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"
26 #include "byte-order.h"
31 #include "dynamic-string.h"
32 #include "fail-open.h"
36 #include "mac-learning.h"
37 #include "meta-flow.h"
38 #include "multipath.h"
39 #include "netdev-vport.h"
46 #include "ofp-actions.h"
47 #include "ofp-parse.h"
48 #include "ofp-print.h"
49 #include "ofproto-dpif-governor.h"
50 #include "ofproto-dpif-ipfix.h"
51 #include "ofproto-dpif-sflow.h"
52 #include "poll-loop.h"
57 #include "unaligned.h"
59 #include "vlan-bitmap.h"
62 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
64 COVERAGE_DEFINE(ofproto_dpif_expired);
65 COVERAGE_DEFINE(ofproto_dpif_xlate);
66 COVERAGE_DEFINE(facet_changed_rule);
67 COVERAGE_DEFINE(facet_revalidate);
68 COVERAGE_DEFINE(facet_unexpected);
69 COVERAGE_DEFINE(facet_suppress);
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 *,
115 const struct flow *);
116 static struct rule_dpif *rule_dpif_lookup__(struct ofproto_dpif *,
119 static struct rule_dpif *rule_dpif_miss_rule(struct ofproto_dpif *ofproto,
120 const struct flow *flow);
122 static void rule_get_stats(struct rule *, uint64_t *packets, uint64_t *bytes);
123 static void rule_credit_stats(struct rule_dpif *,
124 const struct dpif_flow_stats *);
125 static void flow_push_stats(struct facet *, const struct dpif_flow_stats *);
126 static tag_type rule_calculate_tag(const struct flow *,
127 const struct minimask *, uint32_t basis);
128 static void rule_invalidate(const struct rule_dpif *);
130 #define MAX_MIRRORS 32
131 typedef uint32_t mirror_mask_t;
132 #define MIRROR_MASK_C(X) UINT32_C(X)
133 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
135 struct ofproto_dpif *ofproto; /* Owning ofproto. */
136 size_t idx; /* In ofproto's "mirrors" array. */
137 void *aux; /* Key supplied by ofproto's client. */
138 char *name; /* Identifier for log messages. */
140 /* Selection criteria. */
141 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
142 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
143 unsigned long *vlans; /* Bitmap of chosen VLANs, NULL selects all. */
145 /* Output (exactly one of out == NULL and out_vlan == -1 is true). */
146 struct ofbundle *out; /* Output port or NULL. */
147 int out_vlan; /* Output VLAN or -1. */
148 mirror_mask_t dup_mirrors; /* Bitmap of mirrors with the same output. */
151 int64_t packet_count; /* Number of packets sent. */
152 int64_t byte_count; /* Number of bytes sent. */
155 static void mirror_destroy(struct ofmirror *);
156 static void update_mirror_stats(struct ofproto_dpif *ofproto,
157 mirror_mask_t mirrors,
158 uint64_t packets, uint64_t bytes);
161 struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
162 struct ofproto_dpif *ofproto; /* Owning ofproto. */
163 void *aux; /* Key supplied by ofproto's client. */
164 char *name; /* Identifier for log messages. */
167 struct list ports; /* Contains "struct ofport"s. */
168 enum port_vlan_mode vlan_mode; /* VLAN mode */
169 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
170 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
171 * NULL if all VLANs are trunked. */
172 struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */
173 struct bond *bond; /* Nonnull iff more than one port. */
174 bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */
177 bool floodable; /* True if no port has OFPUTIL_PC_NO_FLOOD set. */
179 /* Port mirroring info. */
180 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
181 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
182 mirror_mask_t mirror_out; /* Mirrors that output to this bundle. */
185 static void bundle_remove(struct ofport *);
186 static void bundle_update(struct ofbundle *);
187 static void bundle_destroy(struct ofbundle *);
188 static void bundle_del_port(struct ofport_dpif *);
189 static void bundle_run(struct ofbundle *);
190 static void bundle_wait(struct ofbundle *);
191 static struct ofbundle *lookup_input_bundle(const struct ofproto_dpif *,
192 uint16_t in_port, bool warn,
193 struct ofport_dpif **in_ofportp);
195 /* A controller may use OFPP_NONE as the ingress port to indicate that
196 * it did not arrive on a "real" port. 'ofpp_none_bundle' exists for
197 * when an input bundle is needed for validation (e.g., mirroring or
198 * OFPP_NORMAL processing). It is not connected to an 'ofproto' or have
199 * any 'port' structs, so care must be taken when dealing with it. */
200 static struct ofbundle ofpp_none_bundle = {
202 .vlan_mode = PORT_VLAN_TRUNK
205 static void stp_run(struct ofproto_dpif *ofproto);
206 static void stp_wait(struct ofproto_dpif *ofproto);
207 static int set_stp_port(struct ofport *,
208 const struct ofproto_port_stp_settings *);
210 static bool ofbundle_includes_vlan(const struct ofbundle *, uint16_t vlan);
214 /* Initial values of fields of the packet that may be changed during
215 * flow processing and needed later. */
216 struct initial_vals {
217 /* This is the value of vlan_tci in the packet as actually received from
218 * dpif. This is the same as the facet's flow.vlan_tci unless the packet
219 * was received via a VLAN splinter. In that case, this value is 0
220 * (because the packet as actually received from the dpif had no 802.1Q
221 * tag) but the facet's flow.vlan_tci is set to the VLAN that the splinter
224 * This member should be removed when the VLAN splinters feature is no
230 tag_type tags; /* Tags associated with actions. */
231 enum slow_path_reason slow; /* 0 if fast path may be used. */
232 bool has_learn; /* Actions include NXAST_LEARN? */
233 bool has_normal; /* Actions output to OFPP_NORMAL? */
234 bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */
235 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
236 mirror_mask_t mirrors; /* Bitmap of associated mirrors. */
238 uint64_t odp_actions_stub[256 / 8];
239 struct ofpbuf odp_actions;
243 struct ofproto_dpif *ofproto;
245 /* Flow to which the OpenFlow actions apply. xlate_actions() will modify
246 * this flow when actions change header fields. */
249 struct initial_vals initial_vals;
251 /* The packet corresponding to 'flow', or a null pointer if we are
252 * revalidating without a packet to refer to. */
253 const struct ofpbuf *packet;
255 /* Should OFPP_NORMAL update the MAC learning table? Should "learn"
256 * actions update the flow table?
258 * We want to update these tables if we are actually processing a packet,
259 * or if we are accounting for packets that the datapath has processed, but
260 * not if we are just revalidating. */
263 /* The rule initiating translation or NULL. */
264 struct rule_dpif *rule;
266 /* The actions to translate. If 'rule' is not NULL, these may be NULL. */
267 const struct ofpact *ofpacts;
270 /* Union of the set of TCP flags seen so far in this flow. (Used only by
271 * NXAST_FIN_TIMEOUT. Set to zero to avoid updating updating rules'
275 /* If nonnull, flow translation calls this function just before executing a
276 * resubmit or OFPP_TABLE action. In addition, disables logging of traces
277 * when the recursion depth is exceeded.
279 * 'rule' is the rule being submitted into. It will be null if the
280 * resubmit or OFPP_TABLE action didn't find a matching rule.
282 * This is normally null so the client has to set it manually after
283 * calling xlate_in_init(). */
284 void (*resubmit_hook)(struct xlate_ctx *, struct rule_dpif *rule);
286 /* If nonnull, flow translation calls this function to report some
287 * significant decision, e.g. to explain why OFPP_NORMAL translation
288 * dropped a packet. */
289 void (*report_hook)(struct xlate_ctx *, const char *s);
291 /* If nonnull, flow translation credits the specified statistics to each
292 * rule reached through a resubmit or OFPP_TABLE action.
294 * This is normally null so the client has to set it manually after
295 * calling xlate_in_init(). */
296 const struct dpif_flow_stats *resubmit_stats;
299 /* Context used by xlate_actions() and its callees. */
301 struct xlate_in *xin;
302 struct xlate_out *xout;
304 struct ofproto_dpif *ofproto;
306 /* Flow at the last commit. */
307 struct flow base_flow;
309 /* Tunnel IP destination address as received. This is stored separately
310 * as the base_flow.tunnel is cleared on init to reflect the datapath
311 * behavior. Used to make sure not to send tunneled output to ourselves,
312 * which might lead to an infinite loop. This could happen easily
313 * if a tunnel is marked as 'ip_remote=flow', and the flow does not
314 * actually set the tun_dst field. */
315 ovs_be32 orig_tunnel_ip_dst;
317 /* Stack for the push and pop actions. Each stack element is of type
318 * "union mf_subvalue". */
319 union mf_subvalue init_stack[1024 / sizeof(union mf_subvalue)];
322 /* The rule that we are currently translating, or NULL. */
323 struct rule_dpif *rule;
325 int recurse; /* Recursion level, via xlate_table_action. */
326 bool max_resubmit_trigger; /* Recursed too deeply during translation. */
327 uint32_t orig_skb_priority; /* Priority when packet arrived. */
328 uint8_t table_id; /* OpenFlow table ID where flow was found. */
329 uint32_t sflow_n_outputs; /* Number of output ports. */
330 uint32_t sflow_odp_port; /* Output port for composing sFlow action. */
331 uint16_t user_cookie_offset;/* Used for user_action_cookie fixup. */
332 bool exit; /* No further actions should be processed. */
335 static void xlate_in_init(struct xlate_in *, struct ofproto_dpif *,
336 const struct flow *, const struct initial_vals *,
337 struct rule_dpif *, uint8_t tcp_flags,
338 const struct ofpbuf *);
340 static void xlate_out_uninit(struct xlate_out *);
342 static void xlate_actions(struct xlate_in *, struct xlate_out *);
344 static void xlate_actions_for_side_effects(struct xlate_in *);
346 static void xlate_table_action(struct xlate_ctx *, uint16_t in_port,
347 uint8_t table_id, bool may_packet_in);
349 static size_t put_userspace_action(const struct ofproto_dpif *,
350 struct ofpbuf *odp_actions,
352 const union user_action_cookie *,
355 static void compose_slow_path(const struct ofproto_dpif *, const struct flow *,
356 enum slow_path_reason,
357 uint64_t *stub, size_t stub_size,
358 const struct nlattr **actionsp,
359 size_t *actions_lenp);
361 static void xlate_report(struct xlate_ctx *ctx, const char *s);
363 /* A subfacet (see "struct subfacet" below) has three possible installation
366 * - SF_NOT_INSTALLED: Not installed in the datapath. This will only be the
367 * case just after the subfacet is created, just before the subfacet is
368 * destroyed, or if the datapath returns an error when we try to install a
371 * - SF_FAST_PATH: The subfacet's actions are installed in the datapath.
373 * - SF_SLOW_PATH: An action that sends every packet for the subfacet through
374 * ofproto_dpif is installed in the datapath.
377 SF_NOT_INSTALLED, /* No datapath flow for this subfacet. */
378 SF_FAST_PATH, /* Full actions are installed. */
379 SF_SLOW_PATH, /* Send-to-userspace action is installed. */
382 /* A dpif flow and actions associated with a facet.
384 * See also the large comment on struct facet. */
387 struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */
388 struct list list_node; /* In struct facet's 'facets' list. */
389 struct facet *facet; /* Owning facet. */
391 enum odp_key_fitness key_fitness;
395 long long int used; /* Time last used; time created if not used. */
396 long long int created; /* Time created. */
398 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
399 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
401 enum subfacet_path path; /* Installed in datapath? */
403 /* Datapath port the packet arrived on. This is needed to remove
404 * flows for ports that are no longer part of the bridge. Since the
405 * flow definition only has the OpenFlow port number and the port is
406 * no longer part of the bridge, we can't determine the datapath port
407 * number needed to delete the flow from the datapath. */
408 uint32_t odp_in_port;
411 #define SUBFACET_DESTROY_MAX_BATCH 50
413 static struct subfacet *subfacet_create(struct facet *, struct flow_miss *miss,
415 static struct subfacet *subfacet_find(struct ofproto_dpif *,
416 const struct nlattr *key, size_t key_len,
418 static void subfacet_destroy(struct subfacet *);
419 static void subfacet_destroy__(struct subfacet *);
420 static void subfacet_destroy_batch(struct ofproto_dpif *,
421 struct subfacet **, int n);
422 static void subfacet_reset_dp_stats(struct subfacet *,
423 struct dpif_flow_stats *);
424 static void subfacet_update_time(struct subfacet *, long long int used);
425 static void subfacet_update_stats(struct subfacet *,
426 const struct dpif_flow_stats *);
427 static int subfacet_install(struct subfacet *,
428 const struct ofpbuf *odp_actions,
429 struct dpif_flow_stats *);
430 static void subfacet_uninstall(struct subfacet *);
432 /* An exact-match instantiation of an OpenFlow flow.
434 * A facet associates a "struct flow", which represents the Open vSwitch
435 * userspace idea of an exact-match flow, with one or more subfacets. Each
436 * subfacet tracks the datapath's idea of the exact-match flow equivalent to
437 * the facet. When the kernel module (or other dpif implementation) and Open
438 * vSwitch userspace agree on the definition of a flow key, there is exactly
439 * one subfacet per facet. If the dpif implementation supports more-specific
440 * flow matching than userspace, however, a facet can have more than one
441 * subfacet, each of which corresponds to some distinction in flow that
442 * userspace simply doesn't understand.
444 * Flow expiration works in terms of subfacets, so a facet must have at least
445 * one subfacet or it will never expire, leaking memory. */
448 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
449 struct list list_node; /* In owning rule's 'facets' list. */
450 struct rule_dpif *rule; /* Owning rule. */
453 struct list subfacets;
454 long long int used; /* Time last used; time created if not used. */
461 * - Do include packets and bytes sent "by hand", e.g. with
464 * - Do include packets and bytes that were obtained from the datapath
465 * when a subfacet's statistics were reset (e.g. dpif_flow_put() with
466 * DPIF_FP_ZERO_STATS).
468 * - Do not include packets or bytes that can be obtained from the
469 * datapath for any existing subfacet.
471 uint64_t packet_count; /* Number of packets received. */
472 uint64_t byte_count; /* Number of bytes received. */
474 /* Resubmit statistics. */
475 uint64_t prev_packet_count; /* Number of packets from last stats push. */
476 uint64_t prev_byte_count; /* Number of bytes from last stats push. */
477 long long int prev_used; /* Used time from last stats push. */
480 uint64_t accounted_bytes; /* Bytes processed by facet_account(). */
481 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
482 uint8_t tcp_flags; /* TCP flags seen for this 'rule'. */
484 struct xlate_out xout;
486 /* Initial values of the packet that may be needed later. */
487 struct initial_vals initial_vals;
489 /* Storage for a single subfacet, to reduce malloc() time and space
490 * overhead. (A facet always has at least one subfacet and in the common
491 * case has exactly one subfacet. However, 'one_subfacet' may not
492 * always be valid, since it could have been removed after newer
493 * subfacets were pushed onto the 'subfacets' list.) */
494 struct subfacet one_subfacet;
496 long long int learn_rl; /* Rate limiter for facet_learn(). */
499 static struct facet *facet_create(const struct flow_miss *, uint32_t hash);
500 static void facet_remove(struct facet *);
501 static void facet_free(struct facet *);
503 static struct facet *facet_find(struct ofproto_dpif *,
504 const struct flow *, uint32_t hash);
505 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
506 const struct flow *, uint32_t hash);
507 static bool facet_revalidate(struct facet *);
508 static bool facet_check_consistency(struct facet *);
510 static void facet_flush_stats(struct facet *);
512 static void facet_update_time(struct facet *, long long int used);
513 static void facet_reset_counters(struct facet *);
514 static void facet_push_stats(struct facet *);
515 static void facet_learn(struct facet *);
516 static void facet_account(struct facet *);
517 static void push_all_stats(void);
519 static bool facet_is_controller_flow(struct facet *);
522 struct hmap_node odp_port_node; /* In dpif_backer's "odp_to_ofport_map". */
526 struct ofbundle *bundle; /* Bundle that contains this port, if any. */
527 struct list bundle_node; /* In struct ofbundle's "ports" list. */
528 struct cfm *cfm; /* Connectivity Fault Management, if any. */
529 struct bfd *bfd; /* BFD, if any. */
530 tag_type tag; /* Tag associated with this port. */
531 bool may_enable; /* May be enabled in bonds. */
532 long long int carrier_seq; /* Carrier status changes. */
533 struct tnl_port *tnl_port; /* Tunnel handle, or null. */
536 struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */
537 enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */
538 long long int stp_state_entered;
540 struct hmap priorities; /* Map of attached 'priority_to_dscp's. */
542 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
544 * This is deprecated. It is only for compatibility with broken device
545 * drivers in old versions of Linux that do not properly support VLANs when
546 * VLAN devices are not used. When broken device drivers are no longer in
547 * widespread use, we will delete these interfaces. */
548 uint16_t realdev_ofp_port;
552 /* Node in 'ofport_dpif''s 'priorities' map. Used to maintain a map from
553 * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
554 * traffic egressing the 'ofport' with that priority should be marked with. */
555 struct priority_to_dscp {
556 struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'priorities' map. */
557 uint32_t priority; /* Priority of this queue (see struct flow). */
559 uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */
562 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
564 * This is deprecated. It is only for compatibility with broken device drivers
565 * in old versions of Linux that do not properly support VLANs when VLAN
566 * devices are not used. When broken device drivers are no longer in
567 * widespread use, we will delete these interfaces. */
568 struct vlan_splinter {
569 struct hmap_node realdev_vid_node;
570 struct hmap_node vlandev_node;
571 uint16_t realdev_ofp_port;
572 uint16_t vlandev_ofp_port;
576 static uint32_t vsp_realdev_to_vlandev(const struct ofproto_dpif *,
577 uint32_t realdev, ovs_be16 vlan_tci);
578 static bool vsp_adjust_flow(const struct ofproto_dpif *, struct flow *);
579 static void vsp_remove(struct ofport_dpif *);
580 static void vsp_add(struct ofport_dpif *, uint16_t realdev_ofp_port, int vid);
582 static uint32_t ofp_port_to_odp_port(const struct ofproto_dpif *,
584 static uint16_t odp_port_to_ofp_port(const struct ofproto_dpif *,
587 static struct ofport_dpif *
588 ofport_dpif_cast(const struct ofport *ofport)
590 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
593 static void port_run(struct ofport_dpif *);
594 static void port_run_fast(struct ofport_dpif *);
595 static void port_wait(struct ofport_dpif *);
596 static int set_bfd(struct ofport *, const struct smap *);
597 static int set_cfm(struct ofport *, const struct cfm_settings *);
598 static void ofport_clear_priorities(struct ofport_dpif *);
599 static void run_fast_rl(void);
601 struct dpif_completion {
602 struct list list_node;
603 struct ofoperation *op;
606 /* Extra information about a classifier table.
607 * Currently used just for optimized flow revalidation. */
609 /* If either of these is nonnull, then this table has a form that allows
610 * flows to be tagged to avoid revalidating most flows for the most common
611 * kinds of flow table changes. */
612 struct cls_table *catchall_table; /* Table that wildcards all fields. */
613 struct cls_table *other_table; /* Table with any other wildcard set. */
614 uint32_t basis; /* Keeps each table's tags separate. */
617 /* Reasons that we might need to revalidate every facet, and corresponding
620 * A value of 0 means that there is no need to revalidate.
622 * It would be nice to have some cleaner way to integrate with coverage
623 * counters, but with only a few reasons I guess this is good enough for
625 enum revalidate_reason {
626 REV_RECONFIGURE = 1, /* Switch configuration changed. */
627 REV_STP, /* Spanning tree protocol port status change. */
628 REV_PORT_TOGGLED, /* Port enabled or disabled by CFM, LACP, ...*/
629 REV_FLOW_TABLE, /* Flow table changed. */
630 REV_INCONSISTENCY /* Facet self-check failed. */
632 COVERAGE_DEFINE(rev_reconfigure);
633 COVERAGE_DEFINE(rev_stp);
634 COVERAGE_DEFINE(rev_port_toggled);
635 COVERAGE_DEFINE(rev_flow_table);
636 COVERAGE_DEFINE(rev_inconsistency);
638 /* Drop keys are odp flow keys which have drop flows installed in the kernel.
639 * These are datapath flows which have no associated ofproto, if they did we
640 * would use facets. */
642 struct hmap_node hmap_node;
647 /* All datapaths of a given type share a single dpif backer instance. */
652 struct timer next_expiration;
653 struct hmap odp_to_ofport_map; /* ODP port to ofport mapping. */
655 struct simap tnl_backers; /* Set of dpif ports backing tunnels. */
657 /* Facet revalidation flags applying to facets which use this backer. */
658 enum revalidate_reason need_revalidate; /* Revalidate every facet. */
659 struct tag_set revalidate_set; /* Revalidate only matching facets. */
661 struct hmap drop_keys; /* Set of dropped odp keys. */
664 /* All existing ofproto_backer instances, indexed by ofproto->up.type. */
665 static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers);
667 static void drop_key_clear(struct dpif_backer *);
668 static struct ofport_dpif *
669 odp_port_to_ofport(const struct dpif_backer *, uint32_t odp_port);
671 static void dpif_stats_update_hit_count(struct ofproto_dpif *ofproto,
673 struct avg_subfacet_rates {
674 double add_rate; /* Moving average of new flows created per minute. */
675 double del_rate; /* Moving average of flows deleted per minute. */
677 static void show_dp_rates(struct ds *ds, const char *heading,
678 const struct avg_subfacet_rates *rates);
679 static void exp_mavg(double *avg, int base, double new);
681 struct ofproto_dpif {
682 struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
684 struct dpif_backer *backer;
686 /* Special OpenFlow rules. */
687 struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */
688 struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */
691 struct netflow *netflow;
692 struct dpif_sflow *sflow;
693 struct dpif_ipfix *ipfix;
694 struct hmap bundles; /* Contains "struct ofbundle"s. */
695 struct mac_learning *ml;
696 struct ofmirror *mirrors[MAX_MIRRORS];
698 bool has_bonded_bundles;
702 struct hmap subfacets;
703 struct governor *governor;
704 long long int consistency_rl;
707 struct table_dpif tables[N_TABLES];
709 /* Support for debugging async flow mods. */
710 struct list completions;
712 bool has_bundle_action; /* True when the first bundle action appears. */
713 struct netdev_stats stats; /* To account packets generated and consumed in
718 long long int stp_last_tick;
720 /* VLAN splinters. */
721 struct hmap realdev_vid_map; /* (realdev,vid) -> vlandev. */
722 struct hmap vlandev_map; /* vlandev -> (realdev,vid). */
725 struct sset ports; /* Set of standard port names. */
726 struct sset ghost_ports; /* Ports with no datapath port. */
727 struct sset port_poll_set; /* Queued names for port_poll() reply. */
728 int port_poll_errno; /* Last errno for port_poll() reply. */
730 /* Per ofproto's dpif stats. */
734 /* Subfacet statistics.
736 * These keep track of the total number of subfacets added and deleted and
737 * flow life span. They are useful for computing the flow rates stats
738 * exposed via "ovs-appctl dpif/show". The goal is to learn about
739 * traffic patterns in ways that we can use later to improve Open vSwitch
740 * performance in new situations. */
741 long long int created; /* Time when it is created. */
742 unsigned int max_n_subfacet; /* Maximum number of flows */
744 /* The average number of subfacets... */
745 struct avg_subfacet_rates hourly; /* ...over the last hour. */
746 struct avg_subfacet_rates daily; /* ...over the last day. */
747 long long int last_minute; /* Last time 'hourly' was updated. */
749 /* Number of subfacets added or deleted since 'last_minute'. */
750 unsigned int subfacet_add_count;
751 unsigned int subfacet_del_count;
753 /* Number of subfacets added or deleted from 'created' to 'last_minute.' */
754 unsigned long long int total_subfacet_add_count;
755 unsigned long long int total_subfacet_del_count;
757 /* Sum of the number of milliseconds that each subfacet existed,
758 * over the subfacets that have been added and then later deleted. */
759 unsigned long long int total_subfacet_life_span;
761 /* Incremented by the number of currently existing subfacets, each
762 * time we pull statistics from the kernel. */
763 unsigned long long int total_subfacet_count;
765 /* Number of times we pull statistics from the kernel. */
766 unsigned long long int n_update_stats;
768 static unsigned long long int avg_subfacet_life_span(
769 const struct ofproto_dpif *);
770 static double avg_subfacet_count(const struct ofproto_dpif *ofproto);
771 static void update_moving_averages(struct ofproto_dpif *ofproto);
772 static void dpif_stats_update_hit_count(struct ofproto_dpif *ofproto,
774 static void update_max_subfacet_count(struct ofproto_dpif *ofproto);
776 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
777 * for debugging the asynchronous flow_mod implementation.) */
780 /* All existing ofproto_dpif instances, indexed by ->up.name. */
781 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
783 static void ofproto_dpif_unixctl_init(void);
785 static struct ofproto_dpif *
786 ofproto_dpif_cast(const struct ofproto *ofproto)
788 ovs_assert(ofproto->ofproto_class == &ofproto_dpif_class);
789 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
792 static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *,
794 static struct ofport_dpif *get_odp_port(const struct ofproto_dpif *,
796 static void ofproto_trace(struct ofproto_dpif *, const struct flow *,
797 const struct ofpbuf *,
798 const struct initial_vals *, struct ds *);
800 /* Packet processing. */
801 static void update_learning_table(struct ofproto_dpif *,
802 const struct flow *, int vlan,
805 #define FLOW_MISS_MAX_BATCH 50
806 static int handle_upcalls(struct dpif_backer *, unsigned int max_batch);
808 /* Flow expiration. */
809 static int expire(struct dpif_backer *);
812 static void send_netflow_active_timeouts(struct ofproto_dpif *);
815 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
816 static size_t compose_sflow_action(const struct ofproto_dpif *,
817 struct ofpbuf *odp_actions,
818 const struct flow *, uint32_t odp_port);
819 static void compose_ipfix_action(const struct ofproto_dpif *,
820 struct ofpbuf *odp_actions,
821 const struct flow *);
822 static void add_mirror_actions(struct xlate_ctx *ctx,
823 const struct flow *flow);
824 /* Global variables. */
825 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
827 /* Initial mappings of port to bridge mappings. */
828 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
830 /* Factory functions. */
833 init(const struct shash *iface_hints)
835 struct shash_node *node;
837 /* Make a local copy, since we don't own 'iface_hints' elements. */
838 SHASH_FOR_EACH(node, iface_hints) {
839 const struct iface_hint *orig_hint = node->data;
840 struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
842 new_hint->br_name = xstrdup(orig_hint->br_name);
843 new_hint->br_type = xstrdup(orig_hint->br_type);
844 new_hint->ofp_port = orig_hint->ofp_port;
846 shash_add(&init_ofp_ports, node->name, new_hint);
851 enumerate_types(struct sset *types)
853 dp_enumerate_types(types);
857 enumerate_names(const char *type, struct sset *names)
859 struct ofproto_dpif *ofproto;
862 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
863 if (strcmp(type, ofproto->up.type)) {
866 sset_add(names, ofproto->up.name);
873 del(const char *type, const char *name)
878 error = dpif_open(name, type, &dpif);
880 error = dpif_delete(dpif);
887 port_open_type(const char *datapath_type, const char *port_type)
889 return dpif_port_open_type(datapath_type, port_type);
892 /* Type functions. */
894 static struct ofproto_dpif *
895 lookup_ofproto_dpif_by_port_name(const char *name)
897 struct ofproto_dpif *ofproto;
899 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
900 if (sset_contains(&ofproto->ports, name)) {
909 type_run(const char *type)
911 static long long int push_timer = LLONG_MIN;
912 struct dpif_backer *backer;
916 backer = shash_find_data(&all_dpif_backers, type);
918 /* This is not necessarily a problem, since backers are only
919 * created on demand. */
923 dpif_run(backer->dpif);
925 /* The most natural place to push facet statistics is when they're pulled
926 * from the datapath. However, when there are many flows in the datapath,
927 * this expensive operation can occur so frequently, that it reduces our
928 * ability to quickly set up flows. To reduce the cost, we push statistics
930 if (time_msec() > push_timer) {
931 push_timer = time_msec() + 2000;
935 if (backer->need_revalidate
936 || !tag_set_is_empty(&backer->revalidate_set)) {
937 struct tag_set revalidate_set = backer->revalidate_set;
938 bool need_revalidate = backer->need_revalidate;
939 struct ofproto_dpif *ofproto;
940 struct simap_node *node;
941 struct simap tmp_backers;
943 /* Handle tunnel garbage collection. */
944 simap_init(&tmp_backers);
945 simap_swap(&backer->tnl_backers, &tmp_backers);
947 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
948 struct ofport_dpif *iter;
950 if (backer != ofproto->backer) {
954 HMAP_FOR_EACH (iter, up.hmap_node, &ofproto->up.ports) {
957 if (!iter->tnl_port) {
961 dp_port = netdev_vport_get_dpif_port(iter->up.netdev);
962 node = simap_find(&tmp_backers, dp_port);
964 simap_put(&backer->tnl_backers, dp_port, node->data);
965 simap_delete(&tmp_backers, node);
966 node = simap_find(&backer->tnl_backers, dp_port);
968 node = simap_find(&backer->tnl_backers, dp_port);
970 uint32_t odp_port = UINT32_MAX;
972 if (!dpif_port_add(backer->dpif, iter->up.netdev,
974 simap_put(&backer->tnl_backers, dp_port, odp_port);
975 node = simap_find(&backer->tnl_backers, dp_port);
980 iter->odp_port = node ? node->data : OVSP_NONE;
981 if (tnl_port_reconfigure(&iter->up, iter->odp_port,
983 backer->need_revalidate = REV_RECONFIGURE;
988 SIMAP_FOR_EACH (node, &tmp_backers) {
989 dpif_port_del(backer->dpif, node->data);
991 simap_destroy(&tmp_backers);
993 switch (backer->need_revalidate) {
994 case REV_RECONFIGURE: COVERAGE_INC(rev_reconfigure); break;
995 case REV_STP: COVERAGE_INC(rev_stp); break;
996 case REV_PORT_TOGGLED: COVERAGE_INC(rev_port_toggled); break;
997 case REV_FLOW_TABLE: COVERAGE_INC(rev_flow_table); break;
998 case REV_INCONSISTENCY: COVERAGE_INC(rev_inconsistency); break;
1001 if (backer->need_revalidate) {
1002 /* Clear the drop_keys in case we should now be accepting some
1003 * formerly dropped flows. */
1004 drop_key_clear(backer);
1007 /* Clear the revalidation flags. */
1008 tag_set_init(&backer->revalidate_set);
1009 backer->need_revalidate = 0;
1011 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1012 struct facet *facet, *next;
1014 if (ofproto->backer != backer) {
1018 HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &ofproto->facets) {
1020 || tag_set_intersects(&revalidate_set, facet->xout.tags)) {
1021 facet_revalidate(facet);
1028 if (timer_expired(&backer->next_expiration)) {
1029 int delay = expire(backer);
1030 timer_set_duration(&backer->next_expiration, delay);
1033 /* Check for port changes in the dpif. */
1034 while ((error = dpif_port_poll(backer->dpif, &devname)) == 0) {
1035 struct ofproto_dpif *ofproto;
1036 struct dpif_port port;
1038 /* Don't report on the datapath's device. */
1039 if (!strcmp(devname, dpif_base_name(backer->dpif))) {
1043 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
1044 &all_ofproto_dpifs) {
1045 if (simap_contains(&ofproto->backer->tnl_backers, devname)) {
1050 ofproto = lookup_ofproto_dpif_by_port_name(devname);
1051 if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
1052 /* The port was removed. If we know the datapath,
1053 * report it through poll_set(). If we don't, it may be
1054 * notifying us of a removal we initiated, so ignore it.
