2 * Copyright (c) 2009, 2010, 2011, 2012, 2013 Nicira, Inc.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at:
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
19 #include "ofproto/ofproto-provider.h"
25 #include "byte-order.h"
30 #include "dynamic-string.h"
31 #include "fail-open.h"
35 #include "mac-learning.h"
36 #include "meta-flow.h"
37 #include "multipath.h"
38 #include "netdev-vport.h"
45 #include "ofp-actions.h"
46 #include "ofp-parse.h"
47 #include "ofp-print.h"
48 #include "ofproto-dpif-governor.h"
49 #include "ofproto-dpif-ipfix.h"
50 #include "ofproto-dpif-sflow.h"
51 #include "poll-loop.h"
56 #include "unaligned.h"
58 #include "vlan-bitmap.h"
61 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
63 COVERAGE_DEFINE(ofproto_dpif_expired);
64 COVERAGE_DEFINE(ofproto_dpif_xlate);
65 COVERAGE_DEFINE(facet_changed_rule);
66 COVERAGE_DEFINE(facet_revalidate);
67 COVERAGE_DEFINE(facet_unexpected);
68 COVERAGE_DEFINE(facet_suppress);
70 /* Maximum depth of flow table recursion (due to resubmit actions) in a
71 * flow translation. */
72 #define MAX_RESUBMIT_RECURSION 64
74 /* Number of implemented OpenFlow tables. */
75 enum { N_TABLES = 255 };
76 enum { TBL_INTERNAL = N_TABLES - 1 }; /* Used for internal hidden rules. */
77 BUILD_ASSERT_DECL(N_TABLES >= 2 && N_TABLES <= 255);
89 * - Do include packets and bytes from facets that have been deleted or
90 * whose own statistics have been folded into the rule.
92 * - Do include packets and bytes sent "by hand" that were accounted to
93 * the rule without any facet being involved (this is a rare corner
94 * case in rule_execute()).
96 * - Do not include packet or bytes that can be obtained from any facet's
97 * packet_count or byte_count member or that can be obtained from the
98 * datapath by, e.g., dpif_flow_get() for any subfacet.
100 uint64_t packet_count; /* Number of packets received. */
101 uint64_t byte_count; /* Number of bytes received. */
103 tag_type tag; /* Caches rule_calculate_tag() result. */
105 struct list facets; /* List of "struct facet"s. */
108 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
110 return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
113 static struct rule_dpif *rule_dpif_lookup(struct ofproto_dpif *,
115 struct flow_wildcards *wc);
116 static struct rule_dpif *rule_dpif_lookup__(struct ofproto_dpif *,
118 struct flow_wildcards *wc,
120 static struct rule_dpif *rule_dpif_miss_rule(struct ofproto_dpif *ofproto,
121 const struct flow *flow);
123 static void rule_get_stats(struct rule *, uint64_t *packets, uint64_t *bytes);
124 static void rule_credit_stats(struct rule_dpif *,
125 const struct dpif_flow_stats *);
126 static tag_type rule_calculate_tag(const struct flow *,
127 const struct minimask *, uint32_t basis);
128 static void rule_invalidate(const struct rule_dpif *);
130 #define MAX_MIRRORS 32
131 typedef uint32_t mirror_mask_t;
132 #define MIRROR_MASK_C(X) UINT32_C(X)
133 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
135 struct ofproto_dpif *ofproto; /* Owning ofproto. */
136 size_t idx; /* In ofproto's "mirrors" array. */
137 void *aux; /* Key supplied by ofproto's client. */
138 char *name; /* Identifier for log messages. */
140 /* Selection criteria. */
141 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
142 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
143 unsigned long *vlans; /* Bitmap of chosen VLANs, NULL selects all. */
145 /* Output (exactly one of out == NULL and out_vlan == -1 is true). */
146 struct ofbundle *out; /* Output port or NULL. */
147 int out_vlan; /* Output VLAN or -1. */
148 mirror_mask_t dup_mirrors; /* Bitmap of mirrors with the same output. */
151 int64_t packet_count; /* Number of packets sent. */
152 int64_t byte_count; /* Number of bytes sent. */
155 static void mirror_destroy(struct ofmirror *);
156 static void update_mirror_stats(struct ofproto_dpif *ofproto,
157 mirror_mask_t mirrors,
158 uint64_t packets, uint64_t bytes);
161 struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
162 struct ofproto_dpif *ofproto; /* Owning ofproto. */
163 void *aux; /* Key supplied by ofproto's client. */
164 char *name; /* Identifier for log messages. */
167 struct list ports; /* Contains "struct ofport"s. */
168 enum port_vlan_mode vlan_mode; /* VLAN mode */
169 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
170 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
171 * NULL if all VLANs are trunked. */
172 struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */
173 struct bond *bond; /* Nonnull iff more than one port. */
174 bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */
177 bool floodable; /* True if no port has OFPUTIL_PC_NO_FLOOD set. */
179 /* Port mirroring info. */
180 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
181 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
182 mirror_mask_t mirror_out; /* Mirrors that output to this bundle. */
185 static void bundle_remove(struct ofport *);
186 static void bundle_update(struct ofbundle *);
187 static void bundle_destroy(struct ofbundle *);
188 static void bundle_del_port(struct ofport_dpif *);
189 static void bundle_run(struct ofbundle *);
190 static void bundle_wait(struct ofbundle *);
191 static struct ofbundle *lookup_input_bundle(const struct ofproto_dpif *,
192 uint16_t in_port, bool warn,
193 struct ofport_dpif **in_ofportp);
195 /* A controller may use OFPP_NONE as the ingress port to indicate that
196 * it did not arrive on a "real" port. 'ofpp_none_bundle' exists for
197 * when an input bundle is needed for validation (e.g., mirroring or
198 * OFPP_NORMAL processing). It is not connected to an 'ofproto' or have
199 * any 'port' structs, so care must be taken when dealing with it. */
200 static struct ofbundle ofpp_none_bundle = {
202 .vlan_mode = PORT_VLAN_TRUNK
205 static void stp_run(struct ofproto_dpif *ofproto);
206 static void stp_wait(struct ofproto_dpif *ofproto);
207 static int set_stp_port(struct ofport *,
208 const struct ofproto_port_stp_settings *);
210 static bool ofbundle_includes_vlan(const struct ofbundle *, uint16_t vlan);
214 /* Initial values of fields of the packet that may be changed during
215 * flow processing and needed later. */
216 struct initial_vals {
217 /* This is the value of vlan_tci in the packet as actually received from
218 * dpif. This is the same as the facet's flow.vlan_tci unless the packet
219 * was received via a VLAN splinter. In that case, this value is 0
220 * (because the packet as actually received from the dpif had no 802.1Q
221 * tag) but the facet's flow.vlan_tci is set to the VLAN that the splinter
224 * This member should be removed when the VLAN splinters feature is no
228 /* If received on a tunnel, the IP TOS value of the tunnel. */
229 uint8_t tunnel_ip_tos;
233 /* Wildcards relevant in translation. Any fields that were used to
234 * calculate the action must be set for caching and kernel
235 * wildcarding to work. For example, if the flow lookup involved
236 * performing the "normal" action on IPv4 and ARP packets, 'wc'
237 * would have the 'in_port' (always set), 'dl_type' (flow match),
238 * 'vlan_tci' (normal action), and 'dl_dst' (normal action) fields
240 struct flow_wildcards wc;
242 tag_type tags; /* Tags associated with actions. */
243 enum slow_path_reason slow; /* 0 if fast path may be used. */
244 bool has_learn; /* Actions include NXAST_LEARN? */
245 bool has_normal; /* Actions output to OFPP_NORMAL? */
246 bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */
247 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
248 mirror_mask_t mirrors; /* Bitmap of associated mirrors. */
250 uint64_t odp_actions_stub[256 / 8];
251 struct ofpbuf odp_actions;
255 struct ofproto_dpif *ofproto;
257 /* Flow to which the OpenFlow actions apply. xlate_actions() will modify
258 * this flow when actions change header fields. */
261 struct initial_vals initial_vals;
263 /* The packet corresponding to 'flow', or a null pointer if we are
264 * revalidating without a packet to refer to. */
265 const struct ofpbuf *packet;
267 /* Should OFPP_NORMAL update the MAC learning table? Should "learn"
268 * actions update the flow table?
