_port.
7716 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
7718 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
7719 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
7720 * checked by input_vid_is_valid().
7722 * May also add tags to '*tags', although the current implementation only does
7723 * so in one special case.
7726 is_admissible(struct xlate_ctx *ctx, struct ofport_dpif *in_port,
7729 struct ofproto_dpif *ofproto = ctx->ofproto;
7730 struct flow *flow = &ctx->xin->flow;
7731 struct ofbundle *in_bundle = in_port->bundle;
7733 /* Drop frames for reserved multicast addresses
7734 * only if forward_bpdu option is absent. */
7735 if (!ofproto->up.forward_bpdu && eth_addr_is_reserved(flow->dl_dst)) {
7736 xlate_report(ctx, "packet has reserved destination MAC, dropping");
7740 if (in_bundle->bond) {
7741 struct mac_entry *mac;
7743 switch (bond_check_admissibility(in_bundle->bond, in_port,
7744 flow->dl_dst, &ctx->xout->tags)) {
7749 xlate_report(ctx, "bonding refused admissibility, dropping");
7752 case BV_DROP_IF_MOVED:
7753 mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL);
7754 if (mac && mac->port.p != in_bundle &&
7755 (!is_gratuitous_arp(flow, &ctx->xout->wc)
7756 || mac_entry_is_grat_arp_locked(mac))) {
7757 xlate_report(ctx, "SLB bond thinks this packet looped back, "
7769 xlate_normal(struct xlate_ctx *ctx)
7771 struct flow_wildcards *wc = &ctx->xout->wc;
7772 struct ofport_dpif *in_port;
7773 struct ofbundle *in_bundle;
7774 struct mac_entry *mac;
7778 ctx->xout->has_normal = true;
7780 memset(&wc->masks.dl_src, 0xff, sizeof wc->masks.dl_src);
7781 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
7782 wc->masks.vlan_tci |= htons(VLAN_VID_MASK | VLAN_CFI);
7784 in_bundle = lookup_input_bundle(ctx->ofproto, ctx->xin->flow.in_port,
7785 ctx->xin->packet != NULL, &in_port);
7787 xlate_report(ctx, "no input bundle, dropping");
7791 /* Drop malformed frames. */
7792 if (ctx->xin->flow.dl_type == htons(ETH_TYPE_VLAN) &&
7793 !(ctx->xin->flow.vlan_tci & htons(VLAN_CFI))) {
7794 if (ctx->xin->packet != NULL) {
7795 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7796 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
7797 "VLAN tag received on port %s",
7798 ctx->ofproto->up.name, in_bundle->name);
7800 xlate_report(ctx, "partial VLAN tag, dropping");
7804 /* Drop frames on bundles reserved for mirroring. */
7805 if (in_bundle->mirror_out) {
7806 if (ctx->xin->packet != NULL) {
7807 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7808 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
7809 "%s, which is reserved exclusively for mirroring",
7810 ctx->ofproto->up.name, in_bundle->name);
7812 xlate_report(ctx, "input port is mirror output port, dropping");
7817 vid = vlan_tci_to_vid(ctx->xin->flow.vlan_tci);
7818 if (!input_vid_is_valid(vid, in_bundle, ctx->xin->packet != NULL)) {
7819 xlate_report(ctx, "disallowed VLAN VID for this input port, dropping");
7822 vlan = input_vid_to_vlan(in_bundle, vid);
7824 /* Check other admissibility requirements. */
7825 if (in_port && !is_admissible(ctx, in_port, vlan)) {
7829 /* Learn source MAC. */
7830 if (ctx->xin->may_learn) {
7831 update_learning_table(ctx->ofproto, &ctx->xin->flow, wc,
7835 /* Determine output bundle. */
7836 mac = mac_learning_lookup(ctx->ofproto->ml, ctx->xin->flow.dl_dst, vlan,
7839 if (mac->port.p != in_bundle) {
7840 xlate_report(ctx, "forwarding to learned port");
7841 output_normal(ctx, mac->port.p, vlan);
7843 xlate_report(ctx, "learned port is input port, dropping");
7846 struct ofbundle *bundle;
7848 xlate_report(ctx, "no learned MAC for destination, flooding");
7849 HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
7850 if (bundle != in_bundle
7851 && ofbundle_includes_vlan(bundle, vlan)
7852 && bundle->floodable
7853 && !bundle->mirror_out) {
7854 output_normal(ctx, bundle, vlan);
7857 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
7861 /* Optimized flow revalidation.
7863 * It's a difficult problem, in general, to tell which facets need to have
7864 * their actions recalculated whenever the OpenFlow flow table changes. We
7865 * don't try to solve that general problem: for most kinds of OpenFlow flow
7866 * table changes, we recalculate the actions for every facet. This is
7867 * relatively expensive, but it's good enough if the OpenFlow flow table
7868 * doesn't change very often.
7870 * However, we can expect one particular kind of OpenFlow flow table change to
7871 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
7872 * of CPU on revalidating every facet whenever MAC learning modifies the flow
7873 * table, we add a special case that applies to flow tables in which every rule
7874 * has the same form (that is, the same wildcards), except that the table is
7875 * also allowed to have a single "catch-all" flow that matches all packets. We
7876 * optimize this case by tagging all of the facets that resubmit into the table
7877 * and invalidating the same tag whenever a flow changes in that table. The
7878 * end result is that we revalidate just the facets that need it (and sometimes
7879 * a few more, but not all of the facets or even all of the facets that
7880 * resubmit to the table modified by MAC learning). */
7882 /* Calculates the tag to use for 'flow' and mask 'mask' when it is inserted
7883 * into an OpenFlow table with the given 'basis'. */
7885 rule_calculate_tag(const struct flow *flow, const struct minimask *mask,
7888 if (minimask_is_catchall(mask)) {
7891 uint32_t hash = flow_hash_in_minimask(flow, mask, secret);
7892 return tag_create_deterministic(hash);
7896 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
7897 * taggability of that table.
