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 struct ofpbuf *odp_actions);
428 static int subfacet_install(struct subfacet *,
429 const struct nlattr *actions, size_t actions_len,
430 struct dpif_flow_stats *, enum slow_path_reason);
431 static void subfacet_uninstall(struct subfacet *);
433 static enum subfacet_path subfacet_want_path(enum slow_path_reason);
435 /* An exact-match instantiation of an OpenFlow flow.
437 * A facet associates a "struct flow", which represents the Open vSwitch
438 * userspace idea of an exact-match flow, with one or more subfacets. Each
439 * subfacet tracks the datapath's idea of the exact-match flow equivalent to
440 * the facet. When the kernel module (or other dpif implementation) and Open
441 * vSwitch userspace agree on the definition of a flow key, there is exactly
442 * one subfacet per facet. If the dpif implementation supports more-specific
443 * flow matching than userspace, however, a facet can have more than one
444 * subfacet, each of which corresponds to some distinction in flow that
445 * userspace simply doesn't understand.
447 * Flow expiration works in terms of subfacets, so a facet must have at least
448 * one subfacet or it will never expire, leaking memory. */
451 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
452 struct list list_node; /* In owning rule's 'facets' list. */
453 struct rule_dpif *rule; /* Owning rule. */
456 struct list subfacets;
457 long long int used; /* Time last used; time created if not used. */
464 * - Do include packets and bytes sent "by hand", e.g. with
467 * - Do include packets and bytes that were obtained from the datapath
468 * when a subfacet's statistics were reset (e.g. dpif_flow_put() with
469 * DPIF_FP_ZERO_STATS).
471 * - Do not include packets or bytes that can be obtained from the
472 * datapath for any existing subfacet.
474 uint64_t packet_count; /* Number of packets received. */
475 uint64_t byte_count; /* Number of bytes received. */
477 /* Resubmit statistics. */
478 uint64_t prev_packet_count; /* Number of packets from last stats push. */
479 uint64_t prev_byte_count; /* Number of bytes from last stats push. */
480 long long int prev_used; /* Used time from last stats push. */
483 uint64_t accounted_bytes; /* Bytes processed by facet_account(). */
484 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
485 uint8_t tcp_flags; /* TCP flags seen for this 'rule'. */
487 /* Properties of datapath actions.
489 * Every subfacet has its own actions because actions can differ slightly
490 * between splintered and non-splintered subfacets due to the VLAN tag
491 * being initially different (present vs. absent). All of them have these
492 * properties in common so we just store one copy of them here. */
493 bool has_learn; /* Actions include NXAST_LEARN? */
494 bool has_normal; /* Actions output to OFPP_NORMAL? */
495 bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */
496 tag_type tags; /* Tags that would require revalidation. */
497 mirror_mask_t mirrors; /* Bitmap of dependent mirrors. */
499 /* Storage for a single subfacet, to reduce malloc() time and space
500 * overhead. (A facet always has at least one subfacet and in the common
501 * case has exactly one subfacet. However, 'one_subfacet' may not
502 * always be valid, since it could have been removed after newer
503 * subfacets were pushed onto the 'subfacets' list.) */
504 struct subfacet one_subfacet;
506 long long int learn_rl; /* Rate limiter for facet_learn(). */
509 static struct facet *facet_create(struct rule_dpif *,
510 const struct flow *, uint32_t hash);
511 static void facet_remove(struct facet *);
512 static void facet_free(struct facet *);
514 static struct facet *facet_find(struct ofproto_dpif *,
515 const struct flow *, uint32_t hash);
516 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
517 const struct flow *, uint32_t hash);
518 static void facet_revalidate(struct facet *);
519 static bool facet_check_consistency(struct facet *);
521 static void facet_flush_stats(struct facet *);
523 static void facet_update_time(struct facet *, long long int used);
524 static void facet_reset_counters(struct facet *);
525 static void facet_push_stats(struct facet *);
526 static void facet_learn(struct facet *);
527 static void facet_account(struct facet *);
528 static void push_all_stats(void);
530 static struct subfacet *facet_get_subfacet(struct facet *);
532 static bool facet_is_controller_flow(struct facet *);
535 struct hmap_node odp_port_node; /* In dpif_backer's "odp_to_ofport_map". */
539 struct ofbundle *bundle; /* Bundle that contains this port, if any. */
540 struct list bundle_node; /* In struct ofbundle's "ports" list. */
541 struct cfm *cfm; /* Connectivity Fault Management, if any. */
542 struct bfd *bfd; /* BFD, if any. */
543 tag_type tag; /* Tag associated with this port. */
544 bool may_enable; /* May be enabled in bonds. */
545 long long int carrier_seq; /* Carrier status changes. */
546 struct tnl_port *tnl_port; /* Tunnel handle, or null. */
549 struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */
550 enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */
551 long long int stp_state_entered;
553 struct hmap priorities; /* Map of attached 'priority_to_dscp's. */
555 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
557 * This is deprecated. It is only for compatibility with broken device
558 * drivers in old versions of Linux that do not properly support VLANs when
559 * VLAN devices are not used. When broken device drivers are no longer in
560 * widespread use, we will delete these interfaces. */
561 uint16_t realdev_ofp_port;
565 /* Node in 'ofport_dpif''s 'priorities' map. Used to maintain a map from
566 * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
567 * traffic egressing the 'ofport' with that priority should be marked with. */
568 struct priority_to_dscp {
569 struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'priorities' map. */
570 uint32_t priority; /* Priority of this queue (see struct flow). */
572 uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */
575 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
577 * This is deprecated. It is only for compatibility with broken device drivers
578 * in old versions of Linux that do not properly support VLANs when VLAN
579 * devices are not used. When broken device drivers are no longer in
580 * widespread use, we will delete these interfaces. */
581 struct vlan_splinter {
582 struct hmap_node realdev_vid_node;
583 struct hmap_node vlandev_node;
584 uint16_t realdev_ofp_port;
585 uint16_t vlandev_ofp_port;
589 static uint32_t vsp_realdev_to_vlandev(const struct ofproto_dpif *,
590 uint32_t realdev, ovs_be16 vlan_tci);
591 static bool vsp_adjust_flow(const struct ofproto_dpif *, struct flow *);
592 static void vsp_remove(struct ofport_dpif *);
593 static void vsp_add(struct ofport_dpif *, uint16_t realdev_ofp_port, int vid);
595 static uint32_t ofp_port_to_odp_port(const struct ofproto_dpif *,
597 static uint16_t odp_port_to_ofp_port(const struct ofproto_dpif *,
600 static struct ofport_dpif *
601 ofport_dpif_cast(const struct ofport *ofport)
603 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
606 static void port_run(struct ofport_dpif *);
607 static void port_run_fast(struct ofport_dpif *);
608 static void port_wait(struct ofport_dpif *);
609 static int set_bfd(struct ofport *, const struct smap *);
610 static int set_cfm(struct ofport *, const struct cfm_settings *);
611 static void ofport_clear_priorities(struct ofport_dpif *);
612 static void run_fast_rl(void);
614 struct dpif_completion {
615 struct list list_node;
616 struct ofoperation *op;
619 /* Extra information about a classifier table.
620 * Currently used just for optimized flow revalidation. */
622 /* If either of these is nonnull, then this table has a form that allows
623 * flows to be tagged to avoid revalidating most flows for the most common
624 * kinds of flow table changes. */
625 struct cls_table *catchall_table; /* Table that wildcards all fields. */
626 struct cls_table *other_table; /* Table with any other wildcard set. */
627 uint32_t basis; /* Keeps each table's tags separate. */
630 /* Reasons that we might need to revalidate every facet, and corresponding
633 * A value of 0 means that there is no need to revalidate.
635 * It would be nice to have some cleaner way to integrate with coverage
636 * counters, but with only a few reasons I guess this is good enough for
638 enum revalidate_reason {
639 REV_RECONFIGURE = 1, /* Switch configuration changed. */
640 REV_STP, /* Spanning tree protocol port status change. */
641 REV_PORT_TOGGLED, /* Port enabled or disabled by CFM, LACP, ...*/
642 REV_FLOW_TABLE, /* Flow table changed. */
643 REV_INCONSISTENCY /* Facet self-check failed. */
645 COVERAGE_DEFINE(rev_reconfigure);
646 COVERAGE_DEFINE(rev_stp);
647 COVERAGE_DEFINE(rev_port_toggled);
648 COVERAGE_DEFINE(rev_flow_table);
649 COVERAGE_DEFINE(rev_inconsistency);
651 /* Drop keys are odp flow keys which have drop flows installed in the kernel.
652 * These are datapath flows which have no associated ofproto, if they did we
653 * would use facets. */
655 struct hmap_node hmap_node;
660 /* All datapaths of a given type share a single dpif backer instance. */
665 struct timer next_expiration;
666 struct hmap odp_to_ofport_map; /* ODP port to ofport mapping. */
668 struct simap tnl_backers; /* Set of dpif ports backing tunnels. */
670 /* Facet revalidation flags applying to facets which use this backer. */
671 enum revalidate_reason need_revalidate; /* Revalidate every facet. */
672 struct tag_set revalidate_set; /* Revalidate only matching facets. */
674 struct hmap drop_keys; /* Set of dropped odp keys. */
677 /* All existing ofproto_backer instances, indexed by ofproto->up.type. */
678 static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers);
680 static void drop_key_clear(struct dpif_backer *);
681 static struct ofport_dpif *
682 odp_port_to_ofport(const struct dpif_backer *, uint32_t odp_port);
684 static void dpif_stats_update_hit_count(struct ofproto_dpif *ofproto,
686 struct avg_subfacet_rates {
687 double add_rate; /* Moving average of new flows created per minute. */
688 double del_rate; /* Moving average of flows deleted per minute. */
690 static void show_dp_rates(struct ds *ds, const char *heading,
691 const struct avg_subfacet_rates *rates);
692 static void exp_mavg(double *avg, int base, double new);
694 struct ofproto_dpif {
695 struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
697 struct dpif_backer *backer;
699 /* Special OpenFlow rules. */
700 struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */
701 struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */
707 struct netflow *netflow;
708 struct dpif_sflow *sflow;
709 struct dpif_ipfix *ipfix;
710 struct hmap bundles; /* Contains "struct ofbundle"s. */
711 struct mac_learning *ml;
712 struct ofmirror *mirrors[MAX_MIRRORS];
714 bool has_bonded_bundles;
718 struct hmap subfacets;
719 struct governor *governor;
720 long long int consistency_rl;
723 struct table_dpif tables[N_TABLES];
725 /* Support for debugging async flow mods. */
726 struct list completions;
728 bool has_bundle_action; /* True when the first bundle action appears. */
729 struct netdev_stats stats; /* To account packets generated and consumed in
734 long long int stp_last_tick;
736 /* VLAN splinters. */
737 struct hmap realdev_vid_map; /* (realdev,vid) -> vlandev. */
738 struct hmap vlandev_map; /* vlandev -> (realdev,vid). */
741 struct sset ports; /* Set of standard port names. */
742 struct sset ghost_ports; /* Ports with no datapath port. */
743 struct sset port_poll_set; /* Queued names for port_poll() reply. */
744 int port_poll_errno; /* Last errno for port_poll() reply. */
746 /* Per ofproto's dpif stats. */
750 /* Subfacet statistics.
752 * These keep track of the total number of subfacets added and deleted and
753 * flow life span. They are useful for computing the flow rates stats
754 * exposed via "ovs-appctl dpif/show". The goal is to learn about
755 * traffic patterns in ways that we can use later to improve Open vSwitch
756 * performance in new situations. */
757 long long int created; /* Time when it is created. */
758 unsigned int max_n_subfacet; /* Maximum number of flows */
760 /* The average number of subfacets... */
761 struct avg_subfacet_rates hourly; /* ...over the last hour. */
762 struct avg_subfacet_rates daily; /* ...over the last day. */
763 long long int last_minute; /* Last time 'hourly' was updated. */
765 /* Number of subfacets added or deleted since 'last_minute'. */
766 unsigned int subfacet_add_count;
767 unsigned int subfacet_del_count;
769 /* Number of subfacets added or deleted from 'created' to 'last_minute.' */
770 unsigned long long int total_subfacet_add_count;
771 unsigned long long int total_subfacet_del_count;
773 /* Sum of the number of milliseconds that each subfacet existed,
774 * over the subfacets that have been added and then later deleted. */
775 unsigned long long int total_subfacet_life_span;
777 /* Incremented by the number of currently existing subfacets, each
778 * time we pull statistics from the kernel. */
779 unsigned long long int total_subfacet_count;
781 /* Number of times we pull statistics from the kernel. */
782 unsigned long long int n_update_stats;
784 static unsigned long long int avg_subfacet_life_span(
785 const struct ofproto_dpif *);
786 static double avg_subfacet_count(const struct ofproto_dpif *ofproto);
787 static void update_moving_averages(struct ofproto_dpif *ofproto);
788 static void dpif_stats_update_hit_count(struct ofproto_dpif *ofproto,
790 static void update_max_subfacet_count(struct ofproto_dpif *ofproto);
792 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
793 * for debugging the asynchronous flow_mod implementation.) */
796 /* All existing ofproto_dpif instances, indexed by ->up.name. */
797 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
799 static void ofproto_dpif_unixctl_init(void);
801 static struct ofproto_dpif *
802 ofproto_dpif_cast(const struct ofproto *ofproto)
804 ovs_assert(ofproto->ofproto_class == &ofproto_dpif_class);
805 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
808 static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *,
810 static struct ofport_dpif *get_odp_port(const struct ofproto_dpif *,
812 static void ofproto_trace(struct ofproto_dpif *, const struct flow *,
813 const struct ofpbuf *,
814 const struct initial_vals *, struct ds *);
816 /* Packet processing. */
817 static void update_learning_table(struct ofproto_dpif *,
818 const struct flow *, int vlan,
821 #define FLOW_MISS_MAX_BATCH 50
822 static int handle_upcalls(struct dpif_backer *, unsigned int max_batch);
824 /* Flow expiration. */
825 static int expire(struct dpif_backer *);
828 static void send_netflow_active_timeouts(struct ofproto_dpif *);
831 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
832 static size_t compose_sflow_action(const struct ofproto_dpif *,
833 struct ofpbuf *odp_actions,
834 const struct flow *, uint32_t odp_port);
835 static void compose_ipfix_action(const struct ofproto_dpif *,
836 struct ofpbuf *odp_actions,
837 const struct flow *);
838 static void add_mirror_actions(struct action_xlate_ctx *ctx,
839 const struct flow *flow);
840 /* Global variables. */
841 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
843 /* Initial mappings of port to bridge mappings. */
844 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
846 /* Factory functions. */
849 init(const struct shash *iface_hints)
851 struct shash_node *node;
853 /* Make a local copy, since we don't own 'iface_hints' elements. */
854 SHASH_FOR_EACH(node, iface_hints) {
855 const struct iface_hint *orig_hint = node->data;
856 struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
858 new_hint->br_name = xstrdup(orig_hint->br_name);
859 new_hint->br_type = xstrdup(orig_hint->br_type);
860 new_hint->ofp_port = orig_hint->ofp_port;
862 shash_add(&init_ofp_ports, node->name, new_hint);
867 enumerate_types(struct sset *types)
869 dp_enumerate_types(types);
873 enumerate_names(const char *type, struct sset *names)
875 struct ofproto_dpif *ofproto;
878 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
879 if (strcmp(type, ofproto->up.type)) {
882 sset_add(names, ofproto->up.name);
889 del(const char *type, const char *name)
894 error = dpif_open(name, type, &dpif);
896 error = dpif_delete(dpif);
903 port_open_type(const char *datapath_type, const char *port_type)
905 return dpif_port_open_type(datapath_type, port_type);
908 /* Type functions. */
910 static struct ofproto_dpif *
911 lookup_ofproto_dpif_by_port_name(const char *name)
913 struct ofproto_dpif *ofproto;
915 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
916 if (sset_contains(&ofproto->ports, name)) {
925 type_run(const char *type)
927 static long long int push_timer = LLONG_MIN;
928 struct dpif_backer *backer;
932 backer = shash_find_data(&all_dpif_backers, type);
934 /* This is not necessarily a problem, since backers are only
935 * created on demand. */
939 dpif_run(backer->dpif);
941 /* The most natural place to push facet statistics is when they're pulled
942 * from the datapath. However, when there are many flows in the datapath,
943 * this expensive operation can occur so frequently, that it reduces our
944 * ability to quickly set up flows. To reduce the cost, we push statistics
946 if (time_msec() > push_timer) {
947 push_timer = time_msec() + 2000;
951 if (backer->need_revalidate
952 || !tag_set_is_empty(&backer->revalidate_set)) {
953 struct tag_set revalidate_set = backer->revalidate_set;
954 bool need_revalidate = backer->need_revalidate;
955 struct ofproto_dpif *ofproto;
956 struct simap_node *node;
957 struct simap tmp_backers;
959 /* Handle tunnel garbage collection. */
960 simap_init(&tmp_backers);
961 simap_swap(&backer->tnl_backers, &tmp_backers);
963 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
964 struct ofport_dpif *iter;
966 if (backer != ofproto->backer) {
970 HMAP_FOR_EACH (iter, up.hmap_node, &ofproto->up.ports) {
973 if (!iter->tnl_port) {
977 dp_port = netdev_vport_get_dpif_port(iter->up.netdev);
978 node = simap_find(&tmp_backers, dp_port);
980 simap_put(&backer->tnl_backers, dp_port, node->data);
981 simap_delete(&tmp_backers, node);
982 node = simap_find(&backer->tnl_backers, dp_port);
984 node = simap_find(&backer->tnl_backers, dp_port);
986 uint32_t odp_port = UINT32_MAX;
988 if (!dpif_port_add(backer->dpif, iter->up.netdev,
990 simap_put(&backer->tnl_backers, dp_port, odp_port);
991 node = simap_find(&backer->tnl_backers, dp_port);
996 iter->odp_port = node ? node->data : OVSP_NONE;
997 if (tnl_port_reconfigure(&iter->up, iter->odp_port,
999 backer->need_revalidate = REV_RECONFIGURE;
1004 SIMAP_FOR_EACH (node, &tmp_backers) {
1005 dpif_port_del(backer->dpif, node->data);
1007 simap_destroy(&tmp_backers);
1009 switch (backer->need_revalidate) {
1010 case REV_RECONFIGURE: COVERAGE_INC(rev_reconfigure); break;
1011 case REV_STP: COVERAGE_INC(rev_stp); break;
1012 case REV_PORT_TOGGLED: COVERAGE_INC(rev_port_toggled); break;
1013 case REV_FLOW_TABLE: COVERAGE_INC(rev_flow_table); break;
1014 case REV_INCONSISTENCY: COVERAGE_INC(rev_inconsistency); break;
1017 if (backer->need_revalidate) {
1018 /* Clear the drop_keys in case we should now be accepting some
1019 * formerly dropped flows. */
1020 drop_key_clear(backer);
1023 /* Clear the revalidation flags. */
1024 tag_set_init(&backer->revalidate_set);
1025 backer->need_revalidate = 0;
1027 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1028 struct facet *facet, *next;
1030 if (ofproto->backer != backer) {
1034 HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &ofproto->facets) {
1036 || tag_set_intersects(&revalidate_set, facet->tags)) {
1037 facet_revalidate(facet);
1044 if (timer_expired(&backer->next_expiration)) {
1045 int delay = expire(backer);
1046 timer_set_duration(&backer->next_expiration, delay);
1049 /* Check for port changes in the dpif. */
1050 while ((error = dpif_port_poll(backer->dpif, &devname)) == 0) {
1051 struct ofproto_dpif *ofproto;
1052 struct dpif_port port;
1054 /* Don't report on the datapath's device. */
1055 if (!strcmp(devname, dpif_base_name(backer->dpif))) {
1059 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
1060 &all_ofproto_dpifs) {
1061 if (simap_contains(&ofproto->backer->tnl_backers, devname)) {
1066 ofproto = lookup_ofproto_dpif_by_port_name(devname);
1067 if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
1068 /* The port was removed. If we know the datapath,
1069 * report it through poll_set(). If we don't, it may be
1070 * notifying us of a removal we initiated, so ignore it.
1071 * If there's a pending ENOBUFS, let it stand, since
1072 * everything will be reevaluated. */
1073 if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
1074 sset_add(&ofproto->port_poll_set, devname);
1075 ofproto->port_poll_errno = 0;
1077 } else if (!ofproto) {
1078 /* The port was added, but we don't know with which
1079 * ofproto we should associate it. Delete it. */
1080 dpif_port_del(backer->dpif, port.port_no);
1082 dpif_port_destroy(&port);
1088 if (error != EAGAIN) {
1089 struct ofproto_dpif *ofproto;
1091 /* There was some sort of error, so propagate it to all
1092 * ofprotos that use this backer. */
1093 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
1094 &all_ofproto_dpifs) {
1095 if (ofproto->backer == backer) {
1096 sset_clear(&ofproto->port_poll_set);
1097 ofproto->port_poll_errno = error;
1106 dpif_backer_run_fast(struct dpif_backer *backer, int max_batch)
1110 /* Handle one or more batches of upcalls, until there's nothing left to do
1111 * or until we do a fixed total amount of work.
1113 * We do work in batches because it can be much cheaper to set up a number
1114 * of flows and fire off their patches all at once. We do multiple batches
1115 * because in some cases handling a packet can cause another packet to be
1116 * queued almost immediately as part of the return flow. Both
1117 * optimizations can make major improvements on some benchmarks and
1118 * presumably for real traffic as well. */
1120 while (work < max_batch) {
1121 int retval = handle_upcalls(backer, max_batch - work);
1132 type_run_fast(const char *type)
1134 struct dpif_backer *backer;
1136 backer = shash_find_data(&all_dpif_backers, type);
1138 /* This is not necessarily a problem, since backers are only
1139 * created on demand. */
1143 return dpif_backer_run_fast(backer, FLOW_MISS_MAX_BATCH);
1149 static long long int port_rl = LLONG_MIN;
1150 static unsigned int backer_rl = 0;
1152 if (time_msec() >= port_rl) {
1153 struct ofproto_dpif *ofproto;
1154 struct ofport_dpif *ofport;
1156 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1158 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1159 port_run_fast(ofport);
1162 port_rl = time_msec() + 200;
1165 /* XXX: We have to be careful not to do too much work in this function. If
1166 * we call dpif_backer_run_fast() too often, or with too large a batch,
1167 * performance improves signifcantly, but at a cost. It's possible for the
1168 * number of flows in the datapath to increase without bound, and for poll
1169 * loops to take 10s of seconds. The correct solution to this problem,
1170 * long term, is to separate flow miss handling into it's own thread so it
1171 * isn't affected by revalidations, and expirations. Until then, this is
1172 * the best we can do. */
1173 if (++backer_rl >= 10) {
1174 struct shash_node *node;
1177 SHASH_FOR_EACH (node, &all_dpif_backers) {
1178 dpif_backer_run_fast(node->data, 1);
1184 type_wait(const char *type)
1186 struct dpif_backer *backer;
1188 backer = shash_find_data(&all_dpif_backers, type);
1190 /* This is not necessarily a problem, since backers are only
1191 * created on demand. */
1195 timer_wait(&backer->next_expiration);
1198 /* Basic life-cycle. */
1200 static int add_internal_flows(struct ofproto_dpif *);
1202 static struct ofproto *
1205 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
1206 return &ofproto->up;
1210 dealloc(struct ofproto *ofproto_)
1212 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1217 close_dpif_backer(struct dpif_backer *backer)
1219 struct shash_node *node;
1221 ovs_assert(backer->refcount > 0);
1223 if (--backer->refcount) {
1227 drop_key_clear(backer);
1228 hmap_destroy(&backer->drop_keys);
1230 simap_destroy(&backer->tnl_backers);
1231 hmap_destroy(&backer->odp_to_ofport_map);
1232 node = shash_find(&all_dpif_backers, backer->type);
1234 shash_delete(&all_dpif_backers, node);
1235 dpif_close(backer->dpif);
1240 /* Datapath port slated for removal from datapath. */
1241 struct odp_garbage {
1242 struct list list_node;
1247 open_dpif_backer(const char *type, struct dpif_backer **backerp)
1249 struct dpif_backer *backer;
1250 struct dpif_port_dump port_dump;
1251 struct dpif_port port;
1252 struct shash_node *node;
1253 struct list garbage_list;
1254 struct odp_garbage *garbage, *next;
1260 backer = shash_find_data(&all_dpif_backers, type);
1267 backer_name = xasprintf("ovs-%s", type);
1269 /* Remove any existing datapaths, since we assume we're the only
1270 * userspace controlling the datapath. */
1272 dp_enumerate_names(type, &names);
1273 SSET_FOR_EACH(name, &names) {
1274 struct dpif *old_dpif;
1276 /* Don't remove our backer if it exists. */
1277 if (!strcmp(name, backer_name)) {
1281 if (dpif_open(name, type, &old_dpif)) {
1282 VLOG_WARN("couldn't open old datapath %s to remove it", name);
1284 dpif_delete(old_dpif);
1285 dpif_close(old_dpif);
1288 sset_destroy(&names);
1290 backer = xmalloc(sizeof *backer);
1292 error = dpif_create_and_open(backer_name, type, &backer->dpif);
1295 VLOG_ERR("failed to open datapath of type %s: %s", type,
1301 backer->type = xstrdup(type);
1302 backer->refcount = 1;
1303 hmap_init(&backer->odp_to_ofport_map);
1304 hmap_init(&backer->drop_keys);
1305 timer_set_duration(&backer->next_expiration, 1000);
1306 backer->need_revalidate = 0;
1307 simap_init(&backer->tnl_backers);
1308 tag_set_init(&backer->revalidate_set);
1311 dpif_flow_flush(backer->dpif);
1313 /* Loop through the ports already on the datapath and remove any
1314 * that we don't need anymore. */
1315 list_init(&garbage_list);
1316 dpif_port_dump_start(&port_dump, backer->dpif);
1317 while (dpif_port_dump_next(&port_dump, &port)) {
1318 node = shash_find(&init_ofp_ports, port.name);
1319 if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
1320 garbage = xmalloc(sizeof *garbage);
1321 garbage->odp_port = port.port_no;
1322 list_push_front(&garbage_list, &garbage->list_node);
1325 dpif_port_dump_done(&port_dump);
1327 LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) {
1328 dpif_port_del(backer->dpif, garbage->odp_port);
1329 list_remove(&garbage->list_node);
1333 shash_add(&all_dpif_backers, type, backer);
1335 error = dpif_recv_set(backer->dpif, true);
1337 VLOG_ERR("failed to listen on datapath of type %s: %s",
1338 type, strerror(error));
1339 close_dpif_backer(backer);
1347 construct(struct ofproto *ofproto_)
1349 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1350 struct shash_node *node, *next;
1355 error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1360 max_ports = dpif_get_max_ports(ofproto->backer->dpif);
1361 ofproto_init_max_ports(ofproto_, MIN(max_ports, OFPP_MAX));
1363 ofproto->n_matches = 0;
1365 ofproto->netflow = NULL;
1366 ofproto->sflow = NULL;
1367 ofproto->ipfix = NULL;
1368 ofproto->stp = NULL;
1369 hmap_init(&ofproto->bundles);
1370 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1371 for (i = 0; i < MAX_MIRRORS; i++) {
1372 ofproto->mirrors[i] = NULL;
1374 ofproto->has_bonded_bundles = false;
1376 hmap_init(&ofproto->facets);
1377 hmap_init(&ofproto->subfacets);
1378 ofproto->governor = NULL;
1379 ofproto->consistency_rl = LLONG_MIN;
1381 for (i = 0; i < N_TABLES; i++) {
1382 struct table_dpif *table = &ofproto->tables[i];
1384 table->catchall_table = NULL;
1385 table->other_table = NULL;
1386 table->basis = random_uint32();
1389 list_init(&ofproto->completions);
1391 ofproto_dpif_unixctl_init();
1393 ofproto->has_mirrors = false;
1394 ofproto->has_bundle_action = false;
1396 hmap_init(&ofproto->vlandev_map);
1397 hmap_init(&ofproto->realdev_vid_map);
1399 sset_init(&ofproto->ports);
1400 sset_init(&ofproto->ghost_ports);
1401 sset_init(&ofproto->port_poll_set);
1402 ofproto->port_poll_errno = 0;
1404 SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1405 struct iface_hint *iface_hint = node->data;
1407 if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1408 /* Check if the datapath already has this port. */
1409 if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1410 sset_add(&ofproto->ports, node->name);
1413 free(iface_hint->br_name);
1414 free(iface_hint->br_type);
1416 shash_delete(&init_ofp_ports, node);
1420 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1421 hash_string(ofproto->up.name, 0));
1422 memset(&ofproto->stats, 0, sizeof ofproto->stats);
1424 ofproto_init_tables(ofproto_, N_TABLES);
1425 error = add_internal_flows(ofproto);
1426 ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1429 ofproto->n_missed = 0;
1431 ofproto->max_n_subfacet = 0;
1432 ofproto->created = time_msec();
1433 ofproto->last_minute = ofproto->created;
1434 memset(&ofproto->hourly, 0, sizeof ofproto->hourly);
1435 memset(&ofproto->daily, 0, sizeof ofproto->daily);
1436 ofproto->subfacet_add_count = 0;
1437 ofproto->subfacet_del_count = 0;
1438 ofproto->total_subfacet_add_count = 0;
1439 ofproto->total_subfacet_del_count = 0;
1440 ofproto->total_subfacet_life_span = 0;
1441 ofproto->total_subfacet_count = 0;
1442 ofproto->n_update_stats = 0;
1448 add_internal_flow(struct ofproto_dpif *ofproto, int id,
1449 const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1451 struct ofputil_flow_mod fm;
1454 match_init_catchall(&fm.match);
1456 match_set_reg(&fm.match, 0, id);
1457 fm.new_cookie = htonll(0);
1458 fm.cookie = htonll(0);
1459 fm.cookie_mask = htonll(0);
1460 fm.table_id = TBL_INTERNAL;
1461 fm.command = OFPFC_ADD;
1462 fm.idle_timeout = 0;
1463 fm.hard_timeout = 0;
1467 fm.ofpacts = ofpacts->data;
1468 fm.ofpacts_len = ofpacts->size;
1470 error = ofproto_flow_mod(&ofproto->up, &fm);
1472 VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
1473 id, ofperr_to_string(error));
1477 *rulep = rule_dpif_lookup__(ofproto, &fm.match.flow, TBL_INTERNAL);
1478 ovs_assert(*rulep != NULL);
1484 add_internal_flows(struct ofproto_dpif *ofproto)
1486 struct ofpact_controller *controller;
1487 uint64_t ofpacts_stub[128 / 8];
1488 struct ofpbuf ofpacts;
1492 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1495 controller = ofpact_put_CONTROLLER(&ofpacts);
1496 controller->max_len = UINT16_MAX;
1497 controller->controller_id = 0;
1498 controller->reason = OFPR_NO_MATCH;
1499 ofpact_pad(&ofpacts);
1501 error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule);
1506 ofpbuf_clear(&ofpacts);
1507 error = add_internal_flow(ofproto, id++, &ofpacts,
1508 &ofproto->no_packet_in_rule);
1513 complete_operations(struct ofproto_dpif *ofproto)
1515 struct dpif_completion *c, *next;
1517 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
1518 ofoperation_complete(c->op, 0);
1519 list_remove(&c->list_node);
1525 destruct(struct ofproto *ofproto_)
1527 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1528 struct rule_dpif *rule, *next_rule;
1529 struct oftable *table;
1532 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1533 complete_operations(ofproto);
1535 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1536 struct cls_cursor cursor;
1538 cls_cursor_init(&cursor, &table->cls, NULL);
1539 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1540 ofproto_rule_destroy(&rule->up);
1544 for (i = 0; i < MAX_MIRRORS; i++) {
1545 mirror_destroy(ofproto->mirrors[i]);
1548 netflow_destroy(ofproto->netflow);
1549 dpif_sflow_destroy(ofproto->sflow);
1550 hmap_destroy(&ofproto->bundles);
1551 mac_learning_destroy(ofproto->ml);
1553 hmap_destroy(&ofproto->facets);
1554 hmap_destroy(&ofproto->subfacets);
1555 governor_destroy(ofproto->governor);
1557 hmap_destroy(&ofproto->vlandev_map);
1558 hmap_destroy(&ofproto->realdev_vid_map);
1560 sset_destroy(&ofproto->ports);
1561 sset_destroy(&ofproto->ghost_ports);
1562 sset_destroy(&ofproto->port_poll_set);
1564 close_dpif_backer(ofproto->backer);
1568 run_fast(struct ofproto *ofproto_)
1570 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1571 struct ofport_dpif *ofport;
1573 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1574 port_run_fast(ofport);
1581 run(struct ofproto *ofproto_)
1583 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1584 struct ofport_dpif *ofport;
1585 struct ofbundle *bundle;
1589 complete_operations(ofproto);
1592 error = run_fast(ofproto_);
1597 if (ofproto->netflow) {
1598 if (netflow_run(ofproto->netflow)) {
1599 send_netflow_active_timeouts(ofproto);
1602 if (ofproto->sflow) {
1603 dpif_sflow_run(ofproto->sflow);
1606 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1609 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1614 mac_learning_run(ofproto->ml, &ofproto->backer->revalidate_set);
1616 /* Check the consistency of a random facet, to aid debugging. */
1617 if (time_msec() >= ofproto->consistency_rl
1618 && !hmap_is_empty(&ofproto->facets)
1619 && !ofproto->backer->need_revalidate) {
1620 struct facet *facet;
1622 ofproto->consistency_rl = time_msec() + 250;
1624 facet = CONTAINER_OF(hmap_random_node(&ofproto->facets),
1625 struct facet, hmap_node);
1626 if (!tag_set_intersects(&ofproto->backer->revalidate_set,
1628 if (!facet_check_consistency(facet)) {
1629 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
1634 if (ofproto->governor) {
1637 governor_run(ofproto->governor);
1639 /* If the governor has shrunk to its minimum size and the number of
1640 * subfacets has dwindled, then drop the governor entirely.
