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_credit_stats(struct rule_dpif *,
123 const struct dpif_flow_stats *);
124 static void flow_push_stats(struct facet *, const struct dpif_flow_stats *);
125 static tag_type rule_calculate_tag(const struct flow *,
126 const struct minimask *, uint32_t basis);
127 static void rule_invalidate(const struct rule_dpif *);
129 #define MAX_MIRRORS 32
130 typedef uint32_t mirror_mask_t;
131 #define MIRROR_MASK_C(X) UINT32_C(X)
132 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
134 struct ofproto_dpif *ofproto; /* Owning ofproto. */
135 size_t idx; /* In ofproto's "mirrors" array. */
136 void *aux; /* Key supplied by ofproto's client. */
137 char *name; /* Identifier for log messages. */
139 /* Selection criteria. */
140 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
141 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
142 unsigned long *vlans; /* Bitmap of chosen VLANs, NULL selects all. */
144 /* Output (exactly one of out == NULL and out_vlan == -1 is true). */
145 struct ofbundle *out; /* Output port or NULL. */
146 int out_vlan; /* Output VLAN or -1. */
147 mirror_mask_t dup_mirrors; /* Bitmap of mirrors with the same output. */
150 int64_t packet_count; /* Number of packets sent. */
151 int64_t byte_count; /* Number of bytes sent. */
154 static void mirror_destroy(struct ofmirror *);
155 static void update_mirror_stats(struct ofproto_dpif *ofproto,
156 mirror_mask_t mirrors,
157 uint64_t packets, uint64_t bytes);
160 struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
161 struct ofproto_dpif *ofproto; /* Owning ofproto. */
162 void *aux; /* Key supplied by ofproto's client. */
163 char *name; /* Identifier for log messages. */
166 struct list ports; /* Contains "struct ofport"s. */
167 enum port_vlan_mode vlan_mode; /* VLAN mode */
168 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
169 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
170 * NULL if all VLANs are trunked. */
171 struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */
172 struct bond *bond; /* Nonnull iff more than one port. */
173 bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */
176 bool floodable; /* True if no port has OFPUTIL_PC_NO_FLOOD set. */
178 /* Port mirroring info. */
179 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
180 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
181 mirror_mask_t mirror_out; /* Mirrors that output to this bundle. */
184 static void bundle_remove(struct ofport *);
185 static void bundle_update(struct ofbundle *);
186 static void bundle_destroy(struct ofbundle *);
187 static void bundle_del_port(struct ofport_dpif *);
188 static void bundle_run(struct ofbundle *);
189 static void bundle_wait(struct ofbundle *);
190 static struct ofbundle *lookup_input_bundle(const struct ofproto_dpif *,
191 uint16_t in_port, bool warn,
192 struct ofport_dpif **in_ofportp);
194 /* A controller may use OFPP_NONE as the ingress port to indicate that
195 * it did not arrive on a "real" port. 'ofpp_none_bundle' exists for
196 * when an input bundle is needed for validation (e.g., mirroring or
197 * OFPP_NORMAL processing). It is not connected to an 'ofproto' or have
198 * any 'port' structs, so care must be taken when dealing with it. */
199 static struct ofbundle ofpp_none_bundle = {
201 .vlan_mode = PORT_VLAN_TRUNK
204 static void stp_run(struct ofproto_dpif *ofproto);
205 static void stp_wait(struct ofproto_dpif *ofproto);
206 static int set_stp_port(struct ofport *,
207 const struct ofproto_port_stp_settings *);
209 static bool ofbundle_includes_vlan(const struct ofbundle *, uint16_t vlan);
211 struct action_xlate_ctx {
212 /* action_xlate_ctx_init() initializes these members. */
215 struct ofproto_dpif *ofproto;
217 /* Flow to which the OpenFlow actions apply. xlate_actions() will modify
218 * this flow when actions change header fields. */
221 /* Flow at the last commit. */
222 struct flow base_flow;
224 /* Tunnel IP destination address as received. This is stored separately
225 * as the base_flow.tunnel is cleared on init to reflect the datapath
226 * behavior. Used to make sure not to send tunneled output to ourselves,
227 * which might lead to an infinite loop. This could happen easily
228 * if a tunnel is marked as 'ip_remote=flow', and the flow does not
229 * actually set the tun_dst field. */
230 ovs_be32 orig_tunnel_ip_dst;
232 /* stack for the push and pop actions.
233 * Each stack element is of the type "union mf_subvalue". */
235 union mf_subvalue init_stack[1024 / sizeof(union mf_subvalue)];
237 /* The packet corresponding to 'flow', or a null pointer if we are
238 * revalidating without a packet to refer to. */
239 const struct ofpbuf *packet;
241 /* Should OFPP_NORMAL update the MAC learning table? Should "learn"
242 * actions update the flow table?
244 * We want to update these tables if we are actually processing a packet,
245 * or if we are accounting for packets that the datapath has processed, but
246 * not if we are just revalidating. */
249 /* The rule that we are currently translating, or NULL. */
250 struct rule_dpif *rule;
252 /* Union of the set of TCP flags seen so far in this flow. (Used only by
253 * NXAST_FIN_TIMEOUT. Set to zero to avoid updating updating rules'
257 /* If nonnull, flow translation calls this function just before executing a
258 * resubmit or OFPP_TABLE action. In addition, disables logging of traces
259 * when the recursion depth is exceeded.
261 * 'rule' is the rule being submitted into. It will be null if the
262 * resubmit or OFPP_TABLE action didn't find a matching rule.
264 * This is normally null so the client has to set it manually after
265 * calling action_xlate_ctx_init(). */
266 void (*resubmit_hook)(struct action_xlate_ctx *, struct rule_dpif *rule);
268 /* If nonnull, flow translation calls this function to report some
269 * significant decision, e.g. to explain why OFPP_NORMAL translation
270 * dropped a packet. */
271 void (*report_hook)(struct action_xlate_ctx *, const char *s);
273 /* If nonnull, flow translation credits the specified statistics to each
274 * rule reached through a resubmit or OFPP_TABLE action.
276 * This is normally null so the client has to set it manually after
277 * calling action_xlate_ctx_init(). */
278 const struct dpif_flow_stats *resubmit_stats;
280 /* xlate_actions() initializes and uses these members. The client might want
281 * to look at them after it returns. */
283 struct ofpbuf *odp_actions; /* Datapath actions. */
284 tag_type tags; /* Tags associated with actions. */
285 enum slow_path_reason slow; /* 0 if fast path may be used. */
286 bool has_learn; /* Actions include NXAST_LEARN? */
287 bool has_normal; /* Actions output to OFPP_NORMAL? */
288 bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */
289 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
290 mirror_mask_t mirrors; /* Bitmap of associated mirrors. */
292 /* xlate_actions() initializes and uses these members, but the client has no
293 * reason to look at them. */
295 int recurse; /* Recursion level, via xlate_table_action. */
296 bool max_resubmit_trigger; /* Recursed too deeply during translation. */
297 uint32_t orig_skb_priority; /* Priority when packet arrived. */
298 uint8_t table_id; /* OpenFlow table ID where flow was found. */
299 uint32_t sflow_n_outputs; /* Number of output ports. */
300 uint32_t sflow_odp_port; /* Output port for composing sFlow action. */
301 uint16_t user_cookie_offset;/* Used for user_action_cookie fixup. */
302 bool exit; /* No further actions should be processed. */
305 /* Initial values of fields of the packet that may be changed during
306 * flow processing and needed later. */
307 struct initial_vals {
308 /* This is the value of vlan_tci in the packet as actually received from
309 * dpif. This is the same as the facet's flow.vlan_tci unless the packet
310 * was received via a VLAN splinter. In that case, this value is 0
311 * (because the packet as actually received from the dpif had no 802.1Q
312 * tag) but the facet's flow.vlan_tci is set to the VLAN that the splinter
315 * This member should be removed when the VLAN splinters feature is no
320 static void action_xlate_ctx_init(struct action_xlate_ctx *,
321 struct ofproto_dpif *, const struct flow *,
322 const struct initial_vals *initial_vals,
324 uint8_t tcp_flags, const struct ofpbuf *);
325 static void xlate_actions(struct action_xlate_ctx *,
326 const struct ofpact *ofpacts, size_t ofpacts_len,
327 struct ofpbuf *odp_actions);
328 static void xlate_actions_for_side_effects(struct action_xlate_ctx *,
329 const struct ofpact *ofpacts,
331 static void xlate_table_action(struct action_xlate_ctx *, uint16_t in_port,
332 uint8_t table_id, bool may_packet_in);
334 static size_t put_userspace_action(const struct ofproto_dpif *,
335 struct ofpbuf *odp_actions,
337 const union user_action_cookie *,
340 static void compose_slow_path(const struct ofproto_dpif *, const struct flow *,
341 enum slow_path_reason,
342 uint64_t *stub, size_t stub_size,
343 const struct nlattr **actionsp,
344 size_t *actions_lenp);
346 static void xlate_report(struct action_xlate_ctx *ctx, const char *s);
348 /* A subfacet (see "struct subfacet" below) has three possible installation
351 * - SF_NOT_INSTALLED: Not installed in the datapath. This will only be the
352 * case just after the subfacet is created, just before the subfacet is
353 * destroyed, or if the datapath returns an error when we try to install a
356 * - SF_FAST_PATH: The subfacet's actions are installed in the datapath.
358 * - SF_SLOW_PATH: An action that sends every packet for the subfacet through
359 * ofproto_dpif is installed in the datapath.
362 SF_NOT_INSTALLED, /* No datapath flow for this subfacet. */
363 SF_FAST_PATH, /* Full actions are installed. */
364 SF_SLOW_PATH, /* Send-to-userspace action is installed. */
367 static const char *subfacet_path_to_string(enum subfacet_path);
369 /* A dpif flow and actions associated with a facet.
371 * See also the large comment on struct facet. */
374 struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */
375 struct list list_node; /* In struct facet's 'facets' list. */
376 struct facet *facet; /* Owning facet. */
378 enum odp_key_fitness key_fitness;
382 long long int used; /* Time last used; time created if not used. */
383 long long int created; /* Time created. */
385 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
386 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
390 * These should be essentially identical for every subfacet in a facet, but
391 * may differ in trivial ways due to VLAN splinters. */
392 size_t actions_len; /* Number of bytes in actions[]. */
393 struct nlattr *actions; /* Datapath actions. */
395 enum slow_path_reason slow; /* 0 if fast path may be used. */
396 enum subfacet_path path; /* Installed in datapath? */
398 /* Initial values of the packet that may be needed later. */
399 struct initial_vals initial_vals;
401 /* Datapath port the packet arrived on. This is needed to remove
402 * flows for ports that are no longer part of the bridge. Since the
403 * flow definition only has the OpenFlow port number and the port is
404 * no longer part of the bridge, we can't determine the datapath port
405 * number needed to delete the flow from the datapath. */
406 uint32_t odp_in_port;
409 #define SUBFACET_DESTROY_MAX_BATCH 50
411 static struct subfacet *subfacet_create(struct facet *, struct flow_miss *miss,
413 static struct subfacet *subfacet_find(struct ofproto_dpif *,
414 const struct nlattr *key, size_t key_len,
416 static void subfacet_destroy(struct subfacet *);
417 static void subfacet_destroy__(struct subfacet *);
418 static void subfacet_destroy_batch(struct ofproto_dpif *,
419 struct subfacet **, int n);
420 static void subfacet_reset_dp_stats(struct subfacet *,
421 struct dpif_flow_stats *);
422 static void subfacet_update_time(struct subfacet *, long long int used);
423 static void subfacet_update_stats(struct subfacet *,
424 const struct dpif_flow_stats *);
425 static void subfacet_make_actions(struct subfacet *,
426 const struct ofpbuf *packet);
427 static int subfacet_install(struct subfacet *,
428 const struct nlattr *actions, size_t actions_len,
429 struct dpif_flow_stats *, enum slow_path_reason);
430 static void subfacet_uninstall(struct subfacet *);
432 static enum subfacet_path subfacet_want_path(enum slow_path_reason);
434 /* An exact-match instantiation of an OpenFlow flow.
436 * A facet associates a "struct flow", which represents the Open vSwitch
437 * userspace idea of an exact-match flow, with one or more subfacets. Each
438 * subfacet tracks the datapath's idea of the exact-match flow equivalent to
439 * the facet. When the kernel module (or other dpif implementation) and Open
440 * vSwitch userspace agree on the definition of a flow key, there is exactly
441 * one subfacet per facet. If the dpif implementation supports more-specific
442 * flow matching than userspace, however, a facet can have more than one
443 * subfacet, each of which corresponds to some distinction in flow that
444 * userspace simply doesn't understand.
446 * Flow expiration works in terms of subfacets, so a facet must have at least
447 * one subfacet or it will never expire, leaking memory. */
450 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
451 struct list list_node; /* In owning rule's 'facets' list. */
452 struct rule_dpif *rule; /* Owning rule. */
455 struct list subfacets;
456 long long int used; /* Time last used; time created if not used. */
463 * - Do include packets and bytes sent "by hand", e.g. with
466 * - Do include packets and bytes that were obtained from the datapath
467 * when a subfacet's statistics were reset (e.g. dpif_flow_put() with
468 * DPIF_FP_ZERO_STATS).
470 * - Do not include packets or bytes that can be obtained from the
471 * datapath for any existing subfacet.
473 uint64_t packet_count; /* Number of packets received. */
474 uint64_t byte_count; /* Number of bytes received. */
476 /* Resubmit statistics. */
477 uint64_t prev_packet_count; /* Number of packets from last stats push. */
478 uint64_t prev_byte_count; /* Number of bytes from last stats push. */
479 long long int prev_used; /* Used time from last stats push. */
482 uint64_t accounted_bytes; /* Bytes processed by facet_account(). */
483 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
484 uint8_t tcp_flags; /* TCP flags seen for this 'rule'. */
486 /* Properties of datapath actions.
488 * Every subfacet has its own actions because actions can differ slightly
489 * between splintered and non-splintered subfacets due to the VLAN tag
490 * being initially different (present vs. absent). All of them have these
491 * properties in common so we just store one copy of them here. */
492 bool has_learn; /* Actions include NXAST_LEARN? */
493 bool has_normal; /* Actions output to OFPP_NORMAL? */
494 bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */
495 tag_type tags; /* Tags that would require revalidation. */
496 mirror_mask_t mirrors; /* Bitmap of dependent mirrors. */
498 /* Storage for a single subfacet, to reduce malloc() time and space
499 * overhead. (A facet always has at least one subfacet and in the common
500 * case has exactly one subfacet. However, 'one_subfacet' may not
501 * always be valid, since it could have been removed after newer
502 * subfacets were pushed onto the 'subfacets' list.) */
503 struct subfacet one_subfacet;
505 long long int learn_rl; /* Rate limiter for facet_learn(). */
508 static struct facet *facet_create(struct rule_dpif *,
509 const struct flow *, uint32_t hash);
510 static void facet_remove(struct facet *);
511 static void facet_free(struct facet *);
513 static struct facet *facet_find(struct ofproto_dpif *,
514 const struct flow *, uint32_t hash);
515 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
516 const struct flow *, uint32_t hash);
517 static bool facet_revalidate(struct facet *);
518 static bool facet_check_consistency(struct facet *);
520 static void facet_flush_stats(struct facet *);
522 static void facet_update_time(struct facet *, long long int used);
523 static void facet_reset_counters(struct facet *);
524 static void facet_push_stats(struct facet *);
525 static void facet_learn(struct facet *);
526 static void facet_account(struct facet *);
527 static void push_all_stats(void);
529 static struct subfacet *facet_get_subfacet(struct facet *);
531 static bool facet_is_controller_flow(struct facet *);
534 struct hmap_node odp_port_node; /* In dpif_backer's "odp_to_ofport_map". */
538 struct ofbundle *bundle; /* Bundle that contains this port, if any. */
539 struct list bundle_node; /* In struct ofbundle's "ports" list. */
540 struct cfm *cfm; /* Connectivity Fault Management, if any. */
541 struct bfd *bfd; /* BFD, if any. */
542 tag_type tag; /* Tag associated with this port. */
543 bool may_enable; /* May be enabled in bonds. */
544 long long int carrier_seq; /* Carrier status changes. */
545 struct tnl_port *tnl_port; /* Tunnel handle, or null. */
548 struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */
549 enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */
550 long long int stp_state_entered;
552 struct hmap priorities; /* Map of attached 'priority_to_dscp's. */
554 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
556 * This is deprecated. It is only for compatibility with broken device
557 * drivers in old versions of Linux that do not properly support VLANs when
558 * VLAN devices are not used. When broken device drivers are no longer in
559 * widespread use, we will delete these interfaces. */
560 uint16_t realdev_ofp_port;
564 /* Node in 'ofport_dpif''s 'priorities' map. Used to maintain a map from
565 * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
566 * traffic egressing the 'ofport' with that priority should be marked with. */
567 struct priority_to_dscp {
568 struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'priorities' map. */
569 uint32_t priority; /* Priority of this queue (see struct flow). */
571 uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */
574 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
576 * This is deprecated. It is only for compatibility with broken device drivers
577 * in old versions of Linux that do not properly support VLANs when VLAN
578 * devices are not used. When broken device drivers are no longer in
579 * widespread use, we will delete these interfaces. */
580 struct vlan_splinter {
581 struct hmap_node realdev_vid_node;
582 struct hmap_node vlandev_node;
583 uint16_t realdev_ofp_port;
584 uint16_t vlandev_ofp_port;
588 static uint32_t vsp_realdev_to_vlandev(const struct ofproto_dpif *,
589 uint32_t realdev, ovs_be16 vlan_tci);
590 static bool vsp_adjust_flow(const struct ofproto_dpif *, struct flow *);
591 static void vsp_remove(struct ofport_dpif *);
592 static void vsp_add(struct ofport_dpif *, uint16_t realdev_ofp_port, int vid);
594 static uint32_t ofp_port_to_odp_port(const struct ofproto_dpif *,
596 static uint16_t odp_port_to_ofp_port(const struct ofproto_dpif *,
599 static struct ofport_dpif *
600 ofport_dpif_cast(const struct ofport *ofport)
602 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
605 static void port_run(struct ofport_dpif *);
606 static void port_run_fast(struct ofport_dpif *);
607 static void port_wait(struct ofport_dpif *);
608 static int set_bfd(struct ofport *, const struct smap *);
609 static int set_cfm(struct ofport *, const struct cfm_settings *);
610 static void ofport_clear_priorities(struct ofport_dpif *);
611 static void run_fast_rl(void);
613 struct dpif_completion {
614 struct list list_node;
615 struct ofoperation *op;
618 /* Extra information about a classifier table.
619 * Currently used just for optimized flow revalidation. */
621 /* If either of these is nonnull, then this table has a form that allows
622 * flows to be tagged to avoid revalidating most flows for the most common
623 * kinds of flow table changes. */
624 struct cls_table *catchall_table; /* Table that wildcards all fields. */
625 struct cls_table *other_table; /* Table with any other wildcard set. */
626 uint32_t basis; /* Keeps each table's tags separate. */
629 /* Reasons that we might need to revalidate every facet, and corresponding
632 * A value of 0 means that there is no need to revalidate.
634 * It would be nice to have some cleaner way to integrate with coverage
635 * counters, but with only a few reasons I guess this is good enough for
637 enum revalidate_reason {
638 REV_RECONFIGURE = 1, /* Switch configuration changed. */
639 REV_STP, /* Spanning tree protocol port status change. */
640 REV_PORT_TOGGLED, /* Port enabled or disabled by CFM, LACP, ...*/
641 REV_FLOW_TABLE, /* Flow table changed. */
642 REV_INCONSISTENCY /* Facet self-check failed. */
644 COVERAGE_DEFINE(rev_reconfigure);
645 COVERAGE_DEFINE(rev_stp);
646 COVERAGE_DEFINE(rev_port_toggled);
647 COVERAGE_DEFINE(rev_flow_table);
648 COVERAGE_DEFINE(rev_inconsistency);
650 /* Drop keys are odp flow keys which have drop flows installed in the kernel.
651 * These are datapath flows which have no associated ofproto, if they did we
652 * would use facets. */
654 struct hmap_node hmap_node;
659 /* All datapaths of a given type share a single dpif backer instance. */
664 struct timer next_expiration;
665 struct hmap odp_to_ofport_map; /* ODP port to ofport mapping. */
667 struct simap tnl_backers; /* Set of dpif ports backing tunnels. */
669 /* Facet revalidation flags applying to facets which use this backer. */
670 enum revalidate_reason need_revalidate; /* Revalidate every facet. */
671 struct tag_set revalidate_set; /* Revalidate only matching facets. */
673 struct hmap drop_keys; /* Set of dropped odp keys. */
676 /* All existing ofproto_backer instances, indexed by ofproto->up.type. */
677 static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers);
679 static void drop_key_clear(struct dpif_backer *);
680 static struct ofport_dpif *
681 odp_port_to_ofport(const struct dpif_backer *, uint32_t odp_port);
683 static void dpif_stats_update_hit_count(struct ofproto_dpif *ofproto,
685 struct avg_subfacet_rates {
686 double add_rate; /* Moving average of new flows created per minute. */
687 double del_rate; /* Moving average of flows deleted per minute. */
689 static void show_dp_rates(struct ds *ds, const char *heading,
690 const struct avg_subfacet_rates *rates);
691 static void exp_mavg(double *avg, int base, double new);
693 struct ofproto_dpif {
694 struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
696 struct dpif_backer *backer;
698 /* Special OpenFlow rules. */
699 struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */
700 struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */
706 struct netflow *netflow;
707 struct dpif_sflow *sflow;
708 struct dpif_ipfix *ipfix;
709 struct hmap bundles; /* Contains "struct ofbundle"s. */
710 struct mac_learning *ml;
711 struct ofmirror *mirrors[MAX_MIRRORS];
713 bool has_bonded_bundles;
717 struct hmap subfacets;
718 struct governor *governor;
719 long long int consistency_rl;
722 struct table_dpif tables[N_TABLES];
724 /* Support for debugging async flow mods. */
725 struct list completions;
727 bool has_bundle_action; /* True when the first bundle action appears. */
728 struct netdev_stats stats; /* To account packets generated and consumed in
733 long long int stp_last_tick;
735 /* VLAN splinters. */
736 struct hmap realdev_vid_map; /* (realdev,vid) -> vlandev. */
737 struct hmap vlandev_map; /* vlandev -> (realdev,vid). */
740 struct sset ports; /* Set of standard port names. */
741 struct sset ghost_ports; /* Ports with no datapath port. */
742 struct sset port_poll_set; /* Queued names for port_poll() reply. */
743 int port_poll_errno; /* Last errno for port_poll() reply. */
745 /* Per ofproto's dpif stats. */
749 /* Subfacet statistics.
751 * These keep track of the total number of subfacets added and deleted and
752 * flow life span. They are useful for computing the flow rates stats
753 * exposed via "ovs-appctl dpif/show". The goal is to learn about
754 * traffic patterns in ways that we can use later to improve Open vSwitch
755 * performance in new situations. */
756 long long int created; /* Time when it is created. */
757 unsigned int max_n_subfacet; /* Maximum number of flows */
759 /* The average number of subfacets... */
760 struct avg_subfacet_rates hourly; /* ...over the last hour. */
761 struct avg_subfacet_rates daily; /* ...over the last day. */
762 long long int last_minute; /* Last time 'hourly' was updated. */
764 /* Number of subfacets added or deleted since 'last_minute'. */
765 unsigned int subfacet_add_count;
766 unsigned int subfacet_del_count;
768 /* Number of subfacets added or deleted from 'created' to 'last_minute.' */
769 unsigned long long int total_subfacet_add_count;
770 unsigned long long int total_subfacet_del_count;
772 /* Sum of the number of milliseconds that each subfacet existed,
773 * over the subfacets that have been added and then later deleted. */
774 unsigned long long int total_subfacet_life_span;
776 /* Incremented by the number of currently existing subfacets, each
777 * time we pull statistics from the kernel. */
778 unsigned long long int total_subfacet_count;
780 /* Number of times we pull statistics from the kernel. */
781 unsigned long long int n_update_stats;
783 static unsigned long long int avg_subfacet_life_span(
784 const struct ofproto_dpif *);
785 static double avg_subfacet_count(const struct ofproto_dpif *ofproto);
786 static void update_moving_averages(struct ofproto_dpif *ofproto);
787 static void dpif_stats_update_hit_count(struct ofproto_dpif *ofproto,
789 static void update_max_subfacet_count(struct ofproto_dpif *ofproto);
791 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
792 * for debugging the asynchronous flow_mod implementation.) */
795 /* All existing ofproto_dpif instances, indexed by ->up.name. */
796 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
798 static void ofproto_dpif_unixctl_init(void);
800 static struct ofproto_dpif *
801 ofproto_dpif_cast(const struct ofproto *ofproto)
803 ovs_assert(ofproto->ofproto_class == &ofproto_dpif_class);
804 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
807 static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *,
809 static struct ofport_dpif *get_odp_port(const struct ofproto_dpif *,
811 static void ofproto_trace(struct ofproto_dpif *, const struct flow *,
812 const struct ofpbuf *,
813 const struct initial_vals *, struct ds *);
815 /* Packet processing. */
816 static void update_learning_table(struct ofproto_dpif *,
817 const struct flow *, int vlan,
820 #define FLOW_MISS_MAX_BATCH 50
821 static int handle_upcalls(struct dpif_backer *, unsigned int max_batch);
823 /* Flow expiration. */
824 static int expire(struct dpif_backer *);
827 static void send_netflow_active_timeouts(struct ofproto_dpif *);
830 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
831 static size_t compose_sflow_action(const struct ofproto_dpif *,
832 struct ofpbuf *odp_actions,
833 const struct flow *, uint32_t odp_port);
834 static void compose_ipfix_action(const struct ofproto_dpif *,
835 struct ofpbuf *odp_actions,
836 const struct flow *);
837 static void add_mirror_actions(struct action_xlate_ctx *ctx,
838 const struct flow *flow);
839 /* Global variables. */
840 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
842 /* Initial mappings of port to bridge mappings. */
843 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
845 /* Factory functions. */
848 init(const struct shash *iface_hints)
850 struct shash_node *node;
852 /* Make a local copy, since we don't own 'iface_hints' elements. */
853 SHASH_FOR_EACH(node, iface_hints) {
854 const struct iface_hint *orig_hint = node->data;
855 struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
857 new_hint->br_name = xstrdup(orig_hint->br_name);
858 new_hint->br_type = xstrdup(orig_hint->br_type);
859 new_hint->ofp_port = orig_hint->ofp_port;
861 shash_add(&init_ofp_ports, node->name, new_hint);
866 enumerate_types(struct sset *types)
868 dp_enumerate_types(types);
872 enumerate_names(const char *type, struct sset *names)
874 struct ofproto_dpif *ofproto;
877 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
878 if (strcmp(type, ofproto->up.type)) {
881 sset_add(names, ofproto->up.name);
888 del(const char *type, const char *name)
893 error = dpif_open(name, type, &dpif);
895 error = dpif_delete(dpif);
902 port_open_type(const char *datapath_type, const char *port_type)
904 return dpif_port_open_type(datapath_type, port_type);
907 /* Type functions. */
909 static struct ofproto_dpif *
910 lookup_ofproto_dpif_by_port_name(const char *name)
912 struct ofproto_dpif *ofproto;
914 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
915 if (sset_contains(&ofproto->ports, name)) {
924 type_run(const char *type)
926 static long long int push_timer = LLONG_MIN;
927 struct dpif_backer *backer;
931 backer = shash_find_data(&all_dpif_backers, type);
933 /* This is not necessarily a problem, since backers are only
934 * created on demand. */
938 dpif_run(backer->dpif);
940 /* The most natural place to push facet statistics is when they're pulled
941 * from the datapath. However, when there are many flows in the datapath,
942 * this expensive operation can occur so frequently, that it reduces our
943 * ability to quickly set up flows. To reduce the cost, we push statistics
945 if (time_msec() > push_timer) {
946 push_timer = time_msec() + 2000;
950 if (backer->need_revalidate
951 || !tag_set_is_empty(&backer->revalidate_set)) {
952 struct tag_set revalidate_set = backer->revalidate_set;
953 bool need_revalidate = backer->need_revalidate;
954 struct ofproto_dpif *ofproto;
955 struct simap_node *node;
956 struct simap tmp_backers;
958 /* Handle tunnel garbage collection. */
959 simap_init(&tmp_backers);
960 simap_swap(&backer->tnl_backers, &tmp_backers);
962 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
963 struct ofport_dpif *iter;
965 if (backer != ofproto->backer) {
969 HMAP_FOR_EACH (iter, up.hmap_node, &ofproto->up.ports) {
972 if (!iter->tnl_port) {
976 dp_port = netdev_vport_get_dpif_port(iter->up.netdev);
977 node = simap_find(&tmp_backers, dp_port);
979 simap_put(&backer->tnl_backers, dp_port, node->data);
980 simap_delete(&tmp_backers, node);
981 node = simap_find(&backer->tnl_backers, dp_port);
983 node = simap_find(&backer->tnl_backers, dp_port);
985 uint32_t odp_port = UINT32_MAX;
987 if (!dpif_port_add(backer->dpif, iter->up.netdev,
989 simap_put(&backer->tnl_backers, dp_port, odp_port);
990 node = simap_find(&backer->tnl_backers, dp_port);
995 iter->odp_port = node ? node->data : OVSP_NONE;
996 if (tnl_port_reconfigure(&iter->up, iter->odp_port,
998 backer->need_revalidate = REV_RECONFIGURE;
1003 SIMAP_FOR_EACH (node, &tmp_backers) {
1004 dpif_port_del(backer->dpif, node->data);
1006 simap_destroy(&tmp_backers);
1008 switch (backer->need_revalidate) {
1009 case REV_RECONFIGURE: COVERAGE_INC(rev_reconfigure); break;
1010 case REV_STP: COVERAGE_INC(rev_stp); break;
1011 case REV_PORT_TOGGLED: COVERAGE_INC(rev_port_toggled); break;
1012 case REV_FLOW_TABLE: COVERAGE_INC(rev_flow_table); break;
1013 case REV_INCONSISTENCY: COVERAGE_INC(rev_inconsistency); break;
1016 if (backer->need_revalidate) {
1017 /* Clear the drop_keys in case we should now be accepting some
1018 * formerly dropped flows. */
1019 drop_key_clear(backer);
1022 /* Clear the revalidation flags. */
1023 tag_set_init(&backer->revalidate_set);
1024 backer->need_revalidate = 0;
1026 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1027 struct facet *facet, *next;
1029 if (ofproto->backer != backer) {
1033 HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &ofproto->facets) {
1035 || tag_set_intersects(&revalidate_set, facet->tags)) {
1036 facet_revalidate(facet);
1043 if (timer_expired(&backer->next_expiration)) {
1044 int delay = expire(backer);
1045 timer_set_duration(&backer->next_expiration, delay);
1048 /* Check for port changes in the dpif. */
1049 while ((error = dpif_port_poll(backer->dpif, &devname)) == 0) {
1050 struct ofproto_dpif *ofproto;
1051 struct dpif_port port;
1053 /* Don't report on the datapath's device. */
1054 if (!strcmp(devname, dpif_base_name(backer->dpif))) {
1058 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
1059 &all_ofproto_dpifs) {
1060 if (simap_contains(&ofproto->backer->tnl_backers, devname)) {
1065 ofproto = lookup_ofproto_dpif_by_port_name(devname);
1066 if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
1067 /* The port was removed. If we know the datapath,
1068 * report it through poll_set(). If we don't, it may be
1069 * notifying us of a removal we initiated, so ignore it.