7899 * This function must be called after *each* change to a flow table. If you
7900 * skip calling it on some changes then the pointer comparisons at the end can
7901 * be invalid if you get unlucky. For example, if a flow removal causes a
7902 * cls_table to be destroyed and then a flow insertion causes a cls_table with
7903 * different wildcards to be created with the same address, then this function
7904 * will incorrectly skip revalidation. */
7906 table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
7908 struct table_dpif *table = &ofproto->tables[table_id];
7909 const struct oftable *oftable = &ofproto->up.tables[table_id];
7910 struct cls_table *catchall, *other;
7911 struct cls_table *t;
7913 catchall = other = NULL;
7915 switch (hmap_count(&oftable->cls.tables)) {
7917 /* We could tag this OpenFlow table but it would make the logic a
7918 * little harder and it's a corner case that doesn't seem worth it
7924 HMAP_FOR_EACH (t, hmap_node, &oftable->cls.tables) {
7925 if (cls_table_is_catchall(t)) {
7927 } else if (!other) {
7930 /* Indicate that we can't tag this by setting both tables to
7931 * NULL. (We know that 'catchall' is already NULL.) */
7938 /* Can't tag this table. */
7942 if (table->catchall_table != catchall || table->other_table != other) {
7943 table->catchall_table = catchall;
7944 table->other_table = other;
7945 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
7949 /* Given 'rule' that has changed in some way (either it is a rule being
7950 * inserted, a rule being deleted, or a rule whose actions are being
7951 * modified), marks facets for revalidation to ensure that packets will be
7952 * forwarded correctly according to the new state of the flow table.
7954 * This function must be called after *each* change to a flow table. See
7955 * the comment on table_update_taggable() for more information. */
7957 rule_invalidate(const struct rule_dpif *rule)
7959 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
7961 table_update_taggable(ofproto, rule->up.table_id);
7963 if (!ofproto->backer->need_revalidate) {
7964 struct table_dpif *table = &ofproto->tables[rule->up.table_id];
7966 if (table->other_table && rule->tag) {
7967 tag_set_add(&ofproto->backer->revalidate_set, rule->tag);
7969 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
7975 set_frag_handling(struct ofproto *ofproto_,
7976 enum ofp_config_flags frag_handling)
7978 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7979 if (frag_handling != OFPC_FRAG_REASM) {
7980 ofproto->backer->need_revalidate = REV_RECONFIGURE;
7988 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
7989 const struct flow *flow,
7990 const struct ofpact *ofpacts, size_t ofpacts_len)
7992 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7993 struct initial_vals initial_vals;
7994 struct odputil_keybuf keybuf;
7995 struct dpif_flow_stats stats;
7996 struct xlate_out xout;
7997 struct xlate_in xin;
8001 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
8002 odp_flow_key_from_flow(&key, flow,
8003 ofp_port_to_odp_port(ofproto, flow->in_port));
8005 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
8007 initial_vals.vlan_tci = flow->vlan_tci;
8008 initial_vals.tunnel_ip_tos = 0;
8009 xlate_in_init(&xin, ofproto, flow, &initial_vals, NULL, stats.tcp_flags,
8011 xin.resubmit_stats = &stats;
8012 xin.ofpacts_len = ofpacts_len;
8013 xin.ofpacts = ofpacts;
8015 xlate_actions(&xin, &xout);
8016 dpif_execute(ofproto->backer->dpif, key.data, key.size,
8017 xout.odp_actions.data, xout.odp_actions.size, packet);
8018 xlate_out_uninit(&xout);
8026 set_netflow(struct ofproto *ofproto_,
8027 const struct netflow_options *netflow_options)
8029 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
8031 if (netflow_options) {
8032 if (!ofproto->netflow) {
8033 ofproto->netflow = netflow_create();
8035 return netflow_set_options(ofproto->netflow, netflow_options);
8037 netflow_destroy(ofproto->netflow);
8038 ofproto->netflow = NULL;
8044 get_netflow_ids(const struct ofproto *ofproto_,
8045 uint8_t *engine_type, uint8_t *engine_id)
8047 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
8049 dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
8053 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
8055 if (!facet_is_controller_flow(facet) &&
8056 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
8057 struct subfacet *subfacet;
8058 struct ofexpired expired;
8060 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
8061 if (subfacet->path == SF_FAST_PATH) {
8062 struct dpif_flow_stats stats;
8064 subfacet_install(subfacet, &facet->xout.odp_actions,
8066 subfacet_update_stats(subfacet, &stats);
8070 expired.flow = facet->flow;
8071 expired.packet_count = facet->packet_count;
8072 expired.byte_count = facet->byte_count;
8073 expired.used = facet->used;
8074 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
8079 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
8081 struct cls_cursor cursor;
8082 struct facet *facet;