1642 * For hysteresis, the number of subfacets to drop the governor is
1643 * smaller than the number needed to trigger its creation. */
1644 n_subfacets = hmap_count(&ofproto->subfacets);
1645 if (n_subfacets * 4 < ofproto->up.flow_eviction_threshold
1646 && governor_is_idle(ofproto->governor)) {
1647 governor_destroy(ofproto->governor);
1648 ofproto->governor = NULL;
1656 wait(struct ofproto *ofproto_)
1658 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1659 struct ofport_dpif *ofport;
1660 struct ofbundle *bundle;
1662 if (!clogged && !list_is_empty(&ofproto->completions)) {
1663 poll_immediate_wake();
1666 dpif_wait(ofproto->backer->dpif);
1667 dpif_recv_wait(ofproto->backer->dpif);
1668 if (ofproto->sflow) {
1669 dpif_sflow_wait(ofproto->sflow);
1671 if (!tag_set_is_empty(&ofproto->backer->revalidate_set)) {
1672 poll_immediate_wake();
1674 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1677 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1678 bundle_wait(bundle);
1680 if (ofproto->netflow) {
1681 netflow_wait(ofproto->netflow);
1683 mac_learning_wait(ofproto->ml);
1685 if (ofproto->backer->need_revalidate) {
1686 /* Shouldn't happen, but if it does just go around again. */
1687 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1688 poll_immediate_wake();
1690 if (ofproto->governor) {
1691 governor_wait(ofproto->governor);
1696 get_memory_usage(const struct ofproto *ofproto_, struct simap *usage)
1698 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1700 simap_increase(usage, "facets", hmap_count(&ofproto->facets));
1701 simap_increase(usage, "subfacets", hmap_count(&ofproto->subfacets));
1705 flush(struct ofproto *ofproto_)
1707 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1708 struct subfacet *subfacet, *next_subfacet;
1709 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
1713 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
1714 &ofproto->subfacets) {
1715 if (subfacet->path != SF_NOT_INSTALLED) {
1716 batch[n_batch++] = subfacet;
1717 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
1718 subfacet_destroy_batch(ofproto, batch, n_batch);
1722 subfacet_destroy(subfacet);
1727 subfacet_destroy_batch(ofproto, batch, n_batch);
1732 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1733 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1735 *arp_match_ip = true;
1736 *actions = (OFPUTIL_A_OUTPUT |
1737 OFPUTIL_A_SET_VLAN_VID |
1738 OFPUTIL_A_SET_VLAN_PCP |
1739 OFPUTIL_A_STRIP_VLAN |
1740 OFPUTIL_A_SET_DL_SRC |
1741 OFPUTIL_A_SET_DL_DST |
1742 OFPUTIL_A_SET_NW_SRC |
1743 OFPUTIL_A_SET_NW_DST |
1744 OFPUTIL_A_SET_NW_TOS |
1745 OFPUTIL_A_SET_TP_SRC |
1746 OFPUTIL_A_SET_TP_DST |
1751 get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots)
1753 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1754 struct dpif_dp_stats s;
1756 strcpy(ots->name, "classifier");
1758 dpif_get_dp_stats(ofproto->backer->dpif, &s);
1760 ots->lookup_count = htonll(s.n_hit + s.n_missed);
1761 ots->matched_count = htonll(s.n_hit + ofproto->n_matches);
1764 static struct ofport *
1767 struct ofport_dpif *port = xmalloc(sizeof *port);
1772 port_dealloc(struct ofport *port_)
1774 struct ofport_dpif *port = ofport_dpif_cast(port_);
1779 port_construct(struct ofport *port_)
1781 struct ofport_dpif *port = ofport_dpif_cast(port_);
1782 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1783 const struct netdev *netdev = port->up.netdev;
1784 struct dpif_port dpif_port;
1787 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1788 port->bundle = NULL;
1791 port->tag = tag_create_random();
1792 port->may_enable = true;
1793 port->stp_port = NULL;
1794 port->stp_state = STP_DISABLED;
1795 port->tnl_port = NULL;
1796 hmap_init(&port->priorities);
1797 port->realdev_ofp_port = 0;
1798 port->vlandev_vid = 0;
1799 port->carrier_seq = netdev_get_carrier_resets(netdev);
1801 if (netdev_vport_is_patch(netdev)) {
1802 /* By bailing out here, we don't submit the port to the sFlow module
1803 * to be considered for counter polling export. This is correct
1804 * because the patch port represents an interface that sFlow considers
1805 * to be "internal" to the switch as a whole, and therefore not an
1806 * candidate for counter polling. */
1807 port->odp_port = OVSP_NONE;
1811 error = dpif_port_query_by_name(ofproto->backer->dpif,
1812 netdev_vport_get_dpif_port(netdev),
1818 port->odp_port = dpif_port.port_no;
1820 if (netdev_get_tunnel_config(netdev)) {
1821 port->tnl_port = tnl_port_add(&port->up, port->odp_port);
1823 /* Sanity-check that a mapping doesn't already exist. This
1824 * shouldn't happen for non-tunnel ports. */
1825 if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1826 VLOG_ERR("port %s already has an OpenFlow port number",
1828 dpif_port_destroy(&dpif_port);
1832 hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1833 hash_int(port->odp_port, 0));
1835 dpif_port_destroy(&dpif_port);
1837 if (ofproto->sflow) {
1838 dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1845 port_destruct(struct ofport *port_)
1847 struct ofport_dpif *port = ofport_dpif_cast(port_);
1848 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1849 const char *dp_port_name = netdev_vport_get_dpif_port(port->up.netdev);
1850 const char *devname = netdev_get_name(port->up.netdev);
1852 if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1853 /* The underlying device is still there, so delete it. This
1854 * happens when the ofproto is being destroyed, since the caller
1855 * assumes that removal of attached ports will happen as part of
1857 if (!port->tnl_port) {
1858 dpif_port_del(ofproto->backer->dpif, port->odp_port);
1860 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1863 if (port->odp_port != OVSP_NONE && !port->tnl_port) {
1864 hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1867 tnl_port_del(port->tnl_port);
1868 sset_find_and_delete(&ofproto->ports, devname);
1869 sset_find_and_delete(&ofproto->ghost_ports, devname);
1870 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1871 bundle_remove(port_);
1872 set_cfm(port_, NULL);
1873 set_bfd(port_, NULL);
1874 if (ofproto->sflow) {
1875 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1878 ofport_clear_priorities(port);
1879 hmap_destroy(&port->priorities);
1883 port_modified(struct ofport *port_)
1885 struct ofport_dpif *port = ofport_dpif_cast(port_);
1887 if (port->bundle && port->bundle->bond) {
1888 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1893 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1895 struct ofport_dpif *port = ofport_dpif_cast(port_);
1896 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1897 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1899 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1900 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1901 OFPUTIL_PC_NO_PACKET_IN)) {
1902 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1904 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1905 bundle_update(port->bundle);
1911 set_sflow(struct ofproto *ofproto_,
1912 const struct ofproto_sflow_options *sflow_options)
1914 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1915 struct dpif_sflow *ds = ofproto->sflow;
1917 if (sflow_options) {
1919 struct ofport_dpif *ofport;
1921 ds = ofproto->sflow = dpif_sflow_create();
1922 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1923 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1925 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1927 dpif_sflow_set_options(ds, sflow_options);
1930 dpif_sflow_destroy(ds);
1931 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1932 ofproto->sflow = NULL;
1940 struct ofproto *ofproto_,
1941 const struct ofproto_ipfix_bridge_exporter_options *bridge_exporter_options,
1942 const struct ofproto_ipfix_flow_exporter_options *flow_exporters_options,
1943 size_t n_flow_exporters_options)
1945 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1946 struct dpif_ipfix *di = ofproto->ipfix;
1948 if (bridge_exporter_options || flow_exporters_options) {
1950 di = ofproto->ipfix = dpif_ipfix_create();
1952 dpif_ipfix_set_options(
1953 di, bridge_exporter_options, flow_exporters_options,
1954 n_flow_exporters_options);
1957 dpif_ipfix_destroy(di);
1958 ofproto->ipfix = NULL;
1965 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1967 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1974 struct ofproto_dpif *ofproto;
1976 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1977 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1978 ofport->cfm = cfm_create(ofport->up.netdev);
1981 if (cfm_configure(ofport->cfm, s)) {
1987 cfm_destroy(ofport->cfm);
1993 get_cfm_status(const struct ofport *ofport_,
1994 struct ofproto_cfm_status *status)
1996 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1999 status->faults = cfm_get_fault(ofport->cfm);
2000 status->remote_opstate = cfm_get_opup(ofport->cfm);
2001 status->health = cfm_get_health(ofport->cfm);
2002 cfm_get_remote_mpids(ofport->cfm, &status->rmps, &status->n_rmps);
2010 set_bfd(struct ofport *ofport_, const struct smap *cfg)
2012 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
2013 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2017 ofport->bfd = bfd_configure(old, netdev_get_name(ofport->up.netdev), cfg);
2018 if (ofport->bfd != old) {
2019 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2026 get_bfd_status(struct ofport *ofport_, struct smap *smap)
2028 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2031 bfd_get_status(ofport->bfd, smap);
2038 /* Spanning Tree. */
2041 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
2043 struct ofproto_dpif *ofproto = ofproto_;
2044 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
2045 struct ofport_dpif *ofport;
2047 ofport = stp_port_get_aux(sp);
2049 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
2050 ofproto->up.name, port_num);
2052 struct eth_header *eth = pkt->l2;
2054 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
2055 if (eth_addr_is_zero(eth->eth_src)) {
2056 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
2057 "with unknown MAC", ofproto->up.name, port_num);
2059 send_packet(ofport, pkt);
2065 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
2067 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
2069 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2071 /* Only revalidate flows if the configuration changed. */
2072 if (!s != !ofproto->stp) {
2073 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2077 if (!ofproto->stp) {
2078 ofproto->stp = stp_create(ofproto_->name, s->system_id,
2079 send_bpdu_cb, ofproto);
2080 ofproto->stp_last_tick = time_msec();
2083 stp_set_bridge_id(ofproto->stp, s->system_id);
2084 stp_set_bridge_priority(ofproto->stp, s->priority);
2085 stp_set_hello_time(ofproto->stp, s->hello_time);
2086 stp_set_max_age(ofproto->stp, s->max_age);
2087 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
2089 struct ofport *ofport;
2091 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
2092 set_stp_port(ofport, NULL);
2095 stp_destroy(ofproto->stp);
2096 ofproto->stp = NULL;
2103 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
2105 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2109 s->bridge_id = stp_get_bridge_id(ofproto->stp);
2110 s->designated_root = stp_get_designated_root(ofproto->stp);
2111 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
2120 update_stp_port_state(struct ofport_dpif *ofport)
2122 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2123 enum stp_state state;
2125 /* Figure out new state. */
2126 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
2130 if (ofport->stp_state != state) {
2131 enum ofputil_port_state of_state;
2134 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
2135 netdev_get_name(ofport->up.netdev),
2136 stp_state_name(ofport->stp_state),
2137 stp_state_name(state));
2138 if (stp_learn_in_state(ofport->stp_state)
2139 != stp_learn_in_state(state)) {
2140 /* xxx Learning action flows should also be flushed. */
2141 mac_learning_flush(ofproto->ml,
2142 &ofproto->backer->revalidate_set);
2144 fwd_change = stp_forward_in_state(ofport->stp_state)
2145 != stp_forward_in_state(state);
2147 ofproto->backer->need_revalidate = REV_STP;
2148 ofport->stp_state = state;
2149 ofport->stp_state_entered = time_msec();
2151 if (fwd_change && ofport->bundle) {
2152 bundle_update(ofport->bundle);
2155 /* Update the STP state bits in the OpenFlow port description. */
2156 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
2157 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
2158 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
2159 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
2160 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
2162 ofproto_port_set_state(&ofport->up, of_state);
2166 /* Configures STP on 'ofport_' using the settings defined in 's'. The
2167 * caller is responsible for assigning STP port numbers and ensuring
2168 * there are no duplicates. */
2170 set_stp_port(struct ofport *ofport_,
2171 const struct ofproto_port_stp_settings *s)
2173 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2174 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2175 struct stp_port *sp = ofport->stp_port;
2177 if (!s || !s->enable) {
2179 ofport->stp_port = NULL;
2180 stp_port_disable(sp);
2181 update_stp_port_state(ofport);
2184 } else if (sp && stp_port_no(sp) != s->port_num
2185 && ofport == stp_port_get_aux(sp)) {
2186 /* The port-id changed, so disable the old one if it's not
2187 * already in use by another port. */
2188 stp_port_disable(sp);
2191 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
2192 stp_port_enable(sp);
2194 stp_port_set_aux(sp, ofport);
2195 stp_port_set_priority(sp, s->priority);
2196 stp_port_set_path_cost(sp, s->path_cost);
2198 update_stp_port_state(ofport);
2204 get_stp_port_status(struct ofport *ofport_,
2205 struct ofproto_port_stp_status *s)
2207 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2208 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2209 struct stp_port *sp = ofport->stp_port;
2211 if (!ofproto->stp || !sp) {
2217 s->port_id = stp_port_get_id(sp);
2218 s->state = stp_port_get_state(sp);
2219 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
2220 s->role = stp_port_get_role(sp);
2221 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
2227 stp_run(struct ofproto_dpif *ofproto)
2230 long long int now = time_msec();
2231 long long int elapsed = now - ofproto->stp_last_tick;
2232 struct stp_port *sp;
2235 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
2236 ofproto->stp_last_tick = now;
2238 while (stp_get_changed_port(ofproto->stp, &sp)) {
2239 struct ofport_dpif *ofport = stp_port_get_aux(sp);
2242 update_stp_port_state(ofport);
2246 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
2247 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
2253 stp_wait(struct ofproto_dpif *ofproto)
2256 poll_timer_wait(1000);
2260 /* Returns true if STP should process 'flow'. */
2262 stp_should_process_flow(const struct flow *flow)
2264 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
2268 stp_process_packet(const struct ofport_dpif *ofport,
2269 const struct ofpbuf *packet)
2271 struct ofpbuf payload = *packet;
2272 struct eth_header *eth = payload.data;
2273 struct stp_port *sp = ofport->stp_port;
2275 /* Sink packets on ports that have STP disabled when the bridge has
2277 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
2281 /* Trim off padding on payload. */
2282 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
2283 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
2286 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
2287 stp_received_bpdu(sp, payload.data, payload.size);
2291 static struct priority_to_dscp *
2292 get_priority(const struct ofport_dpif *ofport, uint32_t priority)
2294 struct priority_to_dscp *pdscp;
2297 hash = hash_int(priority, 0);
2298 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) {
2299 if (pdscp->priority == priority) {
2307 ofport_clear_priorities(struct ofport_dpif *ofport)
2309 struct priority_to_dscp *pdscp, *next;
2311 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) {
2312 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
2318 set_queues(struct ofport *ofport_,
2319 const struct ofproto_port_queue *qdscp_list,
2322 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2323 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2324 struct hmap new = HMAP_INITIALIZER(&new);
2327 for (i = 0; i < n_qdscp; i++) {
2328 struct priority_to_dscp *pdscp;
2332 dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
2333 if (dpif_queue_to_priority(ofproto->backer->dpif, qdscp_list[i].queue,
2338 pdscp = get_priority(ofport, priority);
2340 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
2342 pdscp = xmalloc(sizeof *pdscp);
2343 pdscp->priority = priority;
2345 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2348 if (pdscp->dscp != dscp) {
2350 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2353 hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0));
2356 if (!hmap_is_empty(&ofport->priorities)) {
2357 ofport_clear_priorities(ofport);
2358 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2361 hmap_swap(&new, &ofport->priorities);
2369 /* Expires all MAC learning entries associated with 'bundle' and forces its
2370 * ofproto to revalidate every flow.