1070 * If there's a pending ENOBUFS, let it stand, since
1071 * everything will be reevaluated. */
1072 if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
1073 sset_add(&ofproto->port_poll_set, devname);
1074 ofproto->port_poll_errno = 0;
1076 } else if (!ofproto) {
1077 /* The port was added, but we don't know with which
1078 * ofproto we should associate it. Delete it. */
1079 dpif_port_del(backer->dpif, port.port_no);
1081 dpif_port_destroy(&port);
1087 if (error != EAGAIN) {
1088 struct ofproto_dpif *ofproto;
1090 /* There was some sort of error, so propagate it to all
1091 * ofprotos that use this backer. */
1092 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
1093 &all_ofproto_dpifs) {
1094 if (ofproto->backer == backer) {
1095 sset_clear(&ofproto->port_poll_set);
1096 ofproto->port_poll_errno = error;
1105 dpif_backer_run_fast(struct dpif_backer *backer, int max_batch)
1109 /* Handle one or more batches of upcalls, until there's nothing left to do
1110 * or until we do a fixed total amount of work.
1112 * We do work in batches because it can be much cheaper to set up a number
1113 * of flows and fire off their patches all at once. We do multiple batches
1114 * because in some cases handling a packet can cause another packet to be
1115 * queued almost immediately as part of the return flow. Both
1116 * optimizations can make major improvements on some benchmarks and
1117 * presumably for real traffic as well. */
1119 while (work < max_batch) {
1120 int retval = handle_upcalls(backer, max_batch - work);
1131 type_run_fast(const char *type)
1133 struct dpif_backer *backer;
1135 backer = shash_find_data(&all_dpif_backers, type);
1137 /* This is not necessarily a problem, since backers are only
1138 * created on demand. */
1142 return dpif_backer_run_fast(backer, FLOW_MISS_MAX_BATCH);
1148 static long long int port_rl = LLONG_MIN;
1149 static unsigned int backer_rl = 0;
1151 if (time_msec() >= port_rl) {
1152 struct ofproto_dpif *ofproto;
1153 struct ofport_dpif *ofport;
1155 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1157 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1158 port_run_fast(ofport);
1161 port_rl = time_msec() + 200;
1164 /* XXX: We have to be careful not to do too much work in this function. If
1165 * we call dpif_backer_run_fast() too often, or with too large a batch,
1166 * performance improves signifcantly, but at a cost. It's possible for the
1167 * number of flows in the datapath to increase without bound, and for poll
1168 * loops to take 10s of seconds. The correct solution to this problem,
1169 * long term, is to separate flow miss handling into it's own thread so it
1170 * isn't affected by revalidations, and expirations. Until then, this is
1171 * the best we can do. */
1172 if (++backer_rl >= 10) {
1173 struct shash_node *node;
1176 SHASH_FOR_EACH (node, &all_dpif_backers) {
1177 dpif_backer_run_fast(node->data, 1);
1183 type_wait(const char *type)
1185 struct dpif_backer *backer;
1187 backer = shash_find_data(&all_dpif_backers, type);
1189 /* This is not necessarily a problem, since backers are only
1190 * created on demand. */
1194 timer_wait(&backer->next_expiration);
1197 /* Basic life-cycle. */
1199 static int add_internal_flows(struct ofproto_dpif *);
1201 static struct ofproto *
1204 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
1205 return &ofproto->up;
1209 dealloc(struct ofproto *ofproto_)
1211 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1216 close_dpif_backer(struct dpif_backer *backer)
1218 struct shash_node *node;
1220 ovs_assert(backer->refcount > 0);
1222 if (--backer->refcount) {
1226 drop_key_clear(backer);
1227 hmap_destroy(&backer->drop_keys);
1229 simap_destroy(&backer->tnl_backers);
1230 hmap_destroy(&backer->odp_to_ofport_map);
1231 node = shash_find(&all_dpif_backers, backer->type);
1233 shash_delete(&all_dpif_backers, node);
1234 dpif_close(backer->dpif);
1239 /* Datapath port slated for removal from datapath. */
1240 struct odp_garbage {
1241 struct list list_node;
1246 open_dpif_backer(const char *type, struct dpif_backer **backerp)
1248 struct dpif_backer *backer;
1249 struct dpif_port_dump port_dump;
1250 struct dpif_port port;
1251 struct shash_node *node;
1252 struct list garbage_list;
1253 struct odp_garbage *garbage, *next;
1259 backer = shash_find_data(&all_dpif_backers, type);
1266 backer_name = xasprintf("ovs-%s", type);
1268 /* Remove any existing datapaths, since we assume we're the only
1269 * userspace controlling the datapath. */
1271 dp_enumerate_names(type, &names);
1272 SSET_FOR_EACH(name, &names) {
1273 struct dpif *old_dpif;
1275 /* Don't remove our backer if it exists. */
1276 if (!strcmp(name, backer_name)) {
1280 if (dpif_open(name, type, &old_dpif)) {
1281 VLOG_WARN("couldn't open old datapath %s to remove it", name);
1283 dpif_delete(old_dpif);
1284 dpif_close(old_dpif);
1287 sset_destroy(&names);
1289 backer = xmalloc(sizeof *backer);
1291 error = dpif_create_and_open(backer_name, type, &backer->dpif);
1294 VLOG_ERR("failed to open datapath of type %s: %s", type,
1300 backer->type = xstrdup(type);
1301 backer->refcount = 1;
1302 hmap_init(&backer->odp_to_ofport_map);
1303 hmap_init(&backer->drop_keys);
1304 timer_set_duration(&backer->next_expiration, 1000);
1305 backer->need_revalidate = 0;
1306 simap_init(&backer->tnl_backers);
1307 tag_set_init(&backer->revalidate_set);
1310 dpif_flow_flush(backer->dpif);
1312 /* Loop through the ports already on the datapath and remove any
1313 * that we don't need anymore. */
1314 list_init(&garbage_list);
1315 dpif_port_dump_start(&port_dump, backer->dpif);
1316 while (dpif_port_dump_next(&port_dump, &port)) {
1317 node = shash_find(&init_ofp_ports, port.name);
1318 if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
1319 garbage = xmalloc(sizeof *garbage);
1320 garbage->odp_port = port.port_no;
1321 list_push_front(&garbage_list, &garbage->list_node);
1324 dpif_port_dump_done(&port_dump);
1326 LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) {
1327 dpif_port_del(backer->dpif, garbage->odp_port);
1328 list_remove(&garbage->list_node);
1332 shash_add(&all_dpif_backers, type, backer);
1334 error = dpif_recv_set(backer->dpif, true);
1336 VLOG_ERR("failed to listen on datapath of type %s: %s",
1337 type, strerror(error));
1338 close_dpif_backer(backer);
1346 construct(struct ofproto *ofproto_)
1348 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1349 struct shash_node *node, *next;
1354 error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1359 max_ports = dpif_get_max_ports(ofproto->backer->dpif);
1360 ofproto_init_max_ports(ofproto_, MIN(max_ports, OFPP_MAX));
1362 ofproto->n_matches = 0;
1364 ofproto->netflow = NULL;
1365 ofproto->sflow = NULL;
1366 ofproto->ipfix = NULL;
1367 ofproto->stp = NULL;
1368 hmap_init(&ofproto->bundles);
1369 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1370 for (i = 0; i < MAX_MIRRORS; i++) {
1371 ofproto->mirrors[i] = NULL;
1373 ofproto->has_bonded_bundles = false;
1375 hmap_init(&ofproto->facets);
1376 hmap_init(&ofproto->subfacets);
1377 ofproto->governor = NULL;
1378 ofproto->consistency_rl = LLONG_MIN;
1380 for (i = 0; i < N_TABLES; i++) {
1381 struct table_dpif *table = &ofproto->tables[i];
1383 table->catchall_table = NULL;
1384 table->other_table = NULL;
1385 table->basis = random_uint32();
1388 list_init(&ofproto->completions);
1390 ofproto_dpif_unixctl_init();
1392 ofproto->has_mirrors = false;
1393 ofproto->has_bundle_action = false;
1395 hmap_init(&ofproto->vlandev_map);
1396 hmap_init(&ofproto->realdev_vid_map);
1398 sset_init(&ofproto->ports);
1399 sset_init(&ofproto->ghost_ports);
1400 sset_init(&ofproto->port_poll_set);
1401 ofproto->port_poll_errno = 0;
1403 SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1404 struct iface_hint *iface_hint = node->data;
1406 if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1407 /* Check if the datapath already has this port. */
1408 if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1409 sset_add(&ofproto->ports, node->name);
1412 free(iface_hint->br_name);
1413 free(iface_hint->br_type);
1415 shash_delete(&init_ofp_ports, node);
1419 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1420 hash_string(ofproto->up.name, 0));
1421 memset(&ofproto->stats, 0, sizeof ofproto->stats);
1423 ofproto_init_tables(ofproto_, N_TABLES);
1424 error = add_internal_flows(ofproto);
1425 ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1428 ofproto->n_missed = 0;
1430 ofproto->max_n_subfacet = 0;
1431 ofproto->created = time_msec();
1432 ofproto->last_minute = ofproto->created;
1433 memset(&ofproto->hourly, 0, sizeof ofproto->hourly);
1434 memset(&ofproto->daily, 0, sizeof ofproto->daily);
1435 ofproto->subfacet_add_count = 0;
1436 ofproto->subfacet_del_count = 0;
1437 ofproto->total_subfacet_add_count = 0;
1438 ofproto->total_subfacet_del_count = 0;
1439 ofproto->total_subfacet_life_span = 0;
1440 ofproto->total_subfacet_count = 0;
1441 ofproto->n_update_stats = 0;
1447 add_internal_flow(struct ofproto_dpif *ofproto, int id,
1448 const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1450 struct ofputil_flow_mod fm;
1453 match_init_catchall(&fm.match);
1455 match_set_reg(&fm.match, 0, id);
1456 fm.new_cookie = htonll(0);
1457 fm.cookie = htonll(0);
1458 fm.cookie_mask = htonll(0);
1459 fm.table_id = TBL_INTERNAL;
1460 fm.command = OFPFC_ADD;
1461 fm.idle_timeout = 0;
1462 fm.hard_timeout = 0;
1466 fm.ofpacts = ofpacts->data;
1467 fm.ofpacts_len = ofpacts->size;
1469 error = ofproto_flow_mod(&ofproto->up, &fm);
1471 VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
1472 id, ofperr_to_string(error));
1476 *rulep = rule_dpif_lookup__(ofproto, &fm.match.flow, TBL_INTERNAL);
1477 ovs_assert(*rulep != NULL);
1483 add_internal_flows(struct ofproto_dpif *ofproto)
1485 struct ofpact_controller *controller;
1486 uint64_t ofpacts_stub[128 / 8];
1487 struct ofpbuf ofpacts;
1491 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1494 controller = ofpact_put_CONTROLLER(&ofpacts);
1495 controller->max_len = UINT16_MAX;
1496 controller->controller_id = 0;
1497 controller->reason = OFPR_NO_MATCH;
1498 ofpact_pad(&ofpacts);
1500 error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule);
1505 ofpbuf_clear(&ofpacts);
1506 error = add_internal_flow(ofproto, id++, &ofpacts,
1507 &ofproto->no_packet_in_rule);
1512 complete_operations(struct ofproto_dpif *ofproto)
1514 struct dpif_completion *c, *next;
1516 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
1517 ofoperation_complete(c->op, 0);
1518 list_remove(&c->list_node);
1524 destruct(struct ofproto *ofproto_)
1526 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1527 struct rule_dpif *rule, *next_rule;
1528 struct oftable *table;
1531 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1532 complete_operations(ofproto);
1534 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1535 struct cls_cursor cursor;
1537 cls_cursor_init(&cursor, &table->cls, NULL);
1538 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1539 ofproto_rule_destroy(&rule->up);
1543 for (i = 0; i < MAX_MIRRORS; i++) {
1544 mirror_destroy(ofproto->mirrors[i]);
1547 netflow_destroy(ofproto->netflow);
1548 dpif_sflow_destroy(ofproto->sflow);
1549 hmap_destroy(&ofproto->bundles);
1550 mac_learning_destroy(ofproto->ml);
1552 hmap_destroy(&ofproto->facets);
1553 hmap_destroy(&ofproto->subfacets);
1554 governor_destroy(ofproto->governor);
1556 hmap_destroy(&ofproto->vlandev_map);
1557 hmap_destroy(&ofproto->realdev_vid_map);
1559 sset_destroy(&ofproto->ports);
1560 sset_destroy(&ofproto->ghost_ports);
1561 sset_destroy(&ofproto->port_poll_set);
1563 close_dpif_backer(ofproto->backer);
1567 run_fast(struct ofproto *ofproto_)
1569 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1570 struct ofport_dpif *ofport;
1572 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1573 port_run_fast(ofport);
1580 run(struct ofproto *ofproto_)
1582 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1583 struct ofport_dpif *ofport;
1584 struct ofbundle *bundle;
1588 complete_operations(ofproto);
1591 error = run_fast(ofproto_);
1596 if (ofproto->netflow) {
1597 if (netflow_run(ofproto->netflow)) {
1598 send_netflow_active_timeouts(ofproto);
1601 if (ofproto->sflow) {
1602 dpif_sflow_run(ofproto->sflow);
1605 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1608 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1613 mac_learning_run(ofproto->ml, &ofproto->backer->revalidate_set);
1615 /* Check the consistency of a random facet, to aid debugging. */
1616 if (time_msec() >= ofproto->consistency_rl
1617 && !hmap_is_empty(&ofproto->facets)
1618 && !ofproto->backer->need_revalidate) {
1619 struct facet *facet;
1621 ofproto->consistency_rl = time_msec() + 250;
1623 facet = CONTAINER_OF(hmap_random_node(&ofproto->facets),
1624 struct facet, hmap_node);
1625 if (!tag_set_intersects(&ofproto->backer->revalidate_set,
1627 if (!facet_check_consistency(facet)) {
1628 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
1633 if (ofproto->governor) {
1636 governor_run(ofproto->governor);
1638 /* If the governor has shrunk to its minimum size and the number of
1639 * subfacets has dwindled, then drop the governor entirely.
1641 * For hysteresis, the number of subfacets to drop the governor is
1642 * smaller than the number needed to trigger its creation. */
1643 n_subfacets = hmap_count(&ofproto->subfacets);
1644 if (n_subfacets * 4 < ofproto->up.flow_eviction_threshold
1645 && governor_is_idle(ofproto->governor)) {
1646 governor_destroy(ofproto->governor);
1647 ofproto->governor = NULL;
1655 wait(struct ofproto *ofproto_)
1657 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1658 struct ofport_dpif *ofport;
1659 struct ofbundle *bundle;
1661 if (!clogged && !list_is_empty(&ofproto->completions)) {
1662 poll_immediate_wake();
1665 dpif_wait(ofproto->backer->dpif);
1666 dpif_recv_wait(ofproto->backer->dpif);
1667 if (ofproto->sflow) {
1668 dpif_sflow_wait(ofproto->sflow);
1670 if (!tag_set_is_empty(&ofproto->backer->revalidate_set)) {
1671 poll_immediate_wake();
1673 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1676 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1677 bundle_wait(bundle);
1679 if (ofproto->netflow) {
1680 netflow_wait(ofproto->netflow);
1682 mac_learning_wait(ofproto->ml);
1684 if (ofproto->backer->need_revalidate) {
1685 /* Shouldn't happen, but if it does just go around again. */
1686 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1687 poll_immediate_wake();
1689 if (ofproto->governor) {
1690 governor_wait(ofproto->governor);
1695 get_memory_usage(const struct ofproto *ofproto_, struct simap *usage)
1697 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1699 simap_increase(usage, "facets", hmap_count(&ofproto->facets));
1700 simap_increase(usage, "subfacets", hmap_count(&ofproto->subfacets));
1704 flush(struct ofproto *ofproto_)
1706 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1707 struct subfacet *subfacet, *next_subfacet;
1708 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
1712 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
1713 &ofproto->subfacets) {
1714 if (subfacet->path != SF_NOT_INSTALLED) {
1715 batch[n_batch++] = subfacet;
1716 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
1717 subfacet_destroy_batch(ofproto, batch, n_batch);
1721 subfacet_destroy(subfacet);
1726 subfacet_destroy_batch(ofproto, batch, n_batch);
1731 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1732 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1734 *arp_match_ip = true;
1735 *actions = (OFPUTIL_A_OUTPUT |
1736 OFPUTIL_A_SET_VLAN_VID |
1737 OFPUTIL_A_SET_VLAN_PCP |
1738 OFPUTIL_A_STRIP_VLAN |
1739 OFPUTIL_A_SET_DL_SRC |
1740 OFPUTIL_A_SET_DL_DST |
1741 OFPUTIL_A_SET_NW_SRC |
1742 OFPUTIL_A_SET_NW_DST |
1743 OFPUTIL_A_SET_NW_TOS |
1744 OFPUTIL_A_SET_TP_SRC |
1745 OFPUTIL_A_SET_TP_DST |
1750 get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots)
1752 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1753 struct dpif_dp_stats s;
1755 strcpy(ots->name, "classifier");
1757 dpif_get_dp_stats(ofproto->backer->dpif, &s);
1759 ots->lookup_count = htonll(s.n_hit + s.n_missed);
1760 ots->matched_count = htonll(s.n_hit + ofproto->n_matches);
1763 static struct ofport *
1766 struct ofport_dpif *port = xmalloc(sizeof *port);
1771 port_dealloc(struct ofport *port_)
1773 struct ofport_dpif *port = ofport_dpif_cast(port_);
1778 port_construct(struct ofport *port_)
1780 struct ofport_dpif *port = ofport_dpif_cast(port_);
1781 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1782 const struct netdev *netdev = port->up.netdev;
1783 struct dpif_port dpif_port;
1786 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1787 port->bundle = NULL;
1790 port->tag = tag_create_random();
1791 port->may_enable = true;
1792 port->stp_port = NULL;
1793 port->stp_state = STP_DISABLED;
1794 port->tnl_port = NULL;
1795 hmap_init(&port->priorities);
1796 port->realdev_ofp_port = 0;
1797 port->vlandev_vid = 0;
1798 port->carrier_seq = netdev_get_carrier_resets(netdev);
1800 if (netdev_vport_is_patch(netdev)) {
1801 /* By bailing out here, we don't submit the port to the sFlow module
1802 * to be considered for counter polling export. This is correct
1803 * because the patch port represents an interface that sFlow considers
1804 * to be "internal" to the switch as a whole, and therefore not an
1805 * candidate for counter polling. */
1806 port->odp_port = OVSP_NONE;
1810 error = dpif_port_query_by_name(ofproto->backer->dpif,
1811 netdev_vport_get_dpif_port(netdev),
1817 port->odp_port = dpif_port.port_no;
1819 if (netdev_get_tunnel_config(netdev)) {
1820 port->tnl_port = tnl_port_add(&port->up, port->odp_port);
1822 /* Sanity-check that a mapping doesn't already exist. This
1823 * shouldn't happen for non-tunnel ports. */
1824 if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1825 VLOG_ERR("port %s already has an OpenFlow port number",
1827 dpif_port_destroy(&dpif_port);
1831 hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1832 hash_int(port->odp_port, 0));
1834 dpif_port_destroy(&dpif_port);
1836 if (ofproto->sflow) {
1837 dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1844 port_destruct(struct ofport *port_)
1846 struct ofport_dpif *port = ofport_dpif_cast(port_);
1847 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1848 const char *dp_port_name = netdev_vport_get_dpif_port(port->up.netdev);
1849 const char *devname = netdev_get_name(port->up.netdev);
1851 if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1852 /* The underlying device is still there, so delete it. This
1853 * happens when the ofproto is being destroyed, since the caller
1854 * assumes that removal of attached ports will happen as part of
1856 if (!port->tnl_port) {
1857 dpif_port_del(ofproto->backer->dpif, port->odp_port);
1859 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1862 if (port->odp_port != OVSP_NONE && !port->tnl_port) {
1863 hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1866 tnl_port_del(port->tnl_port);
1867 sset_find_and_delete(&ofproto->ports, devname);
1868 sset_find_and_delete(&ofproto->ghost_ports, devname);
1869 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1870 bundle_remove(port_);
1871 set_cfm(port_, NULL);
1872 set_bfd(port_, NULL);
1873 if (ofproto->sflow) {
1874 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1877 ofport_clear_priorities(port);
1878 hmap_destroy(&port->priorities);
1882 port_modified(struct ofport *port_)
1884 struct ofport_dpif *port = ofport_dpif_cast(port_);
1886 if (port->bundle && port->bundle->bond) {
1887 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1892 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1894 struct ofport_dpif *port = ofport_dpif_cast(port_);
1895 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1896 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1898 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1899 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1900 OFPUTIL_PC_NO_PACKET_IN)) {
1901 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1903 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1904 bundle_update(port->bundle);
1910 set_sflow(struct ofproto *ofproto_,
1911 const struct ofproto_sflow_options *sflow_options)
1913 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1914 struct dpif_sflow *ds = ofproto->sflow;
1916 if (sflow_options) {
1918 struct ofport_dpif *ofport;
1920 ds = ofproto->sflow = dpif_sflow_create();
1921 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1922 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1924 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1926 dpif_sflow_set_options(ds, sflow_options);
1929 dpif_sflow_destroy(ds);
1930 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1931 ofproto->sflow = NULL;
1939 struct ofproto *ofproto_,
1940 const struct ofproto_ipfix_bridge_exporter_options *bridge_exporter_options,
1941 const struct ofproto_ipfix_flow_exporter_options *flow_exporters_options,
1942 size_t n_flow_exporters_options)
1944 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1945 struct dpif_ipfix *di = ofproto->ipfix;
1947 if (bridge_exporter_options || flow_exporters_options) {
1949 di = ofproto->ipfix = dpif_ipfix_create();
1951 dpif_ipfix_set_options(
1952 di, bridge_exporter_options, flow_exporters_options,
1953 n_flow_exporters_options);
1956 dpif_ipfix_destroy(di);
1957 ofproto->ipfix = NULL;
1964 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1966 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1973 struct ofproto_dpif *ofproto;
1975 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1976 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1977 ofport->cfm = cfm_create(ofport->up.netdev);
1980 if (cfm_configure(ofport->cfm, s)) {
1986 cfm_destroy(ofport->cfm);
1992 get_cfm_status(const struct ofport *ofport_,
1993 struct ofproto_cfm_status *status)
1995 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1998 status->faults = cfm_get_fault(ofport->cfm);
1999 status->remote_opstate = cfm_get_opup(ofport->cfm);
2000 status->health = cfm_get_health(ofport->cfm);
2001 cfm_get_remote_mpids(ofport->cfm, &status->rmps, &status->n_rmps);
2009 set_bfd(struct ofport *ofport_, const struct smap *cfg)
2011 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
2012 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2016 ofport->bfd = bfd_configure(old, netdev_get_name(ofport->up.netdev), cfg);
2017 if (ofport->bfd != old) {
2018 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2025 get_bfd_status(struct ofport *ofport_, struct smap *smap)
2027 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2030 bfd_get_status(ofport->bfd, smap);
2037 /* Spanning Tree. */
2040 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
2042 struct ofproto_dpif *ofproto = ofproto_;
2043 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
2044 struct ofport_dpif *ofport;
2046 ofport = stp_port_get_aux(sp);
2048 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
2049 ofproto->up.name, port_num);
2051 struct eth_header *eth = pkt->l2;
2053 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
2054 if (eth_addr_is_zero(eth->eth_src)) {
2055 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
2056 "with unknown MAC", ofproto->up.name, port_num);
2058 send_packet(ofport, pkt);
2064 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
2066 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
2068 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2070 /* Only revalidate flows if the configuration changed. */
2071 if (!s != !ofproto->stp) {
2072 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2076 if (!ofproto->stp) {
2077 ofproto->stp = stp_create(ofproto_->name, s->system_id,
2078 send_bpdu_cb, ofproto);
2079 ofproto->stp_last_tick = time_msec();
2082 stp_set_bridge_id(ofproto->stp, s->system_id);
2083 stp_set_bridge_priority(ofproto->stp, s->priority);
2084 stp_set_hello_time(ofproto->stp, s->hello_time);
2085 stp_set_max_age(ofproto->stp, s->max_age);
2086 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
2088 struct ofport *ofport;
2090 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
2091 set_stp_port(ofport, NULL);
2094 stp_destroy(ofproto->stp);
2095 ofproto->stp = NULL;
2102 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
2104 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2108 s->bridge_id = stp_get_bridge_id(ofproto->stp);
2109 s->designated_root = stp_get_designated_root(ofproto->stp);
2110 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
2119 update_stp_port_state(struct ofport_dpif *ofport)
2121 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2122 enum stp_state state;
2124 /* Figure out new state. */
2125 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
2129 if (ofport->stp_state != state) {
2130 enum ofputil_port_state of_state;
2133 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
2134 netdev_get_name(ofport->up.netdev),
2135 stp_state_name(ofport->stp_state),
2136 stp_state_name(state));
2137 if (stp_learn_in_state(ofport->stp_state)
2138 != stp_learn_in_state(state)) {
2139 /* xxx Learning action flows should also be flushed. */
2140 mac_learning_flush(ofproto->ml,
2141 &ofproto->backer->revalidate_set);
2143 fwd_change = stp_forward_in_state(ofport->stp_state)
2144 != stp_forward_in_state(state);
2146 ofproto->backer->need_revalidate = REV_STP;
2147 ofport->stp_state = state;
2148 ofport->stp_state_entered = time_msec();
2150 if (fwd_change && ofport->bundle) {
2151 bundle_update(ofport->bundle);
2154 /* Update the STP state bits in the OpenFlow port description. */
2155 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
2156 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
2157 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
2158 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
2159 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
2161 ofproto_port_set_state(&ofport->up, of_state);
2165 /* Configures STP on 'ofport_' using the settings defined in 's'. The
2166 * caller is responsible for assigning STP port numbers and ensuring
2167 * there are no duplicates. */
2169 set_stp_port(struct ofport *ofport_,
2170 const struct ofproto_port_stp_settings *s)
2172 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2173 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2174 struct stp_port *sp = ofport->stp_port;
2176 if (!s || !s->enable) {
2178 ofport->stp_port = NULL;
2179 stp_port_disable(sp);
2180 update_stp_port_state(ofport);
2183 } else if (sp && stp_port_no(sp) != s->port_num
2184 && ofport == stp_port_get_aux(sp)) {
2185 /* The port-id changed, so disable the old one if it's not
2186 * already in use by another port. */
2187 stp_port_disable(sp);
2190 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
2191 stp_port_enable(sp);
2193 stp_port_set_aux(sp, ofport);
2194 stp_port_set_priority(sp, s->priority);
2195 stp_port_set_path_cost(sp, s->path_cost);
2197 update_stp_port_state(ofport);
2203 get_stp_port_status(struct ofport *ofport_,
2204 struct ofproto_port_stp_status *s)
2206 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2207 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2208 struct stp_port *sp = ofport->stp_port;
2210 if (!ofproto->stp || !sp) {
2216 s->port_id = stp_port_get_id(sp);
2217 s->state = stp_port_get_state(sp);
2218 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
2219 s->role = stp_port_get_role(sp);
2220 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
2226 stp_run(struct ofproto_dpif *ofproto)
2229 long long int now = time_msec();
2230 long long int elapsed = now - ofproto->stp_last_tick;
2231 struct stp_port *sp;
2234 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
2235 ofproto->stp_last_tick = now;
2237 while (stp_get_changed_port(ofproto->stp, &sp)) {
2238 struct ofport_dpif *ofport = stp_port_get_aux(sp);
2241 update_stp_port_state(ofport);
2245 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
2246 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
2252 stp_wait(struct ofproto_dpif *ofproto)
2255 poll_timer_wait(1000);
2259 /* Returns true if STP should process 'flow'. */
2261 stp_should_process_flow(const struct flow *flow)
2263 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
2267 stp_process_packet(const struct ofport_dpif *ofport,
2268 const struct ofpbuf *packet)
2270 struct ofpbuf payload = *packet;
2271 struct eth_header *eth = payload.data;
2272 struct stp_port *sp = ofport->stp_port;
2274 /* Sink packets on ports that have STP disabled when the bridge has
2276 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
2280 /* Trim off padding on payload. */
2281 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
2282 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
2285 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
2286 stp_received_bpdu(sp, payload.data, payload.size);
2290 static struct priority_to_dscp *
2291 get_priority(const struct ofport_dpif *ofport, uint32_t priority)
2293 struct priority_to_dscp *pdscp;
2296 hash = hash_int(priority, 0);
2297 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) {
2298 if (pdscp->priority == priority) {
2306 ofport_clear_priorities(struct ofport_dpif *ofport)
2308 struct priority_to_dscp *pdscp, *next;
2310 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) {
2311 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
2317 set_queues(struct ofport *ofport_,
2318 const struct ofproto_port_queue *qdscp_list,
2321 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2322 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2323 struct hmap new = HMAP_INITIALIZER(&new);
2326 for (i = 0; i < n_qdscp; i++) {
2327 struct priority_to_dscp *pdscp;
2331 dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
2332 if (dpif_queue_to_priority(ofproto->backer->dpif, qdscp_list[i].queue,
2337 pdscp = get_priority(ofport, priority);
2339 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
2341 pdscp = xmalloc(sizeof *pdscp);
2342 pdscp->priority = priority;
2344 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2347 if (pdscp->dscp != dscp) {
2349 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2352 hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0));
2355 if (!hmap_is_empty(&ofport->priorities)) {
2356 ofport_clear_priorities(ofport);
2357 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2360 hmap_swap(&new, &ofport->priorities);
2368 /* Expires all MAC learning entries associated with 'bundle' and forces its
2369 * ofproto to revalidate every flow.