8084 cls_cursor_init(&cursor, &ofproto->facets, NULL);
8085 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
8086 send_active_timeout(ofproto, facet);
8090 static struct ofproto_dpif *
8091 ofproto_dpif_lookup(const char *name)
8093 struct ofproto_dpif *ofproto;
8095 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
8096 hash_string(name, 0), &all_ofproto_dpifs) {
8097 if (!strcmp(ofproto->up.name, name)) {
8105 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
8106 const char *argv[], void *aux OVS_UNUSED)
8108 struct ofproto_dpif *ofproto;
8111 ofproto = ofproto_dpif_lookup(argv[1]);
8113 unixctl_command_reply_error(conn, "no such bridge");
8116 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
8118 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
8119 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
8123 unixctl_command_reply(conn, "table successfully flushed");
8127 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
8128 const char *argv[], void *aux OVS_UNUSED)
8130 struct ds ds = DS_EMPTY_INITIALIZER;
8131 const struct ofproto_dpif *ofproto;
8132 const struct mac_entry *e;
8134 ofproto = ofproto_dpif_lookup(argv[1]);
8136 unixctl_command_reply_error(conn, "no such bridge");
8140 ds_put_cstr(&ds, " port VLAN MAC Age\n");
8141 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
8142 struct ofbundle *bundle = e->port.p;
8143 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
8144 ofbundle_get_a_port(bundle)->odp_port,
8145 e->vlan, ETH_ADDR_ARGS(e->mac),
8146 mac_entry_age(ofproto->ml, e));
8148 unixctl_command_reply(conn, ds_cstr(&ds));
8153 struct xlate_out xout;
8154 struct xlate_in xin;
8160 trace_format_rule(struct ds *result, uint8_t table_id, int level,
8161 const struct rule_dpif *rule)
8163 ds_put_char_multiple(result, '\t', level);
8165 ds_put_cstr(result, "No match\n");
8169 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
8170 table_id, ntohll(rule->up.flow_cookie));
8171 cls_rule_format(&rule->up.cr, result);
8172 ds_put_char(result, '\n');
8174 ds_put_char_multiple(result, '\t', level);
8175 ds_put_cstr(result, "OpenFlow ");
8176 ofpacts_format(rule->up.ofpacts, rule->up.ofpacts_len, result);
8177 ds_put_char(result, '\n');
8181 trace_format_flow(struct ds *result, int level, const char *title,
8182 struct trace_ctx *trace)
8184 ds_put_char_multiple(result, '\t', level);
8185 ds_put_format(result, "%s: ", title);
8186 if (flow_equal(&trace->xin.flow, &trace->flow)) {
8187 ds_put_cstr(result, "unchanged");
8189 flow_format(result, &trace->xin.flow);
8190 trace->flow = trace->xin.flow;
8192 ds_put_char(result, '\n');
8196 trace_format_regs(struct ds *result, int level, const char *title,
8197 struct trace_ctx *trace)
8201 ds_put_char_multiple(result, '\t', level);
8202 ds_put_format(result, "%s:", title);
8203 for (i = 0; i < FLOW_N_REGS; i++) {
8204 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
8206 ds_put_char(result, '\n');
8210 trace_format_odp(struct ds *result, int level, const char *title,
8211 struct trace_ctx *trace)
8213 struct ofpbuf *odp_actions = &trace->xout.odp_actions;
8215 ds_put_char_multiple(result, '\t', level);
8216 ds_put_format(result, "%s: ", title);
8217 format_odp_actions(result, odp_actions->data, odp_actions->size);
8218 ds_put_char(result, '\n');
8222 trace_resubmit(struct xlate_ctx *ctx, struct rule_dpif *rule)
8224 struct trace_ctx *trace = CONTAINER_OF(ctx->xin, struct trace_ctx, xin);
8225 struct ds *result = trace->result;
8227 ds_put_char(result, '\n');
8228 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
8229 trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace);
8230 trace_format_odp(result, ctx->recurse + 1, "Resubmitted odp", trace);
8231 trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule);
8235 trace_report(struct xlate_ctx *ctx, const char *s)
8237 struct trace_ctx *trace = CONTAINER_OF(ctx->xin, struct trace_ctx, xin);
8238 struct ds *result = trace->result;
8240 ds_put_char_multiple(result, '\t', ctx->recurse);
8241 ds_put_cstr(result, s);
8242 ds_put_char(result, '\n');
8246 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
8247 void *aux OVS_UNUSED)
8249 const char *dpname = argv[1];
8250 struct ofproto_dpif *ofproto;
8251 struct ofpbuf odp_key, odp_mask;
8252 struct ofpbuf *packet;
8253 struct initial_vals initial_vals;
8259 ofpbuf_init(&odp_key, 0);
8261 ofpbuf_init(&odp_mask, 0);
8263 ofproto = ofproto_dpif_lookup(dpname);
8265 unixctl_command_reply_error(conn, "Unknown ofproto (use ofproto/list "
8269 if (argc == 3 || (argc == 4 && !strcmp(argv[3], "-generate"))) {
8270 /* ofproto/trace dpname flow [-generate] */
8271 const char *flow_s = argv[2];
8272 const char *generate_s = argv[3];
8274 /* Allow 'flow_s' to be either a datapath flow or an OpenFlow-like
8275 * flow. We guess which type it is based on whether 'flow_s' contains
8276 * an '(', since a datapath flow always contains '(') but an
8277 * OpenFlow-like flow should not (in fact it's allowed but I believe
8278 * that's not documented anywhere).