270 * We want to update these tables if we are actually processing a packet,
271 * or if we are accounting for packets that the datapath has processed, but
272 * not if we are just revalidating. */
275 /* The rule initiating translation or NULL. */
276 struct rule_dpif *rule;
278 /* The actions to translate. If 'rule' is not NULL, these may be NULL. */
279 const struct ofpact *ofpacts;
282 /* Union of the set of TCP flags seen so far in this flow. (Used only by
283 * NXAST_FIN_TIMEOUT. Set to zero to avoid updating updating rules'
287 /* If nonnull, flow translation calls this function just before executing a
288 * resubmit or OFPP_TABLE action. In addition, disables logging of traces
289 * when the recursion depth is exceeded.
291 * 'rule' is the rule being submitted into. It will be null if the
292 * resubmit or OFPP_TABLE action didn't find a matching rule.
294 * This is normally null so the client has to set it manually after
295 * calling xlate_in_init(). */
296 void (*resubmit_hook)(struct xlate_ctx *, struct rule_dpif *rule);
298 /* If nonnull, flow translation calls this function to report some
299 * significant decision, e.g. to explain why OFPP_NORMAL translation
300 * dropped a packet. */
301 void (*report_hook)(struct xlate_ctx *, const char *s);
303 /* If nonnull, flow translation credits the specified statistics to each
304 * rule reached through a resubmit or OFPP_TABLE action.
306 * This is normally null so the client has to set it manually after
307 * calling xlate_in_init(). */
308 const struct dpif_flow_stats *resubmit_stats;
311 /* Context used by xlate_actions() and its callees. */
313 struct xlate_in *xin;
314 struct xlate_out *xout;
316 struct ofproto_dpif *ofproto;
318 /* Flow at the last commit. */
319 struct flow base_flow;
321 /* Stack for the push and pop actions. Each stack element is of type
322 * "union mf_subvalue". */
323 union mf_subvalue init_stack[1024 / sizeof(union mf_subvalue)];
326 /* The rule that we are currently translating, or NULL. */
327 struct rule_dpif *rule;
329 int recurse; /* Recursion level, via xlate_table_action. */
330 bool max_resubmit_trigger; /* Recursed too deeply during translation. */
331 uint32_t orig_skb_priority; /* Priority when packet arrived. */
332 uint8_t table_id; /* OpenFlow table ID where flow was found. */
333 uint32_t sflow_n_outputs; /* Number of output ports. */
334 uint32_t sflow_odp_port; /* Output port for composing sFlow action. */
335 uint16_t user_cookie_offset;/* Used for user_action_cookie fixup. */
336 bool exit; /* No further actions should be processed. */
339 static void xlate_in_init(struct xlate_in *, struct ofproto_dpif *,
340 const struct flow *, const struct initial_vals *,
341 struct rule_dpif *, uint8_t tcp_flags,
342 const struct ofpbuf *);
344 static void xlate_out_uninit(struct xlate_out *);
346 static void xlate_actions(struct xlate_in *, struct xlate_out *);
348 static void xlate_actions_for_side_effects(struct xlate_in *);
350 static void xlate_table_action(struct xlate_ctx *, uint16_t in_port,
351 uint8_t table_id, bool may_packet_in);
353 static size_t put_userspace_action(const struct ofproto_dpif *,
354 struct ofpbuf *odp_actions,
356 const union user_action_cookie *,
359 static void compose_slow_path(const struct ofproto_dpif *, const struct flow *,
360 enum slow_path_reason,
361 uint64_t *stub, size_t stub_size,
362 const struct nlattr **actionsp,
363 size_t *actions_lenp);
365 static void xlate_report(struct xlate_ctx *ctx, const char *s);
367 static void xlate_out_copy(struct xlate_out *dst, const struct xlate_out *src);
369 /* A subfacet (see "struct subfacet" below) has three possible installation
372 * - SF_NOT_INSTALLED: Not installed in the datapath. This will only be the
373 * case just after the subfacet is created, just before the subfacet is
374 * destroyed, or if the datapath returns an error when we try to install a
377 * - SF_FAST_PATH: The subfacet's actions are installed in the datapath.
379 * - SF_SLOW_PATH: An action that sends every packet for the subfacet through
380 * ofproto_dpif is installed in the datapath.
383 SF_NOT_INSTALLED, /* No datapath flow for this subfacet. */
384 SF_FAST_PATH, /* Full actions are installed. */
385 SF_SLOW_PATH, /* Send-to-userspace action is installed. */
388 /* A dpif flow and actions associated with a facet.
390 * See also the large comment on struct facet. */
393 struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */
394 struct list list_node; /* In struct facet's 'facets' list. */
395 struct facet *facet; /* Owning facet. */
397 enum odp_key_fitness key_fitness;
401 long long int used; /* Time last used; time created if not used. */
402 long long int created; /* Time created. */
404 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
405 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
407 enum subfacet_path path; /* Installed in datapath? */
409 /* Datapath port the packet arrived on. This is needed to remove
410 * flows for ports that are no longer part of the bridge. Since the
411 * flow definition only has the OpenFlow port number and the port is
412 * no longer part of the bridge, we can't determine the datapath port
413 * number needed to delete the flow from the datapath. */
414 uint32_t odp_in_port;
417 #define SUBFACET_DESTROY_MAX_BATCH 50
419 static struct subfacet *subfacet_create(struct facet *, struct flow_miss *miss,
421 static struct subfacet *subfacet_find(struct ofproto_dpif *,
422 const struct nlattr *key, size_t key_len,
424 static void subfacet_destroy(struct subfacet *);
425 static void subfacet_destroy__(struct subfacet *);
426 static void subfacet_destroy_batch(struct ofproto_dpif *,
427 struct subfacet **, int n);
428 static void subfacet_reset_dp_stats(struct subfacet *,
429 struct dpif_flow_stats *);
430 static void subfacet_update_stats(struct subfacet *,
431 const struct dpif_flow_stats *);
432 static int subfacet_install(struct subfacet *,
433 const struct ofpbuf *odp_actions,
434 struct dpif_flow_stats *);
435 static void subfacet_uninstall(struct subfacet *);
437 /* A unique, non-overlapping instantiation of an OpenFlow flow.
439 * A facet associates a "struct flow", which represents the Open vSwitch
440 * userspace idea of an exact-match flow, with one or more subfacets.
441 * While the facet is created based on an exact-match flow, it is stored
442 * within the ofproto based on the wildcards that could be expressed
443 * based on the flow table and other configuration. (See the 'wc'
444 * description in "struct xlate_out" for more details.)
446 * Each subfacet tracks the datapath's idea of the flow equivalent to
447 * the facet. When the kernel module (or other dpif implementation) and
448 * Open vSwitch userspace agree on the definition of a flow key, there
449 * is exactly one subfacet per facet. If the dpif implementation
450 * supports more-specific flow matching than userspace, however, a facet
451 * can have more than one subfacet. Examples include the dpif
452 * implementation not supporting the same wildcards as userspace or some
453 * distinction in flow that userspace simply doesn't understand.
455 * Flow expiration works in terms of subfacets, so a facet must have at
456 * least one subfacet or it will never expire, leaking memory. */
459 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
460 struct list list_node; /* In owning rule's 'facets' list. */
461 struct rule_dpif *rule; /* Owning rule. */
464 struct list subfacets;
465 long long int used; /* Time last used; time created if not used. */
468 struct flow flow; /* Flow of the creating subfacet. */
469 struct cls_rule cr; /* In 'ofproto_dpif's facets classifier. */
473 * - Do include packets and bytes sent "by hand", e.g. with
476 * - Do include packets and bytes that were obtained from the datapath
477 * when a subfacet's statistics were reset (e.g. dpif_flow_put() with
478 * DPIF_FP_ZERO_STATS).
480 * - Do not include packets or bytes that can be obtained from the
481 * datapath for any existing subfacet.