1055 * If there's a pending ENOBUFS, let it stand, since
1056 * everything will be reevaluated. */
1057 if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
1058 sset_add(&ofproto->port_poll_set, devname);
1059 ofproto->port_poll_errno = 0;
1061 } else if (!ofproto) {
1062 /* The port was added, but we don't know with which
1063 * ofproto we should associate it. Delete it. */
1064 dpif_port_del(backer->dpif, port.port_no);
1066 dpif_port_destroy(&port);
1072 if (error != EAGAIN) {
1073 struct ofproto_dpif *ofproto;
1075 /* There was some sort of error, so propagate it to all
1076 * ofprotos that use this backer. */
1077 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
1078 &all_ofproto_dpifs) {
1079 if (ofproto->backer == backer) {
1080 sset_clear(&ofproto->port_poll_set);
1081 ofproto->port_poll_errno = error;
1090 dpif_backer_run_fast(struct dpif_backer *backer, int max_batch)
1094 /* Handle one or more batches of upcalls, until there's nothing left to do
1095 * or until we do a fixed total amount of work.
1097 * We do work in batches because it can be much cheaper to set up a number
1098 * of flows and fire off their patches all at once. We do multiple batches
1099 * because in some cases handling a packet can cause another packet to be
1100 * queued almost immediately as part of the return flow. Both
1101 * optimizations can make major improvements on some benchmarks and
1102 * presumably for real traffic as well. */
1104 while (work < max_batch) {
1105 int retval = handle_upcalls(backer, max_batch - work);
1116 type_run_fast(const char *type)
1118 struct dpif_backer *backer;
1120 backer = shash_find_data(&all_dpif_backers, type);
1122 /* This is not necessarily a problem, since backers are only
1123 * created on demand. */
1127 return dpif_backer_run_fast(backer, FLOW_MISS_MAX_BATCH);
1133 static long long int port_rl = LLONG_MIN;
1134 static unsigned int backer_rl = 0;
1136 if (time_msec() >= port_rl) {
1137 struct ofproto_dpif *ofproto;
1138 struct ofport_dpif *ofport;
1140 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1142 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1143 port_run_fast(ofport);
1146 port_rl = time_msec() + 200;
1149 /* XXX: We have to be careful not to do too much work in this function. If
1150 * we call dpif_backer_run_fast() too often, or with too large a batch,
1151 * performance improves signifcantly, but at a cost. It's possible for the
1152 * number of flows in the datapath to increase without bound, and for poll
1153 * loops to take 10s of seconds. The correct solution to this problem,
1154 * long term, is to separate flow miss handling into it's own thread so it
1155 * isn't affected by revalidations, and expirations. Until then, this is
1156 * the best we can do. */
1157 if (++backer_rl >= 10) {
1158 struct shash_node *node;
1161 SHASH_FOR_EACH (node, &all_dpif_backers) {
1162 dpif_backer_run_fast(node->data, 1);
1168 type_wait(const char *type)
1170 struct dpif_backer *backer;
1172 backer = shash_find_data(&all_dpif_backers, type);
1174 /* This is not necessarily a problem, since backers are only
1175 * created on demand. */
1179 timer_wait(&backer->next_expiration);
1182 /* Basic life-cycle. */
1184 static int add_internal_flows(struct ofproto_dpif *);
1186 static struct ofproto *
1189 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
1190 return &ofproto->up;
1194 dealloc(struct ofproto *ofproto_)
1196 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1201 close_dpif_backer(struct dpif_backer *backer)
1203 struct shash_node *node;
1205 ovs_assert(backer->refcount > 0);
1207 if (--backer->refcount) {
1211 drop_key_clear(backer);
1212 hmap_destroy(&backer->drop_keys);
1214 simap_destroy(&backer->tnl_backers);
1215 hmap_destroy(&backer->odp_to_ofport_map);
1216 node = shash_find(&all_dpif_backers, backer->type);
1218 shash_delete(&all_dpif_backers, node);
1219 dpif_close(backer->dpif);
1224 /* Datapath port slated for removal from datapath. */
1225 struct odp_garbage {
1226 struct list list_node;
1231 open_dpif_backer(const char *type, struct dpif_backer **backerp)
1233 struct dpif_backer *backer;
1234 struct dpif_port_dump port_dump;
1235 struct dpif_port port;
1236 struct shash_node *node;
1237 struct list garbage_list;
1238 struct odp_garbage *garbage, *next;
1244 backer = shash_find_data(&all_dpif_backers, type);
1251 backer_name = xasprintf("ovs-%s", type);
1253 /* Remove any existing datapaths, since we assume we're the only
1254 * userspace controlling the datapath. */
1256 dp_enumerate_names(type, &names);
1257 SSET_FOR_EACH(name, &names) {
1258 struct dpif *old_dpif;
1260 /* Don't remove our backer if it exists. */
1261 if (!strcmp(name, backer_name)) {
1265 if (dpif_open(name, type, &old_dpif)) {
1266 VLOG_WARN("couldn't open old datapath %s to remove it", name);
1268 dpif_delete(old_dpif);
1269 dpif_close(old_dpif);
1272 sset_destroy(&names);
1274 backer = xmalloc(sizeof *backer);
1276 error = dpif_create_and_open(backer_name, type, &backer->dpif);
1279 VLOG_ERR("failed to open datapath of type %s: %s", type,
1285 backer->type = xstrdup(type);
1286 backer->refcount = 1;
1287 hmap_init(&backer->odp_to_ofport_map);
1288 hmap_init(&backer->drop_keys);
1289 timer_set_duration(&backer->next_expiration, 1000);
1290 backer->need_revalidate = 0;
1291 simap_init(&backer->tnl_backers);
1292 tag_set_init(&backer->revalidate_set);
1295 dpif_flow_flush(backer->dpif);
1297 /* Loop through the ports already on the datapath and remove any
1298 * that we don't need anymore. */
1299 list_init(&garbage_list);
1300 dpif_port_dump_start(&port_dump, backer->dpif);
1301 while (dpif_port_dump_next(&port_dump, &port)) {
1302 node = shash_find(&init_ofp_ports, port.name);
1303 if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
1304 garbage = xmalloc(sizeof *garbage);
1305 garbage->odp_port = port.port_no;
1306 list_push_front(&garbage_list, &garbage->list_node);
1309 dpif_port_dump_done(&port_dump);
1311 LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) {
1312 dpif_port_del(backer->dpif, garbage->odp_port);
1313 list_remove(&garbage->list_node);
1317 shash_add(&all_dpif_backers, type, backer);
1319 error = dpif_recv_set(backer->dpif, true);
1321 VLOG_ERR("failed to listen on datapath of type %s: %s",
1322 type, strerror(error));
1323 close_dpif_backer(backer);
1331 construct(struct ofproto *ofproto_)
1333 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1334 struct shash_node *node, *next;
1339 error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1344 max_ports = dpif_get_max_ports(ofproto->backer->dpif);
1345 ofproto_init_max_ports(ofproto_, MIN(max_ports, OFPP_MAX));
1347 ofproto->netflow = NULL;
1348 ofproto->sflow = NULL;
1349 ofproto->ipfix = NULL;
1350 ofproto->stp = NULL;
1351 hmap_init(&ofproto->bundles);
1352 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1353 for (i = 0; i < MAX_MIRRORS; i++) {
1354 ofproto->mirrors[i] = NULL;
1356 ofproto->has_bonded_bundles = false;
1358 hmap_init(&ofproto->facets);
1359 hmap_init(&ofproto->subfacets);
1360 ofproto->governor = NULL;
1361 ofproto->consistency_rl = LLONG_MIN;
1363 for (i = 0; i < N_TABLES; i++) {
1364 struct table_dpif *table = &ofproto->tables[i];
1366 table->catchall_table = NULL;
1367 table->other_table = NULL;
1368 table->basis = random_uint32();
1371 list_init(&ofproto->completions);
1373 ofproto_dpif_unixctl_init();
1375 ofproto->has_mirrors = false;
1376 ofproto->has_bundle_action = false;
1378 hmap_init(&ofproto->vlandev_map);
1379 hmap_init(&ofproto->realdev_vid_map);
1381 sset_init(&ofproto->ports);
1382 sset_init(&ofproto->ghost_ports);
1383 sset_init(&ofproto->port_poll_set);
1384 ofproto->port_poll_errno = 0;
1386 SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1387 struct iface_hint *iface_hint = node->data;
1389 if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1390 /* Check if the datapath already has this port. */
1391 if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1392 sset_add(&ofproto->ports, node->name);
1395 free(iface_hint->br_name);
1396 free(iface_hint->br_type);
1398 shash_delete(&init_ofp_ports, node);
1402 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1403 hash_string(ofproto->up.name, 0));
1404 memset(&ofproto->stats, 0, sizeof ofproto->stats);
1406 ofproto_init_tables(ofproto_, N_TABLES);
1407 error = add_internal_flows(ofproto);
1408 ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1411 ofproto->n_missed = 0;
1413 ofproto->max_n_subfacet = 0;
1414 ofproto->created = time_msec();
1415 ofproto->last_minute = ofproto->created;
1416 memset(&ofproto->hourly, 0, sizeof ofproto->hourly);
1417 memset(&ofproto->daily, 0, sizeof ofproto->daily);
1418 ofproto->subfacet_add_count = 0;
1419 ofproto->subfacet_del_count = 0;
1420 ofproto->total_subfacet_add_count = 0;
1421 ofproto->total_subfacet_del_count = 0;
1422 ofproto->total_subfacet_life_span = 0;
1423 ofproto->total_subfacet_count = 0;
1424 ofproto->n_update_stats = 0;
1430 add_internal_flow(struct ofproto_dpif *ofproto, int id,
1431 const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1433 struct ofputil_flow_mod fm;
1436 match_init_catchall(&fm.match);
1438 match_set_reg(&fm.match, 0, id);
1439 fm.new_cookie = htonll(0);
1440 fm.cookie = htonll(0);
1441 fm.cookie_mask = htonll(0);
1442 fm.table_id = TBL_INTERNAL;
1443 fm.command = OFPFC_ADD;
1444 fm.idle_timeout = 0;
1445 fm.hard_timeout = 0;
1449 fm.ofpacts = ofpacts->data;
1450 fm.ofpacts_len = ofpacts->size;
1452 error = ofproto_flow_mod(&ofproto->up, &fm);
1454 VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
1455 id, ofperr_to_string(error));
1459 *rulep = rule_dpif_lookup__(ofproto, &fm.match.flow, TBL_INTERNAL);
1460 ovs_assert(*rulep != NULL);
1466 add_internal_flows(struct ofproto_dpif *ofproto)
1468 struct ofpact_controller *controller;
1469 uint64_t ofpacts_stub[128 / 8];
1470 struct ofpbuf ofpacts;
1474 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1477 controller = ofpact_put_CONTROLLER(&ofpacts);
1478 controller->max_len = UINT16_MAX;
1479 controller->controller_id = 0;
1480 controller->reason = OFPR_NO_MATCH;
1481 ofpact_pad(&ofpacts);
1483 error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule);
1488 ofpbuf_clear(&ofpacts);
1489 error = add_internal_flow(ofproto, id++, &ofpacts,
1490 &ofproto->no_packet_in_rule);
1495 complete_operations(struct ofproto_dpif *ofproto)
1497 struct dpif_completion *c, *next;
1499 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
1500 ofoperation_complete(c->op, 0);
1501 list_remove(&c->list_node);
1507 destruct(struct ofproto *ofproto_)
1509 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1510 struct rule_dpif *rule, *next_rule;
1511 struct oftable *table;
1514 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1515 complete_operations(ofproto);
1517 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1518 struct cls_cursor cursor;
1520 cls_cursor_init(&cursor, &table->cls, NULL);
1521 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1522 ofproto_rule_destroy(&rule->up);
1526 for (i = 0; i < MAX_MIRRORS; i++) {
1527 mirror_destroy(ofproto->mirrors[i]);
1530 netflow_destroy(ofproto->netflow);
1531 dpif_sflow_destroy(ofproto->sflow);
1532 hmap_destroy(&ofproto->bundles);
1533 mac_learning_destroy(ofproto->ml);
1535 hmap_destroy(&ofproto->facets);
1536 hmap_destroy(&ofproto->subfacets);
1537 governor_destroy(ofproto->governor);
1539 hmap_destroy(&ofproto->vlandev_map);
1540 hmap_destroy(&ofproto->realdev_vid_map);
1542 sset_destroy(&ofproto->ports);
1543 sset_destroy(&ofproto->ghost_ports);
1544 sset_destroy(&ofproto->port_poll_set);
1546 close_dpif_backer(ofproto->backer);
1550 run_fast(struct ofproto *ofproto_)
1552 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1553 struct ofport_dpif *ofport;
1555 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1556 port_run_fast(ofport);
1563 run(struct ofproto *ofproto_)
1565 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1566 struct ofport_dpif *ofport;
1567 struct ofbundle *bundle;
1571 complete_operations(ofproto);
1574 error = run_fast(ofproto_);
1579 if (ofproto->netflow) {
1580 if (netflow_run(ofproto->netflow)) {
1581 send_netflow_active_timeouts(ofproto);
1584 if (ofproto->sflow) {
1585 dpif_sflow_run(ofproto->sflow);
1588 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1591 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1596 mac_learning_run(ofproto->ml, &ofproto->backer->revalidate_set);
1598 /* Check the consistency of a random facet, to aid debugging. */
1599 if (time_msec() >= ofproto->consistency_rl
1600 && !hmap_is_empty(&ofproto->facets)
1601 && !ofproto->backer->need_revalidate) {
1602 struct facet *facet;
1604 ofproto->consistency_rl = time_msec() + 250;
1606 facet = CONTAINER_OF(hmap_random_node(&ofproto->facets),
1607 struct facet, hmap_node);
1608 if (!tag_set_intersects(&ofproto->backer->revalidate_set,
1609 facet->xout.tags)) {
1610 if (!facet_check_consistency(facet)) {
1611 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
1616 if (ofproto->governor) {
1619 governor_run(ofproto->governor);
1621 /* If the governor has shrunk to its minimum size and the number of
1622 * subfacets has dwindled, then drop the governor entirely.
1624 * For hysteresis, the number of subfacets to drop the governor is
1625 * smaller than the number needed to trigger its creation. */
1626 n_subfacets = hmap_count(&ofproto->subfacets);
1627 if (n_subfacets * 4 < ofproto->up.flow_eviction_threshold
1628 && governor_is_idle(ofproto->governor)) {
1629 governor_destroy(ofproto->governor);
1630 ofproto->governor = NULL;
1638 wait(struct ofproto *ofproto_)
1640 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1641 struct ofport_dpif *ofport;
1642 struct ofbundle *bundle;
1644 if (!clogged && !list_is_empty(&ofproto->completions)) {
1645 poll_immediate_wake();
1648 dpif_wait(ofproto->backer->dpif);
1649 dpif_recv_wait(ofproto->backer->dpif);
1650 if (ofproto->sflow) {
1651 dpif_sflow_wait(ofproto->sflow);
1653 if (!tag_set_is_empty(&ofproto->backer->revalidate_set)) {
1654 poll_immediate_wake();
1656 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1659 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1660 bundle_wait(bundle);
1662 if (ofproto->netflow) {
1663 netflow_wait(ofproto->netflow);
1665 mac_learning_wait(ofproto->ml);
1667 if (ofproto->backer->need_revalidate) {
1668 /* Shouldn't happen, but if it does just go around again. */
1669 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1670 poll_immediate_wake();
1672 if (ofproto->governor) {
1673 governor_wait(ofproto->governor);
1678 get_memory_usage(const struct ofproto *ofproto_, struct simap *usage)
1680 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1682 simap_increase(usage, "facets", hmap_count(&ofproto->facets));
1683 simap_increase(usage, "subfacets", hmap_count(&ofproto->subfacets));
1687 flush(struct ofproto *ofproto_)
1689 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1690 struct subfacet *subfacet, *next_subfacet;
1691 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
1695 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
1696 &ofproto->subfacets) {
1697 if (subfacet->path != SF_NOT_INSTALLED) {
1698 batch[n_batch++] = subfacet;
1699 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
1700 subfacet_destroy_batch(ofproto, batch, n_batch);
1704 subfacet_destroy(subfacet);
1709 subfacet_destroy_batch(ofproto, batch, n_batch);
1714 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1715 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1717 *arp_match_ip = true;
1718 *actions = (OFPUTIL_A_OUTPUT |
1719 OFPUTIL_A_SET_VLAN_VID |
1720 OFPUTIL_A_SET_VLAN_PCP |
1721 OFPUTIL_A_STRIP_VLAN |
1722 OFPUTIL_A_SET_DL_SRC |
1723 OFPUTIL_A_SET_DL_DST |
1724 OFPUTIL_A_SET_NW_SRC |
1725 OFPUTIL_A_SET_NW_DST |
1726 OFPUTIL_A_SET_NW_TOS |
1727 OFPUTIL_A_SET_TP_SRC |
1728 OFPUTIL_A_SET_TP_DST |
1733 get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots)
1735 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1736 struct dpif_dp_stats s;
1737 uint64_t n_miss, n_no_pkt_in, n_bytes;
1740 strcpy(ots->name, "classifier");
1742 dpif_get_dp_stats(ofproto->backer->dpif, &s);
1743 rule_get_stats(&ofproto->miss_rule->up, &n_miss, &n_bytes);
1744 rule_get_stats(&ofproto->no_packet_in_rule->up, &n_no_pkt_in, &n_bytes);
1746 n_lookup = s.n_hit + s.n_missed;
1747 ots->lookup_count = htonll(n_lookup);
1748 ots->matched_count = htonll(n_lookup - n_miss - n_no_pkt_in);
1751 static struct ofport *
1754 struct ofport_dpif *port = xmalloc(sizeof *port);
1759 port_dealloc(struct ofport *port_)
1761 struct ofport_dpif *port = ofport_dpif_cast(port_);
1766 port_construct(struct ofport *port_)
1768 struct ofport_dpif *port = ofport_dpif_cast(port_);
1769 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1770 const struct netdev *netdev = port->up.netdev;
1771 struct dpif_port dpif_port;
1774 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1775 port->bundle = NULL;
1778 port->tag = tag_create_random();
1779 port->may_enable = true;
1780 port->stp_port = NULL;
1781 port->stp_state = STP_DISABLED;
1782 port->tnl_port = NULL;
1783 hmap_init(&port->priorities);
1784 port->realdev_ofp_port = 0;
1785 port->vlandev_vid = 0;
1786 port->carrier_seq = netdev_get_carrier_resets(netdev);
1788 if (netdev_vport_is_patch(netdev)) {
1789 /* By bailing out here, we don't submit the port to the sFlow module
1790 * to be considered for counter polling export. This is correct
1791 * because the patch port represents an interface that sFlow considers
1792 * to be "internal" to the switch as a whole, and therefore not an
1793 * candidate for counter polling. */
1794 port->odp_port = OVSP_NONE;
1798 error = dpif_port_query_by_name(ofproto->backer->dpif,
1799 netdev_vport_get_dpif_port(netdev),
1805 port->odp_port = dpif_port.port_no;
1807 if (netdev_get_tunnel_config(netdev)) {
1808 port->tnl_port = tnl_port_add(&port->up, port->odp_port);
1810 /* Sanity-check that a mapping doesn't already exist. This
1811 * shouldn't happen for non-tunnel ports. */
1812 if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1813 VLOG_ERR("port %s already has an OpenFlow port number",
1815 dpif_port_destroy(&dpif_port);
1819 hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1820 hash_int(port->odp_port, 0));
1822 dpif_port_destroy(&dpif_port);
1824 if (ofproto->sflow) {
1825 dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1832 port_destruct(struct ofport *port_)
1834 struct ofport_dpif *port = ofport_dpif_cast(port_);
1835 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1836 const char *dp_port_name = netdev_vport_get_dpif_port(port->up.netdev);
1837 const char *devname = netdev_get_name(port->up.netdev);
1839 if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1840 /* The underlying device is still there, so delete it. This
1841 * happens when the ofproto is being destroyed, since the caller
1842 * assumes that removal of attached ports will happen as part of
1844 if (!port->tnl_port) {
1845 dpif_port_del(ofproto->backer->dpif, port->odp_port);
1847 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1850 if (port->odp_port != OVSP_NONE && !port->tnl_port) {
1851 hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1854 tnl_port_del(port->tnl_port);
1855 sset_find_and_delete(&ofproto->ports, devname);
1856 sset_find_and_delete(&ofproto->ghost_ports, devname);
1857 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1858 bundle_remove(port_);
1859 set_cfm(port_, NULL);
1860 set_bfd(port_, NULL);
1861 if (ofproto->sflow) {
1862 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1865 ofport_clear_priorities(port);
1866 hmap_destroy(&port->priorities);
1870 port_modified(struct ofport *port_)
1872 struct ofport_dpif *port = ofport_dpif_cast(port_);
1874 if (port->bundle && port->bundle->bond) {
1875 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1880 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1882 struct ofport_dpif *port = ofport_dpif_cast(port_);
1883 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1884 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1886 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1887 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1888 OFPUTIL_PC_NO_PACKET_IN)) {
1889 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1891 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1892 bundle_update(port->bundle);
1898 set_sflow(struct ofproto *ofproto_,
1899 const struct ofproto_sflow_options *sflow_options)
1901 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1902 struct dpif_sflow *ds = ofproto->sflow;
1904 if (sflow_options) {
1906 struct ofport_dpif *ofport;
1908 ds = ofproto->sflow = dpif_sflow_create();
1909 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1910 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1912 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1914 dpif_sflow_set_options(ds, sflow_options);
1917 dpif_sflow_destroy(ds);
1918 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1919 ofproto->sflow = NULL;
1927 struct ofproto *ofproto_,
1928 const struct ofproto_ipfix_bridge_exporter_options *bridge_exporter_options,
1929 const struct ofproto_ipfix_flow_exporter_options *flow_exporters_options,
1930 size_t n_flow_exporters_options)
1932 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1933 struct dpif_ipfix *di = ofproto->ipfix;
1935 if (bridge_exporter_options || flow_exporters_options) {
1937 di = ofproto->ipfix = dpif_ipfix_create();
1939 dpif_ipfix_set_options(
1940 di, bridge_exporter_options, flow_exporters_options,
1941 n_flow_exporters_options);
1944 dpif_ipfix_destroy(di);
1945 ofproto->ipfix = NULL;
1952 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1954 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1961 struct ofproto_dpif *ofproto;
1963 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1964 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1965 ofport->cfm = cfm_create(ofport->up.netdev);
1968 if (cfm_configure(ofport->cfm, s)) {
1974 cfm_destroy(ofport->cfm);
1980 get_cfm_status(const struct ofport *ofport_,
1981 struct ofproto_cfm_status *status)
1983 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1986 status->faults = cfm_get_fault(ofport->cfm);
1987 status->remote_opstate = cfm_get_opup(ofport->cfm);
1988 status->health = cfm_get_health(ofport->cfm);
1989 cfm_get_remote_mpids(ofport->cfm, &status->rmps, &status->n_rmps);
1997 set_bfd(struct ofport *ofport_, const struct smap *cfg)
1999 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
2000 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2004 ofport->bfd = bfd_configure(old, netdev_get_name(ofport->up.netdev), cfg);
2005 if (ofport->bfd != old) {
2006 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2013 get_bfd_status(struct ofport *ofport_, struct smap *smap)
2015 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2018 bfd_get_status(ofport->bfd, smap);
2025 /* Spanning Tree. */
2028 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
2030 struct ofproto_dpif *ofproto = ofproto_;
2031 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
2032 struct ofport_dpif *ofport;
2034 ofport = stp_port_get_aux(sp);
2036 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
2037 ofproto->up.name, port_num);
2039 struct eth_header *eth = pkt->l2;
2041 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
2042 if (eth_addr_is_zero(eth->eth_src)) {
2043 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
2044 "with unknown MAC", ofproto->up.name, port_num);
2046 send_packet(ofport, pkt);
2052 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
2054 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
2056 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2058 /* Only revalidate flows if the configuration changed. */
2059 if (!s != !ofproto->stp) {
2060 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2064 if (!ofproto->stp) {
2065 ofproto->stp = stp_create(ofproto_->name, s->system_id,
2066 send_bpdu_cb, ofproto);
2067 ofproto->stp_last_tick = time_msec();
2070 stp_set_bridge_id(ofproto->stp, s->system_id);
2071 stp_set_bridge_priority(ofproto->stp, s->priority);
2072 stp_set_hello_time(ofproto->stp, s->hello_time);
2073 stp_set_max_age(ofproto->stp, s->max_age);
2074 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
2076 struct ofport *ofport;
2078 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
2079 set_stp_port(ofport, NULL);
2082 stp_destroy(ofproto->stp);
2083 ofproto->stp = NULL;
2090 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
2092 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2096 s->bridge_id = stp_get_bridge_id(ofproto->stp);
2097 s->designated_root = stp_get_designated_root(ofproto->stp);
2098 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
2107 update_stp_port_state(struct ofport_dpif *ofport)
2109 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2110 enum stp_state state;
2112 /* Figure out new state. */
2113 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
2117 if (ofport->stp_state != state) {
2118 enum ofputil_port_state of_state;
2121 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
2122 netdev_get_name(ofport->up.netdev),
2123 stp_state_name(ofport->stp_state),
2124 stp_state_name(state));
2125 if (stp_learn_in_state(ofport->stp_state)
2126 != stp_learn_in_state(state)) {
2127 /* xxx Learning action flows should also be flushed. */
2128 mac_learning_flush(ofproto->ml,
2129 &ofproto->backer->revalidate_set);
2131 fwd_change = stp_forward_in_state(ofport->stp_state)
2132 != stp_forward_in_state(state);
2134 ofproto->backer->need_revalidate = REV_STP;
2135 ofport->stp_state = state;
2136 ofport->stp_state_entered = time_msec();
2138 if (fwd_change && ofport->bundle) {
2139 bundle_update(ofport->bundle);
2142 /* Update the STP state bits in the OpenFlow port description. */
2143 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
2144 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
2145 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
2146 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
2147 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
2149 ofproto_port_set_state(&ofport->up, of_state);
2153 /* Configures STP on 'ofport_' using the settings defined in 's'. The
2154 * caller is responsible for assigning STP port numbers and ensuring
2155 * there are no duplicates. */
2157 set_stp_port(struct ofport *ofport_,
2158 const struct ofproto_port_stp_settings *s)
2160 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2161 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2162 struct stp_port *sp = ofport->stp_port;
2164 if (!s || !s->enable) {
2166 ofport->stp_port = NULL;
2167 stp_port_disable(sp);
2168 update_stp_port_state(ofport);
2171 } else if (sp && stp_port_no(sp) != s->port_num
2172 && ofport == stp_port_get_aux(sp)) {
2173 /* The port-id changed, so disable the old one if it's not
2174 * already in use by another port. */
2175 stp_port_disable(sp);
2178 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
2179 stp_port_enable(sp);
2181 stp_port_set_aux(sp, ofport);
2182 stp_port_set_priority(sp, s->priority);
2183 stp_port_set_path_cost(sp, s->path_cost);
2185 update_stp_port_state(ofport);
2191 get_stp_port_status(struct ofport *ofport_,
2192 struct ofproto_port_stp_status *s)
2194 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2195 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2196 struct stp_port *sp = ofport->stp_port;
2198 if (!ofproto->stp || !sp) {
2204 s->port_id = stp_port_get_id(sp);
2205 s->state = stp_port_get_state(sp);
2206 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
2207 s->role = stp_port_get_role(sp);
2208 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
2214 stp_run(struct ofproto_dpif *ofproto)
2217 long long int now = time_msec();
2218 long long int elapsed = now - ofproto->stp_last_tick;
2219 struct stp_port *sp;
2222 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
2223 ofproto->stp_last_tick = now;
2225 while (stp_get_changed_port(ofproto->stp, &sp)) {
2226 struct ofport_dpif *ofport = stp_port_get_aux(sp);
2229 update_stp_port_state(ofport);
2233 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
2234 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
2240 stp_wait(struct ofproto_dpif *ofproto)
2243 poll_timer_wait(1000);
2247 /* Returns true if STP should process 'flow'. */
2249 stp_should_process_flow(const struct flow *flow)
2251 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
2255 stp_process_packet(const struct ofport_dpif *ofport,
2256 const struct ofpbuf *packet)
2258 struct ofpbuf payload = *packet;
2259 struct eth_header *eth = payload.data;
2260 struct stp_port *sp = ofport->stp_port;
2262 /* Sink packets on ports that have STP disabled when the bridge has
2264 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
2268 /* Trim off padding on payload. */
2269 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
2270 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
2273 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
2274 stp_received_bpdu(sp, payload.data, payload.size);
2278 static struct priority_to_dscp *
2279 get_priority(const struct ofport_dpif *ofport, uint32_t priority)
2281 struct priority_to_dscp *pdscp;
2284 hash = hash_int(priority, 0);
2285 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) {
2286 if (pdscp->priority == priority) {
2294 ofport_clear_priorities(struct ofport_dpif *ofport)
2296 struct priority_to_dscp *pdscp, *next;
2298 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) {
2299 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
2305 set_queues(struct ofport *ofport_,
2306 const struct ofproto_port_queue *qdscp_list,
2309 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2310 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2311 struct hmap new = HMAP_INITIALIZER(&new);
2314 for (i = 0; i < n_qdscp; i++) {
2315 struct priority_to_dscp *pdscp;
2319 dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
2320 if (dpif_queue_to_priority(ofproto->backer->dpif, qdscp_list[i].queue,
2325 pdscp = get_priority(ofport, priority);
2327 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
2329 pdscp = xmalloc(sizeof *pdscp);
2330 pdscp->priority = priority;
2332 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2335 if (pdscp->dscp != dscp) {
2337 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2340 hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0));
2343 if (!hmap_is_empty(&ofport->priorities)) {
2344 ofport_clear_priorities(ofport);
2345 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2348 hmap_swap(&new, &ofport->priorities);
2356 /* Expires all MAC learning entries associated with 'bundle' and forces its
2357 * ofproto to revalidate every flow.