2371 * Normally MAC learning entries are removed only from the ofproto associated
2372 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2373 * are removed from every ofproto. When patch ports and SLB bonds are in use
2374 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2375 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2376 * with the host from which it migrated. */
2378 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2380 struct ofproto_dpif *ofproto = bundle->ofproto;
2381 struct mac_learning *ml = ofproto->ml;
2382 struct mac_entry *mac, *next_mac;
2384 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2385 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2386 if (mac->port.p == bundle) {
2388 struct ofproto_dpif *o;
2390 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2392 struct mac_entry *e;
2394 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan,
2397 mac_learning_expire(o->ml, e);
2403 mac_learning_expire(ml, mac);
2408 static struct ofbundle *
2409 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2411 struct ofbundle *bundle;
2413 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2414 &ofproto->bundles) {
2415 if (bundle->aux == aux) {
2422 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
2423 * ones that are found to 'bundles'. */
2425 bundle_lookup_multiple(struct ofproto_dpif *ofproto,
2426 void **auxes, size_t n_auxes,
2427 struct hmapx *bundles)
2431 hmapx_init(bundles);
2432 for (i = 0; i < n_auxes; i++) {
2433 struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
2435 hmapx_add(bundles, bundle);
2441 bundle_update(struct ofbundle *bundle)
2443 struct ofport_dpif *port;
2445 bundle->floodable = true;
2446 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2447 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2448 || !stp_forward_in_state(port->stp_state)) {
2449 bundle->floodable = false;
2456 bundle_del_port(struct ofport_dpif *port)
2458 struct ofbundle *bundle = port->bundle;
2460 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2462 list_remove(&port->bundle_node);
2463 port->bundle = NULL;
2466 lacp_slave_unregister(bundle->lacp, port);
2469 bond_slave_unregister(bundle->bond, port);
2472 bundle_update(bundle);
2476 bundle_add_port(struct ofbundle *bundle, uint16_t ofp_port,
2477 struct lacp_slave_settings *lacp)
2479 struct ofport_dpif *port;
2481 port = get_ofp_port(bundle->ofproto, ofp_port);
2486 if (port->bundle != bundle) {
2487 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2489 bundle_del_port(port);
2492 port->bundle = bundle;
2493 list_push_back(&bundle->ports, &port->bundle_node);
2494 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2495 || !stp_forward_in_state(port->stp_state)) {
2496 bundle->floodable = false;
2500 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2501 lacp_slave_register(bundle->lacp, port, lacp);
2508 bundle_destroy(struct ofbundle *bundle)
2510 struct ofproto_dpif *ofproto;
2511 struct ofport_dpif *port, *next_port;
2518 ofproto = bundle->ofproto;
2519 for (i = 0; i < MAX_MIRRORS; i++) {
2520 struct ofmirror *m = ofproto->mirrors[i];
2522 if (m->out == bundle) {
2524 } else if (hmapx_find_and_delete(&m->srcs, bundle)
2525 || hmapx_find_and_delete(&m->dsts, bundle)) {
2526 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2531 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2532 bundle_del_port(port);
2535 bundle_flush_macs(bundle, true);
2536 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2538 free(bundle->trunks);
2539 lacp_destroy(bundle->lacp);
2540 bond_destroy(bundle->bond);
2545 bundle_set(struct ofproto *ofproto_, void *aux,
2546 const struct ofproto_bundle_settings *s)
2548 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2549 bool need_flush = false;
2550 struct ofport_dpif *port;
2551 struct ofbundle *bundle;
2552 unsigned long *trunks;
2558 bundle_destroy(bundle_lookup(ofproto, aux));
2562 ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2563 ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2565 bundle = bundle_lookup(ofproto, aux);
2567 bundle = xmalloc(sizeof *bundle);
2569 bundle->ofproto = ofproto;
2570 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2571 hash_pointer(aux, 0));
2573 bundle->name = NULL;
2575 list_init(&bundle->ports);
2576 bundle->vlan_mode = PORT_VLAN_TRUNK;
2578 bundle->trunks = NULL;
2579 bundle->use_priority_tags = s->use_priority_tags;
2580 bundle->lacp = NULL;
2581 bundle->bond = NULL;
2583 bundle->floodable = true;
2585 bundle->src_mirrors = 0;
2586 bundle->dst_mirrors = 0;
2587 bundle->mirror_out = 0;
2590 if (!bundle->name || strcmp(s->name, bundle->name)) {
2592 bundle->name = xstrdup(s->name);
2597 if (!bundle->lacp) {
2598 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2599 bundle->lacp = lacp_create();
2601 lacp_configure(bundle->lacp, s->lacp);
2603 lacp_destroy(bundle->lacp);
2604 bundle->lacp = NULL;
2607 /* Update set of ports. */
2609 for (i = 0; i < s->n_slaves; i++) {
2610 if (!bundle_add_port(bundle, s->slaves[i],
2611 s->lacp ? &s->lacp_slaves[i] : NULL)) {
2615 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2616 struct ofport_dpif *next_port;
2618 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2619 for (i = 0; i < s->n_slaves; i++) {
2620 if (s->slaves[i] == port->up.ofp_port) {
2625 bundle_del_port(port);
2629 ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2631 if (list_is_empty(&bundle->ports)) {
2632 bundle_destroy(bundle);
2636 /* Set VLAN tagging mode */
2637 if (s->vlan_mode != bundle->vlan_mode
2638 || s->use_priority_tags != bundle->use_priority_tags) {
2639 bundle->vlan_mode = s->vlan_mode;
2640 bundle->use_priority_tags = s->use_priority_tags;
2645 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2646 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2648 if (vlan != bundle->vlan) {
2649 bundle->vlan = vlan;
2653 /* Get trunked VLANs. */
2654 switch (s->vlan_mode) {
2655 case PORT_VLAN_ACCESS:
2659 case PORT_VLAN_TRUNK:
2660 trunks = CONST_CAST(unsigned long *, s->trunks);
2663 case PORT_VLAN_NATIVE_UNTAGGED:
2664 case PORT_VLAN_NATIVE_TAGGED:
2665 if (vlan != 0 && (!s->trunks
2666 || !bitmap_is_set(s->trunks, vlan)
2667 || bitmap_is_set(s->trunks, 0))) {
2668 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2670 trunks = bitmap_clone(s->trunks, 4096);
2672 trunks = bitmap_allocate1(4096);
2674 bitmap_set1(trunks, vlan);
2675 bitmap_set0(trunks, 0);
2677 trunks = CONST_CAST(unsigned long *, s->trunks);
2684 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2685 free(bundle->trunks);
2686 if (trunks == s->trunks) {
2687 bundle->trunks = vlan_bitmap_clone(trunks);
2689 bundle->trunks = trunks;
2694 if (trunks != s->trunks) {
2699 if (!list_is_short(&bundle->ports)) {
2700 bundle->ofproto->has_bonded_bundles = true;
2702 if (bond_reconfigure(bundle->bond, s->bond)) {
2703 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2706 bundle->bond = bond_create(s->bond);
2707 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2710 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2711 bond_slave_register(bundle->bond, port, port->up.netdev);
2714 bond_destroy(bundle->bond);
2715 bundle->bond = NULL;
2718 /* If we changed something that would affect MAC learning, un-learn
2719 * everything on this port and force flow revalidation. */
2721 bundle_flush_macs(bundle, false);
2728 bundle_remove(struct ofport *port_)
2730 struct ofport_dpif *port = ofport_dpif_cast(port_);
2731 struct ofbundle *bundle = port->bundle;
2734 bundle_del_port(port);
2735 if (list_is_empty(&bundle->ports)) {
2736 bundle_destroy(bundle);
2737 } else if (list_is_short(&bundle->ports)) {
2738 bond_destroy(bundle->bond);
2739 bundle->bond = NULL;
2745 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2747 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2748 struct ofport_dpif *port = port_;
2749 uint8_t ea[ETH_ADDR_LEN];
2752 error = netdev_get_etheraddr(port->up.netdev, ea);
2754 struct ofpbuf packet;
2757 ofpbuf_init(&packet, 0);
2758 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2760 memcpy(packet_pdu, pdu, pdu_size);
2762 send_packet(port, &packet);
2763 ofpbuf_uninit(&packet);
2765 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2766 "%s (%s)", port->bundle->name,
2767 netdev_get_name(port->up.netdev), strerror(error));
2772 bundle_send_learning_packets(struct ofbundle *bundle)
2774 struct ofproto_dpif *ofproto = bundle->ofproto;
2775 int error, n_packets, n_errors;
2776 struct mac_entry *e;
2778 error = n_packets = n_errors = 0;
2779 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2780 if (e->port.p != bundle) {
2781 struct ofpbuf *learning_packet;
2782 struct ofport_dpif *port;
2786 /* The assignment to "port" is unnecessary but makes "grep"ing for
2787 * struct ofport_dpif more effective. */
2788 learning_packet = bond_compose_learning_packet(bundle->bond,
2792 ret = send_packet(port, learning_packet);
2793 ofpbuf_delete(learning_packet);
2803 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2804 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2805 "packets, last error was: %s",
2806 bundle->name, n_errors, n_packets, strerror(error));
2808 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2809 bundle->name, n_packets);
2814 bundle_run(struct ofbundle *bundle)
2817 lacp_run(bundle->lacp, send_pdu_cb);
2820 struct ofport_dpif *port;
2822 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2823 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2826 bond_run(bundle->bond, &bundle->ofproto->backer->revalidate_set,
2827 lacp_status(bundle->lacp));
2828 if (bond_should_send_learning_packets(bundle->bond)) {
2829 bundle_send_learning_packets(bundle);
2835 bundle_wait(struct ofbundle *bundle)
2838 lacp_wait(bundle->lacp);
2841 bond_wait(bundle->bond);
2848 mirror_scan(struct ofproto_dpif *ofproto)
2852 for (idx = 0; idx < MAX_MIRRORS; idx++) {
2853 if (!ofproto->mirrors[idx]) {
2860 static struct ofmirror *
2861 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
2865 for (i = 0; i < MAX_MIRRORS; i++) {
2866 struct ofmirror *mirror = ofproto->mirrors[i];
2867 if (mirror && mirror->aux == aux) {
2875 /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */
2877 mirror_update_dups(struct ofproto_dpif *ofproto)
2881 for (i = 0; i < MAX_MIRRORS; i++) {
2882 struct ofmirror *m = ofproto->mirrors[i];
2885 m->dup_mirrors = MIRROR_MASK_C(1) << i;
2889 for (i = 0; i < MAX_MIRRORS; i++) {
2890 struct ofmirror *m1 = ofproto->mirrors[i];
2897 for (j = i + 1; j < MAX_MIRRORS; j++) {
2898 struct ofmirror *m2 = ofproto->mirrors[j];
2900 if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) {
2901 m1->dup_mirrors |= MIRROR_MASK_C(1) << j;
2902 m2->dup_mirrors |= m1->dup_mirrors;
2909 mirror_set(struct ofproto *ofproto_, void *aux,
2910 const struct ofproto_mirror_settings *s)
2912 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2913 mirror_mask_t mirror_bit;
2914 struct ofbundle *bundle;
2915 struct ofmirror *mirror;
2916 struct ofbundle *out;
2917 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
2918 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
2921 mirror = mirror_lookup(ofproto, aux);
2923 mirror_destroy(mirror);
2929 idx = mirror_scan(ofproto);
2931 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
2933 ofproto->up.name, MAX_MIRRORS, s->name);
2937 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
2938 mirror->ofproto = ofproto;
2941 mirror->out_vlan = -1;
2942 mirror->name = NULL;
2945 if (!mirror->name || strcmp(s->name, mirror->name)) {
2947 mirror->name = xstrdup(s->name);
2950 /* Get the new configuration. */
2951 if (s->out_bundle) {
2952 out = bundle_lookup(ofproto, s->out_bundle);
2954 mirror_destroy(mirror);
2960 out_vlan = s->out_vlan;
2962 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
2963 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
2965 /* If the configuration has not changed, do nothing. */
2966 if (hmapx_equals(&srcs, &mirror->srcs)
2967 && hmapx_equals(&dsts, &mirror->dsts)
2968 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
2969 && mirror->out == out
2970 && mirror->out_vlan == out_vlan)
2972 hmapx_destroy(&srcs);
2973 hmapx_destroy(&dsts);
2977 hmapx_swap(&srcs, &mirror->srcs);
2978 hmapx_destroy(&srcs);
2980 hmapx_swap(&dsts, &mirror->dsts);
2981 hmapx_destroy(&dsts);
2983 free(mirror->vlans);
2984 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
2987 mirror->out_vlan = out_vlan;
2989 /* Update bundles. */
2990 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2991 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
2992 if (hmapx_contains(&mirror->srcs, bundle)) {
2993 bundle->src_mirrors |= mirror_bit;
2995 bundle->src_mirrors &= ~mirror_bit;
2998 if (hmapx_contains(&mirror->dsts, bundle)) {
2999 bundle->dst_mirrors |= mirror_bit;
3001 bundle->dst_mirrors &= ~mirror_bit;
3004 if (mirror->out == bundle) {
3005 bundle->mirror_out |= mirror_bit;
3007 bundle->mirror_out &= ~mirror_bit;
3011 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3012 ofproto->has_mirrors = true;
3013 mac_learning_flush(ofproto->ml,
3014 &ofproto->backer->revalidate_set);
3015 mirror_update_dups(ofproto);
3021 mirror_destroy(struct ofmirror *mirror)
3023 struct ofproto_dpif *ofproto;
3024 mirror_mask_t mirror_bit;
3025 struct ofbundle *bundle;
3032 ofproto = mirror->ofproto;
3033 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3034 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
3036 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
3037 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
3038 bundle->src_mirrors &= ~mirror_bit;
3039 bundle->dst_mirrors &= ~mirror_bit;
3040 bundle->mirror_out &= ~mirror_bit;
3043 hmapx_destroy(&mirror->srcs);
3044 hmapx_destroy(&mirror->dsts);
3045 free(mirror->vlans);
3047 ofproto->mirrors[mirror->idx] = NULL;
3051 mirror_update_dups(ofproto);
3053 ofproto->has_mirrors = false;
3054 for (i = 0; i < MAX_MIRRORS; i++) {
3055 if (ofproto->mirrors[i]) {
3056 ofproto->has_mirrors = true;
3063 mirror_get_stats(struct ofproto *ofproto_, void *aux,
3064 uint64_t *packets, uint64_t *bytes)
3066 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3067 struct ofmirror *mirror = mirror_lookup(ofproto, aux);
3070 *packets = *bytes = UINT64_MAX;
3076 *packets = mirror->packet_count;
3077 *bytes = mirror->byte_count;
3083 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
3085 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3086 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
3087 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
3093 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
3095 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3096 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
3097 return bundle && bundle->mirror_out != 0;
3101 forward_bpdu_changed(struct ofproto *ofproto_)
3103 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3104 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3108 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
3111 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3112 mac_learning_set_idle_time(ofproto->ml, idle_time);
3113 mac_learning_set_max_entries(ofproto->ml, max_entries);
3118 static struct ofport_dpif *
3119 get_ofp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
3121 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
3122 return ofport ? ofport_dpif_cast(ofport) : NULL;
3125 static struct ofport_dpif *
3126 get_odp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
3128 struct ofport_dpif *port = odp_port_to_ofport(ofproto->backer, odp_port);
3129 return port && &ofproto->up == port->up.ofproto ? port : NULL;
3133 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
3134 struct ofproto_port *ofproto_port,
3135 struct dpif_port *dpif_port)
3137 ofproto_port->name = dpif_port->name;
3138 ofproto_port->type = dpif_port->type;
3139 ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
3142 static struct ofport_dpif *
3143 ofport_get_peer(const struct ofport_dpif *ofport_dpif)
3145 const struct ofproto_dpif *ofproto;
3148 peer = netdev_vport_patch_peer(ofport_dpif->up.netdev);
3153 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
3154 struct ofport *ofport;
3156 ofport = shash_find_data(&ofproto->up.port_by_name, peer);
3157 if (ofport && ofport->ofproto->ofproto_class == &ofproto_dpif_class) {
3158 return ofport_dpif_cast(ofport);
3165 port_run_fast(struct ofport_dpif *ofport)
3167 if (ofport->cfm && cfm_should_send_ccm(ofport->cfm)) {
3168 struct ofpbuf packet;
3170 ofpbuf_init(&packet, 0);
3171 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
3172 send_packet(ofport, &packet);
3173 ofpbuf_uninit(&packet);
3176 if (ofport->bfd && bfd_should_send_packet(ofport->bfd)) {
3177 struct ofpbuf packet;
3179 ofpbuf_init(&packet, 0);
3180 bfd_put_packet(ofport->bfd, &packet, ofport->up.pp.hw_addr);
3181 send_packet(ofport, &packet);
3182 ofpbuf_uninit(&packet);
3187 port_run(struct ofport_dpif *ofport)
3189 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
3190 bool carrier_changed = carrier_seq != ofport->carrier_seq;
3191 bool enable = netdev_get_carrier(ofport->up.netdev);
3193 ofport->carrier_seq = carrier_seq;
3195 port_run_fast(ofport);
3197 if (ofport->tnl_port
3198 && tnl_port_reconfigure(&ofport->up, ofport->odp_port,
3199 &ofport->tnl_port)) {
3200 ofproto_dpif_cast(ofport->up.ofproto)->backer->need_revalidate = true;
3204 int cfm_opup = cfm_get_opup(ofport->cfm);
3206 cfm_run(ofport->cfm);
3207 enable = enable && !cfm_get_fault(ofport->cfm);
3209 if (cfm_opup >= 0) {
3210 enable = enable && cfm_opup;
3215 bfd_run(ofport->bfd);
3216 enable = enable && bfd_forwarding(ofport->bfd);
3219 if (ofport->bundle) {
3220 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
3221 if (carrier_changed) {
3222 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
3226 if (ofport->may_enable != enable) {
3227 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3229 if (ofproto->has_bundle_action) {
3230 ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
3234 ofport->may_enable = enable;
3238 port_wait(struct ofport_dpif *ofport)
3241 cfm_wait(ofport->cfm);
3245 bfd_wait(ofport->bfd);
3250 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
3251 struct ofproto_port *ofproto_port)
3253 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3254 struct dpif_port dpif_port;
3257 if (sset_contains(&ofproto->ghost_ports, devname)) {
3258 const char *type = netdev_get_type_from_name(devname);
3260 /* We may be called before ofproto->up.port_by_name is populated with
3261 * the appropriate ofport. For this reason, we must get the name and
3262 * type from the netdev layer directly. */
3264 const struct ofport *ofport;
3266 ofport = shash_find_data(&ofproto->up.port_by_name, devname);
3267 ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
3268 ofproto_port->name = xstrdup(devname);
3269 ofproto_port->type = xstrdup(type);
3275 if (!sset_contains(&ofproto->ports, devname)) {
3278 error = dpif_port_query_by_name(ofproto->backer->dpif,
3279 devname, &dpif_port);
3281 ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
3287 port_add(struct ofproto *ofproto_, struct netdev *netdev)
3289 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3290 const char *dp_port_name = netdev_vport_get_dpif_port(netdev);
3291 const char *devname = netdev_get_name(netdev);
3293 if (netdev_vport_is_patch(netdev)) {
3294 sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
3298 if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
3299 uint32_t port_no = UINT32_MAX;
3302 error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
3306 if (netdev_get_tunnel_config(netdev)) {
3307 simap_put(&ofproto->backer->tnl_backers, dp_port_name, port_no);
3311 if (netdev_get_tunnel_config(netdev)) {
3312 sset_add(&ofproto->ghost_ports, devname);
3314 sset_add(&ofproto->ports, devname);
3320 port_del(struct ofproto *ofproto_, uint16_t ofp_port)
3322 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3323 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
3330 sset_find_and_delete(&ofproto->ghost_ports,
3331 netdev_get_name(ofport->up.netdev));
3332 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3333 if (!ofport->tnl_port) {
3334 error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
3336 /* The caller is going to close ofport->up.netdev. If this is a
3337 * bonded port, then the bond is using that netdev, so remove it
3338 * from the bond. The client will need to reconfigure everything
3339 * after deleting ports, so then the slave will get re-added. */
3340 bundle_remove(&ofport->up);
3347 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
3349 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3354 error = netdev_get_stats(ofport->up.netdev, stats);
3356 if (!error && ofport_->ofp_port == OFPP_LOCAL) {
3357 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3359 /* ofproto->stats.tx_packets represents packets that we created
3360 * internally and sent to some port (e.g. packets sent with
3361 * send_packet()). Account for them as if they had come from
3362 * OFPP_LOCAL and got forwarded. */
3364 if (stats->rx_packets != UINT64_MAX) {
3365 stats->rx_packets += ofproto->stats.tx_packets;
3368 if (stats->rx_bytes != UINT64_MAX) {
3369 stats->rx_bytes += ofproto->stats.tx_bytes;
3372 /* ofproto->stats.rx_packets represents packets that were received on
3373 * some port and we processed internally and dropped (e.g. STP).
3374 * Account for them as if they had been forwarded to OFPP_LOCAL. */
3376 if (stats->tx_packets != UINT64_MAX) {
3377 stats->tx_packets += ofproto->stats.rx_packets;
3380 if (stats->tx_bytes != UINT64_MAX) {
3381 stats->tx_bytes += ofproto->stats.rx_bytes;
3388 struct port_dump_state {
3393 struct ofproto_port port;
3398 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
3400 *statep = xzalloc(sizeof(struct port_dump_state));
3405 port_dump_next(const struct ofproto *ofproto_, void *state_,
3406 struct ofproto_port *port)
3408 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3409 struct port_dump_state *state = state_;
3410 const struct sset *sset;
3411 struct sset_node *node;
3413 if (state->has_port) {
3414 ofproto_port_destroy(&state->port);
3415 state->has_port = false;
3417 sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
3418 while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
3421 error = port_query_by_name(ofproto_, node->name, &state->port);
3423 *port = state->port;
3424 state->has_port = true;
3426 } else if (error != ENODEV) {
3431 if (!state->ghost) {
3432 state->ghost = true;
3435 return port_dump_next(ofproto_, state_, port);
3442 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3444 struct port_dump_state *state = state_;
3446 if (state->has_port) {
3447 ofproto_port_destroy(&state->port);
3454 port_poll(const struct ofproto *ofproto_, char **devnamep)
3456 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3458 if (ofproto->port_poll_errno) {
3459 int error = ofproto->port_poll_errno;
3460 ofproto->port_poll_errno = 0;
3464 if (sset_is_empty(&ofproto->port_poll_set)) {
3468 *devnamep = sset_pop(&ofproto->port_poll_set);
3473 port_poll_wait(const struct ofproto *ofproto_)
3475 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3476 dpif_port_poll_wait(ofproto->backer->dpif);
3480 port_is_lacp_current(const struct ofport *ofport_)
3482 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3483 return (ofport->bundle && ofport->bundle->lacp
3484 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3488 /* Upcall handling. */
3490 /* Flow miss batching.
3492 * Some dpifs implement operations faster when you hand them off in a batch.
3493 * To allow batching, "struct flow_miss" queues the dpif-related work needed
3494 * for a given flow. Each "struct flow_miss" corresponds to sending one or
3495 * more packets, plus possibly installing the flow in the dpif.
3497 * So far we only batch the operations that affect flow setup time the most.
3498 * It's possible to batch more than that, but the benefit might be minimal. */
3500 struct hmap_node hmap_node;
3501 struct ofproto_dpif *ofproto;
3503 enum odp_key_fitness key_fitness;
3504 const struct nlattr *key;
3506 struct initial_vals initial_vals;
3507 struct list packets;
3508 enum dpif_upcall_type upcall_type;
3509 uint32_t odp_in_port;
3512 struct flow_miss_op {
3513 struct dpif_op dpif_op;
3514 void *garbage; /* Pointer to pass to free(), NULL if none. */
3515 uint64_t stub[1024 / 8]; /* Temporary buffer. */
3518 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
3519 * OpenFlow controller as necessary according to their individual
3520 * configurations. */
3522 send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet,
3523 const struct flow *flow)
3525 struct ofputil_packet_in pin;
3527 pin.packet = packet->data;
3528 pin.packet_len = packet->size;
3529 pin.reason = OFPR_NO_MATCH;
3530 pin.controller_id = 0;
3535 pin.send_len = 0; /* not used for flow table misses */
3537 flow_get_metadata(flow, &pin.fmd);
3539 connmgr_send_packet_in(ofproto->up.connmgr, &pin);
3542 static enum slow_path_reason
3543 process_special(struct ofproto_dpif *ofproto, const struct flow *flow,
3544 const struct ofport_dpif *ofport, const struct ofpbuf *packet)
3548 } else if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) {
3550 cfm_process_heartbeat(ofport->cfm, packet);
3553 } else if (ofport->bfd && bfd_should_process_flow(flow)) {
3555 bfd_process_packet(ofport->bfd, flow, packet);
3558 } else if (ofport->bundle && ofport->bundle->lacp
3559 && flow->dl_type == htons(ETH_TYPE_LACP)) {
3561 lacp_process_packet(ofport->bundle->lacp, ofport, packet);
3564 } else if (ofproto->stp && stp_should_process_flow(flow)) {
3566 stp_process_packet(ofport, packet);
3574 static struct flow_miss *
3575 flow_miss_find(struct hmap *todo, const struct ofproto_dpif *ofproto,
3576 const struct flow *flow, uint32_t hash)
3578 struct flow_miss *miss;
3580 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
3581 if (miss->ofproto == ofproto && flow_equal(&miss->flow, flow)) {
3589 /* Partially Initializes 'op' as an "execute" operation for 'miss' and
3590 * 'packet'. The caller must initialize op->actions and op->actions_len. If
3591 * 'miss' is associated with a subfacet the caller must also initialize the
3592 * returned op->subfacet, and if anything needs to be freed after processing
3593 * the op, the caller must initialize op->garbage also. */
3595 init_flow_miss_execute_op(struct flow_miss *miss, struct ofpbuf *packet,
3596 struct flow_miss_op *op)
3598 if (miss->flow.vlan_tci != miss->initial_vals.vlan_tci) {
3599 /* This packet was received on a VLAN splinter port. We
3600 * added a VLAN to the packet to make the packet resemble
3601 * the flow, but the actions were composed assuming that
3602 * the packet contained no VLAN. So, we must remove the
3603 * VLAN header from the packet before trying to execute the
3605 eth_pop_vlan(packet);
3609 op->dpif_op.type = DPIF_OP_EXECUTE;
3610 op->dpif_op.u.execute.key = miss->key;
3611 op->dpif_op.u.execute.key_len = miss->key_len;
3612 op->dpif_op.u.execute.packet = packet;
3615 /* Helper for handle_flow_miss_without_facet() and
3616 * handle_flow_miss_with_facet(). */
3618 handle_flow_miss_common(struct rule_dpif *rule,
3619 struct ofpbuf *packet, const struct flow *flow)
3621 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3623 ofproto->n_matches++;
3625 if (rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
3627 * Extra-special case for fail-open mode.
3629 * We are in fail-open mode and the packet matched the fail-open
3630 * rule, but we are connected to a controller too. We should send
3631 * the packet up to the controller in the hope that it will try to
3632 * set up a flow and thereby allow us to exit fail-open.
3634 * See the top-level comment in fail-open.c for more information.
3636 send_packet_in_miss(ofproto, packet, flow);
3640 /* Figures out whether a flow that missed in 'ofproto', whose details are in
3641 * 'miss', is likely to be worth tracking in detail in userspace and (usually)
3642 * installing a datapath flow. The answer is usually "yes" (a return value of
3643 * true). However, for short flows the cost of bookkeeping is much higher than
3644 * the benefits, so when the datapath holds a large number of flows we impose
3645 * some heuristics to decide which flows are likely to be worth tracking. */
3647 flow_miss_should_make_facet(struct ofproto_dpif *ofproto,
3648 struct flow_miss *miss, uint32_t hash)
3650 if (!ofproto->governor) {
3653 n_subfacets = hmap_count(&ofproto->subfacets);
3654 if (n_subfacets * 2 <= ofproto->up.flow_eviction_threshold) {
3658 ofproto->governor = governor_create(ofproto->up.name);
3661 return governor_should_install_flow(ofproto->governor, hash,
3662 list_size(&miss->packets));
3665 /* Handles 'miss', which matches 'rule', without creating a facet or subfacet
3666 * or creating any datapath flow. May add an "execute" operation to 'ops' and
3667 * increment '*n_ops'. */
3669 handle_flow_miss_without_facet(struct flow_miss *miss,
3670 struct rule_dpif *rule,
3671 struct flow_miss_op *ops, size_t *n_ops)
3673 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3674 long long int now = time_msec();
3675 struct action_xlate_ctx ctx;
3676 struct ofpbuf *packet;
3678 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3679 struct flow_miss_op *op = &ops[*n_ops];
3680 struct dpif_flow_stats stats;
3681 struct ofpbuf odp_actions;
3683 COVERAGE_INC(facet_suppress);
3685 handle_flow_miss_common(rule, packet, &miss->flow);
3687 ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub);
3689 dpif_flow_stats_extract(&miss->flow, packet, now, &stats);
3690 rule_credit_stats(rule, &stats);
3692 action_xlate_ctx_init(&ctx, ofproto, &miss->flow, &miss->initial_vals,
3693 rule, stats.tcp_flags, packet);
3694 ctx.resubmit_stats = &stats;
3695 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len,
3698 if (odp_actions.size) {
3699 struct dpif_execute *execute = &op->dpif_op.u.execute;
3701 init_flow_miss_execute_op(miss, packet, op);
3702 execute->actions = odp_actions.data;
3703 execute->actions_len = odp_actions.size;
3704 op->garbage = ofpbuf_get_uninit_pointer(&odp_actions);
3708 ofpbuf_uninit(&odp_actions);
3713 /* Handles 'miss', which matches 'facet'. May add any required datapath
3714 * operations to 'ops', incrementing '*n_ops' for each new op.