483 uint64_t packet_count; /* Number of packets received. */
484 uint64_t byte_count; /* Number of bytes received. */
486 /* Resubmit statistics. */
487 uint64_t prev_packet_count; /* Number of packets from last stats push. */
488 uint64_t prev_byte_count; /* Number of bytes from last stats push. */
489 long long int prev_used; /* Used time from last stats push. */
492 uint64_t accounted_bytes; /* Bytes processed by facet_account(). */
493 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
494 uint8_t tcp_flags; /* TCP flags seen for this 'rule'. */
496 struct xlate_out xout;
498 /* Initial values of the packet that may be needed later. */
499 struct initial_vals initial_vals;
501 /* Storage for a single subfacet, to reduce malloc() time and space
502 * overhead. (A facet always has at least one subfacet and in the common
503 * case has exactly one subfacet. However, 'one_subfacet' may not
504 * always be valid, since it could have been removed after newer
505 * subfacets were pushed onto the 'subfacets' list.) */
506 struct subfacet one_subfacet;
508 long long int learn_rl; /* Rate limiter for facet_learn(). */
511 static struct facet *facet_create(const struct flow_miss *, struct rule_dpif *,
513 struct dpif_flow_stats *);
514 static void facet_remove(struct facet *);
515 static void facet_free(struct facet *);
517 static struct facet *facet_find(struct ofproto_dpif *, const struct flow *);
518 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
519 const struct flow *);
520 static bool facet_revalidate(struct facet *);
521 static bool facet_check_consistency(struct facet *);
523 static void facet_flush_stats(struct facet *);
525 static void facet_reset_counters(struct facet *);
526 static void facet_push_stats(struct facet *, bool may_learn);
527 static void facet_learn(struct facet *);
528 static void facet_account(struct facet *);
529 static void push_all_stats(void);
531 static bool facet_is_controller_flow(struct facet *);
534 struct hmap_node odp_port_node; /* In dpif_backer's "odp_to_ofport_map". */
538 struct ofbundle *bundle; /* Bundle that contains this port, if any. */
539 struct list bundle_node; /* In struct ofbundle's "ports" list. */
540 struct cfm *cfm; /* Connectivity Fault Management, if any. */
541 tag_type tag; /* Tag associated with this port. */
542 bool may_enable; /* May be enabled in bonds. */
543 long long int carrier_seq; /* Carrier status changes. */
544 struct tnl_port *tnl_port; /* Tunnel handle, or null. */
547 struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */
548 enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */
549 long long int stp_state_entered;
551 struct hmap priorities; /* Map of attached 'priority_to_dscp's. */
553 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
555 * This is deprecated. It is only for compatibility with broken device
556 * drivers in old versions of Linux that do not properly support VLANs when
557 * VLAN devices are not used. When broken device drivers are no longer in
558 * widespread use, we will delete these interfaces. */
559 uint16_t realdev_ofp_port;
563 /* Node in 'ofport_dpif''s 'priorities' map. Used to maintain a map from
564 * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
565 * traffic egressing the 'ofport' with that priority should be marked with. */
566 struct priority_to_dscp {
567 struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'priorities' map. */
568 uint32_t priority; /* Priority of this queue (see struct flow). */
570 uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */
573 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
575 * This is deprecated. It is only for compatibility with broken device drivers
576 * in old versions of Linux that do not properly support VLANs when VLAN
577 * devices are not used. When broken device drivers are no longer in
578 * widespread use, we will delete these interfaces. */
579 struct vlan_splinter {
580 struct hmap_node realdev_vid_node;
581 struct hmap_node vlandev_node;
582 uint16_t realdev_ofp_port;
583 uint16_t vlandev_ofp_port;
587 static uint16_t vsp_realdev_to_vlandev(const struct ofproto_dpif *,
588 uint16_t realdev_ofp_port,
590 static bool vsp_adjust_flow(const struct ofproto_dpif *, struct flow *);
591 static void vsp_remove(struct ofport_dpif *);
592 static void vsp_add(struct ofport_dpif *, uint16_t realdev_ofp_port, int vid);
594 static uint32_t ofp_port_to_odp_port(const struct ofproto_dpif *,
596 static uint16_t odp_port_to_ofp_port(const struct ofproto_dpif *,
599 static struct ofport_dpif *
600 ofport_dpif_cast(const struct ofport *ofport)
602 ovs_assert(ofport->ofproto->ofproto_class == &ofproto_dpif_class);
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_cfm(struct ofport *, const struct cfm_settings *);
610 static void ofport_clear_priorities(struct ofport_dpif *);
611 static void run_fast_rl(void);
613 struct dpif_completion {
614 struct list list_node;
615 struct ofoperation *op;
618 /* Extra information about a classifier table.
619 * Currently used just for optimized flow revalidation. */
621 /* If either of these is nonnull, then this table has a form that allows
622 * flows to be tagged to avoid revalidating most flows for the most common
623 * kinds of flow table changes. */
624 struct cls_table *catchall_table; /* Table that wildcards all fields. */
625 struct cls_table *other_table; /* Table with any other wildcard set. */
626 uint32_t basis; /* Keeps each table's tags separate. */
629 /* Reasons that we might need to revalidate every facet, and corresponding
632 * A value of 0 means that there is no need to revalidate.
634 * It would be nice to have some cleaner way to integrate with coverage
635 * counters, but with only a few reasons I guess this is good enough for
637 enum revalidate_reason {
638 REV_RECONFIGURE = 1, /* Switch configuration changed. */
639 REV_STP, /* Spanning tree protocol port status change. */
640 REV_PORT_TOGGLED, /* Port enabled or disabled by CFM, LACP, ...*/
641 REV_FLOW_TABLE, /* Flow table changed. */
642 REV_INCONSISTENCY /* Facet self-check failed. */
644 COVERAGE_DEFINE(rev_reconfigure);
645 COVERAGE_DEFINE(rev_stp);
646 COVERAGE_DEFINE(rev_port_toggled);
647 COVERAGE_DEFINE(rev_flow_table);
648 COVERAGE_DEFINE(rev_inconsistency);
650 /* Drop keys are odp flow keys which have drop flows installed in the kernel.
651 * These are datapath flows which have no associated ofproto, if they did we
652 * would use facets. */
654 struct hmap_node hmap_node;
659 /* All datapaths of a given type share a single dpif backer instance. */
664 struct timer next_expiration;
665 struct hmap odp_to_ofport_map; /* ODP port to ofport mapping. */
667 struct simap tnl_backers; /* Set of dpif ports backing tunnels. */
669 /* Facet revalidation flags applying to facets which use this backer. */
670 enum revalidate_reason need_revalidate; /* Revalidate every facet. */
671 struct tag_set revalidate_set; /* Revalidate only matching facets. */
673 struct hmap drop_keys; /* Set of dropped odp keys. */
674 bool recv_set_enable; /* Enables or disables receiving packets. */
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. */
702 struct rule_dpif *drop_frags_rule; /* Used in OFPC_FRAG_DROP mode. */
705 struct netflow *netflow;
706 struct dpif_sflow *sflow;
707 struct dpif_ipfix *ipfix;
708 struct hmap bundles; /* Contains "struct ofbundle"s. */
709 struct mac_learning *ml;
710 struct ofmirror *mirrors[MAX_MIRRORS];
712 bool has_bonded_bundles;
715 struct classifier facets; /* Contains 'struct facet's. */
716 struct hmap subfacets;
717 struct governor *governor;
718 long long int consistency_rl;
721 struct table_dpif tables[N_TABLES];
723 /* Support for debugging async flow mods. */
724 struct list completions;
726 bool has_bundle_action; /* True when the first bundle action appears. */
727 struct netdev_stats stats; /* To account packets generated and consumed in
732 long long int stp_last_tick;
734 /* VLAN splinters. */
735 struct hmap realdev_vid_map; /* (realdev,vid) -> vlandev. */
736 struct hmap vlandev_map; /* vlandev -> (realdev,vid). */
739 struct sset ports; /* Set of standard port names. */
740 struct sset ghost_ports; /* Ports with no datapath port. */
741 struct sset port_poll_set; /* Queued names for port_poll() reply. */
742 int port_poll_errno; /* Last errno for port_poll() reply. */
744 /* Per ofproto's dpif stats. */
748 /* Subfacet statistics.