2359 * Normally MAC learning entries are removed only from the ofproto associated
2360 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2361 * are removed from every ofproto. When patch ports and SLB bonds are in use
2362 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2363 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2364 * with the host from which it migrated. */
2366 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2368 struct ofproto_dpif *ofproto = bundle->ofproto;
2369 struct mac_learning *ml = ofproto->ml;
2370 struct mac_entry *mac, *next_mac;
2372 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2373 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2374 if (mac->port.p == bundle) {
2376 struct ofproto_dpif *o;
2378 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2380 struct mac_entry *e;
2382 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan,
2385 mac_learning_expire(o->ml, e);
2391 mac_learning_expire(ml, mac);
2396 static struct ofbundle *
2397 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2399 struct ofbundle *bundle;
2401 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2402 &ofproto->bundles) {
2403 if (bundle->aux == aux) {
2410 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
2411 * ones that are found to 'bundles'. */
2413 bundle_lookup_multiple(struct ofproto_dpif *ofproto,
2414 void **auxes, size_t n_auxes,
2415 struct hmapx *bundles)
2419 hmapx_init(bundles);
2420 for (i = 0; i < n_auxes; i++) {
2421 struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
2423 hmapx_add(bundles, bundle);
2429 bundle_update(struct ofbundle *bundle)
2431 struct ofport_dpif *port;
2433 bundle->floodable = true;
2434 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2435 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2436 || !stp_forward_in_state(port->stp_state)) {
2437 bundle->floodable = false;
2444 bundle_del_port(struct ofport_dpif *port)
2446 struct ofbundle *bundle = port->bundle;
2448 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2450 list_remove(&port->bundle_node);
2451 port->bundle = NULL;
2454 lacp_slave_unregister(bundle->lacp, port);
2457 bond_slave_unregister(bundle->bond, port);
2460 bundle_update(bundle);
2464 bundle_add_port(struct ofbundle *bundle, uint16_t ofp_port,
2465 struct lacp_slave_settings *lacp)
2467 struct ofport_dpif *port;
2469 port = get_ofp_port(bundle->ofproto, ofp_port);
2474 if (port->bundle != bundle) {
2475 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2477 bundle_del_port(port);
2480 port->bundle = bundle;
2481 list_push_back(&bundle->ports, &port->bundle_node);
2482 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2483 || !stp_forward_in_state(port->stp_state)) {
2484 bundle->floodable = false;
2488 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2489 lacp_slave_register(bundle->lacp, port, lacp);
2496 bundle_destroy(struct ofbundle *bundle)
2498 struct ofproto_dpif *ofproto;
2499 struct ofport_dpif *port, *next_port;
2506 ofproto = bundle->ofproto;
2507 for (i = 0; i < MAX_MIRRORS; i++) {
2508 struct ofmirror *m = ofproto->mirrors[i];
2510 if (m->out == bundle) {
2512 } else if (hmapx_find_and_delete(&m->srcs, bundle)
2513 || hmapx_find_and_delete(&m->dsts, bundle)) {
2514 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2519 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2520 bundle_del_port(port);
2523 bundle_flush_macs(bundle, true);
2524 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2526 free(bundle->trunks);
2527 lacp_destroy(bundle->lacp);
2528 bond_destroy(bundle->bond);
2533 bundle_set(struct ofproto *ofproto_, void *aux,
2534 const struct ofproto_bundle_settings *s)
2536 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2537 bool need_flush = false;
2538 struct ofport_dpif *port;
2539 struct ofbundle *bundle;
2540 unsigned long *trunks;
2546 bundle_destroy(bundle_lookup(ofproto, aux));
2550 ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2551 ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2553 bundle = bundle_lookup(ofproto, aux);
2555 bundle = xmalloc(sizeof *bundle);
2557 bundle->ofproto = ofproto;
2558 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2559 hash_pointer(aux, 0));
2561 bundle->name = NULL;
2563 list_init(&bundle->ports);
2564 bundle->vlan_mode = PORT_VLAN_TRUNK;
2566 bundle->trunks = NULL;
2567 bundle->use_priority_tags = s->use_priority_tags;
2568 bundle->lacp = NULL;
2569 bundle->bond = NULL;
2571 bundle->floodable = true;
2573 bundle->src_mirrors = 0;
2574 bundle->dst_mirrors = 0;
2575 bundle->mirror_out = 0;
2578 if (!bundle->name || strcmp(s->name, bundle->name)) {
2580 bundle->name = xstrdup(s->name);
2585 if (!bundle->lacp) {
2586 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2587 bundle->lacp = lacp_create();
2589 lacp_configure(bundle->lacp, s->lacp);
2591 lacp_destroy(bundle->lacp);
2592 bundle->lacp = NULL;
2595 /* Update set of ports. */
2597 for (i = 0; i < s->n_slaves; i++) {
2598 if (!bundle_add_port(bundle, s->slaves[i],
2599 s->lacp ? &s->lacp_slaves[i] : NULL)) {
2603 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2604 struct ofport_dpif *next_port;
2606 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2607 for (i = 0; i < s->n_slaves; i++) {
2608 if (s->slaves[i] == port->up.ofp_port) {
2613 bundle_del_port(port);
2617 ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2619 if (list_is_empty(&bundle->ports)) {
2620 bundle_destroy(bundle);
2624 /* Set VLAN tagging mode */
2625 if (s->vlan_mode != bundle->vlan_mode
2626 || s->use_priority_tags != bundle->use_priority_tags) {
2627 bundle->vlan_mode = s->vlan_mode;
2628 bundle->use_priority_tags = s->use_priority_tags;
2633 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2634 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2636 if (vlan != bundle->vlan) {
2637 bundle->vlan = vlan;
2641 /* Get trunked VLANs. */
2642 switch (s->vlan_mode) {
2643 case PORT_VLAN_ACCESS:
2647 case PORT_VLAN_TRUNK:
2648 trunks = CONST_CAST(unsigned long *, s->trunks);
2651 case PORT_VLAN_NATIVE_UNTAGGED:
2652 case PORT_VLAN_NATIVE_TAGGED:
2653 if (vlan != 0 && (!s->trunks
2654 || !bitmap_is_set(s->trunks, vlan)
2655 || bitmap_is_set(s->trunks, 0))) {
2656 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2658 trunks = bitmap_clone(s->trunks, 4096);
2660 trunks = bitmap_allocate1(4096);
2662 bitmap_set1(trunks, vlan);
2663 bitmap_set0(trunks, 0);
2665 trunks = CONST_CAST(unsigned long *, s->trunks);
2672 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2673 free(bundle->trunks);
2674 if (trunks == s->trunks) {
2675 bundle->trunks = vlan_bitmap_clone(trunks);
2677 bundle->trunks = trunks;
2682 if (trunks != s->trunks) {
2687 if (!list_is_short(&bundle->ports)) {
2688 bundle->ofproto->has_bonded_bundles = true;
2690 if (bond_reconfigure(bundle->bond, s->bond)) {
2691 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2694 bundle->bond = bond_create(s->bond);
2695 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2698 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2699 bond_slave_register(bundle->bond, port, port->up.netdev);
2702 bond_destroy(bundle->bond);
2703 bundle->bond = NULL;
2706 /* If we changed something that would affect MAC learning, un-learn
2707 * everything on this port and force flow revalidation. */
2709 bundle_flush_macs(bundle, false);
2716 bundle_remove(struct ofport *port_)
2718 struct ofport_dpif *port = ofport_dpif_cast(port_);
2719 struct ofbundle *bundle = port->bundle;
2722 bundle_del_port(port);
2723 if (list_is_empty(&bundle->ports)) {
2724 bundle_destroy(bundle);
2725 } else if (list_is_short(&bundle->ports)) {
2726 bond_destroy(bundle->bond);
2727 bundle->bond = NULL;
2733 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2735 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2736 struct ofport_dpif *port = port_;
2737 uint8_t ea[ETH_ADDR_LEN];
2740 error = netdev_get_etheraddr(port->up.netdev, ea);
2742 struct ofpbuf packet;
2745 ofpbuf_init(&packet, 0);
2746 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2748 memcpy(packet_pdu, pdu, pdu_size);
2750 send_packet(port, &packet);
2751 ofpbuf_uninit(&packet);
2753 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2754 "%s (%s)", port->bundle->name,
2755 netdev_get_name(port->up.netdev), strerror(error));
2760 bundle_send_learning_packets(struct ofbundle *bundle)
2762 struct ofproto_dpif *ofproto = bundle->ofproto;
2763 int error, n_packets, n_errors;
2764 struct mac_entry *e;
2766 error = n_packets = n_errors = 0;
2767 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2768 if (e->port.p != bundle) {
2769 struct ofpbuf *learning_packet;
2770 struct ofport_dpif *port;
2774 /* The assignment to "port" is unnecessary but makes "grep"ing for
2775 * struct ofport_dpif more effective. */
2776 learning_packet = bond_compose_learning_packet(bundle->bond,
2780 ret = send_packet(port, learning_packet);
2781 ofpbuf_delete(learning_packet);
2791 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2792 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2793 "packets, last error was: %s",
2794 bundle->name, n_errors, n_packets, strerror(error));
2796 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2797 bundle->name, n_packets);
2802 bundle_run(struct ofbundle *bundle)
2805 lacp_run(bundle->lacp, send_pdu_cb);
2808 struct ofport_dpif *port;
2810 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2811 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2814 bond_run(bundle->bond, &bundle->ofproto->backer->revalidate_set,
2815 lacp_status(bundle->lacp));
2816 if (bond_should_send_learning_packets(bundle->bond)) {
2817 bundle_send_learning_packets(bundle);
2823 bundle_wait(struct ofbundle *bundle)
2826 lacp_wait(bundle->lacp);
2829 bond_wait(bundle->bond);
2836 mirror_scan(struct ofproto_dpif *ofproto)
2840 for (idx = 0; idx < MAX_MIRRORS; idx++) {
2841 if (!ofproto->mirrors[idx]) {
2848 static struct ofmirror *
2849 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
2853 for (i = 0; i < MAX_MIRRORS; i++) {
2854 struct ofmirror *mirror = ofproto->mirrors[i];
2855 if (mirror && mirror->aux == aux) {
2863 /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */
2865 mirror_update_dups(struct ofproto_dpif *ofproto)
2869 for (i = 0; i < MAX_MIRRORS; i++) {
2870 struct ofmirror *m = ofproto->mirrors[i];
2873 m->dup_mirrors = MIRROR_MASK_C(1) << i;
2877 for (i = 0; i < MAX_MIRRORS; i++) {
2878 struct ofmirror *m1 = ofproto->mirrors[i];
2885 for (j = i + 1; j < MAX_MIRRORS; j++) {
2886 struct ofmirror *m2 = ofproto->mirrors[j];
2888 if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) {
2889 m1->dup_mirrors |= MIRROR_MASK_C(1) << j;
2890 m2->dup_mirrors |= m1->dup_mirrors;
2897 mirror_set(struct ofproto *ofproto_, void *aux,
2898 const struct ofproto_mirror_settings *s)
2900 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2901 mirror_mask_t mirror_bit;
2902 struct ofbundle *bundle;
2903 struct ofmirror *mirror;
2904 struct ofbundle *out;
2905 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
2906 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
2909 mirror = mirror_lookup(ofproto, aux);
2911 mirror_destroy(mirror);
2917 idx = mirror_scan(ofproto);
2919 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
2921 ofproto->up.name, MAX_MIRRORS, s->name);
2925 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
2926 mirror->ofproto = ofproto;
2929 mirror->out_vlan = -1;
2930 mirror->name = NULL;
2933 if (!mirror->name || strcmp(s->name, mirror->name)) {
2935 mirror->name = xstrdup(s->name);
2938 /* Get the new configuration. */
2939 if (s->out_bundle) {
2940 out = bundle_lookup(ofproto, s->out_bundle);
2942 mirror_destroy(mirror);
2948 out_vlan = s->out_vlan;
2950 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
2951 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
2953 /* If the configuration has not changed, do nothing. */
2954 if (hmapx_equals(&srcs, &mirror->srcs)
2955 && hmapx_equals(&dsts, &mirror->dsts)
2956 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
2957 && mirror->out == out
2958 && mirror->out_vlan == out_vlan)
2960 hmapx_destroy(&srcs);
2961 hmapx_destroy(&dsts);
2965 hmapx_swap(&srcs, &mirror->srcs);
2966 hmapx_destroy(&srcs);
2968 hmapx_swap(&dsts, &mirror->dsts);
2969 hmapx_destroy(&dsts);
2971 free(mirror->vlans);
2972 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
2975 mirror->out_vlan = out_vlan;
2977 /* Update bundles. */
2978 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2979 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
2980 if (hmapx_contains(&mirror->srcs, bundle)) {
2981 bundle->src_mirrors |= mirror_bit;
2983 bundle->src_mirrors &= ~mirror_bit;
2986 if (hmapx_contains(&mirror->dsts, bundle)) {
2987 bundle->dst_mirrors |= mirror_bit;
2989 bundle->dst_mirrors &= ~mirror_bit;
2992 if (mirror->out == bundle) {
2993 bundle->mirror_out |= mirror_bit;
2995 bundle->mirror_out &= ~mirror_bit;
2999 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3000 ofproto->has_mirrors = true;
3001 mac_learning_flush(ofproto->ml,
3002 &ofproto->backer->revalidate_set);
3003 mirror_update_dups(ofproto);
3009 mirror_destroy(struct ofmirror *mirror)
3011 struct ofproto_dpif *ofproto;
3012 mirror_mask_t mirror_bit;
3013 struct ofbundle *bundle;
3020 ofproto = mirror->ofproto;
3021 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3022 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
3024 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
3025 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
3026 bundle->src_mirrors &= ~mirror_bit;
3027 bundle->dst_mirrors &= ~mirror_bit;
3028 bundle->mirror_out &= ~mirror_bit;
3031 hmapx_destroy(&mirror->srcs);
3032 hmapx_destroy(&mirror->dsts);
3033 free(mirror->vlans);
3035 ofproto->mirrors[mirror->idx] = NULL;
3039 mirror_update_dups(ofproto);
3041 ofproto->has_mirrors = false;
3042 for (i = 0; i < MAX_MIRRORS; i++) {
3043 if (ofproto->mirrors[i]) {
3044 ofproto->has_mirrors = true;
3051 mirror_get_stats(struct ofproto *ofproto_, void *aux,
3052 uint64_t *packets, uint64_t *bytes)
3054 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3055 struct ofmirror *mirror = mirror_lookup(ofproto, aux);
3058 *packets = *bytes = UINT64_MAX;
3064 *packets = mirror->packet_count;
3065 *bytes = mirror->byte_count;
3071 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
3073 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3074 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
3075 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
3081 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
3083 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3084 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
3085 return bundle && bundle->mirror_out != 0;
3089 forward_bpdu_changed(struct ofproto *ofproto_)
3091 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3092 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3096 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
3099 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3100 mac_learning_set_idle_time(ofproto->ml, idle_time);
3101 mac_learning_set_max_entries(ofproto->ml, max_entries);
3106 static struct ofport_dpif *
3107 get_ofp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
3109 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
3110 return ofport ? ofport_dpif_cast(ofport) : NULL;
3113 static struct ofport_dpif *
3114 get_odp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
3116 struct ofport_dpif *port = odp_port_to_ofport(ofproto->backer, odp_port);
3117 return port && &ofproto->up == port->up.ofproto ? port : NULL;
3121 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
3122 struct ofproto_port *ofproto_port,
3123 struct dpif_port *dpif_port)
3125 ofproto_port->name = dpif_port->name;
3126 ofproto_port->type = dpif_port->type;
3127 ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
3130 static struct ofport_dpif *
3131 ofport_get_peer(const struct ofport_dpif *ofport_dpif)
3133 const struct ofproto_dpif *ofproto;
3136 peer = netdev_vport_patch_peer(ofport_dpif->up.netdev);
3141 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
3142 struct ofport *ofport;
3144 ofport = shash_find_data(&ofproto->up.port_by_name, peer);
3145 if (ofport && ofport->ofproto->ofproto_class == &ofproto_dpif_class) {
3146 return ofport_dpif_cast(ofport);
3153 port_run_fast(struct ofport_dpif *ofport)
3155 if (ofport->cfm && cfm_should_send_ccm(ofport->cfm)) {
3156 struct ofpbuf packet;
3158 ofpbuf_init(&packet, 0);
3159 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
3160 send_packet(ofport, &packet);
3161 ofpbuf_uninit(&packet);
3164 if (ofport->bfd && bfd_should_send_packet(ofport->bfd)) {
3165 struct ofpbuf packet;
3167 ofpbuf_init(&packet, 0);
3168 bfd_put_packet(ofport->bfd, &packet, ofport->up.pp.hw_addr);
3169 send_packet(ofport, &packet);
3170 ofpbuf_uninit(&packet);
3175 port_run(struct ofport_dpif *ofport)
3177 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
3178 bool carrier_changed = carrier_seq != ofport->carrier_seq;
3179 bool enable = netdev_get_carrier(ofport->up.netdev);
3181 ofport->carrier_seq = carrier_seq;
3183 port_run_fast(ofport);
3185 if (ofport->tnl_port
3186 && tnl_port_reconfigure(&ofport->up, ofport->odp_port,
3187 &ofport->tnl_port)) {
3188 ofproto_dpif_cast(ofport->up.ofproto)->backer->need_revalidate = true;
3192 int cfm_opup = cfm_get_opup(ofport->cfm);
3194 cfm_run(ofport->cfm);
3195 enable = enable && !cfm_get_fault(ofport->cfm);
3197 if (cfm_opup >= 0) {
3198 enable = enable && cfm_opup;
3203 bfd_run(ofport->bfd);
3204 enable = enable && bfd_forwarding(ofport->bfd);
3207 if (ofport->bundle) {
3208 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
3209 if (carrier_changed) {
3210 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
3214 if (ofport->may_enable != enable) {
3215 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3217 if (ofproto->has_bundle_action) {
3218 ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
3222 ofport->may_enable = enable;
3226 port_wait(struct ofport_dpif *ofport)
3229 cfm_wait(ofport->cfm);
3233 bfd_wait(ofport->bfd);
3238 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
3239 struct ofproto_port *ofproto_port)
3241 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3242 struct dpif_port dpif_port;
3245 if (sset_contains(&ofproto->ghost_ports, devname)) {
3246 const char *type = netdev_get_type_from_name(devname);
3248 /* We may be called before ofproto->up.port_by_name is populated with
3249 * the appropriate ofport. For this reason, we must get the name and
3250 * type from the netdev layer directly. */
3252 const struct ofport *ofport;
3254 ofport = shash_find_data(&ofproto->up.port_by_name, devname);
3255 ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
3256 ofproto_port->name = xstrdup(devname);
3257 ofproto_port->type = xstrdup(type);
3263 if (!sset_contains(&ofproto->ports, devname)) {
3266 error = dpif_port_query_by_name(ofproto->backer->dpif,
3267 devname, &dpif_port);
3269 ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
3275 port_add(struct ofproto *ofproto_, struct netdev *netdev)
3277 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3278 const char *dp_port_name = netdev_vport_get_dpif_port(netdev);
3279 const char *devname = netdev_get_name(netdev);
3281 if (netdev_vport_is_patch(netdev)) {
3282 sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
3286 if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
3287 uint32_t port_no = UINT32_MAX;
3290 error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
3294 if (netdev_get_tunnel_config(netdev)) {
3295 simap_put(&ofproto->backer->tnl_backers, dp_port_name, port_no);
3299 if (netdev_get_tunnel_config(netdev)) {
3300 sset_add(&ofproto->ghost_ports, devname);
3302 sset_add(&ofproto->ports, devname);
3308 port_del(struct ofproto *ofproto_, uint16_t ofp_port)
3310 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3311 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
3318 sset_find_and_delete(&ofproto->ghost_ports,
3319 netdev_get_name(ofport->up.netdev));
3320 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3321 if (!ofport->tnl_port) {
3322 error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
3324 /* The caller is going to close ofport->up.netdev. If this is a
3325 * bonded port, then the bond is using that netdev, so remove it
3326 * from the bond. The client will need to reconfigure everything
3327 * after deleting ports, so then the slave will get re-added. */
3328 bundle_remove(&ofport->up);
3335 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
3337 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3342 error = netdev_get_stats(ofport->up.netdev, stats);
3344 if (!error && ofport_->ofp_port == OFPP_LOCAL) {
3345 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3347 /* ofproto->stats.tx_packets represents packets that we created
3348 * internally and sent to some port (e.g. packets sent with
3349 * send_packet()). Account for them as if they had come from
3350 * OFPP_LOCAL and got forwarded. */
3352 if (stats->rx_packets != UINT64_MAX) {
3353 stats->rx_packets += ofproto->stats.tx_packets;
3356 if (stats->rx_bytes != UINT64_MAX) {
3357 stats->rx_bytes += ofproto->stats.tx_bytes;
3360 /* ofproto->stats.rx_packets represents packets that were received on
3361 * some port and we processed internally and dropped (e.g. STP).
3362 * Account for them as if they had been forwarded to OFPP_LOCAL. */
3364 if (stats->tx_packets != UINT64_MAX) {
3365 stats->tx_packets += ofproto->stats.rx_packets;
3368 if (stats->tx_bytes != UINT64_MAX) {
3369 stats->tx_bytes += ofproto->stats.rx_bytes;
3376 struct port_dump_state {
3381 struct ofproto_port port;
3386 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
3388 *statep = xzalloc(sizeof(struct port_dump_state));
3393 port_dump_next(const struct ofproto *ofproto_, void *state_,
3394 struct ofproto_port *port)
3396 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3397 struct port_dump_state *state = state_;
3398 const struct sset *sset;
3399 struct sset_node *node;
3401 if (state->has_port) {
3402 ofproto_port_destroy(&state->port);
3403 state->has_port = false;
3405 sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
3406 while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
3409 error = port_query_by_name(ofproto_, node->name, &state->port);
3411 *port = state->port;
3412 state->has_port = true;
3414 } else if (error != ENODEV) {
3419 if (!state->ghost) {
3420 state->ghost = true;
3423 return port_dump_next(ofproto_, state_, port);
3430 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3432 struct port_dump_state *state = state_;
3434 if (state->has_port) {
3435 ofproto_port_destroy(&state->port);
3442 port_poll(const struct ofproto *ofproto_, char **devnamep)
3444 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3446 if (ofproto->port_poll_errno) {
3447 int error = ofproto->port_poll_errno;
3448 ofproto->port_poll_errno = 0;
3452 if (sset_is_empty(&ofproto->port_poll_set)) {
3456 *devnamep = sset_pop(&ofproto->port_poll_set);
3461 port_poll_wait(const struct ofproto *ofproto_)
3463 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3464 dpif_port_poll_wait(ofproto->backer->dpif);
3468 port_is_lacp_current(const struct ofport *ofport_)
3470 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3471 return (ofport->bundle && ofport->bundle->lacp
3472 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3476 /* Upcall handling. */
3478 /* Flow miss batching.
3480 * Some dpifs implement operations faster when you hand them off in a batch.
3481 * To allow batching, "struct flow_miss" queues the dpif-related work needed
3482 * for a given flow. Each "struct flow_miss" corresponds to sending one or
3483 * more packets, plus possibly installing the flow in the dpif.
3485 * So far we only batch the operations that affect flow setup time the most.
3486 * It's possible to batch more than that, but the benefit might be minimal. */
3488 struct hmap_node hmap_node;
3489 struct ofproto_dpif *ofproto;
3491 enum odp_key_fitness key_fitness;
3492 const struct nlattr *key;
3494 struct initial_vals initial_vals;
3495 struct list packets;
3496 enum dpif_upcall_type upcall_type;
3497 uint32_t odp_in_port;
3500 struct flow_miss_op {
3501 struct dpif_op dpif_op;
3503 uint64_t slow_stub[128 / 8]; /* Buffer for compose_slow_path() */
3504 struct xlate_out xout;
3505 bool xout_garbage; /* 'xout' needs to be uninitialized? */
3508 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
3509 * OpenFlow controller as necessary according to their individual
3510 * configurations. */
3512 send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet,
3513 const struct flow *flow)
3515 struct ofputil_packet_in pin;
3517 pin.packet = packet->data;
3518 pin.packet_len = packet->size;
3519 pin.reason = OFPR_NO_MATCH;
3520 pin.controller_id = 0;
3525 pin.send_len = 0; /* not used for flow table misses */
3527 flow_get_metadata(flow, &pin.fmd);
3529 connmgr_send_packet_in(ofproto->up.connmgr, &pin);
3532 static enum slow_path_reason
3533 process_special(struct ofproto_dpif *ofproto, const struct flow *flow,
3534 const struct ofport_dpif *ofport, const struct ofpbuf *packet)
3538 } else if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) {
3540 cfm_process_heartbeat(ofport->cfm, packet);
3543 } else if (ofport->bfd && bfd_should_process_flow(flow)) {
3545 bfd_process_packet(ofport->bfd, flow, packet);
3548 } else if (ofport->bundle && ofport->bundle->lacp
3549 && flow->dl_type == htons(ETH_TYPE_LACP)) {
3551 lacp_process_packet(ofport->bundle->lacp, ofport, packet);
3554 } else if (ofproto->stp && stp_should_process_flow(flow)) {
3556 stp_process_packet(ofport, packet);
3564 static struct flow_miss *
3565 flow_miss_find(struct hmap *todo, const struct ofproto_dpif *ofproto,
3566 const struct flow *flow, uint32_t hash)
3568 struct flow_miss *miss;
3570 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
3571 if (miss->ofproto == ofproto && flow_equal(&miss->flow, flow)) {
3579 /* Partially Initializes 'op' as an "execute" operation for 'miss' and
3580 * 'packet'. The caller must initialize op->actions and op->actions_len. If
3581 * 'miss' is associated with a subfacet the caller must also initialize the
3582 * returned op->subfacet, and if anything needs to be freed after processing
3583 * the op, the caller must initialize op->garbage also. */
3585 init_flow_miss_execute_op(struct flow_miss *miss, struct ofpbuf *packet,
3586 struct flow_miss_op *op)
3588 if (miss->flow.vlan_tci != miss->initial_vals.vlan_tci) {
3589 /* This packet was received on a VLAN splinter port. We
3590 * added a VLAN to the packet to make the packet resemble
3591 * the flow, but the actions were composed assuming that
3592 * the packet contained no VLAN. So, we must remove the
3593 * VLAN header from the packet before trying to execute the
3595 eth_pop_vlan(packet);
3598 op->xout_garbage = false;
3599 op->dpif_op.type = DPIF_OP_EXECUTE;
3600 op->dpif_op.u.execute.key = miss->key;
3601 op->dpif_op.u.execute.key_len = miss->key_len;
3602 op->dpif_op.u.execute.packet = packet;
3605 /* Helper for handle_flow_miss_without_facet() and
3606 * handle_flow_miss_with_facet(). */
3608 handle_flow_miss_common(struct rule_dpif *rule,
3609 struct ofpbuf *packet, const struct flow *flow)
3611 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3613 if (rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
3615 * Extra-special case for fail-open mode.
3617 * We are in fail-open mode and the packet matched the fail-open
3618 * rule, but we are connected to a controller too. We should send
3619 * the packet up to the controller in the hope that it will try to
3620 * set up a flow and thereby allow us to exit fail-open.
3622 * See the top-level comment in fail-open.c for more information.
3624 send_packet_in_miss(ofproto, packet, flow);
3628 /* Figures out whether a flow that missed in 'ofproto', whose details are in
3629 * 'miss', is likely to be worth tracking in detail in userspace and (usually)
3630 * installing a datapath flow. The answer is usually "yes" (a return value of
3631 * true). However, for short flows the cost of bookkeeping is much higher than
3632 * the benefits, so when the datapath holds a large number of flows we impose
3633 * some heuristics to decide which flows are likely to be worth tracking. */
3635 flow_miss_should_make_facet(struct ofproto_dpif *ofproto,
3636 struct flow_miss *miss, uint32_t hash)
3638 if (!ofproto->governor) {
3641 n_subfacets = hmap_count(&ofproto->subfacets);
3642 if (n_subfacets * 2 <= ofproto->up.flow_eviction_threshold) {
3646 ofproto->governor = governor_create(ofproto->up.name);
3649 return governor_should_install_flow(ofproto->governor, hash,
3650 list_size(&miss->packets));
3653 /* Handles 'miss', which matches 'rule', without creating a facet or subfacet
3654 * or creating any datapath flow. May add an "execute" operation to 'ops' and
3655 * increment '*n_ops'. */
3657 handle_flow_miss_without_facet(struct flow_miss *miss,
3658 struct flow_miss_op *ops, size_t *n_ops)
3660 struct rule_dpif *rule = rule_dpif_lookup(miss->ofproto, &miss->flow);
3661 long long int now = time_msec();
3662 struct ofpbuf *packet;
3663 struct xlate_in xin;
3665 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3666 struct flow_miss_op *op = &ops[*n_ops];
3667 struct dpif_flow_stats stats;
3669 COVERAGE_INC(facet_suppress);
3671 handle_flow_miss_common(rule, packet, &miss->flow);
3673 dpif_flow_stats_extract(&miss->flow, packet, now, &stats);
3674 rule_credit_stats(rule, &stats);
3676 xlate_in_init(&xin, miss->ofproto, &miss->flow, &miss->initial_vals,
3677 rule, stats.tcp_flags, packet);
3678 xin.resubmit_stats = &stats;
3679 xlate_actions(&xin, &op->xout);
3681 if (op->xout.odp_actions.size) {
3682 struct dpif_execute *execute = &op->dpif_op.u.execute;
3684 init_flow_miss_execute_op(miss, packet, op);
3685 execute->actions = op->xout.odp_actions.data;
3686 execute->actions_len = op->xout.odp_actions.size;
3687 op->xout_garbage = true;
3691 xlate_out_uninit(&op->xout);
3696 /* Handles 'miss', which matches 'facet'. May add any required datapath
3697 * operations to 'ops', incrementing '*n_ops' for each new op.