8280 * An alternative would be to try to parse 'flow_s' both ways, but then
8281 * it would be tricky giving a sensible error message. After all, do
8282 * you just say "syntax error" or do you present both error messages?
8283 * Both choices seem lousy. */
8284 if (strchr(flow_s, '(')) {
8287 /* Convert string to datapath key. */
8288 ofpbuf_init(&odp_key, 0);
8289 error = odp_flow_from_string(flow_s, NULL, &odp_key, &odp_mask);
8291 unixctl_command_reply_error(conn, "Bad flow syntax");
8295 /* The user might have specified the wrong ofproto but within the
8296 * same backer. That's OK, ofproto_receive() can find the right
8298 if (ofproto_receive(ofproto->backer, NULL, odp_key.data,
8299 odp_key.size, &flow, NULL, &ofproto, NULL,
8301 unixctl_command_reply_error(conn, "Invalid flow");
8304 ds_put_format(&result, "Bridge: %s\n", ofproto->up.name);
8308 error_s = parse_ofp_exact_flow(&flow, argv[2]);
8310 unixctl_command_reply_error(conn, error_s);
8315 initial_vals.vlan_tci = flow.vlan_tci;
8316 initial_vals.tunnel_ip_tos = flow.tunnel.ip_tos;
8319 /* Generate a packet, if requested. */
8321 packet = ofpbuf_new(0);
8322 flow_compose(packet, &flow);
8324 } else if (argc == 7) {
8325 /* ofproto/trace dpname priority tun_id in_port mark packet */
8326 const char *priority_s = argv[2];
8327 const char *tun_id_s = argv[3];
8328 const char *in_port_s = argv[4];
8329 const char *mark_s = argv[5];
8330 const char *packet_s = argv[6];
8331 uint32_t in_port = atoi(in_port_s);
8332 ovs_be64 tun_id = htonll(strtoull(tun_id_s, NULL, 0));
8333 uint32_t priority = atoi(priority_s);
8334 uint32_t mark = atoi(mark_s);
8337 msg = eth_from_hex(packet_s, &packet);
8339 unixctl_command_reply_error(conn, msg);
8343 ds_put_cstr(&result, "Packet: ");
8344 s = ofp_packet_to_string(packet->data, packet->size);
8345 ds_put_cstr(&result, s);
8348 flow_extract(packet, priority, mark, NULL, in_port, &flow);
8349 flow.tunnel.tun_id = tun_id;
8350 initial_vals.vlan_tci = flow.vlan_tci;
8351 initial_vals.tunnel_ip_tos = flow.tunnel.ip_tos;
8353 unixctl_command_reply_error(conn, "Bad command syntax");
8357 ofproto_trace(ofproto, &flow, packet, &initial_vals, &result);
8358 unixctl_command_reply(conn, ds_cstr(&result));
8361 ds_destroy(&result);
8362 ofpbuf_delete(packet);
8363 ofpbuf_uninit(&odp_key);
8364 ofpbuf_uninit(&odp_mask);
8368 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
8369 const struct ofpbuf *packet,
8370 const struct initial_vals *initial_vals, struct ds *ds)
8372 struct rule_dpif *rule;
8374 ds_put_cstr(ds, "Flow: ");
8375 flow_format(ds, flow);
8376 ds_put_char(ds, '\n');
8378 rule = rule_dpif_lookup(ofproto, flow, NULL);
8380 trace_format_rule(ds, 0, 0, rule);
8381 if (rule == ofproto->miss_rule) {
8382 ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
8383 } else if (rule == ofproto->no_packet_in_rule) {
8384 ds_put_cstr(ds, "\nNo match, packets dropped because "
8385 "OFPPC_NO_PACKET_IN is set on in_port.\n");
8386 } else if (rule == ofproto->drop_frags_rule) {
8387 ds_put_cstr(ds, "\nPackets dropped because they are IP fragments "
8388 "and the fragment handling mode is \"drop\".\n");
8392 uint64_t odp_actions_stub[1024 / 8];
8393 struct ofpbuf odp_actions;
8394 struct trace_ctx trace;
8398 tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
8401 ofpbuf_use_stub(&odp_actions,
8402 odp_actions_stub, sizeof odp_actions_stub);
8403 xlate_in_init(&trace.xin, ofproto, flow, initial_vals, rule, tcp_flags,
8405 trace.xin.resubmit_hook = trace_resubmit;
8406 trace.xin.report_hook = trace_report;
8408 xlate_actions(&trace.xin, &trace.xout);
8410 ds_put_char(ds, '\n');
8411 trace_format_flow(ds, 0, "Final flow", &trace);
8413 match_init(&match, flow, &trace.xout.wc);
8414 ds_put_cstr(ds, "Relevant fields: ");
8415 match_format(&match, ds, OFP_DEFAULT_PRIORITY);
8416 ds_put_char(ds, '\n');
8418 ds_put_cstr(ds, "Datapath actions: ");
8419 format_odp_actions(ds, trace.xout.odp_actions.data,
8420 trace.xout.odp_actions.size);
8422 if (trace.xout.slow) {
8423 ds_put_cstr(ds, "\nThis flow is handled by the userspace "
8424 "slow path because it:");
8425 switch (trace.xout.slow) {
8427 ds_put_cstr(ds, "\n\t- Consists of CFM packets.");
8430 ds_put_cstr(ds, "\n\t- Consists of LACP packets.");
8433 ds_put_cstr(ds, "\n\t- Consists of STP packets.");
8435 case SLOW_CONTROLLER:
8436 ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages "
8437 "to the OpenFlow controller.");
8444 xlate_out_uninit(&trace.xout);
8449 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
8450 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
8453 unixctl_command_reply(conn, NULL);
8457 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
8458 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
8461 unixctl_command_reply(conn, NULL);
8464 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
8465 * 'reply' describing the results. */
8467 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
8469 struct cls_cursor cursor;
8470 struct facet *facet;
8474 cls_cursor_init(&cursor, &ofproto->facets, NULL);
8475 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
8476 if (!facet_check_consistency(facet)) {
8481 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