3716 * All of the packets in 'miss' are considered to have arrived at time 'now'.
3717 * This is really important only for new facets: if we just called time_msec()
3718 * here, then the new subfacet or its packets could look (occasionally) as
3719 * though it was used some time after the facet was used. That can make a
3720 * one-packet flow look like it has a nonzero duration, which looks odd in
3721 * e.g. NetFlow statistics. */
3723 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
3725 struct flow_miss_op *ops, size_t *n_ops)
3727 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3728 enum subfacet_path want_path;
3729 struct subfacet *subfacet;
3730 struct ofpbuf *packet;
3732 subfacet = subfacet_create(facet, miss, now);
3734 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3735 struct flow_miss_op *op = &ops[*n_ops];
3736 struct dpif_flow_stats stats;
3738 handle_flow_miss_common(facet->rule, packet, &miss->flow);
3740 if (!subfacet->actions) {
3741 subfacet_make_actions(subfacet, packet);
3742 } else if (subfacet->slow) {
3743 struct action_xlate_ctx ctx;
3745 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
3746 &subfacet->initial_vals, facet->rule, 0,
3748 xlate_actions_for_side_effects(&ctx, facet->rule->up.ofpacts,
3749 facet->rule->up.ofpacts_len);
3752 dpif_flow_stats_extract(&facet->flow, packet, now, &stats);
3753 subfacet_update_stats(subfacet, &stats);
3755 if (subfacet->actions_len) {
3756 struct dpif_execute *execute = &op->dpif_op.u.execute;
3758 init_flow_miss_execute_op(miss, packet, op);
3759 execute->actions = subfacet->actions;
3760 execute->actions_len = subfacet->actions_len;
3766 want_path = subfacet_want_path(subfacet->slow);
3767 if (miss->upcall_type == DPIF_UC_MISS || subfacet->path != want_path) {
3768 struct flow_miss_op *op = &ops[(*n_ops)++];
3769 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
3771 subfacet->path = want_path;
3774 op->dpif_op.type = DPIF_OP_FLOW_PUT;
3775 put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
3776 put->key = miss->key;
3777 put->key_len = miss->key_len;
3778 if (want_path == SF_FAST_PATH) {
3779 put->actions = subfacet->actions;
3780 put->actions_len = subfacet->actions_len;
3782 compose_slow_path(ofproto, &facet->flow, subfacet->slow,
3783 op->stub, sizeof op->stub,
3784 &put->actions, &put->actions_len);
3790 /* Handles flow miss 'miss'. May add any required datapath operations
3791 * to 'ops', incrementing '*n_ops' for each new op. */
3793 handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops,
3796 struct ofproto_dpif *ofproto = miss->ofproto;
3797 struct facet *facet;
3801 /* The caller must ensure that miss->hmap_node.hash contains
3802 * flow_hash(miss->flow, 0). */
3803 hash = miss->hmap_node.hash;
3805 facet = facet_lookup_valid(ofproto, &miss->flow, hash);
3807 struct rule_dpif *rule = rule_dpif_lookup(ofproto, &miss->flow);
3809 /* There does not exist a bijection between 'struct flow' and datapath
3810 * flow keys with fitness ODP_FIT_TO_LITTLE. This breaks a fundamental
3811 * assumption used throughout the facet and subfacet handling code.
3812 * Since we have to handle these misses in userspace anyway, we simply
3813 * skip facet creation, avoiding the problem alltogether. */
3814 if (miss->key_fitness == ODP_FIT_TOO_LITTLE
3815 || !flow_miss_should_make_facet(ofproto, miss, hash)) {
3816 handle_flow_miss_without_facet(miss, rule, ops, n_ops);
3820 facet = facet_create(rule, &miss->flow, hash);
3825 handle_flow_miss_with_facet(miss, facet, now, ops, n_ops);
3828 static struct drop_key *
3829 drop_key_lookup(const struct dpif_backer *backer, const struct nlattr *key,
3832 struct drop_key *drop_key;
3834 HMAP_FOR_EACH_WITH_HASH (drop_key, hmap_node, hash_bytes(key, key_len, 0),
3835 &backer->drop_keys) {
3836 if (drop_key->key_len == key_len
3837 && !memcmp(drop_key->key, key, key_len)) {
3845 drop_key_clear(struct dpif_backer *backer)
3847 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3848 struct drop_key *drop_key, *next;
3850 HMAP_FOR_EACH_SAFE (drop_key, next, hmap_node, &backer->drop_keys) {
3853 error = dpif_flow_del(backer->dpif, drop_key->key, drop_key->key_len,
3855 if (error && !VLOG_DROP_WARN(&rl)) {
3856 struct ds ds = DS_EMPTY_INITIALIZER;
3857 odp_flow_key_format(drop_key->key, drop_key->key_len, &ds);
3858 VLOG_WARN("Failed to delete drop key (%s) (%s)", strerror(error),
3863 hmap_remove(&backer->drop_keys, &drop_key->hmap_node);
3864 free(drop_key->key);
3869 /* Given a datpath, packet, and flow metadata ('backer', 'packet', and 'key'
3870 * respectively), populates 'flow' with the result of odp_flow_key_to_flow().
3871 * Optionally, if nonnull, populates 'fitnessp' with the fitness of 'flow' as
3872 * returned by odp_flow_key_to_flow(). Also, optionally populates 'ofproto'
3873 * with the ofproto_dpif, and 'odp_in_port' with the datapath in_port, that
3874 * 'packet' ingressed.
3876 * If 'ofproto' is nonnull, requires 'flow''s in_port to exist. Otherwise sets
3877 * 'flow''s in_port to OFPP_NONE.
3879 * This function does post-processing on data returned from
3880 * odp_flow_key_to_flow() to help make VLAN splinters transparent to the rest
3881 * of the upcall processing logic. In particular, if the extracted in_port is
3882 * a VLAN splinter port, it replaces flow->in_port by the "real" port, sets
3883 * flow->vlan_tci correctly for the VLAN of the VLAN splinter port, and pushes
3884 * a VLAN header onto 'packet' (if it is nonnull).
3886 * Optionally, if 'initial_vals' is nonnull, sets 'initial_vals->vlan_tci'
3887 * to the VLAN TCI with which the packet was really received, that is, the
3888 * actual VLAN TCI extracted by odp_flow_key_to_flow(). (This differs from
3889 * the value returned in flow->vlan_tci only for packets received on
3892 * Similarly, this function also includes some logic to help with tunnels. It
3893 * may modify 'flow' as necessary to make the tunneling implementation
3894 * transparent to the upcall processing logic.
3896 * Returns 0 if successful, ENODEV if the parsed flow has no associated ofport,
3897 * or some other positive errno if there are other problems. */
3899 ofproto_receive(const struct dpif_backer *backer, struct ofpbuf *packet,
3900 const struct nlattr *key, size_t key_len,
3901 struct flow *flow, enum odp_key_fitness *fitnessp,
3902 struct ofproto_dpif **ofproto, uint32_t *odp_in_port,
3903 struct initial_vals *initial_vals)
3905 const struct ofport_dpif *port;
3906 enum odp_key_fitness fitness;
3909 fitness = odp_flow_key_to_flow(key, key_len, flow);
3910 if (fitness == ODP_FIT_ERROR) {
3916 initial_vals->vlan_tci = flow->vlan_tci;
3920 *odp_in_port = flow->in_port;
3923 port = (tnl_port_should_receive(flow)
3924 ? ofport_dpif_cast(tnl_port_receive(flow))
3925 : odp_port_to_ofport(backer, flow->in_port));
3926 flow->in_port = port ? port->up.ofp_port : OFPP_NONE;
3931 /* XXX: Since the tunnel module is not scoped per backer, for a tunnel port
3932 * it's theoretically possible that we'll receive an ofport belonging to an
3933 * entirely different datapath. In practice, this can't happen because no
3934 * platforms has two separate datapaths which each support tunneling. */
3935 ovs_assert(ofproto_dpif_cast(port->up.ofproto)->backer == backer);
3937 if (vsp_adjust_flow(ofproto_dpif_cast(port->up.ofproto), flow)) {
3939 /* Make the packet resemble the flow, so that it gets sent to
3940 * an OpenFlow controller properly, so that it looks correct
3941 * for sFlow, and so that flow_extract() will get the correct
3942 * vlan_tci if it is called on 'packet'.
3944 * The allocated space inside 'packet' probably also contains
3945 * 'key', that is, both 'packet' and 'key' are probably part of
3946 * a struct dpif_upcall (see the large comment on that
3947 * structure definition), so pushing data on 'packet' is in
3948 * general not a good idea since it could overwrite 'key' or
3949 * free it as a side effect. However, it's OK in this special
3950 * case because we know that 'packet' is inside a Netlink
3951 * attribute: pushing 4 bytes will just overwrite the 4-byte
3952 * "struct nlattr", which is fine since we don't need that
3953 * header anymore. */
3954 eth_push_vlan(packet, flow->vlan_tci);
3956 /* We can't reproduce 'key' from 'flow'. */
3957 fitness = fitness == ODP_FIT_PERFECT ? ODP_FIT_TOO_MUCH : fitness;
3962 *ofproto = ofproto_dpif_cast(port->up.ofproto);
3967 *fitnessp = fitness;
3973 handle_miss_upcalls(struct dpif_backer *backer, struct dpif_upcall *upcalls,
3976 struct dpif_upcall *upcall;
3977 struct flow_miss *miss;
3978 struct flow_miss misses[FLOW_MISS_MAX_BATCH];
3979 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
3980 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
3990 /* Construct the to-do list.
3992 * This just amounts to extracting the flow from each packet and sticking
3993 * the packets that have the same flow in the same "flow_miss" structure so
3994 * that we can process them together. */
3997 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
3998 struct flow_miss *miss = &misses[n_misses];
3999 struct flow_miss *existing_miss;
4000 struct ofproto_dpif *ofproto;
4001 uint32_t odp_in_port;
4006 error = ofproto_receive(backer, upcall->packet, upcall->key,
4007 upcall->key_len, &flow, &miss->key_fitness,
4008 &ofproto, &odp_in_port, &miss->initial_vals);
4009 if (error == ENODEV) {
4010 struct drop_key *drop_key;
4012 /* Received packet on port for which we couldn't associate
4013 * an ofproto. This can happen if a port is removed while
4014 * traffic is being received. Print a rate-limited message
4015 * in case it happens frequently. Install a drop flow so
4016 * that future packets of the flow are inexpensively dropped
4018 VLOG_INFO_RL(&rl, "received packet on unassociated port %"PRIu32,
4021 drop_key = drop_key_lookup(backer, upcall->key, upcall->key_len);
4023 drop_key = xmalloc(sizeof *drop_key);
4024 drop_key->key = xmemdup(upcall->key, upcall->key_len);
4025 drop_key->key_len = upcall->key_len;
4027 hmap_insert(&backer->drop_keys, &drop_key->hmap_node,
4028 hash_bytes(drop_key->key, drop_key->key_len, 0));
4029 dpif_flow_put(backer->dpif, DPIF_FP_CREATE | DPIF_FP_MODIFY,
4030 drop_key->key, drop_key->key_len, NULL, 0, NULL);
4038 ofproto->n_missed++;
4039 flow_extract(upcall->packet, flow.skb_priority, flow.skb_mark,
4040 &flow.tunnel, flow.in_port, &miss->flow);
4042 /* Add other packets to a to-do list. */
4043 hash = flow_hash(&miss->flow, 0);
4044 existing_miss = flow_miss_find(&todo, ofproto, &miss->flow, hash);
4045 if (!existing_miss) {
4046 hmap_insert(&todo, &miss->hmap_node, hash);
4047 miss->ofproto = ofproto;
4048 miss->key = upcall->key;
4049 miss->key_len = upcall->key_len;
4050 miss->upcall_type = upcall->type;
4051 miss->odp_in_port = odp_in_port;
4052 list_init(&miss->packets);
4056 miss = existing_miss;
4058 list_push_back(&miss->packets, &upcall->packet->list_node);
4061 /* Process each element in the to-do list, constructing the set of
4062 * operations to batch. */
4064 HMAP_FOR_EACH (miss, hmap_node, &todo) {
4065 handle_flow_miss(miss, flow_miss_ops, &n_ops);
4067 ovs_assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
4069 /* Execute batch. */
4070 for (i = 0; i < n_ops; i++) {
4071 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
4073 dpif_operate(backer->dpif, dpif_ops, n_ops);
4076 for (i = 0; i < n_ops; i++) {
4077 free(flow_miss_ops[i].garbage);
4079 hmap_destroy(&todo);
4082 static enum { SFLOW_UPCALL, MISS_UPCALL, BAD_UPCALL, FLOW_SAMPLE_UPCALL,
4084 classify_upcall(const struct dpif_upcall *upcall)
4086 size_t userdata_len;
4087 union user_action_cookie cookie;
4089 /* First look at the upcall type. */
4090 switch (upcall->type) {
4091 case DPIF_UC_ACTION:
4097 case DPIF_N_UC_TYPES:
4099 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
4103 /* "action" upcalls need a closer look. */
4104 if (!upcall->userdata) {
4105 VLOG_WARN_RL(&rl, "action upcall missing cookie");
4108 userdata_len = nl_attr_get_size(upcall->userdata);
4109 if (userdata_len < sizeof cookie.type
4110 || userdata_len > sizeof cookie) {
4111 VLOG_WARN_RL(&rl, "action upcall cookie has unexpected size %zu",
4115 memset(&cookie, 0, sizeof cookie);
4116 memcpy(&cookie, nl_attr_get(upcall->userdata), userdata_len);
4117 if (userdata_len == sizeof cookie.sflow
4118 && cookie.type == USER_ACTION_COOKIE_SFLOW) {
4119 return SFLOW_UPCALL;
4120 } else if (userdata_len == sizeof cookie.slow_path
4121 && cookie.type == USER_ACTION_COOKIE_SLOW_PATH) {
4123 } else if (userdata_len == sizeof cookie.flow_sample
4124 && cookie.type == USER_ACTION_COOKIE_FLOW_SAMPLE) {
4125 return FLOW_SAMPLE_UPCALL;
4126 } else if (userdata_len == sizeof cookie.ipfix
4127 && cookie.type == USER_ACTION_COOKIE_IPFIX) {
4128 return IPFIX_UPCALL;
4130 VLOG_WARN_RL(&rl, "invalid user cookie of type %"PRIu16
4131 " and size %zu", cookie.type, userdata_len);
4137 handle_sflow_upcall(struct dpif_backer *backer,
4138 const struct dpif_upcall *upcall)
4140 struct ofproto_dpif *ofproto;
4141 union user_action_cookie cookie;
4143 uint32_t odp_in_port;
4145 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
4146 &flow, NULL, &ofproto, &odp_in_port, NULL)
4147 || !ofproto->sflow) {
4151 memset(&cookie, 0, sizeof cookie);
4152 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.sflow);
4153 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
4154 odp_in_port, &cookie);
4158 handle_flow_sample_upcall(struct dpif_backer *backer,
4159 const struct dpif_upcall *upcall)
4161 struct ofproto_dpif *ofproto;
4162 union user_action_cookie cookie;
4165 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
4166 &flow, NULL, &ofproto, NULL, NULL)
4167 || !ofproto->ipfix) {
4171 memset(&cookie, 0, sizeof cookie);
4172 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.flow_sample);
4174 /* The flow reflects exactly the contents of the packet. Sample
4175 * the packet using it. */
4176 dpif_ipfix_flow_sample(ofproto->ipfix, upcall->packet, &flow,
4177 cookie.flow_sample.collector_set_id,
4178 cookie.flow_sample.probability,
4179 cookie.flow_sample.obs_domain_id,
4180 cookie.flow_sample.obs_point_id);
4184 handle_ipfix_upcall(struct dpif_backer *backer,
4185 const struct dpif_upcall *upcall)
4187 struct ofproto_dpif *ofproto;
4190 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
4191 &flow, NULL, &ofproto, NULL, NULL)
4192 || !ofproto->ipfix) {
4196 /* The flow reflects exactly the contents of the packet. Sample
4197 * the packet using it. */
4198 dpif_ipfix_bridge_sample(ofproto->ipfix, upcall->packet, &flow);
4202 handle_upcalls(struct dpif_backer *backer, unsigned int max_batch)
4204 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
4205 struct ofpbuf miss_bufs[FLOW_MISS_MAX_BATCH];
4206 uint64_t miss_buf_stubs[FLOW_MISS_MAX_BATCH][4096 / 8];
4211 ovs_assert(max_batch <= FLOW_MISS_MAX_BATCH);
4214 for (n_processed = 0; n_processed < max_batch; n_processed++) {
4215 struct dpif_upcall *upcall = &misses[n_misses];
4216 struct ofpbuf *buf = &miss_bufs[n_misses];
4219 ofpbuf_use_stub(buf, miss_buf_stubs[n_misses],
4220 sizeof miss_buf_stubs[n_misses]);
4221 error = dpif_recv(backer->dpif, upcall, buf);
4227 switch (classify_upcall(upcall)) {
4229 /* Handle it later. */
4234 handle_sflow_upcall(backer, upcall);
4238 case FLOW_SAMPLE_UPCALL:
4239 handle_flow_sample_upcall(backer, upcall);
4244 handle_ipfix_upcall(backer, upcall);
4254 /* Handle deferred MISS_UPCALL processing. */
4255 handle_miss_upcalls(backer, misses, n_misses);
4256 for (i = 0; i < n_misses; i++) {
4257 ofpbuf_uninit(&miss_bufs[i]);
4263 /* Flow expiration. */
4265 static int subfacet_max_idle(const struct ofproto_dpif *);
4266 static void update_stats(struct dpif_backer *);
4267 static void rule_expire(struct rule_dpif *);
4268 static void expire_subfacets(struct ofproto_dpif *, int dp_max_idle);
4270 /* This function is called periodically by run(). Its job is to collect
4271 * updates for the flows that have been installed into the datapath, most
4272 * importantly when they last were used, and then use that information to
4273 * expire flows that have not been used recently.
4275 * Returns the number of milliseconds after which it should be called again. */
4277 expire(struct dpif_backer *backer)
4279 struct ofproto_dpif *ofproto;
4280 int max_idle = INT32_MAX;
4282 /* Periodically clear out the drop keys in an effort to keep them
4283 * relatively few. */
4284 drop_key_clear(backer);
4286 /* Update stats for each flow in the backer. */
4287 update_stats(backer);
4289 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4290 struct rule *rule, *next_rule;
4293 if (ofproto->backer != backer) {
4297 /* Keep track of the max number of flows per ofproto_dpif. */
4298 update_max_subfacet_count(ofproto);
4300 /* Expire subfacets that have been idle too long. */
4301 dp_max_idle = subfacet_max_idle(ofproto);
4302 expire_subfacets(ofproto, dp_max_idle);
4304 max_idle = MIN(max_idle, dp_max_idle);
4306 /* Expire OpenFlow flows whose idle_timeout or hard_timeout
4308 LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
4309 &ofproto->up.expirable) {
4310 rule_expire(rule_dpif_cast(rule));
4313 /* All outstanding data in existing flows has been accounted, so it's a
4314 * good time to do bond rebalancing. */
4315 if (ofproto->has_bonded_bundles) {
4316 struct ofbundle *bundle;
4318 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
4320 bond_rebalance(bundle->bond, &backer->revalidate_set);
4326 return MIN(max_idle, 1000);
4329 /* Updates flow table statistics given that the datapath just reported 'stats'
4330 * as 'subfacet''s statistics. */
4332 update_subfacet_stats(struct subfacet *subfacet,
4333 const struct dpif_flow_stats *stats)
4335 struct facet *facet = subfacet->facet;
4337 if (stats->n_packets >= subfacet->dp_packet_count) {
4338 uint64_t extra = stats->n_packets - subfacet->dp_packet_count;
4339 facet->packet_count += extra;
4341 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
4344 if (stats->n_bytes >= subfacet->dp_byte_count) {
4345 facet->byte_count += stats->n_bytes - subfacet->dp_byte_count;
4347 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
4350 subfacet->dp_packet_count = stats->n_packets;
4351 subfacet->dp_byte_count = stats->n_bytes;
4353 facet->tcp_flags |= stats->tcp_flags;
4355 subfacet_update_time(subfacet, stats->used);
4356 if (facet->accounted_bytes < facet->byte_count) {
4358 facet_account(facet);
4359 facet->accounted_bytes = facet->byte_count;
4363 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
4364 * about, or a flow that shouldn't be installed but was anyway. Delete it. */
4366 delete_unexpected_flow(struct ofproto_dpif *ofproto,
4367 const struct nlattr *key, size_t key_len)
4369 if (!VLOG_DROP_WARN(&rl)) {
4373 odp_flow_key_format(key, key_len, &s);
4374 VLOG_WARN("unexpected flow on %s: %s", ofproto->up.name, ds_cstr(&s));
4378 COVERAGE_INC(facet_unexpected);
4379 dpif_flow_del(ofproto->backer->dpif, key, key_len, NULL);
4382 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
4384 * This function also pushes statistics updates to rules which each facet
4385 * resubmits into. Generally these statistics will be accurate. However, if a
4386 * facet changes the rule it resubmits into at some time in between
4387 * update_stats() runs, it is possible that statistics accrued to the
4388 * old rule will be incorrectly attributed to the new rule. This could be
4389 * avoided by calling update_stats() whenever rules are created or
4390 * deleted. However, the performance impact of making so many calls to the
4391 * datapath do not justify the benefit of having perfectly accurate statistics.
4393 * In addition, this function maintains per ofproto flow hit counts. The patch
4394 * port is not treated specially. e.g. A packet ingress from br0 patched into
4395 * br1 will increase the hit count of br0 by 1, however, does not affect
4396 * the hit or miss counts of br1.
4399 update_stats(struct dpif_backer *backer)
4401 const struct dpif_flow_stats *stats;
4402 struct dpif_flow_dump dump;
4403 const struct nlattr *key;
4404 struct ofproto_dpif *ofproto;
4407 dpif_flow_dump_start(&dump, backer->dpif);
4408 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
4410 struct subfacet *subfacet;
4411 struct ofport_dpif *ofport;
4414 if (ofproto_receive(backer, NULL, key, key_len, &flow, NULL, &ofproto,
4419 ofproto->total_subfacet_count += hmap_count(&ofproto->subfacets);
4420 ofproto->n_update_stats++;
4422 ofport = get_ofp_port(ofproto, flow.in_port);
4423 if (ofport && ofport->tnl_port) {
4424 netdev_vport_inc_rx(ofport->up.netdev, stats);
4427 key_hash = odp_flow_key_hash(key, key_len);
4428 subfacet = subfacet_find(ofproto, key, key_len, key_hash);
4429 switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
4431 /* Update ofproto_dpif's hit count. */
4432 if (stats->n_packets > subfacet->dp_packet_count) {
4433 uint64_t delta = stats->n_packets - subfacet->dp_packet_count;
4434 dpif_stats_update_hit_count(ofproto, delta);
4437 update_subfacet_stats(subfacet, stats);
4441 /* Stats are updated per-packet. */
4444 case SF_NOT_INSTALLED:
4446 delete_unexpected_flow(ofproto, key, key_len);
4451 dpif_flow_dump_done(&dump);
4453 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4454 update_moving_averages(ofproto);
4459 /* Calculates and returns the number of milliseconds of idle time after which
4460 * subfacets should expire from the datapath. When a subfacet expires, we fold
4461 * its statistics into its facet, and when a facet's last subfacet expires, we
4462 * fold its statistic into its rule. */
4464 subfacet_max_idle(const struct ofproto_dpif *ofproto)
4467 * Idle time histogram.
4469 * Most of the time a switch has a relatively small number of subfacets.
4470 * When this is the case we might as well keep statistics for all of them
4471 * in userspace and to cache them in the kernel datapath for performance as
4474 * As the number of subfacets increases, the memory required to maintain
4475 * statistics about them in userspace and in the kernel becomes
4476 * significant. However, with a large number of subfacets it is likely
4477 * that only a few of them are "heavy hitters" that consume a large amount
4478 * of bandwidth. At this point, only heavy hitters are worth caching in
4479 * the kernel and maintaining in userspaces; other subfacets we can
4482 * The technique used to compute the idle time is to build a histogram with
4483 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
4484 * that is installed in the kernel gets dropped in the appropriate bucket.
4485 * After the histogram has been built, we compute the cutoff so that only
4486 * the most-recently-used 1% of subfacets (but at least
4487 * ofproto->up.flow_eviction_threshold flows) are kept cached. At least
4488 * the most-recently-used bucket of subfacets is kept, so actually an
4489 * arbitrary number of subfacets can be kept in any given expiration run
4490 * (though the next run will delete most of those unless they receive
4493 * This requires a second pass through the subfacets, in addition to the
4494 * pass made by update_stats(), because the former function never looks at
4495 * uninstallable subfacets.
4497 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
4498 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
4499 int buckets[N_BUCKETS] = { 0 };
4500 int total, subtotal, bucket;
4501 struct subfacet *subfacet;
4505 total = hmap_count(&ofproto->subfacets);
4506 if (total <= ofproto->up.flow_eviction_threshold) {
4507 return N_BUCKETS * BUCKET_WIDTH;
4510 /* Build histogram. */
4512 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
4513 long long int idle = now - subfacet->used;
4514 int bucket = (idle <= 0 ? 0
4515 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
4516 : (unsigned int) idle / BUCKET_WIDTH);
4520 /* Find the first bucket whose flows should be expired. */
4521 subtotal = bucket = 0;
4523 subtotal += buckets[bucket++];
4524 } while (bucket < N_BUCKETS &&
4525 subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100));
4527 if (VLOG_IS_DBG_ENABLED()) {
4531 ds_put_cstr(&s, "keep");
4532 for (i = 0; i < N_BUCKETS; i++) {
4534 ds_put_cstr(&s, ", drop");
4537 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
4540 VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s));
4544 return bucket * BUCKET_WIDTH;
4548 expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle)
4550 /* Cutoff time for most flows. */
4551 long long int normal_cutoff = time_msec() - dp_max_idle;
4553 /* We really want to keep flows for special protocols around, so use a more
4554 * conservative cutoff. */
4555 long long int special_cutoff = time_msec() - 10000;
4557 struct subfacet *subfacet, *next_subfacet;
4558 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
4562 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
4563 &ofproto->subfacets) {
4564 long long int cutoff;
4566 cutoff = (subfacet->slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP | SLOW_STP)
4569 if (subfacet->used < cutoff) {
4570 if (subfacet->path != SF_NOT_INSTALLED) {
4571 batch[n_batch++] = subfacet;
4572 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
4573 subfacet_destroy_batch(ofproto, batch, n_batch);
4577 subfacet_destroy(subfacet);
4583 subfacet_destroy_batch(ofproto, batch, n_batch);
4587 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
4588 * then delete it entirely. */
4590 rule_expire(struct rule_dpif *rule)
4592 struct facet *facet, *next_facet;
4596 if (rule->up.pending) {
4597 /* We'll have to expire it later. */
4601 /* Has 'rule' expired? */
4603 if (rule->up.hard_timeout
4604 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
4605 reason = OFPRR_HARD_TIMEOUT;
4606 } else if (rule->up.idle_timeout
4607 && now > rule->up.used + rule->up.idle_timeout * 1000) {
4608 reason = OFPRR_IDLE_TIMEOUT;
4613 COVERAGE_INC(ofproto_dpif_expired);
4615 /* Update stats. (This is a no-op if the rule expired due to an idle
4616 * timeout, because that only happens when the rule has no facets left.) */
4617 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4618 facet_remove(facet);
4621 /* Get rid of the rule. */
4622 ofproto_rule_expire(&rule->up, reason);
4627 /* Creates and returns a new facet owned by 'rule', given a 'flow'.
4629 * The caller must already have determined that no facet with an identical
4630 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
4631 * the ofproto's classifier table.
4633 * 'hash' must be the return value of flow_hash(flow, 0).
4635 * The facet will initially have no subfacets. The caller should create (at
4636 * least) one subfacet with subfacet_create(). */
4637 static struct facet *
4638 facet_create(struct rule_dpif *rule, const struct flow *flow, uint32_t hash)
4640 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4641 struct facet *facet;
4643 facet = xzalloc(sizeof *facet);
4644 facet->used = time_msec();
4645 hmap_insert(&ofproto->facets, &facet->hmap_node, hash);
4646 list_push_back(&rule->facets, &facet->list_node);
4648 facet->flow = *flow;
4649 list_init(&facet->subfacets);
4650 netflow_flow_init(&facet->nf_flow);
4651 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
4653 facet->learn_rl = time_msec() + 500;
4659 facet_free(struct facet *facet)
4664 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
4665 * 'packet', which arrived on 'in_port'. */
4667 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
4668 const struct nlattr *odp_actions, size_t actions_len,
4669 struct ofpbuf *packet)
4671 struct odputil_keybuf keybuf;
4675 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4676 odp_flow_key_from_flow(&key, flow,
4677 ofp_port_to_odp_port(ofproto, flow->in_port));
4679 error = dpif_execute(ofproto->backer->dpif, key.data, key.size,
4680 odp_actions, actions_len, packet);
4684 /* Remove 'facet' from 'ofproto' and free up the associated memory:
4686 * - If 'facet' was installed in the datapath, uninstalls it and updates its
4687 * rule's statistics, via subfacet_uninstall().
4689 * - Removes 'facet' from its rule and from ofproto->facets.
4692 facet_remove(struct facet *facet)
4694 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4695 struct subfacet *subfacet, *next_subfacet;
4697 ovs_assert(!list_is_empty(&facet->subfacets));
4699 /* First uninstall all of the subfacets to get final statistics. */
4700 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4701 subfacet_uninstall(subfacet);
4704 /* Flush the final stats to the rule.