3699 * All of the packets in 'miss' are considered to have arrived at time 'now'.
3700 * This is really important only for new facets: if we just called time_msec()
3701 * here, then the new subfacet or its packets could look (occasionally) as
3702 * though it was used some time after the facet was used. That can make a
3703 * one-packet flow look like it has a nonzero duration, which looks odd in
3704 * e.g. NetFlow statistics. */
3706 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
3708 struct flow_miss_op *ops, size_t *n_ops)
3710 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3711 enum subfacet_path want_path;
3712 struct subfacet *subfacet;
3713 struct ofpbuf *packet;
3715 subfacet = subfacet_create(facet, miss, now);
3716 want_path = subfacet->facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
3718 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3719 struct flow_miss_op *op = &ops[*n_ops];
3720 struct dpif_flow_stats stats;
3722 handle_flow_miss_common(facet->rule, packet, &miss->flow);
3724 if (want_path != SF_FAST_PATH) {
3725 struct xlate_in xin;
3727 xlate_in_init(&xin, ofproto, &facet->flow, &facet->initial_vals,
3728 facet->rule, 0, packet);
3729 xlate_actions_for_side_effects(&xin);
3732 dpif_flow_stats_extract(&facet->flow, packet, now, &stats);
3733 subfacet_update_stats(subfacet, &stats);
3735 if (facet->xout.odp_actions.size) {
3736 struct dpif_execute *execute = &op->dpif_op.u.execute;
3738 init_flow_miss_execute_op(miss, packet, op);
3739 execute->actions = facet->xout.odp_actions.data,
3740 execute->actions_len = facet->xout.odp_actions.size;
3745 if (miss->upcall_type == DPIF_UC_MISS || subfacet->path != want_path) {
3746 struct flow_miss_op *op = &ops[(*n_ops)++];
3747 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
3749 subfacet->path = want_path;
3751 op->xout_garbage = false;
3752 op->dpif_op.type = DPIF_OP_FLOW_PUT;
3753 put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
3754 put->key = miss->key;
3755 put->key_len = miss->key_len;
3756 if (want_path == SF_FAST_PATH) {
3757 put->actions = facet->xout.odp_actions.data;
3758 put->actions_len = facet->xout.odp_actions.size;
3760 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
3761 op->slow_stub, sizeof op->slow_stub,
3762 &put->actions, &put->actions_len);
3768 /* Handles flow miss 'miss'. May add any required datapath operations
3769 * to 'ops', incrementing '*n_ops' for each new op. */
3771 handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops,
3774 struct ofproto_dpif *ofproto = miss->ofproto;
3775 struct facet *facet;
3779 /* The caller must ensure that miss->hmap_node.hash contains
3780 * flow_hash(miss->flow, 0). */
3781 hash = miss->hmap_node.hash;
3783 facet = facet_lookup_valid(ofproto, &miss->flow, hash);
3785 /* There does not exist a bijection between 'struct flow' and datapath
3786 * flow keys with fitness ODP_FIT_TO_LITTLE. This breaks a fundamental
3787 * assumption used throughout the facet and subfacet handling code.
3788 * Since we have to handle these misses in userspace anyway, we simply
3789 * skip facet creation, avoiding the problem alltogether. */
3790 if (miss->key_fitness == ODP_FIT_TOO_LITTLE
3791 || !flow_miss_should_make_facet(ofproto, miss, hash)) {
3792 handle_flow_miss_without_facet(miss, ops, n_ops);
3796 facet = facet_create(miss, hash);
3801 handle_flow_miss_with_facet(miss, facet, now, ops, n_ops);
3804 static struct drop_key *
3805 drop_key_lookup(const struct dpif_backer *backer, const struct nlattr *key,
3808 struct drop_key *drop_key;
3810 HMAP_FOR_EACH_WITH_HASH (drop_key, hmap_node, hash_bytes(key, key_len, 0),
3811 &backer->drop_keys) {
3812 if (drop_key->key_len == key_len
3813 && !memcmp(drop_key->key, key, key_len)) {
3821 drop_key_clear(struct dpif_backer *backer)
3823 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3824 struct drop_key *drop_key, *next;
3826 HMAP_FOR_EACH_SAFE (drop_key, next, hmap_node, &backer->drop_keys) {
3829 error = dpif_flow_del(backer->dpif, drop_key->key, drop_key->key_len,
3831 if (error && !VLOG_DROP_WARN(&rl)) {
3832 struct ds ds = DS_EMPTY_INITIALIZER;
3833 odp_flow_key_format(drop_key->key, drop_key->key_len, &ds);
3834 VLOG_WARN("Failed to delete drop key (%s) (%s)", strerror(error),
3839 hmap_remove(&backer->drop_keys, &drop_key->hmap_node);
3840 free(drop_key->key);
3845 /* Given a datpath, packet, and flow metadata ('backer', 'packet', and 'key'
3846 * respectively), populates 'flow' with the result of odp_flow_key_to_flow().
3847 * Optionally, if nonnull, populates 'fitnessp' with the fitness of 'flow' as
3848 * returned by odp_flow_key_to_flow(). Also, optionally populates 'ofproto'
3849 * with the ofproto_dpif, and 'odp_in_port' with the datapath in_port, that
3850 * 'packet' ingressed.
3852 * If 'ofproto' is nonnull, requires 'flow''s in_port to exist. Otherwise sets
3853 * 'flow''s in_port to OFPP_NONE.
3855 * This function does post-processing on data returned from
3856 * odp_flow_key_to_flow() to help make VLAN splinters transparent to the rest
3857 * of the upcall processing logic. In particular, if the extracted in_port is
3858 * a VLAN splinter port, it replaces flow->in_port by the "real" port, sets
3859 * flow->vlan_tci correctly for the VLAN of the VLAN splinter port, and pushes
3860 * a VLAN header onto 'packet' (if it is nonnull).
3862 * Optionally, if 'initial_vals' is nonnull, sets 'initial_vals->vlan_tci'
3863 * to the VLAN TCI with which the packet was really received, that is, the
3864 * actual VLAN TCI extracted by odp_flow_key_to_flow(). (This differs from
3865 * the value returned in flow->vlan_tci only for packets received on
3868 * Similarly, this function also includes some logic to help with tunnels. It
3869 * may modify 'flow' as necessary to make the tunneling implementation
3870 * transparent to the upcall processing logic.
3872 * Returns 0 if successful, ENODEV if the parsed flow has no associated ofport,
3873 * or some other positive errno if there are other problems. */
3875 ofproto_receive(const struct dpif_backer *backer, struct ofpbuf *packet,
3876 const struct nlattr *key, size_t key_len,
3877 struct flow *flow, enum odp_key_fitness *fitnessp,
3878 struct ofproto_dpif **ofproto, uint32_t *odp_in_port,
3879 struct initial_vals *initial_vals)
3881 const struct ofport_dpif *port;
3882 enum odp_key_fitness fitness;
3885 fitness = odp_flow_key_to_flow(key, key_len, flow);
3886 if (fitness == ODP_FIT_ERROR) {
3892 initial_vals->vlan_tci = flow->vlan_tci;
3896 *odp_in_port = flow->in_port;
3899 port = (tnl_port_should_receive(flow)
3900 ? ofport_dpif_cast(tnl_port_receive(flow))
3901 : odp_port_to_ofport(backer, flow->in_port));
3902 flow->in_port = port ? port->up.ofp_port : OFPP_NONE;
3907 /* XXX: Since the tunnel module is not scoped per backer, for a tunnel port
3908 * it's theoretically possible that we'll receive an ofport belonging to an
3909 * entirely different datapath. In practice, this can't happen because no
3910 * platforms has two separate datapaths which each support tunneling. */
3911 ovs_assert(ofproto_dpif_cast(port->up.ofproto)->backer == backer);
3913 if (vsp_adjust_flow(ofproto_dpif_cast(port->up.ofproto), flow)) {
3915 /* Make the packet resemble the flow, so that it gets sent to
3916 * an OpenFlow controller properly, so that it looks correct
3917 * for sFlow, and so that flow_extract() will get the correct
3918 * vlan_tci if it is called on 'packet'.
3920 * The allocated space inside 'packet' probably also contains
3921 * 'key', that is, both 'packet' and 'key' are probably part of
3922 * a struct dpif_upcall (see the large comment on that
3923 * structure definition), so pushing data on 'packet' is in
3924 * general not a good idea since it could overwrite 'key' or
3925 * free it as a side effect. However, it's OK in this special
3926 * case because we know that 'packet' is inside a Netlink
3927 * attribute: pushing 4 bytes will just overwrite the 4-byte
3928 * "struct nlattr", which is fine since we don't need that
3929 * header anymore. */
3930 eth_push_vlan(packet, flow->vlan_tci);
3932 /* We can't reproduce 'key' from 'flow'. */
3933 fitness = fitness == ODP_FIT_PERFECT ? ODP_FIT_TOO_MUCH : fitness;
3938 *ofproto = ofproto_dpif_cast(port->up.ofproto);
3943 *fitnessp = fitness;
3949 handle_miss_upcalls(struct dpif_backer *backer, struct dpif_upcall *upcalls,
3952 struct dpif_upcall *upcall;
3953 struct flow_miss *miss;
3954 struct flow_miss misses[FLOW_MISS_MAX_BATCH];
3955 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
3956 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
3966 /* Construct the to-do list.
3968 * This just amounts to extracting the flow from each packet and sticking
3969 * the packets that have the same flow in the same "flow_miss" structure so
3970 * that we can process them together. */
3973 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
3974 struct flow_miss *miss = &misses[n_misses];
3975 struct flow_miss *existing_miss;
3976 struct ofproto_dpif *ofproto;
3977 uint32_t odp_in_port;
3982 error = ofproto_receive(backer, upcall->packet, upcall->key,
3983 upcall->key_len, &flow, &miss->key_fitness,
3984 &ofproto, &odp_in_port, &miss->initial_vals);
3985 if (error == ENODEV) {
3986 struct drop_key *drop_key;
3988 /* Received packet on port for which we couldn't associate
3989 * an ofproto. This can happen if a port is removed while
3990 * traffic is being received. Print a rate-limited message
3991 * in case it happens frequently. Install a drop flow so
3992 * that future packets of the flow are inexpensively dropped
3994 VLOG_INFO_RL(&rl, "received packet on unassociated port %"PRIu32,
3997 drop_key = drop_key_lookup(backer, upcall->key, upcall->key_len);
3999 drop_key = xmalloc(sizeof *drop_key);
4000 drop_key->key = xmemdup(upcall->key, upcall->key_len);
4001 drop_key->key_len = upcall->key_len;
4003 hmap_insert(&backer->drop_keys, &drop_key->hmap_node,
4004 hash_bytes(drop_key->key, drop_key->key_len, 0));
4005 dpif_flow_put(backer->dpif, DPIF_FP_CREATE | DPIF_FP_MODIFY,
4006 drop_key->key, drop_key->key_len, NULL, 0, NULL);
4014 ofproto->n_missed++;
4015 flow_extract(upcall->packet, flow.skb_priority, flow.skb_mark,
4016 &flow.tunnel, flow.in_port, &miss->flow);
4018 /* Add other packets to a to-do list. */
4019 hash = flow_hash(&miss->flow, 0);
4020 existing_miss = flow_miss_find(&todo, ofproto, &miss->flow, hash);
4021 if (!existing_miss) {
4022 hmap_insert(&todo, &miss->hmap_node, hash);
4023 miss->ofproto = ofproto;
4024 miss->key = upcall->key;
4025 miss->key_len = upcall->key_len;
4026 miss->upcall_type = upcall->type;
4027 miss->odp_in_port = odp_in_port;
4028 list_init(&miss->packets);
4032 miss = existing_miss;
4034 list_push_back(&miss->packets, &upcall->packet->list_node);
4037 /* Process each element in the to-do list, constructing the set of
4038 * operations to batch. */
4040 HMAP_FOR_EACH (miss, hmap_node, &todo) {
4041 handle_flow_miss(miss, flow_miss_ops, &n_ops);
4043 ovs_assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
4045 /* Execute batch. */
4046 for (i = 0; i < n_ops; i++) {
4047 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
4049 dpif_operate(backer->dpif, dpif_ops, n_ops);
4052 for (i = 0; i < n_ops; i++) {
4053 if (flow_miss_ops[i].xout_garbage) {
4054 xlate_out_uninit(&flow_miss_ops[i].xout);
4057 hmap_destroy(&todo);
4060 static enum { SFLOW_UPCALL, MISS_UPCALL, BAD_UPCALL, FLOW_SAMPLE_UPCALL,
4062 classify_upcall(const struct dpif_upcall *upcall)
4064 size_t userdata_len;
4065 union user_action_cookie cookie;
4067 /* First look at the upcall type. */
4068 switch (upcall->type) {
4069 case DPIF_UC_ACTION:
4075 case DPIF_N_UC_TYPES:
4077 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
4081 /* "action" upcalls need a closer look. */
4082 if (!upcall->userdata) {
4083 VLOG_WARN_RL(&rl, "action upcall missing cookie");
4086 userdata_len = nl_attr_get_size(upcall->userdata);
4087 if (userdata_len < sizeof cookie.type
4088 || userdata_len > sizeof cookie) {
4089 VLOG_WARN_RL(&rl, "action upcall cookie has unexpected size %zu",
4093 memset(&cookie, 0, sizeof cookie);
4094 memcpy(&cookie, nl_attr_get(upcall->userdata), userdata_len);
4095 if (userdata_len == sizeof cookie.sflow
4096 && cookie.type == USER_ACTION_COOKIE_SFLOW) {
4097 return SFLOW_UPCALL;
4098 } else if (userdata_len == sizeof cookie.slow_path
4099 && cookie.type == USER_ACTION_COOKIE_SLOW_PATH) {
4101 } else if (userdata_len == sizeof cookie.flow_sample
4102 && cookie.type == USER_ACTION_COOKIE_FLOW_SAMPLE) {
4103 return FLOW_SAMPLE_UPCALL;
4104 } else if (userdata_len == sizeof cookie.ipfix
4105 && cookie.type == USER_ACTION_COOKIE_IPFIX) {
4106 return IPFIX_UPCALL;
4108 VLOG_WARN_RL(&rl, "invalid user cookie of type %"PRIu16
4109 " and size %zu", cookie.type, userdata_len);
4115 handle_sflow_upcall(struct dpif_backer *backer,
4116 const struct dpif_upcall *upcall)
4118 struct ofproto_dpif *ofproto;
4119 union user_action_cookie cookie;
4121 uint32_t odp_in_port;
4123 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
4124 &flow, NULL, &ofproto, &odp_in_port, NULL)
4125 || !ofproto->sflow) {
4129 memset(&cookie, 0, sizeof cookie);
4130 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.sflow);
4131 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
4132 odp_in_port, &cookie);
4136 handle_flow_sample_upcall(struct dpif_backer *backer,
4137 const struct dpif_upcall *upcall)
4139 struct ofproto_dpif *ofproto;
4140 union user_action_cookie cookie;
4143 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
4144 &flow, NULL, &ofproto, NULL, NULL)
4145 || !ofproto->ipfix) {
4149 memset(&cookie, 0, sizeof cookie);
4150 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.flow_sample);
4152 /* The flow reflects exactly the contents of the packet. Sample
4153 * the packet using it. */
4154 dpif_ipfix_flow_sample(ofproto->ipfix, upcall->packet, &flow,
4155 cookie.flow_sample.collector_set_id,
4156 cookie.flow_sample.probability,
4157 cookie.flow_sample.obs_domain_id,
4158 cookie.flow_sample.obs_point_id);
4162 handle_ipfix_upcall(struct dpif_backer *backer,
4163 const struct dpif_upcall *upcall)
4165 struct ofproto_dpif *ofproto;
4168 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
4169 &flow, NULL, &ofproto, NULL, NULL)
4170 || !ofproto->ipfix) {
4174 /* The flow reflects exactly the contents of the packet. Sample
4175 * the packet using it. */
4176 dpif_ipfix_bridge_sample(ofproto->ipfix, upcall->packet, &flow);
4180 handle_upcalls(struct dpif_backer *backer, unsigned int max_batch)
4182 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
4183 struct ofpbuf miss_bufs[FLOW_MISS_MAX_BATCH];
4184 uint64_t miss_buf_stubs[FLOW_MISS_MAX_BATCH][4096 / 8];
4189 ovs_assert(max_batch <= FLOW_MISS_MAX_BATCH);
4192 for (n_processed = 0; n_processed < max_batch; n_processed++) {
4193 struct dpif_upcall *upcall = &misses[n_misses];
4194 struct ofpbuf *buf = &miss_bufs[n_misses];
4197 ofpbuf_use_stub(buf, miss_buf_stubs[n_misses],
4198 sizeof miss_buf_stubs[n_misses]);
4199 error = dpif_recv(backer->dpif, upcall, buf);
4205 switch (classify_upcall(upcall)) {
4207 /* Handle it later. */
4212 handle_sflow_upcall(backer, upcall);
4216 case FLOW_SAMPLE_UPCALL:
4217 handle_flow_sample_upcall(backer, upcall);
4222 handle_ipfix_upcall(backer, upcall);
4232 /* Handle deferred MISS_UPCALL processing. */
4233 handle_miss_upcalls(backer, misses, n_misses);
4234 for (i = 0; i < n_misses; i++) {
4235 ofpbuf_uninit(&miss_bufs[i]);
4241 /* Flow expiration. */
4243 static int subfacet_max_idle(const struct ofproto_dpif *);
4244 static void update_stats(struct dpif_backer *);
4245 static void rule_expire(struct rule_dpif *);
4246 static void expire_subfacets(struct ofproto_dpif *, int dp_max_idle);
4248 /* This function is called periodically by run(). Its job is to collect
4249 * updates for the flows that have been installed into the datapath, most
4250 * importantly when they last were used, and then use that information to
4251 * expire flows that have not been used recently.
4253 * Returns the number of milliseconds after which it should be called again. */
4255 expire(struct dpif_backer *backer)
4257 struct ofproto_dpif *ofproto;
4258 int max_idle = INT32_MAX;
4260 /* Periodically clear out the drop keys in an effort to keep them
4261 * relatively few. */
4262 drop_key_clear(backer);
4264 /* Update stats for each flow in the backer. */
4265 update_stats(backer);
4267 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4268 struct rule *rule, *next_rule;
4271 if (ofproto->backer != backer) {
4275 /* Keep track of the max number of flows per ofproto_dpif. */
4276 update_max_subfacet_count(ofproto);
4278 /* Expire subfacets that have been idle too long. */
4279 dp_max_idle = subfacet_max_idle(ofproto);
4280 expire_subfacets(ofproto, dp_max_idle);
4282 max_idle = MIN(max_idle, dp_max_idle);
4284 /* Expire OpenFlow flows whose idle_timeout or hard_timeout
4286 LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
4287 &ofproto->up.expirable) {
4288 rule_expire(rule_dpif_cast(rule));
4291 /* All outstanding data in existing flows has been accounted, so it's a
4292 * good time to do bond rebalancing. */
4293 if (ofproto->has_bonded_bundles) {
4294 struct ofbundle *bundle;
4296 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
4298 bond_rebalance(bundle->bond, &backer->revalidate_set);
4304 return MIN(max_idle, 1000);
4307 /* Updates flow table statistics given that the datapath just reported 'stats'
4308 * as 'subfacet''s statistics. */
4310 update_subfacet_stats(struct subfacet *subfacet,
4311 const struct dpif_flow_stats *stats)
4313 struct facet *facet = subfacet->facet;
4315 if (stats->n_packets >= subfacet->dp_packet_count) {
4316 uint64_t extra = stats->n_packets - subfacet->dp_packet_count;
4317 facet->packet_count += extra;
4319 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
4322 if (stats->n_bytes >= subfacet->dp_byte_count) {
4323 facet->byte_count += stats->n_bytes - subfacet->dp_byte_count;
4325 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
4328 subfacet->dp_packet_count = stats->n_packets;
4329 subfacet->dp_byte_count = stats->n_bytes;
4331 facet->tcp_flags |= stats->tcp_flags;
4333 subfacet_update_time(subfacet, stats->used);
4334 if (facet->accounted_bytes < facet->byte_count) {
4336 facet_account(facet);
4337 facet->accounted_bytes = facet->byte_count;
4341 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
4342 * about, or a flow that shouldn't be installed but was anyway. Delete it. */
4344 delete_unexpected_flow(struct ofproto_dpif *ofproto,
4345 const struct nlattr *key, size_t key_len)
4347 if (!VLOG_DROP_WARN(&rl)) {
4351 odp_flow_key_format(key, key_len, &s);
4352 VLOG_WARN("unexpected flow on %s: %s", ofproto->up.name, ds_cstr(&s));
4356 COVERAGE_INC(facet_unexpected);
4357 dpif_flow_del(ofproto->backer->dpif, key, key_len, NULL);
4360 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
4362 * This function also pushes statistics updates to rules which each facet
4363 * resubmits into. Generally these statistics will be accurate. However, if a
4364 * facet changes the rule it resubmits into at some time in between
4365 * update_stats() runs, it is possible that statistics accrued to the
4366 * old rule will be incorrectly attributed to the new rule. This could be
4367 * avoided by calling update_stats() whenever rules are created or
4368 * deleted. However, the performance impact of making so many calls to the
4369 * datapath do not justify the benefit of having perfectly accurate statistics.
4371 * In addition, this function maintains per ofproto flow hit counts. The patch
4372 * port is not treated specially. e.g. A packet ingress from br0 patched into
4373 * br1 will increase the hit count of br0 by 1, however, does not affect
4374 * the hit or miss counts of br1.
4377 update_stats(struct dpif_backer *backer)
4379 const struct dpif_flow_stats *stats;
4380 struct dpif_flow_dump dump;
4381 const struct nlattr *key;
4382 struct ofproto_dpif *ofproto;
4385 dpif_flow_dump_start(&dump, backer->dpif);
4386 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
4388 struct subfacet *subfacet;
4389 struct ofport_dpif *ofport;
4392 if (ofproto_receive(backer, NULL, key, key_len, &flow, NULL, &ofproto,
4397 ofproto->total_subfacet_count += hmap_count(&ofproto->subfacets);
4398 ofproto->n_update_stats++;
4400 ofport = get_ofp_port(ofproto, flow.in_port);
4401 if (ofport && ofport->tnl_port) {
4402 netdev_vport_inc_rx(ofport->up.netdev, stats);
4405 key_hash = odp_flow_key_hash(key, key_len);
4406 subfacet = subfacet_find(ofproto, key, key_len, key_hash);
4407 switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
4409 /* Update ofproto_dpif's hit count. */
4410 if (stats->n_packets > subfacet->dp_packet_count) {
4411 uint64_t delta = stats->n_packets - subfacet->dp_packet_count;
4412 dpif_stats_update_hit_count(ofproto, delta);
4415 update_subfacet_stats(subfacet, stats);
4419 /* Stats are updated per-packet. */
4422 case SF_NOT_INSTALLED:
4424 delete_unexpected_flow(ofproto, key, key_len);
4429 dpif_flow_dump_done(&dump);
4431 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4432 update_moving_averages(ofproto);
4437 /* Calculates and returns the number of milliseconds of idle time after which
4438 * subfacets should expire from the datapath. When a subfacet expires, we fold
4439 * its statistics into its facet, and when a facet's last subfacet expires, we
4440 * fold its statistic into its rule. */
4442 subfacet_max_idle(const struct ofproto_dpif *ofproto)
4445 * Idle time histogram.
4447 * Most of the time a switch has a relatively small number of subfacets.
4448 * When this is the case we might as well keep statistics for all of them
4449 * in userspace and to cache them in the kernel datapath for performance as
4452 * As the number of subfacets increases, the memory required to maintain
4453 * statistics about them in userspace and in the kernel becomes
4454 * significant. However, with a large number of subfacets it is likely
4455 * that only a few of them are "heavy hitters" that consume a large amount
4456 * of bandwidth. At this point, only heavy hitters are worth caching in
4457 * the kernel and maintaining in userspaces; other subfacets we can
4460 * The technique used to compute the idle time is to build a histogram with
4461 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
4462 * that is installed in the kernel gets dropped in the appropriate bucket.
4463 * After the histogram has been built, we compute the cutoff so that only
4464 * the most-recently-used 1% of subfacets (but at least
4465 * ofproto->up.flow_eviction_threshold flows) are kept cached. At least
4466 * the most-recently-used bucket of subfacets is kept, so actually an
4467 * arbitrary number of subfacets can be kept in any given expiration run
4468 * (though the next run will delete most of those unless they receive
4471 * This requires a second pass through the subfacets, in addition to the
4472 * pass made by update_stats(), because the former function never looks at
4473 * uninstallable subfacets.
4475 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
4476 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
4477 int buckets[N_BUCKETS] = { 0 };
4478 int total, subtotal, bucket;
4479 struct subfacet *subfacet;
4483 total = hmap_count(&ofproto->subfacets);
4484 if (total <= ofproto->up.flow_eviction_threshold) {
4485 return N_BUCKETS * BUCKET_WIDTH;
4488 /* Build histogram. */
4490 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
4491 long long int idle = now - subfacet->used;
4492 int bucket = (idle <= 0 ? 0
4493 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
4494 : (unsigned int) idle / BUCKET_WIDTH);
4498 /* Find the first bucket whose flows should be expired. */
4499 subtotal = bucket = 0;
4501 subtotal += buckets[bucket++];
4502 } while (bucket < N_BUCKETS &&
4503 subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100));
4505 if (VLOG_IS_DBG_ENABLED()) {
4509 ds_put_cstr(&s, "keep");
4510 for (i = 0; i < N_BUCKETS; i++) {
4512 ds_put_cstr(&s, ", drop");
4515 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
4518 VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s));
4522 return bucket * BUCKET_WIDTH;
4526 expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle)
4528 /* Cutoff time for most flows. */
4529 long long int normal_cutoff = time_msec() - dp_max_idle;
4531 /* We really want to keep flows for special protocols around, so use a more
4532 * conservative cutoff. */
4533 long long int special_cutoff = time_msec() - 10000;
4535 struct subfacet *subfacet, *next_subfacet;
4536 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
4540 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
4541 &ofproto->subfacets) {
4542 long long int cutoff;
4544 cutoff = (subfacet->facet->xout.slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP
4548 if (subfacet->used < cutoff) {
4549 if (subfacet->path != SF_NOT_INSTALLED) {
4550 batch[n_batch++] = subfacet;
4551 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
4552 subfacet_destroy_batch(ofproto, batch, n_batch);
4556 subfacet_destroy(subfacet);
4562 subfacet_destroy_batch(ofproto, batch, n_batch);
4566 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
4567 * then delete it entirely. */
4569 rule_expire(struct rule_dpif *rule)
4571 struct facet *facet, *next_facet;
4575 if (rule->up.pending) {
4576 /* We'll have to expire it later. */
4580 /* Has 'rule' expired? */
4582 if (rule->up.hard_timeout
4583 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
4584 reason = OFPRR_HARD_TIMEOUT;
4585 } else if (rule->up.idle_timeout
4586 && now > rule->up.used + rule->up.idle_timeout * 1000) {
4587 reason = OFPRR_IDLE_TIMEOUT;
4592 COVERAGE_INC(ofproto_dpif_expired);
4594 /* Update stats. (This is a no-op if the rule expired due to an idle
4595 * timeout, because that only happens when the rule has no facets left.) */
4596 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4597 facet_remove(facet);
4600 /* Get rid of the rule. */
4601 ofproto_rule_expire(&rule->up, reason);
4606 /* Creates and returns a new facet based on 'miss'.
4608 * The caller must already have determined that no facet with an identical
4609 * 'miss->flow' exists in 'miss->ofproto'.
4611 * 'hash' must be the return value of flow_hash(miss->flow, 0).
4613 * The facet will initially have no subfacets. The caller should create (at
4614 * least) one subfacet with subfacet_create(). */
4615 static struct facet *
4616 facet_create(const struct flow_miss *miss, uint32_t hash)
4618 struct ofproto_dpif *ofproto = miss->ofproto;
4619 struct xlate_in xin;
4620 struct facet *facet;
4622 facet = xzalloc(sizeof *facet);
4623 facet->used = time_msec();
4624 facet->flow = miss->flow;
4625 facet->initial_vals = miss->initial_vals;
4626 facet->rule = rule_dpif_lookup(ofproto, &facet->flow);
4627 facet->learn_rl = time_msec() + 500;
4629 hmap_insert(&ofproto->facets, &facet->hmap_node, hash);
4630 list_push_back(&facet->rule->facets, &facet->list_node);
4631 list_init(&facet->subfacets);
4632 netflow_flow_init(&facet->nf_flow);
4633 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
4635 xlate_in_init(&xin, ofproto, &facet->flow, &facet->initial_vals,
4636 facet->rule, 0, NULL);
4637 xin.may_learn = true;
4638 xlate_actions(&xin, &facet->xout);
4639 facet->nf_flow.output_iface = facet->xout.nf_output_iface;
4645 facet_free(struct facet *facet)
4648 xlate_out_uninit(&facet->xout);
4653 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
4654 * 'packet', which arrived on 'in_port'. */
4656 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
4657 const struct nlattr *odp_actions, size_t actions_len,
4658 struct ofpbuf *packet)
4660 struct odputil_keybuf keybuf;
4664 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4665 odp_flow_key_from_flow(&key, flow,
4666 ofp_port_to_odp_port(ofproto, flow->in_port));
4668 error = dpif_execute(ofproto->backer->dpif, key.data, key.size,
4669 odp_actions, actions_len, packet);
4673 /* Remove 'facet' from 'ofproto' and free up the associated memory:
4675 * - If 'facet' was installed in the datapath, uninstalls it and updates its
4676 * rule's statistics, via subfacet_uninstall().