2372 * Normally MAC learning entries are removed only from the ofproto associated
2373 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2374 * are removed from every ofproto. When patch ports and SLB bonds are in use
2375 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2376 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2377 * with the host from which it migrated. */
2379 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2381 struct ofproto_dpif *ofproto = bundle->ofproto;
2382 struct mac_learning *ml = ofproto->ml;
2383 struct mac_entry *mac, *next_mac;
2385 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2386 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2387 if (mac->port.p == bundle) {
2389 struct ofproto_dpif *o;
2391 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2393 struct mac_entry *e;
2395 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan,
2398 mac_learning_expire(o->ml, e);
2404 mac_learning_expire(ml, mac);
2409 static struct ofbundle *
2410 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2412 struct ofbundle *bundle;
2414 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2415 &ofproto->bundles) {
2416 if (bundle->aux == aux) {
2423 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
2424 * ones that are found to 'bundles'. */
2426 bundle_lookup_multiple(struct ofproto_dpif *ofproto,
2427 void **auxes, size_t n_auxes,
2428 struct hmapx *bundles)
2432 hmapx_init(bundles);
2433 for (i = 0; i < n_auxes; i++) {
2434 struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
2436 hmapx_add(bundles, bundle);
2442 bundle_update(struct ofbundle *bundle)
2444 struct ofport_dpif *port;
2446 bundle->floodable = true;
2447 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2448 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2449 || !stp_forward_in_state(port->stp_state)) {
2450 bundle->floodable = false;
2457 bundle_del_port(struct ofport_dpif *port)
2459 struct ofbundle *bundle = port->bundle;
2461 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2463 list_remove(&port->bundle_node);
2464 port->bundle = NULL;
2467 lacp_slave_unregister(bundle->lacp, port);
2470 bond_slave_unregister(bundle->bond, port);
2473 bundle_update(bundle);
2477 bundle_add_port(struct ofbundle *bundle, uint16_t ofp_port,
2478 struct lacp_slave_settings *lacp)
2480 struct ofport_dpif *port;
2482 port = get_ofp_port(bundle->ofproto, ofp_port);
2487 if (port->bundle != bundle) {
2488 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2490 bundle_del_port(port);
2493 port->bundle = bundle;
2494 list_push_back(&bundle->ports, &port->bundle_node);
2495 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2496 || !stp_forward_in_state(port->stp_state)) {
2497 bundle->floodable = false;
2501 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2502 lacp_slave_register(bundle->lacp, port, lacp);
2509 bundle_destroy(struct ofbundle *bundle)
2511 struct ofproto_dpif *ofproto;
2512 struct ofport_dpif *port, *next_port;
2519 ofproto = bundle->ofproto;
2520 for (i = 0; i < MAX_MIRRORS; i++) {
2521 struct ofmirror *m = ofproto->mirrors[i];
2523 if (m->out == bundle) {
2525 } else if (hmapx_find_and_delete(&m->srcs, bundle)
2526 || hmapx_find_and_delete(&m->dsts, bundle)) {
2527 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2532 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2533 bundle_del_port(port);
2536 bundle_flush_macs(bundle, true);
2537 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2539 free(bundle->trunks);
2540 lacp_destroy(bundle->lacp);
2541 bond_destroy(bundle->bond);
2546 bundle_set(struct ofproto *ofproto_, void *aux,
2547 const struct ofproto_bundle_settings *s)
2549 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2550 bool need_flush = false;
2551 struct ofport_dpif *port;
2552 struct ofbundle *bundle;
2553 unsigned long *trunks;
2559 bundle_destroy(bundle_lookup(ofproto, aux));
2563 ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2564 ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2566 bundle = bundle_lookup(ofproto, aux);
2568 bundle = xmalloc(sizeof *bundle);
2570 bundle->ofproto = ofproto;
2571 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2572 hash_pointer(aux, 0));
2574 bundle->name = NULL;
2576 list_init(&bundle->ports);
2577 bundle->vlan_mode = PORT_VLAN_TRUNK;
2579 bundle->trunks = NULL;
2580 bundle->use_priority_tags = s->use_priority_tags;
2581 bundle->lacp = NULL;
2582 bundle->bond = NULL;
2584 bundle->floodable = true;
2586 bundle->src_mirrors = 0;
2587 bundle->dst_mirrors = 0;
2588 bundle->mirror_out = 0;
2591 if (!bundle->name || strcmp(s->name, bundle->name)) {
2593 bundle->name = xstrdup(s->name);
2598 if (!bundle->lacp) {
2599 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2600 bundle->lacp = lacp_create();
2602 lacp_configure(bundle->lacp, s->lacp);
2604 lacp_destroy(bundle->lacp);
2605 bundle->lacp = NULL;
2608 /* Update set of ports. */
2610 for (i = 0; i < s->n_slaves; i++) {
2611 if (!bundle_add_port(bundle, s->slaves[i],
2612 s->lacp ? &s->lacp_slaves[i] : NULL)) {
2616 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2617 struct ofport_dpif *next_port;
2619 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2620 for (i = 0; i < s->n_slaves; i++) {
2621 if (s->slaves[i] == port->up.ofp_port) {
2626 bundle_del_port(port);
2630 ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2632 if (list_is_empty(&bundle->ports)) {
2633 bundle_destroy(bundle);
2637 /* Set VLAN tagging mode */
2638 if (s->vlan_mode != bundle->vlan_mode
2639 || s->use_priority_tags != bundle->use_priority_tags) {
2640 bundle->vlan_mode = s->vlan_mode;
2641 bundle->use_priority_tags = s->use_priority_tags;
2646 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2647 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2649 if (vlan != bundle->vlan) {
2650 bundle->vlan = vlan;
2654 /* Get trunked VLANs. */
2655 switch (s->vlan_mode) {
2656 case PORT_VLAN_ACCESS:
2660 case PORT_VLAN_TRUNK:
2661 trunks = CONST_CAST(unsigned long *, s->trunks);
2664 case PORT_VLAN_NATIVE_UNTAGGED:
2665 case PORT_VLAN_NATIVE_TAGGED:
2666 if (vlan != 0 && (!s->trunks
2667 || !bitmap_is_set(s->trunks, vlan)
2668 || bitmap_is_set(s->trunks, 0))) {
2669 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2671 trunks = bitmap_clone(s->trunks, 4096);
2673 trunks = bitmap_allocate1(4096);
2675 bitmap_set1(trunks, vlan);
2676 bitmap_set0(trunks, 0);
2678 trunks = CONST_CAST(unsigned long *, s->trunks);
2685 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2686 free(bundle->trunks);
2687 if (trunks == s->trunks) {
2688 bundle->trunks = vlan_bitmap_clone(trunks);
2690 bundle->trunks = trunks;
2695 if (trunks != s->trunks) {
2700 if (!list_is_short(&bundle->ports)) {
2701 bundle->ofproto->has_bonded_bundles = true;
2703 if (bond_reconfigure(bundle->bond, s->bond)) {
2704 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2707 bundle->bond = bond_create(s->bond);
2708 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2711 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2712 bond_slave_register(bundle->bond, port, port->up.netdev);
2715 bond_destroy(bundle->bond);
2716 bundle->bond = NULL;
2719 /* If we changed something that would affect MAC learning, un-learn
2720 * everything on this port and force flow revalidation. */
2722 bundle_flush_macs(bundle, false);
2729 bundle_remove(struct ofport *port_)
2731 struct ofport_dpif *port = ofport_dpif_cast(port_);
2732 struct ofbundle *bundle = port->bundle;
2735 bundle_del_port(port);
2736 if (list_is_empty(&bundle->ports)) {
2737 bundle_destroy(bundle);
2738 } else if (list_is_short(&bundle->ports)) {
2739 bond_destroy(bundle->bond);
2740 bundle->bond = NULL;
2746 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2748 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2749 struct ofport_dpif *port = port_;
2750 uint8_t ea[ETH_ADDR_LEN];
2753 error = netdev_get_etheraddr(port->up.netdev, ea);
2755 struct ofpbuf packet;
2758 ofpbuf_init(&packet, 0);
2759 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2761 memcpy(packet_pdu, pdu, pdu_size);
2763 send_packet(port, &packet);
2764 ofpbuf_uninit(&packet);
2766 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2767 "%s (%s)", port->bundle->name,
2768 netdev_get_name(port->up.netdev), strerror(error));
2773 bundle_send_learning_packets(struct ofbundle *bundle)
2775 struct ofproto_dpif *ofproto = bundle->ofproto;
2776 int error, n_packets, n_errors;
2777 struct mac_entry *e;
2779 error = n_packets = n_errors = 0;
2780 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2781 if (e->port.p != bundle) {
2782 struct ofpbuf *learning_packet;
2783 struct ofport_dpif *port;
2787 /* The assignment to "port" is unnecessary but makes "grep"ing for
2788 * struct ofport_dpif more effective. */
2789 learning_packet = bond_compose_learning_packet(bundle->bond,
2793 ret = send_packet(port, learning_packet);
2794 ofpbuf_delete(learning_packet);
2804 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2805 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2806 "packets, last error was: %s",
2807 bundle->name, n_errors, n_packets, strerror(error));
2809 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2810 bundle->name, n_packets);
2815 bundle_run(struct ofbundle *bundle)
2818 lacp_run(bundle->lacp, send_pdu_cb);
2821 struct ofport_dpif *port;
2823 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2824 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2827 bond_run(bundle->bond, &bundle->ofproto->backer->revalidate_set,
2828 lacp_status(bundle->lacp));
2829 if (bond_should_send_learning_packets(bundle->bond)) {
2830 bundle_send_learning_packets(bundle);
2836 bundle_wait(struct ofbundle *bundle)
2839 lacp_wait(bundle->lacp);
2842 bond_wait(bundle->bond);
2849 mirror_scan(struct ofproto_dpif *ofproto)
2853 for (idx = 0; idx < MAX_MIRRORS; idx++) {
2854 if (!ofproto->mirrors[idx]) {
2861 static struct ofmirror *
2862 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
2866 for (i = 0; i < MAX_MIRRORS; i++) {
2867 struct ofmirror *mirror = ofproto->mirrors[i];
2868 if (mirror && mirror->aux == aux) {
2876 /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */
2878 mirror_update_dups(struct ofproto_dpif *ofproto)
2882 for (i = 0; i < MAX_MIRRORS; i++) {
2883 struct ofmirror *m = ofproto->mirrors[i];
2886 m->dup_mirrors = MIRROR_MASK_C(1) << i;
2890 for (i = 0; i < MAX_MIRRORS; i++) {
2891 struct ofmirror *m1 = ofproto->mirrors[i];
2898 for (j = i + 1; j < MAX_MIRRORS; j++) {
2899 struct ofmirror *m2 = ofproto->mirrors[j];
2901 if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) {
2902 m1->dup_mirrors |= MIRROR_MASK_C(1) << j;
2903 m2->dup_mirrors |= m1->dup_mirrors;
2910 mirror_set(struct ofproto *ofproto_, void *aux,
2911 const struct ofproto_mirror_settings *s)
2913 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2914 mirror_mask_t mirror_bit;
2915 struct ofbundle *bundle;
2916 struct ofmirror *mirror;
2917 struct ofbundle *out;
2918 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
2919 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
2922 mirror = mirror_lookup(ofproto, aux);
2924 mirror_destroy(mirror);
2930 idx = mirror_scan(ofproto);
2932 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
2934 ofproto->up.name, MAX_MIRRORS, s->name);
2938 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
2939 mirror->ofproto = ofproto;
2942 mirror->out_vlan = -1;
2943 mirror->name = NULL;
2946 if (!mirror->name || strcmp(s->name, mirror->name)) {
2948 mirror->name = xstrdup(s->name);
2951 /* Get the new configuration. */
2952 if (s->out_bundle) {
2953 out = bundle_lookup(ofproto, s->out_bundle);
2955 mirror_destroy(mirror);
2961 out_vlan = s->out_vlan;
2963 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
2964 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
2966 /* If the configuration has not changed, do nothing. */
2967 if (hmapx_equals(&srcs, &mirror->srcs)
2968 && hmapx_equals(&dsts, &mirror->dsts)
2969 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
2970 && mirror->out == out
2971 && mirror->out_vlan == out_vlan)
2973 hmapx_destroy(&srcs);
2974 hmapx_destroy(&dsts);
2978 hmapx_swap(&srcs, &mirror->srcs);
2979 hmapx_destroy(&srcs);
2981 hmapx_swap(&dsts, &mirror->dsts);
2982 hmapx_destroy(&dsts);
2984 free(mirror->vlans);
2985 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
2988 mirror->out_vlan = out_vlan;
2990 /* Update bundles. */
2991 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2992 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
2993 if (hmapx_contains(&mirror->srcs, bundle)) {
2994 bundle->src_mirrors |= mirror_bit;
2996 bundle->src_mirrors &= ~mirror_bit;
2999 if (hmapx_contains(&mirror->dsts, bundle)) {
3000 bundle->dst_mirrors |= mirror_bit;
3002 bundle->dst_mirrors &= ~mirror_bit;
3005 if (mirror->out == bundle) {
3006 bundle->mirror_out |= mirror_bit;
3008 bundle->mirror_out &= ~mirror_bit;
3012 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3013 ofproto->has_mirrors = true;
3014 mac_learning_flush(ofproto->ml,
3015 &ofproto->backer->revalidate_set);
3016 mirror_update_dups(ofproto);
3022 mirror_destroy(struct ofmirror *mirror)
3024 struct ofproto_dpif *ofproto;
3025 mirror_mask_t mirror_bit;
3026 struct ofbundle *bundle;
3033 ofproto = mirror->ofproto;
3034 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3035 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
3037 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
3038 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
3039 bundle->src_mirrors &= ~mirror_bit;
3040 bundle->dst_mirrors &= ~mirror_bit;
3041 bundle->mirror_out &= ~mirror_bit;
3044 hmapx_destroy(&mirror->srcs);
3045 hmapx_destroy(&mirror->dsts);
3046 free(mirror->vlans);
3048 ofproto->mirrors[mirror->idx] = NULL;
3052 mirror_update_dups(ofproto);
3054 ofproto->has_mirrors = false;
3055 for (i = 0; i < MAX_MIRRORS; i++) {
3056 if (ofproto->mirrors[i]) {
3057 ofproto->has_mirrors = true;
3064 mirror_get_stats(struct ofproto *ofproto_, void *aux,
3065 uint64_t *packets, uint64_t *bytes)
3067 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3068 struct ofmirror *mirror = mirror_lookup(ofproto, aux);
3071 *packets = *bytes = UINT64_MAX;
3077 *packets = mirror->packet_count;
3078 *bytes = mirror->byte_count;
3084 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
3086 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3087 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
3088 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
3094 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
3096 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3097 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
3098 return bundle && bundle->mirror_out != 0;
3102 forward_bpdu_changed(struct ofproto *ofproto_)
3104 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3105 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3109 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
3112 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3113 mac_learning_set_idle_time(ofproto->ml, idle_time);
3114 mac_learning_set_max_entries(ofproto->ml, max_entries);
3119 static struct ofport_dpif *
3120 get_ofp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
3122 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
3123 return ofport ? ofport_dpif_cast(ofport) : NULL;
3126 static struct ofport_dpif *
3127 get_odp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
3129 struct ofport_dpif *port = odp_port_to_ofport(ofproto->backer, odp_port);
3130 return port && &ofproto->up == port->up.ofproto ? port : NULL;
3134 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
3135 struct ofproto_port *ofproto_port,
3136 struct dpif_port *dpif_port)
3138 ofproto_port->name = dpif_port->name;
3139 ofproto_port->type = dpif_port->type;
3140 ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
3143 static struct ofport_dpif *
3144 ofport_get_peer(const struct ofport_dpif *ofport_dpif)
3146 const struct ofproto_dpif *ofproto;
3149 peer = netdev_vport_patch_peer(ofport_dpif->up.netdev);
3154 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
3155 struct ofport *ofport;
3157 ofport = shash_find_data(&ofproto->up.port_by_name, peer);
3158 if (ofport && ofport->ofproto->ofproto_class == &ofproto_dpif_class) {
3159 return ofport_dpif_cast(ofport);
3166 port_run_fast(struct ofport_dpif *ofport)
3168 if (ofport->cfm && cfm_should_send_ccm(ofport->cfm)) {
3169 struct ofpbuf packet;
3171 ofpbuf_init(&packet, 0);
3172 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
3173 send_packet(ofport, &packet);
3174 ofpbuf_uninit(&packet);
3177 if (ofport->bfd && bfd_should_send_packet(ofport->bfd)) {
3178 struct ofpbuf packet;
3180 ofpbuf_init(&packet, 0);
3181 bfd_put_packet(ofport->bfd, &packet, ofport->up.pp.hw_addr);
3182 send_packet(ofport, &packet);
3183 ofpbuf_uninit(&packet);
3188 port_run(struct ofport_dpif *ofport)
3190 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
3191 bool carrier_changed = carrier_seq != ofport->carrier_seq;
3192 bool enable = netdev_get_carrier(ofport->up.netdev);
3194 ofport->carrier_seq = carrier_seq;
3196 port_run_fast(ofport);
3198 if (ofport->tnl_port
3199 && tnl_port_reconfigure(&ofport->up, ofport->odp_port,
3200 &ofport->tnl_port)) {
3201 ofproto_dpif_cast(ofport->up.ofproto)->backer->need_revalidate = true;
3205 int cfm_opup = cfm_get_opup(ofport->cfm);
3207 cfm_run(ofport->cfm);
3208 enable = enable && !cfm_get_fault(ofport->cfm);
3210 if (cfm_opup >= 0) {
3211 enable = enable && cfm_opup;
3216 bfd_run(ofport->bfd);
3217 enable = enable && bfd_forwarding(ofport->bfd);
3220 if (ofport->bundle) {
3221 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
3222 if (carrier_changed) {
3223 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
3227 if (ofport->may_enable != enable) {
3228 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3230 if (ofproto->has_bundle_action) {
3231 ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
3235 ofport->may_enable = enable;
3239 port_wait(struct ofport_dpif *ofport)
3242 cfm_wait(ofport->cfm);
3246 bfd_wait(ofport->bfd);
3251 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
3252 struct ofproto_port *ofproto_port)
3254 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3255 struct dpif_port dpif_port;
3258 if (sset_contains(&ofproto->ghost_ports, devname)) {
3259 const char *type = netdev_get_type_from_name(devname);
3261 /* We may be called before ofproto->up.port_by_name is populated with
3262 * the appropriate ofport. For this reason, we must get the name and
3263 * type from the netdev layer directly. */
3265 const struct ofport *ofport;
3267 ofport = shash_find_data(&ofproto->up.port_by_name, devname);
3268 ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
3269 ofproto_port->name = xstrdup(devname);
3270 ofproto_port->type = xstrdup(type);
3276 if (!sset_contains(&ofproto->ports, devname)) {
3279 error = dpif_port_query_by_name(ofproto->backer->dpif,
3280 devname, &dpif_port);
3282 ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
3288 port_add(struct ofproto *ofproto_, struct netdev *netdev)
3290 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3291 const char *dp_port_name = netdev_vport_get_dpif_port(netdev);
3292 const char *devname = netdev_get_name(netdev);
3294 if (netdev_vport_is_patch(netdev)) {
3295 sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
3299 if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
3300 uint32_t port_no = UINT32_MAX;
3303 error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
3307 if (netdev_get_tunnel_config(netdev)) {
3308 simap_put(&ofproto->backer->tnl_backers, dp_port_name, port_no);
3312 if (netdev_get_tunnel_config(netdev)) {
3313 sset_add(&ofproto->ghost_ports, devname);
3315 sset_add(&ofproto->ports, devname);
3321 port_del(struct ofproto *ofproto_, uint16_t ofp_port)
3323 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3324 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
3331 sset_find_and_delete(&ofproto->ghost_ports,
3332 netdev_get_name(ofport->up.netdev));
3333 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3334 if (!ofport->tnl_port) {
3335 error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
3337 /* The caller is going to close ofport->up.netdev. If this is a
3338 * bonded port, then the bond is using that netdev, so remove it
3339 * from the bond. The client will need to reconfigure everything
3340 * after deleting ports, so then the slave will get re-added. */
3341 bundle_remove(&ofport->up);
3348 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
3350 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3355 error = netdev_get_stats(ofport->up.netdev, stats);
3357 if (!error && ofport_->ofp_port == OFPP_LOCAL) {
3358 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3360 /* ofproto->stats.tx_packets represents packets that we created
3361 * internally and sent to some port (e.g. packets sent with
3362 * send_packet()). Account for them as if they had come from
3363 * OFPP_LOCAL and got forwarded. */
3365 if (stats->rx_packets != UINT64_MAX) {
3366 stats->rx_packets += ofproto->stats.tx_packets;
3369 if (stats->rx_bytes != UINT64_MAX) {
3370 stats->rx_bytes += ofproto->stats.tx_bytes;
3373 /* ofproto->stats.rx_packets represents packets that were received on
3374 * some port and we processed internally and dropped (e.g. STP).
3375 * Account for them as if they had been forwarded to OFPP_LOCAL. */
3377 if (stats->tx_packets != UINT64_MAX) {
3378 stats->tx_packets += ofproto->stats.rx_packets;
3381 if (stats->tx_bytes != UINT64_MAX) {
3382 stats->tx_bytes += ofproto->stats.rx_bytes;
3389 struct port_dump_state {
3394 struct ofproto_port port;
3399 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
3401 *statep = xzalloc(sizeof(struct port_dump_state));
3406 port_dump_next(const struct ofproto *ofproto_, void *state_,
3407 struct ofproto_port *port)
3409 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3410 struct port_dump_state *state = state_;
3411 const struct sset *sset;
3412 struct sset_node *node;
3414 if (state->has_port) {
3415 ofproto_port_destroy(&state->port);
3416 state->has_port = false;
3418 sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
3419 while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
3422 error = port_query_by_name(ofproto_, node->name, &state->port);
3424 *port = state->port;
3425 state->has_port = true;
3427 } else if (error != ENODEV) {
3432 if (!state->ghost) {
3433 state->ghost = true;
3436 return port_dump_next(ofproto_, state_, port);
3443 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3445 struct port_dump_state *state = state_;
3447 if (state->has_port) {
3448 ofproto_port_destroy(&state->port);
3455 port_poll(const struct ofproto *ofproto_, char **devnamep)
3457 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3459 if (ofproto->port_poll_errno) {
3460 int error = ofproto->port_poll_errno;
3461 ofproto->port_poll_errno = 0;
3465 if (sset_is_empty(&ofproto->port_poll_set)) {
3469 *devnamep = sset_pop(&ofproto->port_poll_set);
3474 port_poll_wait(const struct ofproto *ofproto_)
3476 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3477 dpif_port_poll_wait(ofproto->backer->dpif);
3481 port_is_lacp_current(const struct ofport *ofport_)
3483 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3484 return (ofport->bundle && ofport->bundle->lacp
3485 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3489 /* Upcall handling. */
3491 /* Flow miss batching.
3493 * Some dpifs implement operations faster when you hand them off in a batch.
3494 * To allow batching, "struct flow_miss" queues the dpif-related work needed
3495 * for a given flow. Each "struct flow_miss" corresponds to sending one or
3496 * more packets, plus possibly installing the flow in the dpif.
3498 * So far we only batch the operations that affect flow setup time the most.
3499 * It's possible to batch more than that, but the benefit might be minimal. */
3501 struct hmap_node hmap_node;
3502 struct ofproto_dpif *ofproto;
3504 enum odp_key_fitness key_fitness;
3505 const struct nlattr *key;
3507 struct initial_vals initial_vals;
3508 struct list packets;
3509 enum dpif_upcall_type upcall_type;
3510 uint32_t odp_in_port;
3513 struct flow_miss_op {
3514 struct dpif_op dpif_op;
3515 void *garbage; /* Pointer to pass to free(), NULL if none. */
3516 uint64_t stub[1024 / 8]; /* Temporary buffer. */
3519 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
3520 * OpenFlow controller as necessary according to their individual
3521 * configurations. */
3523 send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet,
3524 const struct flow *flow)
3526 struct ofputil_packet_in pin;
3528 pin.packet = packet->data;
3529 pin.packet_len = packet->size;
3530 pin.reason = OFPR_NO_MATCH;
3531 pin.controller_id = 0;
3536 pin.send_len = 0; /* not used for flow table misses */
3538 flow_get_metadata(flow, &pin.fmd);
3540 connmgr_send_packet_in(ofproto->up.connmgr, &pin);
3543 static enum slow_path_reason
3544 process_special(struct ofproto_dpif *ofproto, const struct flow *flow,
3545 const struct ofport_dpif *ofport, const struct ofpbuf *packet)
3549 } else if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) {
3551 cfm_process_heartbeat(ofport->cfm, packet);
3554 } else if (ofport->bfd && bfd_should_process_flow(flow)) {
3556 bfd_process_packet(ofport->bfd, flow, packet);
3559 } else if (ofport->bundle && ofport->bundle->lacp
3560 && flow->dl_type == htons(ETH_TYPE_LACP)) {
3562 lacp_process_packet(ofport->bundle->lacp, ofport, packet);
3565 } else if (ofproto->stp && stp_should_process_flow(flow)) {
3567 stp_process_packet(ofport, packet);
3575 static struct flow_miss *
3576 flow_miss_find(struct hmap *todo, const struct ofproto_dpif *ofproto,
3577 const struct flow *flow, uint32_t hash)
3579 struct flow_miss *miss;
3581 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
3582 if (miss->ofproto == ofproto && flow_equal(&miss->flow, flow)) {
3590 /* Partially Initializes 'op' as an "execute" operation for 'miss' and
3591 * 'packet'. The caller must initialize op->actions and op->actions_len. If
3592 * 'miss' is associated with a subfacet the caller must also initialize the
3593 * returned op->subfacet, and if anything needs to be freed after processing
3594 * the op, the caller must initialize op->garbage also. */
3596 init_flow_miss_execute_op(struct flow_miss *miss, struct ofpbuf *packet,
3597 struct flow_miss_op *op)
3599 if (miss->flow.vlan_tci != miss->initial_vals.vlan_tci) {
3600 /* This packet was received on a VLAN splinter port. We
3601 * added a VLAN to the packet to make the packet resemble
3602 * the flow, but the actions were composed assuming that
3603 * the packet contained no VLAN. So, we must remove the
3604 * VLAN header from the packet before trying to execute the
3606 eth_pop_vlan(packet);
3610 op->dpif_op.type = DPIF_OP_EXECUTE;
3611 op->dpif_op.u.execute.key = miss->key;
3612 op->dpif_op.u.execute.key_len = miss->key_len;
3613 op->dpif_op.u.execute.packet = packet;
3616 /* Helper for handle_flow_miss_without_facet() and
3617 * handle_flow_miss_with_facet(). */
3619 handle_flow_miss_common(struct rule_dpif *rule,
3620 struct ofpbuf *packet, const struct flow *flow)
3622 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3624 ofproto->n_matches++;
3626 if (rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
3628 * Extra-special case for fail-open mode.
3630 * We are in fail-open mode and the packet matched the fail-open
3631 * rule, but we are connected to a controller too. We should send
3632 * the packet up to the controller in the hope that it will try to
3633 * set up a flow and thereby allow us to exit fail-open.
3635 * See the top-level comment in fail-open.c for more information.
3637 send_packet_in_miss(ofproto, packet, flow);
3641 /* Figures out whether a flow that missed in 'ofproto', whose details are in
3642 * 'miss', is likely to be worth tracking in detail in userspace and (usually)
3643 * installing a datapath flow. The answer is usually "yes" (a return value of
3644 * true). However, for short flows the cost of bookkeeping is much higher than
3645 * the benefits, so when the datapath holds a large number of flows we impose
3646 * some heuristics to decide which flows are likely to be worth tracking. */
3648 flow_miss_should_make_facet(struct ofproto_dpif *ofproto,
3649 struct flow_miss *miss, uint32_t hash)
3651 if (!ofproto->governor) {
3654 n_subfacets = hmap_count(&ofproto->subfacets);
3655 if (n_subfacets * 2 <= ofproto->up.flow_eviction_threshold) {
3659 ofproto->governor = governor_create(ofproto->up.name);
3662 return governor_should_install_flow(ofproto->governor, hash,
3663 list_size(&miss->packets));
3666 /* Handles 'miss', which matches 'rule', without creating a facet or subfacet
3667 * or creating any datapath flow. May add an "execute" operation to 'ops' and
3668 * increment '*n_ops'. */
3670 handle_flow_miss_without_facet(struct flow_miss *miss,
3671 struct rule_dpif *rule,
3672 struct flow_miss_op *ops, size_t *n_ops)
3674 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3675 long long int now = time_msec();
3676 struct action_xlate_ctx ctx;
3677 struct ofpbuf *packet;
3679 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3680 struct flow_miss_op *op = &ops[*n_ops];
3681 struct dpif_flow_stats stats;
3682 struct ofpbuf odp_actions;
3684 COVERAGE_INC(facet_suppress);
3686 ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub);
3688 dpif_flow_stats_extract(&miss->flow, packet, now, &stats);
3689 rule_credit_stats(rule, &stats);
3691 action_xlate_ctx_init(&ctx, ofproto, &miss->flow, &miss->initial_vals,
3692 rule, stats.tcp_flags, packet);
3693 ctx.resubmit_stats = &stats;
3694 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len,
3697 if (odp_actions.size) {
3698 struct dpif_execute *execute = &op->dpif_op.u.execute;
3700 init_flow_miss_execute_op(miss, packet, op);
3701 execute->actions = odp_actions.data;
3702 execute->actions_len = odp_actions.size;
3703 op->garbage = ofpbuf_get_uninit_pointer(&odp_actions);
3707 ofpbuf_uninit(&odp_actions);
3712 /* Handles 'miss', which matches 'facet'. May add any required datapath
3713 * operations to 'ops', incrementing '*n_ops' for each new op.
3715 * All of the packets in 'miss' are considered to have arrived at time 'now'.
3716 * This is really important only for new facets: if we just called time_msec()
3717 * here, then the new subfacet or its packets could look (occasionally) as
3718 * though it was used some time after the facet was used. That can make a
3719 * one-packet flow look like it has a nonzero duration, which looks odd in
3720 * e.g. NetFlow statistics. */
3722 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
3724 struct flow_miss_op *ops, size_t *n_ops)
3726 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3727 enum subfacet_path want_path;
3728 struct subfacet *subfacet;
3729 struct ofpbuf *packet;
3731 subfacet = subfacet_create(facet, miss, now);
3733 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3734 struct flow_miss_op *op = &ops[*n_ops];
3735 struct dpif_flow_stats stats;
3736 struct ofpbuf odp_actions;
3738 handle_flow_miss_common(facet->rule, packet, &miss->flow);
3740 ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub);
3741 if (!subfacet->actions || subfacet->slow) {
3742 subfacet_make_actions(subfacet, packet, &odp_actions);
3745 dpif_flow_stats_extract(&facet->flow, packet, now, &stats);
3746 subfacet_update_stats(subfacet, &stats);
3748 if (subfacet->actions_len) {
3749 struct dpif_execute *execute = &op->dpif_op.u.execute;
3751 init_flow_miss_execute_op(miss, packet, op);
3752 if (!subfacet->slow) {
3753 execute->actions = subfacet->actions;
3754 execute->actions_len = subfacet->actions_len;
3755 ofpbuf_uninit(&odp_actions);
3757 execute->actions = odp_actions.data;
3758 execute->actions_len = odp_actions.size;
3759 op->garbage = ofpbuf_get_uninit_pointer(&odp_actions);
3764 ofpbuf_uninit(&odp_actions);
3768 want_path = subfacet_want_path(subfacet->slow);
3769 if (miss->upcall_type == DPIF_UC_MISS || subfacet->path != want_path) {
3770 struct flow_miss_op *op = &ops[(*n_ops)++];
3771 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
3773 subfacet->path = want_path;
3776 op->dpif_op.type = DPIF_OP_FLOW_PUT;
3777 put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
3778 put->key = miss->key;
3779 put->key_len = miss->key_len;
3780 if (want_path == SF_FAST_PATH) {
3781 put->actions = subfacet->actions;
3782 put->actions_len = subfacet->actions_len;
3784 compose_slow_path(ofproto, &facet->flow, subfacet->slow,
3785 op->stub, sizeof op->stub,
3786 &put->actions, &put->actions_len);
3792 /* Handles flow miss 'miss'. May add any required datapath operations
3793 * to 'ops', incrementing '*n_ops' for each new op. */
3795 handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops,
3798 struct ofproto_dpif *ofproto = miss->ofproto;
3799 struct facet *facet;
3803 /* The caller must ensure that miss->hmap_node.hash contains
3804 * flow_hash(miss->flow, 0). */
3805 hash = miss->hmap_node.hash;
3807 facet = facet_lookup_valid(ofproto, &miss->flow, hash);
3809 struct rule_dpif *rule = rule_dpif_lookup(ofproto, &miss->flow);
3811 if (!flow_miss_should_make_facet(ofproto, miss, hash)) {
3812 handle_flow_miss_without_facet(miss, rule, ops, n_ops);
3816 facet = facet_create(rule, &miss->flow, hash);
3821 handle_flow_miss_with_facet(miss, facet, now, ops, n_ops);
3824 static struct drop_key *
3825 drop_key_lookup(const struct dpif_backer *backer, const struct nlattr *key,
3828 struct drop_key *drop_key;
3830 HMAP_FOR_EACH_WITH_HASH (drop_key, hmap_node, hash_bytes(key, key_len, 0),
3831 &backer->drop_keys) {
3832 if (drop_key->key_len == key_len
3833 && !memcmp(drop_key->key, key, key_len)) {
3841 drop_key_clear(struct dpif_backer *backer)
3843 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3844 struct drop_key *drop_key, *next;
3846 HMAP_FOR_EACH_SAFE (drop_key, next, hmap_node, &backer->drop_keys) {
3849 error = dpif_flow_del(backer->dpif, drop_key->key, drop_key->key_len,
3851 if (error && !VLOG_DROP_WARN(&rl)) {
3852 struct ds ds = DS_EMPTY_INITIALIZER;
3853 odp_flow_key_format(drop_key->key, drop_key->key_len, &ds);
3854 VLOG_WARN("Failed to delete drop key (%s) (%s)", strerror(error),
3859 hmap_remove(&backer->drop_keys, &drop_key->hmap_node);
3860 free(drop_key->key);
3865 /* Given a datpath, packet, and flow metadata ('backer', 'packet', and 'key'
3866 * respectively), populates 'flow' with the result of odp_flow_key_to_flow().
3867 * Optionally, if nonnull, populates 'fitnessp' with the fitness of 'flow' as
3868 * returned by odp_flow_key_to_flow(). Also, optionally populates 'ofproto'
3869 * with the ofproto_dpif, and 'odp_in_port' with the datapath in_port, that
3870 * 'packet' ingressed.
3872 * If 'ofproto' is nonnull, requires 'flow''s in_port to exist. Otherwise sets
3873 * 'flow''s in_port to OFPP_NONE.
3875 * This function does post-processing on data returned from
3876 * odp_flow_key_to_flow() to help make VLAN splinters transparent to the rest
3877 * of the upcall processing logic. In particular, if the extracted in_port is
3878 * a VLAN splinter port, it replaces flow->in_port by the "real" port, sets
3879 * flow->vlan_tci correctly for the VLAN of the VLAN splinter port, and pushes
3880 * a VLAN header onto 'packet' (if it is nonnull).
3882 * Optionally, if 'initial_vals' is nonnull, sets 'initial_vals->vlan_tci'
3883 * to the VLAN TCI with which the packet was really received, that is, the
3884 * actual VLAN TCI extracted by odp_flow_key_to_flow(). (This differs from
3885 * the value returned in flow->vlan_tci only for packets received on
3888 * Similarly, this function also includes some logic to help with tunnels. It
3889 * may modify 'flow' as necessary to make the tunneling implementation
3890 * transparent to the upcall processing logic.
3892 * Returns 0 if successful, ENODEV if the parsed flow has no associated ofport,
3893 * or some other positive errno if there are other problems. */
3895 ofproto_receive(const struct dpif_backer *backer, struct ofpbuf *packet,
3896 const struct nlattr *key, size_t key_len,
3897 struct flow *flow, enum odp_key_fitness *fitnessp,
3898 struct ofproto_dpif **ofproto, uint32_t *odp_in_port,
3899 struct initial_vals *initial_vals)
3901 const struct ofport_dpif *port;
3902 enum odp_key_fitness fitness;
3905 fitness = odp_flow_key_to_flow(key, key_len, flow);
3906 if (fitness == ODP_FIT_ERROR) {
3912 initial_vals->vlan_tci = flow->vlan_tci;
3916 *odp_in_port = flow->in_port;
3919 port = (tnl_port_should_receive(flow)
3920 ? ofport_dpif_cast(tnl_port_receive(flow))
3921 : odp_port_to_ofport(backer, flow->in_port));
3922 flow->in_port = port ? port->up.ofp_port : OFPP_NONE;
3927 /* XXX: Since the tunnel module is not scoped per backer, for a tunnel port
3928 * it's theoretically possible that we'll receive an ofport belonging to an
3929 * entirely different datapath. In practice, this can't happen because no
3930 * platforms has two separate datapaths which each support tunneling. */
3931 ovs_assert(ofproto_dpif_cast(port->up.ofproto)->backer == backer);
3933 if (vsp_adjust_flow(ofproto_dpif_cast(port->up.ofproto), flow)) {
3935 /* Make the packet resemble the flow, so that it gets sent to
3936 * an OpenFlow controller properly, so that it looks correct
3937 * for sFlow, and so that flow_extract() will get the correct
3938 * vlan_tci if it is called on 'packet'.