4706 * This might require us to have at least one subfacet around so that we
4707 * can use its actions for accounting in facet_account(), which is why we
4708 * have uninstalled but not yet destroyed the subfacets. */
4709 facet_flush_stats(facet);
4711 /* Now we're really all done so destroy everything. */
4712 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
4713 &facet->subfacets) {
4714 subfacet_destroy__(subfacet);
4716 hmap_remove(&ofproto->facets, &facet->hmap_node);
4717 list_remove(&facet->list_node);
4721 /* Feed information from 'facet' back into the learning table to keep it in
4722 * sync with what is actually flowing through the datapath. */
4724 facet_learn(struct facet *facet)
4726 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4727 struct subfacet *subfacet= CONTAINER_OF(list_front(&facet->subfacets),
4728 struct subfacet, list_node);
4729 long long int now = time_msec();
4730 struct action_xlate_ctx ctx;
4732 if (!facet->has_fin_timeout && now < facet->learn_rl) {
4736 facet->learn_rl = now + 500;
4738 if (!facet->has_learn
4739 && !facet->has_normal
4740 && (!facet->has_fin_timeout
4741 || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
4745 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
4746 &subfacet->initial_vals,
4747 facet->rule, facet->tcp_flags, NULL);
4748 ctx.may_learn = true;
4749 xlate_actions_for_side_effects(&ctx, facet->rule->up.ofpacts,
4750 facet->rule->up.ofpacts_len);
4754 facet_account(struct facet *facet)
4756 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4757 struct subfacet *subfacet = facet_get_subfacet(facet);
4758 const struct nlattr *a;
4763 if (!facet->has_normal || !ofproto->has_bonded_bundles) {
4766 n_bytes = facet->byte_count - facet->accounted_bytes;
4768 /* This loop feeds byte counters to bond_account() for rebalancing to use
4769 * as a basis. We also need to track the actual VLAN on which the packet
4770 * is going to be sent to ensure that it matches the one passed to
4771 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
4774 * We use the actions from an arbitrary subfacet because they should all
4775 * be equally valid for our purpose. */
4776 vlan_tci = facet->flow.vlan_tci;
4777 NL_ATTR_FOR_EACH_UNSAFE (a, left,
4778 subfacet->actions, subfacet->actions_len) {
4779 const struct ovs_action_push_vlan *vlan;
4780 struct ofport_dpif *port;
4782 switch (nl_attr_type(a)) {
4783 case OVS_ACTION_ATTR_OUTPUT:
4784 port = get_odp_port(ofproto, nl_attr_get_u32(a));
4785 if (port && port->bundle && port->bundle->bond) {
4786 bond_account(port->bundle->bond, &facet->flow,
4787 vlan_tci_to_vid(vlan_tci), n_bytes);
4791 case OVS_ACTION_ATTR_POP_VLAN:
4792 vlan_tci = htons(0);
4795 case OVS_ACTION_ATTR_PUSH_VLAN:
4796 vlan = nl_attr_get(a);
4797 vlan_tci = vlan->vlan_tci;
4803 /* Returns true if the only action for 'facet' is to send to the controller.
4804 * (We don't report NetFlow expiration messages for such facets because they
4805 * are just part of the control logic for the network, not real traffic). */
4807 facet_is_controller_flow(struct facet *facet)
4810 const struct rule *rule = &facet->rule->up;
4811 const struct ofpact *ofpacts = rule->ofpacts;
4812 size_t ofpacts_len = rule->ofpacts_len;
4814 if (ofpacts_len > 0 &&
4815 ofpacts->type == OFPACT_CONTROLLER &&
4816 ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len)) {
4823 /* Folds all of 'facet''s statistics into its rule. Also updates the
4824 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
4825 * 'facet''s statistics in the datapath should have been zeroed and folded into
4826 * its packet and byte counts before this function is called. */
4828 facet_flush_stats(struct facet *facet)
4830 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4831 struct subfacet *subfacet;
4833 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4834 ovs_assert(!subfacet->dp_byte_count);
4835 ovs_assert(!subfacet->dp_packet_count);
4838 facet_push_stats(facet);
4839 if (facet->accounted_bytes < facet->byte_count) {
4840 facet_account(facet);
4841 facet->accounted_bytes = facet->byte_count;
4844 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
4845 struct ofexpired expired;
4846 expired.flow = facet->flow;
4847 expired.packet_count = facet->packet_count;
4848 expired.byte_count = facet->byte_count;
4849 expired.used = facet->used;
4850 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4853 /* Reset counters to prevent double counting if 'facet' ever gets
4855 facet_reset_counters(facet);
4857 netflow_flow_clear(&facet->nf_flow);
4858 facet->tcp_flags = 0;
4861 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
4862 * Returns it if found, otherwise a null pointer.
4864 * 'hash' must be the return value of flow_hash(flow, 0).
4866 * The returned facet might need revalidation; use facet_lookup_valid()
4867 * instead if that is important. */
4868 static struct facet *
4869 facet_find(struct ofproto_dpif *ofproto,
4870 const struct flow *flow, uint32_t hash)
4872 struct facet *facet;
4874 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, hash, &ofproto->facets) {
4875 if (flow_equal(flow, &facet->flow)) {
4883 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
4884 * Returns it if found, otherwise a null pointer.
4886 * 'hash' must be the return value of flow_hash(flow, 0).
4888 * The returned facet is guaranteed to be valid. */
4889 static struct facet *
4890 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow,
4893 struct facet *facet;
4895 facet = facet_find(ofproto, flow, hash);
4897 && (ofproto->backer->need_revalidate
4898 || tag_set_intersects(&ofproto->backer->revalidate_set,
4900 && !facet_revalidate(facet)) {
4901 facet = facet_find(ofproto, flow, hash);
4907 /* Return a subfacet from 'facet'. A facet consists of one or more
4908 * subfacets, and this function returns one of them. */
4909 static struct subfacet *facet_get_subfacet(struct facet *facet)
4911 return CONTAINER_OF(list_front(&facet->subfacets), struct subfacet,
4916 subfacet_path_to_string(enum subfacet_path path)
4919 case SF_NOT_INSTALLED:
4920 return "not installed";
4922 return "in fast path";
4924 return "in slow path";
4930 /* Returns the path in which a subfacet should be installed if its 'slow'
4931 * member has the specified value. */
4932 static enum subfacet_path
4933 subfacet_want_path(enum slow_path_reason slow)
4935 return slow ? SF_SLOW_PATH : SF_FAST_PATH;
4938 /* Returns true if 'subfacet' needs to have its datapath flow updated,
4939 * supposing that its actions have been recalculated as 'want_actions' and that
4940 * 'slow' is nonzero iff 'subfacet' should be in the slow path. */
4942 subfacet_should_install(struct subfacet *subfacet, enum slow_path_reason slow,
4943 const struct ofpbuf *want_actions)
4945 enum subfacet_path want_path = subfacet_want_path(slow);
4946 return (want_path != subfacet->path
4947 || (want_path == SF_FAST_PATH
4948 && (subfacet->actions_len != want_actions->size
4949 || memcmp(subfacet->actions, want_actions->data,
4950 subfacet->actions_len))));
4954 facet_check_consistency(struct facet *facet)
4956 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
4958 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4960 uint64_t odp_actions_stub[1024 / 8];
4961 struct ofpbuf odp_actions;
4963 struct rule_dpif *rule;
4964 struct subfacet *subfacet;
4965 bool may_log = false;
4968 /* Check the rule for consistency. */
4969 rule = rule_dpif_lookup(ofproto, &facet->flow);
4970 ok = rule == facet->rule;
4972 may_log = !VLOG_DROP_WARN(&rl);
4977 flow_format(&s, &facet->flow);
4978 ds_put_format(&s, ": facet associated with wrong rule (was "
4979 "table=%"PRIu8",", facet->rule->up.table_id);
4980 cls_rule_format(&facet->rule->up.cr, &s);
4981 ds_put_format(&s, ") (should have been table=%"PRIu8",",
4983 cls_rule_format(&rule->up.cr, &s);
4984 ds_put_char(&s, ')');
4986 VLOG_WARN("%s", ds_cstr(&s));
4991 /* Check the datapath actions for consistency. */
4992 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
4993 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4994 enum subfacet_path want_path;
4995 struct action_xlate_ctx ctx;
4998 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
4999 &subfacet->initial_vals, rule, 0, NULL);
5000 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len,
5003 if (subfacet->path == SF_NOT_INSTALLED) {
5004 /* This only happens if the datapath reported an error when we
5005 * tried to install the flow. Don't flag another error here. */
5009 want_path = subfacet_want_path(subfacet->slow);
5011 if (!subfacet_should_install(subfacet, subfacet->slow, &odp_actions)) {
5015 /* Inconsistency! */
5017 may_log = !VLOG_DROP_WARN(&rl);
5021 /* Rate-limited, skip reporting. */
5026 odp_flow_key_format(subfacet->key, subfacet->key_len, &s);
5028 ds_put_cstr(&s, ": inconsistency in subfacet");
5029 if (want_path != subfacet->path) {
5030 enum odp_key_fitness fitness = subfacet->key_fitness;
5032 ds_put_format(&s, " (%s, fitness=%s)",
5033 subfacet_path_to_string(subfacet->path),
5034 odp_key_fitness_to_string(fitness));
5035 ds_put_format(&s, " (should have been %s)",
5036 subfacet_path_to_string(want_path));
5037 } else if (want_path == SF_FAST_PATH) {
5038 ds_put_cstr(&s, " (actions were: ");
5039 format_odp_actions(&s, subfacet->actions,
5040 subfacet->actions_len);
5041 ds_put_cstr(&s, ") (correct actions: ");
5042 format_odp_actions(&s, odp_actions.data, odp_actions.size);
5043 ds_put_char(&s, ')');
5045 ds_put_cstr(&s, " (actions: ");
5046 format_odp_actions(&s, subfacet->actions,
5047 subfacet->actions_len);
5048 ds_put_char(&s, ')');
5050 VLOG_WARN("%s", ds_cstr(&s));
5053 ofpbuf_uninit(&odp_actions);
5058 /* Re-searches the classifier for 'facet':
5060 * - If the rule found is different from 'facet''s current rule, moves
5061 * 'facet' to the new rule and recompiles its actions.
5063 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
5064 * where it is and recompiles its actions anyway.
5066 * - If any of 'facet''s subfacets correspond to a new flow according to
5067 * ofproto_receive(), 'facet' is removed.
5069 * Returns true if 'facet' is still valid. False if 'facet' was removed. */
5071 facet_revalidate(struct facet *facet)
5073 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5075 struct nlattr *odp_actions;
5078 struct actions *new_actions;
5080 struct action_xlate_ctx ctx;
5081 uint64_t odp_actions_stub[1024 / 8];
5082 struct ofpbuf odp_actions;
5084 struct rule_dpif *new_rule;
5085 struct subfacet *subfacet;
5088 COVERAGE_INC(facet_revalidate);
5090 /* Check that child subfacets still correspond to this facet. Tunnel
5091 * configuration changes could cause a subfacet's OpenFlow in_port to
5093 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
5094 struct ofproto_dpif *recv_ofproto;
5095 struct flow recv_flow;
5098 error = ofproto_receive(ofproto->backer, NULL, subfacet->key,
5099 subfacet->key_len, &recv_flow, NULL,
5100 &recv_ofproto, NULL, NULL);
5102 || recv_ofproto != ofproto
5103 || memcmp(&recv_flow, &facet->flow, sizeof recv_flow)) {
5104 facet_remove(facet);
5109 new_rule = rule_dpif_lookup(ofproto, &facet->flow);
5111 /* Calculate new datapath actions.
5113 * We do not modify any 'facet' state yet, because we might need to, e.g.,
5114 * emit a NetFlow expiration and, if so, we need to have the old state
5115 * around to properly compose it. */
5117 /* If the datapath actions changed or the installability changed,
5118 * then we need to talk to the datapath. */
5121 memset(&ctx, 0, sizeof ctx);
5122 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5123 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
5124 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
5125 &subfacet->initial_vals, new_rule, 0, NULL);
5126 xlate_actions(&ctx, new_rule->up.ofpacts, new_rule->up.ofpacts_len,
5129 if (subfacet_should_install(subfacet, ctx.slow, &odp_actions)) {
5130 struct dpif_flow_stats stats;
5132 subfacet_install(subfacet, odp_actions.data, odp_actions.size,
5134 subfacet_update_stats(subfacet, &stats);
5137 new_actions = xcalloc(list_size(&facet->subfacets),
5138 sizeof *new_actions);
5140 new_actions[i].odp_actions = xmemdup(odp_actions.data,
5142 new_actions[i].actions_len = odp_actions.size;
5147 ofpbuf_uninit(&odp_actions);
5150 facet_flush_stats(facet);
5153 /* Update 'facet' now that we've taken care of all the old state. */
5154 facet->tags = ctx.tags;
5155 facet->nf_flow.output_iface = ctx.nf_output_iface;
5156 facet->has_learn = ctx.has_learn;
5157 facet->has_normal = ctx.has_normal;
5158 facet->has_fin_timeout = ctx.has_fin_timeout;
5159 facet->mirrors = ctx.mirrors;
5162 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
5163 subfacet->slow = ctx.slow;
5165 if (new_actions && new_actions[i].odp_actions) {
5166 free(subfacet->actions);
5167 subfacet->actions = new_actions[i].odp_actions;
5168 subfacet->actions_len = new_actions[i].actions_len;
5174 if (facet->rule != new_rule) {
5175 COVERAGE_INC(facet_changed_rule);
5176 list_remove(&facet->list_node);
5177 list_push_back(&new_rule->facets, &facet->list_node);
5178 facet->rule = new_rule;
5179 facet->used = new_rule->up.created;
5180 facet->prev_used = facet->used;
5186 /* Updates 'facet''s used time. Caller is responsible for calling
5187 * facet_push_stats() to update the flows which 'facet' resubmits into. */
5189 facet_update_time(struct facet *facet, long long int used)
5191 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5192 if (used > facet->used) {
5194 ofproto_rule_update_used(&facet->rule->up, used);
5195 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
5200 facet_reset_counters(struct facet *facet)
5202 facet->packet_count = 0;
5203 facet->byte_count = 0;
5204 facet->prev_packet_count = 0;
5205 facet->prev_byte_count = 0;
5206 facet->accounted_bytes = 0;
5210 facet_push_stats(struct facet *facet)
5212 struct dpif_flow_stats stats;
5214 ovs_assert(facet->packet_count >= facet->prev_packet_count);
5215 ovs_assert(facet->byte_count >= facet->prev_byte_count);
5216 ovs_assert(facet->used >= facet->prev_used);
5218 stats.n_packets = facet->packet_count - facet->prev_packet_count;
5219 stats.n_bytes = facet->byte_count - facet->prev_byte_count;
5220 stats.used = facet->used;
5221 stats.tcp_flags = 0;
5223 if (stats.n_packets || stats.n_bytes || facet->used > facet->prev_used) {
5224 facet->prev_packet_count = facet->packet_count;
5225 facet->prev_byte_count = facet->byte_count;
5226 facet->prev_used = facet->used;
5228 rule_credit_stats(facet->rule, &stats);
5229 flow_push_stats(facet, &stats);
5231 update_mirror_stats(ofproto_dpif_cast(facet->rule->up.ofproto),
5232 facet->mirrors, stats.n_packets, stats.n_bytes);
5237 push_all_stats__(bool run_fast)
5239 static long long int rl = LLONG_MIN;
5240 struct ofproto_dpif *ofproto;
5242 if (time_msec() < rl) {
5246 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5247 struct facet *facet;
5249 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
5250 facet_push_stats(facet);
5257 rl = time_msec() + 100;
5261 push_all_stats(void)
5263 push_all_stats__(true);
5267 rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats)
5269 rule->packet_count += stats->n_packets;
5270 rule->byte_count += stats->n_bytes;
5271 ofproto_rule_update_used(&rule->up, stats->used);
5274 /* Pushes flow statistics to the rules which 'facet->flow' resubmits
5275 * into given 'facet->rule''s actions and mirrors. */
5277 flow_push_stats(struct facet *facet, const struct dpif_flow_stats *stats)
5279 struct rule_dpif *rule = facet->rule;
5280 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5281 struct subfacet *subfacet = facet_get_subfacet(facet);
5282 struct action_xlate_ctx ctx;
5284 ofproto_rule_update_used(&rule->up, stats->used);
5286 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
5287 &subfacet->initial_vals, rule, 0, NULL);
5288 ctx.resubmit_stats = stats;
5289 xlate_actions_for_side_effects(&ctx, rule->up.ofpacts,
5290 rule->up.ofpacts_len);
5295 static struct subfacet *
5296 subfacet_find(struct ofproto_dpif *ofproto,
5297 const struct nlattr *key, size_t key_len, uint32_t key_hash)
5299 struct subfacet *subfacet;
5301 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
5302 &ofproto->subfacets) {
5303 if (subfacet->key_len == key_len
5304 && !memcmp(key, subfacet->key, key_len)) {
5312 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
5313 * 'key_fitness', 'key', and 'key_len' members in 'miss'. Returns the
5314 * existing subfacet if there is one, otherwise creates and returns a
5317 * If the returned subfacet is new, then subfacet->actions will be NULL, in
5318 * which case the caller must populate the actions with
5319 * subfacet_make_actions(). */
5320 static struct subfacet *
5321 subfacet_create(struct facet *facet, struct flow_miss *miss,
5324 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5325 enum odp_key_fitness key_fitness = miss->key_fitness;
5326 const struct nlattr *key = miss->key;
5327 size_t key_len = miss->key_len;
5329 struct subfacet *subfacet;
5331 key_hash = odp_flow_key_hash(key, key_len);
5333 if (list_is_empty(&facet->subfacets)) {
5334 subfacet = &facet->one_subfacet;
5336 subfacet = subfacet_find(ofproto, key, key_len, key_hash);
5338 if (subfacet->facet == facet) {
5342 /* This shouldn't happen. */
5343 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
5344 subfacet_destroy(subfacet);
5347 subfacet = xmalloc(sizeof *subfacet);
5350 hmap_insert(&ofproto->subfacets, &subfacet->hmap_node, key_hash);
5351 list_push_back(&facet->subfacets, &subfacet->list_node);
5352 subfacet->facet = facet;
5353 subfacet->key_fitness = key_fitness;
5354 subfacet->key = xmemdup(key, key_len);
5355 subfacet->key_len = key_len;
5356 subfacet->used = now;
5357 subfacet->created = now;
5358 subfacet->dp_packet_count = 0;
5359 subfacet->dp_byte_count = 0;
5360 subfacet->actions_len = 0;
5361 subfacet->actions = NULL;
5363 subfacet->path = SF_NOT_INSTALLED;
5364 subfacet->initial_vals = miss->initial_vals;
5365 subfacet->odp_in_port = miss->odp_in_port;
5367 ofproto->subfacet_add_count++;
5371 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
5372 * its facet within 'ofproto', and frees it. */
5374 subfacet_destroy__(struct subfacet *subfacet)
5376 struct facet *facet = subfacet->facet;
5377 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5379 /* Update ofproto stats before uninstall the subfacet. */
5380 ofproto->subfacet_del_count++;
5381 ofproto->total_subfacet_life_span += (time_msec() - subfacet->created);
5383 subfacet_uninstall(subfacet);
5384 hmap_remove(&ofproto->subfacets, &subfacet->hmap_node);
5385 list_remove(&subfacet->list_node);
5386 free(subfacet->key);
5387 free(subfacet->actions);
5388 if (subfacet != &facet->one_subfacet) {
5393 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
5394 * last remaining subfacet in its facet destroys the facet too. */
5396 subfacet_destroy(struct subfacet *subfacet)
5398 struct facet *facet = subfacet->facet;
5400 if (list_is_singleton(&facet->subfacets)) {
5401 /* facet_remove() needs at least one subfacet (it will remove it). */
5402 facet_remove(facet);
5404 subfacet_destroy__(subfacet);
5409 subfacet_destroy_batch(struct ofproto_dpif *ofproto,
5410 struct subfacet **subfacets, int n)
5412 struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH];
5413 struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH];
5414 struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH];
5417 for (i = 0; i < n; i++) {
5418 ops[i].type = DPIF_OP_FLOW_DEL;
5419 ops[i].u.flow_del.key = subfacets[i]->key;
5420 ops[i].u.flow_del.key_len = subfacets[i]->key_len;
5421 ops[i].u.flow_del.stats = &stats[i];
5425 dpif_operate(ofproto->backer->dpif, opsp, n);
5426 for (i = 0; i < n; i++) {
5427 subfacet_reset_dp_stats(subfacets[i], &stats[i]);
5428 subfacets[i]->path = SF_NOT_INSTALLED;
5429 subfacet_destroy(subfacets[i]);
5434 /* Composes the datapath actions for 'subfacet' based on its rule's actions. */
5436 subfacet_make_actions(struct subfacet *subfacet, const struct ofpbuf *packet)
5438 struct facet *facet = subfacet->facet;
5439 struct rule_dpif *rule = facet->rule;
5440 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5442 struct action_xlate_ctx ctx;
5443 struct ofpbuf odp_actions;
5444 uint64_t stub[1024 / 8];
5446 ofpbuf_use_stub(&odp_actions, stub, sizeof stub);
5447 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
5448 &subfacet->initial_vals, rule, 0, packet);
5449 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, &odp_actions);
5450 facet->tags = ctx.tags;
5451 facet->has_learn = ctx.has_learn;
5452 facet->has_normal = ctx.has_normal;
5453 facet->has_fin_timeout = ctx.has_fin_timeout;
5454 facet->nf_flow.output_iface = ctx.nf_output_iface;
5455 facet->mirrors = ctx.mirrors;
5457 subfacet->slow = ctx.slow;
5459 ovs_assert(!subfacet->actions);
5460 subfacet->actions_len = odp_actions.size;
5461 subfacet->actions = ofpbuf_steal_data(&odp_actions);
5464 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
5465 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
5466 * in the datapath will be zeroed and 'stats' will be updated with traffic new
5467 * since 'subfacet' was last updated.
5469 * Returns 0 if successful, otherwise a positive errno value. */
5471 subfacet_install(struct subfacet *subfacet,
5472 const struct nlattr *actions, size_t actions_len,
5473 struct dpif_flow_stats *stats,
5474 enum slow_path_reason slow)
5476 struct facet *facet = subfacet->facet;
5477 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5478 enum subfacet_path path = subfacet_want_path(slow);
5479 uint64_t slow_path_stub[128 / 8];
5480 enum dpif_flow_put_flags flags;
5483 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
5485 flags |= DPIF_FP_ZERO_STATS;
5488 if (path == SF_SLOW_PATH) {
5489 compose_slow_path(ofproto, &facet->flow, slow,
5490 slow_path_stub, sizeof slow_path_stub,
5491 &actions, &actions_len);
5494 ret = dpif_flow_put(ofproto->backer->dpif, flags, subfacet->key,
5495 subfacet->key_len, actions, actions_len, stats);
5498 subfacet_reset_dp_stats(subfacet, stats);
5502 subfacet->path = path;
5508 subfacet_reinstall(struct subfacet *subfacet, struct dpif_flow_stats *stats)
5510 return subfacet_install(subfacet, subfacet->actions, subfacet->actions_len,
5511 stats, subfacet->slow);
5514 /* If 'subfacet' is installed in the datapath, uninstalls it. */
5516 subfacet_uninstall(struct subfacet *subfacet)
5518 if (subfacet->path != SF_NOT_INSTALLED) {
5519 struct rule_dpif *rule = subfacet->facet->rule;
5520 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5521 struct dpif_flow_stats stats;
5524 error = dpif_flow_del(ofproto->backer->dpif, subfacet->key,
5525 subfacet->key_len, &stats);
5526 subfacet_reset_dp_stats(subfacet, &stats);
5528 subfacet_update_stats(subfacet, &stats);
5530 subfacet->path = SF_NOT_INSTALLED;
5532 ovs_assert(subfacet->dp_packet_count == 0);
5533 ovs_assert(subfacet->dp_byte_count == 0);
5537 /* Resets 'subfacet''s datapath statistics counters. This should be called
5538 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
5539 * non-null, it should contain the statistics returned by dpif when 'subfacet'
5540 * was reset in the datapath. 'stats' will be modified to include only
5541 * statistics new since 'subfacet' was last updated. */
5543 subfacet_reset_dp_stats(struct subfacet *subfacet,
5544 struct dpif_flow_stats *stats)
5547 && subfacet->dp_packet_count <= stats->n_packets
5548 && subfacet->dp_byte_count <= stats->n_bytes) {
5549 stats->n_packets -= subfacet->dp_packet_count;
5550 stats->n_bytes -= subfacet->dp_byte_count;
5553 subfacet->dp_packet_count = 0;
5554 subfacet->dp_byte_count = 0;
5557 /* Updates 'subfacet''s used time. The caller is responsible for calling
5558 * facet_push_stats() to update the flows which 'subfacet' resubmits into. */
5560 subfacet_update_time(struct subfacet *subfacet, long long int used)
5562 if (used > subfacet->used) {
5563 subfacet->used = used;
5564 facet_update_time(subfacet->facet, used);
5568 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
5570 * Because of the meaning of a subfacet's counters, it only makes sense to do
5571 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
5572 * represents a packet that was sent by hand or if it represents statistics
5573 * that have been cleared out of the datapath. */
5575 subfacet_update_stats(struct subfacet *subfacet,
5576 const struct dpif_flow_stats *stats)
5578 if (stats->n_packets || stats->used > subfacet->used) {
5579 struct facet *facet = subfacet->facet;
5581 subfacet_update_time(subfacet, stats->used);
5582 facet->packet_count += stats->n_packets;
5583 facet->byte_count += stats->n_bytes;
5584 facet->tcp_flags |= stats->tcp_flags;
5585 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
5591 static struct rule_dpif *
5592 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow)
5594 struct rule_dpif *rule;
5596 rule = rule_dpif_lookup__(ofproto, flow, 0);
5601 return rule_dpif_miss_rule(ofproto, flow);
5604 static struct rule_dpif *
5605 rule_dpif_lookup__(struct ofproto_dpif *ofproto, const struct flow *flow,
5608 struct cls_rule *cls_rule;
5609 struct classifier *cls;
5611 if (table_id >= N_TABLES) {
5615 cls = &ofproto->up.tables[table_id].cls;
5616 if (flow->nw_frag & FLOW_NW_FRAG_ANY
5617 && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
5618 /* For OFPC_NORMAL frag_handling, we must pretend that transport ports
5619 * are unavailable. */
5620 struct flow ofpc_normal_flow = *flow;
5621 ofpc_normal_flow.tp_src = htons(0);
5622 ofpc_normal_flow.tp_dst = htons(0);
5623 cls_rule = classifier_lookup(cls, &ofpc_normal_flow);
5625 cls_rule = classifier_lookup(cls, flow);
5627 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
5630 static struct rule_dpif *
5631 rule_dpif_miss_rule(struct ofproto_dpif *ofproto, const struct flow *flow)
5633 struct ofport_dpif *port;
5635 port = get_ofp_port(ofproto, flow->in_port);
5637 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, flow->in_port);
5638 return ofproto->miss_rule;
5641 if (port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN) {
5642 return ofproto->no_packet_in_rule;
5644 return ofproto->miss_rule;
5648 complete_operation(struct rule_dpif *rule)
5650 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5652 rule_invalidate(rule);
5654 struct dpif_completion *c = xmalloc(sizeof *c);
5655 c->op = rule->up.pending;
5656 list_push_back(&ofproto->completions, &c->list_node);
5658 ofoperation_complete(rule->up.pending, 0);
5662 static struct rule *
5665 struct rule_dpif *rule = xmalloc(sizeof *rule);
5670 rule_dealloc(struct rule *rule_)
5672 struct rule_dpif *rule = rule_dpif_cast(rule_);
5677 rule_construct(struct rule *rule_)
5679 struct rule_dpif *rule = rule_dpif_cast(rule_);
5680 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5681 struct rule_dpif *victim;
5684 rule->packet_count = 0;
5685 rule->byte_count = 0;
5687 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
5688 if (victim && !list_is_empty(&victim->facets)) {
5689 struct facet *facet;
5691 rule->facets = victim->facets;
5692 list_moved(&rule->facets);
5693 LIST_FOR_EACH (facet, list_node, &rule->facets) {
5694 /* XXX: We're only clearing our local counters here. It's possible
5695 * that quite a few packets are unaccounted for in the datapath
5696 * statistics. These will be accounted to the new rule instead of
5697 * cleared as required. This could be fixed by clearing out the
5698 * datapath statistics for this facet, but currently it doesn't
5700 facet_reset_counters(facet);
5704 /* Must avoid list_moved() in this case. */
5705 list_init(&rule->facets);
5708 table_id = rule->up.table_id;
5710 rule->tag = victim->tag;
5711 } else if (table_id == 0) {
5716 miniflow_expand(&rule->up.cr.match.flow, &flow);
5717 rule->tag = rule_calculate_tag(&flow, &rule->up.cr.match.mask,
5718 ofproto->tables[table_id].basis);
5721 complete_operation(rule);
5726 rule_destruct(struct rule *rule_)
5728 struct rule_dpif *rule = rule_dpif_cast(rule_);
5729 struct facet *facet, *next_facet;
5731 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
5732 facet_revalidate(facet);
5735 complete_operation(rule);
5739 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
5741 struct rule_dpif *rule = rule_dpif_cast(rule_);
5743 /* push_all_stats() can handle flow misses which, when using the learn
5744 * action, can cause rules to be added and deleted. This can corrupt our
5745 * caller's datastructures which assume that rule_get_stats() doesn't have
5746 * an impact on the flow table. To be safe, we disable miss handling. */
5747 push_all_stats__(false);
5749 /* Start from historical data for 'rule' itself that are no longer tracked
5750 * in facets. This counts, for example, facets that have expired. */
5751 *packets = rule->packet_count;
5752 *bytes = rule->byte_count;
5756 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
5757 struct ofpbuf *packet)
5759 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5760 struct initial_vals initial_vals;
5761 struct dpif_flow_stats stats;
5762 struct action_xlate_ctx ctx;
5763 uint64_t odp_actions_stub[1024 / 8];
5764 struct ofpbuf odp_actions;
5766 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
5767 rule_credit_stats(rule, &stats);
5769 initial_vals.vlan_tci = flow->vlan_tci;
5770 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5771 action_xlate_ctx_init(&ctx, ofproto, flow, &initial_vals,
5772 rule, stats.tcp_flags, packet);
5773 ctx.resubmit_stats = &stats;
5774 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, &odp_actions);
5776 execute_odp_actions(ofproto, flow, odp_actions.data,
5777 odp_actions.size, packet);
5779 ofpbuf_uninit(&odp_actions);
5783 rule_execute(struct rule *rule, const struct flow *flow,
5784 struct ofpbuf *packet)
5786 rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
5787 ofpbuf_delete(packet);
5792 rule_modify_actions(struct rule *rule_)
5794 struct rule_dpif *rule = rule_dpif_cast(rule_);
5796 complete_operation(rule);
5799 /* Sends 'packet' out 'ofport'.