4678 * - Removes 'facet' from its rule and from ofproto->facets.
4681 facet_remove(struct facet *facet)
4683 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4684 struct subfacet *subfacet, *next_subfacet;
4686 ovs_assert(!list_is_empty(&facet->subfacets));
4688 /* First uninstall all of the subfacets to get final statistics. */
4689 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4690 subfacet_uninstall(subfacet);
4693 /* Flush the final stats to the rule.
4695 * This might require us to have at least one subfacet around so that we
4696 * can use its actions for accounting in facet_account(), which is why we
4697 * have uninstalled but not yet destroyed the subfacets. */
4698 facet_flush_stats(facet);
4700 /* Now we're really all done so destroy everything. */
4701 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
4702 &facet->subfacets) {
4703 subfacet_destroy__(subfacet);
4705 hmap_remove(&ofproto->facets, &facet->hmap_node);
4706 list_remove(&facet->list_node);
4710 /* Feed information from 'facet' back into the learning table to keep it in
4711 * sync with what is actually flowing through the datapath. */
4713 facet_learn(struct facet *facet)
4715 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4716 long long int now = time_msec();
4717 struct xlate_in xin;
4719 if (!facet->xout.has_fin_timeout && now < facet->learn_rl) {
4723 facet->learn_rl = now + 500;
4725 if (!facet->xout.has_learn
4726 && !facet->xout.has_normal
4727 && (!facet->xout.has_fin_timeout
4728 || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
4732 xlate_in_init(&xin, ofproto, &facet->flow, &facet->initial_vals,
4733 facet->rule, facet->tcp_flags, NULL);
4734 xin.may_learn = true;
4735 xlate_actions_for_side_effects(&xin);
4739 facet_account(struct facet *facet)
4741 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4742 const struct nlattr *a;
4747 if (!facet->xout.has_normal || !ofproto->has_bonded_bundles) {
4750 n_bytes = facet->byte_count - facet->accounted_bytes;
4752 /* This loop feeds byte counters to bond_account() for rebalancing to use
4753 * as a basis. We also need to track the actual VLAN on which the packet
4754 * is going to be sent to ensure that it matches the one passed to
4755 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
4758 * We use the actions from an arbitrary subfacet because they should all
4759 * be equally valid for our purpose. */
4760 vlan_tci = facet->flow.vlan_tci;
4761 NL_ATTR_FOR_EACH_UNSAFE (a, left, facet->xout.odp_actions.data,
4762 facet->xout.odp_actions.size) {
4763 const struct ovs_action_push_vlan *vlan;
4764 struct ofport_dpif *port;
4766 switch (nl_attr_type(a)) {
4767 case OVS_ACTION_ATTR_OUTPUT:
4768 port = get_odp_port(ofproto, nl_attr_get_u32(a));
4769 if (port && port->bundle && port->bundle->bond) {
4770 bond_account(port->bundle->bond, &facet->flow,
4771 vlan_tci_to_vid(vlan_tci), n_bytes);
4775 case OVS_ACTION_ATTR_POP_VLAN:
4776 vlan_tci = htons(0);
4779 case OVS_ACTION_ATTR_PUSH_VLAN:
4780 vlan = nl_attr_get(a);
4781 vlan_tci = vlan->vlan_tci;
4787 /* Returns true if the only action for 'facet' is to send to the controller.
4788 * (We don't report NetFlow expiration messages for such facets because they
4789 * are just part of the control logic for the network, not real traffic). */
4791 facet_is_controller_flow(struct facet *facet)
4794 const struct rule *rule = &facet->rule->up;
4795 const struct ofpact *ofpacts = rule->ofpacts;
4796 size_t ofpacts_len = rule->ofpacts_len;
4798 if (ofpacts_len > 0 &&
4799 ofpacts->type == OFPACT_CONTROLLER &&
4800 ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len)) {
4807 /* Folds all of 'facet''s statistics into its rule. Also updates the
4808 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
4809 * 'facet''s statistics in the datapath should have been zeroed and folded into
4810 * its packet and byte counts before this function is called. */
4812 facet_flush_stats(struct facet *facet)
4814 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4815 struct subfacet *subfacet;
4817 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4818 ovs_assert(!subfacet->dp_byte_count);
4819 ovs_assert(!subfacet->dp_packet_count);
4822 facet_push_stats(facet);
4823 if (facet->accounted_bytes < facet->byte_count) {
4824 facet_account(facet);
4825 facet->accounted_bytes = facet->byte_count;
4828 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
4829 struct ofexpired expired;
4830 expired.flow = facet->flow;
4831 expired.packet_count = facet->packet_count;
4832 expired.byte_count = facet->byte_count;
4833 expired.used = facet->used;
4834 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4837 /* Reset counters to prevent double counting if 'facet' ever gets
4839 facet_reset_counters(facet);
4841 netflow_flow_clear(&facet->nf_flow);
4842 facet->tcp_flags = 0;
4845 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
4846 * Returns it if found, otherwise a null pointer.
4848 * 'hash' must be the return value of flow_hash(flow, 0).
4850 * The returned facet might need revalidation; use facet_lookup_valid()
4851 * instead if that is important. */
4852 static struct facet *
4853 facet_find(struct ofproto_dpif *ofproto,
4854 const struct flow *flow, uint32_t hash)
4856 struct facet *facet;
4858 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, hash, &ofproto->facets) {
4859 if (flow_equal(flow, &facet->flow)) {
4867 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
4868 * Returns it if found, otherwise a null pointer.
4870 * 'hash' must be the return value of flow_hash(flow, 0).
4872 * The returned facet is guaranteed to be valid. */
4873 static struct facet *
4874 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow,
4877 struct facet *facet;
4879 facet = facet_find(ofproto, flow, hash);
4881 && (ofproto->backer->need_revalidate
4882 || tag_set_intersects(&ofproto->backer->revalidate_set,
4884 && !facet_revalidate(facet)) {
4892 facet_check_consistency(struct facet *facet)
4894 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
4896 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4898 struct xlate_out xout;
4899 struct xlate_in xin;
4901 struct rule_dpif *rule;
4904 /* Check the rule for consistency. */
4905 rule = rule_dpif_lookup(ofproto, &facet->flow);
4906 if (rule != facet->rule) {
4907 if (!VLOG_DROP_WARN(&rl)) {
4908 struct ds s = DS_EMPTY_INITIALIZER;
4910 flow_format(&s, &facet->flow);
4911 ds_put_format(&s, ": facet associated with wrong rule (was "
4912 "table=%"PRIu8",", facet->rule->up.table_id);
4913 cls_rule_format(&facet->rule->up.cr, &s);
4914 ds_put_format(&s, ") (should have been table=%"PRIu8",",
4916 cls_rule_format(&rule->up.cr, &s);
4917 ds_put_cstr(&s, ")\n");
4924 /* Check the datapath actions for consistency. */
4925 xlate_in_init(&xin, ofproto, &facet->flow, &facet->initial_vals, rule,
4927 xlate_actions(&xin, &xout);
4929 ok = ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)
4930 && facet->xout.slow == xout.slow;
4931 if (!ok && !VLOG_DROP_WARN(&rl)) {
4932 struct ds s = DS_EMPTY_INITIALIZER;
4934 flow_format(&s, &facet->flow);
4935 ds_put_cstr(&s, ": inconsistency in facet");
4937 if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4938 ds_put_cstr(&s, " (actions were: ");
4939 format_odp_actions(&s, facet->xout.odp_actions.data,
4940 facet->xout.odp_actions.size);
4941 ds_put_cstr(&s, ") (correct actions: ");
4942 format_odp_actions(&s, xout.odp_actions.data,
4943 xout.odp_actions.size);
4944 ds_put_cstr(&s, ")");
4947 if (facet->xout.slow != xout.slow) {
4948 ds_put_format(&s, " slow path incorrect. should be %d", xout.slow);
4953 xlate_out_uninit(&xout);
4958 /* Re-searches the classifier for 'facet':
4960 * - If the rule found is different from 'facet''s current rule, moves
4961 * 'facet' to the new rule and recompiles its actions.
4963 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
4964 * where it is and recompiles its actions anyway.
4966 * - If any of 'facet''s subfacets correspond to a new flow according to
4967 * ofproto_receive(), 'facet' is removed.
4969 * Returns true if 'facet' is still valid. False if 'facet' was removed. */
4971 facet_revalidate(struct facet *facet)
4973 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4974 struct rule_dpif *new_rule;
4975 struct subfacet *subfacet;
4976 struct xlate_out xout;
4977 struct xlate_in xin;
4979 COVERAGE_INC(facet_revalidate);
4981 /* Check that child subfacets still correspond to this facet. Tunnel
4982 * configuration changes could cause a subfacet's OpenFlow in_port to
4984 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4985 struct ofproto_dpif *recv_ofproto;
4986 struct flow recv_flow;
4989 error = ofproto_receive(ofproto->backer, NULL, subfacet->key,
4990 subfacet->key_len, &recv_flow, NULL,
4991 &recv_ofproto, NULL, NULL);
4993 || recv_ofproto != ofproto
4994 || memcmp(&recv_flow, &facet->flow, sizeof recv_flow)) {
4995 facet_remove(facet);
5000 new_rule = rule_dpif_lookup(ofproto, &facet->flow);
5002 /* Calculate new datapath actions.
5004 * We do not modify any 'facet' state yet, because we might need to, e.g.,
5005 * emit a NetFlow expiration and, if so, we need to have the old state
5006 * around to properly compose it. */
5007 xlate_in_init(&xin, ofproto, &facet->flow, &facet->initial_vals, new_rule,
5009 xlate_actions(&xin, &xout);
5011 /* A facet's slow path reason should only change under dramatic
5012 * circumstances. Rather than try to update everything, it's simpler to
5013 * remove the facet and start over. */
5014 if (facet->xout.slow != xout.slow) {
5015 facet_remove(facet);
5016 xlate_out_uninit(&xout);
5020 if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
5021 LIST_FOR_EACH(subfacet, list_node, &facet->subfacets) {
5022 if (subfacet->path == SF_FAST_PATH) {
5023 struct dpif_flow_stats stats;
5025 subfacet_install(subfacet, &xout.odp_actions, &stats);
5026 subfacet_update_stats(subfacet, &stats);
5030 facet_flush_stats(facet);
5032 ofpbuf_clear(&facet->xout.odp_actions);
5033 ofpbuf_put(&facet->xout.odp_actions, xout.odp_actions.data,
5034 xout.odp_actions.size);
5037 /* Update 'facet' now that we've taken care of all the old state. */
5038 facet->xout.tags = xout.tags;
5039 facet->xout.slow = xout.slow;
5040 facet->xout.has_learn = xout.has_learn;
5041 facet->xout.has_normal = xout.has_normal;
5042 facet->xout.has_fin_timeout = xout.has_fin_timeout;
5043 facet->xout.nf_output_iface = xout.nf_output_iface;
5044 facet->xout.mirrors = xout.mirrors;
5045 facet->nf_flow.output_iface = facet->xout.nf_output_iface;
5047 if (facet->rule != new_rule) {
5048 COVERAGE_INC(facet_changed_rule);
5049 list_remove(&facet->list_node);
5050 list_push_back(&new_rule->facets, &facet->list_node);
5051 facet->rule = new_rule;
5052 facet->used = new_rule->up.created;
5053 facet->prev_used = facet->used;
5056 xlate_out_uninit(&xout);
5060 /* Updates 'facet''s used time. Caller is responsible for calling
5061 * facet_push_stats() to update the flows which 'facet' resubmits into. */
5063 facet_update_time(struct facet *facet, long long int used)
5065 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5066 if (used > facet->used) {
5068 ofproto_rule_update_used(&facet->rule->up, used);
5069 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
5074 facet_reset_counters(struct facet *facet)
5076 facet->packet_count = 0;
5077 facet->byte_count = 0;
5078 facet->prev_packet_count = 0;
5079 facet->prev_byte_count = 0;
5080 facet->accounted_bytes = 0;
5084 facet_push_stats(struct facet *facet)
5086 struct dpif_flow_stats stats;
5088 ovs_assert(facet->packet_count >= facet->prev_packet_count);
5089 ovs_assert(facet->byte_count >= facet->prev_byte_count);
5090 ovs_assert(facet->used >= facet->prev_used);
5092 stats.n_packets = facet->packet_count - facet->prev_packet_count;
5093 stats.n_bytes = facet->byte_count - facet->prev_byte_count;
5094 stats.used = facet->used;
5095 stats.tcp_flags = 0;
5097 if (stats.n_packets || stats.n_bytes || facet->used > facet->prev_used) {
5098 facet->prev_packet_count = facet->packet_count;
5099 facet->prev_byte_count = facet->byte_count;
5100 facet->prev_used = facet->used;
5102 rule_credit_stats(facet->rule, &stats);
5103 flow_push_stats(facet, &stats);
5105 update_mirror_stats(ofproto_dpif_cast(facet->rule->up.ofproto),
5106 facet->xout.mirrors, stats.n_packets,
5112 push_all_stats__(bool run_fast)
5114 static long long int rl = LLONG_MIN;
5115 struct ofproto_dpif *ofproto;
5117 if (time_msec() < rl) {
5121 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5122 struct facet *facet;
5124 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
5125 facet_push_stats(facet);
5132 rl = time_msec() + 100;
5136 push_all_stats(void)
5138 push_all_stats__(true);
5142 rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats)
5144 rule->packet_count += stats->n_packets;
5145 rule->byte_count += stats->n_bytes;
5146 ofproto_rule_update_used(&rule->up, stats->used);
5149 /* Pushes flow statistics to the rules which 'facet->flow' resubmits
5150 * into given 'facet->rule''s actions and mirrors. */
5152 flow_push_stats(struct facet *facet, const struct dpif_flow_stats *stats)
5154 struct rule_dpif *rule = facet->rule;
5155 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5156 struct xlate_in xin;
5158 ofproto_rule_update_used(&rule->up, stats->used);
5160 xlate_in_init(&xin, ofproto, &facet->flow, &facet->initial_vals, rule,
5162 xin.resubmit_stats = stats;
5163 xlate_actions_for_side_effects(&xin);
5168 static struct subfacet *
5169 subfacet_find(struct ofproto_dpif *ofproto,
5170 const struct nlattr *key, size_t key_len, uint32_t key_hash)
5172 struct subfacet *subfacet;
5174 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
5175 &ofproto->subfacets) {
5176 if (subfacet->key_len == key_len
5177 && !memcmp(key, subfacet->key, key_len)) {
5185 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
5186 * 'key_fitness', 'key', and 'key_len' members in 'miss'. Returns the
5187 * existing subfacet if there is one, otherwise creates and returns a
5189 static struct subfacet *
5190 subfacet_create(struct facet *facet, struct flow_miss *miss,
5193 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5194 enum odp_key_fitness key_fitness = miss->key_fitness;
5195 const struct nlattr *key = miss->key;
5196 size_t key_len = miss->key_len;
5198 struct subfacet *subfacet;
5200 key_hash = odp_flow_key_hash(key, key_len);
5202 if (list_is_empty(&facet->subfacets)) {
5203 subfacet = &facet->one_subfacet;
5205 subfacet = subfacet_find(ofproto, key, key_len, key_hash);
5207 if (subfacet->facet == facet) {
5211 /* This shouldn't happen. */
5212 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
5213 subfacet_destroy(subfacet);
5216 subfacet = xmalloc(sizeof *subfacet);
5219 hmap_insert(&ofproto->subfacets, &subfacet->hmap_node, key_hash);
5220 list_push_back(&facet->subfacets, &subfacet->list_node);
5221 subfacet->facet = facet;
5222 subfacet->key_fitness = key_fitness;
5223 subfacet->key = xmemdup(key, key_len);
5224 subfacet->key_len = key_len;
5225 subfacet->used = now;
5226 subfacet->created = now;
5227 subfacet->dp_packet_count = 0;
5228 subfacet->dp_byte_count = 0;
5229 subfacet->path = SF_NOT_INSTALLED;
5230 subfacet->odp_in_port = miss->odp_in_port;
5232 ofproto->subfacet_add_count++;
5236 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
5237 * its facet within 'ofproto', and frees it. */
5239 subfacet_destroy__(struct subfacet *subfacet)
5241 struct facet *facet = subfacet->facet;
5242 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5244 /* Update ofproto stats before uninstall the subfacet. */
5245 ofproto->subfacet_del_count++;
5246 ofproto->total_subfacet_life_span += (time_msec() - subfacet->created);
5248 subfacet_uninstall(subfacet);
5249 hmap_remove(&ofproto->subfacets, &subfacet->hmap_node);
5250 list_remove(&subfacet->list_node);
5251 free(subfacet->key);
5252 if (subfacet != &facet->one_subfacet) {
5257 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
5258 * last remaining subfacet in its facet destroys the facet too. */
5260 subfacet_destroy(struct subfacet *subfacet)
5262 struct facet *facet = subfacet->facet;
5264 if (list_is_singleton(&facet->subfacets)) {
5265 /* facet_remove() needs at least one subfacet (it will remove it). */
5266 facet_remove(facet);
5268 subfacet_destroy__(subfacet);
5273 subfacet_destroy_batch(struct ofproto_dpif *ofproto,
5274 struct subfacet **subfacets, int n)
5276 struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH];
5277 struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH];
5278 struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH];
5281 for (i = 0; i < n; i++) {
5282 ops[i].type = DPIF_OP_FLOW_DEL;
5283 ops[i].u.flow_del.key = subfacets[i]->key;
5284 ops[i].u.flow_del.key_len = subfacets[i]->key_len;
5285 ops[i].u.flow_del.stats = &stats[i];
5289 dpif_operate(ofproto->backer->dpif, opsp, n);
5290 for (i = 0; i < n; i++) {
5291 subfacet_reset_dp_stats(subfacets[i], &stats[i]);
5292 subfacets[i]->path = SF_NOT_INSTALLED;
5293 subfacet_destroy(subfacets[i]);
5298 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
5299 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
5300 * in the datapath will be zeroed and 'stats' will be updated with traffic new
5301 * since 'subfacet' was last updated.
5303 * Returns 0 if successful, otherwise a positive errno value. */
5305 subfacet_install(struct subfacet *subfacet, const struct ofpbuf *odp_actions,
5306 struct dpif_flow_stats *stats)
5308 struct facet *facet = subfacet->facet;
5309 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5310 enum subfacet_path path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
5311 const struct nlattr *actions = odp_actions->data;
5312 size_t actions_len = odp_actions->size;
5314 uint64_t slow_path_stub[128 / 8];
5315 enum dpif_flow_put_flags flags;
5318 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
5320 flags |= DPIF_FP_ZERO_STATS;
5323 if (path == SF_SLOW_PATH) {
5324 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
5325 slow_path_stub, sizeof slow_path_stub,
5326 &actions, &actions_len);
5329 ret = dpif_flow_put(ofproto->backer->dpif, flags, subfacet->key,
5330 subfacet->key_len, actions, actions_len, stats);
5333 subfacet_reset_dp_stats(subfacet, stats);
5337 subfacet->path = path;
5342 /* If 'subfacet' is installed in the datapath, uninstalls it. */
5344 subfacet_uninstall(struct subfacet *subfacet)
5346 if (subfacet->path != SF_NOT_INSTALLED) {
5347 struct rule_dpif *rule = subfacet->facet->rule;
5348 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5349 struct dpif_flow_stats stats;
5352 error = dpif_flow_del(ofproto->backer->dpif, subfacet->key,
5353 subfacet->key_len, &stats);
5354 subfacet_reset_dp_stats(subfacet, &stats);
5356 subfacet_update_stats(subfacet, &stats);
5358 subfacet->path = SF_NOT_INSTALLED;
5360 ovs_assert(subfacet->dp_packet_count == 0);
5361 ovs_assert(subfacet->dp_byte_count == 0);
5365 /* Resets 'subfacet''s datapath statistics counters. This should be called
5366 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
5367 * non-null, it should contain the statistics returned by dpif when 'subfacet'
5368 * was reset in the datapath. 'stats' will be modified to include only
5369 * statistics new since 'subfacet' was last updated. */
5371 subfacet_reset_dp_stats(struct subfacet *subfacet,
5372 struct dpif_flow_stats *stats)
5375 && subfacet->dp_packet_count <= stats->n_packets
5376 && subfacet->dp_byte_count <= stats->n_bytes) {
5377 stats->n_packets -= subfacet->dp_packet_count;
5378 stats->n_bytes -= subfacet->dp_byte_count;
5381 subfacet->dp_packet_count = 0;
5382 subfacet->dp_byte_count = 0;
5385 /* Updates 'subfacet''s used time. The caller is responsible for calling
5386 * facet_push_stats() to update the flows which 'subfacet' resubmits into. */
5388 subfacet_update_time(struct subfacet *subfacet, long long int used)
5390 if (used > subfacet->used) {
5391 subfacet->used = used;
5392 facet_update_time(subfacet->facet, used);
5396 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
5398 * Because of the meaning of a subfacet's counters, it only makes sense to do
5399 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
5400 * represents a packet that was sent by hand or if it represents statistics
5401 * that have been cleared out of the datapath. */
5403 subfacet_update_stats(struct subfacet *subfacet,
5404 const struct dpif_flow_stats *stats)
5406 if (stats->n_packets || stats->used > subfacet->used) {
5407 struct facet *facet = subfacet->facet;
5409 subfacet_update_time(subfacet, stats->used);
5410 facet->packet_count += stats->n_packets;
5411 facet->byte_count += stats->n_bytes;
5412 facet->tcp_flags |= stats->tcp_flags;
5413 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
5419 static struct rule_dpif *
5420 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow)
5422 struct rule_dpif *rule;
5424 rule = rule_dpif_lookup__(ofproto, flow, 0);
5429 return rule_dpif_miss_rule(ofproto, flow);
5432 static struct rule_dpif *
5433 rule_dpif_lookup__(struct ofproto_dpif *ofproto, const struct flow *flow,
5436 struct cls_rule *cls_rule;
5437 struct classifier *cls;
5439 if (table_id >= N_TABLES) {
5443 cls = &ofproto->up.tables[table_id].cls;
5444 if (flow->nw_frag & FLOW_NW_FRAG_ANY
5445 && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
5446 /* For OFPC_NORMAL frag_handling, we must pretend that transport ports
5447 * are unavailable. */
5448 struct flow ofpc_normal_flow = *flow;
5449 ofpc_normal_flow.tp_src = htons(0);
5450 ofpc_normal_flow.tp_dst = htons(0);
5451 cls_rule = classifier_lookup(cls, &ofpc_normal_flow);
5453 cls_rule = classifier_lookup(cls, flow);
5455 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
5458 static struct rule_dpif *
5459 rule_dpif_miss_rule(struct ofproto_dpif *ofproto, const struct flow *flow)
5461 struct ofport_dpif *port;
5463 port = get_ofp_port(ofproto, flow->in_port);
5465 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, flow->in_port);
5466 return ofproto->miss_rule;
5469 if (port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN) {
5470 return ofproto->no_packet_in_rule;
5472 return ofproto->miss_rule;
5476 complete_operation(struct rule_dpif *rule)
5478 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5480 rule_invalidate(rule);
5482 struct dpif_completion *c = xmalloc(sizeof *c);
5483 c->op = rule->up.pending;
5484 list_push_back(&ofproto->completions, &c->list_node);
5486 ofoperation_complete(rule->up.pending, 0);
5490 static struct rule *
5493 struct rule_dpif *rule = xmalloc(sizeof *rule);
5498 rule_dealloc(struct rule *rule_)
5500 struct rule_dpif *rule = rule_dpif_cast(rule_);
5505 rule_construct(struct rule *rule_)
5507 struct rule_dpif *rule = rule_dpif_cast(rule_);
5508 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5509 struct rule_dpif *victim;
5512 rule->packet_count = 0;
5513 rule->byte_count = 0;
5515 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
5516 if (victim && !list_is_empty(&victim->facets)) {
5517 struct facet *facet;
5519 rule->facets = victim->facets;
5520 list_moved(&rule->facets);
5521 LIST_FOR_EACH (facet, list_node, &rule->facets) {
5522 /* XXX: We're only clearing our local counters here. It's possible
5523 * that quite a few packets are unaccounted for in the datapath
5524 * statistics. These will be accounted to the new rule instead of
5525 * cleared as required. This could be fixed by clearing out the
5526 * datapath statistics for this facet, but currently it doesn't
5528 facet_reset_counters(facet);
5532 /* Must avoid list_moved() in this case. */
5533 list_init(&rule->facets);
5536 table_id = rule->up.table_id;
5538 rule->tag = victim->tag;
5539 } else if (table_id == 0) {
5544 miniflow_expand(&rule->up.cr.match.flow, &flow);
5545 rule->tag = rule_calculate_tag(&flow, &rule->up.cr.match.mask,
5546 ofproto->tables[table_id].basis);
5549 complete_operation(rule);
5554 rule_destruct(struct rule *rule_)
5556 struct rule_dpif *rule = rule_dpif_cast(rule_);
5557 struct facet *facet, *next_facet;
5559 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
5560 facet_revalidate(facet);
5563 complete_operation(rule);
5567 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
5569 struct rule_dpif *rule = rule_dpif_cast(rule_);
5571 /* push_all_stats() can handle flow misses which, when using the learn
5572 * action, can cause rules to be added and deleted. This can corrupt our
5573 * caller's datastructures which assume that rule_get_stats() doesn't have
5574 * an impact on the flow table. To be safe, we disable miss handling. */
5575 push_all_stats__(false);
5577 /* Start from historical data for 'rule' itself that are no longer tracked
5578 * in facets. This counts, for example, facets that have expired. */
5579 *packets = rule->packet_count;
5580 *bytes = rule->byte_count;
5584 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
5585 struct ofpbuf *packet)
5587 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5588 struct initial_vals initial_vals;
5589 struct dpif_flow_stats stats;
5590 struct xlate_out xout;
5591 struct xlate_in xin;
5593 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
5594 rule_credit_stats(rule, &stats);
5596 initial_vals.vlan_tci = flow->vlan_tci;
5597 xlate_in_init(&xin, ofproto, flow, &initial_vals, rule, stats.tcp_flags,
5599 xin.resubmit_stats = &stats;
5600 xlate_actions(&xin, &xout);
5602 execute_odp_actions(ofproto, flow, xout.odp_actions.data,
5603 xout.odp_actions.size, packet);
5605 xlate_out_uninit(&xout);
5609 rule_execute(struct rule *rule, const struct flow *flow,
5610 struct ofpbuf *packet)
5612 rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
5613 ofpbuf_delete(packet);
5618 rule_modify_actions(struct rule *rule_)
5620 struct rule_dpif *rule = rule_dpif_cast(rule_);
5622 complete_operation(rule);
5625 /* Sends 'packet' out 'ofport'.
5626 * May modify 'packet'.
5627 * Returns 0 if successful, otherwise a positive errno value. */
5629 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
5631 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
5632 uint64_t odp_actions_stub[1024 / 8];
5633 struct ofpbuf key, odp_actions;
5634 struct dpif_flow_stats stats;
5635 struct odputil_keybuf keybuf;
5636 struct ofpact_output output;
5637 struct xlate_out xout;
5638 struct xlate_in xin;
5642 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5643 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5645 /* Use OFPP_NONE as the in_port to avoid special packet processing. */
5646 flow_extract(packet, 0, 0, NULL, OFPP_NONE, &flow);
5647 odp_flow_key_from_flow(&key, &flow, ofp_port_to_odp_port(ofproto,
5649 dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
5651 ofpact_init(&output.ofpact, OFPACT_OUTPUT, sizeof output);
5652 output.port = ofport->up.ofp_port;
5655 xlate_in_init(&xin, ofproto, &flow, NULL, NULL, 0, packet);
5656 xin.ofpacts_len = sizeof output;
5657 xin.ofpacts = &output.ofpact;
5658 xin.resubmit_stats = &stats;
5659 xlate_actions(&xin, &xout);
5661 error = dpif_execute(ofproto->backer->dpif,
5663 xout.odp_actions.data, xout.odp_actions.size,
5665 xlate_out_uninit(&xout);
5668 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %s (%s)",
5669 ofproto->up.name, netdev_get_name(ofport->up.netdev),
5673 ofproto->stats.tx_packets++;
5674 ofproto->stats.tx_bytes += packet->size;
5678 /* OpenFlow to datapath action translation. */
5680 static bool may_receive(const struct ofport_dpif *, struct xlate_ctx *);
5681 static void do_xlate_actions(const struct ofpact *, size_t ofpacts_len,
5682 struct xlate_ctx *);
5683 static void xlate_normal(struct xlate_ctx *);
5685 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
5686 * The action will state 'slow' as the reason that the action is in the slow
5687 * path. (This is purely informational: it allows a human viewing "ovs-dpctl
5688 * dump-flows" output to see why a flow is in the slow path.)
5690 * The 'stub_size' bytes in 'stub' will be used to store the action.
5691 * 'stub_size' must be large enough for the action.
5693 * The action and its size will be stored in '*actionsp' and '*actions_lenp',
5696 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
5697 enum slow_path_reason slow,
5698 uint64_t *stub, size_t stub_size,
5699 const struct nlattr **actionsp, size_t *actions_lenp)
5701 union user_action_cookie cookie;
5704 cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
5705 cookie.slow_path.unused = 0;
5706 cookie.slow_path.reason = slow;
5708 ofpbuf_use_stack(&buf, stub, stub_size);
5709 if (slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP | SLOW_STP)) {
5710 uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif, UINT32_MAX);
5711 odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
5713 put_userspace_action(ofproto, &buf, flow, &cookie,
5714 sizeof cookie.slow_path);
5716 *actionsp = buf.data;
5717 *actions_lenp = buf.size;
5721 put_userspace_action(const struct ofproto_dpif *ofproto,
5722 struct ofpbuf *odp_actions,
5723 const struct flow *flow,
5724 const union user_action_cookie *cookie,
5725 const size_t cookie_size)
5729 pid = dpif_port_get_pid(ofproto->backer->dpif,
5730 ofp_port_to_odp_port(ofproto, flow->in_port));
5732 return odp_put_userspace_action(pid, cookie, cookie_size, odp_actions);
5735 /* Compose SAMPLE action for sFlow or IPFIX. The given probability is
5736 * the number of packets out of UINT32_MAX to sample. The given
5737 * cookie is passed back in the callback for each sampled packet.