750 * These keep track of the total number of subfacets added and deleted and
751 * flow life span. They are useful for computing the flow rates stats
752 * exposed via "ovs-appctl dpif/show". The goal is to learn about
753 * traffic patterns in ways that we can use later to improve Open vSwitch
754 * performance in new situations. */
755 long long int created; /* Time when it is created. */
756 unsigned int max_n_subfacet; /* Maximum number of flows */
758 /* The average number of subfacets... */
759 struct avg_subfacet_rates hourly; /* ...over the last hour. */
760 struct avg_subfacet_rates daily; /* ...over the last day. */
761 long long int last_minute; /* Last time 'hourly' was updated. */
763 /* Number of subfacets added or deleted since 'last_minute'. */
764 unsigned int subfacet_add_count;
765 unsigned int subfacet_del_count;
767 /* Number of subfacets added or deleted from 'created' to 'last_minute.' */
768 unsigned long long int total_subfacet_add_count;
769 unsigned long long int total_subfacet_del_count;
771 /* Sum of the number of milliseconds that each subfacet existed,
772 * over the subfacets that have been added and then later deleted. */
773 unsigned long long int total_subfacet_life_span;
775 /* Incremented by the number of currently existing subfacets, each
776 * time we pull statistics from the kernel. */
777 unsigned long long int total_subfacet_count;
779 /* Number of times we pull statistics from the kernel. */
780 unsigned long long int n_update_stats;
782 static unsigned long long int avg_subfacet_life_span(
783 const struct ofproto_dpif *);
784 static double avg_subfacet_count(const struct ofproto_dpif *ofproto);
785 static void update_moving_averages(struct ofproto_dpif *ofproto);
786 static void dpif_stats_update_hit_count(struct ofproto_dpif *ofproto,
788 static void update_max_subfacet_count(struct ofproto_dpif *ofproto);
790 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
791 * for debugging the asynchronous flow_mod implementation.) */
794 /* All existing ofproto_dpif instances, indexed by ->up.name. */
795 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
797 static void ofproto_dpif_unixctl_init(void);
799 static struct ofproto_dpif *
800 ofproto_dpif_cast(const struct ofproto *ofproto)
802 ovs_assert(ofproto->ofproto_class == &ofproto_dpif_class);
803 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
806 static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *,
808 static struct ofport_dpif *get_odp_port(const struct ofproto_dpif *,
810 static void ofproto_trace(struct ofproto_dpif *, const struct flow *,
811 const struct ofpbuf *,
812 const struct initial_vals *, struct ds *);
814 /* Packet processing. */
815 static void update_learning_table(struct ofproto_dpif *, const struct flow *,
816 struct flow_wildcards *, int vlan,
819 #define FLOW_MISS_MAX_BATCH 50
820 static int handle_upcalls(struct dpif_backer *, unsigned int max_batch);
822 /* Flow expiration. */
823 static int expire(struct dpif_backer *);
826 static void send_netflow_active_timeouts(struct ofproto_dpif *);
829 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
830 static size_t compose_sflow_action(const struct ofproto_dpif *,
831 struct ofpbuf *odp_actions,
832 const struct flow *, uint32_t odp_port);
833 static void compose_ipfix_action(const struct ofproto_dpif *,
834 struct ofpbuf *odp_actions,
835 const struct flow *);
836 static void add_mirror_actions(struct xlate_ctx *ctx,
837 const struct flow *flow);
838 /* Global variables. */
839 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
841 /* Initial mappings of port to bridge mappings. */
842 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
844 /* Factory functions. */
847 init(const struct shash *iface_hints)
849 struct shash_node *node;
851 /* Make a local copy, since we don't own 'iface_hints' elements. */
852 SHASH_FOR_EACH(node, iface_hints) {
853 const struct iface_hint *orig_hint = node->data;
854 struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
856 new_hint->br_name = xstrdup(orig_hint->br_name);
857 new_hint->br_type = xstrdup(orig_hint->br_type);
858 new_hint->ofp_port = orig_hint->ofp_port;
860 shash_add(&init_ofp_ports, node->name, new_hint);
865 enumerate_types(struct sset *types)
867 dp_enumerate_types(types);
871 enumerate_names(const char *type, struct sset *names)
873 struct ofproto_dpif *ofproto;
876 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
877 if (strcmp(type, ofproto->up.type)) {
880 sset_add(names, ofproto->up.name);
887 del(const char *type, const char *name)
892 error = dpif_open(name, type, &dpif);
894 error = dpif_delete(dpif);
901 port_open_type(const char *datapath_type, const char *port_type)
903 return dpif_port_open_type(datapath_type, port_type);
906 /* Type functions. */
908 static struct ofproto_dpif *
909 lookup_ofproto_dpif_by_port_name(const char *name)
911 struct ofproto_dpif *ofproto;
913 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
914 if (sset_contains(&ofproto->ports, name)) {
923 type_run(const char *type)
925 static long long int push_timer = LLONG_MIN;
926 struct dpif_backer *backer;
930 backer = shash_find_data(&all_dpif_backers, type);
932 /* This is not necessarily a problem, since backers are only
933 * created on demand. */
937 dpif_run(backer->dpif);
939 /* The most natural place to push facet statistics is when they're pulled
940 * from the datapath. However, when there are many flows in the datapath,
941 * this expensive operation can occur so frequently, that it reduces our
942 * ability to quickly set up flows. To reduce the cost, we push statistics
944 if (time_msec() > push_timer) {
945 push_timer = time_msec() + 2000;
949 /* If vswitchd started with other_config:flow_restore_wait set as "true",
950 * and the configuration has now changed to "false", enable receiving
951 * packets from the datapath. */
952 if (!backer->recv_set_enable && !ofproto_get_flow_restore_wait()) {
953 backer->recv_set_enable = true;
955 error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
957 VLOG_ERR("Failed to enable receiving packets in dpif.");
960 dpif_flow_flush(backer->dpif);
961 backer->need_revalidate = REV_RECONFIGURE;
964 if (backer->need_revalidate
965 || !tag_set_is_empty(&backer->revalidate_set)) {
966 struct tag_set revalidate_set = backer->revalidate_set;
967 bool need_revalidate = backer->need_revalidate;
968 struct ofproto_dpif *ofproto;
969 struct simap_node *node;
970 struct simap tmp_backers;
972 /* Handle tunnel garbage collection. */
973 simap_init(&tmp_backers);
974 simap_swap(&backer->tnl_backers, &tmp_backers);
976 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
977 struct ofport_dpif *iter;
979 if (backer != ofproto->backer) {
983 HMAP_FOR_EACH (iter, up.hmap_node, &ofproto->up.ports) {
986 if (!iter->tnl_port) {
990 dp_port = netdev_vport_get_dpif_port(iter->up.netdev);
991 node = simap_find(&tmp_backers, dp_port);
993 simap_put(&backer->tnl_backers, dp_port, node->data);
994 simap_delete(&tmp_backers, node);
995 node = simap_find(&backer->tnl_backers, dp_port);
997 node = simap_find(&backer->tnl_backers, dp_port);
999 uint32_t odp_port = UINT32_MAX;
1001 if (!dpif_port_add(backer->dpif, iter->up.netdev,
1003 simap_put(&backer->tnl_backers, dp_port, odp_port);
1004 node = simap_find(&backer->tnl_backers, dp_port);
1009 iter->odp_port = node ? node->data : OVSP_NONE;
1010 if (tnl_port_reconfigure(&iter->up, iter->odp_port,
1012 backer->need_revalidate = REV_RECONFIGURE;
1017 SIMAP_FOR_EACH (node, &tmp_backers) {
1018 dpif_port_del(backer->dpif, node->data);
1020 simap_destroy(&tmp_backers);
1022 switch (backer->need_revalidate) {
1023 case REV_RECONFIGURE: COVERAGE_INC(rev_reconfigure); break;
1024 case REV_STP: COVERAGE_INC(rev_stp); break;
1025 case REV_PORT_TOGGLED: COVERAGE_INC(rev_port_toggled); break;
1026 case REV_FLOW_TABLE: COVERAGE_INC(rev_flow_table); break;
1027 case REV_INCONSISTENCY: COVERAGE_INC(rev_inconsistency); break;
1030 if (backer->need_revalidate) {
1031 /* Clear the drop_keys in case we should now be accepting some
1032 * formerly dropped flows. */
1033 drop_key_clear(backer);
1036 /* Clear the revalidation flags. */
1037 tag_set_init(&backer->revalidate_set);
1038 backer->need_revalidate = 0;
1040 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1041 struct facet *facet, *next;
1042 struct cls_cursor cursor;
1044 if (ofproto->backer != backer) {
1048 cls_cursor_init(&cursor, &ofproto->facets, NULL);
1049 CLS_CURSOR_FOR_EACH_SAFE (facet, next, cr, &cursor) {
1051 || tag_set_intersects(&revalidate_set, facet->xout.tags)) {
1052 facet_revalidate(facet);
1059 if (!backer->recv_set_enable) {
1060 /* Wake up before a max of 1000ms. */
1061 timer_set_duration(&backer->next_expiration, 1000);
1062 } else if (timer_expired(&backer->next_expiration)) {
1063 int delay = expire(backer);
1064 timer_set_duration(&backer->next_expiration, delay);
1067 /* Check for port changes in the dpif. */
1068 while ((error = dpif_port_poll(backer->dpif, &devname)) == 0) {
1069 struct ofproto_dpif *ofproto;
1070 struct dpif_port port;
1072 /* Don't report on the datapath's device. */
1073 if (!strcmp(devname, dpif_base_name(backer->dpif))) {
1077 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
1078 &all_ofproto_dpifs) {
1079 if (simap_contains(&ofproto->backer->tnl_backers, devname)) {
1084 ofproto = lookup_ofproto_dpif_by_port_name(devname);
1085 if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
1086 /* The port was removed. If we know the datapath,
1087 * report it through poll_set(). If we don't, it may be
1088 * notifying us of a removal we initiated, so ignore it.