3940 * The allocated space inside 'packet' probably also contains
3941 * 'key', that is, both 'packet' and 'key' are probably part of
3942 * a struct dpif_upcall (see the large comment on that
3943 * structure definition), so pushing data on 'packet' is in
3944 * general not a good idea since it could overwrite 'key' or
3945 * free it as a side effect. However, it's OK in this special
3946 * case because we know that 'packet' is inside a Netlink
3947 * attribute: pushing 4 bytes will just overwrite the 4-byte
3948 * "struct nlattr", which is fine since we don't need that
3949 * header anymore. */
3950 eth_push_vlan(packet, flow->vlan_tci);
3952 /* We can't reproduce 'key' from 'flow'. */
3953 fitness = fitness == ODP_FIT_PERFECT ? ODP_FIT_TOO_MUCH : fitness;
3958 *ofproto = ofproto_dpif_cast(port->up.ofproto);
3963 *fitnessp = fitness;
3969 handle_miss_upcalls(struct dpif_backer *backer, struct dpif_upcall *upcalls,
3972 struct dpif_upcall *upcall;
3973 struct flow_miss *miss;
3974 struct flow_miss misses[FLOW_MISS_MAX_BATCH];
3975 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
3976 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
3986 /* Construct the to-do list.
3988 * This just amounts to extracting the flow from each packet and sticking
3989 * the packets that have the same flow in the same "flow_miss" structure so
3990 * that we can process them together. */
3993 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
3994 struct flow_miss *miss = &misses[n_misses];
3995 struct flow_miss *existing_miss;
3996 struct ofproto_dpif *ofproto;
3997 uint32_t odp_in_port;
4002 error = ofproto_receive(backer, upcall->packet, upcall->key,
4003 upcall->key_len, &flow, &miss->key_fitness,
4004 &ofproto, &odp_in_port, &miss->initial_vals);
4005 if (error == ENODEV) {
4006 struct drop_key *drop_key;
4008 /* Received packet on port for which we couldn't associate
4009 * an ofproto. This can happen if a port is removed while
4010 * traffic is being received. Print a rate-limited message
4011 * in case it happens frequently. Install a drop flow so
4012 * that future packets of the flow are inexpensively dropped
4014 VLOG_INFO_RL(&rl, "received packet on unassociated port %"PRIu32,
4017 drop_key = drop_key_lookup(backer, upcall->key, upcall->key_len);
4019 drop_key = xmalloc(sizeof *drop_key);
4020 drop_key->key = xmemdup(upcall->key, upcall->key_len);
4021 drop_key->key_len = upcall->key_len;
4023 hmap_insert(&backer->drop_keys, &drop_key->hmap_node,
4024 hash_bytes(drop_key->key, drop_key->key_len, 0));
4025 dpif_flow_put(backer->dpif, DPIF_FP_CREATE | DPIF_FP_MODIFY,
4026 drop_key->key, drop_key->key_len, NULL, 0, NULL);
4034 ofproto->n_missed++;
4035 flow_extract(upcall->packet, flow.skb_priority, flow.skb_mark,
4036 &flow.tunnel, flow.in_port, &miss->flow);
4038 /* Add other packets to a to-do list. */
4039 hash = flow_hash(&miss->flow, 0);
4040 existing_miss = flow_miss_find(&todo, ofproto, &miss->flow, hash);
4041 if (!existing_miss) {
4042 hmap_insert(&todo, &miss->hmap_node, hash);
4043 miss->ofproto = ofproto;
4044 miss->key = upcall->key;
4045 miss->key_len = upcall->key_len;
4046 miss->upcall_type = upcall->type;
4047 miss->odp_in_port = odp_in_port;
4048 list_init(&miss->packets);
4052 miss = existing_miss;
4054 list_push_back(&miss->packets, &upcall->packet->list_node);
4057 /* Process each element in the to-do list, constructing the set of
4058 * operations to batch. */
4060 HMAP_FOR_EACH (miss, hmap_node, &todo) {
4061 handle_flow_miss(miss, flow_miss_ops, &n_ops);
4063 ovs_assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
4065 /* Execute batch. */
4066 for (i = 0; i < n_ops; i++) {
4067 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
4069 dpif_operate(backer->dpif, dpif_ops, n_ops);
4072 for (i = 0; i < n_ops; i++) {
4073 free(flow_miss_ops[i].garbage);
4075 hmap_destroy(&todo);
4078 static enum { SFLOW_UPCALL, MISS_UPCALL, BAD_UPCALL, FLOW_SAMPLE_UPCALL,
4080 classify_upcall(const struct dpif_upcall *upcall)
4082 size_t userdata_len;
4083 union user_action_cookie cookie;
4085 /* First look at the upcall type. */
4086 switch (upcall->type) {
4087 case DPIF_UC_ACTION:
4093 case DPIF_N_UC_TYPES:
4095 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
4099 /* "action" upcalls need a closer look. */
4100 if (!upcall->userdata) {
4101 VLOG_WARN_RL(&rl, "action upcall missing cookie");
4104 userdata_len = nl_attr_get_size(upcall->userdata);
4105 if (userdata_len < sizeof cookie.type
4106 || userdata_len > sizeof cookie) {
4107 VLOG_WARN_RL(&rl, "action upcall cookie has unexpected size %zu",
4111 memset(&cookie, 0, sizeof cookie);
4112 memcpy(&cookie, nl_attr_get(upcall->userdata), userdata_len);
4113 if (userdata_len == sizeof cookie.sflow
4114 && cookie.type == USER_ACTION_COOKIE_SFLOW) {
4115 return SFLOW_UPCALL;
4116 } else if (userdata_len == sizeof cookie.slow_path
4117 && cookie.type == USER_ACTION_COOKIE_SLOW_PATH) {
4119 } else if (userdata_len == sizeof cookie.flow_sample
4120 && cookie.type == USER_ACTION_COOKIE_FLOW_SAMPLE) {
4121 return FLOW_SAMPLE_UPCALL;
4122 } else if (userdata_len == sizeof cookie.ipfix
4123 && cookie.type == USER_ACTION_COOKIE_IPFIX) {
4124 return IPFIX_UPCALL;
4126 VLOG_WARN_RL(&rl, "invalid user cookie of type %"PRIu16
4127 " and size %zu", cookie.type, userdata_len);
4133 handle_sflow_upcall(struct dpif_backer *backer,
4134 const struct dpif_upcall *upcall)
4136 struct ofproto_dpif *ofproto;
4137 union user_action_cookie cookie;
4139 uint32_t odp_in_port;
4141 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
4142 &flow, NULL, &ofproto, &odp_in_port, NULL)
4143 || !ofproto->sflow) {
4147 memset(&cookie, 0, sizeof cookie);
4148 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.sflow);
4149 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
4150 odp_in_port, &cookie);
4154 handle_flow_sample_upcall(struct dpif_backer *backer,
4155 const struct dpif_upcall *upcall)
4157 struct ofproto_dpif *ofproto;
4158 union user_action_cookie cookie;
4161 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
4162 &flow, NULL, &ofproto, NULL, NULL)
4163 || !ofproto->ipfix) {
4167 memset(&cookie, 0, sizeof cookie);
4168 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.flow_sample);
4170 /* The flow reflects exactly the contents of the packet. Sample
4171 * the packet using it. */
4172 dpif_ipfix_flow_sample(ofproto->ipfix, upcall->packet, &flow,
4173 cookie.flow_sample.collector_set_id,
4174 cookie.flow_sample.probability,
4175 cookie.flow_sample.obs_domain_id,
4176 cookie.flow_sample.obs_point_id);
4180 handle_ipfix_upcall(struct dpif_backer *backer,
4181 const struct dpif_upcall *upcall)
4183 struct ofproto_dpif *ofproto;
4186 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
4187 &flow, NULL, &ofproto, NULL, NULL)
4188 || !ofproto->ipfix) {
4192 /* The flow reflects exactly the contents of the packet. Sample
4193 * the packet using it. */
4194 dpif_ipfix_bridge_sample(ofproto->ipfix, upcall->packet, &flow);
4198 handle_upcalls(struct dpif_backer *backer, unsigned int max_batch)
4200 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
4201 struct ofpbuf miss_bufs[FLOW_MISS_MAX_BATCH];
4202 uint64_t miss_buf_stubs[FLOW_MISS_MAX_BATCH][4096 / 8];
4207 ovs_assert(max_batch <= FLOW_MISS_MAX_BATCH);
4210 for (n_processed = 0; n_processed < max_batch; n_processed++) {
4211 struct dpif_upcall *upcall = &misses[n_misses];
4212 struct ofpbuf *buf = &miss_bufs[n_misses];
4215 ofpbuf_use_stub(buf, miss_buf_stubs[n_misses],
4216 sizeof miss_buf_stubs[n_misses]);
4217 error = dpif_recv(backer->dpif, upcall, buf);
4223 switch (classify_upcall(upcall)) {
4225 /* Handle it later. */
4230 handle_sflow_upcall(backer, upcall);
4234 case FLOW_SAMPLE_UPCALL:
4235 handle_flow_sample_upcall(backer, upcall);
4240 handle_ipfix_upcall(backer, upcall);
4250 /* Handle deferred MISS_UPCALL processing. */
4251 handle_miss_upcalls(backer, misses, n_misses);
4252 for (i = 0; i < n_misses; i++) {
4253 ofpbuf_uninit(&miss_bufs[i]);
4259 /* Flow expiration. */
4261 static int subfacet_max_idle(const struct ofproto_dpif *);
4262 static void update_stats(struct dpif_backer *);
4263 static void rule_expire(struct rule_dpif *);
4264 static void expire_subfacets(struct ofproto_dpif *, int dp_max_idle);
4266 /* This function is called periodically by run(). Its job is to collect
4267 * updates for the flows that have been installed into the datapath, most
4268 * importantly when they last were used, and then use that information to
4269 * expire flows that have not been used recently.
4271 * Returns the number of milliseconds after which it should be called again. */
4273 expire(struct dpif_backer *backer)
4275 struct ofproto_dpif *ofproto;
4276 int max_idle = INT32_MAX;
4278 /* Periodically clear out the drop keys in an effort to keep them
4279 * relatively few. */
4280 drop_key_clear(backer);
4282 /* Update stats for each flow in the backer. */
4283 update_stats(backer);
4285 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4286 struct rule *rule, *next_rule;
4289 if (ofproto->backer != backer) {
4293 /* Keep track of the max number of flows per ofproto_dpif. */
4294 update_max_subfacet_count(ofproto);
4296 /* Expire subfacets that have been idle too long. */
4297 dp_max_idle = subfacet_max_idle(ofproto);
4298 expire_subfacets(ofproto, dp_max_idle);
4300 max_idle = MIN(max_idle, dp_max_idle);
4302 /* Expire OpenFlow flows whose idle_timeout or hard_timeout
4304 LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
4305 &ofproto->up.expirable) {
4306 rule_expire(rule_dpif_cast(rule));
4309 /* All outstanding data in existing flows has been accounted, so it's a
4310 * good time to do bond rebalancing. */
4311 if (ofproto->has_bonded_bundles) {
4312 struct ofbundle *bundle;
4314 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
4316 bond_rebalance(bundle->bond, &backer->revalidate_set);
4322 return MIN(max_idle, 1000);
4325 /* Updates flow table statistics given that the datapath just reported 'stats'
4326 * as 'subfacet''s statistics. */
4328 update_subfacet_stats(struct subfacet *subfacet,
4329 const struct dpif_flow_stats *stats)
4331 struct facet *facet = subfacet->facet;
4333 if (stats->n_packets >= subfacet->dp_packet_count) {
4334 uint64_t extra = stats->n_packets - subfacet->dp_packet_count;
4335 facet->packet_count += extra;
4337 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
4340 if (stats->n_bytes >= subfacet->dp_byte_count) {
4341 facet->byte_count += stats->n_bytes - subfacet->dp_byte_count;
4343 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
4346 subfacet->dp_packet_count = stats->n_packets;
4347 subfacet->dp_byte_count = stats->n_bytes;
4349 facet->tcp_flags |= stats->tcp_flags;
4351 subfacet_update_time(subfacet, stats->used);
4352 if (facet->accounted_bytes < facet->byte_count) {
4354 facet_account(facet);
4355 facet->accounted_bytes = facet->byte_count;
4359 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
4360 * about, or a flow that shouldn't be installed but was anyway. Delete it. */
4362 delete_unexpected_flow(struct ofproto_dpif *ofproto,
4363 const struct nlattr *key, size_t key_len)
4365 if (!VLOG_DROP_WARN(&rl)) {
4369 odp_flow_key_format(key, key_len, &s);
4370 VLOG_WARN("unexpected flow on %s: %s", ofproto->up.name, ds_cstr(&s));
4374 COVERAGE_INC(facet_unexpected);
4375 dpif_flow_del(ofproto->backer->dpif, key, key_len, NULL);
4378 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
4380 * This function also pushes statistics updates to rules which each facet
4381 * resubmits into. Generally these statistics will be accurate. However, if a
4382 * facet changes the rule it resubmits into at some time in between
4383 * update_stats() runs, it is possible that statistics accrued to the
4384 * old rule will be incorrectly attributed to the new rule. This could be
4385 * avoided by calling update_stats() whenever rules are created or
4386 * deleted. However, the performance impact of making so many calls to the
4387 * datapath do not justify the benefit of having perfectly accurate statistics.
4389 * In addition, this function maintains per ofproto flow hit counts. The patch
4390 * port is not treated specially. e.g. A packet ingress from br0 patched into
4391 * br1 will increase the hit count of br0 by 1, however, does not affect
4392 * the hit or miss counts of br1.
4395 update_stats(struct dpif_backer *backer)
4397 const struct dpif_flow_stats *stats;
4398 struct dpif_flow_dump dump;
4399 const struct nlattr *key;
4400 struct ofproto_dpif *ofproto;
4403 dpif_flow_dump_start(&dump, backer->dpif);
4404 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
4406 struct subfacet *subfacet;
4407 struct ofport_dpif *ofport;
4410 if (ofproto_receive(backer, NULL, key, key_len, &flow, NULL, &ofproto,
4415 ofproto->total_subfacet_count += hmap_count(&ofproto->subfacets);
4416 ofproto->n_update_stats++;
4418 ofport = get_ofp_port(ofproto, flow.in_port);
4419 if (ofport && ofport->tnl_port) {
4420 netdev_vport_inc_rx(ofport->up.netdev, stats);
4423 key_hash = odp_flow_key_hash(key, key_len);
4424 subfacet = subfacet_find(ofproto, key, key_len, key_hash);
4425 switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
4427 /* Update ofproto_dpif's hit count. */
4428 if (stats->n_packets > subfacet->dp_packet_count) {
4429 uint64_t delta = stats->n_packets - subfacet->dp_packet_count;
4430 dpif_stats_update_hit_count(ofproto, delta);
4433 update_subfacet_stats(subfacet, stats);
4437 /* Stats are updated per-packet. */
4440 case SF_NOT_INSTALLED:
4442 delete_unexpected_flow(ofproto, key, key_len);
4447 dpif_flow_dump_done(&dump);
4449 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4450 update_moving_averages(ofproto);
4455 /* Calculates and returns the number of milliseconds of idle time after which
4456 * subfacets should expire from the datapath. When a subfacet expires, we fold
4457 * its statistics into its facet, and when a facet's last subfacet expires, we
4458 * fold its statistic into its rule. */
4460 subfacet_max_idle(const struct ofproto_dpif *ofproto)
4463 * Idle time histogram.
4465 * Most of the time a switch has a relatively small number of subfacets.
4466 * When this is the case we might as well keep statistics for all of them
4467 * in userspace and to cache them in the kernel datapath for performance as
4470 * As the number of subfacets increases, the memory required to maintain
4471 * statistics about them in userspace and in the kernel becomes
4472 * significant. However, with a large number of subfacets it is likely
4473 * that only a few of them are "heavy hitters" that consume a large amount
4474 * of bandwidth. At this point, only heavy hitters are worth caching in
4475 * the kernel and maintaining in userspaces; other subfacets we can
4478 * The technique used to compute the idle time is to build a histogram with
4479 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
4480 * that is installed in the kernel gets dropped in the appropriate bucket.
4481 * After the histogram has been built, we compute the cutoff so that only
4482 * the most-recently-used 1% of subfacets (but at least
4483 * ofproto->up.flow_eviction_threshold flows) are kept cached. At least
4484 * the most-recently-used bucket of subfacets is kept, so actually an
4485 * arbitrary number of subfacets can be kept in any given expiration run
4486 * (though the next run will delete most of those unless they receive
4489 * This requires a second pass through the subfacets, in addition to the
4490 * pass made by update_stats(), because the former function never looks at
4491 * uninstallable subfacets.
4493 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
4494 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
4495 int buckets[N_BUCKETS] = { 0 };
4496 int total, subtotal, bucket;
4497 struct subfacet *subfacet;
4501 total = hmap_count(&ofproto->subfacets);
4502 if (total <= ofproto->up.flow_eviction_threshold) {
4503 return N_BUCKETS * BUCKET_WIDTH;
4506 /* Build histogram. */
4508 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
4509 long long int idle = now - subfacet->used;
4510 int bucket = (idle <= 0 ? 0
4511 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
4512 : (unsigned int) idle / BUCKET_WIDTH);
4516 /* Find the first bucket whose flows should be expired. */
4517 subtotal = bucket = 0;
4519 subtotal += buckets[bucket++];
4520 } while (bucket < N_BUCKETS &&
4521 subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100));
4523 if (VLOG_IS_DBG_ENABLED()) {
4527 ds_put_cstr(&s, "keep");
4528 for (i = 0; i < N_BUCKETS; i++) {
4530 ds_put_cstr(&s, ", drop");
4533 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
4536 VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s));
4540 return bucket * BUCKET_WIDTH;
4544 expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle)
4546 /* Cutoff time for most flows. */
4547 long long int normal_cutoff = time_msec() - dp_max_idle;
4549 /* We really want to keep flows for special protocols around, so use a more
4550 * conservative cutoff. */
4551 long long int special_cutoff = time_msec() - 10000;
4553 struct subfacet *subfacet, *next_subfacet;
4554 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
4558 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
4559 &ofproto->subfacets) {
4560 long long int cutoff;
4562 cutoff = (subfacet->slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP | SLOW_STP)
4565 if (subfacet->used < cutoff) {
4566 if (subfacet->path != SF_NOT_INSTALLED) {
4567 batch[n_batch++] = subfacet;
4568 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
4569 subfacet_destroy_batch(ofproto, batch, n_batch);
4573 subfacet_destroy(subfacet);
4579 subfacet_destroy_batch(ofproto, batch, n_batch);
4583 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
4584 * then delete it entirely. */
4586 rule_expire(struct rule_dpif *rule)
4588 struct facet *facet, *next_facet;
4592 if (rule->up.pending) {
4593 /* We'll have to expire it later. */
4597 /* Has 'rule' expired? */
4599 if (rule->up.hard_timeout
4600 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
4601 reason = OFPRR_HARD_TIMEOUT;
4602 } else if (rule->up.idle_timeout
4603 && now > rule->up.used + rule->up.idle_timeout * 1000) {
4604 reason = OFPRR_IDLE_TIMEOUT;
4609 COVERAGE_INC(ofproto_dpif_expired);
4611 /* Update stats. (This is a no-op if the rule expired due to an idle
4612 * timeout, because that only happens when the rule has no facets left.) */
4613 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4614 facet_remove(facet);
4617 /* Get rid of the rule. */
4618 ofproto_rule_expire(&rule->up, reason);
4623 /* Creates and returns a new facet owned by 'rule', given a 'flow'.
4625 * The caller must already have determined that no facet with an identical
4626 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
4627 * the ofproto's classifier table.
4629 * 'hash' must be the return value of flow_hash(flow, 0).
4631 * The facet will initially have no subfacets. The caller should create (at
4632 * least) one subfacet with subfacet_create(). */
4633 static struct facet *
4634 facet_create(struct rule_dpif *rule, const struct flow *flow, uint32_t hash)
4636 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4637 struct facet *facet;
4639 facet = xzalloc(sizeof *facet);
4640 facet->used = time_msec();
4641 hmap_insert(&ofproto->facets, &facet->hmap_node, hash);
4642 list_push_back(&rule->facets, &facet->list_node);
4644 facet->flow = *flow;
4645 list_init(&facet->subfacets);
4646 netflow_flow_init(&facet->nf_flow);
4647 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
4649 facet->learn_rl = time_msec() + 500;
4655 facet_free(struct facet *facet)
4660 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
4661 * 'packet', which arrived on 'in_port'. */
4663 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
4664 const struct nlattr *odp_actions, size_t actions_len,
4665 struct ofpbuf *packet)
4667 struct odputil_keybuf keybuf;
4671 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4672 odp_flow_key_from_flow(&key, flow,
4673 ofp_port_to_odp_port(ofproto, flow->in_port));
4675 error = dpif_execute(ofproto->backer->dpif, key.data, key.size,
4676 odp_actions, actions_len, packet);
4680 /* Remove 'facet' from 'ofproto' and free up the associated memory:
4682 * - If 'facet' was installed in the datapath, uninstalls it and updates its
4683 * rule's statistics, via subfacet_uninstall().
4685 * - Removes 'facet' from its rule and from ofproto->facets.
4688 facet_remove(struct facet *facet)
4690 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4691 struct subfacet *subfacet, *next_subfacet;
4693 ovs_assert(!list_is_empty(&facet->subfacets));
4695 /* First uninstall all of the subfacets to get final statistics. */
4696 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4697 subfacet_uninstall(subfacet);
4700 /* Flush the final stats to the rule.
4702 * This might require us to have at least one subfacet around so that we
4703 * can use its actions for accounting in facet_account(), which is why we
4704 * have uninstalled but not yet destroyed the subfacets. */
4705 facet_flush_stats(facet);
4707 /* Now we're really all done so destroy everything. */
4708 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
4709 &facet->subfacets) {
4710 subfacet_destroy__(subfacet);
4712 hmap_remove(&ofproto->facets, &facet->hmap_node);
4713 list_remove(&facet->list_node);
4717 /* Feed information from 'facet' back into the learning table to keep it in
4718 * sync with what is actually flowing through the datapath. */
4720 facet_learn(struct facet *facet)
4722 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4723 struct subfacet *subfacet= CONTAINER_OF(list_front(&facet->subfacets),
4724 struct subfacet, list_node);
4725 long long int now = time_msec();
4726 struct action_xlate_ctx ctx;
4728 if (!facet->has_fin_timeout && now < facet->learn_rl) {
4732 facet->learn_rl = now + 500;
4734 if (!facet->has_learn
4735 && !facet->has_normal
4736 && (!facet->has_fin_timeout
4737 || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
4741 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
4742 &subfacet->initial_vals,
4743 facet->rule, facet->tcp_flags, NULL);
4744 ctx.may_learn = true;
4745 xlate_actions_for_side_effects(&ctx, facet->rule->up.ofpacts,
4746 facet->rule->up.ofpacts_len);
4750 facet_account(struct facet *facet)
4752 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4753 struct subfacet *subfacet = facet_get_subfacet(facet);
4754 const struct nlattr *a;
4759 if (!facet->has_normal || !ofproto->has_bonded_bundles) {
4762 n_bytes = facet->byte_count - facet->accounted_bytes;
4764 /* This loop feeds byte counters to bond_account() for rebalancing to use
4765 * as a basis. We also need to track the actual VLAN on which the packet
4766 * is going to be sent to ensure that it matches the one passed to
4767 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
4770 * We use the actions from an arbitrary subfacet because they should all
4771 * be equally valid for our purpose. */
4772 vlan_tci = facet->flow.vlan_tci;
4773 NL_ATTR_FOR_EACH_UNSAFE (a, left,
4774 subfacet->actions, subfacet->actions_len) {
4775 const struct ovs_action_push_vlan *vlan;
4776 struct ofport_dpif *port;
4778 switch (nl_attr_type(a)) {
4779 case OVS_ACTION_ATTR_OUTPUT:
4780 port = get_odp_port(ofproto, nl_attr_get_u32(a));
4781 if (port && port->bundle && port->bundle->bond) {
4782 bond_account(port->bundle->bond, &facet->flow,
4783 vlan_tci_to_vid(vlan_tci), n_bytes);
4787 case OVS_ACTION_ATTR_POP_VLAN:
4788 vlan_tci = htons(0);
4791 case OVS_ACTION_ATTR_PUSH_VLAN:
4792 vlan = nl_attr_get(a);
4793 vlan_tci = vlan->vlan_tci;
4799 /* Returns true if the only action for 'facet' is to send to the controller.
4800 * (We don't report NetFlow expiration messages for such facets because they
4801 * are just part of the control logic for the network, not real traffic). */
4803 facet_is_controller_flow(struct facet *facet)
4806 const struct rule *rule = &facet->rule->up;
4807 const struct ofpact *ofpacts = rule->ofpacts;
4808 size_t ofpacts_len = rule->ofpacts_len;
4810 if (ofpacts_len > 0 &&
4811 ofpacts->type == OFPACT_CONTROLLER &&
4812 ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len)) {
4819 /* Folds all of 'facet''s statistics into its rule. Also updates the
4820 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
4821 * 'facet''s statistics in the datapath should have been zeroed and folded into
4822 * its packet and byte counts before this function is called. */
4824 facet_flush_stats(struct facet *facet)
4826 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4827 struct subfacet *subfacet;
4829 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4830 ovs_assert(!subfacet->dp_byte_count);
4831 ovs_assert(!subfacet->dp_packet_count);
4834 facet_push_stats(facet);
4835 if (facet->accounted_bytes < facet->byte_count) {
4836 facet_account(facet);
4837 facet->accounted_bytes = facet->byte_count;
4840 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
4841 struct ofexpired expired;
4842 expired.flow = facet->flow;
4843 expired.packet_count = facet->packet_count;
4844 expired.byte_count = facet->byte_count;
4845 expired.used = facet->used;
4846 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4849 /* Reset counters to prevent double counting if 'facet' ever gets
4851 facet_reset_counters(facet);
4853 netflow_flow_clear(&facet->nf_flow);
4854 facet->tcp_flags = 0;
4857 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
4858 * Returns it if found, otherwise a null pointer.
4860 * 'hash' must be the return value of flow_hash(flow, 0).
4862 * The returned facet might need revalidation; use facet_lookup_valid()
4863 * instead if that is important. */
4864 static struct facet *
4865 facet_find(struct ofproto_dpif *ofproto,
4866 const struct flow *flow, uint32_t hash)
4868 struct facet *facet;
4870 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, hash, &ofproto->facets) {
4871 if (flow_equal(flow, &facet->flow)) {
4879 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
4880 * Returns it if found, otherwise a null pointer.
4882 * 'hash' must be the return value of flow_hash(flow, 0).