5740 compose_sample_action(const struct ofproto_dpif *ofproto,
5741 struct ofpbuf *odp_actions,
5742 const struct flow *flow,
5743 const uint32_t probability,
5744 const union user_action_cookie *cookie,
5745 const size_t cookie_size)
5747 size_t sample_offset, actions_offset;
5750 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
5752 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
5754 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
5755 cookie_offset = put_userspace_action(ofproto, odp_actions, flow, cookie,
5758 nl_msg_end_nested(odp_actions, actions_offset);
5759 nl_msg_end_nested(odp_actions, sample_offset);
5760 return cookie_offset;
5764 compose_sflow_cookie(const struct ofproto_dpif *ofproto,
5765 ovs_be16 vlan_tci, uint32_t odp_port,
5766 unsigned int n_outputs, union user_action_cookie *cookie)
5770 cookie->type = USER_ACTION_COOKIE_SFLOW;
5771 cookie->sflow.vlan_tci = vlan_tci;
5773 /* See http://www.sflow.org/sflow_version_5.txt (search for "Input/output
5774 * port information") for the interpretation of cookie->output. */
5775 switch (n_outputs) {
5777 /* 0x40000000 | 256 means "packet dropped for unknown reason". */
5778 cookie->sflow.output = 0x40000000 | 256;
5782 ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port);
5784 cookie->sflow.output = ifindex;
5789 /* 0x80000000 means "multiple output ports. */
5790 cookie->sflow.output = 0x80000000 | n_outputs;
5795 /* Compose SAMPLE action for sFlow bridge sampling. */
5797 compose_sflow_action(const struct ofproto_dpif *ofproto,
5798 struct ofpbuf *odp_actions,
5799 const struct flow *flow,
5802 uint32_t probability;
5803 union user_action_cookie cookie;
5805 if (!ofproto->sflow || flow->in_port == OFPP_NONE) {
5809 probability = dpif_sflow_get_probability(ofproto->sflow);
5810 compose_sflow_cookie(ofproto, htons(0), odp_port,
5811 odp_port == OVSP_NONE ? 0 : 1, &cookie);
5813 return compose_sample_action(ofproto, odp_actions, flow, probability,
5814 &cookie, sizeof cookie.sflow);
5818 compose_flow_sample_cookie(uint16_t probability, uint32_t collector_set_id,
5819 uint32_t obs_domain_id, uint32_t obs_point_id,
5820 union user_action_cookie *cookie)
5822 cookie->type = USER_ACTION_COOKIE_FLOW_SAMPLE;
5823 cookie->flow_sample.probability = probability;
5824 cookie->flow_sample.collector_set_id = collector_set_id;
5825 cookie->flow_sample.obs_domain_id = obs_domain_id;
5826 cookie->flow_sample.obs_point_id = obs_point_id;
5830 compose_ipfix_cookie(union user_action_cookie *cookie)
5832 cookie->type = USER_ACTION_COOKIE_IPFIX;
5835 /* Compose SAMPLE action for IPFIX bridge sampling. */
5837 compose_ipfix_action(const struct ofproto_dpif *ofproto,
5838 struct ofpbuf *odp_actions,
5839 const struct flow *flow)
5841 uint32_t probability;
5842 union user_action_cookie cookie;
5844 if (!ofproto->ipfix || flow->in_port == OFPP_NONE) {
5848 probability = dpif_ipfix_get_bridge_exporter_probability(ofproto->ipfix);
5849 compose_ipfix_cookie(&cookie);
5851 compose_sample_action(ofproto, odp_actions, flow, probability,
5852 &cookie, sizeof cookie.ipfix);
5855 /* SAMPLE action for sFlow must be first action in any given list of
5856 * actions. At this point we do not have all information required to
5857 * build it. So try to build sample action as complete as possible. */
5859 add_sflow_action(struct xlate_ctx *ctx)
5861 ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto,
5862 &ctx->xout->odp_actions,
5863 &ctx->xin->flow, OVSP_NONE);
5864 ctx->sflow_odp_port = 0;
5865 ctx->sflow_n_outputs = 0;
5868 /* SAMPLE action for IPFIX must be 1st or 2nd action in any given list
5869 * of actions, eventually after the SAMPLE action for sFlow. */
5871 add_ipfix_action(struct xlate_ctx *ctx)
5873 compose_ipfix_action(ctx->ofproto, &ctx->xout->odp_actions,
5877 /* Fix SAMPLE action according to data collected while composing ODP actions.
5878 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
5879 * USERSPACE action's user-cookie which is required for sflow. */
5881 fix_sflow_action(struct xlate_ctx *ctx)
5883 const struct flow *base = &ctx->base_flow;
5884 union user_action_cookie *cookie;
5886 if (!ctx->user_cookie_offset) {
5890 cookie = ofpbuf_at(&ctx->xout->odp_actions, ctx->user_cookie_offset,
5891 sizeof cookie->sflow);
5892 ovs_assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
5894 compose_sflow_cookie(ctx->ofproto, base->vlan_tci,
5895 ctx->sflow_odp_port, ctx->sflow_n_outputs, cookie);
5899 compose_output_action__(struct xlate_ctx *ctx, uint16_t ofp_port,
5902 const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
5903 ovs_be16 flow_vlan_tci;
5904 uint32_t flow_skb_mark;
5905 uint8_t flow_nw_tos;
5906 struct priority_to_dscp *pdscp;
5907 uint32_t out_port, odp_port;
5909 /* If 'struct flow' gets additional metadata, we'll need to zero it out
5910 * before traversing a patch port. */
5911 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 20);
5914 xlate_report(ctx, "Nonexistent output port");
5916 } else if (ofport->up.pp.config & OFPUTIL_PC_NO_FWD) {
5917 xlate_report(ctx, "OFPPC_NO_FWD set, skipping output");
5919 } else if (check_stp && !stp_forward_in_state(ofport->stp_state)) {
5920 xlate_report(ctx, "STP not in forwarding state, skipping output");
5924 if (netdev_vport_is_patch(ofport->up.netdev)) {
5925 struct ofport_dpif *peer = ofport_get_peer(ofport);
5926 struct flow old_flow = ctx->xin->flow;
5927 const struct ofproto_dpif *peer_ofproto;
5928 enum slow_path_reason special;
5929 struct ofport_dpif *in_port;
5932 xlate_report(ctx, "Nonexistent patch port peer");
5936 peer_ofproto = ofproto_dpif_cast(peer->up.ofproto);
5937 if (peer_ofproto->backer != ctx->ofproto->backer) {
5938 xlate_report(ctx, "Patch port peer on a different datapath");
5942 ctx->ofproto = ofproto_dpif_cast(peer->up.ofproto);
5943 ctx->xin->flow.in_port = peer->up.ofp_port;
5944 ctx->xin->flow.metadata = htonll(0);
5945 memset(&ctx->xin->flow.tunnel, 0, sizeof ctx->xin->flow.tunnel);
5946 memset(ctx->xin->flow.regs, 0, sizeof ctx->xin->flow.regs);
5948 in_port = get_ofp_port(ctx->ofproto, ctx->xin->flow.in_port);
5949 special = process_special(ctx->ofproto, &ctx->xin->flow, in_port,
5952 ctx->xout->slow = special;
5953 } else if (!in_port || may_receive(in_port, ctx)) {
5954 if (!in_port || stp_forward_in_state(in_port->stp_state)) {
5955 xlate_table_action(ctx, ctx->xin->flow.in_port, 0, true);
5957 /* Forwarding is disabled by STP. Let OFPP_NORMAL and the
5958 * learning action look at the packet, then drop it. */
5959 struct flow old_base_flow = ctx->base_flow;
5960 size_t old_size = ctx->xout->odp_actions.size;
5961 xlate_table_action(ctx, ctx->xin->flow.in_port, 0, true);
5962 ctx->base_flow = old_base_flow;
5963 ctx->xout->odp_actions.size = old_size;
5967 ctx->xin->flow = old_flow;
5968 ctx->ofproto = ofproto_dpif_cast(ofport->up.ofproto);
5970 if (ctx->xin->resubmit_stats) {
5971 netdev_vport_inc_tx(ofport->up.netdev, ctx->xin->resubmit_stats);
5972 netdev_vport_inc_rx(peer->up.netdev, ctx->xin->resubmit_stats);
5978 flow_vlan_tci = ctx->xin->flow.vlan_tci;
5979 flow_skb_mark = ctx->xin->flow.skb_mark;
5980 flow_nw_tos = ctx->xin->flow.nw_tos;
5982 pdscp = get_priority(ofport, ctx->xin->flow.skb_priority);
5984 ctx->xin->flow.nw_tos &= ~IP_DSCP_MASK;
5985 ctx->xin->flow.nw_tos |= pdscp->dscp;
5988 if (ofport->tnl_port) {
5989 /* Save tunnel metadata so that changes made due to
5990 * the Logical (tunnel) Port are not visible for any further
5991 * matches, while explicit set actions on tunnel metadata are.
5993 struct flow_tnl flow_tnl = ctx->xin->flow.tunnel;
5994 odp_port = tnl_port_send(ofport->tnl_port, &ctx->xin->flow);
5995 if (odp_port == OVSP_NONE) {
5996 xlate_report(ctx, "Tunneling decided against output");
5997 goto out; /* restore flow_nw_tos */
5999 if (ctx->xin->flow.tunnel.ip_dst == ctx->orig_tunnel_ip_dst) {
6000 xlate_report(ctx, "Not tunneling to our own address");
6001 goto out; /* restore flow_nw_tos */
6003 if (ctx->xin->resubmit_stats) {
6004 netdev_vport_inc_tx(ofport->up.netdev, ctx->xin->resubmit_stats);
6006 out_port = odp_port;
6007 commit_odp_tunnel_action(&ctx->xin->flow, &ctx->base_flow,
6008 &ctx->xout->odp_actions);
6009 ctx->xin->flow.tunnel = flow_tnl; /* Restore tunnel metadata */
6011 odp_port = ofport->odp_port;
6012 out_port = vsp_realdev_to_vlandev(ctx->ofproto, odp_port,
6013 ctx->xin->flow.vlan_tci);
6014 if (out_port != odp_port) {
6015 ctx->xin->flow.vlan_tci = htons(0);
6017 ctx->xin->flow.skb_mark &= ~IPSEC_MARK;
6019 commit_odp_actions(&ctx->xin->flow, &ctx->base_flow,
6020 &ctx->xout->odp_actions);
6021 nl_msg_put_u32(&ctx->xout->odp_actions, OVS_ACTION_ATTR_OUTPUT, out_port);
6023 ctx->sflow_odp_port = odp_port;
6024 ctx->sflow_n_outputs++;
6025 ctx->xout->nf_output_iface = ofp_port;
6028 ctx->xin->flow.vlan_tci = flow_vlan_tci;
6029 ctx->xin->flow.skb_mark = flow_skb_mark;
6031 ctx->xin->flow.nw_tos = flow_nw_tos;
6035 compose_output_action(struct xlate_ctx *ctx, uint16_t ofp_port)
6037 compose_output_action__(ctx, ofp_port, true);
6041 tag_the_flow(struct xlate_ctx *ctx, struct rule_dpif *rule)
6043 struct ofproto_dpif *ofproto = ctx->ofproto;
6044 uint8_t table_id = ctx->table_id;
6046 if (table_id > 0 && table_id < N_TABLES) {
6047 struct table_dpif *table = &ofproto->tables[table_id];
6048 if (table->other_table) {
6049 ctx->xout->tags |= (rule && rule->tag
6051 : rule_calculate_tag(&ctx->xin->flow,
6052 &table->other_table->mask,
6058 /* Common rule processing in one place to avoid duplicating code. */
6059 static struct rule_dpif *
6060 ctx_rule_hooks(struct xlate_ctx *ctx, struct rule_dpif *rule,
6063 if (ctx->xin->resubmit_hook) {
6064 ctx->xin->resubmit_hook(ctx, rule);
6066 if (rule == NULL && may_packet_in) {
6068 * check if table configuration flags
6069 * OFPTC_TABLE_MISS_CONTROLLER, default.
6070 * OFPTC_TABLE_MISS_CONTINUE,
6071 * OFPTC_TABLE_MISS_DROP
6072 * When OF1.0, OFPTC_TABLE_MISS_CONTINUE is used. What to do?
6074 rule = rule_dpif_miss_rule(ctx->ofproto, &ctx->xin->flow);
6076 if (rule && ctx->xin->resubmit_stats) {
6077 rule_credit_stats(rule, ctx->xin->resubmit_stats);
6083 xlate_table_action(struct xlate_ctx *ctx,
6084 uint16_t in_port, uint8_t table_id, bool may_packet_in)
6086 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
6087 struct rule_dpif *rule;
6088 uint16_t old_in_port = ctx->xin->flow.in_port;
6089 uint8_t old_table_id = ctx->table_id;
6091 ctx->table_id = table_id;
6093 /* Look up a flow with 'in_port' as the input port. */
6094 ctx->xin->flow.in_port = in_port;
6095 rule = rule_dpif_lookup__(ctx->ofproto, &ctx->xin->flow, table_id);
6097 tag_the_flow(ctx, rule);
6099 /* Restore the original input port. Otherwise OFPP_NORMAL and
6100 * OFPP_IN_PORT will have surprising behavior. */
6101 ctx->xin->flow.in_port = old_in_port;
6103 rule = ctx_rule_hooks(ctx, rule, may_packet_in);
6106 struct rule_dpif *old_rule = ctx->rule;
6110 do_xlate_actions(rule->up.ofpacts, rule->up.ofpacts_len, ctx);
6111 ctx->rule = old_rule;
6115 ctx->table_id = old_table_id;
6117 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
6119 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
6120 MAX_RESUBMIT_RECURSION);
6121 ctx->max_resubmit_trigger = true;
6126 xlate_ofpact_resubmit(struct xlate_ctx *ctx,
6127 const struct ofpact_resubmit *resubmit)
6132 in_port = resubmit->in_port;
6133 if (in_port == OFPP_IN_PORT) {
6134 in_port = ctx->xin->flow.in_port;
6137 table_id = resubmit->table_id;
6138 if (table_id == 255) {
6139 table_id = ctx->table_id;
6142 xlate_table_action(ctx, in_port, table_id, false);
6146 flood_packets(struct xlate_ctx *ctx, bool all)
6148 struct ofport_dpif *ofport;
6150 HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) {
6151 uint16_t ofp_port = ofport->up.ofp_port;
6153 if (ofp_port == ctx->xin->flow.in_port) {
6158 compose_output_action__(ctx, ofp_port, false);
6159 } else if (!(ofport->up.pp.config & OFPUTIL_PC_NO_FLOOD)) {
6160 compose_output_action(ctx, ofp_port);
6164 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
6168 execute_controller_action(struct xlate_ctx *ctx, int len,
6169 enum ofp_packet_in_reason reason,
6170 uint16_t controller_id)
6172 struct ofputil_packet_in pin;
6173 struct ofpbuf *packet;
6175 ovs_assert(!ctx->xout->slow || ctx->xout->slow == SLOW_CONTROLLER);
6176 ctx->xout->slow = SLOW_CONTROLLER;
6177 if (!ctx->xin->packet) {
6181 packet = ofpbuf_clone(ctx->xin->packet);
6183 if (packet->l2 && packet->l3) {
6184 struct eth_header *eh;
6185 uint16_t mpls_depth;
6187 eth_pop_vlan(packet);
6190 memcpy(eh->eth_src, ctx->xin->flow.dl_src, sizeof eh->eth_src);
6191 memcpy(eh->eth_dst, ctx->xin->flow.dl_dst, sizeof eh->eth_dst);
6193 if (ctx->xin->flow.vlan_tci & htons(VLAN_CFI)) {
6194 eth_push_vlan(packet, ctx->xin->flow.vlan_tci);
6197 mpls_depth = eth_mpls_depth(packet);
6199 if (mpls_depth < ctx->xin->flow.mpls_depth) {
6200 push_mpls(packet, ctx->xin->flow.dl_type, ctx->xin->flow.mpls_lse);
6201 } else if (mpls_depth > ctx->xin->flow.mpls_depth) {
6202 pop_mpls(packet, ctx->xin->flow.dl_type);
6203 } else if (mpls_depth) {
6204 set_mpls_lse(packet, ctx->xin->flow.mpls_lse);
6208 if (ctx->xin->flow.dl_type == htons(ETH_TYPE_IP)) {
6209 packet_set_ipv4(packet, ctx->xin->flow.nw_src,
6210 ctx->xin->flow.nw_dst, ctx->xin->flow.nw_tos,
6211 ctx->xin->flow.nw_ttl);
6215 if (ctx->xin->flow.nw_proto == IPPROTO_TCP) {
6216 packet_set_tcp_port(packet, ctx->xin->flow.tp_src,
6217 ctx->xin->flow.tp_dst);
6218 } else if (ctx->xin->flow.nw_proto == IPPROTO_UDP) {
6219 packet_set_udp_port(packet, ctx->xin->flow.tp_src,
6220 ctx->xin->flow.tp_dst);
6226 pin.packet = packet->data;
6227 pin.packet_len = packet->size;
6228 pin.reason = reason;
6229 pin.controller_id = controller_id;
6230 pin.table_id = ctx->table_id;
6231 pin.cookie = ctx->rule ? ctx->rule->up.flow_cookie : 0;
6234 flow_get_metadata(&ctx->xin->flow, &pin.fmd);
6236 connmgr_send_packet_in(ctx->ofproto->up.connmgr, &pin);
6237 ofpbuf_delete(packet);
6241 execute_mpls_push_action(struct xlate_ctx *ctx, ovs_be16 eth_type)
6243 ovs_assert(eth_type_mpls(eth_type));
6245 if (ctx->base_flow.mpls_depth) {
6246 ctx->xin->flow.mpls_lse &= ~htonl(MPLS_BOS_MASK);
6247 ctx->xin->flow.mpls_depth++;
6252 if (ctx->xin->flow.dl_type == htons(ETH_TYPE_IPV6)) {
6253 label = htonl(0x2); /* IPV6 Explicit Null. */
6255 label = htonl(0x0); /* IPV4 Explicit Null. */
6257 tc = (ctx->xin->flow.nw_tos & IP_DSCP_MASK) >> 2;
6258 ttl = ctx->xin->flow.nw_ttl ? ctx->xin->flow.nw_ttl : 0x40;
6259 ctx->xin->flow.mpls_lse = set_mpls_lse_values(ttl, tc, 1, label);
6260 ctx->xin->flow.mpls_depth = 1;
6262 ctx->xin->flow.dl_type = eth_type;
6266 execute_mpls_pop_action(struct xlate_ctx *ctx, ovs_be16 eth_type)
6268 ovs_assert(eth_type_mpls(ctx->xin->flow.dl_type));
6269 ovs_assert(!eth_type_mpls(eth_type));
6271 if (ctx->xin->flow.mpls_depth) {
6272 ctx->xin->flow.mpls_depth--;
6273 ctx->xin->flow.mpls_lse = htonl(0);
6274 if (!ctx->xin->flow.mpls_depth) {
6275 ctx->xin->flow.dl_type = eth_type;
6281 compose_dec_ttl(struct xlate_ctx *ctx, struct ofpact_cnt_ids *ids)
6283 if (ctx->xin->flow.dl_type != htons(ETH_TYPE_IP) &&
6284 ctx->xin->flow.dl_type != htons(ETH_TYPE_IPV6)) {
6288 if (ctx->xin->flow.nw_ttl > 1) {
6289 ctx->xin->flow.nw_ttl--;
6294 for (i = 0; i < ids->n_controllers; i++) {
6295 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL,
6299 /* Stop processing for current table. */
6305 execute_set_mpls_ttl_action(struct xlate_ctx *ctx, uint8_t ttl)
6307 if (!eth_type_mpls(ctx->xin->flow.dl_type)) {
6311 set_mpls_lse_ttl(&ctx->xin->flow.mpls_lse, ttl);
6316 execute_dec_mpls_ttl_action(struct xlate_ctx *ctx)
6318 uint8_t ttl = mpls_lse_to_ttl(ctx->xin->flow.mpls_lse);
6320 if (!eth_type_mpls(ctx->xin->flow.dl_type)) {
6326 set_mpls_lse_ttl(&ctx->xin->flow.mpls_lse, ttl);
6329 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL, 0);
6331 /* Stop processing for current table. */
6337 xlate_output_action(struct xlate_ctx *ctx,
6338 uint16_t port, uint16_t max_len, bool may_packet_in)
6340 uint16_t prev_nf_output_iface = ctx->xout->nf_output_iface;
6342 ctx->xout->nf_output_iface = NF_OUT_DROP;
6346 compose_output_action(ctx, ctx->xin->flow.in_port);
6349 xlate_table_action(ctx, ctx->xin->flow.in_port, 0, may_packet_in);
6355 flood_packets(ctx, false);
6358 flood_packets(ctx, true);
6360 case OFPP_CONTROLLER:
6361 execute_controller_action(ctx, max_len, OFPR_ACTION, 0);
6367 if (port != ctx->xin->flow.in_port) {
6368 compose_output_action(ctx, port);
6370 xlate_report(ctx, "skipping output to input port");
6375 if (prev_nf_output_iface == NF_OUT_FLOOD) {
6376 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
6377 } else if (ctx->xout->nf_output_iface == NF_OUT_DROP) {
6378 ctx->xout->nf_output_iface = prev_nf_output_iface;
6379 } else if (prev_nf_output_iface != NF_OUT_DROP &&
6380 ctx->xout->nf_output_iface != NF_OUT_FLOOD) {
6381 ctx->xout->nf_output_iface = NF_OUT_MULTI;
6386 xlate_output_reg_action(struct xlate_ctx *ctx,
6387 const struct ofpact_output_reg *or)
6389 uint64_t port = mf_get_subfield(&or->src, &ctx->xin->flow);
6390 if (port <= UINT16_MAX) {
6391 xlate_output_action(ctx, port, or->max_len, false);
6396 xlate_enqueue_action(struct xlate_ctx *ctx,
6397 const struct ofpact_enqueue *enqueue)
6399 uint16_t ofp_port = enqueue->port;
6400 uint32_t queue_id = enqueue->queue;
6401 uint32_t flow_priority, priority;
6404 /* Translate queue to priority. */
6405 error = dpif_queue_to_priority(ctx->ofproto->backer->dpif,
6406 queue_id, &priority);
6408 /* Fall back to ordinary output action. */
6409 xlate_output_action(ctx, enqueue->port, 0, false);
6413 /* Check output port. */
6414 if (ofp_port == OFPP_IN_PORT) {
6415 ofp_port = ctx->xin->flow.in_port;
6416 } else if (ofp_port == ctx->xin->flow.in_port) {
6420 /* Add datapath actions. */
6421 flow_priority = ctx->xin->flow.skb_priority;
6422 ctx->xin->flow.skb_priority = priority;
6423 compose_output_action(ctx, ofp_port);
6424 ctx->xin->flow.skb_priority = flow_priority;
6426 /* Update NetFlow output port. */
6427 if (ctx->xout->nf_output_iface == NF_OUT_DROP) {
6428 ctx->xout->nf_output_iface = ofp_port;
6429 } else if (ctx->xout->nf_output_iface != NF_OUT_FLOOD) {
6430 ctx->xout->nf_output_iface = NF_OUT_MULTI;
6435 xlate_set_queue_action(struct xlate_ctx *ctx, uint32_t queue_id)
6437 uint32_t skb_priority;
6439 if (!dpif_queue_to_priority(ctx->ofproto->backer->dpif,
6440 queue_id, &skb_priority)) {
6441 ctx->xin->flow.skb_priority = skb_priority;
6443 /* Couldn't translate queue to a priority. Nothing to do. A warning
6444 * has already been logged. */
6449 slave_enabled_cb(uint16_t ofp_port, void *ofproto_)
6451 struct ofproto_dpif *ofproto = ofproto_;
6452 struct ofport_dpif *port;
6462 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
6465 port = get_ofp_port(ofproto, ofp_port);
6466 return port ? port->may_enable : false;
6471 xlate_bundle_action(struct xlate_ctx *ctx,
6472 const struct ofpact_bundle *bundle)
6476 port = bundle_execute(bundle, &ctx->xin->flow, slave_enabled_cb,
6478 if (bundle->dst.field) {
6479 nxm_reg_load(&bundle->dst, port, &ctx->xin->flow);
6481 xlate_output_action(ctx, port, 0, false);
6486 xlate_learn_action(struct xlate_ctx *ctx,
6487 const struct ofpact_learn *learn)
6489 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
6490 struct ofputil_flow_mod fm;
6491 uint64_t ofpacts_stub[1024 / 8];
6492 struct ofpbuf ofpacts;
6495 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
6496 learn_execute(learn, &ctx->xin->flow, &fm, &ofpacts);
6498 error = ofproto_flow_mod(&ctx->ofproto->up, &fm);
6499 if (error && !VLOG_DROP_WARN(&rl)) {
6500 VLOG_WARN("learning action failed to modify flow table (%s)",
6501 ofperr_get_name(error));
6504 ofpbuf_uninit(&ofpacts);
6507 /* Reduces '*timeout' to no more than 'max'. A value of zero in either case
6508 * means "infinite". */
6510 reduce_timeout(uint16_t max, uint16_t *timeout)
6512 if (max && (!*timeout || *timeout > max)) {
6518 xlate_fin_timeout(struct xlate_ctx *ctx,
6519 const struct ofpact_fin_timeout *oft)
6521 if (ctx->xin->tcp_flags & (TCP_FIN | TCP_RST) && ctx->rule) {
6522 struct rule_dpif *rule = ctx->rule;
6524 reduce_timeout(oft->fin_idle_timeout, &rule->up.idle_timeout);
6525 reduce_timeout(oft->fin_hard_timeout, &rule->up.hard_timeout);
6530 xlate_sample_action(struct xlate_ctx *ctx,
6531 const struct ofpact_sample *os)
6533 union user_action_cookie cookie;
6534 /* Scale the probability from 16-bit to 32-bit while representing
6535 * the same percentage. */
6536 uint32_t probability = (os->probability << 16) | os->probability;
6538 commit_odp_actions(&ctx->xin->flow, &ctx->base_flow,
6539 &ctx->xout->odp_actions);
6541 compose_flow_sample_cookie(os->probability, os->collector_set_id,
6542 os->obs_domain_id, os->obs_point_id, &cookie);
6543 compose_sample_action(ctx->ofproto, &ctx->xout->odp_actions, &ctx->xin->flow,
6544 probability, &cookie, sizeof cookie.flow_sample);
6548 may_receive(const struct ofport_dpif *port, struct xlate_ctx *ctx)
6550 if (port->up.pp.config & (eth_addr_equals(ctx->xin->flow.dl_dst,
6552 ? OFPUTIL_PC_NO_RECV_STP
6553 : OFPUTIL_PC_NO_RECV)) {
6557 /* Only drop packets here if both forwarding and learning are
6558 * disabled. If just learning is enabled, we need to have
6559 * OFPP_NORMAL and the learning action have a look at the packet
6560 * before we can drop it. */
6561 if (!stp_forward_in_state(port->stp_state)
6562 && !stp_learn_in_state(port->stp_state)) {
6570 tunnel_ecn_ok(struct xlate_ctx *ctx)
6572 if (is_ip_any(&ctx->base_flow)
6573 && (ctx->xin->flow.tunnel.ip_tos & IP_ECN_MASK) == IP_ECN_CE) {
6574 if ((ctx->base_flow.nw_tos & IP_ECN_MASK) == IP_ECN_NOT_ECT) {
6575 VLOG_WARN_RL(&rl, "dropping tunnel packet marked ECN CE"
6576 " but is not ECN capable");
6579 /* Set the ECN CE value in the tunneled packet. */
6580 ctx->xin->flow.nw_tos |= IP_ECN_CE;
6588 do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len,
6589 struct xlate_ctx *ctx)
6591 bool was_evictable = true;
6592 const struct ofpact *a;
6595 /* Don't let the rule we're working on get evicted underneath us. */
6596 was_evictable = ctx->rule->up.evictable;
6597 ctx->rule->up.evictable = false;
6600 do_xlate_actions_again:
6601 OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) {
6602 struct ofpact_controller *controller;
6603 const struct ofpact_metadata *metadata;
6611 xlate_output_action(ctx, ofpact_get_OUTPUT(a)->port,
6612 ofpact_get_OUTPUT(a)->max_len, true);
6615 case OFPACT_CONTROLLER:
6616 controller = ofpact_get_CONTROLLER(a);
6617 execute_controller_action(ctx, controller->max_len,
6619 controller->controller_id);
6622 case OFPACT_ENQUEUE:
6623 xlate_enqueue_action(ctx, ofpact_get_ENQUEUE(a));
6626 case OFPACT_SET_VLAN_VID:
6627 ctx->xin->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
6628 ctx->xin->flow.vlan_tci |=
6629 (htons(ofpact_get_SET_VLAN_VID(a)->vlan_vid)
6633 case OFPACT_SET_VLAN_PCP:
6634 ctx->xin->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
6635 ctx->xin->flow.vlan_tci |=
6636 htons((ofpact_get_SET_VLAN_PCP(a)->vlan_pcp << VLAN_PCP_SHIFT)
6640 case OFPACT_STRIP_VLAN:
6641 ctx->xin->flow.vlan_tci = htons(0);
6644 case OFPACT_PUSH_VLAN:
6645 /* XXX 802.1AD(QinQ) */
6646 ctx->xin->flow.vlan_tci = htons(VLAN_CFI);
6649 case OFPACT_SET_ETH_SRC:
6650 memcpy(ctx->xin->flow.dl_src, ofpact_get_SET_ETH_SRC(a)->mac,
6654 case OFPACT_SET_ETH_DST:
6655 memcpy(ctx->xin->flow.dl_dst, ofpact_get_SET_ETH_DST(a)->mac,
6659 case OFPACT_SET_IPV4_SRC:
6660 if (ctx->xin->flow.dl_type == htons(ETH_TYPE_IP)) {
6661 ctx->xin->flow.nw_src = ofpact_get_SET_IPV4_SRC(a)->ipv4;
6665 case OFPACT_SET_IPV4_DST:
6666 if (ctx->xin->flow.dl_type == htons(ETH_TYPE_IP)) {
6667 ctx->xin->flow.nw_dst = ofpact_get_SET_IPV4_DST(a)->ipv4;
6671 case OFPACT_SET_IPV4_DSCP:
6672 /* OpenFlow 1.0 only supports IPv4. */
6673 if (ctx->xin->flow.dl_type == htons(ETH_TYPE_IP)) {
6674 ctx->xin->flow.nw_tos &= ~IP_DSCP_MASK;
6675 ctx->xin->flow.nw_tos |= ofpact_get_SET_IPV4_DSCP(a)->dscp;
6679 case OFPACT_SET_L4_SRC_PORT:
6680 if (is_ip_any(&ctx->xin->flow)) {
6681 ctx->xin->flow.tp_src =
6682 htons(ofpact_get_SET_L4_SRC_PORT(a)->port);
6686 case OFPACT_SET_L4_DST_PORT:
6687 if (is_ip_any(&ctx->xin->flow)) {
6688 ctx->xin->flow.tp_dst =
6689 htons(ofpact_get_SET_L4_DST_PORT(a)->port);
6693 case OFPACT_RESUBMIT:
6694 xlate_ofpact_resubmit(ctx, ofpact_get_RESUBMIT(a));
6697 case OFPACT_SET_TUNNEL:
6698 ctx->xin->flow.tunnel.tun_id =
6699 htonll(ofpact_get_SET_TUNNEL(a)->tun_id);
6702 case OFPACT_SET_QUEUE:
6703 xlate_set_queue_action(ctx, ofpact_get_SET_QUEUE(a)->queue_id);
6706 case OFPACT_POP_QUEUE:
6707 ctx->xin->flow.skb_priority = ctx->orig_skb_priority;
6710 case OFPACT_REG_MOVE:
6711 nxm_execute_reg_move(ofpact_get_REG_MOVE(a), &ctx->xin->flow);
6714 case OFPACT_REG_LOAD:
6715 nxm_execute_reg_load(ofpact_get_REG_LOAD(a), &ctx->xin->flow);
6718 case OFPACT_STACK_PUSH:
6719 nxm_execute_stack_push(ofpact_get_STACK_PUSH(a), &ctx->xin->flow,
6723 case OFPACT_STACK_POP:
6724 nxm_execute_stack_pop(ofpact_get_STACK_POP(a), &ctx->xin->flow,
6728 case OFPACT_PUSH_MPLS:
6729 execute_mpls_push_action(ctx, ofpact_get_PUSH_MPLS(a)->ethertype);
6732 case OFPACT_POP_MPLS:
6733 execute_mpls_pop_action(ctx, ofpact_get_POP_MPLS(a)->ethertype);
6736 case OFPACT_SET_MPLS_TTL:
6737 if (execute_set_mpls_ttl_action(ctx,
6738 ofpact_get_SET_MPLS_TTL(a)->ttl)) {
6743 case OFPACT_DEC_MPLS_TTL:
6744 if (execute_dec_mpls_ttl_action(ctx)) {
6749 case OFPACT_DEC_TTL:
6750 if (compose_dec_ttl(ctx, ofpact_get_DEC_TTL(a))) {
6756 /* Nothing to do. */
6759 case OFPACT_MULTIPATH:
6760 multipath_execute(ofpact_get_MULTIPATH(a), &ctx->xin->flow);
6764 ctx->ofproto->has_bundle_action = true;
6765 xlate_bundle_action(ctx, ofpact_get_BUNDLE(a));
6768 case OFPACT_OUTPUT_REG:
6769 xlate_output_reg_action(ctx, ofpact_get_OUTPUT_REG(a));
6773 ctx->xout->has_learn = true;
6774 if (ctx->xin->may_learn) {
6775 xlate_learn_action(ctx, ofpact_get_LEARN(a));
6783 case OFPACT_FIN_TIMEOUT:
6784 ctx->xout->has_fin_timeout = true;
6785 xlate_fin_timeout(ctx, ofpact_get_FIN_TIMEOUT(a));
6788 case OFPACT_CLEAR_ACTIONS:
6790 * Nothing to do because writa-actions is not supported for now.