5800 * May modify 'packet'.
5801 * Returns 0 if successful, otherwise a positive errno value. */
5803 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
5805 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
5806 uint64_t odp_actions_stub[1024 / 8];
5807 struct ofpbuf key, odp_actions;
5808 struct dpif_flow_stats stats;
5809 struct odputil_keybuf keybuf;
5810 struct ofpact_output output;
5811 struct action_xlate_ctx ctx;
5815 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5816 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5818 /* Use OFPP_NONE as the in_port to avoid special packet processing. */
5819 flow_extract(packet, 0, 0, NULL, OFPP_NONE, &flow);
5820 odp_flow_key_from_flow(&key, &flow, ofp_port_to_odp_port(ofproto,
5822 dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
5824 ofpact_init(&output.ofpact, OFPACT_OUTPUT, sizeof output);
5825 output.port = ofport->up.ofp_port;
5828 action_xlate_ctx_init(&ctx, ofproto, &flow, NULL, NULL, 0, packet);
5829 ctx.resubmit_stats = &stats;
5830 xlate_actions(&ctx, &output.ofpact, sizeof output, &odp_actions);
5832 error = dpif_execute(ofproto->backer->dpif,
5834 odp_actions.data, odp_actions.size,
5836 ofpbuf_uninit(&odp_actions);
5839 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %s (%s)",
5840 ofproto->up.name, netdev_get_name(ofport->up.netdev),
5844 ofproto->stats.tx_packets++;
5845 ofproto->stats.tx_bytes += packet->size;
5849 /* OpenFlow to datapath action translation. */
5851 static bool may_receive(const struct ofport_dpif *, struct action_xlate_ctx *);
5852 static void do_xlate_actions(const struct ofpact *, size_t ofpacts_len,
5853 struct action_xlate_ctx *);
5854 static void xlate_normal(struct action_xlate_ctx *);
5856 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
5857 * The action will state 'slow' as the reason that the action is in the slow
5858 * path. (This is purely informational: it allows a human viewing "ovs-dpctl
5859 * dump-flows" output to see why a flow is in the slow path.)
5861 * The 'stub_size' bytes in 'stub' will be used to store the action.
5862 * 'stub_size' must be large enough for the action.
5864 * The action and its size will be stored in '*actionsp' and '*actions_lenp',
5867 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
5868 enum slow_path_reason slow,
5869 uint64_t *stub, size_t stub_size,
5870 const struct nlattr **actionsp, size_t *actions_lenp)
5872 union user_action_cookie cookie;
5875 cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
5876 cookie.slow_path.unused = 0;
5877 cookie.slow_path.reason = slow;
5879 ofpbuf_use_stack(&buf, stub, stub_size);
5880 if (slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP | SLOW_STP)) {
5881 uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif, UINT32_MAX);
5882 odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
5884 put_userspace_action(ofproto, &buf, flow, &cookie,
5885 sizeof cookie.slow_path);
5887 *actionsp = buf.data;
5888 *actions_lenp = buf.size;
5892 put_userspace_action(const struct ofproto_dpif *ofproto,
5893 struct ofpbuf *odp_actions,
5894 const struct flow *flow,
5895 const union user_action_cookie *cookie,
5896 const size_t cookie_size)
5900 pid = dpif_port_get_pid(ofproto->backer->dpif,
5901 ofp_port_to_odp_port(ofproto, flow->in_port));
5903 return odp_put_userspace_action(pid, cookie, cookie_size, odp_actions);
5906 /* Compose SAMPLE action for sFlow or IPFIX. The given probability is
5907 * the number of packets out of UINT32_MAX to sample. The given
5908 * cookie is passed back in the callback for each sampled packet.
5911 compose_sample_action(const struct ofproto_dpif *ofproto,
5912 struct ofpbuf *odp_actions,
5913 const struct flow *flow,
5914 const uint32_t probability,
5915 const union user_action_cookie *cookie,
5916 const size_t cookie_size)
5918 size_t sample_offset, actions_offset;
5921 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
5923 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
5925 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
5926 cookie_offset = put_userspace_action(ofproto, odp_actions, flow, cookie,
5929 nl_msg_end_nested(odp_actions, actions_offset);
5930 nl_msg_end_nested(odp_actions, sample_offset);
5931 return cookie_offset;
5935 compose_sflow_cookie(const struct ofproto_dpif *ofproto,
5936 ovs_be16 vlan_tci, uint32_t odp_port,
5937 unsigned int n_outputs, union user_action_cookie *cookie)
5941 cookie->type = USER_ACTION_COOKIE_SFLOW;
5942 cookie->sflow.vlan_tci = vlan_tci;
5944 /* See http://www.sflow.org/sflow_version_5.txt (search for "Input/output
5945 * port information") for the interpretation of cookie->output. */
5946 switch (n_outputs) {
5948 /* 0x40000000 | 256 means "packet dropped for unknown reason". */
5949 cookie->sflow.output = 0x40000000 | 256;
5953 ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port);
5955 cookie->sflow.output = ifindex;
5960 /* 0x80000000 means "multiple output ports. */
5961 cookie->sflow.output = 0x80000000 | n_outputs;
5966 /* Compose SAMPLE action for sFlow bridge sampling. */
5968 compose_sflow_action(const struct ofproto_dpif *ofproto,
5969 struct ofpbuf *odp_actions,
5970 const struct flow *flow,
5973 uint32_t probability;
5974 union user_action_cookie cookie;
5976 if (!ofproto->sflow || flow->in_port == OFPP_NONE) {
5980 probability = dpif_sflow_get_probability(ofproto->sflow);
5981 compose_sflow_cookie(ofproto, htons(0), odp_port,
5982 odp_port == OVSP_NONE ? 0 : 1, &cookie);
5984 return compose_sample_action(ofproto, odp_actions, flow, probability,
5985 &cookie, sizeof cookie.sflow);
5989 compose_flow_sample_cookie(uint16_t probability, uint32_t collector_set_id,
5990 uint32_t obs_domain_id, uint32_t obs_point_id,
5991 union user_action_cookie *cookie)
5993 cookie->type = USER_ACTION_COOKIE_FLOW_SAMPLE;
5994 cookie->flow_sample.probability = probability;
5995 cookie->flow_sample.collector_set_id = collector_set_id;
5996 cookie->flow_sample.obs_domain_id = obs_domain_id;
5997 cookie->flow_sample.obs_point_id = obs_point_id;
6001 compose_ipfix_cookie(union user_action_cookie *cookie)
6003 cookie->type = USER_ACTION_COOKIE_IPFIX;
6006 /* Compose SAMPLE action for IPFIX bridge sampling. */
6008 compose_ipfix_action(const struct ofproto_dpif *ofproto,
6009 struct ofpbuf *odp_actions,
6010 const struct flow *flow)
6012 uint32_t probability;
6013 union user_action_cookie cookie;
6015 if (!ofproto->ipfix || flow->in_port == OFPP_NONE) {
6019 probability = dpif_ipfix_get_bridge_exporter_probability(ofproto->ipfix);
6020 compose_ipfix_cookie(&cookie);
6022 compose_sample_action(ofproto, odp_actions, flow, probability,
6023 &cookie, sizeof cookie.ipfix);
6026 /* SAMPLE action for sFlow must be first action in any given list of
6027 * actions. At this point we do not have all information required to
6028 * build it. So try to build sample action as complete as possible. */
6030 add_sflow_action(struct action_xlate_ctx *ctx)
6032 ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto,
6034 &ctx->flow, OVSP_NONE);
6035 ctx->sflow_odp_port = 0;
6036 ctx->sflow_n_outputs = 0;
6039 /* SAMPLE action for IPFIX must be 1st or 2nd action in any given list
6040 * of actions, eventually after the SAMPLE action for sFlow. */
6042 add_ipfix_action(struct action_xlate_ctx *ctx)
6044 compose_ipfix_action(ctx->ofproto, ctx->odp_actions, &ctx->flow);
6047 /* Fix SAMPLE action according to data collected while composing ODP actions.
6048 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
6049 * USERSPACE action's user-cookie which is required for sflow. */
6051 fix_sflow_action(struct action_xlate_ctx *ctx)
6053 const struct flow *base = &ctx->base_flow;
6054 union user_action_cookie *cookie;
6056 if (!ctx->user_cookie_offset) {
6060 cookie = ofpbuf_at(ctx->odp_actions, ctx->user_cookie_offset,
6061 sizeof cookie->sflow);
6062 ovs_assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
6064 compose_sflow_cookie(ctx->ofproto, base->vlan_tci,
6065 ctx->sflow_odp_port, ctx->sflow_n_outputs, cookie);
6069 compose_output_action__(struct action_xlate_ctx *ctx, uint16_t ofp_port,
6072 const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
6073 ovs_be16 flow_vlan_tci;
6074 uint32_t flow_skb_mark;
6075 uint8_t flow_nw_tos;
6076 struct priority_to_dscp *pdscp;
6077 uint32_t out_port, odp_port;
6079 /* If 'struct flow' gets additional metadata, we'll need to zero it out
6080 * before traversing a patch port. */
6081 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 20);
6084 xlate_report(ctx, "Nonexistent output port");
6086 } else if (ofport->up.pp.config & OFPUTIL_PC_NO_FWD) {
6087 xlate_report(ctx, "OFPPC_NO_FWD set, skipping output");
6089 } else if (check_stp && !stp_forward_in_state(ofport->stp_state)) {
6090 xlate_report(ctx, "STP not in forwarding state, skipping output");
6094 if (netdev_vport_is_patch(ofport->up.netdev)) {
6095 struct ofport_dpif *peer = ofport_get_peer(ofport);
6096 struct flow old_flow = ctx->flow;
6097 const struct ofproto_dpif *peer_ofproto;
6098 enum slow_path_reason special;
6099 struct ofport_dpif *in_port;
6102 xlate_report(ctx, "Nonexistent patch port peer");
6106 peer_ofproto = ofproto_dpif_cast(peer->up.ofproto);
6107 if (peer_ofproto->backer != ctx->ofproto->backer) {
6108 xlate_report(ctx, "Patch port peer on a different datapath");
6112 ctx->ofproto = ofproto_dpif_cast(peer->up.ofproto);
6113 ctx->flow.in_port = peer->up.ofp_port;
6114 ctx->flow.metadata = htonll(0);
6115 memset(&ctx->flow.tunnel, 0, sizeof ctx->flow.tunnel);
6116 memset(ctx->flow.regs, 0, sizeof ctx->flow.regs);
6118 in_port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
6119 special = process_special(ctx->ofproto, &ctx->flow, in_port,
6122 ctx->slow = special;
6123 } else if (!in_port || may_receive(in_port, ctx)) {
6124 if (!in_port || stp_forward_in_state(in_port->stp_state)) {
6125 xlate_table_action(ctx, ctx->flow.in_port, 0, true);
6127 /* Forwarding is disabled by STP. Let OFPP_NORMAL and the
6128 * learning action look at the packet, then drop it. */
6129 struct flow old_base_flow = ctx->base_flow;
6130 size_t old_size = ctx->odp_actions->size;
6131 xlate_table_action(ctx, ctx->flow.in_port, 0, true);
6132 ctx->base_flow = old_base_flow;
6133 ctx->odp_actions->size = old_size;
6137 ctx->flow = old_flow;
6138 ctx->ofproto = ofproto_dpif_cast(ofport->up.ofproto);
6140 if (ctx->resubmit_stats) {
6141 netdev_vport_inc_tx(ofport->up.netdev, ctx->resubmit_stats);
6142 netdev_vport_inc_rx(peer->up.netdev, ctx->resubmit_stats);
6148 flow_vlan_tci = ctx->flow.vlan_tci;
6149 flow_skb_mark = ctx->flow.skb_mark;
6150 flow_nw_tos = ctx->flow.nw_tos;
6152 pdscp = get_priority(ofport, ctx->flow.skb_priority);
6154 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
6155 ctx->flow.nw_tos |= pdscp->dscp;
6158 if (ofport->tnl_port) {
6159 /* Save tunnel metadata so that changes made due to
6160 * the Logical (tunnel) Port are not visible for any further
6161 * matches, while explicit set actions on tunnel metadata are.
6163 struct flow_tnl flow_tnl = ctx->flow.tunnel;
6164 odp_port = tnl_port_send(ofport->tnl_port, &ctx->flow);
6165 if (odp_port == OVSP_NONE) {
6166 xlate_report(ctx, "Tunneling decided against output");
6167 goto out; /* restore flow_nw_tos */
6169 if (ctx->flow.tunnel.ip_dst == ctx->orig_tunnel_ip_dst) {
6170 xlate_report(ctx, "Not tunneling to our own address");
6171 goto out; /* restore flow_nw_tos */
6173 if (ctx->resubmit_stats) {
6174 netdev_vport_inc_tx(ofport->up.netdev, ctx->resubmit_stats);
6176 out_port = odp_port;
6177 commit_odp_tunnel_action(&ctx->flow, &ctx->base_flow,
6179 ctx->flow.tunnel = flow_tnl; /* Restore tunnel metadata */
6181 odp_port = ofport->odp_port;
6182 out_port = vsp_realdev_to_vlandev(ctx->ofproto, odp_port,
6183 ctx->flow.vlan_tci);
6184 if (out_port != odp_port) {
6185 ctx->flow.vlan_tci = htons(0);
6187 ctx->flow.skb_mark &= ~IPSEC_MARK;
6189 commit_odp_actions(&ctx->flow, &ctx->base_flow, ctx->odp_actions);
6190 nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_OUTPUT, out_port);
6192 ctx->sflow_odp_port = odp_port;
6193 ctx->sflow_n_outputs++;
6194 ctx->nf_output_iface = ofp_port;
6197 ctx->flow.vlan_tci = flow_vlan_tci;
6198 ctx->flow.skb_mark = flow_skb_mark;
6200 ctx->flow.nw_tos = flow_nw_tos;
6204 compose_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port)
6206 compose_output_action__(ctx, ofp_port, true);
6210 tag_the_flow(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
6212 struct ofproto_dpif *ofproto = ctx->ofproto;
6213 uint8_t table_id = ctx->table_id;
6215 if (table_id > 0 && table_id < N_TABLES) {
6216 struct table_dpif *table = &ofproto->tables[table_id];
6217 if (table->other_table) {
6218 ctx->tags |= (rule && rule->tag
6220 : rule_calculate_tag(&ctx->flow,
6221 &table->other_table->mask,
6227 /* Common rule processing in one place to avoid duplicating code. */
6228 static struct rule_dpif *
6229 ctx_rule_hooks(struct action_xlate_ctx *ctx, struct rule_dpif *rule,
6232 if (ctx->resubmit_hook) {
6233 ctx->resubmit_hook(ctx, rule);
6235 if (rule == NULL && may_packet_in) {
6237 * check if table configuration flags
6238 * OFPTC_TABLE_MISS_CONTROLLER, default.
6239 * OFPTC_TABLE_MISS_CONTINUE,
6240 * OFPTC_TABLE_MISS_DROP
6241 * When OF1.0, OFPTC_TABLE_MISS_CONTINUE is used. What to do?
6243 rule = rule_dpif_miss_rule(ctx->ofproto, &ctx->flow);
6245 if (rule && ctx->resubmit_stats) {
6246 rule_credit_stats(rule, ctx->resubmit_stats);
6252 xlate_table_action(struct action_xlate_ctx *ctx,
6253 uint16_t in_port, uint8_t table_id, bool may_packet_in)
6255 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
6256 struct rule_dpif *rule;
6257 uint16_t old_in_port = ctx->flow.in_port;
6258 uint8_t old_table_id = ctx->table_id;
6260 ctx->table_id = table_id;
6262 /* Look up a flow with 'in_port' as the input port. */
6263 ctx->flow.in_port = in_port;
6264 rule = rule_dpif_lookup__(ctx->ofproto, &ctx->flow, table_id);
6266 tag_the_flow(ctx, rule);
6268 /* Restore the original input port. Otherwise OFPP_NORMAL and
6269 * OFPP_IN_PORT will have surprising behavior. */
6270 ctx->flow.in_port = old_in_port;
6272 rule = ctx_rule_hooks(ctx, rule, may_packet_in);
6275 struct rule_dpif *old_rule = ctx->rule;
6279 do_xlate_actions(rule->up.ofpacts, rule->up.ofpacts_len, ctx);
6280 ctx->rule = old_rule;
6284 ctx->table_id = old_table_id;
6286 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
6288 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
6289 MAX_RESUBMIT_RECURSION);
6290 ctx->max_resubmit_trigger = true;
6295 xlate_ofpact_resubmit(struct action_xlate_ctx *ctx,
6296 const struct ofpact_resubmit *resubmit)
6301 in_port = resubmit->in_port;
6302 if (in_port == OFPP_IN_PORT) {
6303 in_port = ctx->flow.in_port;
6306 table_id = resubmit->table_id;
6307 if (table_id == 255) {
6308 table_id = ctx->table_id;
6311 xlate_table_action(ctx, in_port, table_id, false);
6315 flood_packets(struct action_xlate_ctx *ctx, bool all)
6317 struct ofport_dpif *ofport;
6319 HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) {
6320 uint16_t ofp_port = ofport->up.ofp_port;
6322 if (ofp_port == ctx->flow.in_port) {
6327 compose_output_action__(ctx, ofp_port, false);
6328 } else if (!(ofport->up.pp.config & OFPUTIL_PC_NO_FLOOD)) {
6329 compose_output_action(ctx, ofp_port);
6333 ctx->nf_output_iface = NF_OUT_FLOOD;
6337 execute_controller_action(struct action_xlate_ctx *ctx, int len,
6338 enum ofp_packet_in_reason reason,
6339 uint16_t controller_id)
6341 struct ofputil_packet_in pin;
6342 struct ofpbuf *packet;
6344 ovs_assert(!ctx->slow || ctx->slow == SLOW_CONTROLLER);
6345 ctx->slow = SLOW_CONTROLLER;
6350 packet = ofpbuf_clone(ctx->packet);
6352 if (packet->l2 && packet->l3) {
6353 struct eth_header *eh;
6354 uint16_t mpls_depth;
6356 eth_pop_vlan(packet);
6359 memcpy(eh->eth_src, ctx->flow.dl_src, sizeof eh->eth_src);
6360 memcpy(eh->eth_dst, ctx->flow.dl_dst, sizeof eh->eth_dst);
6362 if (ctx->flow.vlan_tci & htons(VLAN_CFI)) {
6363 eth_push_vlan(packet, ctx->flow.vlan_tci);
6366 mpls_depth = eth_mpls_depth(packet);
6368 if (mpls_depth < ctx->flow.mpls_depth) {
6369 push_mpls(packet, ctx->flow.dl_type, ctx->flow.mpls_lse);
6370 } else if (mpls_depth > ctx->flow.mpls_depth) {
6371 pop_mpls(packet, ctx->flow.dl_type);
6372 } else if (mpls_depth) {
6373 set_mpls_lse(packet, ctx->flow.mpls_lse);
6377 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
6378 packet_set_ipv4(packet, ctx->flow.nw_src, ctx->flow.nw_dst,
6379 ctx->flow.nw_tos, ctx->flow.nw_ttl);
6383 if (ctx->flow.nw_proto == IPPROTO_TCP) {
6384 packet_set_tcp_port(packet, ctx->flow.tp_src,
6386 } else if (ctx->flow.nw_proto == IPPROTO_UDP) {
6387 packet_set_udp_port(packet, ctx->flow.tp_src,
6394 pin.packet = packet->data;
6395 pin.packet_len = packet->size;
6396 pin.reason = reason;
6397 pin.controller_id = controller_id;
6398 pin.table_id = ctx->table_id;
6399 pin.cookie = ctx->rule ? ctx->rule->up.flow_cookie : 0;
6402 flow_get_metadata(&ctx->flow, &pin.fmd);
6404 connmgr_send_packet_in(ctx->ofproto->up.connmgr, &pin);
6405 ofpbuf_delete(packet);
6409 execute_mpls_push_action(struct action_xlate_ctx *ctx, ovs_be16 eth_type)
6411 ovs_assert(eth_type_mpls(eth_type));
6413 if (ctx->base_flow.mpls_depth) {
6414 ctx->flow.mpls_lse &= ~htonl(MPLS_BOS_MASK);
6415 ctx->flow.mpls_depth++;
6420 if (ctx->flow.dl_type == htons(ETH_TYPE_IPV6)) {
6421 label = htonl(0x2); /* IPV6 Explicit Null. */
6423 label = htonl(0x0); /* IPV4 Explicit Null. */
6425 tc = (ctx->flow.nw_tos & IP_DSCP_MASK) >> 2;
6426 ttl = ctx->flow.nw_ttl ? ctx->flow.nw_ttl : 0x40;
6427 ctx->flow.mpls_lse = set_mpls_lse_values(ttl, tc, 1, label);
6428 ctx->flow.mpls_depth = 1;
6430 ctx->flow.dl_type = eth_type;
6434 execute_mpls_pop_action(struct action_xlate_ctx *ctx, ovs_be16 eth_type)
6436 ovs_assert(eth_type_mpls(ctx->flow.dl_type));
6437 ovs_assert(!eth_type_mpls(eth_type));
6439 if (ctx->flow.mpls_depth) {
6440 ctx->flow.mpls_depth--;
6441 ctx->flow.mpls_lse = htonl(0);
6442 if (!ctx->flow.mpls_depth) {
6443 ctx->flow.dl_type = eth_type;
6449 compose_dec_ttl(struct action_xlate_ctx *ctx, struct ofpact_cnt_ids *ids)
6451 if (ctx->flow.dl_type != htons(ETH_TYPE_IP) &&
6452 ctx->flow.dl_type != htons(ETH_TYPE_IPV6)) {
6456 if (ctx->flow.nw_ttl > 1) {
6462 for (i = 0; i < ids->n_controllers; i++) {
6463 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL,
6467 /* Stop processing for current table. */
6473 execute_set_mpls_ttl_action(struct action_xlate_ctx *ctx, uint8_t ttl)
6475 if (!eth_type_mpls(ctx->flow.dl_type)) {
6479 set_mpls_lse_ttl(&ctx->flow.mpls_lse, ttl);
6484 execute_dec_mpls_ttl_action(struct action_xlate_ctx *ctx)
6486 uint8_t ttl = mpls_lse_to_ttl(ctx->flow.mpls_lse);
6488 if (!eth_type_mpls(ctx->flow.dl_type)) {
6494 set_mpls_lse_ttl(&ctx->flow.mpls_lse, ttl);
6497 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL, 0);
6499 /* Stop processing for current table. */
6505 xlate_output_action(struct action_xlate_ctx *ctx,
6506 uint16_t port, uint16_t max_len, bool may_packet_in)
6508 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
6510 ctx->nf_output_iface = NF_OUT_DROP;
6514 compose_output_action(ctx, ctx->flow.in_port);
6517 xlate_table_action(ctx, ctx->flow.in_port, 0, may_packet_in);
6523 flood_packets(ctx, false);
6526 flood_packets(ctx, true);
6528 case OFPP_CONTROLLER:
6529 execute_controller_action(ctx, max_len, OFPR_ACTION, 0);
6535 if (port != ctx->flow.in_port) {
6536 compose_output_action(ctx, port);
6538 xlate_report(ctx, "skipping output to input port");
6543 if (prev_nf_output_iface == NF_OUT_FLOOD) {
6544 ctx->nf_output_iface = NF_OUT_FLOOD;
6545 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
6546 ctx->nf_output_iface = prev_nf_output_iface;
6547 } else if (prev_nf_output_iface != NF_OUT_DROP &&
6548 ctx->nf_output_iface != NF_OUT_FLOOD) {
6549 ctx->nf_output_iface = NF_OUT_MULTI;
6554 xlate_output_reg_action(struct action_xlate_ctx *ctx,
6555 const struct ofpact_output_reg *or)
6557 uint64_t port = mf_get_subfield(&or->src, &ctx->flow);
6558 if (port <= UINT16_MAX) {
6559 xlate_output_action(ctx, port, or->max_len, false);
6564 xlate_enqueue_action(struct action_xlate_ctx *ctx,
6565 const struct ofpact_enqueue *enqueue)
6567 uint16_t ofp_port = enqueue->port;
6568 uint32_t queue_id = enqueue->queue;
6569 uint32_t flow_priority, priority;
6572 /* Translate queue to priority. */
6573 error = dpif_queue_to_priority(ctx->ofproto->backer->dpif,
6574 queue_id, &priority);
6576 /* Fall back to ordinary output action. */
6577 xlate_output_action(ctx, enqueue->port, 0, false);
6581 /* Check output port. */
6582 if (ofp_port == OFPP_IN_PORT) {
6583 ofp_port = ctx->flow.in_port;
6584 } else if (ofp_port == ctx->flow.in_port) {
6588 /* Add datapath actions. */
6589 flow_priority = ctx->flow.skb_priority;
6590 ctx->flow.skb_priority = priority;
6591 compose_output_action(ctx, ofp_port);
6592 ctx->flow.skb_priority = flow_priority;
6594 /* Update NetFlow output port. */
6595 if (ctx->nf_output_iface == NF_OUT_DROP) {
6596 ctx->nf_output_iface = ofp_port;
6597 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
6598 ctx->nf_output_iface = NF_OUT_MULTI;
6603 xlate_set_queue_action(struct action_xlate_ctx *ctx, uint32_t queue_id)
6605 uint32_t skb_priority;
6607 if (!dpif_queue_to_priority(ctx->ofproto->backer->dpif,
6608 queue_id, &skb_priority)) {
6609 ctx->flow.skb_priority = skb_priority;
6611 /* Couldn't translate queue to a priority. Nothing to do. A warning
6612 * has already been logged. */
6617 slave_enabled_cb(uint16_t ofp_port, void *ofproto_)
6619 struct ofproto_dpif *ofproto = ofproto_;
6620 struct ofport_dpif *port;
6630 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
6633 port = get_ofp_port(ofproto, ofp_port);
6634 return port ? port->may_enable : false;
6639 xlate_bundle_action(struct action_xlate_ctx *ctx,
6640 const struct ofpact_bundle *bundle)
6644 port = bundle_execute(bundle, &ctx->flow, slave_enabled_cb, ctx->ofproto);
6645 if (bundle->dst.field) {
6646 nxm_reg_load(&bundle->dst, port, &ctx->flow);
6648 xlate_output_action(ctx, port, 0, false);
6653 xlate_learn_action(struct action_xlate_ctx *ctx,
6654 const struct ofpact_learn *learn)
6656 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
6657 struct ofputil_flow_mod fm;
6658 uint64_t ofpacts_stub[1024 / 8];
6659 struct ofpbuf ofpacts;
6662 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
6663 learn_execute(learn, &ctx->flow, &fm, &ofpacts);
6665 error = ofproto_flow_mod(&ctx->ofproto->up, &fm);
6666 if (error && !VLOG_DROP_WARN(&rl)) {
6667 VLOG_WARN("learning action failed to modify flow table (%s)",
6668 ofperr_get_name(error));
6671 ofpbuf_uninit(&ofpacts);
6674 /* Reduces '*timeout' to no more than 'max'. A value of zero in either case
6675 * means "infinite". */
6677 reduce_timeout(uint16_t max, uint16_t *timeout)
6679 if (max && (!*timeout || *timeout > max)) {
6685 xlate_fin_timeout(struct action_xlate_ctx *ctx,
6686 const struct ofpact_fin_timeout *oft)
6688 if (ctx->tcp_flags & (TCP_FIN | TCP_RST) && ctx->rule) {
6689 struct rule_dpif *rule = ctx->rule;
6691 reduce_timeout(oft->fin_idle_timeout, &rule->up.idle_timeout);
6692 reduce_timeout(oft->fin_hard_timeout, &rule->up.hard_timeout);
6697 xlate_sample_action(struct action_xlate_ctx *ctx,
6698 const struct ofpact_sample *os)
6700 union user_action_cookie cookie;
6701 /* Scale the probability from 16-bit to 32-bit while representing
6702 * the same percentage. */
6703 uint32_t probability = (os->probability << 16) | os->probability;
6705 commit_odp_actions(&ctx->flow, &ctx->base_flow, ctx->odp_actions);
6707 compose_flow_sample_cookie(os->probability, os->collector_set_id,
6708 os->obs_domain_id, os->obs_point_id, &cookie);
6709 compose_sample_action(ctx->ofproto, ctx->odp_actions, &ctx->flow,
6710 probability, &cookie, sizeof cookie.flow_sample);
6714 may_receive(const struct ofport_dpif *port, struct action_xlate_ctx *ctx)
6716 if (port->up.pp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
6717 ? OFPUTIL_PC_NO_RECV_STP
6718 : OFPUTIL_PC_NO_RECV)) {
6722 /* Only drop packets here if both forwarding and learning are
6723 * disabled. If just learning is enabled, we need to have
6724 * OFPP_NORMAL and the learning action have a look at the packet
6725 * before we can drop it. */
6726 if (!stp_forward_in_state(port->stp_state)
6727 && !stp_learn_in_state(port->stp_state)) {
6735 tunnel_ecn_ok(struct action_xlate_ctx *ctx)
6737 if (is_ip_any(&ctx->base_flow)
6738 && (ctx->flow.tunnel.ip_tos & IP_ECN_MASK) == IP_ECN_CE) {
6739 if ((ctx->base_flow.nw_tos & IP_ECN_MASK) == IP_ECN_NOT_ECT) {
6740 VLOG_WARN_RL(&rl, "dropping tunnel packet marked ECN CE"
6741 " but is not ECN capable");
6744 /* Set the ECN CE value in the tunneled packet. */
6745 ctx->flow.nw_tos |= IP_ECN_CE;
6753 do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len,
6754 struct action_xlate_ctx *ctx)
6756 bool was_evictable = true;
6757 const struct ofpact *a;
6760 /* Don't let the rule we're working on get evicted underneath us. */
6761 was_evictable = ctx->rule->up.evictable;
6762 ctx->rule->up.evictable = false;
6765 do_xlate_actions_again:
6766 OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) {
6767 struct ofpact_controller *controller;
6768 const struct ofpact_metadata *metadata;
6776 xlate_output_action(ctx, ofpact_get_OUTPUT(a)->port,
6777 ofpact_get_OUTPUT(a)->max_len, true);
6780 case OFPACT_CONTROLLER:
6781 controller = ofpact_get_CONTROLLER(a);
6782 execute_controller_action(ctx, controller->max_len,
6784 controller->controller_id);
6787 case OFPACT_ENQUEUE:
6788 xlate_enqueue_action(ctx, ofpact_get_ENQUEUE(a));
6791 case OFPACT_SET_VLAN_VID:
6792 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
6793 ctx->flow.vlan_tci |= (htons(ofpact_get_SET_VLAN_VID(a)->vlan_vid)
6797 case OFPACT_SET_VLAN_PCP:
6798 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
6799 ctx->flow.vlan_tci |= htons((ofpact_get_SET_VLAN_PCP(a)->vlan_pcp
6804 case OFPACT_STRIP_VLAN:
6805 ctx->flow.vlan_tci = htons(0);
6808 case OFPACT_PUSH_VLAN:
6809 /* XXX 802.1AD(QinQ) */
6810 ctx->flow.vlan_tci = htons(VLAN_CFI);
6813 case OFPACT_SET_ETH_SRC:
6814 memcpy(ctx->flow.dl_src, ofpact_get_SET_ETH_SRC(a)->mac,
6818 case OFPACT_SET_ETH_DST:
6819 memcpy(ctx->flow.dl_dst, ofpact_get_SET_ETH_DST(a)->mac,
6823 case OFPACT_SET_IPV4_SRC:
6824 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
6825 ctx->flow.nw_src = ofpact_get_SET_IPV4_SRC(a)->ipv4;
6829 case OFPACT_SET_IPV4_DST:
6830 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
6831 ctx->flow.nw_dst = ofpact_get_SET_IPV4_DST(a)->ipv4;
6835 case OFPACT_SET_IPV4_DSCP:
6836 /* OpenFlow 1.0 only supports IPv4. */
6837 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
6838 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
6839 ctx->flow.nw_tos |= ofpact_get_SET_IPV4_DSCP(a)->dscp;
6843 case OFPACT_SET_L4_SRC_PORT:
6844 if (is_ip_any(&ctx->flow)) {
6845 ctx->flow.tp_src = htons(ofpact_get_SET_L4_SRC_PORT(a)->port);
6849 case OFPACT_SET_L4_DST_PORT:
6850 if (is_ip_any(&ctx->flow)) {
6851 ctx->flow.tp_dst = htons(ofpact_get_SET_L4_DST_PORT(a)->port);
6855 case OFPACT_RESUBMIT:
6856 xlate_ofpact_resubmit(ctx, ofpact_get_RESUBMIT(a));
6859 case OFPACT_SET_TUNNEL:
6860 ctx->flow.tunnel.tun_id = htonll(ofpact_get_SET_TUNNEL(a)->tun_id);
6863 case OFPACT_SET_QUEUE:
6864 xlate_set_queue_action(ctx, ofpact_get_SET_QUEUE(a)->queue_id);
6867 case OFPACT_POP_QUEUE:
6868 ctx->flow.skb_priority = ctx->orig_skb_priority;
6871 case OFPACT_REG_MOVE:
6872 nxm_execute_reg_move(ofpact_get_REG_MOVE(a), &ctx->flow);
6875 case OFPACT_REG_LOAD:
6876 nxm_execute_reg_load(ofpact_get_REG_LOAD(a), &ctx->flow);
6879 case OFPACT_STACK_PUSH:
6880 nxm_execute_stack_push(ofpact_get_STACK_PUSH(a), &ctx->flow,
6884 case OFPACT_STACK_POP:
6885 nxm_execute_stack_pop(ofpact_get_STACK_POP(a), &ctx->flow,
6889 case OFPACT_PUSH_MPLS:
6890 execute_mpls_push_action(ctx, ofpact_get_PUSH_MPLS(a)->ethertype);
6893 case OFPACT_POP_MPLS:
6894 execute_mpls_pop_action(ctx, ofpact_get_POP_MPLS(a)->ethertype);
6897 case OFPACT_SET_MPLS_TTL:
6898 if (execute_set_mpls_ttl_action(ctx, ofpact_get_SET_MPLS_TTL(a)->ttl)) {
6903 case OFPACT_DEC_MPLS_TTL:
6904 if (execute_dec_mpls_ttl_action(ctx)) {
6909 case OFPACT_DEC_TTL:
6910 if (compose_dec_ttl(ctx, ofpact_get_DEC_TTL(a))) {
6916 /* Nothing to do. */
6919 case OFPACT_MULTIPATH:
6920 multipath_execute(ofpact_get_MULTIPATH(a), &ctx->flow);
6924 ctx->ofproto->has_bundle_action = true;
6925 xlate_bundle_action(ctx, ofpact_get_BUNDLE(a));
6928 case OFPACT_OUTPUT_REG:
6929 xlate_output_reg_action(ctx, ofpact_get_OUTPUT_REG(a));
6933 ctx->has_learn = true;
6934 if (ctx->may_learn) {
6935 xlate_learn_action(ctx, ofpact_get_LEARN(a));
6943 case OFPACT_FIN_TIMEOUT:
6944 ctx->has_fin_timeout = true;
6945 xlate_fin_timeout(ctx, ofpact_get_FIN_TIMEOUT(a));
6948 case OFPACT_CLEAR_ACTIONS:
6950 * Nothing to do because writa-actions is not supported for now.