1089 * If there's a pending ENOBUFS, let it stand, since
1090 * everything will be reevaluated. */
1091 if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
1092 sset_add(&ofproto->port_poll_set, devname);
1093 ofproto->port_poll_errno = 0;
1095 } else if (!ofproto) {
1096 /* The port was added, but we don't know with which
1097 * ofproto we should associate it. Delete it. */
1098 dpif_port_del(backer->dpif, port.port_no);
1100 dpif_port_destroy(&port);
1106 if (error != EAGAIN) {
1107 struct ofproto_dpif *ofproto;
1109 /* There was some sort of error, so propagate it to all
1110 * ofprotos that use this backer. */
1111 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
1112 &all_ofproto_dpifs) {
1113 if (ofproto->backer == backer) {
1114 sset_clear(&ofproto->port_poll_set);
1115 ofproto->port_poll_errno = error;
1124 dpif_backer_run_fast(struct dpif_backer *backer, int max_batch)
1128 /* If recv_set_enable is false, we should not handle upcalls. */
1129 if (!backer->recv_set_enable) {
1133 /* Handle one or more batches of upcalls, until there's nothing left to do
1134 * or until we do a fixed total amount of work.
1136 * We do work in batches because it can be much cheaper to set up a number
1137 * of flows and fire off their patches all at once. We do multiple batches
1138 * because in some cases handling a packet can cause another packet to be
1139 * queued almost immediately as part of the return flow. Both
1140 * optimizations can make major improvements on some benchmarks and
1141 * presumably for real traffic as well. */
1143 while (work < max_batch) {
1144 int retval = handle_upcalls(backer, max_batch - work);
1155 type_run_fast(const char *type)
1157 struct dpif_backer *backer;
1159 backer = shash_find_data(&all_dpif_backers, type);
1161 /* This is not necessarily a problem, since backers are only
1162 * created on demand. */
1166 return dpif_backer_run_fast(backer, FLOW_MISS_MAX_BATCH);
1172 static long long int port_rl = LLONG_MIN;
1173 static unsigned int backer_rl = 0;
1175 if (time_msec() >= port_rl) {
1176 struct ofproto_dpif *ofproto;
1177 struct ofport_dpif *ofport;
1179 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1181 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1182 port_run_fast(ofport);
1185 port_rl = time_msec() + 200;
1188 /* XXX: We have to be careful not to do too much work in this function. If
1189 * we call dpif_backer_run_fast() too often, or with too large a batch,
1190 * performance improves signifcantly, but at a cost. It's possible for the
1191 * number of flows in the datapath to increase without bound, and for poll
1192 * loops to take 10s of seconds. The correct solution to this problem,
1193 * long term, is to separate flow miss handling into it's own thread so it
1194 * isn't affected by revalidations, and expirations. Until then, this is
1195 * the best we can do. */
1196 if (++backer_rl >= 10) {
1197 struct shash_node *node;
1200 SHASH_FOR_EACH (node, &all_dpif_backers) {
1201 dpif_backer_run_fast(node->data, 1);
1207 type_wait(const char *type)
1209 struct dpif_backer *backer;
1211 backer = shash_find_data(&all_dpif_backers, type);
1213 /* This is not necessarily a problem, since backers are only
1214 * created on demand. */
1218 timer_wait(&backer->next_expiration);
1221 /* Basic life-cycle. */
1223 static int add_internal_flows(struct ofproto_dpif *);
1225 static struct ofproto *
1228 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
1229 return &ofproto->up;
1233 dealloc(struct ofproto *ofproto_)
1235 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1240 close_dpif_backer(struct dpif_backer *backer)
1242 struct shash_node *node;
1244 ovs_assert(backer->refcount > 0);
1246 if (--backer->refcount) {
1250 drop_key_clear(backer);
1251 hmap_destroy(&backer->drop_keys);
1253 simap_destroy(&backer->tnl_backers);
1254 hmap_destroy(&backer->odp_to_ofport_map);
1255 node = shash_find(&all_dpif_backers, backer->type);
1257 shash_delete(&all_dpif_backers, node);
1258 dpif_close(backer->dpif);
1263 /* Datapath port slated for removal from datapath. */
1264 struct odp_garbage {
1265 struct list list_node;
1270 open_dpif_backer(const char *type, struct dpif_backer **backerp)
1272 struct dpif_backer *backer;
1273 struct dpif_port_dump port_dump;
1274 struct dpif_port port;
1275 struct shash_node *node;
1276 struct list garbage_list;
1277 struct odp_garbage *garbage, *next;
1283 backer = shash_find_data(&all_dpif_backers, type);
1290 backer_name = xasprintf("ovs-%s", type);
1292 /* Remove any existing datapaths, since we assume we're the only
1293 * userspace controlling the datapath. */
1295 dp_enumerate_names(type, &names);
1296 SSET_FOR_EACH(name, &names) {
1297 struct dpif *old_dpif;
1299 /* Don't remove our backer if it exists. */
1300 if (!strcmp(name, backer_name)) {
1304 if (dpif_open(name, type, &old_dpif)) {
1305 VLOG_WARN("couldn't open old datapath %s to remove it", name);
1307 dpif_delete(old_dpif);
1308 dpif_close(old_dpif);
1311 sset_destroy(&names);
1313 backer = xmalloc(sizeof *backer);
1315 error = dpif_create_and_open(backer_name, type, &backer->dpif);
1318 VLOG_ERR("failed to open datapath of type %s: %s", type,
1324 backer->type = xstrdup(type);
1325 backer->refcount = 1;
1326 hmap_init(&backer->odp_to_ofport_map);
1327 hmap_init(&backer->drop_keys);
1328 timer_set_duration(&backer->next_expiration, 1000);
1329 backer->need_revalidate = 0;
1330 simap_init(&backer->tnl_backers);
1331 tag_set_init(&backer->revalidate_set);
1332 backer->recv_set_enable = !ofproto_get_flow_restore_wait();
1335 if (backer->recv_set_enable) {
1336 dpif_flow_flush(backer->dpif);
1339 /* Loop through the ports already on the datapath and remove any
1340 * that we don't need anymore. */
1341 list_init(&garbage_list);
1342 dpif_port_dump_start(&port_dump, backer->dpif);
1343 while (dpif_port_dump_next(&port_dump, &port)) {
1344 node = shash_find(&init_ofp_ports, port.name);
1345 if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
1346 garbage = xmalloc(sizeof *garbage);
1347 garbage->odp_port = port.port_no;
1348 list_push_front(&garbage_list, &garbage->list_node);
1351 dpif_port_dump_done(&port_dump);
1353 LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) {
1354 dpif_port_del(backer->dpif, garbage->odp_port);
1355 list_remove(&garbage->list_node);
1359 shash_add(&all_dpif_backers, type, backer);
1361 error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
1363 VLOG_ERR("failed to listen on datapath of type %s: %s",
1364 type, strerror(error));
1365 close_dpif_backer(backer);
1373 construct(struct ofproto *ofproto_)
1375 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1376 struct shash_node *node, *next;
1381 error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1386 max_ports = dpif_get_max_ports(ofproto->backer->dpif);
1387 ofproto_init_max_ports(ofproto_, MIN(max_ports, OFPP_MAX));
1389 ofproto->netflow = NULL;
1390 ofproto->sflow = NULL;
1391 ofproto->ipfix = NULL;
1392 ofproto->stp = NULL;
1393 hmap_init(&ofproto->bundles);