6791 * When writa-actions is supported, clear-actions also must
6792 * be supported at the same time.
6796 case OFPACT_WRITE_METADATA:
6797 metadata = ofpact_get_WRITE_METADATA(a);
6798 ctx->xin->flow.metadata &= ~metadata->mask;
6799 ctx->xin->flow.metadata |= metadata->metadata & metadata->mask;
6802 case OFPACT_GOTO_TABLE: {
6803 /* It is assumed that goto-table is the last action. */
6804 struct ofpact_goto_table *ogt = ofpact_get_GOTO_TABLE(a);
6805 struct rule_dpif *rule;
6807 ovs_assert(ctx->table_id < ogt->table_id);
6809 ctx->table_id = ogt->table_id;
6811 /* Look up a flow from the new table. */
6812 rule = rule_dpif_lookup__(ctx->ofproto, &ctx->xin->flow, ctx->table_id);
6814 tag_the_flow(ctx, rule);
6816 rule = ctx_rule_hooks(ctx, rule, true);
6820 ctx->rule->up.evictable = was_evictable;
6823 was_evictable = rule->up.evictable;
6824 rule->up.evictable = false;
6826 /* Tail recursion removal. */
6827 ofpacts = rule->up.ofpacts;
6828 ofpacts_len = rule->up.ofpacts_len;
6829 goto do_xlate_actions_again;
6835 xlate_sample_action(ctx, ofpact_get_SAMPLE(a));
6842 ctx->rule->up.evictable = was_evictable;
6847 xlate_in_init(struct xlate_in *xin, struct ofproto_dpif *ofproto,
6848 const struct flow *flow,
6849 const struct initial_vals *initial_vals,
6850 struct rule_dpif *rule, uint8_t tcp_flags,
6851 const struct ofpbuf *packet)
6853 xin->ofproto = ofproto;
6855 xin->packet = packet;
6856 xin->may_learn = packet != NULL;
6858 xin->ofpacts = NULL;
6859 xin->ofpacts_len = 0;
6860 xin->tcp_flags = tcp_flags;
6861 xin->resubmit_hook = NULL;
6862 xin->report_hook = NULL;
6863 xin->resubmit_stats = NULL;
6866 xin->initial_vals = *initial_vals;
6868 xin->initial_vals.vlan_tci = xin->flow.vlan_tci;
6873 xlate_out_uninit(struct xlate_out *xout)
6876 ofpbuf_uninit(&xout->odp_actions);
6880 /* Translates the 'ofpacts_len' bytes of "struct ofpacts" starting at 'ofpacts'
6881 * into datapath actions in 'odp_actions', using 'ctx'. */
6883 xlate_actions(struct xlate_in *xin, struct xlate_out *xout)
6885 /* Normally false. Set to true if we ever hit MAX_RESUBMIT_RECURSION, so
6886 * that in the future we always keep a copy of the original flow for
6887 * tracing purposes. */
6888 static bool hit_resubmit_limit;
6890 enum slow_path_reason special;
6891 const struct ofpact *ofpacts;
6892 struct ofport_dpif *in_port;
6893 struct flow orig_flow;
6894 struct xlate_ctx ctx;
6897 COVERAGE_INC(ofproto_dpif_xlate);
6899 /* Flow initialization rules:
6900 * - 'base_flow' must match the kernel's view of the packet at the
6901 * time that action processing starts. 'flow' represents any
6902 * transformations we wish to make through actions.
6903 * - By default 'base_flow' and 'flow' are the same since the input
6904 * packet matches the output before any actions are applied.
6905 * - When using VLAN splinters, 'base_flow''s VLAN is set to the value
6906 * of the received packet as seen by the kernel. If we later output
6907 * to another device without any modifications this will cause us to
6908 * insert a new tag since the original one was stripped off by the
6910 * - Tunnel metadata as received is retained in 'flow'. This allows
6911 * tunnel metadata matching also in later tables.
6912 * Since a kernel action for setting the tunnel metadata will only be
6913 * generated with actual tunnel output, changing the tunnel metadata
6914 * values in 'flow' (such as tun_id) will only have effect with a later
6915 * tunnel output action.
6916 * - Tunnel 'base_flow' is completely cleared since that is what the
6917 * kernel does. If we wish to maintain the original values an action
6918 * needs to be generated. */
6923 ctx.ofproto = xin->ofproto;
6924 ctx.rule = xin->rule;
6926 ctx.base_flow = ctx.xin->flow;
6927 ctx.base_flow.vlan_tci = xin->initial_vals.vlan_tci;
6928 memset(&ctx.base_flow.tunnel, 0, sizeof ctx.base_flow.tunnel);
6929 ctx.orig_tunnel_ip_dst = ctx.xin->flow.tunnel.ip_dst;
6933 ctx.xout->has_learn = false;
6934 ctx.xout->has_normal = false;
6935 ctx.xout->has_fin_timeout = false;
6936 ctx.xout->nf_output_iface = NF_OUT_DROP;
6937 ctx.xout->mirrors = 0;
6939 ofpbuf_use_stub(&ctx.xout->odp_actions, ctx.xout->odp_actions_stub,
6940 sizeof ctx.xout->odp_actions_stub);
6941 ofpbuf_reserve(&ctx.xout->odp_actions, NL_A_U32_SIZE);
6944 ctx.max_resubmit_trigger = false;
6945 ctx.orig_skb_priority = ctx.xin->flow.skb_priority;
6950 ofpacts = xin->ofpacts;
6951 ofpacts_len = xin->ofpacts_len;
6952 } else if (xin->rule) {
6953 ofpacts = xin->rule->up.ofpacts;
6954 ofpacts_len = xin->rule->up.ofpacts_len;
6959 ofpbuf_use_stub(&ctx.stack, ctx.init_stack, sizeof ctx.init_stack);
6961 if (ctx.ofproto->has_mirrors || hit_resubmit_limit) {
6962 /* Do this conditionally because the copy is expensive enough that it
6963 * shows up in profiles. */
6964 orig_flow = ctx.xin->flow;
6967 if (ctx.xin->flow.nw_frag & FLOW_NW_FRAG_ANY) {
6968 switch (ctx.ofproto->up.frag_handling) {
6969 case OFPC_FRAG_NORMAL:
6970 /* We must pretend that transport ports are unavailable. */
6971 ctx.xin->flow.tp_src = ctx.base_flow.tp_src = htons(0);
6972 ctx.xin->flow.tp_dst = ctx.base_flow.tp_dst = htons(0);
6975 case OFPC_FRAG_DROP:
6978 case OFPC_FRAG_REASM:
6981 case OFPC_FRAG_NX_MATCH:
6982 /* Nothing to do. */
6985 case OFPC_INVALID_TTL_TO_CONTROLLER:
6990 in_port = get_ofp_port(ctx.ofproto, ctx.xin->flow.in_port);
6991 special = process_special(ctx.ofproto, &ctx.xin->flow, in_port,
6994 ctx.xout->slow = special;
6996 static struct vlog_rate_limit trace_rl = VLOG_RATE_LIMIT_INIT(1, 1);
6997 struct initial_vals initial_vals;
6998 size_t sample_actions_len;
6999 uint32_t local_odp_port;
7001 initial_vals.vlan_tci = ctx.base_flow.vlan_tci;
7003 add_sflow_action(&ctx);
7004 add_ipfix_action(&ctx);
7005 sample_actions_len = ctx.xout->odp_actions.size;
7007 if (tunnel_ecn_ok(&ctx) && (!in_port || may_receive(in_port, &ctx))) {
7008 do_xlate_actions(ofpacts, ofpacts_len, &ctx);
7010 /* We've let OFPP_NORMAL and the learning action look at the
7011 * packet, so drop it now if forwarding is disabled. */
7012 if (in_port && !stp_forward_in_state(in_port->stp_state)) {
7013 ctx.xout->odp_actions.size = sample_actions_len;
7017 if (ctx.max_resubmit_trigger && !ctx.xin->resubmit_hook) {
7018 if (!hit_resubmit_limit) {
7019 /* We didn't record the original flow. Make sure we do from
7021 hit_resubmit_limit = true;
7022 } else if (!VLOG_DROP_ERR(&trace_rl)) {
7023 struct ds ds = DS_EMPTY_INITIALIZER;
7025 ofproto_trace(ctx.ofproto, &orig_flow, ctx.xin->packet,
7026 &initial_vals, &ds);
7027 VLOG_ERR("Trace triggered by excessive resubmit "
7028 "recursion:\n%s", ds_cstr(&ds));
7033 local_odp_port = ofp_port_to_odp_port(ctx.ofproto, OFPP_LOCAL);
7034 if (!connmgr_must_output_local(ctx.ofproto->up.connmgr, &ctx.xin->flow,
7036 ctx.xout->odp_actions.data,
7037 ctx.xout->odp_actions.size)) {
7038 compose_output_action(&ctx, OFPP_LOCAL);
7040 if (ctx.ofproto->has_mirrors) {
7041 add_mirror_actions(&ctx, &orig_flow);
7043 fix_sflow_action(&ctx);
7046 ofpbuf_uninit(&ctx.stack);
7049 /* Translates the 'ofpacts_len' bytes of "struct ofpact"s starting at 'ofpacts'
7050 * into datapath actions, using 'ctx', and discards the datapath actions. */
7052 xlate_actions_for_side_effects(struct xlate_in *xin)
7054 struct xlate_out xout;
7056 xlate_actions(xin, &xout);
7057 xlate_out_uninit(&xout);
7061 xlate_report(struct xlate_ctx *ctx, const char *s)
7063 if (ctx->xin->report_hook) {
7064 ctx->xin->report_hook(ctx, s);
7068 /* OFPP_NORMAL implementation. */
7070 static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *);
7072 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
7073 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle',
7074 * the bundle on which the packet was received, returns the VLAN to which the
7077 * Both 'vid' and the return value are in the range 0...4095. */
7079 input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid)
7081 switch (in_bundle->vlan_mode) {
7082 case PORT_VLAN_ACCESS:
7083 return in_bundle->vlan;
7086 case PORT_VLAN_TRUNK:
7089 case PORT_VLAN_NATIVE_UNTAGGED:
7090 case PORT_VLAN_NATIVE_TAGGED:
7091 return vid ? vid : in_bundle->vlan;
7098 /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'.
7099 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
7102 * 'vid' should be the VID obtained from the 802.1Q header that was received as
7103 * part of a packet (specify 0 if there was no 802.1Q header), in the range
7106 input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn)
7108 /* Allow any VID on the OFPP_NONE port. */
7109 if (in_bundle == &ofpp_none_bundle) {
7113 switch (in_bundle->vlan_mode) {
7114 case PORT_VLAN_ACCESS:
7117 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7118 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
7119 "packet received on port %s configured as VLAN "
7120 "%"PRIu16" access port",
7121 in_bundle->ofproto->up.name, vid,
7122 in_bundle->name, in_bundle->vlan);
7128 case PORT_VLAN_NATIVE_UNTAGGED:
7129 case PORT_VLAN_NATIVE_TAGGED:
7131 /* Port must always carry its native VLAN. */
7135 case PORT_VLAN_TRUNK:
7136 if (!ofbundle_includes_vlan(in_bundle, vid)) {
7138 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7139 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet "
7140 "received on port %s not configured for trunking "
7142 in_bundle->ofproto->up.name, vid,
7143 in_bundle->name, vid);
7155 /* Given 'vlan', the VLAN that a packet belongs to, and
7156 * 'out_bundle', a bundle on which the packet is to be output, returns the VID
7157 * that should be included in the 802.1Q header. (If the return value is 0,
7158 * then the 802.1Q header should only be included in the packet if there is a
7161 * Both 'vlan' and the return value are in the range 0...4095. */
7163 output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan)
7165 switch (out_bundle->vlan_mode) {
7166 case PORT_VLAN_ACCESS:
7169 case PORT_VLAN_TRUNK:
7170 case PORT_VLAN_NATIVE_TAGGED:
7173 case PORT_VLAN_NATIVE_UNTAGGED:
7174 return vlan == out_bundle->vlan ? 0 : vlan;
7182 output_normal(struct xlate_ctx *ctx, const struct ofbundle *out_bundle,
7185 struct ofport_dpif *port;
7187 ovs_be16 tci, old_tci;
7189 vid = output_vlan_to_vid(out_bundle, vlan);
7190 if (!out_bundle->bond) {
7191 port = ofbundle_get_a_port(out_bundle);
7193 port = bond_choose_output_slave(out_bundle->bond, &ctx->xin->flow,
7194 vid, &ctx->xout->tags);
7196 /* No slaves enabled, so drop packet. */
7201 old_tci = ctx->xin->flow.vlan_tci;
7203 if (tci || out_bundle->use_priority_tags) {
7204 tci |= ctx->xin->flow.vlan_tci & htons(VLAN_PCP_MASK);
7206 tci |= htons(VLAN_CFI);
7209 ctx->xin->flow.vlan_tci = tci;
7211 compose_output_action(ctx, port->up.ofp_port);
7212 ctx->xin->flow.vlan_tci = old_tci;
7216 mirror_mask_ffs(mirror_mask_t mask)
7218 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
7223 ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
7225 return (bundle->vlan_mode != PORT_VLAN_ACCESS
7226 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
7230 ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan)
7232 return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan);
7235 /* Returns an arbitrary interface within 'bundle'. */
7236 static struct ofport_dpif *
7237 ofbundle_get_a_port(const struct ofbundle *bundle)
7239 return CONTAINER_OF(list_front(&bundle->ports),
7240 struct ofport_dpif, bundle_node);
7244 vlan_is_mirrored(const struct ofmirror *m, int vlan)
7246 return !m->vlans || bitmap_is_set(m->vlans, vlan);
7250 add_mirror_actions(struct xlate_ctx *ctx, const struct flow *orig_flow)
7252 struct ofproto_dpif *ofproto = ctx->ofproto;
7253 mirror_mask_t mirrors;
7254 struct ofbundle *in_bundle;
7257 const struct nlattr *a;
7260 in_bundle = lookup_input_bundle(ctx->ofproto, orig_flow->in_port,
7261 ctx->xin->packet != NULL, NULL);
7265 mirrors = in_bundle->src_mirrors;
7267 /* Drop frames on bundles reserved for mirroring. */
7268 if (in_bundle->mirror_out) {
7269 if (ctx->xin->packet != NULL) {
7270 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7271 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
7272 "%s, which is reserved exclusively for mirroring",
7273 ctx->ofproto->up.name, in_bundle->name);
7279 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
7280 if (!input_vid_is_valid(vid, in_bundle, ctx->xin->packet != NULL)) {
7283 vlan = input_vid_to_vlan(in_bundle, vid);
7285 /* Look at the output ports to check for destination selections. */
7287 NL_ATTR_FOR_EACH (a, left, ctx->xout->odp_actions.data,
7288 ctx->xout->odp_actions.size) {
7289 enum ovs_action_attr type = nl_attr_type(a);
7290 struct ofport_dpif *ofport;
7292 if (type != OVS_ACTION_ATTR_OUTPUT) {
7296 ofport = get_odp_port(ofproto, nl_attr_get_u32(a));
7297 if (ofport && ofport->bundle) {
7298 mirrors |= ofport->bundle->dst_mirrors;
7306 /* Restore the original packet before adding the mirror actions. */
7307 ctx->xin->flow = *orig_flow;
7312 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
7314 if (!vlan_is_mirrored(m, vlan)) {
7315 mirrors = zero_rightmost_1bit(mirrors);
7319 mirrors &= ~m->dup_mirrors;
7320 ctx->xout->mirrors |= m->dup_mirrors;
7322 output_normal(ctx, m->out, vlan);
7323 } else if (vlan != m->out_vlan
7324 && !eth_addr_is_reserved(orig_flow->dl_dst)) {
7325 struct ofbundle *bundle;
7327 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
7328 if (ofbundle_includes_vlan(bundle, m->out_vlan)
7329 && !bundle->mirror_out) {
7330 output_normal(ctx, bundle, m->out_vlan);
7338 update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors,
7339 uint64_t packets, uint64_t bytes)
7345 for (; mirrors; mirrors = zero_rightmost_1bit(mirrors)) {
7348 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
7351 /* In normal circumstances 'm' will not be NULL. However,
7352 * if mirrors are reconfigured, we can temporarily get out
7353 * of sync in facet_revalidate(). We could "correct" the
7354 * mirror list before reaching here, but doing that would
7355 * not properly account the traffic stats we've currently
7356 * accumulated for previous mirror configuration. */
7360 m->packet_count += packets;
7361 m->byte_count += bytes;
7365 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
7366 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
7367 * indicate this; newer upstream kernels use gratuitous ARP requests. */
7369 is_gratuitous_arp(const struct flow *flow)
7371 return (flow->dl_type == htons(ETH_TYPE_ARP)
7372 && eth_addr_is_broadcast(flow->dl_dst)
7373 && (flow->nw_proto == ARP_OP_REPLY
7374 || (flow->nw_proto == ARP_OP_REQUEST
7375 && flow->nw_src == flow->nw_dst)));
7379 update_learning_table(struct ofproto_dpif *ofproto,
7380 const struct flow *flow, int vlan,
7381 struct ofbundle *in_bundle)
7383 struct mac_entry *mac;
7385 /* Don't learn the OFPP_NONE port. */
7386 if (in_bundle == &ofpp_none_bundle) {
7390 if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
7394 mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan);
7395 if (is_gratuitous_arp(flow)) {
7396 /* We don't want to learn from gratuitous ARP packets that are
7397 * reflected back over bond slaves so we lock the learning table. */
7398 if (!in_bundle->bond) {
7399 mac_entry_set_grat_arp_lock(mac);
7400 } else if (mac_entry_is_grat_arp_locked(mac)) {
7405 if (mac_entry_is_new(mac) || mac->port.p != in_bundle) {
7406 /* The log messages here could actually be useful in debugging,
7407 * so keep the rate limit relatively high. */
7408 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
7409 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
7410 "on port %s in VLAN %d",
7411 ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src),
7412 in_bundle->name, vlan);
7414 mac->port.p = in_bundle;
7415 tag_set_add(&ofproto->backer->revalidate_set,
7416 mac_learning_changed(ofproto->ml, mac));
7420 static struct ofbundle *
7421 lookup_input_bundle(const struct ofproto_dpif *ofproto, uint16_t in_port,
7422 bool warn, struct ofport_dpif **in_ofportp)
7424 struct ofport_dpif *ofport;
7426 /* Find the port and bundle for the received packet. */
7427 ofport = get_ofp_port(ofproto, in_port);
7429 *in_ofportp = ofport;
7431 if (ofport && ofport->bundle) {
7432 return ofport->bundle;
7435 /* Special-case OFPP_NONE, which a controller may use as the ingress
7436 * port for traffic that it is sourcing. */
7437 if (in_port == OFPP_NONE) {
7438 return &ofpp_none_bundle;
7441 /* Odd. A few possible reasons here:
7443 * - We deleted a port but there are still a few packets queued up
7446 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
7447 * we don't know about.
7449 * - The ofproto client didn't configure the port as part of a bundle.
7450 * This is particularly likely to happen if a packet was received on the
7451 * port after it was created, but before the client had a chance to
7452 * configure its bundle.
7455 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7457 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
7458 "port %"PRIu16, ofproto->up.name, in_port);
7463 /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or
7464 * dropped. Returns true if they may be forwarded, false if they should be
7467 * 'in_port' must be the ofport_dpif that corresponds to flow->in_port.
7468 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
7470 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
7471 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
7472 * checked by input_vid_is_valid().
7474 * May also add tags to '*tags', although the current implementation only does
7475 * so in one special case.
7478 is_admissible(struct xlate_ctx *ctx, struct ofport_dpif *in_port,
7481 struct ofproto_dpif *ofproto = ctx->ofproto;
7482 struct flow *flow = &ctx->xin->flow;
7483 struct ofbundle *in_bundle = in_port->bundle;
7485 /* Drop frames for reserved multicast addresses
7486 * only if forward_bpdu option is absent. */
7487 if (!ofproto->up.forward_bpdu && eth_addr_is_reserved(flow->dl_dst)) {
7488 xlate_report(ctx, "packet has reserved destination MAC, dropping");
7492 if (in_bundle->bond) {
7493 struct mac_entry *mac;
7495 switch (bond_check_admissibility(in_bundle->bond, in_port,
7496 flow->dl_dst, &ctx->xout->tags)) {
7501 xlate_report(ctx, "bonding refused admissibility, dropping");
7504 case BV_DROP_IF_MOVED:
7505 mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL);
7506 if (mac && mac->port.p != in_bundle &&
7507 (!is_gratuitous_arp(flow)
7508 || mac_entry_is_grat_arp_locked(mac))) {
7509 xlate_report(ctx, "SLB bond thinks this packet looped back, "
7521 xlate_normal(struct xlate_ctx *ctx)
7523 struct ofport_dpif *in_port;
7524 struct ofbundle *in_bundle;
7525 struct mac_entry *mac;
7529 ctx->xout->has_normal = true;
7531 in_bundle = lookup_input_bundle(ctx->ofproto, ctx->xin->flow.in_port,
7532 ctx->xin->packet != NULL, &in_port);
7534 xlate_report(ctx, "no input bundle, dropping");
7538 /* Drop malformed frames. */
7539 if (ctx->xin->flow.dl_type == htons(ETH_TYPE_VLAN) &&
7540 !(ctx->xin->flow.vlan_tci & htons(VLAN_CFI))) {
7541 if (ctx->xin->packet != NULL) {
7542 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7543 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
7544 "VLAN tag received on port %s",
7545 ctx->ofproto->up.name, in_bundle->name);
7547 xlate_report(ctx, "partial VLAN tag, dropping");
7551 /* Drop frames on bundles reserved for mirroring. */
7552 if (in_bundle->mirror_out) {
7553 if (ctx->xin->packet != NULL) {
7554 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7555 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
7556 "%s, which is reserved exclusively for mirroring",
7557 ctx->ofproto->up.name, in_bundle->name);
7559 xlate_report(ctx, "input port is mirror output port, dropping");
7564 vid = vlan_tci_to_vid(ctx->xin->flow.vlan_tci);
7565 if (!input_vid_is_valid(vid, in_bundle, ctx->xin->packet != NULL)) {
7566 xlate_report(ctx, "disallowed VLAN VID for this input port, dropping");
7569 vlan = input_vid_to_vlan(in_bundle, vid);
7571 /* Check other admissibility requirements. */
7572 if (in_port && !is_admissible(ctx, in_port, vlan)) {
7576 /* Learn source MAC. */
7577 if (ctx->xin->may_learn) {
7578 update_learning_table(ctx->ofproto, &ctx->xin->flow, vlan, in_bundle);
7581 /* Determine output bundle. */
7582 mac = mac_learning_lookup(ctx->ofproto->ml, ctx->xin->flow.dl_dst, vlan,
7585 if (mac->port.p != in_bundle) {
7586 xlate_report(ctx, "forwarding to learned port");
7587 output_normal(ctx, mac->port.p, vlan);
7589 xlate_report(ctx, "learned port is input port, dropping");
7592 struct ofbundle *bundle;
7594 xlate_report(ctx, "no learned MAC for destination, flooding");
7595 HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
7596 if (bundle != in_bundle
7597 && ofbundle_includes_vlan(bundle, vlan)
7598 && bundle->floodable
7599 && !bundle->mirror_out) {
7600 output_normal(ctx, bundle, vlan);
7603 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
7607 /* Optimized flow revalidation.
7609 * It's a difficult problem, in general, to tell which facets need to have
7610 * their actions recalculated whenever the OpenFlow flow table changes. We
7611 * don't try to solve that general problem: for most kinds of OpenFlow flow
7612 * table changes, we recalculate the actions for every facet. This is
7613 * relatively expensive, but it's good enough if the OpenFlow flow table
7614 * doesn't change very often.
7616 * However, we can expect one particular kind of OpenFlow flow table change to
7617 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
7618 * of CPU on revalidating every facet whenever MAC learning modifies the flow
7619 * table, we add a special case that applies to flow tables in which every rule
7620 * has the same form (that is, the same wildcards), except that the table is
7621 * also allowed to have a single "catch-all" flow that matches all packets. We
7622 * optimize this case by tagging all of the facets that resubmit into the table
7623 * and invalidating the same tag whenever a flow changes in that table. The
7624 * end result is that we revalidate just the facets that need it (and sometimes
7625 * a few more, but not all of the facets or even all of the facets that
7626 * resubmit to the table modified by MAC learning). */
7628 /* Calculates the tag to use for 'flow' and mask 'mask' when it is inserted
7629 * into an OpenFlow table with the given 'basis'. */
7631 rule_calculate_tag(const struct flow *flow, const struct minimask *mask,
7634 if (minimask_is_catchall(mask)) {
7637 uint32_t hash = flow_hash_in_minimask(flow, mask, secret);
7638 return tag_create_deterministic(hash);
7642 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
7643 * taggability of that table.
7645 * This function must be called after *each* change to a flow table. If you
7646 * skip calling it on some changes then the pointer comparisons at the end can
7647 * be invalid if you get unlucky. For example, if a flow removal causes a
7648 * cls_table to be destroyed and then a flow insertion causes a cls_table with
7649 * different wildcards to be created with the same address, then this function
7650 * will incorrectly skip revalidation. */
7652 table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
7654 struct table_dpif *table = &ofproto->tables[table_id];
7655 const struct oftable *oftable = &ofproto->up.tables[table_id];
7656 struct cls_table *catchall, *other;
7657 struct cls_table *t;
7659 catchall = other = NULL;
7661 switch (hmap_count(&oftable->cls.tables)) {
7663 /* We could tag this OpenFlow table but it would make the logic a
7664 * little harder and it's a corner case that doesn't seem worth it
7670 HMAP_FOR_EACH (t, hmap_node, &oftable->cls.tables) {
7671 if (cls_table_is_catchall(t)) {
7673 } else if (!other) {
7676 /* Indicate that we can't tag this by setting both tables to
7677 * NULL. (We know that 'catchall' is already NULL.) */
7684 /* Can't tag this table. */
7688 if (table->catchall_table != catchall || table->other_table != other) {
7689 table->catchall_table = catchall;
7690 table->other_table = other;
7691 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
7695 /* Given 'rule' that has changed in some way (either it is a rule being
7696 * inserted, a rule being deleted, or a rule whose actions are being
7697 * modified), marks facets for revalidation to ensure that packets will be
7698 * forwarded correctly according to the new state of the flow table.