6951 * When writa-actions is supported, clear-actions also must
6952 * be supported at the same time.
6956 case OFPACT_WRITE_METADATA:
6957 metadata = ofpact_get_WRITE_METADATA(a);
6958 ctx->flow.metadata &= ~metadata->mask;
6959 ctx->flow.metadata |= metadata->metadata & metadata->mask;
6962 case OFPACT_GOTO_TABLE: {
6963 /* It is assumed that goto-table is the last action. */
6964 struct ofpact_goto_table *ogt = ofpact_get_GOTO_TABLE(a);
6965 struct rule_dpif *rule;
6967 ovs_assert(ctx->table_id < ogt->table_id);
6969 ctx->table_id = ogt->table_id;
6971 /* Look up a flow from the new table. */
6972 rule = rule_dpif_lookup__(ctx->ofproto, &ctx->flow, ctx->table_id);
6974 tag_the_flow(ctx, rule);
6976 rule = ctx_rule_hooks(ctx, rule, true);
6980 ctx->rule->up.evictable = was_evictable;
6983 was_evictable = rule->up.evictable;
6984 rule->up.evictable = false;
6986 /* Tail recursion removal. */
6987 ofpacts = rule->up.ofpacts;
6988 ofpacts_len = rule->up.ofpacts_len;
6989 goto do_xlate_actions_again;
6995 xlate_sample_action(ctx, ofpact_get_SAMPLE(a));
7002 ctx->rule->up.evictable = was_evictable;
7007 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
7008 struct ofproto_dpif *ofproto, const struct flow *flow,
7009 const struct initial_vals *initial_vals,
7010 struct rule_dpif *rule,
7011 uint8_t tcp_flags, const struct ofpbuf *packet)
7013 /* Flow initialization rules:
7014 * - 'base_flow' must match the kernel's view of the packet at the
7015 * time that action processing starts. 'flow' represents any
7016 * transformations we wish to make through actions.
7017 * - By default 'base_flow' and 'flow' are the same since the input
7018 * packet matches the output before any actions are applied.
7019 * - When using VLAN splinters, 'base_flow''s VLAN is set to the value
7020 * of the received packet as seen by the kernel. If we later output
7021 * to another device without any modifications this will cause us to
7022 * insert a new tag since the original one was stripped off by the
7024 * - Tunnel metadata as received is retained in 'flow'. This allows
7025 * tunnel metadata matching also in later tables.
7026 * Since a kernel action for setting the tunnel metadata will only be
7027 * generated with actual tunnel output, changing the tunnel metadata
7028 * values in 'flow' (such as tun_id) will only have effect with a later
7029 * tunnel output action.
7030 * - Tunnel 'base_flow' is completely cleared since that is what the
7031 * kernel does. If we wish to maintain the original values an action
7032 * needs to be generated. */
7034 ctx->ofproto = ofproto;
7036 ctx->base_flow = ctx->flow;
7037 memset(&ctx->base_flow.tunnel, 0, sizeof ctx->base_flow.tunnel);
7038 ctx->orig_tunnel_ip_dst = flow->tunnel.ip_dst;
7040 ctx->packet = packet;
7041 ctx->may_learn = packet != NULL;
7042 ctx->tcp_flags = tcp_flags;
7043 ctx->resubmit_hook = NULL;
7044 ctx->report_hook = NULL;
7045 ctx->resubmit_stats = NULL;
7048 ctx->base_flow.vlan_tci = initial_vals->vlan_tci;
7052 /* Translates the 'ofpacts_len' bytes of "struct ofpacts" starting at 'ofpacts'
7053 * into datapath actions in 'odp_actions', using 'ctx'. */
7055 xlate_actions(struct action_xlate_ctx *ctx,
7056 const struct ofpact *ofpacts, size_t ofpacts_len,
7057 struct ofpbuf *odp_actions)
7059 /* Normally false. Set to true if we ever hit MAX_RESUBMIT_RECURSION, so
7060 * that in the future we always keep a copy of the original flow for
7061 * tracing purposes. */
7062 static bool hit_resubmit_limit;
7064 enum slow_path_reason special;
7065 struct ofport_dpif *in_port;
7066 struct flow orig_flow;
7068 COVERAGE_INC(ofproto_dpif_xlate);
7070 ofpbuf_clear(odp_actions);
7071 ofpbuf_reserve(odp_actions, NL_A_U32_SIZE);
7073 ctx->odp_actions = odp_actions;
7076 ctx->has_learn = false;
7077 ctx->has_normal = false;
7078 ctx->has_fin_timeout = false;
7079 ctx->nf_output_iface = NF_OUT_DROP;
7082 ctx->max_resubmit_trigger = false;
7083 ctx->orig_skb_priority = ctx->flow.skb_priority;
7087 ofpbuf_use_stub(&ctx->stack, ctx->init_stack, sizeof ctx->init_stack);
7089 if (ctx->ofproto->has_mirrors || hit_resubmit_limit) {
7090 /* Do this conditionally because the copy is expensive enough that it
7091 * shows up in profiles. */
7092 orig_flow = ctx->flow;
7095 if (ctx->flow.nw_frag & FLOW_NW_FRAG_ANY) {
7096 switch (ctx->ofproto->up.frag_handling) {
7097 case OFPC_FRAG_NORMAL:
7098 /* We must pretend that transport ports are unavailable. */
7099 ctx->flow.tp_src = ctx->base_flow.tp_src = htons(0);
7100 ctx->flow.tp_dst = ctx->base_flow.tp_dst = htons(0);
7103 case OFPC_FRAG_DROP:
7106 case OFPC_FRAG_REASM:
7109 case OFPC_FRAG_NX_MATCH:
7110 /* Nothing to do. */
7113 case OFPC_INVALID_TTL_TO_CONTROLLER:
7118 in_port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
7119 special = process_special(ctx->ofproto, &ctx->flow, in_port, ctx->packet);
7121 ctx->slow = special;
7123 static struct vlog_rate_limit trace_rl = VLOG_RATE_LIMIT_INIT(1, 1);
7124 struct initial_vals initial_vals;
7125 size_t sample_actions_len;
7126 uint32_t local_odp_port;
7128 initial_vals.vlan_tci = ctx->base_flow.vlan_tci;
7130 add_sflow_action(ctx);
7131 add_ipfix_action(ctx);
7132 sample_actions_len = ctx->odp_actions->size;
7134 if (tunnel_ecn_ok(ctx) && (!in_port || may_receive(in_port, ctx))) {
7135 do_xlate_actions(ofpacts, ofpacts_len, ctx);
7137 /* We've let OFPP_NORMAL and the learning action look at the
7138 * packet, so drop it now if forwarding is disabled. */
7139 if (in_port && !stp_forward_in_state(in_port->stp_state)) {
7140 ctx->odp_actions->size = sample_actions_len;
7144 if (ctx->max_resubmit_trigger && !ctx->resubmit_hook) {
7145 if (!hit_resubmit_limit) {
7146 /* We didn't record the original flow. Make sure we do from
7148 hit_resubmit_limit = true;
7149 } else if (!VLOG_DROP_ERR(&trace_rl)) {
7150 struct ds ds = DS_EMPTY_INITIALIZER;
7152 ofproto_trace(ctx->ofproto, &orig_flow, ctx->packet,
7153 &initial_vals, &ds);
7154 VLOG_ERR("Trace triggered by excessive resubmit "
7155 "recursion:\n%s", ds_cstr(&ds));
7160 local_odp_port = ofp_port_to_odp_port(ctx->ofproto, OFPP_LOCAL);
7161 if (!connmgr_must_output_local(ctx->ofproto->up.connmgr, &ctx->flow,
7163 ctx->odp_actions->data,
7164 ctx->odp_actions->size)) {
7165 compose_output_action(ctx, OFPP_LOCAL);
7167 if (ctx->ofproto->has_mirrors) {
7168 add_mirror_actions(ctx, &orig_flow);
7170 fix_sflow_action(ctx);
7173 ofpbuf_uninit(&ctx->stack);
7176 /* Translates the 'ofpacts_len' bytes of "struct ofpact"s starting at 'ofpacts'
7177 * into datapath actions, using 'ctx', and discards the datapath actions. */
7179 xlate_actions_for_side_effects(struct action_xlate_ctx *ctx,
7180 const struct ofpact *ofpacts,
7183 uint64_t odp_actions_stub[1024 / 8];
7184 struct ofpbuf odp_actions;
7186 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
7187 xlate_actions(ctx, ofpacts, ofpacts_len, &odp_actions);
7188 ofpbuf_uninit(&odp_actions);
7192 xlate_report(struct action_xlate_ctx *ctx, const char *s)
7194 if (ctx->report_hook) {
7195 ctx->report_hook(ctx, s);
7199 /* OFPP_NORMAL implementation. */
7201 static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *);
7203 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
7204 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle',
7205 * the bundle on which the packet was received, returns the VLAN to which the
7208 * Both 'vid' and the return value are in the range 0...4095. */
7210 input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid)
7212 switch (in_bundle->vlan_mode) {
7213 case PORT_VLAN_ACCESS:
7214 return in_bundle->vlan;
7217 case PORT_VLAN_TRUNK:
7220 case PORT_VLAN_NATIVE_UNTAGGED:
7221 case PORT_VLAN_NATIVE_TAGGED:
7222 return vid ? vid : in_bundle->vlan;
7229 /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'.
7230 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
7233 * 'vid' should be the VID obtained from the 802.1Q header that was received as
7234 * part of a packet (specify 0 if there was no 802.1Q header), in the range
7237 input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn)
7239 /* Allow any VID on the OFPP_NONE port. */
7240 if (in_bundle == &ofpp_none_bundle) {
7244 switch (in_bundle->vlan_mode) {
7245 case PORT_VLAN_ACCESS:
7248 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7249 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
7250 "packet received on port %s configured as VLAN "
7251 "%"PRIu16" access port",
7252 in_bundle->ofproto->up.name, vid,
7253 in_bundle->name, in_bundle->vlan);
7259 case PORT_VLAN_NATIVE_UNTAGGED:
7260 case PORT_VLAN_NATIVE_TAGGED:
7262 /* Port must always carry its native VLAN. */
7266 case PORT_VLAN_TRUNK:
7267 if (!ofbundle_includes_vlan(in_bundle, vid)) {
7269 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7270 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet "
7271 "received on port %s not configured for trunking "
7273 in_bundle->ofproto->up.name, vid,
7274 in_bundle->name, vid);
7286 /* Given 'vlan', the VLAN that a packet belongs to, and
7287 * 'out_bundle', a bundle on which the packet is to be output, returns the VID
7288 * that should be included in the 802.1Q header. (If the return value is 0,
7289 * then the 802.1Q header should only be included in the packet if there is a
7292 * Both 'vlan' and the return value are in the range 0...4095. */
7294 output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan)
7296 switch (out_bundle->vlan_mode) {
7297 case PORT_VLAN_ACCESS:
7300 case PORT_VLAN_TRUNK:
7301 case PORT_VLAN_NATIVE_TAGGED:
7304 case PORT_VLAN_NATIVE_UNTAGGED:
7305 return vlan == out_bundle->vlan ? 0 : vlan;
7313 output_normal(struct action_xlate_ctx *ctx, const struct ofbundle *out_bundle,
7316 struct ofport_dpif *port;
7318 ovs_be16 tci, old_tci;
7320 vid = output_vlan_to_vid(out_bundle, vlan);
7321 if (!out_bundle->bond) {
7322 port = ofbundle_get_a_port(out_bundle);
7324 port = bond_choose_output_slave(out_bundle->bond, &ctx->flow,
7327 /* No slaves enabled, so drop packet. */
7332 old_tci = ctx->flow.vlan_tci;
7334 if (tci || out_bundle->use_priority_tags) {
7335 tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK);
7337 tci |= htons(VLAN_CFI);
7340 ctx->flow.vlan_tci = tci;
7342 compose_output_action(ctx, port->up.ofp_port);
7343 ctx->flow.vlan_tci = old_tci;
7347 mirror_mask_ffs(mirror_mask_t mask)
7349 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
7354 ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
7356 return (bundle->vlan_mode != PORT_VLAN_ACCESS
7357 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
7361 ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan)
7363 return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan);
7366 /* Returns an arbitrary interface within 'bundle'. */
7367 static struct ofport_dpif *
7368 ofbundle_get_a_port(const struct ofbundle *bundle)
7370 return CONTAINER_OF(list_front(&bundle->ports),
7371 struct ofport_dpif, bundle_node);
7375 vlan_is_mirrored(const struct ofmirror *m, int vlan)
7377 return !m->vlans || bitmap_is_set(m->vlans, vlan);
7381 add_mirror_actions(struct action_xlate_ctx *ctx, const struct flow *orig_flow)
7383 struct ofproto_dpif *ofproto = ctx->ofproto;
7384 mirror_mask_t mirrors;
7385 struct ofbundle *in_bundle;
7388 const struct nlattr *a;
7391 in_bundle = lookup_input_bundle(ctx->ofproto, orig_flow->in_port,
7392 ctx->packet != NULL, NULL);
7396 mirrors = in_bundle->src_mirrors;
7398 /* Drop frames on bundles reserved for mirroring. */
7399 if (in_bundle->mirror_out) {
7400 if (ctx->packet != NULL) {
7401 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7402 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
7403 "%s, which is reserved exclusively for mirroring",
7404 ctx->ofproto->up.name, in_bundle->name);
7410 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
7411 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
7414 vlan = input_vid_to_vlan(in_bundle, vid);
7416 /* Look at the output ports to check for destination selections. */
7418 NL_ATTR_FOR_EACH (a, left, ctx->odp_actions->data,
7419 ctx->odp_actions->size) {
7420 enum ovs_action_attr type = nl_attr_type(a);
7421 struct ofport_dpif *ofport;
7423 if (type != OVS_ACTION_ATTR_OUTPUT) {
7427 ofport = get_odp_port(ofproto, nl_attr_get_u32(a));
7428 if (ofport && ofport->bundle) {
7429 mirrors |= ofport->bundle->dst_mirrors;
7437 /* Restore the original packet before adding the mirror actions. */
7438 ctx->flow = *orig_flow;
7443 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
7445 if (!vlan_is_mirrored(m, vlan)) {
7446 mirrors = zero_rightmost_1bit(mirrors);
7450 mirrors &= ~m->dup_mirrors;
7451 ctx->mirrors |= m->dup_mirrors;
7453 output_normal(ctx, m->out, vlan);
7454 } else if (vlan != m->out_vlan
7455 && !eth_addr_is_reserved(orig_flow->dl_dst)) {
7456 struct ofbundle *bundle;
7458 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
7459 if (ofbundle_includes_vlan(bundle, m->out_vlan)
7460 && !bundle->mirror_out) {
7461 output_normal(ctx, bundle, m->out_vlan);
7469 update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors,
7470 uint64_t packets, uint64_t bytes)
7476 for (; mirrors; mirrors = zero_rightmost_1bit(mirrors)) {
7479 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
7482 /* In normal circumstances 'm' will not be NULL. However,
7483 * if mirrors are reconfigured, we can temporarily get out
7484 * of sync in facet_revalidate(). We could "correct" the
7485 * mirror list before reaching here, but doing that would
7486 * not properly account the traffic stats we've currently
7487 * accumulated for previous mirror configuration. */
7491 m->packet_count += packets;
7492 m->byte_count += bytes;
7496 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
7497 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
7498 * indicate this; newer upstream kernels use gratuitous ARP requests. */
7500 is_gratuitous_arp(const struct flow *flow)
7502 return (flow->dl_type == htons(ETH_TYPE_ARP)
7503 && eth_addr_is_broadcast(flow->dl_dst)
7504 && (flow->nw_proto == ARP_OP_REPLY
7505 || (flow->nw_proto == ARP_OP_REQUEST
7506 && flow->nw_src == flow->nw_dst)));
7510 update_learning_table(struct ofproto_dpif *ofproto,
7511 const struct flow *flow, int vlan,
7512 struct ofbundle *in_bundle)
7514 struct mac_entry *mac;
7516 /* Don't learn the OFPP_NONE port. */
7517 if (in_bundle == &ofpp_none_bundle) {
7521 if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
7525 mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan);
7526 if (is_gratuitous_arp(flow)) {
7527 /* We don't want to learn from gratuitous ARP packets that are
7528 * reflected back over bond slaves so we lock the learning table. */
7529 if (!in_bundle->bond) {
7530 mac_entry_set_grat_arp_lock(mac);
7531 } else if (mac_entry_is_grat_arp_locked(mac)) {
7536 if (mac_entry_is_new(mac) || mac->port.p != in_bundle) {
7537 /* The log messages here could actually be useful in debugging,
7538 * so keep the rate limit relatively high. */
7539 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
7540 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
7541 "on port %s in VLAN %d",
7542 ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src),
7543 in_bundle->name, vlan);
7545 mac->port.p = in_bundle;
7546 tag_set_add(&ofproto->backer->revalidate_set,
7547 mac_learning_changed(ofproto->ml, mac));
7551 static struct ofbundle *
7552 lookup_input_bundle(const struct ofproto_dpif *ofproto, uint16_t in_port,
7553 bool warn, struct ofport_dpif **in_ofportp)
7555 struct ofport_dpif *ofport;
7557 /* Find the port and bundle for the received packet. */
7558 ofport = get_ofp_port(ofproto, in_port);
7560 *in_ofportp = ofport;
7562 if (ofport && ofport->bundle) {
7563 return ofport->bundle;
7566 /* Special-case OFPP_NONE, which a controller may use as the ingress
7567 * port for traffic that it is sourcing. */
7568 if (in_port == OFPP_NONE) {
7569 return &ofpp_none_bundle;
7572 /* Odd. A few possible reasons here:
7574 * - We deleted a port but there are still a few packets queued up
7577 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
7578 * we don't know about.
7580 * - The ofproto client didn't configure the port as part of a bundle.
7581 * This is particularly likely to happen if a packet was received on the
7582 * port after it was created, but before the client had a chance to
7583 * configure its bundle.
7586 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7588 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
7589 "port %"PRIu16, ofproto->up.name, in_port);
7594 /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or
7595 * dropped. Returns true if they may be forwarded, false if they should be
7598 * 'in_port' must be the ofport_dpif that corresponds to flow->in_port.
7599 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
7601 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
7602 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
7603 * checked by input_vid_is_valid().
7605 * May also add tags to '*tags', although the current implementation only does
7606 * so in one special case.
7609 is_admissible(struct action_xlate_ctx *ctx, struct ofport_dpif *in_port,
7612 struct ofproto_dpif *ofproto = ctx->ofproto;
7613 struct flow *flow = &ctx->flow;
7614 struct ofbundle *in_bundle = in_port->bundle;
7616 /* Drop frames for reserved multicast addresses
7617 * only if forward_bpdu option is absent. */
7618 if (!ofproto->up.forward_bpdu && eth_addr_is_reserved(flow->dl_dst)) {
7619 xlate_report(ctx, "packet has reserved destination MAC, dropping");
7623 if (in_bundle->bond) {
7624 struct mac_entry *mac;
7626 switch (bond_check_admissibility(in_bundle->bond, in_port,
7627 flow->dl_dst, &ctx->tags)) {
7632 xlate_report(ctx, "bonding refused admissibility, dropping");
7635 case BV_DROP_IF_MOVED:
7636 mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL);
7637 if (mac && mac->port.p != in_bundle &&
7638 (!is_gratuitous_arp(flow)
7639 || mac_entry_is_grat_arp_locked(mac))) {
7640 xlate_report(ctx, "SLB bond thinks this packet looped back, "
7652 xlate_normal(struct action_xlate_ctx *ctx)
7654 struct ofport_dpif *in_port;
7655 struct ofbundle *in_bundle;
7656 struct mac_entry *mac;
7660 ctx->has_normal = true;
7662 in_bundle = lookup_input_bundle(ctx->ofproto, ctx->flow.in_port,
7663 ctx->packet != NULL, &in_port);
7665 xlate_report(ctx, "no input bundle, dropping");
7669 /* Drop malformed frames. */
7670 if (ctx->flow.dl_type == htons(ETH_TYPE_VLAN) &&
7671 !(ctx->flow.vlan_tci & htons(VLAN_CFI))) {
7672 if (ctx->packet != NULL) {
7673 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7674 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
7675 "VLAN tag received on port %s",
7676 ctx->ofproto->up.name, in_bundle->name);
7678 xlate_report(ctx, "partial VLAN tag, dropping");
7682 /* Drop frames on bundles reserved for mirroring. */
7683 if (in_bundle->mirror_out) {
7684 if (ctx->packet != NULL) {
7685 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7686 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
7687 "%s, which is reserved exclusively for mirroring",
7688 ctx->ofproto->up.name, in_bundle->name);
7690 xlate_report(ctx, "input port is mirror output port, dropping");
7695 vid = vlan_tci_to_vid(ctx->flow.vlan_tci);
7696 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
7697 xlate_report(ctx, "disallowed VLAN VID for this input port, dropping");
7700 vlan = input_vid_to_vlan(in_bundle, vid);
7702 /* Check other admissibility requirements. */
7703 if (in_port && !is_admissible(ctx, in_port, vlan)) {
7707 /* Learn source MAC. */
7708 if (ctx->may_learn) {
7709 update_learning_table(ctx->ofproto, &ctx->flow, vlan, in_bundle);
7712 /* Determine output bundle. */
7713 mac = mac_learning_lookup(ctx->ofproto->ml, ctx->flow.dl_dst, vlan,
7716 if (mac->port.p != in_bundle) {
7717 xlate_report(ctx, "forwarding to learned port");
7718 output_normal(ctx, mac->port.p, vlan);
7720 xlate_report(ctx, "learned port is input port, dropping");
7723 struct ofbundle *bundle;
7725 xlate_report(ctx, "no learned MAC for destination, flooding");
7726 HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
7727 if (bundle != in_bundle
7728 && ofbundle_includes_vlan(bundle, vlan)
7729 && bundle->floodable
7730 && !bundle->mirror_out) {
7731 output_normal(ctx, bundle, vlan);
7734 ctx->nf_output_iface = NF_OUT_FLOOD;
7738 /* Optimized flow revalidation.
7740 * It's a difficult problem, in general, to tell which facets need to have
7741 * their actions recalculated whenever the OpenFlow flow table changes. We
7742 * don't try to solve that general problem: for most kinds of OpenFlow flow
7743 * table changes, we recalculate the actions for every facet. This is
7744 * relatively expensive, but it's good enough if the OpenFlow flow table
7745 * doesn't change very often.
7747 * However, we can expect one particular kind of OpenFlow flow table change to
7748 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
7749 * of CPU on revalidating every facet whenever MAC learning modifies the flow
7750 * table, we add a special case that applies to flow tables in which every rule
7751 * has the same form (that is, the same wildcards), except that the table is
7752 * also allowed to have a single "catch-all" flow that matches all packets. We
7753 * optimize this case by tagging all of the facets that resubmit into the table
7754 * and invalidating the same tag whenever a flow changes in that table. The
7755 * end result is that we revalidate just the facets that need it (and sometimes
7756 * a few more, but not all of the facets or even all of the facets that
7757 * resubmit to the table modified by MAC learning). */
7759 /* Calculates the tag to use for 'flow' and mask 'mask' when it is inserted
7760 * into an OpenFlow table with the given 'basis'. */
7762 rule_calculate_tag(const struct flow *flow, const struct minimask *mask,
7765 if (minimask_is_catchall(mask)) {
7768 uint32_t hash = flow_hash_in_minimask(flow, mask, secret);
7769 return tag_create_deterministic(hash);
7773 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
7774 * taggability of that table.
7776 * This function must be called after *each* change to a flow table. If you
7777 * skip calling it on some changes then the pointer comparisons at the end can
7778 * be invalid if you get unlucky. For example, if a flow removal causes a
7779 * cls_table to be destroyed and then a flow insertion causes a cls_table with
7780 * different wildcards to be created with the same address, then this function
7781 * will incorrectly skip revalidation. */
7783 table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
7785 struct table_dpif *table = &ofproto->tables[table_id];
7786 const struct oftable *oftable = &ofproto->up.tables[table_id];
7787 struct cls_table *catchall, *other;
7788 struct cls_table *t;
7790 catchall = other = NULL;
7792 switch (hmap_count(&oftable->cls.tables)) {
7794 /* We could tag this OpenFlow table but it would make the logic a
7795 * little harder and it's a corner case that doesn't seem worth it
7801 HMAP_FOR_EACH (t, hmap_node, &oftable->cls.tables) {
7802 if (cls_table_is_catchall(t)) {
7804 } else if (!other) {
7807 /* Indicate that we can't tag this by setting both tables to
7808 * NULL. (We know that 'catchall' is already NULL.) */
7815 /* Can't tag this table. */
7819 if (table->catchall_table != catchall || table->other_table != other) {
7820 table->catchall_table = catchall;
7821 table->other_table = other;
7822 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
7826 /* Given 'rule' that has changed in some way (either it is a rule being
7827 * inserted, a rule being deleted, or a rule whose actions are being
7828 * modified), marks facets for revalidation to ensure that packets will be
7829 * forwarded correctly according to the new state of the flow table.