8485 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
8486 ofproto->up.name, errors);
8488 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
8493 ofproto_dpif_self_check(struct unixctl_conn *conn,
8494 int argc, const char *argv[], void *aux OVS_UNUSED)
8496 struct ds reply = DS_EMPTY_INITIALIZER;
8497 struct ofproto_dpif *ofproto;
8500 ofproto = ofproto_dpif_lookup(argv[1]);
8502 unixctl_command_reply_error(conn, "Unknown ofproto (use "
8503 "ofproto/list for help)");
8506 ofproto_dpif_self_check__(ofproto, &reply);
8508 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
8509 ofproto_dpif_self_check__(ofproto, &reply);
8513 unixctl_command_reply(conn, ds_cstr(&reply));
8517 /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list
8518 * of the 'ofproto_shash' nodes. It is the responsibility of the caller
8519 * to destroy 'ofproto_shash' and free the returned value. */
8520 static const struct shash_node **
8521 get_ofprotos(struct shash *ofproto_shash)
8523 const struct ofproto_dpif *ofproto;
8525 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
8526 char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
8527 shash_add_nocopy(ofproto_shash, name, ofproto);
8530 return shash_sort(ofproto_shash);
8534 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
8535 const char *argv[] OVS_UNUSED,
8536 void *aux OVS_UNUSED)
8538 struct ds ds = DS_EMPTY_INITIALIZER;
8539 struct shash ofproto_shash;
8540 const struct shash_node **sorted_ofprotos;
8543 shash_init(&ofproto_shash);
8544 sorted_ofprotos = get_ofprotos(&ofproto_shash);
8545 for (i = 0; i < shash_count(&ofproto_shash); i++) {
8546 const struct shash_node *node = sorted_ofprotos[i];
8547 ds_put_format(&ds, "%s\n", node->name);
8550 shash_destroy(&ofproto_shash);
8551 free(sorted_ofprotos);
8553 unixctl_command_reply(conn, ds_cstr(&ds));
8558 show_dp_format(const struct ofproto_dpif *ofproto, struct ds *ds)
8560 const struct shash_node **ports;
8562 struct avg_subfacet_rates lifetime;
8563 unsigned long long int minutes;
8564 const int min_ms = 60 * 1000; /* milliseconds in one minute. */
8566 minutes = (time_msec() - ofproto->created) / min_ms;
8569 lifetime.add_rate = (double)ofproto->total_subfacet_add_count
8571 lifetime.del_rate = (double)ofproto->total_subfacet_del_count
8574 lifetime.add_rate = 0.0;
8575 lifetime.del_rate = 0.0;
8578 ds_put_format(ds, "%s (%s):\n", ofproto->up.name,
8579 dpif_name(ofproto->backer->dpif));
8581 "\tlookups: hit:%"PRIu64" missed:%"PRIu64"\n",
8582 ofproto->n_hit, ofproto->n_missed);
8583 ds_put_format(ds, "\tflows: cur: %zu, avg: %5.3f, max: %d,"
8584 " life span: %llu(ms)\n",
8585 hmap_count(&ofproto->subfacets),
8586 avg_subfacet_count(ofproto),
8587 ofproto->max_n_subfacet,
8588 avg_subfacet_life_span(ofproto));
8589 if (minutes >= 60) {
8590 show_dp_rates(ds, "\t\thourly avg:", &ofproto->hourly);
8592 if (minutes >= 60 * 24) {
8593 show_dp_rates(ds, "\t\tdaily avg:", &ofproto->daily);
8595 show_dp_rates(ds, "\t\toverall avg:", &lifetime);
8597 ports = shash_sort(&ofproto->up.port_by_name);
8598 for (i = 0; i < shash_count(&ofproto->up.port_by_name); i++) {
8599 const struct shash_node *node = ports[i];
8600 struct ofport *ofport = node->data;
8601 const char *name = netdev_get_name(ofport->netdev);
8602 const char *type = netdev_get_type(ofport->netdev);
8605 ds_put_format(ds, "\t%s %u/", name, ofport->ofp_port);
8607 odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
8608 if (odp_port != OVSP_NONE) {
8609 ds_put_format(ds, "%"PRIu32":", odp_port);
8611 ds_put_cstr(ds, "none:");
8614 if (strcmp(type, "system")) {
8615 struct netdev *netdev;
8618 ds_put_format(ds, " (%s", type);
8620 error = netdev_open(name, type, &netdev);
8625 error = netdev_get_config(netdev, &config);
8627 const struct smap_node **nodes;
8630 nodes = smap_sort(&config);
8631 for (i = 0; i < smap_count(&config); i++) {
8632 const struct smap_node *node = nodes[i];
8633 ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
8634 node->key, node->value);
8638 smap_destroy(&config);
8640 netdev_close(netdev);
8642 ds_put_char(ds, ')');
8644 ds_put_char(ds, '\n');
8650 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc,
8651 const char *argv[], void *aux OVS_UNUSED)
8653 struct ds ds = DS_EMPTY_INITIALIZER;
8654 const struct ofproto_dpif *ofproto;
8658 for (i = 1; i < argc; i++) {
8659 ofproto = ofproto_dpif_lookup(argv[i]);
8661 ds_put_format(&ds, "Unknown bridge %s (use dpif/dump-dps "
8662 "for help)", argv[i]);
8663 unixctl_command_reply_error(conn, ds_cstr(&ds));
8666 show_dp_format(ofproto, &ds);
8669 struct shash ofproto_shash;
8670 const struct shash_node **sorted_ofprotos;
8673 shash_init(&ofproto_shash);
8674 sorted_ofprotos = get_ofprotos(&ofproto_shash);
8675 for (i = 0; i < shash_count(&ofproto_shash); i++) {
8676 const struct shash_node *node = sorted_ofprotos[i];
8677 show_dp_format(node->data, &ds);
8680 shash_destroy(&ofproto_shash);
8681 free(sorted_ofprotos);
8684 unixctl_command_reply(conn, ds_cstr(&ds));
8688 /* Dump the megaflow (facet) cache. This is useful to check the
8689 * correctness of flow wildcarding, since the same mechanism is used for
8690 * both xlate caching and kernel wildcarding.