4884 * The returned facet is guaranteed to be valid. */
4885 static struct facet *
4886 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow,
4889 struct facet *facet;
4891 facet = facet_find(ofproto, flow, hash);
4893 && (ofproto->backer->need_revalidate
4894 || tag_set_intersects(&ofproto->backer->revalidate_set,
4896 facet_revalidate(facet);
4898 /* facet_revalidate() may have destroyed 'facet'. */
4899 facet = facet_find(ofproto, flow, hash);
4905 /* Return a subfacet from 'facet'. A facet consists of one or more
4906 * subfacets, and this function returns one of them. */
4907 static struct subfacet *facet_get_subfacet(struct facet *facet)
4909 return CONTAINER_OF(list_front(&facet->subfacets), struct subfacet,
4914 subfacet_path_to_string(enum subfacet_path path)
4917 case SF_NOT_INSTALLED:
4918 return "not installed";
4920 return "in fast path";
4922 return "in slow path";
4928 /* Returns the path in which a subfacet should be installed if its 'slow'
4929 * member has the specified value. */
4930 static enum subfacet_path
4931 subfacet_want_path(enum slow_path_reason slow)
4933 return slow ? SF_SLOW_PATH : SF_FAST_PATH;
4936 /* Returns true if 'subfacet' needs to have its datapath flow updated,
4937 * supposing that its actions have been recalculated as 'want_actions' and that
4938 * 'slow' is nonzero iff 'subfacet' should be in the slow path. */
4940 subfacet_should_install(struct subfacet *subfacet, enum slow_path_reason slow,
4941 const struct ofpbuf *want_actions)
4943 enum subfacet_path want_path = subfacet_want_path(slow);
4944 return (want_path != subfacet->path
4945 || (want_path == SF_FAST_PATH
4946 && (subfacet->actions_len != want_actions->size
4947 || memcmp(subfacet->actions, want_actions->data,
4948 subfacet->actions_len))));
4952 facet_check_consistency(struct facet *facet)
4954 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
4956 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4958 uint64_t odp_actions_stub[1024 / 8];
4959 struct ofpbuf odp_actions;
4961 struct rule_dpif *rule;
4962 struct subfacet *subfacet;
4963 bool may_log = false;
4966 /* Check the rule for consistency. */
4967 rule = rule_dpif_lookup(ofproto, &facet->flow);
4968 ok = rule == facet->rule;
4970 may_log = !VLOG_DROP_WARN(&rl);
4975 flow_format(&s, &facet->flow);
4976 ds_put_format(&s, ": facet associated with wrong rule (was "
4977 "table=%"PRIu8",", facet->rule->up.table_id);
4978 cls_rule_format(&facet->rule->up.cr, &s);
4979 ds_put_format(&s, ") (should have been table=%"PRIu8",",
4981 cls_rule_format(&rule->up.cr, &s);
4982 ds_put_char(&s, ')');
4984 VLOG_WARN("%s", ds_cstr(&s));
4989 /* Check the datapath actions for consistency. */
4990 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
4991 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4992 enum subfacet_path want_path;
4993 struct action_xlate_ctx ctx;
4996 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
4997 &subfacet->initial_vals, rule, 0, NULL);
4998 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len,
5001 if (subfacet->path == SF_NOT_INSTALLED) {
5002 /* This only happens if the datapath reported an error when we
5003 * tried to install the flow. Don't flag another error here. */
5007 want_path = subfacet_want_path(subfacet->slow);
5008 if (want_path == SF_SLOW_PATH && subfacet->path == SF_SLOW_PATH) {
5009 /* The actions for slow-path flows may legitimately vary from one
5010 * packet to the next. We're done. */
5014 if (!subfacet_should_install(subfacet, subfacet->slow, &odp_actions)) {
5018 /* Inconsistency! */
5020 may_log = !VLOG_DROP_WARN(&rl);
5024 /* Rate-limited, skip reporting. */
5029 odp_flow_key_format(subfacet->key, subfacet->key_len, &s);
5031 ds_put_cstr(&s, ": inconsistency in subfacet");
5032 if (want_path != subfacet->path) {
5033 enum odp_key_fitness fitness = subfacet->key_fitness;
5035 ds_put_format(&s, " (%s, fitness=%s)",
5036 subfacet_path_to_string(subfacet->path),
5037 odp_key_fitness_to_string(fitness));
5038 ds_put_format(&s, " (should have been %s)",
5039 subfacet_path_to_string(want_path));
5040 } else if (want_path == SF_FAST_PATH) {
5041 ds_put_cstr(&s, " (actions were: ");
5042 format_odp_actions(&s, subfacet->actions,
5043 subfacet->actions_len);
5044 ds_put_cstr(&s, ") (correct actions: ");
5045 format_odp_actions(&s, odp_actions.data, odp_actions.size);
5046 ds_put_char(&s, ')');
5048 ds_put_cstr(&s, " (actions: ");
5049 format_odp_actions(&s, subfacet->actions,
5050 subfacet->actions_len);
5051 ds_put_char(&s, ')');
5053 VLOG_WARN("%s", ds_cstr(&s));
5056 ofpbuf_uninit(&odp_actions);
5061 /* Re-searches the classifier for 'facet':
5063 * - If the rule found is different from 'facet''s current rule, moves
5064 * 'facet' to the new rule and recompiles its actions.
5066 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
5067 * where it is and recompiles its actions anyway.
5069 * - If any of 'facet''s subfacets correspond to a new flow according to
5070 * ofproto_receive(), 'facet' is removed. */
5072 facet_revalidate(struct facet *facet)
5074 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5076 struct nlattr *odp_actions;
5079 struct actions *new_actions;
5081 struct action_xlate_ctx ctx;
5082 uint64_t odp_actions_stub[1024 / 8];
5083 struct ofpbuf odp_actions;
5085 struct rule_dpif *new_rule;
5086 struct subfacet *subfacet;
5089 COVERAGE_INC(facet_revalidate);
5091 /* Check that child subfacets still correspond to this facet. Tunnel
5092 * configuration changes could cause a subfacet's OpenFlow in_port to
5094 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
5095 struct ofproto_dpif *recv_ofproto;
5096 struct flow recv_flow;
5099 error = ofproto_receive(ofproto->backer, NULL, subfacet->key,
5100 subfacet->key_len, &recv_flow, NULL,
5101 &recv_ofproto, NULL, NULL);
5103 || recv_ofproto != ofproto
5104 || memcmp(&recv_flow, &facet->flow, sizeof recv_flow)) {
5105 facet_remove(facet);
5110 new_rule = rule_dpif_lookup(ofproto, &facet->flow);
5112 /* Calculate new datapath actions.
5114 * We do not modify any 'facet' state yet, because we might need to, e.g.,
5115 * emit a NetFlow expiration and, if so, we need to have the old state
5116 * around to properly compose it. */
5118 /* If the datapath actions changed or the installability changed,
5119 * then we need to talk to the datapath. */
5122 memset(&ctx, 0, sizeof ctx);
5123 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5124 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
5125 enum slow_path_reason slow;
5127 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
5128 &subfacet->initial_vals, new_rule, 0, NULL);
5129 xlate_actions(&ctx, new_rule->up.ofpacts, new_rule->up.ofpacts_len,
5132 slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
5133 if (subfacet_should_install(subfacet, slow, &odp_actions)) {
5134 struct dpif_flow_stats stats;
5136 subfacet_install(subfacet,
5137 odp_actions.data, odp_actions.size, &stats, slow);
5138 subfacet_update_stats(subfacet, &stats);
5141 new_actions = xcalloc(list_size(&facet->subfacets),
5142 sizeof *new_actions);
5144 new_actions[i].odp_actions = xmemdup(odp_actions.data,
5146 new_actions[i].actions_len = odp_actions.size;
5151 ofpbuf_uninit(&odp_actions);
5154 facet_flush_stats(facet);
5157 /* Update 'facet' now that we've taken care of all the old state. */
5158 facet->tags = ctx.tags;
5159 facet->nf_flow.output_iface = ctx.nf_output_iface;
5160 facet->has_learn = ctx.has_learn;
5161 facet->has_normal = ctx.has_normal;
5162 facet->has_fin_timeout = ctx.has_fin_timeout;
5163 facet->mirrors = ctx.mirrors;
5166 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
5167 subfacet->slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
5169 if (new_actions && new_actions[i].odp_actions) {
5170 free(subfacet->actions);
5171 subfacet->actions = new_actions[i].odp_actions;
5172 subfacet->actions_len = new_actions[i].actions_len;
5178 if (facet->rule != new_rule) {
5179 COVERAGE_INC(facet_changed_rule);
5180 list_remove(&facet->list_node);
5181 list_push_back(&new_rule->facets, &facet->list_node);
5182 facet->rule = new_rule;
5183 facet->used = new_rule->up.created;
5184 facet->prev_used = facet->used;
5188 /* Updates 'facet''s used time. Caller is responsible for calling
5189 * facet_push_stats() to update the flows which 'facet' resubmits into. */
5191 facet_update_time(struct facet *facet, long long int used)
5193 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5194 if (used > facet->used) {
5196 ofproto_rule_update_used(&facet->rule->up, used);
5197 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
5202 facet_reset_counters(struct facet *facet)
5204 facet->packet_count = 0;
5205 facet->byte_count = 0;
5206 facet->prev_packet_count = 0;
5207 facet->prev_byte_count = 0;
5208 facet->accounted_bytes = 0;
5212 facet_push_stats(struct facet *facet)
5214 struct dpif_flow_stats stats;
5216 ovs_assert(facet->packet_count >= facet->prev_packet_count);
5217 ovs_assert(facet->byte_count >= facet->prev_byte_count);
5218 ovs_assert(facet->used >= facet->prev_used);
5220 stats.n_packets = facet->packet_count - facet->prev_packet_count;
5221 stats.n_bytes = facet->byte_count - facet->prev_byte_count;
5222 stats.used = facet->used;
5223 stats.tcp_flags = 0;
5225 if (stats.n_packets || stats.n_bytes || facet->used > facet->prev_used) {
5226 facet->prev_packet_count = facet->packet_count;
5227 facet->prev_byte_count = facet->byte_count;
5228 facet->prev_used = facet->used;
5230 rule_credit_stats(facet->rule, &stats);
5231 flow_push_stats(facet, &stats);
5233 update_mirror_stats(ofproto_dpif_cast(facet->rule->up.ofproto),
5234 facet->mirrors, stats.n_packets, stats.n_bytes);
5239 push_all_stats__(bool run_fast)
5241 static long long int rl = LLONG_MIN;
5242 struct ofproto_dpif *ofproto;
5244 if (time_msec() < rl) {
5248 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5249 struct facet *facet;
5251 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
5252 facet_push_stats(facet);
5259 rl = time_msec() + 100;
5263 push_all_stats(void)
5265 push_all_stats__(true);
5269 rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats)
5271 rule->packet_count += stats->n_packets;
5272 rule->byte_count += stats->n_bytes;
5273 ofproto_rule_update_used(&rule->up, stats->used);
5276 /* Pushes flow statistics to the rules which 'facet->flow' resubmits
5277 * into given 'facet->rule''s actions and mirrors. */
5279 flow_push_stats(struct facet *facet, const struct dpif_flow_stats *stats)
5281 struct rule_dpif *rule = facet->rule;
5282 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5283 struct subfacet *subfacet = facet_get_subfacet(facet);
5284 struct action_xlate_ctx ctx;
5286 ofproto_rule_update_used(&rule->up, stats->used);
5288 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
5289 &subfacet->initial_vals, rule, 0, NULL);
5290 ctx.resubmit_stats = stats;
5291 xlate_actions_for_side_effects(&ctx, rule->up.ofpacts,
5292 rule->up.ofpacts_len);
5297 static struct subfacet *
5298 subfacet_find(struct ofproto_dpif *ofproto,
5299 const struct nlattr *key, size_t key_len, uint32_t key_hash)
5301 struct subfacet *subfacet;
5303 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
5304 &ofproto->subfacets) {
5305 if (subfacet->key_len == key_len
5306 && !memcmp(key, subfacet->key, key_len)) {
5314 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
5315 * 'key_fitness', 'key', and 'key_len' members in 'miss'. Returns the
5316 * existing subfacet if there is one, otherwise creates and returns a
5319 * If the returned subfacet is new, then subfacet->actions will be NULL, in
5320 * which case the caller must populate the actions with
5321 * subfacet_make_actions(). */
5322 static struct subfacet *
5323 subfacet_create(struct facet *facet, struct flow_miss *miss,
5326 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5327 enum odp_key_fitness key_fitness = miss->key_fitness;
5328 const struct nlattr *key = miss->key;
5329 size_t key_len = miss->key_len;
5331 struct subfacet *subfacet;
5333 key_hash = odp_flow_key_hash(key, key_len);
5335 if (list_is_empty(&facet->subfacets)) {
5336 subfacet = &facet->one_subfacet;
5338 subfacet = subfacet_find(ofproto, key, key_len, key_hash);
5340 if (subfacet->facet == facet) {
5344 /* This shouldn't happen. */
5345 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
5346 subfacet_destroy(subfacet);
5349 subfacet = xmalloc(sizeof *subfacet);
5352 hmap_insert(&ofproto->subfacets, &subfacet->hmap_node, key_hash);
5353 list_push_back(&facet->subfacets, &subfacet->list_node);
5354 subfacet->facet = facet;
5355 subfacet->key_fitness = key_fitness;
5356 subfacet->key = xmemdup(key, key_len);
5357 subfacet->key_len = key_len;
5358 subfacet->used = now;
5359 subfacet->created = now;
5360 subfacet->dp_packet_count = 0;
5361 subfacet->dp_byte_count = 0;
5362 subfacet->actions_len = 0;
5363 subfacet->actions = NULL;
5364 subfacet->slow = (subfacet->key_fitness == ODP_FIT_TOO_LITTLE
5367 subfacet->path = SF_NOT_INSTALLED;
5368 subfacet->initial_vals = miss->initial_vals;
5369 subfacet->odp_in_port = miss->odp_in_port;
5371 ofproto->subfacet_add_count++;
5375 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
5376 * its facet within 'ofproto', and frees it. */
5378 subfacet_destroy__(struct subfacet *subfacet)
5380 struct facet *facet = subfacet->facet;
5381 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5383 /* Update ofproto stats before uninstall the subfacet. */
5384 ofproto->subfacet_del_count++;
5385 ofproto->total_subfacet_life_span += (time_msec() - subfacet->created);
5387 subfacet_uninstall(subfacet);
5388 hmap_remove(&ofproto->subfacets, &subfacet->hmap_node);
5389 list_remove(&subfacet->list_node);
5390 free(subfacet->key);
5391 free(subfacet->actions);
5392 if (subfacet != &facet->one_subfacet) {
5397 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
5398 * last remaining subfacet in its facet destroys the facet too. */
5400 subfacet_destroy(struct subfacet *subfacet)
5402 struct facet *facet = subfacet->facet;
5404 if (list_is_singleton(&facet->subfacets)) {
5405 /* facet_remove() needs at least one subfacet (it will remove it). */
5406 facet_remove(facet);
5408 subfacet_destroy__(subfacet);
5413 subfacet_destroy_batch(struct ofproto_dpif *ofproto,
5414 struct subfacet **subfacets, int n)
5416 struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH];
5417 struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH];
5418 struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH];
5421 for (i = 0; i < n; i++) {
5422 ops[i].type = DPIF_OP_FLOW_DEL;
5423 ops[i].u.flow_del.key = subfacets[i]->key;
5424 ops[i].u.flow_del.key_len = subfacets[i]->key_len;
5425 ops[i].u.flow_del.stats = &stats[i];
5429 dpif_operate(ofproto->backer->dpif, opsp, n);
5430 for (i = 0; i < n; i++) {
5431 subfacet_reset_dp_stats(subfacets[i], &stats[i]);
5432 subfacets[i]->path = SF_NOT_INSTALLED;
5433 subfacet_destroy(subfacets[i]);
5438 /* Composes the datapath actions for 'subfacet' based on its rule's actions.
5439 * Translates the actions into 'odp_actions', which the caller must have
5440 * initialized and is responsible for uninitializing. */
5442 subfacet_make_actions(struct subfacet *subfacet, const struct ofpbuf *packet,
5443 struct ofpbuf *odp_actions)
5445 struct facet *facet = subfacet->facet;
5446 struct rule_dpif *rule = facet->rule;
5447 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5449 struct action_xlate_ctx ctx;
5451 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
5452 &subfacet->initial_vals, rule, 0, packet);
5453 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, odp_actions);
5454 facet->tags = ctx.tags;
5455 facet->has_learn = ctx.has_learn;
5456 facet->has_normal = ctx.has_normal;
5457 facet->has_fin_timeout = ctx.has_fin_timeout;
5458 facet->nf_flow.output_iface = ctx.nf_output_iface;
5459 facet->mirrors = ctx.mirrors;
5461 subfacet->slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
5462 if (subfacet->actions_len != odp_actions->size
5463 || memcmp(subfacet->actions, odp_actions->data, odp_actions->size)) {
5464 free(subfacet->actions);
5465 subfacet->actions_len = odp_actions->size;
5466 subfacet->actions = xmemdup(odp_actions->data, odp_actions->size);
5470 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
5471 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
5472 * in the datapath will be zeroed and 'stats' will be updated with traffic new
5473 * since 'subfacet' was last updated.
5475 * Returns 0 if successful, otherwise a positive errno value. */
5477 subfacet_install(struct subfacet *subfacet,
5478 const struct nlattr *actions, size_t actions_len,
5479 struct dpif_flow_stats *stats,
5480 enum slow_path_reason slow)
5482 struct facet *facet = subfacet->facet;
5483 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5484 enum subfacet_path path = subfacet_want_path(slow);
5485 uint64_t slow_path_stub[128 / 8];
5486 enum dpif_flow_put_flags flags;
5489 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
5491 flags |= DPIF_FP_ZERO_STATS;
5494 if (path == SF_SLOW_PATH) {
5495 compose_slow_path(ofproto, &facet->flow, slow,
5496 slow_path_stub, sizeof slow_path_stub,
5497 &actions, &actions_len);
5500 ret = dpif_flow_put(ofproto->backer->dpif, flags, subfacet->key,
5501 subfacet->key_len, actions, actions_len, stats);
5504 subfacet_reset_dp_stats(subfacet, stats);
5508 subfacet->path = path;
5514 subfacet_reinstall(struct subfacet *subfacet, struct dpif_flow_stats *stats)
5516 return subfacet_install(subfacet, subfacet->actions, subfacet->actions_len,
5517 stats, subfacet->slow);
5520 /* If 'subfacet' is installed in the datapath, uninstalls it. */
5522 subfacet_uninstall(struct subfacet *subfacet)
5524 if (subfacet->path != SF_NOT_INSTALLED) {
5525 struct rule_dpif *rule = subfacet->facet->rule;
5526 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5527 struct dpif_flow_stats stats;
5530 error = dpif_flow_del(ofproto->backer->dpif, subfacet->key,
5531 subfacet->key_len, &stats);
5532 subfacet_reset_dp_stats(subfacet, &stats);
5534 subfacet_update_stats(subfacet, &stats);
5536 subfacet->path = SF_NOT_INSTALLED;
5538 ovs_assert(subfacet->dp_packet_count == 0);
5539 ovs_assert(subfacet->dp_byte_count == 0);
5543 /* Resets 'subfacet''s datapath statistics counters. This should be called
5544 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
5545 * non-null, it should contain the statistics returned by dpif when 'subfacet'
5546 * was reset in the datapath. 'stats' will be modified to include only
5547 * statistics new since 'subfacet' was last updated. */
5549 subfacet_reset_dp_stats(struct subfacet *subfacet,
5550 struct dpif_flow_stats *stats)
5553 && subfacet->dp_packet_count <= stats->n_packets
5554 && subfacet->dp_byte_count <= stats->n_bytes) {
5555 stats->n_packets -= subfacet->dp_packet_count;
5556 stats->n_bytes -= subfacet->dp_byte_count;
5559 subfacet->dp_packet_count = 0;
5560 subfacet->dp_byte_count = 0;
5563 /* Updates 'subfacet''s used time. The caller is responsible for calling
5564 * facet_push_stats() to update the flows which 'subfacet' resubmits into. */
5566 subfacet_update_time(struct subfacet *subfacet, long long int used)
5568 if (used > subfacet->used) {
5569 subfacet->used = used;
5570 facet_update_time(subfacet->facet, used);
5574 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
5576 * Because of the meaning of a subfacet's counters, it only makes sense to do
5577 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
5578 * represents a packet that was sent by hand or if it represents statistics
5579 * that have been cleared out of the datapath. */
5581 subfacet_update_stats(struct subfacet *subfacet,
5582 const struct dpif_flow_stats *stats)
5584 if (stats->n_packets || stats->used > subfacet->used) {
5585 struct facet *facet = subfacet->facet;
5587 subfacet_update_time(subfacet, stats->used);
5588 facet->packet_count += stats->n_packets;
5589 facet->byte_count += stats->n_bytes;
5590 facet->tcp_flags |= stats->tcp_flags;
5591 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
5597 static struct rule_dpif *
5598 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow)
5600 struct rule_dpif *rule;
5602 rule = rule_dpif_lookup__(ofproto, flow, 0);
5607 return rule_dpif_miss_rule(ofproto, flow);
5610 static struct rule_dpif *
5611 rule_dpif_lookup__(struct ofproto_dpif *ofproto, const struct flow *flow,
5614 struct cls_rule *cls_rule;
5615 struct classifier *cls;
5617 if (table_id >= N_TABLES) {
5621 cls = &ofproto->up.tables[table_id].cls;
5622 if (flow->nw_frag & FLOW_NW_FRAG_ANY
5623 && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
5624 /* For OFPC_NORMAL frag_handling, we must pretend that transport ports
5625 * are unavailable. */
5626 struct flow ofpc_normal_flow = *flow;
5627 ofpc_normal_flow.tp_src = htons(0);
5628 ofpc_normal_flow.tp_dst = htons(0);
5629 cls_rule = classifier_lookup(cls, &ofpc_normal_flow);
5631 cls_rule = classifier_lookup(cls, flow);
5633 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
5636 static struct rule_dpif *
5637 rule_dpif_miss_rule(struct ofproto_dpif *ofproto, const struct flow *flow)
5639 struct ofport_dpif *port;
5641 port = get_ofp_port(ofproto, flow->in_port);
5643 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, flow->in_port);
5644 return ofproto->miss_rule;
5647 if (port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN) {
5648 return ofproto->no_packet_in_rule;
5650 return ofproto->miss_rule;
5654 complete_operation(struct rule_dpif *rule)
5656 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5658 rule_invalidate(rule);
5660 struct dpif_completion *c = xmalloc(sizeof *c);
5661 c->op = rule->up.pending;
5662 list_push_back(&ofproto->completions, &c->list_node);
5664 ofoperation_complete(rule->up.pending, 0);
5668 static struct rule *
5671 struct rule_dpif *rule = xmalloc(sizeof *rule);
5676 rule_dealloc(struct rule *rule_)
5678 struct rule_dpif *rule = rule_dpif_cast(rule_);
5683 rule_construct(struct rule *rule_)
5685 struct rule_dpif *rule = rule_dpif_cast(rule_);
5686 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5687 struct rule_dpif *victim;
5690 rule->packet_count = 0;
5691 rule->byte_count = 0;
5693 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
5694 if (victim && !list_is_empty(&victim->facets)) {
5695 struct facet *facet;
5697 rule->facets = victim->facets;
5698 list_moved(&rule->facets);
5699 LIST_FOR_EACH (facet, list_node, &rule->facets) {
5700 /* XXX: We're only clearing our local counters here. It's possible
5701 * that quite a few packets are unaccounted for in the datapath
5702 * statistics. These will be accounted to the new rule instead of
5703 * cleared as required. This could be fixed by clearing out the
5704 * datapath statistics for this facet, but currently it doesn't
5706 facet_reset_counters(facet);
5710 /* Must avoid list_moved() in this case. */
5711 list_init(&rule->facets);
5714 table_id = rule->up.table_id;
5716 rule->tag = victim->tag;
5717 } else if (table_id == 0) {
5722 miniflow_expand(&rule->up.cr.match.flow, &flow);
5723 rule->tag = rule_calculate_tag(&flow, &rule->up.cr.match.mask,
5724 ofproto->tables[table_id].basis);
5727 complete_operation(rule);
5732 rule_destruct(struct rule *rule_)
5734 struct rule_dpif *rule = rule_dpif_cast(rule_);
5735 struct facet *facet, *next_facet;
5737 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
5738 facet_revalidate(facet);
5741 complete_operation(rule);
5745 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
5747 struct rule_dpif *rule = rule_dpif_cast(rule_);
5749 /* push_all_stats() can handle flow misses which, when using the learn
5750 * action, can cause rules to be added and deleted. This can corrupt our
5751 * caller's datastructures which assume that rule_get_stats() doesn't have
5752 * an impact on the flow table. To be safe, we disable miss handling. */
5753 push_all_stats__(false);
5755 /* Start from historical data for 'rule' itself that are no longer tracked
5756 * in facets. This counts, for example, facets that have expired. */
5757 *packets = rule->packet_count;
5758 *bytes = rule->byte_count;
5762 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
5763 struct ofpbuf *packet)
5765 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5766 struct initial_vals initial_vals;
5767 struct dpif_flow_stats stats;
5768 struct action_xlate_ctx ctx;
5769 uint64_t odp_actions_stub[1024 / 8];
5770 struct ofpbuf odp_actions;
5772 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
5773 rule_credit_stats(rule, &stats);
5775 initial_vals.vlan_tci = flow->vlan_tci;
5776 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5777 action_xlate_ctx_init(&ctx, ofproto, flow, &initial_vals,
5778 rule, stats.tcp_flags, packet);
5779 ctx.resubmit_stats = &stats;
5780 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, &odp_actions);
5782 execute_odp_actions(ofproto, flow, odp_actions.data,
5783 odp_actions.size, packet);
5785 ofpbuf_uninit(&odp_actions);
5789 rule_execute(struct rule *rule, const struct flow *flow,
5790 struct ofpbuf *packet)
5792 rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
5793 ofpbuf_delete(packet);
5798 rule_modify_actions(struct rule *rule_)
5800 struct rule_dpif *rule = rule_dpif_cast(rule_);
5802 complete_operation(rule);
5805 /* Sends 'packet' out 'ofport'.
5806 * May modify 'packet'.
5807 * Returns 0 if successful, otherwise a positive errno value. */
5809 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
5811 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
5812 uint64_t odp_actions_stub[1024 / 8];
5813 struct ofpbuf key, odp_actions;
5814 struct dpif_flow_stats stats;
5815 struct odputil_keybuf keybuf;
5816 struct ofpact_output output;
5817 struct action_xlate_ctx ctx;
5821 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5822 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5824 /* Use OFPP_NONE as the in_port to avoid special packet processing. */
5825 flow_extract(packet, 0, 0, NULL, OFPP_NONE, &flow);
5826 odp_flow_key_from_flow(&key, &flow, ofp_port_to_odp_port(ofproto,
5828 dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
5830 ofpact_init(&output.ofpact, OFPACT_OUTPUT, sizeof output);
5831 output.port = ofport->up.ofp_port;
5834 action_xlate_ctx_init(&ctx, ofproto, &flow, NULL, NULL, 0, packet);
5835 ctx.resubmit_stats = &stats;
5836 xlate_actions(&ctx, &output.ofpact, sizeof output, &odp_actions);
5838 error = dpif_execute(ofproto->backer->dpif,
5840 odp_actions.data, odp_actions.size,
5842 ofpbuf_uninit(&odp_actions);
5845 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %s (%s)",
5846 ofproto->up.name, netdev_get_name(ofport->up.netdev),
5850 ofproto->stats.tx_packets++;
5851 ofproto->stats.tx_bytes += packet->size;
5855 /* OpenFlow to datapath action translation. */
5857 static bool may_receive(const struct ofport_dpif *, struct action_xlate_ctx *);
5858 static void do_xlate_actions(const struct ofpact *, size_t ofpacts_len,
5859 struct action_xlate_ctx *);
5860 static void xlate_normal(struct action_xlate_ctx *);
5862 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
5863 * The action will state 'slow' as the reason that the action is in the slow
5864 * path. (This is purely informational: it allows a human viewing "ovs-dpctl
5865 * dump-flows" output to see why a flow is in the slow path.)
5867 * The 'stub_size' bytes in 'stub' will be used to store the action.
5868 * 'stub_size' must be large enough for the action.
5870 * The action and its size will be stored in '*actionsp' and '*actions_lenp',
5873 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
5874 enum slow_path_reason slow,
5875 uint64_t *stub, size_t stub_size,
5876 const struct nlattr **actionsp, size_t *actions_lenp)
5878 union user_action_cookie cookie;
5881 cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
5882 cookie.slow_path.unused = 0;
5883 cookie.slow_path.reason = slow;
5885 ofpbuf_use_stack(&buf, stub, stub_size);
5886 if (slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP | SLOW_STP)) {
5887 uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif, UINT32_MAX);
5888 odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
5890 put_userspace_action(ofproto, &buf, flow, &cookie,
5891 sizeof cookie.slow_path);
5893 *actionsp = buf.data;
5894 *actions_lenp = buf.size;
5898 put_userspace_action(const struct ofproto_dpif *ofproto,
5899 struct ofpbuf *odp_actions,
5900 const struct flow *flow,
5901 const union user_action_cookie *cookie,
5902 const size_t cookie_size)
5906 pid = dpif_port_get_pid(ofproto->backer->dpif,
5907 ofp_port_to_odp_port(ofproto, flow->in_port));
5909 return odp_put_userspace_action(pid, cookie, cookie_size, odp_actions);
5912 /* Compose SAMPLE action for sFlow or IPFIX. The given probability is
5913 * the number of packets out of UINT32_MAX to sample. The given
5914 * cookie is passed back in the callback for each sampled packet.