1394 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1395 for (i = 0; i < MAX_MIRRORS; i++) {
1396 ofproto->mirrors[i] = NULL;
1398 ofproto->has_bonded_bundles = false;
1400 classifier_init(&ofproto->facets);
1401 hmap_init(&ofproto->subfacets);
1402 ofproto->governor = NULL;
1403 ofproto->consistency_rl = LLONG_MIN;
1405 for (i = 0; i < N_TABLES; i++) {
1406 struct table_dpif *table = &ofproto->tables[i];
1408 table->catchall_table = NULL;
1409 table->other_table = NULL;
1410 table->basis = random_uint32();
1413 list_init(&ofproto->completions);
1415 ofproto_dpif_unixctl_init();
1417 ofproto->has_mirrors = false;
1418 ofproto->has_bundle_action = false;
1420 hmap_init(&ofproto->vlandev_map);
1421 hmap_init(&ofproto->realdev_vid_map);
1423 sset_init(&ofproto->ports);
1424 sset_init(&ofproto->ghost_ports);
1425 sset_init(&ofproto->port_poll_set);
1426 ofproto->port_poll_errno = 0;
1428 SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1429 struct iface_hint *iface_hint = node->data;
1431 if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1432 /* Check if the datapath already has this port. */
1433 if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1434 sset_add(&ofproto->ports, node->name);
1437 free(iface_hint->br_name);
1438 free(iface_hint->br_type);
1440 shash_delete(&init_ofp_ports, node);
1444 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1445 hash_string(ofproto->up.name, 0));
1446 memset(&ofproto->stats, 0, sizeof ofproto->stats);
1448 ofproto_init_tables(ofproto_, N_TABLES);
1449 error = add_internal_flows(ofproto);
1450 ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1453 ofproto->n_missed = 0;
1455 ofproto->max_n_subfacet = 0;
1456 ofproto->created = time_msec();
1457 ofproto->last_minute = ofproto->created;
1458 memset(&ofproto->hourly, 0, sizeof ofproto->hourly);
1459 memset(&ofproto->daily, 0, sizeof ofproto->daily);
1460 ofproto->subfacet_add_count = 0;
1461 ofproto->subfacet_del_count = 0;
1462 ofproto->total_subfacet_add_count = 0;
1463 ofproto->total_subfacet_del_count = 0;
1464 ofproto->total_subfacet_life_span = 0;
1465 ofproto->total_subfacet_count = 0;
1466 ofproto->n_update_stats = 0;
1472 add_internal_flow(struct ofproto_dpif *ofproto, int id,
1473 const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1475 struct ofputil_flow_mod fm;
1478 match_init_catchall(&fm.match);
1480 match_set_reg(&fm.match, 0, id);
1481 fm.new_cookie = htonll(0);
1482 fm.cookie = htonll(0);
1483 fm.cookie_mask = htonll(0);
1484 fm.table_id = TBL_INTERNAL;
1485 fm.command = OFPFC_ADD;
1486 fm.idle_timeout = 0;
1487 fm.hard_timeout = 0;
1491 fm.ofpacts = ofpacts->data;
1492 fm.ofpacts_len = ofpacts->size;
1494 error = ofproto_flow_mod(&ofproto->up, &fm);
1496 VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
1497 id, ofperr_to_string(error));
1501 *rulep = rule_dpif_lookup__(ofproto, &fm.match.flow, NULL, TBL_INTERNAL);
1502 ovs_assert(*rulep != NULL);
1508 add_internal_flows(struct ofproto_dpif *ofproto)
1510 struct ofpact_controller *controller;
1511 uint64_t ofpacts_stub[128 / 8];
1512 struct ofpbuf ofpacts;
1516 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1519 controller = ofpact_put_CONTROLLER(&ofpacts);
1520 controller->max_len = UINT16_MAX;
1521 controller->controller_id = 0;
1522 controller->reason = OFPR_NO_MATCH;
1523 ofpact_pad(&ofpacts);
1525 error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule);
1530 ofpbuf_clear(&ofpacts);
1531 error = add_internal_flow(ofproto, id++, &ofpacts,
1532 &ofproto->no_packet_in_rule);
1537 error = add_internal_flow(ofproto, id++, &ofpacts,
1538 &ofproto->drop_frags_rule);
1543 complete_operations(struct ofproto_dpif *ofproto)
1545 struct dpif_completion *c, *next;
1547 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
1548 ofoperation_complete(c->op, 0);
1549 list_remove(&c->list_node);
1555 destruct(struct ofproto *ofproto_)
1557 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1558 struct rule_dpif *rule, *next_rule;
1559 struct oftable *table;
1562 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1563 complete_operations(ofproto);
1565 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1566 struct cls_cursor cursor;
1568 cls_cursor_init(&cursor, &table->cls, NULL);
1569 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1570 ofproto_rule_destroy(&rule->up);
1574 for (i = 0; i < MAX_MIRRORS; i++) {
1575 mirror_destroy(ofproto->mirrors[i]);
1578 netflow_destroy(ofproto->netflow);
1579 dpif_sflow_destroy(ofproto->sflow);
1580 hmap_destroy(&ofproto->bundles);
1581 mac_learning_destroy(ofproto->ml);
1583 classifier_destroy(&ofproto->facets);
1584 hmap_destroy(&ofproto->subfacets);
1585 governor_destroy(ofproto->governor);
1587 hmap_destroy(&ofproto->vlandev_map);
1588 hmap_destroy(&ofproto->realdev_vid_map);
1590 sset_destroy(&ofproto->ports);
1591 sset_destroy(&ofproto->ghost_ports);
1592 sset_destroy(&ofproto->port_poll_set);
1594 close_dpif_backer(ofproto->backer);
1598 run_fast(struct ofproto *ofproto_)
1600 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1601 struct ofport_dpif *ofport;
1603 /* Do not perform any periodic activity required by 'ofproto' while
1604 * waiting for flow restore to complete. */
1605 if (ofproto_get_flow_restore_wait()) {
1609 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1610 port_run_fast(ofport);
1617 run(struct ofproto *ofproto_)
1619 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1620 struct ofport_dpif *ofport;
1621 struct ofbundle *bundle;
1625 complete_operations(ofproto);
1628 /* Do not perform any periodic activity below required by 'ofproto' while
1629 * waiting for flow restore to complete. */
1630 if (ofproto_get_flow_restore_wait()) {
1634 error = run_fast(ofproto_);
1639 if (ofproto->netflow) {
1640 if (netflow_run(ofproto->netflow)) {
1641 send_netflow_active_timeouts(ofproto);
1644 if (ofproto->sflow) {
1645 dpif_sflow_run(ofproto->sflow);
1648 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1651 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1656 mac_learning_run(ofproto->ml, &ofproto->backer->revalidate_set);
1658 /* Check the consistency of a random facet, to aid debugging. */
1659 if (time_msec() >= ofproto->consistency_rl
1660 && !classifier_is_empty(&ofproto->facets)
1661 && !ofproto->backer->need_revalidate) {
1662 struct cls_table *table;
1663 struct cls_rule *cr;
1664 struct facet *facet;
1666 ofproto->consistency_rl = time_msec() + 250;
1668 table = CONTAINER_OF(hmap_random_node(&ofproto->facets.tables),
1669 struct cls_table, hmap_node);
1670 cr = CONTAINER_OF(hmap_random_node(&table->rules), struct cls_rule,
1672 facet = CONTAINER_OF(cr, struct facet, cr);
1674 if (!tag_set_intersects(&ofproto->backer->revalidate_set,
1675 facet->xout.tags)) {
1676 if (!facet_check_consistency(facet)) {
1677 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
1682 if (ofproto->governor) {
1685 governor_run(ofproto->governor);
1687 /* If the governor has shrunk to its minimum size and the number of
1688 * subfacets has dwindled, then drop the governor entirely.