8692 * It's important to note that in the output the flow description uses
8693 * OpenFlow (OFP) ports, but the actions use datapath (ODP) ports.
8695 * This command is only needed for advanced debugging, so it's not
8696 * documented in the man page. */
8698 ofproto_unixctl_dpif_dump_megaflows(struct unixctl_conn *conn,
8699 int argc OVS_UNUSED, const char *argv[],
8700 void *aux OVS_UNUSED)
8702 struct ds ds = DS_EMPTY_INITIALIZER;
8703 const struct ofproto_dpif *ofproto;
8704 long long int now = time_msec();
8705 struct cls_cursor cursor;
8706 struct facet *facet;
8708 ofproto = ofproto_dpif_lookup(argv[1]);
8710 unixctl_command_reply_error(conn, "no such bridge");
8714 cls_cursor_init(&cursor, &ofproto->facets, NULL);
8715 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
8716 cls_rule_format(&facet->cr, &ds);
8717 ds_put_cstr(&ds, ", ");
8718 ds_put_format(&ds, "n_subfacets:%zu, ", list_size(&facet->subfacets));
8719 ds_put_format(&ds, "used:%.3fs, ", (now - facet->used) / 1000.0);
8720 ds_put_cstr(&ds, "Datapath actions: ");
8721 if (facet->xout.slow) {
8722 uint64_t slow_path_stub[128 / 8];
8723 const struct nlattr *actions;
8726 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
8727 slow_path_stub, sizeof slow_path_stub,
8728 &actions, &actions_len);
8729 format_odp_actions(&ds, actions, actions_len);
8731 format_odp_actions(&ds, facet->xout.odp_actions.data,
8732 facet->xout.odp_actions.size);
8734 ds_put_cstr(&ds, "\n");
8737 ds_chomp(&ds, '\n');
8738 unixctl_command_reply(conn, ds_cstr(&ds));
8742 /* Disable using the megaflows.
8744 * This command is only needed for advanced debugging, so it's not
8745 * documented in the man page. */
8747 ofproto_unixctl_dpif_disable_megaflows(struct unixctl_conn *conn,
8748 int argc OVS_UNUSED,
8749 const char *argv[] OVS_UNUSED,
8750 void *aux OVS_UNUSED)
8752 struct ofproto_dpif *ofproto;
8754 enable_megaflows = false;
8756 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
8757 flush(&ofproto->up);
8760 unixctl_command_reply(conn, "megaflows disabled");
8763 /* Re-enable using megaflows.
8765 * This command is only needed for advanced debugging, so it's not
8766 * documented in the man page. */
8768 ofproto_unixctl_dpif_enable_megaflows(struct unixctl_conn *conn,
8769 int argc OVS_UNUSED,
8770 const char *argv[] OVS_UNUSED,
8771 void *aux OVS_UNUSED)
8773 struct ofproto_dpif *ofproto;
8775 enable_megaflows = true;
8777 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
8778 flush(&ofproto->up);
8781 unixctl_command_reply(conn, "megaflows enabled");
8785 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
8786 int argc OVS_UNUSED, const char *argv[],
8787 void *aux OVS_UNUSED)
8789 struct ds ds = DS_EMPTY_INITIALIZER;
8790 const struct ofproto_dpif *ofproto;
8791 struct subfacet *subfacet;
8793 ofproto = ofproto_dpif_lookup(argv[1]);
8795 unixctl_command_reply_error(conn, "no such bridge");
8799 update_stats(ofproto->backer);
8801 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
8802 struct facet *facet = subfacet->facet;
8804 odp_flow_key_format(subfacet->key, subfacet->key_len, &ds);
8806 ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:",
8807 subfacet->dp_packet_count, subfacet->dp_byte_count);
8808 if (subfacet->used) {
8809 ds_put_format(&ds, "%.3fs",
8810 (time_msec() - subfacet->used) / 1000.0);
8812 ds_put_format(&ds, "never");
8814 if (subfacet->facet->tcp_flags) {
8815 ds_put_cstr(&ds, ", flags:");
8816 packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags);
8819 ds_put_cstr(&ds, ", actions:");
8820 if (facet->xout.slow) {
8821 uint64_t slow_path_stub[128 / 8];
8822 const struct nlattr *actions;
8825 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
8826 slow_path_stub, sizeof slow_path_stub,
8827 &actions, &actions_len);
8828 format_odp_actions(&ds, actions, actions_len);
8830 format_odp_actions(&ds, facet->xout.odp_actions.data,
8831 facet->xout.odp_actions.size);
8833 ds_put_char(&ds, '\n');
8836 unixctl_command_reply(conn, ds_cstr(&ds));
8841 ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn,
8842 int argc OVS_UNUSED, const char *argv[],
8843 void *aux OVS_UNUSED)
8845 struct ds ds = DS_EMPTY_INITIALIZER;
8846 struct ofproto_dpif *ofproto;
8848 ofproto = ofproto_dpif_lookup(argv[1]);
8850 unixctl_command_reply_error(conn, "no such bridge");