5917 compose_sample_action(const struct ofproto_dpif *ofproto,
5918 struct ofpbuf *odp_actions,
5919 const struct flow *flow,
5920 const uint32_t probability,
5921 const union user_action_cookie *cookie,
5922 const size_t cookie_size)
5924 size_t sample_offset, actions_offset;
5927 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
5929 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
5931 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
5932 cookie_offset = put_userspace_action(ofproto, odp_actions, flow, cookie,
5935 nl_msg_end_nested(odp_actions, actions_offset);
5936 nl_msg_end_nested(odp_actions, sample_offset);
5937 return cookie_offset;
5941 compose_sflow_cookie(const struct ofproto_dpif *ofproto,
5942 ovs_be16 vlan_tci, uint32_t odp_port,
5943 unsigned int n_outputs, union user_action_cookie *cookie)
5947 cookie->type = USER_ACTION_COOKIE_SFLOW;
5948 cookie->sflow.vlan_tci = vlan_tci;
5950 /* See http://www.sflow.org/sflow_version_5.txt (search for "Input/output
5951 * port information") for the interpretation of cookie->output. */
5952 switch (n_outputs) {
5954 /* 0x40000000 | 256 means "packet dropped for unknown reason". */
5955 cookie->sflow.output = 0x40000000 | 256;
5959 ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port);
5961 cookie->sflow.output = ifindex;
5966 /* 0x80000000 means "multiple output ports. */
5967 cookie->sflow.output = 0x80000000 | n_outputs;
5972 /* Compose SAMPLE action for sFlow bridge sampling. */
5974 compose_sflow_action(const struct ofproto_dpif *ofproto,
5975 struct ofpbuf *odp_actions,
5976 const struct flow *flow,
5979 uint32_t probability;
5980 union user_action_cookie cookie;
5982 if (!ofproto->sflow || flow->in_port == OFPP_NONE) {
5986 probability = dpif_sflow_get_probability(ofproto->sflow);
5987 compose_sflow_cookie(ofproto, htons(0), odp_port,
5988 odp_port == OVSP_NONE ? 0 : 1, &cookie);
5990 return compose_sample_action(ofproto, odp_actions, flow, probability,
5991 &cookie, sizeof cookie.sflow);
5995 compose_flow_sample_cookie(uint16_t probability, uint32_t collector_set_id,
5996 uint32_t obs_domain_id, uint32_t obs_point_id,
5997 union user_action_cookie *cookie)
5999 cookie->type = USER_ACTION_COOKIE_FLOW_SAMPLE;
6000 cookie->flow_sample.probability = probability;
6001 cookie->flow_sample.collector_set_id = collector_set_id;
6002 cookie->flow_sample.obs_domain_id = obs_domain_id;
6003 cookie->flow_sample.obs_point_id = obs_point_id;
6007 compose_ipfix_cookie(union user_action_cookie *cookie)
6009 cookie->type = USER_ACTION_COOKIE_IPFIX;
6012 /* Compose SAMPLE action for IPFIX bridge sampling. */
6014 compose_ipfix_action(const struct ofproto_dpif *ofproto,
6015 struct ofpbuf *odp_actions,
6016 const struct flow *flow)
6018 uint32_t probability;
6019 union user_action_cookie cookie;
6021 if (!ofproto->ipfix || flow->in_port == OFPP_NONE) {
6025 probability = dpif_ipfix_get_bridge_exporter_probability(ofproto->ipfix);
6026 compose_ipfix_cookie(&cookie);
6028 compose_sample_action(ofproto, odp_actions, flow, probability,
6029 &cookie, sizeof cookie.ipfix);
6032 /* SAMPLE action for sFlow must be first action in any given list of
6033 * actions. At this point we do not have all information required to
6034 * build it. So try to build sample action as complete as possible. */
6036 add_sflow_action(struct action_xlate_ctx *ctx)
6038 ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto,
6040 &ctx->flow, OVSP_NONE);
6041 ctx->sflow_odp_port = 0;
6042 ctx->sflow_n_outputs = 0;
6045 /* SAMPLE action for IPFIX must be 1st or 2nd action in any given list
6046 * of actions, eventually after the SAMPLE action for sFlow. */
6048 add_ipfix_action(struct action_xlate_ctx *ctx)
6050 compose_ipfix_action(ctx->ofproto, ctx->odp_actions, &ctx->flow);
6053 /* Fix SAMPLE action according to data collected while composing ODP actions.
6054 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
6055 * USERSPACE action's user-cookie which is required for sflow. */
6057 fix_sflow_action(struct action_xlate_ctx *ctx)
6059 const struct flow *base = &ctx->base_flow;
6060 union user_action_cookie *cookie;
6062 if (!ctx->user_cookie_offset) {
6066 cookie = ofpbuf_at(ctx->odp_actions, ctx->user_cookie_offset,
6067 sizeof cookie->sflow);
6068 ovs_assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
6070 compose_sflow_cookie(ctx->ofproto, base->vlan_tci,
6071 ctx->sflow_odp_port, ctx->sflow_n_outputs, cookie);
6075 compose_output_action__(struct action_xlate_ctx *ctx, uint16_t ofp_port,
6078 const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
6079 ovs_be16 flow_vlan_tci;
6080 uint32_t flow_skb_mark;
6081 uint8_t flow_nw_tos;
6082 struct priority_to_dscp *pdscp;
6083 uint32_t out_port, odp_port;
6085 /* If 'struct flow' gets additional metadata, we'll need to zero it out
6086 * before traversing a patch port. */
6087 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 20);
6090 xlate_report(ctx, "Nonexistent output port");
6092 } else if (ofport->up.pp.config & OFPUTIL_PC_NO_FWD) {
6093 xlate_report(ctx, "OFPPC_NO_FWD set, skipping output");
6095 } else if (check_stp && !stp_forward_in_state(ofport->stp_state)) {
6096 xlate_report(ctx, "STP not in forwarding state, skipping output");
6100 if (netdev_vport_is_patch(ofport->up.netdev)) {
6101 struct ofport_dpif *peer = ofport_get_peer(ofport);
6102 struct flow old_flow = ctx->flow;
6103 const struct ofproto_dpif *peer_ofproto;
6104 enum slow_path_reason special;
6105 struct ofport_dpif *in_port;
6108 xlate_report(ctx, "Nonexistent patch port peer");
6112 peer_ofproto = ofproto_dpif_cast(peer->up.ofproto);
6113 if (peer_ofproto->backer != ctx->ofproto->backer) {
6114 xlate_report(ctx, "Patch port peer on a different datapath");
6118 ctx->ofproto = ofproto_dpif_cast(peer->up.ofproto);
6119 ctx->flow.in_port = peer->up.ofp_port;
6120 ctx->flow.metadata = htonll(0);
6121 memset(&ctx->flow.tunnel, 0, sizeof ctx->flow.tunnel);
6122 memset(ctx->flow.regs, 0, sizeof ctx->flow.regs);
6124 in_port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
6125 special = process_special(ctx->ofproto, &ctx->flow, in_port,
6128 ctx->slow |= special;
6129 } else if (!in_port || may_receive(in_port, ctx)) {
6130 if (!in_port || stp_forward_in_state(in_port->stp_state)) {
6131 xlate_table_action(ctx, ctx->flow.in_port, 0, true);
6133 /* Forwarding is disabled by STP. Let OFPP_NORMAL and the
6134 * learning action look at the packet, then drop it. */
6135 struct flow old_base_flow = ctx->base_flow;
6136 size_t old_size = ctx->odp_actions->size;
6137 xlate_table_action(ctx, ctx->flow.in_port, 0, true);
6138 ctx->base_flow = old_base_flow;
6139 ctx->odp_actions->size = old_size;
6143 ctx->flow = old_flow;
6144 ctx->ofproto = ofproto_dpif_cast(ofport->up.ofproto);
6146 if (ctx->resubmit_stats) {
6147 netdev_vport_inc_tx(ofport->up.netdev, ctx->resubmit_stats);
6148 netdev_vport_inc_rx(peer->up.netdev, ctx->resubmit_stats);
6154 flow_vlan_tci = ctx->flow.vlan_tci;
6155 flow_skb_mark = ctx->flow.skb_mark;
6156 flow_nw_tos = ctx->flow.nw_tos;
6158 pdscp = get_priority(ofport, ctx->flow.skb_priority);
6160 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
6161 ctx->flow.nw_tos |= pdscp->dscp;
6164 if (ofport->tnl_port) {
6165 /* Save tunnel metadata so that changes made due to
6166 * the Logical (tunnel) Port are not visible for any further
6167 * matches, while explicit set actions on tunnel metadata are.
6169 struct flow_tnl flow_tnl = ctx->flow.tunnel;
6170 odp_port = tnl_port_send(ofport->tnl_port, &ctx->flow);
6171 if (odp_port == OVSP_NONE) {
6172 xlate_report(ctx, "Tunneling decided against output");
6173 goto out; /* restore flow_nw_tos */
6175 if (ctx->flow.tunnel.ip_dst == ctx->orig_tunnel_ip_dst) {
6176 xlate_report(ctx, "Not tunneling to our own address");
6177 goto out; /* restore flow_nw_tos */
6179 if (ctx->resubmit_stats) {
6180 netdev_vport_inc_tx(ofport->up.netdev, ctx->resubmit_stats);
6182 out_port = odp_port;
6183 commit_odp_tunnel_action(&ctx->flow, &ctx->base_flow,
6185 ctx->flow.tunnel = flow_tnl; /* Restore tunnel metadata */
6187 odp_port = ofport->odp_port;
6188 out_port = vsp_realdev_to_vlandev(ctx->ofproto, odp_port,
6189 ctx->flow.vlan_tci);
6190 if (out_port != odp_port) {
6191 ctx->flow.vlan_tci = htons(0);
6193 ctx->flow.skb_mark &= ~IPSEC_MARK;
6195 commit_odp_actions(&ctx->flow, &ctx->base_flow, ctx->odp_actions);
6196 nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_OUTPUT, out_port);
6198 ctx->sflow_odp_port = odp_port;
6199 ctx->sflow_n_outputs++;
6200 ctx->nf_output_iface = ofp_port;
6203 ctx->flow.vlan_tci = flow_vlan_tci;
6204 ctx->flow.skb_mark = flow_skb_mark;
6206 ctx->flow.nw_tos = flow_nw_tos;
6210 compose_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port)
6212 compose_output_action__(ctx, ofp_port, true);
6216 tag_the_flow(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
6218 struct ofproto_dpif *ofproto = ctx->ofproto;
6219 uint8_t table_id = ctx->table_id;
6221 if (table_id > 0 && table_id < N_TABLES) {
6222 struct table_dpif *table = &ofproto->tables[table_id];
6223 if (table->other_table) {
6224 ctx->tags |= (rule && rule->tag
6226 : rule_calculate_tag(&ctx->flow,
6227 &table->other_table->mask,
6233 /* Common rule processing in one place to avoid duplicating code. */
6234 static struct rule_dpif *
6235 ctx_rule_hooks(struct action_xlate_ctx *ctx, struct rule_dpif *rule,
6238 if (ctx->resubmit_hook) {
6239 ctx->resubmit_hook(ctx, rule);
6241 if (rule == NULL && may_packet_in) {
6243 * check if table configuration flags
6244 * OFPTC_TABLE_MISS_CONTROLLER, default.
6245 * OFPTC_TABLE_MISS_CONTINUE,
6246 * OFPTC_TABLE_MISS_DROP
6247 * When OF1.0, OFPTC_TABLE_MISS_CONTINUE is used. What to do?
6249 rule = rule_dpif_miss_rule(ctx->ofproto, &ctx->flow);
6251 if (rule && ctx->resubmit_stats) {
6252 rule_credit_stats(rule, ctx->resubmit_stats);
6258 xlate_table_action(struct action_xlate_ctx *ctx,
6259 uint16_t in_port, uint8_t table_id, bool may_packet_in)
6261 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
6262 struct rule_dpif *rule;
6263 uint16_t old_in_port = ctx->flow.in_port;
6264 uint8_t old_table_id = ctx->table_id;
6266 ctx->table_id = table_id;
6268 /* Look up a flow with 'in_port' as the input port. */
6269 ctx->flow.in_port = in_port;
6270 rule = rule_dpif_lookup__(ctx->ofproto, &ctx->flow, table_id);
6272 tag_the_flow(ctx, rule);
6274 /* Restore the original input port. Otherwise OFPP_NORMAL and
6275 * OFPP_IN_PORT will have surprising behavior. */
6276 ctx->flow.in_port = old_in_port;
6278 rule = ctx_rule_hooks(ctx, rule, may_packet_in);
6281 struct rule_dpif *old_rule = ctx->rule;
6285 do_xlate_actions(rule->up.ofpacts, rule->up.ofpacts_len, ctx);
6286 ctx->rule = old_rule;
6290 ctx->table_id = old_table_id;
6292 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
6294 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
6295 MAX_RESUBMIT_RECURSION);
6296 ctx->max_resubmit_trigger = true;
6301 xlate_ofpact_resubmit(struct action_xlate_ctx *ctx,
6302 const struct ofpact_resubmit *resubmit)
6307 in_port = resubmit->in_port;
6308 if (in_port == OFPP_IN_PORT) {
6309 in_port = ctx->flow.in_port;
6312 table_id = resubmit->table_id;
6313 if (table_id == 255) {
6314 table_id = ctx->table_id;
6317 xlate_table_action(ctx, in_port, table_id, false);
6321 flood_packets(struct action_xlate_ctx *ctx, bool all)
6323 struct ofport_dpif *ofport;
6325 HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) {
6326 uint16_t ofp_port = ofport->up.ofp_port;
6328 if (ofp_port == ctx->flow.in_port) {
6333 compose_output_action__(ctx, ofp_port, false);
6334 } else if (!(ofport->up.pp.config & OFPUTIL_PC_NO_FLOOD)) {
6335 compose_output_action(ctx, ofp_port);
6339 ctx->nf_output_iface = NF_OUT_FLOOD;
6343 execute_controller_action(struct action_xlate_ctx *ctx, int len,
6344 enum ofp_packet_in_reason reason,
6345 uint16_t controller_id)
6347 struct ofputil_packet_in pin;
6348 struct ofpbuf *packet;
6350 ctx->slow |= SLOW_CONTROLLER;
6355 packet = ofpbuf_clone(ctx->packet);
6357 if (packet->l2 && packet->l3) {
6358 struct eth_header *eh;
6359 uint16_t mpls_depth;
6361 eth_pop_vlan(packet);
6364 memcpy(eh->eth_src, ctx->flow.dl_src, sizeof eh->eth_src);
6365 memcpy(eh->eth_dst, ctx->flow.dl_dst, sizeof eh->eth_dst);
6367 if (ctx->flow.vlan_tci & htons(VLAN_CFI)) {
6368 eth_push_vlan(packet, ctx->flow.vlan_tci);
6371 mpls_depth = eth_mpls_depth(packet);
6373 if (mpls_depth < ctx->flow.mpls_depth) {
6374 push_mpls(packet, ctx->flow.dl_type, ctx->flow.mpls_lse);
6375 } else if (mpls_depth > ctx->flow.mpls_depth) {
6376 pop_mpls(packet, ctx->flow.dl_type);
6377 } else if (mpls_depth) {
6378 set_mpls_lse(packet, ctx->flow.mpls_lse);
6382 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
6383 packet_set_ipv4(packet, ctx->flow.nw_src, ctx->flow.nw_dst,
6384 ctx->flow.nw_tos, ctx->flow.nw_ttl);
6388 if (ctx->flow.nw_proto == IPPROTO_TCP) {
6389 packet_set_tcp_port(packet, ctx->flow.tp_src,
6391 } else if (ctx->flow.nw_proto == IPPROTO_UDP) {
6392 packet_set_udp_port(packet, ctx->flow.tp_src,
6399 pin.packet = packet->data;
6400 pin.packet_len = packet->size;
6401 pin.reason = reason;
6402 pin.controller_id = controller_id;
6403 pin.table_id = ctx->table_id;
6404 pin.cookie = ctx->rule ? ctx->rule->up.flow_cookie : 0;
6407 flow_get_metadata(&ctx->flow, &pin.fmd);
6409 connmgr_send_packet_in(ctx->ofproto->up.connmgr, &pin);
6410 ofpbuf_delete(packet);
6414 execute_mpls_push_action(struct action_xlate_ctx *ctx, ovs_be16 eth_type)
6416 ovs_assert(eth_type_mpls(eth_type));
6418 if (ctx->base_flow.mpls_depth) {
6419 ctx->flow.mpls_lse &= ~htonl(MPLS_BOS_MASK);
6420 ctx->flow.mpls_depth++;
6425 if (ctx->flow.dl_type == htons(ETH_TYPE_IPV6)) {
6426 label = htonl(0x2); /* IPV6 Explicit Null. */
6428 label = htonl(0x0); /* IPV4 Explicit Null. */
6430 tc = (ctx->flow.nw_tos & IP_DSCP_MASK) >> 2;
6431 ttl = ctx->flow.nw_ttl ? ctx->flow.nw_ttl : 0x40;
6432 ctx->flow.mpls_lse = set_mpls_lse_values(ttl, tc, 1, label);
6433 ctx->flow.mpls_depth = 1;
6435 ctx->flow.dl_type = eth_type;
6439 execute_mpls_pop_action(struct action_xlate_ctx *ctx, ovs_be16 eth_type)
6441 ovs_assert(eth_type_mpls(ctx->flow.dl_type));
6442 ovs_assert(!eth_type_mpls(eth_type));
6444 if (ctx->flow.mpls_depth) {
6445 ctx->flow.mpls_depth--;
6446 ctx->flow.mpls_lse = htonl(0);
6447 if (!ctx->flow.mpls_depth) {
6448 ctx->flow.dl_type = eth_type;
6454 compose_dec_ttl(struct action_xlate_ctx *ctx, struct ofpact_cnt_ids *ids)
6456 if (ctx->flow.dl_type != htons(ETH_TYPE_IP) &&
6457 ctx->flow.dl_type != htons(ETH_TYPE_IPV6)) {
6461 if (ctx->flow.nw_ttl > 1) {
6467 for (i = 0; i < ids->n_controllers; i++) {
6468 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL,
6472 /* Stop processing for current table. */
6478 execute_set_mpls_ttl_action(struct action_xlate_ctx *ctx, uint8_t ttl)
6480 if (!eth_type_mpls(ctx->flow.dl_type)) {
6484 set_mpls_lse_ttl(&ctx->flow.mpls_lse, ttl);
6489 execute_dec_mpls_ttl_action(struct action_xlate_ctx *ctx)
6491 uint8_t ttl = mpls_lse_to_ttl(ctx->flow.mpls_lse);
6493 if (!eth_type_mpls(ctx->flow.dl_type)) {
6499 set_mpls_lse_ttl(&ctx->flow.mpls_lse, ttl);
6502 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL, 0);
6504 /* Stop processing for current table. */
6510 xlate_output_action(struct action_xlate_ctx *ctx,
6511 uint16_t port, uint16_t max_len, bool may_packet_in)
6513 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
6515 ctx->nf_output_iface = NF_OUT_DROP;
6519 compose_output_action(ctx, ctx->flow.in_port);
6522 xlate_table_action(ctx, ctx->flow.in_port, 0, may_packet_in);
6528 flood_packets(ctx, false);
6531 flood_packets(ctx, true);
6533 case OFPP_CONTROLLER:
6534 execute_controller_action(ctx, max_len, OFPR_ACTION, 0);
6540 if (port != ctx->flow.in_port) {
6541 compose_output_action(ctx, port);
6543 xlate_report(ctx, "skipping output to input port");
6548 if (prev_nf_output_iface == NF_OUT_FLOOD) {
6549 ctx->nf_output_iface = NF_OUT_FLOOD;
6550 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
6551 ctx->nf_output_iface = prev_nf_output_iface;
6552 } else if (prev_nf_output_iface != NF_OUT_DROP &&
6553 ctx->nf_output_iface != NF_OUT_FLOOD) {
6554 ctx->nf_output_iface = NF_OUT_MULTI;
6559 xlate_output_reg_action(struct action_xlate_ctx *ctx,
6560 const struct ofpact_output_reg *or)
6562 uint64_t port = mf_get_subfield(&or->src, &ctx->flow);
6563 if (port <= UINT16_MAX) {
6564 xlate_output_action(ctx, port, or->max_len, false);
6569 xlate_enqueue_action(struct action_xlate_ctx *ctx,
6570 const struct ofpact_enqueue *enqueue)
6572 uint16_t ofp_port = enqueue->port;
6573 uint32_t queue_id = enqueue->queue;
6574 uint32_t flow_priority, priority;
6577 /* Translate queue to priority. */
6578 error = dpif_queue_to_priority(ctx->ofproto->backer->dpif,
6579 queue_id, &priority);
6581 /* Fall back to ordinary output action. */
6582 xlate_output_action(ctx, enqueue->port, 0, false);
6586 /* Check output port. */
6587 if (ofp_port == OFPP_IN_PORT) {
6588 ofp_port = ctx->flow.in_port;
6589 } else if (ofp_port == ctx->flow.in_port) {
6593 /* Add datapath actions. */
6594 flow_priority = ctx->flow.skb_priority;
6595 ctx->flow.skb_priority = priority;
6596 compose_output_action(ctx, ofp_port);
6597 ctx->flow.skb_priority = flow_priority;
6599 /* Update NetFlow output port. */
6600 if (ctx->nf_output_iface == NF_OUT_DROP) {
6601 ctx->nf_output_iface = ofp_port;
6602 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
6603 ctx->nf_output_iface = NF_OUT_MULTI;
6608 xlate_set_queue_action(struct action_xlate_ctx *ctx, uint32_t queue_id)
6610 uint32_t skb_priority;
6612 if (!dpif_queue_to_priority(ctx->ofproto->backer->dpif,
6613 queue_id, &skb_priority)) {
6614 ctx->flow.skb_priority = skb_priority;
6616 /* Couldn't translate queue to a priority. Nothing to do. A warning
6617 * has already been logged. */
6622 slave_enabled_cb(uint16_t ofp_port, void *ofproto_)
6624 struct ofproto_dpif *ofproto = ofproto_;
6625 struct ofport_dpif *port;
6635 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
6638 port = get_ofp_port(ofproto, ofp_port);
6639 return port ? port->may_enable : false;
6644 xlate_bundle_action(struct action_xlate_ctx *ctx,
6645 const struct ofpact_bundle *bundle)
6649 port = bundle_execute(bundle, &ctx->flow, slave_enabled_cb, ctx->ofproto);
6650 if (bundle->dst.field) {
6651 nxm_reg_load(&bundle->dst, port, &ctx->flow);
6653 xlate_output_action(ctx, port, 0, false);
6658 xlate_learn_action(struct action_xlate_ctx *ctx,
6659 const struct ofpact_learn *learn)
6661 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
6662 struct ofputil_flow_mod fm;
6663 uint64_t ofpacts_stub[1024 / 8];
6664 struct ofpbuf ofpacts;
6667 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
6668 learn_execute(learn, &ctx->flow, &fm, &ofpacts);
6670 error = ofproto_flow_mod(&ctx->ofproto->up, &fm);
6671 if (error && !VLOG_DROP_WARN(&rl)) {
6672 VLOG_WARN("learning action failed to modify flow table (%s)",
6673 ofperr_get_name(error));
6676 ofpbuf_uninit(&ofpacts);
6679 /* Reduces '*timeout' to no more than 'max'. A value of zero in either case
6680 * means "infinite". */
6682 reduce_timeout(uint16_t max, uint16_t *timeout)
6684 if (max && (!*timeout || *timeout > max)) {
6690 xlate_fin_timeout(struct action_xlate_ctx *ctx,
6691 const struct ofpact_fin_timeout *oft)
6693 if (ctx->tcp_flags & (TCP_FIN | TCP_RST) && ctx->rule) {
6694 struct rule_dpif *rule = ctx->rule;
6696 reduce_timeout(oft->fin_idle_timeout, &rule->up.idle_timeout);
6697 reduce_timeout(oft->fin_hard_timeout, &rule->up.hard_timeout);
6702 xlate_sample_action(struct action_xlate_ctx *ctx,
6703 const struct ofpact_sample *os)
6705 union user_action_cookie cookie;
6706 /* Scale the probability from 16-bit to 32-bit while representing
6707 * the same percentage. */
6708 uint32_t probability = (os->probability << 16) | os->probability;
6710 commit_odp_actions(&ctx->flow, &ctx->base_flow, ctx->odp_actions);
6712 compose_flow_sample_cookie(os->probability, os->collector_set_id,
6713 os->obs_domain_id, os->obs_point_id, &cookie);
6714 compose_sample_action(ctx->ofproto, ctx->odp_actions, &ctx->flow,
6715 probability, &cookie, sizeof cookie.flow_sample);
6719 may_receive(const struct ofport_dpif *port, struct action_xlate_ctx *ctx)
6721 if (port->up.pp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
6722 ? OFPUTIL_PC_NO_RECV_STP
6723 : OFPUTIL_PC_NO_RECV)) {
6727 /* Only drop packets here if both forwarding and learning are
6728 * disabled. If just learning is enabled, we need to have
6729 * OFPP_NORMAL and the learning action have a look at the packet
6730 * before we can drop it. */
6731 if (!stp_forward_in_state(port->stp_state)
6732 && !stp_learn_in_state(port->stp_state)) {
6740 tunnel_ecn_ok(struct action_xlate_ctx *ctx)
6742 if (is_ip_any(&ctx->base_flow)
6743 && (ctx->flow.tunnel.ip_tos & IP_ECN_MASK) == IP_ECN_CE) {
6744 if ((ctx->base_flow.nw_tos & IP_ECN_MASK) == IP_ECN_NOT_ECT) {
6745 VLOG_WARN_RL(&rl, "dropping tunnel packet marked ECN CE"
6746 " but is not ECN capable");
6749 /* Set the ECN CE value in the tunneled packet. */
6750 ctx->flow.nw_tos |= IP_ECN_CE;
6758 do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len,
6759 struct action_xlate_ctx *ctx)
6761 bool was_evictable = true;
6762 const struct ofpact *a;
6765 /* Don't let the rule we're working on get evicted underneath us. */
6766 was_evictable = ctx->rule->up.evictable;
6767 ctx->rule->up.evictable = false;
6770 do_xlate_actions_again:
6771 OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) {
6772 struct ofpact_controller *controller;
6773 const struct ofpact_metadata *metadata;
6781 xlate_output_action(ctx, ofpact_get_OUTPUT(a)->port,
6782 ofpact_get_OUTPUT(a)->max_len, true);
6785 case OFPACT_CONTROLLER:
6786 controller = ofpact_get_CONTROLLER(a);
6787 execute_controller_action(ctx, controller->max_len,
6789 controller->controller_id);
6792 case OFPACT_ENQUEUE:
6793 xlate_enqueue_action(ctx, ofpact_get_ENQUEUE(a));
6796 case OFPACT_SET_VLAN_VID:
6797 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
6798 ctx->flow.vlan_tci |= (htons(ofpact_get_SET_VLAN_VID(a)->vlan_vid)
6802 case OFPACT_SET_VLAN_PCP:
6803 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
6804 ctx->flow.vlan_tci |= htons((ofpact_get_SET_VLAN_PCP(a)->vlan_pcp
6809 case OFPACT_STRIP_VLAN:
6810 ctx->flow.vlan_tci = htons(0);
6813 case OFPACT_PUSH_VLAN:
6814 /* XXX 802.1AD(QinQ) */
6815 ctx->flow.vlan_tci = htons(VLAN_CFI);
6818 case OFPACT_SET_ETH_SRC:
6819 memcpy(ctx->flow.dl_src, ofpact_get_SET_ETH_SRC(a)->mac,
6823 case OFPACT_SET_ETH_DST:
6824 memcpy(ctx->flow.dl_dst, ofpact_get_SET_ETH_DST(a)->mac,
6828 case OFPACT_SET_IPV4_SRC:
6829 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
6830 ctx->flow.nw_src = ofpact_get_SET_IPV4_SRC(a)->ipv4;
6834 case OFPACT_SET_IPV4_DST:
6835 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
6836 ctx->flow.nw_dst = ofpact_get_SET_IPV4_DST(a)->ipv4;
6840 case OFPACT_SET_IPV4_DSCP:
6841 /* OpenFlow 1.0 only supports IPv4. */
6842 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
6843 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
6844 ctx->flow.nw_tos |= ofpact_get_SET_IPV4_DSCP(a)->dscp;
6848 case OFPACT_SET_L4_SRC_PORT:
6849 if (is_ip_any(&ctx->flow)) {
6850 ctx->flow.tp_src = htons(ofpact_get_SET_L4_SRC_PORT(a)->port);
6854 case OFPACT_SET_L4_DST_PORT:
6855 if (is_ip_any(&ctx->flow)) {
6856 ctx->flow.tp_dst = htons(ofpact_get_SET_L4_DST_PORT(a)->port);
6860 case OFPACT_RESUBMIT:
6861 xlate_ofpact_resubmit(ctx, ofpact_get_RESUBMIT(a));
6864 case OFPACT_SET_TUNNEL:
6865 ctx->flow.tunnel.tun_id = htonll(ofpact_get_SET_TUNNEL(a)->tun_id);
6868 case OFPACT_SET_QUEUE:
6869 xlate_set_queue_action(ctx, ofpact_get_SET_QUEUE(a)->queue_id);
6872 case OFPACT_POP_QUEUE:
6873 ctx->flow.skb_priority = ctx->orig_skb_priority;
6876 case OFPACT_REG_MOVE:
6877 nxm_execute_reg_move(ofpact_get_REG_MOVE(a), &ctx->flow);
6880 case OFPACT_REG_LOAD:
6881 nxm_execute_reg_load(ofpact_get_REG_LOAD(a), &ctx->flow);
6884 case OFPACT_STACK_PUSH:
6885 nxm_execute_stack_push(ofpact_get_STACK_PUSH(a), &ctx->flow,
6889 case OFPACT_STACK_POP:
6890 nxm_execute_stack_pop(ofpact_get_STACK_POP(a), &ctx->flow,
6894 case OFPACT_PUSH_MPLS:
6895 execute_mpls_push_action(ctx, ofpact_get_PUSH_MPLS(a)->ethertype);
6898 case OFPACT_POP_MPLS:
6899 execute_mpls_pop_action(ctx, ofpact_get_POP_MPLS(a)->ethertype);
6902 case OFPACT_SET_MPLS_TTL:
6903 if (execute_set_mpls_ttl_action(ctx, ofpact_get_SET_MPLS_TTL(a)->ttl)) {
6908 case OFPACT_DEC_MPLS_TTL:
6909 if (execute_dec_mpls_ttl_action(ctx)) {
6914 case OFPACT_DEC_TTL:
6915 if (compose_dec_ttl(ctx, ofpact_get_DEC_TTL(a))) {
6921 /* Nothing to do. */
6924 case OFPACT_MULTIPATH:
6925 multipath_execute(ofpact_get_MULTIPATH(a), &ctx->flow);
6929 ctx->ofproto->has_bundle_action = true;
6930 xlate_bundle_action(ctx, ofpact_get_BUNDLE(a));
6933 case OFPACT_OUTPUT_REG:
6934 xlate_output_reg_action(ctx, ofpact_get_OUTPUT_REG(a));
6938 ctx->has_learn = true;
6939 if (ctx->may_learn) {
6940 xlate_learn_action(ctx, ofpact_get_LEARN(a));
6948 case OFPACT_FIN_TIMEOUT:
6949 ctx->has_fin_timeout = true;
6950 xlate_fin_timeout(ctx, ofpact_get_FIN_TIMEOUT(a));
6953 case OFPACT_CLEAR_ACTIONS:
6955 * Nothing to do because writa-actions is not supported for now.