7700 * This function must be called after *each* change to a flow table. See
7701 * the comment on table_update_taggable() for more information. */
7703 rule_invalidate(const struct rule_dpif *rule)
7705 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
7707 table_update_taggable(ofproto, rule->up.table_id);
7709 if (!ofproto->backer->need_revalidate) {
7710 struct table_dpif *table = &ofproto->tables[rule->up.table_id];
7712 if (table->other_table && rule->tag) {
7713 tag_set_add(&ofproto->backer->revalidate_set, rule->tag);
7715 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
7721 set_frag_handling(struct ofproto *ofproto_,
7722 enum ofp_config_flags frag_handling)
7724 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7725 if (frag_handling != OFPC_FRAG_REASM) {
7726 ofproto->backer->need_revalidate = REV_RECONFIGURE;
7734 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
7735 const struct flow *flow,
7736 const struct ofpact *ofpacts, size_t ofpacts_len)
7738 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7739 struct initial_vals initial_vals;
7740 struct odputil_keybuf keybuf;
7741 struct dpif_flow_stats stats;
7742 struct xlate_out xout;
7743 struct xlate_in xin;
7747 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
7748 odp_flow_key_from_flow(&key, flow,
7749 ofp_port_to_odp_port(ofproto, flow->in_port));
7751 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
7753 initial_vals.vlan_tci = flow->vlan_tci;
7754 xlate_in_init(&xin, ofproto, flow, &initial_vals, NULL, stats.tcp_flags,
7756 xin.resubmit_stats = &stats;
7757 xin.ofpacts_len = ofpacts_len;
7758 xin.ofpacts = ofpacts;
7760 xlate_actions(&xin, &xout);
7761 dpif_execute(ofproto->backer->dpif, key.data, key.size,
7762 xout.odp_actions.data, xout.odp_actions.size, packet);
7763 xlate_out_uninit(&xout);
7771 set_netflow(struct ofproto *ofproto_,
7772 const struct netflow_options *netflow_options)
7774 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7776 if (netflow_options) {
7777 if (!ofproto->netflow) {
7778 ofproto->netflow = netflow_create();
7780 return netflow_set_options(ofproto->netflow, netflow_options);
7782 netflow_destroy(ofproto->netflow);
7783 ofproto->netflow = NULL;
7789 get_netflow_ids(const struct ofproto *ofproto_,
7790 uint8_t *engine_type, uint8_t *engine_id)
7792 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7794 dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
7798 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
7800 if (!facet_is_controller_flow(facet) &&
7801 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
7802 struct subfacet *subfacet;
7803 struct ofexpired expired;
7805 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
7806 if (subfacet->path == SF_FAST_PATH) {
7807 struct dpif_flow_stats stats;
7809 subfacet_install(subfacet, &facet->xout.odp_actions, &stats);
7810 subfacet_update_stats(subfacet, &stats);
7814 expired.flow = facet->flow;
7815 expired.packet_count = facet->packet_count;
7816 expired.byte_count = facet->byte_count;
7817 expired.used = facet->used;
7818 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
7823 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
7825 struct facet *facet;
7827 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
7828 send_active_timeout(ofproto, facet);
7832 static struct ofproto_dpif *
7833 ofproto_dpif_lookup(const char *name)
7835 struct ofproto_dpif *ofproto;
7837 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
7838 hash_string(name, 0), &all_ofproto_dpifs) {
7839 if (!strcmp(ofproto->up.name, name)) {
7847 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
7848 const char *argv[], void *aux OVS_UNUSED)
7850 struct ofproto_dpif *ofproto;
7853 ofproto = ofproto_dpif_lookup(argv[1]);
7855 unixctl_command_reply_error(conn, "no such bridge");
7858 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
7860 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
7861 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
7865 unixctl_command_reply(conn, "table successfully flushed");
7869 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
7870 const char *argv[], void *aux OVS_UNUSED)
7872 struct ds ds = DS_EMPTY_INITIALIZER;
7873 const struct ofproto_dpif *ofproto;
7874 const struct mac_entry *e;
7876 ofproto = ofproto_dpif_lookup(argv[1]);
7878 unixctl_command_reply_error(conn, "no such bridge");
7882 ds_put_cstr(&ds, " port VLAN MAC Age\n");
7883 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
7884 struct ofbundle *bundle = e->port.p;
7885 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
7886 ofbundle_get_a_port(bundle)->odp_port,
7887 e->vlan, ETH_ADDR_ARGS(e->mac),
7888 mac_entry_age(ofproto->ml, e));
7890 unixctl_command_reply(conn, ds_cstr(&ds));
7895 struct xlate_out xout;
7896 struct xlate_in xin;
7902 trace_format_rule(struct ds *result, uint8_t table_id, int level,
7903 const struct rule_dpif *rule)
7905 ds_put_char_multiple(result, '\t', level);
7907 ds_put_cstr(result, "No match\n");
7911 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
7912 table_id, ntohll(rule->up.flow_cookie));
7913 cls_rule_format(&rule->up.cr, result);
7914 ds_put_char(result, '\n');
7916 ds_put_char_multiple(result, '\t', level);
7917 ds_put_cstr(result, "OpenFlow ");
7918 ofpacts_format(rule->up.ofpacts, rule->up.ofpacts_len, result);
7919 ds_put_char(result, '\n');
7923 trace_format_flow(struct ds *result, int level, const char *title,
7924 struct trace_ctx *trace)
7926 ds_put_char_multiple(result, '\t', level);
7927 ds_put_format(result, "%s: ", title);
7928 if (flow_equal(&trace->xin.flow, &trace->flow)) {
7929 ds_put_cstr(result, "unchanged");
7931 flow_format(result, &trace->xin.flow);
7932 trace->flow = trace->xin.flow;
7934 ds_put_char(result, '\n');
7938 trace_format_regs(struct ds *result, int level, const char *title,
7939 struct trace_ctx *trace)
7943 ds_put_char_multiple(result, '\t', level);
7944 ds_put_format(result, "%s:", title);
7945 for (i = 0; i < FLOW_N_REGS; i++) {
7946 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
7948 ds_put_char(result, '\n');
7952 trace_format_odp(struct ds *result, int level, const char *title,
7953 struct trace_ctx *trace)
7955 struct ofpbuf *odp_actions = &trace->xout.odp_actions;
7957 ds_put_char_multiple(result, '\t', level);
7958 ds_put_format(result, "%s: ", title);
7959 format_odp_actions(result, odp_actions->data, odp_actions->size);
7960 ds_put_char(result, '\n');
7964 trace_resubmit(struct xlate_ctx *ctx, struct rule_dpif *rule)
7966 struct trace_ctx *trace = CONTAINER_OF(ctx->xin, struct trace_ctx, xin);
7967 struct ds *result = trace->result;
7969 ds_put_char(result, '\n');
7970 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
7971 trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace);
7972 trace_format_odp(result, ctx->recurse + 1, "Resubmitted odp", trace);
7973 trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule);
7977 trace_report(struct xlate_ctx *ctx, const char *s)
7979 struct trace_ctx *trace = CONTAINER_OF(ctx->xin, struct trace_ctx, xin);
7980 struct ds *result = trace->result;
7982 ds_put_char_multiple(result, '\t', ctx->recurse);
7983 ds_put_cstr(result, s);
7984 ds_put_char(result, '\n');
7988 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
7989 void *aux OVS_UNUSED)
7991 const struct dpif_backer *backer;
7992 struct ofproto_dpif *ofproto;
7993 struct ofpbuf odp_key;
7994 struct ofpbuf *packet;
7995 struct initial_vals initial_vals;
8003 ofpbuf_init(&odp_key, 0);
8005 /* Handle "-generate" or a hex string as the last argument. */
8006 if (!strcmp(argv[argc - 1], "-generate")) {
8007 packet = ofpbuf_new(0);
8010 const char *error = eth_from_hex(argv[argc - 1], &packet);
8013 } else if (argc == 4) {
8014 /* The 3-argument form must end in "-generate' or a hex string. */
8015 unixctl_command_reply_error(conn, error);
8020 /* Parse the flow and determine whether a datapath or
8021 * bridge is specified. If function odp_flow_key_from_string()
8022 * returns 0, the flow is a odp_flow. If function
8023 * parse_ofp_exact_flow() returns 0, the flow is a br_flow. */
8024 if (!odp_flow_key_from_string(argv[argc - 1], NULL, &odp_key)) {
8025 /* If the odp_flow is the second argument,
8026 * the datapath name is the first argument. */
8028 const char *dp_type;
8029 if (!strncmp(argv[1], "ovs-", 4)) {
8030 dp_type = argv[1] + 4;
8034 backer = shash_find_data(&all_dpif_backers, dp_type);
8036 unixctl_command_reply_error(conn, "Cannot find datapath "
8041 /* No datapath name specified, so there should be only one
8043 struct shash_node *node;
8044 if (shash_count(&all_dpif_backers) != 1) {
8045 unixctl_command_reply_error(conn, "Must specify datapath "
8046 "name, there is more than one type of datapath");
8049 node = shash_first(&all_dpif_backers);
8050 backer = node->data;
8053 /* Extract the ofproto_dpif object from the ofproto_receive()
8055 if (ofproto_receive(backer, NULL, odp_key.data,
8056 odp_key.size, &flow, NULL, &ofproto, NULL,
8058 unixctl_command_reply_error(conn, "Invalid datapath flow");
8061 ds_put_format(&result, "Bridge: %s\n", ofproto->up.name);
8062 } else if (!parse_ofp_exact_flow(&flow, argv[argc - 1])) {
8064 unixctl_command_reply_error(conn, "Must specify bridge name");
8068 ofproto = ofproto_dpif_lookup(argv[1]);
8070 unixctl_command_reply_error(conn, "Unknown bridge name");
8073 initial_vals.vlan_tci = flow.vlan_tci;
8075 unixctl_command_reply_error(conn, "Bad flow syntax");
8079 /* Generate a packet, if requested. */
8081 if (!packet->size) {
8082 flow_compose(packet, &flow);
8084 ds_put_cstr(&result, "Packet: ");
8085 s = ofp_packet_to_string(packet->data, packet->size);
8086 ds_put_cstr(&result, s);
8089 /* Use the metadata from the flow and the packet argument
8090 * to reconstruct the flow. */
8091 flow_extract(packet, flow.skb_priority, flow.skb_mark, NULL,
8092 flow.in_port, &flow);
8093 initial_vals.vlan_tci = flow.vlan_tci;
8097 ofproto_trace(ofproto, &flow, packet, &initial_vals, &result);
8098 unixctl_command_reply(conn, ds_cstr(&result));
8101 ds_destroy(&result);
8102 ofpbuf_delete(packet);
8103 ofpbuf_uninit(&odp_key);
8107 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
8108 const struct ofpbuf *packet,
8109 const struct initial_vals *initial_vals, struct ds *ds)
8111 struct rule_dpif *rule;
8113 ds_put_cstr(ds, "Flow: ");
8114 flow_format(ds, flow);
8115 ds_put_char(ds, '\n');
8117 rule = rule_dpif_lookup(ofproto, flow);
8119 trace_format_rule(ds, 0, 0, rule);
8120 if (rule == ofproto->miss_rule) {
8121 ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
8122 } else if (rule == ofproto->no_packet_in_rule) {
8123 ds_put_cstr(ds, "\nNo match, packets dropped because "
8124 "OFPPC_NO_PACKET_IN is set on in_port.\n");
8128 uint64_t odp_actions_stub[1024 / 8];
8129 struct ofpbuf odp_actions;
8131 struct trace_ctx trace;
8134 tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
8137 ofpbuf_use_stub(&odp_actions,
8138 odp_actions_stub, sizeof odp_actions_stub);
8139 xlate_in_init(&trace.xin, ofproto, flow, initial_vals, rule, tcp_flags,
8141 trace.xin.resubmit_hook = trace_resubmit;
8142 trace.xin.report_hook = trace_report;
8143 xlate_actions(&trace.xin, &trace.xout);
8145 ds_put_char(ds, '\n');
8146 trace_format_flow(ds, 0, "Final flow", &trace);
8147 ds_put_cstr(ds, "Datapath actions: ");
8148 format_odp_actions(ds, trace.xout.odp_actions.data,
8149 trace.xout.odp_actions.size);
8151 if (trace.xout.slow) {
8152 ds_put_cstr(ds, "\nThis flow is handled by the userspace "
8153 "slow path because it:");
8154 switch (trace.xout.slow) {
8156 ds_put_cstr(ds, "\n\t- Consists of CFM packets.");
8159 ds_put_cstr(ds, "\n\t- Consists of LACP packets.");
8162 ds_put_cstr(ds, "\n\t- Consists of STP packets.");
8165 ds_put_cstr(ds, "\n\t- Consists of BFD packets.");
8167 case SLOW_CONTROLLER:
8168 ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages "
8169 "to the OpenFlow controller.");
8176 xlate_out_uninit(&trace.xout);
8181 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
8182 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
8185 unixctl_command_reply(conn, NULL);
8189 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
8190 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
8193 unixctl_command_reply(conn, NULL);
8196 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
8197 * 'reply' describing the results. */
8199 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
8201 struct facet *facet;
8205 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
8206 if (!facet_check_consistency(facet)) {
8211 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
8215 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
8216 ofproto->up.name, errors);
8218 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
8223 ofproto_dpif_self_check(struct unixctl_conn *conn,
8224 int argc, const char *argv[], void *aux OVS_UNUSED)
8226 struct ds reply = DS_EMPTY_INITIALIZER;
8227 struct ofproto_dpif *ofproto;
8230 ofproto = ofproto_dpif_lookup(argv[1]);
8232 unixctl_command_reply_error(conn, "Unknown ofproto (use "
8233 "ofproto/list for help)");
8236 ofproto_dpif_self_check__(ofproto, &reply);
8238 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
8239 ofproto_dpif_self_check__(ofproto, &reply);
8243 unixctl_command_reply(conn, ds_cstr(&reply));
8247 /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list
8248 * of the 'ofproto_shash' nodes. It is the responsibility of the caller
8249 * to destroy 'ofproto_shash' and free the returned value. */
8250 static const struct shash_node **
8251 get_ofprotos(struct shash *ofproto_shash)
8253 const struct ofproto_dpif *ofproto;
8255 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
8256 char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
8257 shash_add_nocopy(ofproto_shash, name, ofproto);
8260 return shash_sort(ofproto_shash);
8264 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
8265 const char *argv[] OVS_UNUSED,
8266 void *aux OVS_UNUSED)
8268 struct ds ds = DS_EMPTY_INITIALIZER;
8269 struct shash ofproto_shash;
8270 const struct shash_node **sorted_ofprotos;
8273 shash_init(&ofproto_shash);
8274 sorted_ofprotos = get_ofprotos(&ofproto_shash);
8275 for (i = 0; i < shash_count(&ofproto_shash); i++) {
8276 const struct shash_node *node = sorted_ofprotos[i];
8277 ds_put_format(&ds, "%s\n", node->name);
8280 shash_destroy(&ofproto_shash);
8281 free(sorted_ofprotos);
8283 unixctl_command_reply(conn, ds_cstr(&ds));
8288 show_dp_format(const struct ofproto_dpif *ofproto, struct ds *ds)
8290 const struct shash_node **ports;
8292 struct avg_subfacet_rates lifetime;
8293 unsigned long long int minutes;
8294 const int min_ms = 60 * 1000; /* milliseconds in one minute. */
8296 minutes = (time_msec() - ofproto->created) / min_ms;
8299 lifetime.add_rate = (double)ofproto->total_subfacet_add_count
8301 lifetime.del_rate = (double)ofproto->total_subfacet_del_count
8304 lifetime.add_rate = 0.0;
8305 lifetime.del_rate = 0.0;
8308 ds_put_format(ds, "%s (%s):\n", ofproto->up.name,
8309 dpif_name(ofproto->backer->dpif));
8311 "\tlookups: hit:%"PRIu64" missed:%"PRIu64"\n",
8312 ofproto->n_hit, ofproto->n_missed);
8313 ds_put_format(ds, "\tflows: cur: %zu, avg: %5.3f, max: %d,"
8314 " life span: %llu(ms)\n",
8315 hmap_count(&ofproto->subfacets),
8316 avg_subfacet_count(ofproto),
8317 ofproto->max_n_subfacet,
8318 avg_subfacet_life_span(ofproto));
8319 if (minutes >= 60) {
8320 show_dp_rates(ds, "\t\thourly avg:", &ofproto->hourly);
8322 if (minutes >= 60 * 24) {
8323 show_dp_rates(ds, "\t\tdaily avg:", &ofproto->daily);
8325 show_dp_rates(ds, "\t\toverall avg:", &lifetime);
8327 ports = shash_sort(&ofproto->up.port_by_name);
8328 for (i = 0; i < shash_count(&ofproto->up.port_by_name); i++) {
8329 const struct shash_node *node = ports[i];
8330 struct ofport *ofport = node->data;
8331 const char *name = netdev_get_name(ofport->netdev);
8332 const char *type = netdev_get_type(ofport->netdev);
8335 ds_put_format(ds, "\t%s %u/", name, ofport->ofp_port);
8337 odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
8338 if (odp_port != OVSP_NONE) {
8339 ds_put_format(ds, "%"PRIu32":", odp_port);
8341 ds_put_cstr(ds, "none:");
8344 if (strcmp(type, "system")) {
8345 struct netdev *netdev;
8348 ds_put_format(ds, " (%s", type);
8350 error = netdev_open(name, type, &netdev);
8355 error = netdev_get_config(netdev, &config);
8357 const struct smap_node **nodes;
8360 nodes = smap_sort(&config);
8361 for (i = 0; i < smap_count(&config); i++) {
8362 const struct smap_node *node = nodes[i];
8363 ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
8364 node->key, node->value);
8368 smap_destroy(&config);
8370 netdev_close(netdev);
8372 ds_put_char(ds, ')');
8374 ds_put_char(ds, '\n');
8380 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc,
8381 const char *argv[], void *aux OVS_UNUSED)
8383 struct ds ds = DS_EMPTY_INITIALIZER;
8384 const struct ofproto_dpif *ofproto;
8388 for (i = 1; i < argc; i++) {
8389 ofproto = ofproto_dpif_lookup(argv[i]);
8391 ds_put_format(&ds, "Unknown bridge %s (use dpif/dump-dps "
8392 "for help)", argv[i]);
8393 unixctl_command_reply_error(conn, ds_cstr(&ds));
8396 show_dp_format(ofproto, &ds);
8399 struct shash ofproto_shash;
8400 const struct shash_node **sorted_ofprotos;
8403 shash_init(&ofproto_shash);
8404 sorted_ofprotos = get_ofprotos(&ofproto_shash);
8405 for (i = 0; i < shash_count(&ofproto_shash); i++) {
8406 const struct shash_node *node = sorted_ofprotos[i];
8407 show_dp_format(node->data, &ds);
8410 shash_destroy(&ofproto_shash);
8411 free(sorted_ofprotos);
8414 unixctl_command_reply(conn, ds_cstr(&ds));
8419 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
8420 int argc OVS_UNUSED, const char *argv[],
8421 void *aux OVS_UNUSED)
8423 struct ds ds = DS_EMPTY_INITIALIZER;
8424 const struct ofproto_dpif *ofproto;
8425 struct subfacet *subfacet;
8427 ofproto = ofproto_dpif_lookup(argv[1]);
8429 unixctl_command_reply_error(conn, "no such bridge");
8433 update_stats(ofproto->backer);
8435 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
8436 struct facet *facet = subfacet->facet;
8438 odp_flow_key_format(subfacet->key, subfacet->key_len, &ds);
8440 ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:",
8441 subfacet->dp_packet_count, subfacet->dp_byte_count);
8442 if (subfacet->used) {
8443 ds_put_format(&ds, "%.3fs",
8444 (time_msec() - subfacet->used) / 1000.0);
8446 ds_put_format(&ds, "never");
8448 if (subfacet->facet->tcp_flags) {
8449 ds_put_cstr(&ds, ", flags:");
8450 packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags);
8453 ds_put_cstr(&ds, ", actions:");
8454 if (facet->xout.slow) {
8455 uint64_t slow_path_stub[128 / 8];
8456 const struct nlattr *actions;
8459 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
8460 slow_path_stub, sizeof slow_path_stub,
8461 &actions, &actions_len);
8462 format_odp_actions(&ds, actions, actions_len);
8464 format_odp_actions(&ds, facet->xout.odp_actions.data,
8465 facet->xout.odp_actions.size);
8467 ds_put_char(&ds, '\n');
8470 unixctl_command_reply(conn, ds_cstr(&ds));
8475 ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn,
8476 int argc OVS_UNUSED, const char *argv[],
8477 void *aux OVS_UNUSED)
8479 struct ds ds = DS_EMPTY_INITIALIZER;
8480 struct ofproto_dpif *ofproto;
8482 ofproto = ofproto_dpif_lookup(argv[1]);
8484 unixctl_command_reply_error(conn, "no such bridge");
8488 flush(&ofproto->up);
8490 unixctl_command_reply(conn, ds_cstr(&ds));
8495 ofproto_dpif_unixctl_init(void)
8497 static bool registered;
8503 unixctl_command_register(
8505 "[dp_name]|bridge odp_flow|br_flow [-generate|packet]",
8506 1, 3, ofproto_unixctl_trace, NULL);
8507 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
8508 ofproto_unixctl_fdb_flush, NULL);
8509 unixctl_command_register("fdb/show", "bridge", 1, 1,
8510 ofproto_unixctl_fdb_show, NULL);
8511 unixctl_command_register("ofproto/clog", "", 0, 0,
8512 ofproto_dpif_clog, NULL);
8513 unixctl_command_register("ofproto/unclog", "", 0, 0,
8514 ofproto_dpif_unclog, NULL);
8515 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
8516 ofproto_dpif_self_check, NULL);
8517 unixctl_command_register("dpif/dump-dps", "", 0, 0,
8518 ofproto_unixctl_dpif_dump_dps, NULL);
8519 unixctl_command_register("dpif/show", "[bridge]", 0, INT_MAX,
8520 ofproto_unixctl_dpif_show, NULL);
8521 unixctl_command_register("dpif/dump-flows", "bridge", 1, 1,
8522 ofproto_unixctl_dpif_dump_flows, NULL);
8523 unixctl_command_register("dpif/del-flows", "bridge", 1, 1,
8524 ofproto_unixctl_dpif_del_flows, NULL);
8527 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
8529 * This is deprecated. It is only for compatibility with broken device drivers
8530 * in old versions of Linux that do not properly support VLANs when VLAN
8531 * devices are not used. When broken device drivers are no longer in
8532 * widespread use, we will delete these interfaces. */
8535 set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid)
8537 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
8538 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
8540 if (realdev_ofp_port == ofport->realdev_ofp_port
8541 && vid == ofport->vlandev_vid) {
8545 ofproto->backer->need_revalidate = REV_RECONFIGURE;
8547 if (ofport->realdev_ofp_port) {
8550 if (realdev_ofp_port && ofport->bundle) {
8551 /* vlandevs are enslaved to their realdevs, so they are not allowed to
8552 * themselves be part of a bundle. */
8553 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
8556 ofport->realdev_ofp_port = realdev_ofp_port;
8557 ofport->vlandev_vid = vid;
8559 if (realdev_ofp_port) {
8560 vsp_add(ofport, realdev_ofp_port, vid);
8567 hash_realdev_vid(uint16_t realdev_ofp_port, int vid)
8569 return hash_2words(realdev_ofp_port, vid);
8572 /* Returns the ODP port number of the Linux VLAN device that corresponds to
8573 * 'vlan_tci' on the network device with port number 'realdev_odp_port' in
8574 * 'ofproto'. For example, given 'realdev_odp_port' of eth0 and 'vlan_tci' 9,
8575 * it would return the port number of eth0.9.
8577 * Unless VLAN splinters are enabled for port 'realdev_odp_port', this
8578 * function just returns its 'realdev_odp_port' argument. */
8580 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
8581 uint32_t realdev_odp_port, ovs_be16 vlan_tci)
8583 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
8584 uint16_t realdev_ofp_port;
8585 int vid = vlan_tci_to_vid(vlan_tci);
8586 const struct vlan_splinter *vsp;
8588 realdev_ofp_port = odp_port_to_ofp_port(ofproto, realdev_odp_port);
8589 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
8590 hash_realdev_vid(realdev_ofp_port, vid),
8591 &ofproto->realdev_vid_map) {
8592 if (vsp->realdev_ofp_port == realdev_ofp_port
8593 && vsp->vid == vid) {
8594 return ofp_port_to_odp_port(ofproto, vsp->vlandev_ofp_port);
8598 return realdev_odp_port;
8601 static struct vlan_splinter *
8602 vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port)
8604 struct vlan_splinter *vsp;
8606 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, hash_int(vlandev_ofp_port, 0),
8607 &ofproto->vlandev_map) {
8608 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
8616 /* Returns the OpenFlow port number of the "real" device underlying the Linux
8617 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
8618 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
8619 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
8620 * eth0 and store 9 in '*vid'.
8622 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
8623 * VLAN device. Unless VLAN splinters are enabled, this is what this function
8626 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
8627 uint16_t vlandev_ofp_port, int *vid)
8629 if (!hmap_is_empty(&ofproto->vlandev_map)) {
8630 const struct vlan_splinter *vsp;
8632 vsp = vlandev_find(ofproto, vlandev_ofp_port);
8637 return vsp->realdev_ofp_port;
8643 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
8644 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
8645 * 'flow->in_port' to the "real" device backing the VLAN device, sets
8646 * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is
8647 * always the case unless VLAN splinters are enabled), returns false without
8648 * making any changes. */
8650 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
8655 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port, &vid);
8660 /* Cause the flow to be processed as if it came in on the real device with
8661 * the VLAN device's VLAN ID. */
8662 flow->in_port = realdev;
8663 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
8668 vsp_remove(struct ofport_dpif *port)
8670 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
8671 struct vlan_splinter *vsp;
8673 vsp = vlandev_find(ofproto, port->up.ofp_port);
8675 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
8676 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
8679 port->realdev_ofp_port = 0;
8681 VLOG_ERR("missing vlan device record");
8686 vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid)
8688 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
8690 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
8691 && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid))
8692 == realdev_ofp_port)) {
8693 struct vlan_splinter *vsp;
8695 vsp = xmalloc(sizeof *vsp);
8696 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
8697 hash_int(port->up.ofp_port, 0));
8698 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
8699 hash_realdev_vid(realdev_ofp_port, vid));
8700 vsp->realdev_ofp_port = realdev_ofp_port;
8701 vsp->vlandev_ofp_port = port->up.ofp_port;
8704 port->realdev_ofp_port = realdev_ofp_port;
8706 VLOG_ERR("duplicate vlan device record");
8711 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
8713 const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
8714 return ofport ? ofport->odp_port : OVSP_NONE;
8717 static struct ofport_dpif *
8718 odp_port_to_ofport(const struct dpif_backer *backer, uint32_t odp_port)
8720 struct ofport_dpif *port;
8722 HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node,
8723 hash_int(odp_port, 0),
8724 &backer->odp_to_ofport_map) {
8725 if (port->odp_port == odp_port) {
8734 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
8736 struct ofport_dpif *port;
8738 port = odp_port_to_ofport(ofproto->backer, odp_port);
8739 if (port && &ofproto->up == port->up.ofproto) {
8740 return port->up.ofp_port;
8745 static unsigned long long int
8746 avg_subfacet_life_span(const struct ofproto_dpif *ofproto)
8748 unsigned long long int dc;
8749 unsigned long long int avg;
8751 dc = ofproto->total_subfacet_del_count + ofproto->subfacet_del_count;
8752 avg = dc ? ofproto->total_subfacet_life_span / dc : 0;
8758 avg_subfacet_count(const struct ofproto_dpif *ofproto)
8762 if (ofproto->n_update_stats) {
8763 avg_c = (double)ofproto->total_subfacet_count
8764 / ofproto->n_update_stats;
8771 show_dp_rates(struct ds *ds, const char *heading,
8772 const struct avg_subfacet_rates *rates)
8774 ds_put_format(ds, "%s add rate: %5.3f/min, del rate: %5.3f/min\n",
8775 heading, rates->add_rate, rates->del_rate);
8779 update_max_subfacet_count(struct ofproto_dpif *ofproto)
8781 ofproto->max_n_subfacet = MAX(ofproto->max_n_subfacet,
8782 hmap_count(&ofproto->subfacets));
8785 /* Compute exponentially weighted moving average, adding 'new' as the newest,
8786 * most heavily weighted element. 'base' designates the rate of decay: after
8787 * 'base' further updates, 'new''s weight in the EWMA decays to about 1/e
8790 exp_mavg(double *avg, int base, double new)
8792 *avg = (*avg * (base - 1) + new) / base;
8796 update_moving_averages(struct ofproto_dpif *ofproto)
8798 const int min_ms = 60 * 1000; /* milliseconds in one minute. */
8800 /* Update hourly averages on the minute boundaries. */
8801 if (time_msec() - ofproto->last_minute >= min_ms) {
8802 exp_mavg(&ofproto->hourly.add_rate, 60, ofproto->subfacet_add_count);
8803 exp_mavg(&ofproto->hourly.del_rate, 60, ofproto->subfacet_del_count);
8805 /* Update daily averages on the hour boundaries. */
8806 if ((ofproto->last_minute - ofproto->created) / min_ms % 60 == 59) {
8807 exp_mavg(&ofproto->daily.add_rate, 24, ofproto->hourly.add_rate);
8808 exp_mavg(&ofproto->daily.del_rate, 24, ofproto->hourly.del_rate);
8811 ofproto->total_subfacet_add_count += ofproto->subfacet_add_count;
8812 ofproto->total_subfacet_del_count += ofproto->subfacet_del_count;
8813 ofproto->subfacet_add_count = 0;
8814 ofproto->subfacet_del_count = 0;
8815 ofproto->last_minute += min_ms;
8820 dpif_stats_update_hit_count(struct ofproto_dpif *ofproto, uint64_t delta)
8822 ofproto->n_hit += delta;
8825 const struct ofproto_class ofproto_dpif_class = {
8860 port_is_lacp_current,
8861 NULL, /* rule_choose_table */
8868 rule_modify_actions,
8882 get_stp_port_status,
8889 is_mirror_output_bundle,
8890 forward_bpdu_changed,
8891 set_mac_table_config,