7831 * This function must be called after *each* change to a flow table. See
7832 * the comment on table_update_taggable() for more information. */
7834 rule_invalidate(const struct rule_dpif *rule)
7836 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
7838 table_update_taggable(ofproto, rule->up.table_id);
7840 if (!ofproto->backer->need_revalidate) {
7841 struct table_dpif *table = &ofproto->tables[rule->up.table_id];
7843 if (table->other_table && rule->tag) {
7844 tag_set_add(&ofproto->backer->revalidate_set, rule->tag);
7846 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
7852 set_frag_handling(struct ofproto *ofproto_,
7853 enum ofp_config_flags frag_handling)
7855 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7856 if (frag_handling != OFPC_FRAG_REASM) {
7857 ofproto->backer->need_revalidate = REV_RECONFIGURE;
7865 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
7866 const struct flow *flow,
7867 const struct ofpact *ofpacts, size_t ofpacts_len)
7869 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7870 struct initial_vals initial_vals;
7871 struct odputil_keybuf keybuf;
7872 struct dpif_flow_stats stats;
7876 struct action_xlate_ctx ctx;
7877 uint64_t odp_actions_stub[1024 / 8];
7878 struct ofpbuf odp_actions;
7880 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
7881 odp_flow_key_from_flow(&key, flow,
7882 ofp_port_to_odp_port(ofproto, flow->in_port));
7884 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
7886 initial_vals.vlan_tci = flow->vlan_tci;
7887 action_xlate_ctx_init(&ctx, ofproto, flow, &initial_vals, NULL,
7888 packet_get_tcp_flags(packet, flow), packet);
7889 ctx.resubmit_stats = &stats;
7891 ofpbuf_use_stub(&odp_actions,
7892 odp_actions_stub, sizeof odp_actions_stub);
7893 xlate_actions(&ctx, ofpacts, ofpacts_len, &odp_actions);
7894 dpif_execute(ofproto->backer->dpif, key.data, key.size,
7895 odp_actions.data, odp_actions.size, packet);
7896 ofpbuf_uninit(&odp_actions);
7904 set_netflow(struct ofproto *ofproto_,
7905 const struct netflow_options *netflow_options)
7907 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7909 if (netflow_options) {
7910 if (!ofproto->netflow) {
7911 ofproto->netflow = netflow_create();
7913 return netflow_set_options(ofproto->netflow, netflow_options);
7915 netflow_destroy(ofproto->netflow);
7916 ofproto->netflow = NULL;
7922 get_netflow_ids(const struct ofproto *ofproto_,
7923 uint8_t *engine_type, uint8_t *engine_id)
7925 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7927 dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
7931 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
7933 if (!facet_is_controller_flow(facet) &&
7934 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
7935 struct subfacet *subfacet;
7936 struct ofexpired expired;
7938 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
7939 if (subfacet->path == SF_FAST_PATH) {
7940 struct dpif_flow_stats stats;
7942 subfacet_reinstall(subfacet, &stats);
7943 subfacet_update_stats(subfacet, &stats);
7947 expired.flow = facet->flow;
7948 expired.packet_count = facet->packet_count;
7949 expired.byte_count = facet->byte_count;
7950 expired.used = facet->used;
7951 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
7956 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
7958 struct facet *facet;
7960 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
7961 send_active_timeout(ofproto, facet);
7965 static struct ofproto_dpif *
7966 ofproto_dpif_lookup(const char *name)
7968 struct ofproto_dpif *ofproto;
7970 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
7971 hash_string(name, 0), &all_ofproto_dpifs) {
7972 if (!strcmp(ofproto->up.name, name)) {
7980 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
7981 const char *argv[], void *aux OVS_UNUSED)
7983 struct ofproto_dpif *ofproto;
7986 ofproto = ofproto_dpif_lookup(argv[1]);
7988 unixctl_command_reply_error(conn, "no such bridge");
7991 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
7993 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
7994 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
7998 unixctl_command_reply(conn, "table successfully flushed");
8002 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
8003 const char *argv[], void *aux OVS_UNUSED)
8005 struct ds ds = DS_EMPTY_INITIALIZER;
8006 const struct ofproto_dpif *ofproto;
8007 const struct mac_entry *e;
8009 ofproto = ofproto_dpif_lookup(argv[1]);
8011 unixctl_command_reply_error(conn, "no such bridge");
8015 ds_put_cstr(&ds, " port VLAN MAC Age\n");
8016 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
8017 struct ofbundle *bundle = e->port.p;
8018 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
8019 ofbundle_get_a_port(bundle)->odp_port,
8020 e->vlan, ETH_ADDR_ARGS(e->mac),
8021 mac_entry_age(ofproto->ml, e));
8023 unixctl_command_reply(conn, ds_cstr(&ds));
8028 struct action_xlate_ctx ctx;
8034 trace_format_rule(struct ds *result, uint8_t table_id, int level,
8035 const struct rule_dpif *rule)
8037 ds_put_char_multiple(result, '\t', level);
8039 ds_put_cstr(result, "No match\n");
8043 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
8044 table_id, ntohll(rule->up.flow_cookie));
8045 cls_rule_format(&rule->up.cr, result);
8046 ds_put_char(result, '\n');
8048 ds_put_char_multiple(result, '\t', level);
8049 ds_put_cstr(result, "OpenFlow ");
8050 ofpacts_format(rule->up.ofpacts, rule->up.ofpacts_len, result);
8051 ds_put_char(result, '\n');
8055 trace_format_flow(struct ds *result, int level, const char *title,
8056 struct trace_ctx *trace)
8058 ds_put_char_multiple(result, '\t', level);
8059 ds_put_format(result, "%s: ", title);
8060 if (flow_equal(&trace->ctx.flow, &trace->flow)) {
8061 ds_put_cstr(result, "unchanged");
8063 flow_format(result, &trace->ctx.flow);
8064 trace->flow = trace->ctx.flow;
8066 ds_put_char(result, '\n');
8070 trace_format_regs(struct ds *result, int level, const char *title,
8071 struct trace_ctx *trace)
8075 ds_put_char_multiple(result, '\t', level);
8076 ds_put_format(result, "%s:", title);
8077 for (i = 0; i < FLOW_N_REGS; i++) {
8078 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
8080 ds_put_char(result, '\n');
8084 trace_format_odp(struct ds *result, int level, const char *title,
8085 struct trace_ctx *trace)
8087 struct ofpbuf *odp_actions = trace->ctx.odp_actions;
8089 ds_put_char_multiple(result, '\t', level);
8090 ds_put_format(result, "%s: ", title);
8091 format_odp_actions(result, odp_actions->data, odp_actions->size);
8092 ds_put_char(result, '\n');
8096 trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
8098 struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx);
8099 struct ds *result = trace->result;
8101 ds_put_char(result, '\n');
8102 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
8103 trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace);
8104 trace_format_odp(result, ctx->recurse + 1, "Resubmitted odp", trace);
8105 trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule);
8109 trace_report(struct action_xlate_ctx *ctx, const char *s)
8111 struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx);
8112 struct ds *result = trace->result;
8114 ds_put_char_multiple(result, '\t', ctx->recurse);
8115 ds_put_cstr(result, s);
8116 ds_put_char(result, '\n');
8120 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
8121 void *aux OVS_UNUSED)
8123 const struct dpif_backer *backer;
8124 struct ofproto_dpif *ofproto;
8125 struct ofpbuf odp_key;
8126 struct ofpbuf *packet;
8127 struct initial_vals initial_vals;
8135 ofpbuf_init(&odp_key, 0);
8137 /* Handle "-generate" or a hex string as the last argument. */
8138 if (!strcmp(argv[argc - 1], "-generate")) {
8139 packet = ofpbuf_new(0);
8142 const char *error = eth_from_hex(argv[argc - 1], &packet);
8145 } else if (argc == 4) {
8146 /* The 3-argument form must end in "-generate' or a hex string. */
8147 unixctl_command_reply_error(conn, error);
8152 /* Parse the flow and determine whether a datapath or
8153 * bridge is specified. If function odp_flow_key_from_string()
8154 * returns 0, the flow is a odp_flow. If function
8155 * parse_ofp_exact_flow() returns 0, the flow is a br_flow. */
8156 if (!odp_flow_key_from_string(argv[argc - 1], NULL, &odp_key)) {
8157 /* If the odp_flow is the second argument,
8158 * the datapath name is the first argument. */
8160 const char *dp_type;
8161 if (!strncmp(argv[1], "ovs-", 4)) {
8162 dp_type = argv[1] + 4;
8166 backer = shash_find_data(&all_dpif_backers, dp_type);
8168 unixctl_command_reply_error(conn, "Cannot find datapath "
8173 /* No datapath name specified, so there should be only one
8175 struct shash_node *node;
8176 if (shash_count(&all_dpif_backers) != 1) {
8177 unixctl_command_reply_error(conn, "Must specify datapath "
8178 "name, there is more than one type of datapath");
8181 node = shash_first(&all_dpif_backers);
8182 backer = node->data;
8185 /* Extract the ofproto_dpif object from the ofproto_receive()
8187 if (ofproto_receive(backer, NULL, odp_key.data,
8188 odp_key.size, &flow, NULL, &ofproto, NULL,
8190 unixctl_command_reply_error(conn, "Invalid datapath flow");
8193 ds_put_format(&result, "Bridge: %s\n", ofproto->up.name);
8194 } else if (!parse_ofp_exact_flow(&flow, argv[argc - 1])) {
8196 unixctl_command_reply_error(conn, "Must specify bridge name");
8200 ofproto = ofproto_dpif_lookup(argv[1]);
8202 unixctl_command_reply_error(conn, "Unknown bridge name");
8205 initial_vals.vlan_tci = flow.vlan_tci;
8207 unixctl_command_reply_error(conn, "Bad flow syntax");
8211 /* Generate a packet, if requested. */
8213 if (!packet->size) {
8214 flow_compose(packet, &flow);
8216 ds_put_cstr(&result, "Packet: ");
8217 s = ofp_packet_to_string(packet->data, packet->size);
8218 ds_put_cstr(&result, s);
8221 /* Use the metadata from the flow and the packet argument
8222 * to reconstruct the flow. */
8223 flow_extract(packet, flow.skb_priority, flow.skb_mark, NULL,
8224 flow.in_port, &flow);
8225 initial_vals.vlan_tci = flow.vlan_tci;
8229 ofproto_trace(ofproto, &flow, packet, &initial_vals, &result);
8230 unixctl_command_reply(conn, ds_cstr(&result));
8233 ds_destroy(&result);
8234 ofpbuf_delete(packet);
8235 ofpbuf_uninit(&odp_key);
8239 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
8240 const struct ofpbuf *packet,
8241 const struct initial_vals *initial_vals, struct ds *ds)
8243 struct rule_dpif *rule;
8245 ds_put_cstr(ds, "Flow: ");
8246 flow_format(ds, flow);
8247 ds_put_char(ds, '\n');
8249 rule = rule_dpif_lookup(ofproto, flow);
8251 trace_format_rule(ds, 0, 0, rule);
8252 if (rule == ofproto->miss_rule) {
8253 ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
8254 } else if (rule == ofproto->no_packet_in_rule) {
8255 ds_put_cstr(ds, "\nNo match, packets dropped because "
8256 "OFPPC_NO_PACKET_IN is set on in_port.\n");
8260 uint64_t odp_actions_stub[1024 / 8];
8261 struct ofpbuf odp_actions;
8263 struct trace_ctx trace;
8266 tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
8269 ofpbuf_use_stub(&odp_actions,
8270 odp_actions_stub, sizeof odp_actions_stub);
8271 action_xlate_ctx_init(&trace.ctx, ofproto, flow, initial_vals,
8272 rule, tcp_flags, packet);
8273 trace.ctx.resubmit_hook = trace_resubmit;
8274 trace.ctx.report_hook = trace_report;
8275 xlate_actions(&trace.ctx, rule->up.ofpacts, rule->up.ofpacts_len,
8278 ds_put_char(ds, '\n');
8279 trace_format_flow(ds, 0, "Final flow", &trace);
8280 ds_put_cstr(ds, "Datapath actions: ");
8281 format_odp_actions(ds, odp_actions.data, odp_actions.size);
8282 ofpbuf_uninit(&odp_actions);
8284 if (trace.ctx.slow) {
8285 ds_put_cstr(ds, "\nThis flow is handled by the userspace "
8286 "slow path because it:");
8287 switch (trace.ctx.slow) {
8289 ds_put_cstr(ds, "\n\t- Consists of CFM packets.");
8292 ds_put_cstr(ds, "\n\t- Consists of LACP packets.");
8295 ds_put_cstr(ds, "\n\t- Consists of STP packets.");
8298 ds_put_cstr(ds, "\n\t- Consists of BFD packets.");
8300 case SLOW_CONTROLLER:
8301 ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages "
8302 "to the OpenFlow controller.");
8312 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
8313 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
8316 unixctl_command_reply(conn, NULL);
8320 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
8321 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
8324 unixctl_command_reply(conn, NULL);
8327 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
8328 * 'reply' describing the results. */
8330 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
8332 struct facet *facet;
8336 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
8337 if (!facet_check_consistency(facet)) {
8342 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
8346 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
8347 ofproto->up.name, errors);
8349 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
8354 ofproto_dpif_self_check(struct unixctl_conn *conn,
8355 int argc, const char *argv[], void *aux OVS_UNUSED)
8357 struct ds reply = DS_EMPTY_INITIALIZER;
8358 struct ofproto_dpif *ofproto;
8361 ofproto = ofproto_dpif_lookup(argv[1]);
8363 unixctl_command_reply_error(conn, "Unknown ofproto (use "
8364 "ofproto/list for help)");
8367 ofproto_dpif_self_check__(ofproto, &reply);
8369 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
8370 ofproto_dpif_self_check__(ofproto, &reply);
8374 unixctl_command_reply(conn, ds_cstr(&reply));
8378 /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list
8379 * of the 'ofproto_shash' nodes. It is the responsibility of the caller
8380 * to destroy 'ofproto_shash' and free the returned value. */
8381 static const struct shash_node **
8382 get_ofprotos(struct shash *ofproto_shash)
8384 const struct ofproto_dpif *ofproto;
8386 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
8387 char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
8388 shash_add_nocopy(ofproto_shash, name, ofproto);
8391 return shash_sort(ofproto_shash);
8395 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
8396 const char *argv[] OVS_UNUSED,
8397 void *aux OVS_UNUSED)
8399 struct ds ds = DS_EMPTY_INITIALIZER;
8400 struct shash ofproto_shash;
8401 const struct shash_node **sorted_ofprotos;
8404 shash_init(&ofproto_shash);
8405 sorted_ofprotos = get_ofprotos(&ofproto_shash);
8406 for (i = 0; i < shash_count(&ofproto_shash); i++) {
8407 const struct shash_node *node = sorted_ofprotos[i];
8408 ds_put_format(&ds, "%s\n", node->name);
8411 shash_destroy(&ofproto_shash);
8412 free(sorted_ofprotos);
8414 unixctl_command_reply(conn, ds_cstr(&ds));
8419 show_dp_format(const struct ofproto_dpif *ofproto, struct ds *ds)
8421 const struct shash_node **ports;
8423 struct avg_subfacet_rates lifetime;
8424 unsigned long long int minutes;
8425 const int min_ms = 60 * 1000; /* milliseconds in one minute. */
8427 minutes = (time_msec() - ofproto->created) / min_ms;
8430 lifetime.add_rate = (double)ofproto->total_subfacet_add_count
8432 lifetime.del_rate = (double)ofproto->total_subfacet_del_count
8435 lifetime.add_rate = 0.0;
8436 lifetime.del_rate = 0.0;
8439 ds_put_format(ds, "%s (%s):\n", ofproto->up.name,
8440 dpif_name(ofproto->backer->dpif));
8442 "\tlookups: hit:%"PRIu64" missed:%"PRIu64"\n",
8443 ofproto->n_hit, ofproto->n_missed);
8444 ds_put_format(ds, "\tflows: cur: %zu, avg: %5.3f, max: %d,"
8445 " life span: %llu(ms)\n",
8446 hmap_count(&ofproto->subfacets),
8447 avg_subfacet_count(ofproto),
8448 ofproto->max_n_subfacet,
8449 avg_subfacet_life_span(ofproto));
8450 if (minutes >= 60) {
8451 show_dp_rates(ds, "\t\thourly avg:", &ofproto->hourly);
8453 if (minutes >= 60 * 24) {
8454 show_dp_rates(ds, "\t\tdaily avg:", &ofproto->daily);
8456 show_dp_rates(ds, "\t\toverall avg:", &lifetime);
8458 ports = shash_sort(&ofproto->up.port_by_name);
8459 for (i = 0; i < shash_count(&ofproto->up.port_by_name); i++) {
8460 const struct shash_node *node = ports[i];
8461 struct ofport *ofport = node->data;
8462 const char *name = netdev_get_name(ofport->netdev);
8463 const char *type = netdev_get_type(ofport->netdev);
8466 ds_put_format(ds, "\t%s %u/", name, ofport->ofp_port);
8468 odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
8469 if (odp_port != OVSP_NONE) {
8470 ds_put_format(ds, "%"PRIu32":", odp_port);
8472 ds_put_cstr(ds, "none:");
8475 if (strcmp(type, "system")) {
8476 struct netdev *netdev;
8479 ds_put_format(ds, " (%s", type);
8481 error = netdev_open(name, type, &netdev);
8486 error = netdev_get_config(netdev, &config);
8488 const struct smap_node **nodes;
8491 nodes = smap_sort(&config);
8492 for (i = 0; i < smap_count(&config); i++) {
8493 const struct smap_node *node = nodes[i];
8494 ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
8495 node->key, node->value);
8499 smap_destroy(&config);
8501 netdev_close(netdev);
8503 ds_put_char(ds, ')');
8505 ds_put_char(ds, '\n');
8511 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc,
8512 const char *argv[], void *aux OVS_UNUSED)
8514 struct ds ds = DS_EMPTY_INITIALIZER;
8515 const struct ofproto_dpif *ofproto;
8519 for (i = 1; i < argc; i++) {
8520 ofproto = ofproto_dpif_lookup(argv[i]);
8522 ds_put_format(&ds, "Unknown bridge %s (use dpif/dump-dps "
8523 "for help)", argv[i]);
8524 unixctl_command_reply_error(conn, ds_cstr(&ds));
8527 show_dp_format(ofproto, &ds);
8530 struct shash ofproto_shash;
8531 const struct shash_node **sorted_ofprotos;
8534 shash_init(&ofproto_shash);
8535 sorted_ofprotos = get_ofprotos(&ofproto_shash);
8536 for (i = 0; i < shash_count(&ofproto_shash); i++) {
8537 const struct shash_node *node = sorted_ofprotos[i];
8538 show_dp_format(node->data, &ds);
8541 shash_destroy(&ofproto_shash);
8542 free(sorted_ofprotos);
8545 unixctl_command_reply(conn, ds_cstr(&ds));
8550 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
8551 int argc OVS_UNUSED, const char *argv[],
8552 void *aux OVS_UNUSED)
8554 struct ds ds = DS_EMPTY_INITIALIZER;
8555 const struct ofproto_dpif *ofproto;
8556 struct subfacet *subfacet;
8558 ofproto = ofproto_dpif_lookup(argv[1]);
8560 unixctl_command_reply_error(conn, "no such bridge");
8564 update_stats(ofproto->backer);
8566 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
8567 odp_flow_key_format(subfacet->key, subfacet->key_len, &ds);
8569 ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:",
8570 subfacet->dp_packet_count, subfacet->dp_byte_count);
8571 if (subfacet->used) {
8572 ds_put_format(&ds, "%.3fs",
8573 (time_msec() - subfacet->used) / 1000.0);
8575 ds_put_format(&ds, "never");
8577 if (subfacet->facet->tcp_flags) {
8578 ds_put_cstr(&ds, ", flags:");
8579 packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags);
8582 ds_put_cstr(&ds, ", actions:");
8583 if (subfacet->slow) {
8584 uint64_t slow_path_stub[128 / 8];
8585 const struct nlattr *actions;
8588 compose_slow_path(ofproto, &subfacet->facet->flow, subfacet->slow,
8589 slow_path_stub, sizeof slow_path_stub,
8590 &actions, &actions_len);
8591 format_odp_actions(&ds, actions, actions_len);
8593 format_odp_actions(&ds, subfacet->actions, subfacet->actions_len);
8595 ds_put_char(&ds, '\n');
8598 unixctl_command_reply(conn, ds_cstr(&ds));
8603 ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn,
8604 int argc OVS_UNUSED, const char *argv[],
8605 void *aux OVS_UNUSED)
8607 struct ds ds = DS_EMPTY_INITIALIZER;
8608 struct ofproto_dpif *ofproto;
8610 ofproto = ofproto_dpif_lookup(argv[1]);
8612 unixctl_command_reply_error(conn, "no such bridge");
8616 flush(&ofproto->up);
8618 unixctl_command_reply(conn, ds_cstr(&ds));
8623 ofproto_dpif_unixctl_init(void)
8625 static bool registered;
8631 unixctl_command_register(
8633 "[dp_name]|bridge odp_flow|br_flow [-generate|packet]",
8634 1, 3, ofproto_unixctl_trace, NULL);
8635 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
8636 ofproto_unixctl_fdb_flush, NULL);
8637 unixctl_command_register("fdb/show", "bridge", 1, 1,
8638 ofproto_unixctl_fdb_show, NULL);
8639 unixctl_command_register("ofproto/clog", "", 0, 0,
8640 ofproto_dpif_clog, NULL);
8641 unixctl_command_register("ofproto/unclog", "", 0, 0,
8642 ofproto_dpif_unclog, NULL);
8643 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
8644 ofproto_dpif_self_check, NULL);
8645 unixctl_command_register("dpif/dump-dps", "", 0, 0,
8646 ofproto_unixctl_dpif_dump_dps, NULL);
8647 unixctl_command_register("dpif/show", "[bridge]", 0, INT_MAX,
8648 ofproto_unixctl_dpif_show, NULL);
8649 unixctl_command_register("dpif/dump-flows", "bridge", 1, 1,
8650 ofproto_unixctl_dpif_dump_flows, NULL);
8651 unixctl_command_register("dpif/del-flows", "bridge", 1, 1,
8652 ofproto_unixctl_dpif_del_flows, NULL);
8655 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
8657 * This is deprecated. It is only for compatibility with broken device drivers
8658 * in old versions of Linux that do not properly support VLANs when VLAN
8659 * devices are not used. When broken device drivers are no longer in
8660 * widespread use, we will delete these interfaces. */
8663 set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid)
8665 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
8666 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
8668 if (realdev_ofp_port == ofport->realdev_ofp_port
8669 && vid == ofport->vlandev_vid) {
8673 ofproto->backer->need_revalidate = REV_RECONFIGURE;
8675 if (ofport->realdev_ofp_port) {
8678 if (realdev_ofp_port && ofport->bundle) {
8679 /* vlandevs are enslaved to their realdevs, so they are not allowed to
8680 * themselves be part of a bundle. */
8681 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
8684 ofport->realdev_ofp_port = realdev_ofp_port;
8685 ofport->vlandev_vid = vid;
8687 if (realdev_ofp_port) {
8688 vsp_add(ofport, realdev_ofp_port, vid);
8695 hash_realdev_vid(uint16_t realdev_ofp_port, int vid)
8697 return hash_2words(realdev_ofp_port, vid);
8700 /* Returns the ODP port number of the Linux VLAN device that corresponds to
8701 * 'vlan_tci' on the network device with port number 'realdev_odp_port' in
8702 * 'ofproto'. For example, given 'realdev_odp_port' of eth0 and 'vlan_tci' 9,
8703 * it would return the port number of eth0.9.
8705 * Unless VLAN splinters are enabled for port 'realdev_odp_port', this
8706 * function just returns its 'realdev_odp_port' argument. */
8708 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
8709 uint32_t realdev_odp_port, ovs_be16 vlan_tci)
8711 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
8712 uint16_t realdev_ofp_port;
8713 int vid = vlan_tci_to_vid(vlan_tci);
8714 const struct vlan_splinter *vsp;
8716 realdev_ofp_port = odp_port_to_ofp_port(ofproto, realdev_odp_port);
8717 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
8718 hash_realdev_vid(realdev_ofp_port, vid),
8719 &ofproto->realdev_vid_map) {
8720 if (vsp->realdev_ofp_port == realdev_ofp_port
8721 && vsp->vid == vid) {
8722 return ofp_port_to_odp_port(ofproto, vsp->vlandev_ofp_port);
8726 return realdev_odp_port;
8729 static struct vlan_splinter *
8730 vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port)
8732 struct vlan_splinter *vsp;
8734 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, hash_int(vlandev_ofp_port, 0),
8735 &ofproto->vlandev_map) {
8736 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
8744 /* Returns the OpenFlow port number of the "real" device underlying the Linux
8745 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
8746 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
8747 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
8748 * eth0 and store 9 in '*vid'.
8750 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
8751 * VLAN device. Unless VLAN splinters are enabled, this is what this function
8754 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
8755 uint16_t vlandev_ofp_port, int *vid)
8757 if (!hmap_is_empty(&ofproto->vlandev_map)) {
8758 const struct vlan_splinter *vsp;
8760 vsp = vlandev_find(ofproto, vlandev_ofp_port);
8765 return vsp->realdev_ofp_port;
8771 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
8772 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
8773 * 'flow->in_port' to the "real" device backing the VLAN device, sets
8774 * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is
8775 * always the case unless VLAN splinters are enabled), returns false without
8776 * making any changes. */
8778 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
8783 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port, &vid);
8788 /* Cause the flow to be processed as if it came in on the real device with
8789 * the VLAN device's VLAN ID. */
8790 flow->in_port = realdev;
8791 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
8796 vsp_remove(struct ofport_dpif *port)
8798 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
8799 struct vlan_splinter *vsp;
8801 vsp = vlandev_find(ofproto, port->up.ofp_port);
8803 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
8804 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
8807 port->realdev_ofp_port = 0;
8809 VLOG_ERR("missing vlan device record");
8814 vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid)
8816 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
8818 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
8819 && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid))
8820 == realdev_ofp_port)) {
8821 struct vlan_splinter *vsp;
8823 vsp = xmalloc(sizeof *vsp);
8824 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
8825 hash_int(port->up.ofp_port, 0));
8826 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
8827 hash_realdev_vid(realdev_ofp_port, vid));
8828 vsp->realdev_ofp_port = realdev_ofp_port;
8829 vsp->vlandev_ofp_port = port->up.ofp_port;
8832 port->realdev_ofp_port = realdev_ofp_port;
8834 VLOG_ERR("duplicate vlan device record");
8839 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
8841 const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
8842 return ofport ? ofport->odp_port : OVSP_NONE;
8845 static struct ofport_dpif *
8846 odp_port_to_ofport(const struct dpif_backer *backer, uint32_t odp_port)
8848 struct ofport_dpif *port;
8850 HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node,
8851 hash_int(odp_port, 0),
8852 &backer->odp_to_ofport_map) {
8853 if (port->odp_port == odp_port) {
8862 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
8864 struct ofport_dpif *port;
8866 port = odp_port_to_ofport(ofproto->backer, odp_port);
8867 if (port && &ofproto->up == port->up.ofproto) {
8868 return port->up.ofp_port;
8873 static unsigned long long int
8874 avg_subfacet_life_span(const struct ofproto_dpif *ofproto)
8876 unsigned long long int dc;
8877 unsigned long long int avg;
8879 dc = ofproto->total_subfacet_del_count + ofproto->subfacet_del_count;
8880 avg = dc ? ofproto->total_subfacet_life_span / dc : 0;
8886 avg_subfacet_count(const struct ofproto_dpif *ofproto)
8890 if (ofproto->n_update_stats) {
8891 avg_c = (double)ofproto->total_subfacet_count
8892 / ofproto->n_update_stats;
8899 show_dp_rates(struct ds *ds, const char *heading,
8900 const struct avg_subfacet_rates *rates)
8902 ds_put_format(ds, "%s add rate: %5.3f/min, del rate: %5.3f/min\n",
8903 heading, rates->add_rate, rates->del_rate);
8907 update_max_subfacet_count(struct ofproto_dpif *ofproto)
8909 ofproto->max_n_subfacet = MAX(ofproto->max_n_subfacet,
8910 hmap_count(&ofproto->subfacets));
8913 /* Compute exponentially weighted moving average, adding 'new' as the newest,
8914 * most heavily weighted element. 'base' designates the rate of decay: after
8915 * 'base' further updates, 'new''s weight in the EWMA decays to about 1/e
8918 exp_mavg(double *avg, int base, double new)
8920 *avg = (*avg * (base - 1) + new) / base;
8924 update_moving_averages(struct ofproto_dpif *ofproto)
8926 const int min_ms = 60 * 1000; /* milliseconds in one minute. */
8928 /* Update hourly averages on the minute boundaries. */
8929 if (time_msec() - ofproto->last_minute >= min_ms) {
8930 exp_mavg(&ofproto->hourly.add_rate, 60, ofproto->subfacet_add_count);
8931 exp_mavg(&ofproto->hourly.del_rate, 60, ofproto->subfacet_del_count);
8933 /* Update daily averages on the hour boundaries. */
8934 if ((ofproto->last_minute - ofproto->created) / min_ms % 60 == 59) {
8935 exp_mavg(&ofproto->daily.add_rate, 24, ofproto->hourly.add_rate);
8936 exp_mavg(&ofproto->daily.del_rate, 24, ofproto->hourly.del_rate);
8939 ofproto->total_subfacet_add_count += ofproto->subfacet_add_count;
8940 ofproto->total_subfacet_del_count += ofproto->subfacet_del_count;
8941 ofproto->subfacet_add_count = 0;
8942 ofproto->subfacet_del_count = 0;
8943 ofproto->last_minute += min_ms;
8948 dpif_stats_update_hit_count(struct ofproto_dpif *ofproto, uint64_t delta)
8950 ofproto->n_hit += delta;
8953 const struct ofproto_class ofproto_dpif_class = {
8988 port_is_lacp_current,
8989 NULL, /* rule_choose_table */
8996 rule_modify_actions,
9010 get_stp_port_status,
9017 is_mirror_output_bundle,
9018 forward_bpdu_changed,
9019 set_mac_table_config,