6956 * When writa-actions is supported, clear-actions also must
6957 * be supported at the same time.
6961 case OFPACT_WRITE_METADATA:
6962 metadata = ofpact_get_WRITE_METADATA(a);
6963 ctx->flow.metadata &= ~metadata->mask;
6964 ctx->flow.metadata |= metadata->metadata & metadata->mask;
6967 case OFPACT_GOTO_TABLE: {
6968 /* It is assumed that goto-table is the last action. */
6969 struct ofpact_goto_table *ogt = ofpact_get_GOTO_TABLE(a);
6970 struct rule_dpif *rule;
6972 ovs_assert(ctx->table_id < ogt->table_id);
6974 ctx->table_id = ogt->table_id;
6976 /* Look up a flow from the new table. */
6977 rule = rule_dpif_lookup__(ctx->ofproto, &ctx->flow, ctx->table_id);
6979 tag_the_flow(ctx, rule);
6981 rule = ctx_rule_hooks(ctx, rule, true);
6985 ctx->rule->up.evictable = was_evictable;
6988 was_evictable = rule->up.evictable;
6989 rule->up.evictable = false;
6991 /* Tail recursion removal. */
6992 ofpacts = rule->up.ofpacts;
6993 ofpacts_len = rule->up.ofpacts_len;
6994 goto do_xlate_actions_again;
7000 xlate_sample_action(ctx, ofpact_get_SAMPLE(a));
7007 ctx->rule->up.evictable = was_evictable;
7012 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
7013 struct ofproto_dpif *ofproto, const struct flow *flow,
7014 const struct initial_vals *initial_vals,
7015 struct rule_dpif *rule,
7016 uint8_t tcp_flags, const struct ofpbuf *packet)
7018 /* Flow initialization rules:
7019 * - 'base_flow' must match the kernel's view of the packet at the
7020 * time that action processing starts. 'flow' represents any
7021 * transformations we wish to make through actions.
7022 * - By default 'base_flow' and 'flow' are the same since the input
7023 * packet matches the output before any actions are applied.
7024 * - When using VLAN splinters, 'base_flow''s VLAN is set to the value
7025 * of the received packet as seen by the kernel. If we later output
7026 * to another device without any modifications this will cause us to
7027 * insert a new tag since the original one was stripped off by the
7029 * - Tunnel metadata as received is retained in 'flow'. This allows
7030 * tunnel metadata matching also in later tables.
7031 * Since a kernel action for setting the tunnel metadata will only be
7032 * generated with actual tunnel output, changing the tunnel metadata
7033 * values in 'flow' (such as tun_id) will only have effect with a later
7034 * tunnel output action.
7035 * - Tunnel 'base_flow' is completely cleared since that is what the
7036 * kernel does. If we wish to maintain the original values an action
7037 * needs to be generated. */
7039 ctx->ofproto = ofproto;
7041 ctx->base_flow = ctx->flow;
7042 memset(&ctx->base_flow.tunnel, 0, sizeof ctx->base_flow.tunnel);
7043 ctx->orig_tunnel_ip_dst = flow->tunnel.ip_dst;
7045 ctx->packet = packet;
7046 ctx->may_learn = packet != NULL;
7047 ctx->tcp_flags = tcp_flags;
7048 ctx->resubmit_hook = NULL;
7049 ctx->report_hook = NULL;
7050 ctx->resubmit_stats = NULL;
7053 ctx->base_flow.vlan_tci = initial_vals->vlan_tci;
7057 /* Translates the 'ofpacts_len' bytes of "struct ofpacts" starting at 'ofpacts'
7058 * into datapath actions in 'odp_actions', using 'ctx'. */
7060 xlate_actions(struct action_xlate_ctx *ctx,
7061 const struct ofpact *ofpacts, size_t ofpacts_len,
7062 struct ofpbuf *odp_actions)
7064 /* Normally false. Set to true if we ever hit MAX_RESUBMIT_RECURSION, so
7065 * that in the future we always keep a copy of the original flow for
7066 * tracing purposes. */
7067 static bool hit_resubmit_limit;
7069 enum slow_path_reason special;
7070 struct ofport_dpif *in_port;
7071 struct flow orig_flow;
7073 COVERAGE_INC(ofproto_dpif_xlate);
7075 ofpbuf_clear(odp_actions);
7076 ofpbuf_reserve(odp_actions, NL_A_U32_SIZE);
7078 ctx->odp_actions = odp_actions;
7081 ctx->has_learn = false;
7082 ctx->has_normal = false;
7083 ctx->has_fin_timeout = false;
7084 ctx->nf_output_iface = NF_OUT_DROP;
7087 ctx->max_resubmit_trigger = false;
7088 ctx->orig_skb_priority = ctx->flow.skb_priority;
7092 ofpbuf_use_stub(&ctx->stack, ctx->init_stack, sizeof ctx->init_stack);
7094 if (ctx->ofproto->has_mirrors || hit_resubmit_limit) {
7095 /* Do this conditionally because the copy is expensive enough that it
7096 * shows up in profiles. */
7097 orig_flow = ctx->flow;
7100 if (ctx->flow.nw_frag & FLOW_NW_FRAG_ANY) {
7101 switch (ctx->ofproto->up.frag_handling) {
7102 case OFPC_FRAG_NORMAL:
7103 /* We must pretend that transport ports are unavailable. */
7104 ctx->flow.tp_src = ctx->base_flow.tp_src = htons(0);
7105 ctx->flow.tp_dst = ctx->base_flow.tp_dst = htons(0);
7108 case OFPC_FRAG_DROP:
7111 case OFPC_FRAG_REASM:
7114 case OFPC_FRAG_NX_MATCH:
7115 /* Nothing to do. */
7118 case OFPC_INVALID_TTL_TO_CONTROLLER:
7123 in_port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
7124 special = process_special(ctx->ofproto, &ctx->flow, in_port, ctx->packet);
7126 ctx->slow |= special;
7128 static struct vlog_rate_limit trace_rl = VLOG_RATE_LIMIT_INIT(1, 1);
7129 struct initial_vals initial_vals;
7130 size_t sample_actions_len;
7131 uint32_t local_odp_port;
7133 initial_vals.vlan_tci = ctx->base_flow.vlan_tci;
7135 add_sflow_action(ctx);
7136 add_ipfix_action(ctx);
7137 sample_actions_len = ctx->odp_actions->size;
7139 if (tunnel_ecn_ok(ctx) && (!in_port || may_receive(in_port, ctx))) {
7140 do_xlate_actions(ofpacts, ofpacts_len, ctx);
7142 /* We've let OFPP_NORMAL and the learning action look at the
7143 * packet, so drop it now if forwarding is disabled. */
7144 if (in_port && !stp_forward_in_state(in_port->stp_state)) {
7145 ctx->odp_actions->size = sample_actions_len;
7149 if (ctx->max_resubmit_trigger && !ctx->resubmit_hook) {
7150 if (!hit_resubmit_limit) {
7151 /* We didn't record the original flow. Make sure we do from
7153 hit_resubmit_limit = true;
7154 } else if (!VLOG_DROP_ERR(&trace_rl)) {
7155 struct ds ds = DS_EMPTY_INITIALIZER;
7157 ofproto_trace(ctx->ofproto, &orig_flow, ctx->packet,
7158 &initial_vals, &ds);
7159 VLOG_ERR("Trace triggered by excessive resubmit "
7160 "recursion:\n%s", ds_cstr(&ds));
7165 local_odp_port = ofp_port_to_odp_port(ctx->ofproto, OFPP_LOCAL);
7166 if (!connmgr_may_set_up_flow(ctx->ofproto->up.connmgr, &ctx->flow,
7168 ctx->odp_actions->data,
7169 ctx->odp_actions->size)) {
7170 ctx->slow |= SLOW_IN_BAND;
7172 && connmgr_msg_in_hook(ctx->ofproto->up.connmgr, &ctx->flow,
7174 compose_output_action(ctx, OFPP_LOCAL);
7177 if (ctx->ofproto->has_mirrors) {
7178 add_mirror_actions(ctx, &orig_flow);
7180 fix_sflow_action(ctx);
7183 ofpbuf_uninit(&ctx->stack);
7186 /* Translates the 'ofpacts_len' bytes of "struct ofpact"s starting at 'ofpacts'
7187 * into datapath actions, using 'ctx', and discards the datapath actions. */
7189 xlate_actions_for_side_effects(struct action_xlate_ctx *ctx,
7190 const struct ofpact *ofpacts,
7193 uint64_t odp_actions_stub[1024 / 8];
7194 struct ofpbuf odp_actions;
7196 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
7197 xlate_actions(ctx, ofpacts, ofpacts_len, &odp_actions);
7198 ofpbuf_uninit(&odp_actions);
7202 xlate_report(struct action_xlate_ctx *ctx, const char *s)
7204 if (ctx->report_hook) {
7205 ctx->report_hook(ctx, s);
7209 /* OFPP_NORMAL implementation. */
7211 static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *);
7213 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
7214 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle',
7215 * the bundle on which the packet was received, returns the VLAN to which the
7218 * Both 'vid' and the return value are in the range 0...4095. */
7220 input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid)
7222 switch (in_bundle->vlan_mode) {
7223 case PORT_VLAN_ACCESS:
7224 return in_bundle->vlan;
7227 case PORT_VLAN_TRUNK:
7230 case PORT_VLAN_NATIVE_UNTAGGED:
7231 case PORT_VLAN_NATIVE_TAGGED:
7232 return vid ? vid : in_bundle->vlan;
7239 /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'.
7240 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
7243 * 'vid' should be the VID obtained from the 802.1Q header that was received as
7244 * part of a packet (specify 0 if there was no 802.1Q header), in the range
7247 input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn)
7249 /* Allow any VID on the OFPP_NONE port. */
7250 if (in_bundle == &ofpp_none_bundle) {
7254 switch (in_bundle->vlan_mode) {
7255 case PORT_VLAN_ACCESS:
7258 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7259 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
7260 "packet received on port %s configured as VLAN "
7261 "%"PRIu16" access port",
7262 in_bundle->ofproto->up.name, vid,
7263 in_bundle->name, in_bundle->vlan);
7269 case PORT_VLAN_NATIVE_UNTAGGED:
7270 case PORT_VLAN_NATIVE_TAGGED:
7272 /* Port must always carry its native VLAN. */
7276 case PORT_VLAN_TRUNK:
7277 if (!ofbundle_includes_vlan(in_bundle, vid)) {
7279 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7280 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet "
7281 "received on port %s not configured for trunking "
7283 in_bundle->ofproto->up.name, vid,
7284 in_bundle->name, vid);
7296 /* Given 'vlan', the VLAN that a packet belongs to, and
7297 * 'out_bundle', a bundle on which the packet is to be output, returns the VID
7298 * that should be included in the 802.1Q header. (If the return value is 0,
7299 * then the 802.1Q header should only be included in the packet if there is a
7302 * Both 'vlan' and the return value are in the range 0...4095. */
7304 output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan)
7306 switch (out_bundle->vlan_mode) {
7307 case PORT_VLAN_ACCESS:
7310 case PORT_VLAN_TRUNK:
7311 case PORT_VLAN_NATIVE_TAGGED:
7314 case PORT_VLAN_NATIVE_UNTAGGED:
7315 return vlan == out_bundle->vlan ? 0 : vlan;
7323 output_normal(struct action_xlate_ctx *ctx, const struct ofbundle *out_bundle,
7326 struct ofport_dpif *port;
7328 ovs_be16 tci, old_tci;
7330 vid = output_vlan_to_vid(out_bundle, vlan);
7331 if (!out_bundle->bond) {
7332 port = ofbundle_get_a_port(out_bundle);
7334 port = bond_choose_output_slave(out_bundle->bond, &ctx->flow,
7337 /* No slaves enabled, so drop packet. */
7342 old_tci = ctx->flow.vlan_tci;
7344 if (tci || out_bundle->use_priority_tags) {
7345 tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK);
7347 tci |= htons(VLAN_CFI);
7350 ctx->flow.vlan_tci = tci;
7352 compose_output_action(ctx, port->up.ofp_port);
7353 ctx->flow.vlan_tci = old_tci;
7357 mirror_mask_ffs(mirror_mask_t mask)
7359 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
7364 ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
7366 return (bundle->vlan_mode != PORT_VLAN_ACCESS
7367 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
7371 ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan)
7373 return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan);
7376 /* Returns an arbitrary interface within 'bundle'. */
7377 static struct ofport_dpif *
7378 ofbundle_get_a_port(const struct ofbundle *bundle)
7380 return CONTAINER_OF(list_front(&bundle->ports),
7381 struct ofport_dpif, bundle_node);
7385 vlan_is_mirrored(const struct ofmirror *m, int vlan)
7387 return !m->vlans || bitmap_is_set(m->vlans, vlan);
7391 add_mirror_actions(struct action_xlate_ctx *ctx, const struct flow *orig_flow)
7393 struct ofproto_dpif *ofproto = ctx->ofproto;
7394 mirror_mask_t mirrors;
7395 struct ofbundle *in_bundle;
7398 const struct nlattr *a;
7401 in_bundle = lookup_input_bundle(ctx->ofproto, orig_flow->in_port,
7402 ctx->packet != NULL, NULL);
7406 mirrors = in_bundle->src_mirrors;
7408 /* Drop frames on bundles reserved for mirroring. */
7409 if (in_bundle->mirror_out) {
7410 if (ctx->packet != NULL) {
7411 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7412 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
7413 "%s, which is reserved exclusively for mirroring",
7414 ctx->ofproto->up.name, in_bundle->name);
7420 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
7421 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
7424 vlan = input_vid_to_vlan(in_bundle, vid);
7426 /* Look at the output ports to check for destination selections. */
7428 NL_ATTR_FOR_EACH (a, left, ctx->odp_actions->data,
7429 ctx->odp_actions->size) {
7430 enum ovs_action_attr type = nl_attr_type(a);
7431 struct ofport_dpif *ofport;
7433 if (type != OVS_ACTION_ATTR_OUTPUT) {
7437 ofport = get_odp_port(ofproto, nl_attr_get_u32(a));
7438 if (ofport && ofport->bundle) {
7439 mirrors |= ofport->bundle->dst_mirrors;
7447 /* Restore the original packet before adding the mirror actions. */
7448 ctx->flow = *orig_flow;
7453 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
7455 if (!vlan_is_mirrored(m, vlan)) {
7456 mirrors = zero_rightmost_1bit(mirrors);
7460 mirrors &= ~m->dup_mirrors;
7461 ctx->mirrors |= m->dup_mirrors;
7463 output_normal(ctx, m->out, vlan);
7464 } else if (vlan != m->out_vlan
7465 && !eth_addr_is_reserved(orig_flow->dl_dst)) {
7466 struct ofbundle *bundle;
7468 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
7469 if (ofbundle_includes_vlan(bundle, m->out_vlan)
7470 && !bundle->mirror_out) {
7471 output_normal(ctx, bundle, m->out_vlan);
7479 update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors,
7480 uint64_t packets, uint64_t bytes)
7486 for (; mirrors; mirrors = zero_rightmost_1bit(mirrors)) {
7489 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
7492 /* In normal circumstances 'm' will not be NULL. However,
7493 * if mirrors are reconfigured, we can temporarily get out
7494 * of sync in facet_revalidate(). We could "correct" the
7495 * mirror list before reaching here, but doing that would
7496 * not properly account the traffic stats we've currently
7497 * accumulated for previous mirror configuration. */
7501 m->packet_count += packets;
7502 m->byte_count += bytes;
7506 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
7507 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
7508 * indicate this; newer upstream kernels use gratuitous ARP requests. */
7510 is_gratuitous_arp(const struct flow *flow)
7512 return (flow->dl_type == htons(ETH_TYPE_ARP)
7513 && eth_addr_is_broadcast(flow->dl_dst)
7514 && (flow->nw_proto == ARP_OP_REPLY
7515 || (flow->nw_proto == ARP_OP_REQUEST
7516 && flow->nw_src == flow->nw_dst)));
7520 update_learning_table(struct ofproto_dpif *ofproto,
7521 const struct flow *flow, int vlan,
7522 struct ofbundle *in_bundle)
7524 struct mac_entry *mac;
7526 /* Don't learn the OFPP_NONE port. */
7527 if (in_bundle == &ofpp_none_bundle) {
7531 if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
7535 mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan);
7536 if (is_gratuitous_arp(flow)) {
7537 /* We don't want to learn from gratuitous ARP packets that are
7538 * reflected back over bond slaves so we lock the learning table. */
7539 if (!in_bundle->bond) {
7540 mac_entry_set_grat_arp_lock(mac);
7541 } else if (mac_entry_is_grat_arp_locked(mac)) {
7546 if (mac_entry_is_new(mac) || mac->port.p != in_bundle) {
7547 /* The log messages here could actually be useful in debugging,
7548 * so keep the rate limit relatively high. */
7549 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
7550 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
7551 "on port %s in VLAN %d",
7552 ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src),
7553 in_bundle->name, vlan);
7555 mac->port.p = in_bundle;
7556 tag_set_add(&ofproto->backer->revalidate_set,
7557 mac_learning_changed(ofproto->ml, mac));
7561 static struct ofbundle *
7562 lookup_input_bundle(const struct ofproto_dpif *ofproto, uint16_t in_port,
7563 bool warn, struct ofport_dpif **in_ofportp)
7565 struct ofport_dpif *ofport;
7567 /* Find the port and bundle for the received packet. */
7568 ofport = get_ofp_port(ofproto, in_port);
7570 *in_ofportp = ofport;
7572 if (ofport && ofport->bundle) {
7573 return ofport->bundle;
7576 /* Special-case OFPP_NONE, which a controller may use as the ingress
7577 * port for traffic that it is sourcing. */
7578 if (in_port == OFPP_NONE) {
7579 return &ofpp_none_bundle;
7582 /* Odd. A few possible reasons here:
7584 * - We deleted a port but there are still a few packets queued up
7587 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
7588 * we don't know about.
7590 * - The ofproto client didn't configure the port as part of a bundle.
7591 * This is particularly likely to happen if a packet was received on the
7592 * port after it was created, but before the client had a chance to
7593 * configure its bundle.
7596 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7598 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
7599 "port %"PRIu16, ofproto->up.name, in_port);
7604 /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or
7605 * dropped. Returns true if they may be forwarded, false if they should be
7608 * 'in_port' must be the ofport_dpif that corresponds to flow->in_port.
7609 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
7611 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
7612 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
7613 * checked by input_vid_is_valid().
7615 * May also add tags to '*tags', although the current implementation only does
7616 * so in one special case.
7619 is_admissible(struct action_xlate_ctx *ctx, struct ofport_dpif *in_port,
7622 struct ofproto_dpif *ofproto = ctx->ofproto;
7623 struct flow *flow = &ctx->flow;
7624 struct ofbundle *in_bundle = in_port->bundle;
7626 /* Drop frames for reserved multicast addresses
7627 * only if forward_bpdu option is absent. */
7628 if (!ofproto->up.forward_bpdu && eth_addr_is_reserved(flow->dl_dst)) {
7629 xlate_report(ctx, "packet has reserved destination MAC, dropping");
7633 if (in_bundle->bond) {
7634 struct mac_entry *mac;
7636 switch (bond_check_admissibility(in_bundle->bond, in_port,
7637 flow->dl_dst, &ctx->tags)) {
7642 xlate_report(ctx, "bonding refused admissibility, dropping");
7645 case BV_DROP_IF_MOVED:
7646 mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL);
7647 if (mac && mac->port.p != in_bundle &&
7648 (!is_gratuitous_arp(flow)
7649 || mac_entry_is_grat_arp_locked(mac))) {
7650 xlate_report(ctx, "SLB bond thinks this packet looped back, "
7662 xlate_normal(struct action_xlate_ctx *ctx)
7664 struct ofport_dpif *in_port;
7665 struct ofbundle *in_bundle;
7666 struct mac_entry *mac;
7670 ctx->has_normal = true;
7672 in_bundle = lookup_input_bundle(ctx->ofproto, ctx->flow.in_port,
7673 ctx->packet != NULL, &in_port);
7675 xlate_report(ctx, "no input bundle, dropping");
7679 /* Drop malformed frames. */
7680 if (ctx->flow.dl_type == htons(ETH_TYPE_VLAN) &&
7681 !(ctx->flow.vlan_tci & htons(VLAN_CFI))) {
7682 if (ctx->packet != NULL) {
7683 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7684 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
7685 "VLAN tag received on port %s",
7686 ctx->ofproto->up.name, in_bundle->name);
7688 xlate_report(ctx, "partial VLAN tag, dropping");
7692 /* Drop frames on bundles reserved for mirroring. */
7693 if (in_bundle->mirror_out) {
7694 if (ctx->packet != NULL) {
7695 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7696 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
7697 "%s, which is reserved exclusively for mirroring",
7698 ctx->ofproto->up.name, in_bundle->name);
7700 xlate_report(ctx, "input port is mirror output port, dropping");
7705 vid = vlan_tci_to_vid(ctx->flow.vlan_tci);
7706 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
7707 xlate_report(ctx, "disallowed VLAN VID for this input port, dropping");
7710 vlan = input_vid_to_vlan(in_bundle, vid);
7712 /* Check other admissibility requirements. */
7713 if (in_port && !is_admissible(ctx, in_port, vlan)) {
7717 /* Learn source MAC. */
7718 if (ctx->may_learn) {
7719 update_learning_table(ctx->ofproto, &ctx->flow, vlan, in_bundle);
7722 /* Determine output bundle. */
7723 mac = mac_learning_lookup(ctx->ofproto->ml, ctx->flow.dl_dst, vlan,
7726 if (mac->port.p != in_bundle) {
7727 xlate_report(ctx, "forwarding to learned port");
7728 output_normal(ctx, mac->port.p, vlan);
7730 xlate_report(ctx, "learned port is input port, dropping");
7733 struct ofbundle *bundle;
7735 xlate_report(ctx, "no learned MAC for destination, flooding");
7736 HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
7737 if (bundle != in_bundle
7738 && ofbundle_includes_vlan(bundle, vlan)
7739 && bundle->floodable
7740 && !bundle->mirror_out) {
7741 output_normal(ctx, bundle, vlan);
7744 ctx->nf_output_iface = NF_OUT_FLOOD;
7748 /* Optimized flow revalidation.
7750 * It's a difficult problem, in general, to tell which facets need to have
7751 * their actions recalculated whenever the OpenFlow flow table changes. We
7752 * don't try to solve that general problem: for most kinds of OpenFlow flow
7753 * table changes, we recalculate the actions for every facet. This is
7754 * relatively expensive, but it's good enough if the OpenFlow flow table
7755 * doesn't change very often.
7757 * However, we can expect one particular kind of OpenFlow flow table change to
7758 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
7759 * of CPU on revalidating every facet whenever MAC learning modifies the flow
7760 * table, we add a special case that applies to flow tables in which every rule
7761 * has the same form (that is, the same wildcards), except that the table is
7762 * also allowed to have a single "catch-all" flow that matches all packets. We
7763 * optimize this case by tagging all of the facets that resubmit into the table
7764 * and invalidating the same tag whenever a flow changes in that table. The
7765 * end result is that we revalidate just the facets that need it (and sometimes
7766 * a few more, but not all of the facets or even all of the facets that
7767 * resubmit to the table modified by MAC learning). */
7769 /* Calculates the tag to use for 'flow' and mask 'mask' when it is inserted
7770 * into an OpenFlow table with the given 'basis'. */
7772 rule_calculate_tag(const struct flow *flow, const struct minimask *mask,
7775 if (minimask_is_catchall(mask)) {
7778 uint32_t hash = flow_hash_in_minimask(flow, mask, secret);
7779 return tag_create_deterministic(hash);
7783 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
7784 * taggability of that table.
7786 * This function must be called after *each* change to a flow table. If you
7787 * skip calling it on some changes then the pointer comparisons at the end can
7788 * be invalid if you get unlucky. For example, if a flow removal causes a
7789 * cls_table to be destroyed and then a flow insertion causes a cls_table with
7790 * different wildcards to be created with the same address, then this function
7791 * will incorrectly skip revalidation. */
7793 table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
7795 struct table_dpif *table = &ofproto->tables[table_id];
7796 const struct oftable *oftable = &ofproto->up.tables[table_id];
7797 struct cls_table *catchall, *other;
7798 struct cls_table *t;
7800 catchall = other = NULL;
7802 switch (hmap_count(&oftable->cls.tables)) {
7804 /* We could tag this OpenFlow table but it would make the logic a
7805 * little harder and it's a corner case that doesn't seem worth it
7811 HMAP_FOR_EACH (t, hmap_node, &oftable->cls.tables) {
7812 if (cls_table_is_catchall(t)) {
7814 } else if (!other) {
7817 /* Indicate that we can't tag this by setting both tables to
7818 * NULL. (We know that 'catchall' is already NULL.) */
7825 /* Can't tag this table. */
7829 if (table->catchall_table != catchall || table->other_table != other) {
7830 table->catchall_table = catchall;
7831 table->other_table = other;
7832 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
7836 /* Given 'rule' that has changed in some way (either it is a rule being
7837 * inserted, a rule being deleted, or a rule whose actions are being
7838 * modified), marks facets for revalidation to ensure that packets will be
7839 * forwarded correctly according to the new state of the flow table.