1690 * For hysteresis, the number of subfacets to drop the governor is
1691 * smaller than the number needed to trigger its creation. */
1692 n_subfacets = hmap_count(&ofproto->subfacets);
1693 if (n_subfacets * 4 < ofproto->up.flow_eviction_threshold
1694 && governor_is_idle(ofproto->governor)) {
1695 governor_destroy(ofproto->governor);
1696 ofproto->governor = NULL;
1704 wait(struct ofproto *ofproto_)
1706 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1707 struct ofport_dpif *ofport;
1708 struct ofbundle *bundle;
1710 if (!clogged && !list_is_empty(&ofproto->completions)) {
1711 poll_immediate_wake();
1714 if (ofproto_get_flow_restore_wait()) {
1718 dpif_wait(ofproto->backer->dpif);
1719 dpif_recv_wait(ofproto->backer->dpif);
1720 if (ofproto->sflow) {
1721 dpif_sflow_wait(ofproto->sflow);
1723 if (!tag_set_is_empty(&ofproto->backer->revalidate_set)) {
1724 poll_immediate_wake();
1726 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1729 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1730 bundle_wait(bundle);
1732 if (ofproto->netflow) {
1733 netflow_wait(ofproto->netflow);
1735 mac_learning_wait(ofproto->ml);
1737 if (ofproto->backer->need_revalidate) {
1738 /* Shouldn't happen, but if it does just go around again. */
1739 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1740 poll_immediate_wake();
1742 if (ofproto->governor) {
1743 governor_wait(ofproto->governor);
1748 get_memory_usage(const struct ofproto *ofproto_, struct simap *usage)
1750 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1752 simap_increase(usage, "facets", classifier_count(&ofproto->facets));
1753 simap_increase(usage, "subfacets", hmap_count(&ofproto->subfacets));
1757 flush(struct ofproto *ofproto_)
1759 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1760 struct subfacet *subfacet, *next_subfacet;
1761 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
1765 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
1766 &ofproto->subfacets) {
1767 if (subfacet->path != SF_NOT_INSTALLED) {
1768 batch[n_batch++] = subfacet;
1769 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
1770 subfacet_destroy_batch(ofproto, batch, n_batch);
1774 subfacet_destroy(subfacet);
1779 subfacet_destroy_batch(ofproto, batch, n_batch);
1784 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1785 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1787 *arp_match_ip = true;
1788 *actions = (OFPUTIL_A_OUTPUT |
1789 OFPUTIL_A_SET_VLAN_VID |
1790 OFPUTIL_A_SET_VLAN_PCP |
1791 OFPUTIL_A_STRIP_VLAN |
1792 OFPUTIL_A_SET_DL_SRC |
1793 OFPUTIL_A_SET_DL_DST |
1794 OFPUTIL_A_SET_NW_SRC |
1795 OFPUTIL_A_SET_NW_DST |
1796 OFPUTIL_A_SET_NW_TOS |
1797 OFPUTIL_A_SET_TP_SRC |
1798 OFPUTIL_A_SET_TP_DST |
1803 get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots)
1805 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1806 struct dpif_dp_stats s;
1807 uint64_t n_miss, n_no_pkt_in, n_bytes, n_dropped_frags;
1810 strcpy(ots->name, "classifier");
1812 dpif_get_dp_stats(ofproto->backer->dpif, &s);
1813 rule_get_stats(&ofproto->miss_rule->up, &n_miss, &n_bytes);
1814 rule_get_stats(&ofproto->no_packet_in_rule->up, &n_no_pkt_in, &n_bytes);
1815 rule_get_stats(&ofproto->drop_frags_rule->up, &n_dropped_frags, &n_bytes);
1817 n_lookup = s.n_hit + s.n_missed - n_dropped_frags;
1818 ots->lookup_count = htonll(n_lookup);
1819 ots->matched_count = htonll(n_lookup - n_miss - n_no_pkt_in);
1822 static struct ofport *
1825 struct ofport_dpif *port = xmalloc(sizeof *port);
1830 port_dealloc(struct ofport *port_)
1832 struct ofport_dpif *port = ofport_dpif_cast(port_);
1837 port_construct(struct ofport *port_)
1839 struct ofport_dpif *port = ofport_dpif_cast(port_);
1840 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1841 const struct netdev *netdev = port->up.netdev;
1842 struct dpif_port dpif_port;
1845 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1846 port->bundle = NULL;
1848 port->tag = tag_create_random();
1849 port->may_enable = true;
1850 port->stp_port = NULL;
1851 port->stp_state = STP_DISABLED;
1852 port->tnl_port = NULL;
1853 hmap_init(&port->priorities);
1854 port->realdev_ofp_port = 0;
1855 port->vlandev_vid = 0;
1856 port->carrier_seq = netdev_get_carrier_resets(netdev);
1858 if (netdev_vport_is_patch(netdev)) {
1859 /* By bailing out here, we don't submit the port to the sFlow module
1860 * to be considered for counter polling export. This is correct
1861 * because the patch port represents an interface that sFlow considers
1862 * to be "internal" to the switch as a whole, and therefore not an
1863 * candidate for counter polling. */
1864 port->odp_port = OVSP_NONE;
1868 error = dpif_port_query_by_name(ofproto->backer->dpif,
1869 netdev_vport_get_dpif_port(netdev),
1875 port->odp_port = dpif_port.port_no;
1877 if (netdev_get_tunnel_config(netdev)) {
1878 port->tnl_port = tnl_port_add(&port->up, port->odp_port);
1880 /* Sanity-check that a mapping doesn't already exist. This
1881 * shouldn't happen for non-tunnel ports. */
1882 if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1883 VLOG_ERR("port %s already has an OpenFlow port number",
1885 dpif_port_destroy(&dpif_port);
1889 hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1890 hash_int(port->odp_port, 0));
1892 dpif_port_destroy(&dpif_port);
1894 if (ofproto->sflow) {
1895 dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1902 port_destruct(struct ofport *port_)
1904 struct ofport_dpif *port = ofport_dpif_cast(port_);
1905 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1906 const char *dp_port_name = netdev_vport_get_dpif_port(port->up.netdev);
1907 const char *devname = netdev_get_name(port->up.netdev);
1909 if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1910 /* The underlying device is still there, so delete it. This
1911 * happens when the ofproto is being destroyed, since the caller
1912 * assumes that removal of attached ports will happen as part of
1914 if (!port->tnl_port) {
1915 dpif_port_del(ofproto->backer->dpif, port->odp_port);
1917 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1920 if (port->odp_port != OVSP_NONE && !port->tnl_port) {
1921 hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1924 tnl_port_del(port->tnl_port);
1925 sset_find_and_delete(&ofproto->ports, devname);
1926 sset_find_and_delete(&ofproto->ghost_ports, devname);
1927 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1928 bundle_remove(port_);
1929 set_cfm(port_, NULL);
1930 if (ofproto->sflow) {
1931 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1934 ofport_clear_priorities(port);
1935 hmap_destroy(&port->priorities);
1939 port_modified(struct ofport *port_)
1941 struct ofport_dpif *port = ofport_dpif_cast(port_);
1943 if (port->bundle && port->bundle->bond) {
1944 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1948 cfm_set_netdev(port->cfm, port->up.netdev);
1953 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1955 struct ofport_dpif *port = ofport_dpif_cast(port_);
1956 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1957 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1959 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1960 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1961 OFPUTIL_PC_NO_PACKET_IN)) {
1962 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1964 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1965 bundle_update(port->bundle);
1971 set_sflow(struct ofproto *ofproto_,
1972 const struct ofproto_sflow_options *sflow_options)
1974 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1975 struct dpif_sflow *ds = ofproto->sflow;
1977 if (sflow_options) {
1979 struct ofport_dpif *ofport;
1981 ds = ofproto->sflow = dpif_sflow_create();
1982 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1983 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1985 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1987 dpif_sflow_set_options(ds, sflow_options);
1990 dpif_sflow_destroy(ds);
1991 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1992 ofproto->sflow = NULL;
2000 struct ofproto *ofproto_,
2001 const struct ofproto_ipfix_bridge_exporter_options *bridge_exporter_options,
2002 const struct ofproto_ipfix_flow_exporter_options *flow_exporters_options,
2003 size_t n_flow_exporters_options)
2005 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2006 struct dpif_ipfix *di = ofproto->ipfix;
2008 if (bridge_exporter_options || flow_exporters_options) {
2010 di = ofproto->ipfix = dpif_ipfix_create();
2012 dpif_ipfix_set_options(
2013 di, bridge_exporter_options, flow_exporters_options,
2014 n_flow_exporters_options);
2017 dpif_ipfix_destroy(di);
2018 ofproto->ipfix = NULL;
2025 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
2027 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2034 struct ofproto_dpif *ofproto;
2036 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2037 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2038 ofport->cfm = cfm_create(ofport->up.netdev);
2041 if (cfm_configure(ofport->cfm, s)) {
2047 cfm_destroy(ofport->cfm);
2053 get_cfm_status(const struct ofport *ofport_,