8854 flush(&ofproto->up);
8856 unixctl_command_reply(conn, ds_cstr(&ds));
8861 ofproto_dpif_unixctl_init(void)
8863 static bool registered;
8869 unixctl_command_register(
8871 "bridge {priority tun_id in_port mark packet | odp_flow [-generate]}",
8872 2, 6, ofproto_unixctl_trace, NULL);
8873 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
8874 ofproto_unixctl_fdb_flush, NULL);
8875 unixctl_command_register("fdb/show", "bridge", 1, 1,
8876 ofproto_unixctl_fdb_show, NULL);
8877 unixctl_command_register("ofproto/clog", "", 0, 0,
8878 ofproto_dpif_clog, NULL);
8879 unixctl_command_register("ofproto/unclog", "", 0, 0,
8880 ofproto_dpif_unclog, NULL);
8881 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
8882 ofproto_dpif_self_check, NULL);
8883 unixctl_command_register("dpif/dump-dps", "", 0, 0,
8884 ofproto_unixctl_dpif_dump_dps, NULL);
8885 unixctl_command_register("dpif/show", "[bridge]", 0, INT_MAX,
8886 ofproto_unixctl_dpif_show, NULL);
8887 unixctl_command_register("dpif/dump-flows", "bridge", 1, 1,
8888 ofproto_unixctl_dpif_dump_flows, NULL);
8889 unixctl_command_register("dpif/del-flows", "bridge", 1, 1,
8890 ofproto_unixctl_dpif_del_flows, NULL);
8891 unixctl_command_register("dpif/dump-megaflows", "bridge", 1, 1,
8892 ofproto_unixctl_dpif_dump_megaflows, NULL);
8893 unixctl_command_register("dpif/disable-megaflows", "", 0, 0,
8894 ofproto_unixctl_dpif_disable_megaflows, NULL);
8895 unixctl_command_register("dpif/enable-megaflows", "", 0, 0,
8896 ofproto_unixctl_dpif_enable_megaflows, NULL);
8899 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
8901 * This is deprecated. It is only for compatibility with broken device drivers
8902 * in old versions of Linux that do not properly support VLANs when VLAN
8903 * devices are not used. When broken device drivers are no longer in
8904 * widespread use, we will delete these interfaces. */
8907 set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid)
8909 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
8910 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
8912 if (realdev_ofp_port == ofport->realdev_ofp_port
8913 && vid == ofport->vlandev_vid) {
8917 ofproto->backer->need_revalidate = REV_RECONFIGURE;
8919 if (ofport->realdev_ofp_port) {
8922 if (realdev_ofp_port && ofport->bundle) {
8923 /* vlandevs are enslaved to their realdevs, so they are not allowed to
8924 * themselves be part of a bundle. */
8925 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
8928 ofport->realdev_ofp_port = realdev_ofp_port;
8929 ofport->vlandev_vid = vid;
8931 if (realdev_ofp_port) {
8932 vsp_add(ofport, realdev_ofp_port, vid);
8939 hash_realdev_vid(uint16_t realdev_ofp_port, int vid)
8941 return hash_2words(realdev_ofp_port, vid);
8944 /* Returns the OFP port number of the Linux VLAN device that corresponds to
8945 * 'vlan_tci' on the network device with port number 'realdev_ofp_port' in
8946 * 'struct ofport_dpif'. For example, given 'realdev_ofp_port' of eth0 and
8947 * 'vlan_tci' 9, it would return the port number of eth0.9.
8949 * Unless VLAN splinters are enabled for port 'realdev_ofp_port', this
8950 * function just returns its 'realdev_ofp_port' argument. */
8952 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
8953 uint16_t realdev_ofp_port, ovs_be16 vlan_tci)
8955 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
8956 int vid = vlan_tci_to_vid(vlan_tci);
8957 const struct vlan_splinter *vsp;
8959 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
8960 hash_realdev_vid(realdev_ofp_port, vid),
8961 &ofproto->realdev_vid_map) {
8962 if (vsp->realdev_ofp_port == realdev_ofp_port
8963 && vsp->vid == vid) {
8964 return vsp->vlandev_ofp_port;
8968 return realdev_ofp_port;
8971 static struct vlan_splinter *
8972 vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port)
8974 struct vlan_splinter *vsp;
8976 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, hash_int(vlandev_ofp_port, 0),
8977 &ofproto->vlandev_map) {
8978 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
8986 /* Returns the OpenFlow port number of the "real" device underlying the Linux
8987 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
8988 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
8989 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
8990 * eth0 and store 9 in '*vid'.