7841 * This function must be called after *each* change to a flow table. See
7842 * the comment on table_update_taggable() for more information. */
7844 rule_invalidate(const struct rule_dpif *rule)
7846 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
7848 table_update_taggable(ofproto, rule->up.table_id);
7850 if (!ofproto->backer->need_revalidate) {
7851 struct table_dpif *table = &ofproto->tables[rule->up.table_id];
7853 if (table->other_table && rule->tag) {
7854 tag_set_add(&ofproto->backer->revalidate_set, rule->tag);
7856 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
7862 set_frag_handling(struct ofproto *ofproto_,
7863 enum ofp_config_flags frag_handling)
7865 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7866 if (frag_handling != OFPC_FRAG_REASM) {
7867 ofproto->backer->need_revalidate = REV_RECONFIGURE;
7875 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
7876 const struct flow *flow,
7877 const struct ofpact *ofpacts, size_t ofpacts_len)
7879 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7880 struct initial_vals initial_vals;
7881 struct odputil_keybuf keybuf;
7882 struct dpif_flow_stats stats;
7886 struct action_xlate_ctx ctx;
7887 uint64_t odp_actions_stub[1024 / 8];
7888 struct ofpbuf odp_actions;
7890 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
7891 odp_flow_key_from_flow(&key, flow,
7892 ofp_port_to_odp_port(ofproto, flow->in_port));
7894 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
7896 initial_vals.vlan_tci = flow->vlan_tci;
7897 action_xlate_ctx_init(&ctx, ofproto, flow, &initial_vals, NULL,
7898 packet_get_tcp_flags(packet, flow), packet);
7899 ctx.resubmit_stats = &stats;
7901 ofpbuf_use_stub(&odp_actions,
7902 odp_actions_stub, sizeof odp_actions_stub);
7903 xlate_actions(&ctx, ofpacts, ofpacts_len, &odp_actions);
7904 dpif_execute(ofproto->backer->dpif, key.data, key.size,
7905 odp_actions.data, odp_actions.size, packet);
7906 ofpbuf_uninit(&odp_actions);
7914 set_netflow(struct ofproto *ofproto_,
7915 const struct netflow_options *netflow_options)
7917 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7919 if (netflow_options) {
7920 if (!ofproto->netflow) {
7921 ofproto->netflow = netflow_create();
7923 return netflow_set_options(ofproto->netflow, netflow_options);
7925 netflow_destroy(ofproto->netflow);
7926 ofproto->netflow = NULL;
7932 get_netflow_ids(const struct ofproto *ofproto_,
7933 uint8_t *engine_type, uint8_t *engine_id)
7935 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7937 dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
7941 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
7943 if (!facet_is_controller_flow(facet) &&
7944 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
7945 struct subfacet *subfacet;
7946 struct ofexpired expired;
7948 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
7949 if (subfacet->path == SF_FAST_PATH) {
7950 struct dpif_flow_stats stats;
7952 subfacet_reinstall(subfacet, &stats);
7953 subfacet_update_stats(subfacet, &stats);
7957 expired.flow = facet->flow;
7958 expired.packet_count = facet->packet_count;
7959 expired.byte_count = facet->byte_count;
7960 expired.used = facet->used;
7961 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
7966 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
7968 struct facet *facet;
7970 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
7971 send_active_timeout(ofproto, facet);
7975 static struct ofproto_dpif *
7976 ofproto_dpif_lookup(const char *name)
7978 struct ofproto_dpif *ofproto;
7980 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
7981 hash_string(name, 0), &all_ofproto_dpifs) {
7982 if (!strcmp(ofproto->up.name, name)) {
7990 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
7991 const char *argv[], void *aux OVS_UNUSED)
7993 struct ofproto_dpif *ofproto;
7996 ofproto = ofproto_dpif_lookup(argv[1]);
7998 unixctl_command_reply_error(conn, "no such bridge");
8001 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
8003 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
8004 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
8008 unixctl_command_reply(conn, "table successfully flushed");
8012 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
8013 const char *argv[], void *aux OVS_UNUSED)
8015 struct ds ds = DS_EMPTY_INITIALIZER;
8016 const struct ofproto_dpif *ofproto;
8017 const struct mac_entry *e;
8019 ofproto = ofproto_dpif_lookup(argv[1]);
8021 unixctl_command_reply_error(conn, "no such bridge");
8025 ds_put_cstr(&ds, " port VLAN MAC Age\n");
8026 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
8027 struct ofbundle *bundle = e->port.p;
8028 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
8029 ofbundle_get_a_port(bundle)->odp_port,
8030 e->vlan, ETH_ADDR_ARGS(e->mac),
8031 mac_entry_age(ofproto->ml, e));
8033 unixctl_command_reply(conn, ds_cstr(&ds));
8038 struct action_xlate_ctx ctx;
8044 trace_format_rule(struct ds *result, uint8_t table_id, int level,
8045 const struct rule_dpif *rule)
8047 ds_put_char_multiple(result, '\t', level);
8049 ds_put_cstr(result, "No match\n");
8053 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
8054 table_id, ntohll(rule->up.flow_cookie));
8055 cls_rule_format(&rule->up.cr, result);
8056 ds_put_char(result, '\n');
8058 ds_put_char_multiple(result, '\t', level);
8059 ds_put_cstr(result, "OpenFlow ");
8060 ofpacts_format(rule->up.ofpacts, rule->up.ofpacts_len, result);
8061 ds_put_char(result, '\n');
8065 trace_format_flow(struct ds *result, int level, const char *title,
8066 struct trace_ctx *trace)
8068 ds_put_char_multiple(result, '\t', level);
8069 ds_put_format(result, "%s: ", title);
8070 if (flow_equal(&trace->ctx.flow, &trace->flow)) {
8071 ds_put_cstr(result, "unchanged");
8073 flow_format(result, &trace->ctx.flow);
8074 trace->flow = trace->ctx.flow;
8076 ds_put_char(result, '\n');
8080 trace_format_regs(struct ds *result, int level, const char *title,
8081 struct trace_ctx *trace)
8085 ds_put_char_multiple(result, '\t', level);
8086 ds_put_format(result, "%s:", title);
8087 for (i = 0; i < FLOW_N_REGS; i++) {
8088 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
8090 ds_put_char(result, '\n');
8094 trace_format_odp(struct ds *result, int level, const char *title,
8095 struct trace_ctx *trace)
8097 struct ofpbuf *odp_actions = trace->ctx.odp_actions;
8099 ds_put_char_multiple(result, '\t', level);
8100 ds_put_format(result, "%s: ", title);
8101 format_odp_actions(result, odp_actions->data, odp_actions->size);
8102 ds_put_char(result, '\n');
8106 trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
8108 struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx);
8109 struct ds *result = trace->result;
8111 ds_put_char(result, '\n');
8112 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
8113 trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace);
8114 trace_format_odp(result, ctx->recurse + 1, "Resubmitted odp", trace);
8115 trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule);
8119 trace_report(struct action_xlate_ctx *ctx, const char *s)
8121 struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx);
8122 struct ds *result = trace->result;
8124 ds_put_char_multiple(result, '\t', ctx->recurse);
8125 ds_put_cstr(result, s);
8126 ds_put_char(result, '\n');
8130 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
8131 void *aux OVS_UNUSED)
8133 const struct dpif_backer *backer;
8134 struct ofproto_dpif *ofproto;
8135 struct ofpbuf odp_key;
8136 struct ofpbuf *packet;
8137 struct initial_vals initial_vals;
8145 ofpbuf_init(&odp_key, 0);
8147 /* Handle "-generate" or a hex string as the last argument. */
8148 if (!strcmp(argv[argc - 1], "-generate")) {
8149 packet = ofpbuf_new(0);
8152 const char *error = eth_from_hex(argv[argc - 1], &packet);
8155 } else if (argc == 4) {
8156 /* The 3-argument form must end in "-generate' or a hex string. */
8157 unixctl_command_reply_error(conn, error);
8162 /* Parse the flow and determine whether a datapath or
8163 * bridge is specified. If function odp_flow_key_from_string()
8164 * returns 0, the flow is a odp_flow. If function
8165 * parse_ofp_exact_flow() returns 0, the flow is a br_flow. */
8166 if (!odp_flow_key_from_string(argv[argc - 1], NULL, &odp_key)) {
8167 /* If the odp_flow is the second argument,
8168 * the datapath name is the first argument. */
8170 const char *dp_type;
8171 if (!strncmp(argv[1], "ovs-", 4)) {
8172 dp_type = argv[1] + 4;
8176 backer = shash_find_data(&all_dpif_backers, dp_type);
8178 unixctl_command_reply_error(conn, "Cannot find datapath "
8183 /* No datapath name specified, so there should be only one
8185 struct shash_node *node;
8186 if (shash_count(&all_dpif_backers) != 1) {
8187 unixctl_command_reply_error(conn, "Must specify datapath "
8188 "name, there is more than one type of datapath");
8191 node = shash_first(&all_dpif_backers);
8192 backer = node->data;
8195 /* Extract the ofproto_dpif object from the ofproto_receive()
8197 if (ofproto_receive(backer, NULL, odp_key.data,
8198 odp_key.size, &flow, NULL, &ofproto, NULL,
8200 unixctl_command_reply_error(conn, "Invalid datapath flow");
8203 ds_put_format(&result, "Bridge: %s\n", ofproto->up.name);
8204 } else if (!parse_ofp_exact_flow(&flow, argv[argc - 1])) {
8206 unixctl_command_reply_error(conn, "Must specify bridge name");
8210 ofproto = ofproto_dpif_lookup(argv[1]);
8212 unixctl_command_reply_error(conn, "Unknown bridge name");
8215 initial_vals.vlan_tci = flow.vlan_tci;
8217 unixctl_command_reply_error(conn, "Bad flow syntax");
8221 /* Generate a packet, if requested. */
8223 if (!packet->size) {
8224 flow_compose(packet, &flow);
8226 ds_put_cstr(&result, "Packet: ");
8227 s = ofp_packet_to_string(packet->data, packet->size);
8228 ds_put_cstr(&result, s);
8231 /* Use the metadata from the flow and the packet argument
8232 * to reconstruct the flow. */
8233 flow_extract(packet, flow.skb_priority, flow.skb_mark, NULL,
8234 flow.in_port, &flow);
8235 initial_vals.vlan_tci = flow.vlan_tci;
8239 ofproto_trace(ofproto, &flow, packet, &initial_vals, &result);
8240 unixctl_command_reply(conn, ds_cstr(&result));
8243 ds_destroy(&result);
8244 ofpbuf_delete(packet);
8245 ofpbuf_uninit(&odp_key);
8249 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
8250 const struct ofpbuf *packet,
8251 const struct initial_vals *initial_vals, struct ds *ds)
8253 struct rule_dpif *rule;
8255 ds_put_cstr(ds, "Flow: ");
8256 flow_format(ds, flow);
8257 ds_put_char(ds, '\n');
8259 rule = rule_dpif_lookup(ofproto, flow);
8261 trace_format_rule(ds, 0, 0, rule);
8262 if (rule == ofproto->miss_rule) {
8263 ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
8264 } else if (rule == ofproto->no_packet_in_rule) {
8265 ds_put_cstr(ds, "\nNo match, packets dropped because "
8266 "OFPPC_NO_PACKET_IN is set on in_port.\n");
8270 uint64_t odp_actions_stub[1024 / 8];
8271 struct ofpbuf odp_actions;
8273 struct trace_ctx trace;
8276 tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
8279 ofpbuf_use_stub(&odp_actions,
8280 odp_actions_stub, sizeof odp_actions_stub);
8281 action_xlate_ctx_init(&trace.ctx, ofproto, flow, initial_vals,
8282 rule, tcp_flags, packet);
8283 trace.ctx.resubmit_hook = trace_resubmit;
8284 trace.ctx.report_hook = trace_report;
8285 xlate_actions(&trace.ctx, rule->up.ofpacts, rule->up.ofpacts_len,
8288 ds_put_char(ds, '\n');
8289 trace_format_flow(ds, 0, "Final flow", &trace);
8290 ds_put_cstr(ds, "Datapath actions: ");
8291 format_odp_actions(ds, odp_actions.data, odp_actions.size);
8292 ofpbuf_uninit(&odp_actions);
8294 if (trace.ctx.slow) {
8295 enum slow_path_reason slow;
8297 ds_put_cstr(ds, "\nThis flow is handled by the userspace "
8298 "slow path because it:");
8299 for (slow = trace.ctx.slow; slow; ) {
8300 enum slow_path_reason bit = rightmost_1bit(slow);
8304 ds_put_cstr(ds, "\n\t- Consists of CFM packets.");
8307 ds_put_cstr(ds, "\n\t- Consists of LACP packets.");
8310 ds_put_cstr(ds, "\n\t- Consists of STP packets.");
8313 ds_put_cstr(ds, "\n\t- Consists of BFD packets.");
8316 ds_put_cstr(ds, "\n\t- Needs in-band special case "
8319 ds_put_cstr(ds, "\n\t (The datapath actions are "
8320 "incomplete--for complete actions, "
8321 "please supply a packet.)");
8324 case SLOW_CONTROLLER:
8325 ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages "
8326 "to the OpenFlow controller.");
8329 ds_put_cstr(ds, "\n\t- Needs more specific matching "
8330 "than the datapath supports.");
8337 if (slow & ~SLOW_MATCH) {
8338 ds_put_cstr(ds, "\nThe datapath actions above do not reflect "
8339 "the special slow-path processing.");
8346 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
8347 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
8350 unixctl_command_reply(conn, NULL);
8354 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
8355 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
8358 unixctl_command_reply(conn, NULL);
8361 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
8362 * 'reply' describing the results. */
8364 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
8366 struct facet *facet;
8370 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
8371 if (!facet_check_consistency(facet)) {
8376 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
8380 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
8381 ofproto->up.name, errors);
8383 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
8388 ofproto_dpif_self_check(struct unixctl_conn *conn,
8389 int argc, const char *argv[], void *aux OVS_UNUSED)
8391 struct ds reply = DS_EMPTY_INITIALIZER;
8392 struct ofproto_dpif *ofproto;
8395 ofproto = ofproto_dpif_lookup(argv[1]);
8397 unixctl_command_reply_error(conn, "Unknown ofproto (use "
8398 "ofproto/list for help)");
8401 ofproto_dpif_self_check__(ofproto, &reply);
8403 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
8404 ofproto_dpif_self_check__(ofproto, &reply);
8408 unixctl_command_reply(conn, ds_cstr(&reply));
8412 /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list
8413 * of the 'ofproto_shash' nodes. It is the responsibility of the caller
8414 * to destroy 'ofproto_shash' and free the returned value. */
8415 static const struct shash_node **
8416 get_ofprotos(struct shash *ofproto_shash)
8418 const struct ofproto_dpif *ofproto;
8420 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
8421 char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
8422 shash_add_nocopy(ofproto_shash, name, ofproto);
8425 return shash_sort(ofproto_shash);
8429 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
8430 const char *argv[] OVS_UNUSED,
8431 void *aux OVS_UNUSED)
8433 struct ds ds = DS_EMPTY_INITIALIZER;
8434 struct shash ofproto_shash;
8435 const struct shash_node **sorted_ofprotos;
8438 shash_init(&ofproto_shash);
8439 sorted_ofprotos = get_ofprotos(&ofproto_shash);
8440 for (i = 0; i < shash_count(&ofproto_shash); i++) {
8441 const struct shash_node *node = sorted_ofprotos[i];
8442 ds_put_format(&ds, "%s\n", node->name);
8445 shash_destroy(&ofproto_shash);
8446 free(sorted_ofprotos);
8448 unixctl_command_reply(conn, ds_cstr(&ds));
8453 show_dp_format(const struct ofproto_dpif *ofproto, struct ds *ds)
8455 const struct shash_node **ports;
8457 struct avg_subfacet_rates lifetime;
8458 unsigned long long int minutes;
8459 const int min_ms = 60 * 1000; /* milliseconds in one minute. */
8461 minutes = (time_msec() - ofproto->created) / min_ms;
8464 lifetime.add_rate = (double)ofproto->total_subfacet_add_count
8466 lifetime.del_rate = (double)ofproto->total_subfacet_del_count
8469 lifetime.add_rate = 0.0;
8470 lifetime.del_rate = 0.0;
8473 ds_put_format(ds, "%s (%s):\n", ofproto->up.name,
8474 dpif_name(ofproto->backer->dpif));
8476 "\tlookups: hit:%"PRIu64" missed:%"PRIu64"\n",
8477 ofproto->n_hit, ofproto->n_missed);
8478 ds_put_format(ds, "\tflows: cur: %zu, avg: %5.3f, max: %d,"
8479 " life span: %llu(ms)\n",
8480 hmap_count(&ofproto->subfacets),
8481 avg_subfacet_count(ofproto),
8482 ofproto->max_n_subfacet,
8483 avg_subfacet_life_span(ofproto));
8484 if (minutes >= 60) {
8485 show_dp_rates(ds, "\t\thourly avg:", &ofproto->hourly);
8487 if (minutes >= 60 * 24) {
8488 show_dp_rates(ds, "\t\tdaily avg:", &ofproto->daily);
8490 show_dp_rates(ds, "\t\toverall avg:", &lifetime);
8492 ports = shash_sort(&ofproto->up.port_by_name);
8493 for (i = 0; i < shash_count(&ofproto->up.port_by_name); i++) {
8494 const struct shash_node *node = ports[i];
8495 struct ofport *ofport = node->data;
8496 const char *name = netdev_get_name(ofport->netdev);
8497 const char *type = netdev_get_type(ofport->netdev);
8500 ds_put_format(ds, "\t%s %u/", name, ofport->ofp_port);
8502 odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
8503 if (odp_port != OVSP_NONE) {
8504 ds_put_format(ds, "%"PRIu32":", odp_port);
8506 ds_put_cstr(ds, "none:");
8509 if (strcmp(type, "system")) {
8510 struct netdev *netdev;
8513 ds_put_format(ds, " (%s", type);
8515 error = netdev_open(name, type, &netdev);
8520 error = netdev_get_config(netdev, &config);
8522 const struct smap_node **nodes;
8525 nodes = smap_sort(&config);
8526 for (i = 0; i < smap_count(&config); i++) {
8527 const struct smap_node *node = nodes[i];
8528 ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
8529 node->key, node->value);
8533 smap_destroy(&config);
8535 netdev_close(netdev);
8537 ds_put_char(ds, ')');
8539 ds_put_char(ds, '\n');
8545 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc,
8546 const char *argv[], void *aux OVS_UNUSED)
8548 struct ds ds = DS_EMPTY_INITIALIZER;
8549 const struct ofproto_dpif *ofproto;
8553 for (i = 1; i < argc; i++) {
8554 ofproto = ofproto_dpif_lookup(argv[i]);
8556 ds_put_format(&ds, "Unknown bridge %s (use dpif/dump-dps "
8557 "for help)", argv[i]);
8558 unixctl_command_reply_error(conn, ds_cstr(&ds));
8561 show_dp_format(ofproto, &ds);
8564 struct shash ofproto_shash;
8565 const struct shash_node **sorted_ofprotos;
8568 shash_init(&ofproto_shash);
8569 sorted_ofprotos = get_ofprotos(&ofproto_shash);
8570 for (i = 0; i < shash_count(&ofproto_shash); i++) {
8571 const struct shash_node *node = sorted_ofprotos[i];
8572 show_dp_format(node->data, &ds);
8575 shash_destroy(&ofproto_shash);
8576 free(sorted_ofprotos);
8579 unixctl_command_reply(conn, ds_cstr(&ds));
8584 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
8585 int argc OVS_UNUSED, const char *argv[],
8586 void *aux OVS_UNUSED)
8588 struct ds ds = DS_EMPTY_INITIALIZER;
8589 const struct ofproto_dpif *ofproto;
8590 struct subfacet *subfacet;
8592 ofproto = ofproto_dpif_lookup(argv[1]);
8594 unixctl_command_reply_error(conn, "no such bridge");
8598 update_stats(ofproto->backer);
8600 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
8601 odp_flow_key_format(subfacet->key, subfacet->key_len, &ds);
8603 ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:",
8604 subfacet->dp_packet_count, subfacet->dp_byte_count);
8605 if (subfacet->used) {
8606 ds_put_format(&ds, "%.3fs",
8607 (time_msec() - subfacet->used) / 1000.0);
8609 ds_put_format(&ds, "never");
8611 if (subfacet->facet->tcp_flags) {
8612 ds_put_cstr(&ds, ", flags:");
8613 packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags);
8616 ds_put_cstr(&ds, ", actions:");
8617 if (subfacet->slow) {
8618 uint64_t slow_path_stub[128 / 8];
8619 const struct nlattr *actions;
8622 compose_slow_path(ofproto, &subfacet->facet->flow, subfacet->slow,
8623 slow_path_stub, sizeof slow_path_stub,
8624 &actions, &actions_len);
8625 format_odp_actions(&ds, actions, actions_len);
8627 format_odp_actions(&ds, subfacet->actions, subfacet->actions_len);
8629 ds_put_char(&ds, '\n');
8632 unixctl_command_reply(conn, ds_cstr(&ds));
8637 ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn,
8638 int argc OVS_UNUSED, const char *argv[],
8639 void *aux OVS_UNUSED)
8641 struct ds ds = DS_EMPTY_INITIALIZER;
8642 struct ofproto_dpif *ofproto;
8644 ofproto = ofproto_dpif_lookup(argv[1]);
8646 unixctl_command_reply_error(conn, "no such bridge");
8650 flush(&ofproto->up);
8652 unixctl_command_reply(conn, ds_cstr(&ds));
8657 ofproto_dpif_unixctl_init(void)
8659 static bool registered;
8665 unixctl_command_register(
8667 "[dp_name]|bridge odp_flow|br_flow [-generate|packet]",
8668 1, 3, ofproto_unixctl_trace, NULL);
8669 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
8670 ofproto_unixctl_fdb_flush, NULL);
8671 unixctl_command_register("fdb/show", "bridge", 1, 1,
8672 ofproto_unixctl_fdb_show, NULL);
8673 unixctl_command_register("ofproto/clog", "", 0, 0,
8674 ofproto_dpif_clog, NULL);
8675 unixctl_command_register("ofproto/unclog", "", 0, 0,
8676 ofproto_dpif_unclog, NULL);
8677 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
8678 ofproto_dpif_self_check, NULL);
8679 unixctl_command_register("dpif/dump-dps", "", 0, 0,
8680 ofproto_unixctl_dpif_dump_dps, NULL);
8681 unixctl_command_register("dpif/show", "[bridge]", 0, INT_MAX,
8682 ofproto_unixctl_dpif_show, NULL);
8683 unixctl_command_register("dpif/dump-flows", "bridge", 1, 1,
8684 ofproto_unixctl_dpif_dump_flows, NULL);
8685 unixctl_command_register("dpif/del-flows", "bridge", 1, 1,
8686 ofproto_unixctl_dpif_del_flows, NULL);
8689 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
8691 * This is deprecated. It is only for compatibility with broken device drivers
8692 * in old versions of Linux that do not properly support VLANs when VLAN
8693 * devices are not used. When broken device drivers are no longer in
8694 * widespread use, we will delete these interfaces. */
8697 set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid)
8699 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
8700 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
8702 if (realdev_ofp_port == ofport->realdev_ofp_port
8703 && vid == ofport->vlandev_vid) {
8707 ofproto->backer->need_revalidate = REV_RECONFIGURE;
8709 if (ofport->realdev_ofp_port) {
8712 if (realdev_ofp_port && ofport->bundle) {
8713 /* vlandevs are enslaved to their realdevs, so they are not allowed to
8714 * themselves be part of a bundle. */
8715 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
8718 ofport->realdev_ofp_port = realdev_ofp_port;
8719 ofport->vlandev_vid = vid;
8721 if (realdev_ofp_port) {
8722 vsp_add(ofport, realdev_ofp_port, vid);
8729 hash_realdev_vid(uint16_t realdev_ofp_port, int vid)
8731 return hash_2words(realdev_ofp_port, vid);
8734 /* Returns the ODP port number of the Linux VLAN device that corresponds to
8735 * 'vlan_tci' on the network device with port number 'realdev_odp_port' in
8736 * 'ofproto'. For example, given 'realdev_odp_port' of eth0 and 'vlan_tci' 9,
8737 * it would return the port number of eth0.9.
8739 * Unless VLAN splinters are enabled for port 'realdev_odp_port', this
8740 * function just returns its 'realdev_odp_port' argument. */
8742 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
8743 uint32_t realdev_odp_port, ovs_be16 vlan_tci)
8745 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
8746 uint16_t realdev_ofp_port;
8747 int vid = vlan_tci_to_vid(vlan_tci);
8748 const struct vlan_splinter *vsp;
8750 realdev_ofp_port = odp_port_to_ofp_port(ofproto, realdev_odp_port);
8751 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
8752 hash_realdev_vid(realdev_ofp_port, vid),
8753 &ofproto->realdev_vid_map) {
8754 if (vsp->realdev_ofp_port == realdev_ofp_port
8755 && vsp->vid == vid) {
8756 return ofp_port_to_odp_port(ofproto, vsp->vlandev_ofp_port);
8760 return realdev_odp_port;
8763 static struct vlan_splinter *
8764 vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port)
8766 struct vlan_splinter *vsp;
8768 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, hash_int(vlandev_ofp_port, 0),
8769 &ofproto->vlandev_map) {
8770 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
8778 /* Returns the OpenFlow port number of the "real" device underlying the Linux
8779 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
8780 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
8781 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
8782 * eth0 and store 9 in '*vid'.
8784 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
8785 * VLAN device. Unless VLAN splinters are enabled, this is what this function
8788 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
8789 uint16_t vlandev_ofp_port, int *vid)
8791 if (!hmap_is_empty(&ofproto->vlandev_map)) {
8792 const struct vlan_splinter *vsp;
8794 vsp = vlandev_find(ofproto, vlandev_ofp_port);
8799 return vsp->realdev_ofp_port;
8805 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
8806 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
8807 * 'flow->in_port' to the "real" device backing the VLAN device, sets
8808 * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is
8809 * always the case unless VLAN splinters are enabled), returns false without
8810 * making any changes. */
8812 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
8817 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port, &vid);
8822 /* Cause the flow to be processed as if it came in on the real device with
8823 * the VLAN device's VLAN ID. */
8824 flow->in_port = realdev;
8825 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
8830 vsp_remove(struct ofport_dpif *port)
8832 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
8833 struct vlan_splinter *vsp;
8835 vsp = vlandev_find(ofproto, port->up.ofp_port);
8837 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
8838 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
8841 port->realdev_ofp_port = 0;
8843 VLOG_ERR("missing vlan device record");
8848 vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid)
8850 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
8852 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
8853 && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid))
8854 == realdev_ofp_port)) {
8855 struct vlan_splinter *vsp;
8857 vsp = xmalloc(sizeof *vsp);
8858 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
8859 hash_int(port->up.ofp_port, 0));
8860 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
8861 hash_realdev_vid(realdev_ofp_port, vid));
8862 vsp->realdev_ofp_port = realdev_ofp_port;
8863 vsp->vlandev_ofp_port = port->up.ofp_port;
8866 port->realdev_ofp_port = realdev_ofp_port;
8868 VLOG_ERR("duplicate vlan device record");
8873 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
8875 const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
8876 return ofport ? ofport->odp_port : OVSP_NONE;
8879 static struct ofport_dpif *
8880 odp_port_to_ofport(const struct dpif_backer *backer, uint32_t odp_port)
8882 struct ofport_dpif *port;
8884 HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node,
8885 hash_int(odp_port, 0),
8886 &backer->odp_to_ofport_map) {
8887 if (port->odp_port == odp_port) {
8896 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
8898 struct ofport_dpif *port;
8900 port = odp_port_to_ofport(ofproto->backer, odp_port);
8901 if (port && &ofproto->up == port->up.ofproto) {
8902 return port->up.ofp_port;
8907 static unsigned long long int
8908 avg_subfacet_life_span(const struct ofproto_dpif *ofproto)
8910 unsigned long long int dc;
8911 unsigned long long int avg;
8913 dc = ofproto->total_subfacet_del_count + ofproto->subfacet_del_count;
8914 avg = dc ? ofproto->total_subfacet_life_span / dc : 0;
8920 avg_subfacet_count(const struct ofproto_dpif *ofproto)
8924 if (ofproto->n_update_stats) {
8925 avg_c = (double)ofproto->total_subfacet_count
8926 / ofproto->n_update_stats;
8933 show_dp_rates(struct ds *ds, const char *heading,
8934 const struct avg_subfacet_rates *rates)
8936 ds_put_format(ds, "%s add rate: %5.3f/min, del rate: %5.3f/min\n",
8937 heading, rates->add_rate, rates->del_rate);
8941 update_max_subfacet_count(struct ofproto_dpif *ofproto)
8943 ofproto->max_n_subfacet = MAX(ofproto->max_n_subfacet,
8944 hmap_count(&ofproto->subfacets));
8947 /* Compute exponentially weighted moving average, adding 'new' as the newest,
8948 * most heavily weighted element. 'base' designates the rate of decay: after
8949 * 'base' further updates, 'new''s weight in the EWMA decays to about 1/e
8952 exp_mavg(double *avg, int base, double new)
8954 *avg = (*avg * (base - 1) + new) / base;
8958 update_moving_averages(struct ofproto_dpif *ofproto)
8960 const int min_ms = 60 * 1000; /* milliseconds in one minute. */
8962 /* Update hourly averages on the minute boundaries. */
8963 if (time_msec() - ofproto->last_minute >= min_ms) {
8964 exp_mavg(&ofproto->hourly.add_rate, 60, ofproto->subfacet_add_count);
8965 exp_mavg(&ofproto->hourly.del_rate, 60, ofproto->subfacet_del_count);
8967 /* Update daily averages on the hour boundaries. */
8968 if ((ofproto->last_minute - ofproto->created) / min_ms % 60 == 59) {
8969 exp_mavg(&ofproto->daily.add_rate, 24, ofproto->hourly.add_rate);
8970 exp_mavg(&ofproto->daily.del_rate, 24, ofproto->hourly.del_rate);
8973 ofproto->total_subfacet_add_count += ofproto->subfacet_add_count;
8974 ofproto->total_subfacet_del_count += ofproto->subfacet_del_count;
8975 ofproto->subfacet_add_count = 0;
8976 ofproto->subfacet_del_count = 0;
8977 ofproto->last_minute += min_ms;
8982 dpif_stats_update_hit_count(struct ofproto_dpif *ofproto, uint64_t delta)
8984 ofproto->n_hit += delta;
8987 const struct ofproto_class ofproto_dpif_class = {
9022 port_is_lacp_current,
9023 NULL, /* rule_choose_table */
9030 rule_modify_actions,
9044 get_stp_port_status,
9051 is_mirror_output_bundle,
9052 forward_bpdu_changed,
9053 set_mac_table_config,