8992 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
8993 * VLAN device. Unless VLAN splinters are enabled, this is what this function
8996 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
8997 uint16_t vlandev_ofp_port, int *vid)
8999 if (!hmap_is_empty(&ofproto->vlandev_map)) {
9000 const struct vlan_splinter *vsp;
9002 vsp = vlandev_find(ofproto, vlandev_ofp_port);
9007 return vsp->realdev_ofp_port;
9013 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
9014 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
9015 * 'flow->in_port' to the "real" device backing the VLAN device, sets
9016 * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is
9017 * always the case unless VLAN splinters are enabled), returns false without
9018 * making any changes. */
9020 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
9025 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port, &vid);
9030 /* Cause the flow to be processed as if it came in on the real device with
9031 * the VLAN device's VLAN ID. */
9032 flow->in_port = realdev;
9033 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
9038 vsp_remove(struct ofport_dpif *port)
9040 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
9041 struct vlan_splinter *vsp;
9043 vsp = vlandev_find(ofproto, port->up.ofp_port);
9045 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
9046 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
9049 port->realdev_ofp_port = 0;
9051 VLOG_ERR("missing vlan device record");
9056 vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid)
9058 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
9060 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
9061 && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid))
9062 == realdev_ofp_port)) {
9063 struct vlan_splinter *vsp;
9065 vsp = xmalloc(sizeof *vsp);
9066 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
9067 hash_int(port->up.ofp_port, 0));
9068 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
9069 hash_realdev_vid(realdev_ofp_port, vid));
9070 vsp->realdev_ofp_port = realdev_ofp_port;
9071 vsp->vlandev_ofp_port = port->up.ofp_port;
9074 port->realdev_ofp_port = realdev_ofp_port;
9076 VLOG_ERR("duplicate vlan device record");
9081 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
9083 const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
9084 return ofport ? ofport->odp_port : OVSP_NONE;
9087 static struct ofport_dpif *
9088 odp_port_to_ofport(const struct dpif_backer *backer, uint32_t odp_port)
9090 struct ofport_dpif *port;
9092 HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node,
9093 hash_int(odp_port, 0),
9094 &backer->odp_to_ofport_map) {
9095 if (port->odp_port == odp_port) {
9104 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
9106 struct ofport_dpif *port;
9108 port = odp_port_to_ofport(ofproto->backer, odp_port);
9109 if (port && &ofproto->up == port->up.ofproto) {
9110 return port->up.ofp_port;
9115 static unsigned long long int
9116 avg_subfacet_life_span(const struct ofproto_dpif *ofproto)
9118 unsigned long long int dc;
9119 unsigned long long int avg;
9121 dc = ofproto->total_subfacet_del_count + ofproto->subfacet_del_count;
9122 avg = dc ? ofproto->total_subfacet_life_span / dc : 0;
9128 avg_subfacet_count(const struct ofproto_dpif *ofproto)
9132 if (ofproto->n_update_stats) {
9133 avg_c = (double)ofproto->total_subfacet_count
9134 / ofproto->n_update_stats;
9141 show_dp_rates(struct ds *ds, const char *heading,
9142 const struct avg_subfacet_rates *rates)
9144 ds_put_format(ds, "%s add rate: %5.3f/min, del rate: %5.3f/min\n",
9145 heading, rates->add_rate, rates->del_rate);
9149 update_max_subfacet_count(struct ofproto_dpif *ofproto)
9151 ofproto->max_n_subfacet = MAX(ofproto->max_n_subfacet,
9152 hmap_count(&ofproto->subfacets));
9155 /* Compute exponentially weighted moving average, adding 'new' as the newest,
9156 * most heavily weighted element. 'base' designates the rate of decay: after
9157 * 'base' further updates, 'new''s weight in the EWMA decays to about 1/e
9160 exp_mavg(double *avg, int base, double new)
9162 *avg = (*avg * (base - 1) + new) / base;
9166 update_moving_averages(struct ofproto_dpif *ofproto)
9168 const int min_ms = 60 * 1000; /* milliseconds in one minute. */
9170 /* Update hourly averages on the minute boundaries. */
9171 if (time_msec() - ofproto->last_minute >= min_ms) {
9172 exp_mavg(&ofproto->hourly.add_rate, 60, ofproto->subfacet_add_count);
9173 exp_mavg(&ofproto->hourly.del_rate, 60, ofproto->subfacet_del_count);
9175 /* Update daily averages on the hour boundaries. */
9176 if ((ofproto->last_minute - ofproto->created) / min_ms % 60 == 59) {
9177 exp_mavg(&ofproto->daily.add_rate, 24, ofproto->hourly.add_rate);
9178 exp_mavg(&ofproto->daily.del_rate, 24, ofproto->hourly.del_rate);
9181 ofproto->total_subfacet_add_count += ofproto->subfacet_add_count;
9182 ofproto->total_subfacet_del_count += ofproto->subfacet_del_count;
9183 ofproto->subfacet_add_count = 0;
9184 ofproto->subfacet_del_count = 0;
9185 ofproto->last_minute += min_ms;
9190 dpif_stats_update_hit_count(struct ofproto_dpif *ofproto, uint64_t delta)
9192 ofproto->n_hit += delta;
9195 const struct ofproto_class ofproto_dpif_class = {
9230 port_is_lacp_current,
9231 NULL, /* rule_choose_table */
9238 rule_modify_actions,
9250 get_stp_port_status,
9257 is_mirror_output_bundle,
9258 forward_bpdu_changed,
9259 set_mac_table_config,