2054 struct ofproto_cfm_status *status)
2056 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2059 status->faults = cfm_get_fault(ofport->cfm);
2060 status->remote_opstate = cfm_get_opup(ofport->cfm);
2061 status->health = cfm_get_health(ofport->cfm);
2062 cfm_get_remote_mpids(ofport->cfm, &status->rmps, &status->n_rmps);
2069 /* Spanning Tree. */
2072 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
2074 struct ofproto_dpif *ofproto = ofproto_;
2075 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
2076 struct ofport_dpif *ofport;
2078 ofport = stp_port_get_aux(sp);
2080 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
2081 ofproto->up.name, port_num);
2083 struct eth_header *eth = pkt->l2;
2085 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
2086 if (eth_addr_is_zero(eth->eth_src)) {
2087 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
2088 "with unknown MAC", ofproto->up.name, port_num);
2090 send_packet(ofport, pkt);
2096 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
2098 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
2100 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2102 /* Only revalidate flows if the configuration changed. */
2103 if (!s != !ofproto->stp) {
2104 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2108 if (!ofproto->stp) {
2109 ofproto->stp = stp_create(ofproto_->name, s->system_id,
2110 send_bpdu_cb, ofproto);
2111 ofproto->stp_last_tick = time_msec();
2114 stp_set_bridge_id(ofproto->stp, s->system_id);
2115 stp_set_bridge_priority(ofproto->stp, s->priority);
2116 stp_set_hello_time(ofproto->stp, s->hello_time);
2117 stp_set_max_age(ofproto->stp, s->max_age);
2118 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
2120 struct ofport *ofport;
2122 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
2123 set_stp_port(ofport, NULL);
2126 stp_destroy(ofproto->stp);
2127 ofproto->stp = NULL;
2134 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
2136 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2140 s->bridge_id = stp_get_bridge_id(ofproto->stp);
2141 s->designated_root = stp_get_designated_root(ofproto->stp);
2142 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
2151 update_stp_port_state(struct ofport_dpif *ofport)
2153 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2154 enum stp_state state;
2156 /* Figure out new state. */
2157 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
2161 if (ofport->stp_state != state) {
2162 enum ofputil_port_state of_state;
2165 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
2166 netdev_get_name(ofport->up.netdev),
2167 stp_state_name(ofport->stp_state),
2168 stp_state_name(state));
2169 if (stp_learn_in_state(ofport->stp_state)
2170 != stp_learn_in_state(state)) {
2171 /* xxx Learning action flows should also be flushed. */
2172 mac_learning_flush(ofproto->ml,
2173 &ofproto->backer->revalidate_set);
2175 fwd_change = stp_forward_in_state(ofport->stp_state)
2176 != stp_forward_in_state(state);
2178 ofproto->backer->need_revalidate = REV_STP;
2179 ofport->stp_state = state;
2180 ofport->stp_state_entered = time_msec();
2182 if (fwd_change && ofport->bundle) {
2183 bundle_update(ofport->bundle);
2186 /* Update the STP state bits in the OpenFlow port description. */
2187 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
2188 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
2189 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
2190 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
2191 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
2193 ofproto_port_set_state(&ofport->up, of_state);
2197 /* Configures STP on 'ofport_' using the settings defined in 's'. The
2198 * caller is responsible for assigning STP port numbers and ensuring
2199 * there are no duplicates. */
2201 set_stp_port(struct ofport *ofport_,
2202 const struct ofproto_port_stp_settings *s)
2204 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2205 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2206 struct stp_port *sp = ofport->stp_port;
2208 if (!s || !s->enable) {
2210 ofport->stp_port = NULL;
2211 stp_port_disable(sp);
2212 update_stp_port_state(ofport);
2215 } else if (sp && stp_port_no(sp) != s->port_num
2216 && ofport == stp_port_get_aux(sp)) {
2217 /* The port-id changed, so disable the old one if it's not
2218 * already in use by another port. */
2219 stp_port_disable(sp);
2222 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
2223 stp_port_enable(sp);
2225 stp_port_set_aux(sp, ofport);
2226 stp_port_set_priority(sp, s->priority);
2227 stp_port_set_path_cost(sp, s->path_cost);
2229 update_stp_port_state(ofport);
2235 get_stp_port_status(struct ofport *ofport_,
2236 struct ofproto_port_stp_status *s)
2238 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2239 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2240 struct stp_port *sp = ofport->stp_port;
2242 if (!ofproto->stp || !sp) {
2248 s->port_id = stp_port_get_id(sp);
2249 s->state = stp_port_get_state(sp);
2250 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
2251 s->role = stp_port_get_role(sp);
2252 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
2258 stp_run(struct ofproto_dpif *ofproto)
2261 long long int now = time_msec();
2262 long long int elapsed = now - ofproto->stp_last_tick;
2263 struct stp_port *sp;
2266 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
2267 ofproto->stp_last_tick = now;
2269 while (stp_get_changed_port(ofproto->stp, &sp)) {
2270 struct ofport_dpif *ofport = stp_port_get_aux(sp);
2273 update_stp_port_state(ofport);
2277 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
2278 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
2284 stp_wait(struct ofproto_dpif *ofproto)
2287 poll_timer_wait(1000);
2291 /* Returns true if STP should process 'flow'. Sets fields in 'wc' that
2292 * were used to make the determination.*/
2294 stp_should_process_flow(const struct flow *flow, struct flow_wildcards *wc)
2296 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
2297 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
2301 stp_process_packet(const struct ofport_dpif *ofport,
2302 const struct ofpbuf *packet)
2304 struct ofpbuf payload = *packet;
2305 struct eth_header *eth = payload.data;
2306 struct stp_port *sp = ofport->stp_port;
2308 /* Sink packets on ports that have STP disabled when the bridge has
2310 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
2314 /* Trim off padding on payload. */
2315 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
2316 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
2319 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
2320 stp_received_bpdu(sp, payload.data, payload.size);
2324 static struct priority_to_dscp *
2325 get_priority(const struct ofport_dpif *ofport, uint32_t priority)
2327 struct priority_to_dscp *pdscp;
2330 hash = hash_int(priority, 0);
2331 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) {
2332 if (pdscp->priority == priority) {
2340 ofport_clear_priorities(struct ofport_dpif *ofport)
2342 struct priority_to_dscp *pdscp, *next;
2344 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) {
2345 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
2351 set_queues(struct ofport *ofport_,
2352 const struct ofproto_port_queue *qdscp_list,
2355 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2356 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2357 struct hmap new = HMAP_INITIALIZER(&new);
2360 for (i = 0; i < n_qdscp; i++) {
2361 struct priority_to_dscp *pdscp;
2365 dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
2366 if (dpif_queue_to_priority(ofproto->backer->dpif, qdscp_list[i].queue,
2371 pdscp = get_priority(ofport, priority);
2373 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
2375 pdscp = xmalloc(sizeof *pdscp);
2376 pdscp->priority = priority;
2378 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2381 if (pdscp->dscp != dscp) {
2383 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2386 hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0));
2389 if (!hmap_is_empty(&ofport->priorities)) {
2390 ofport_clear_priorities(ofport);
2391 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2394 hmap_swap(&new, &ofport->priorities);
2402 /* Expires all MAC learning entries associated with 'bundle' and forces its
2403 * ofproto to revalidate every flow.
2405 * Normally MAC learning entries are removed only from the ofproto associated
2406 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2407 * are removed from every ofproto. When patch ports and SLB bonds are in use
2408 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2409 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2410 * with the host from which it migrated. */
2412 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2414 struct ofproto_dpif *ofproto = bundle->ofproto;
2415 struct mac_learning *ml = ofproto->ml;
2416 struct mac_entry *mac, *next_mac;
2418 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2419 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2420 if (mac->port.p == bundle) {
2422 struct ofproto_dpif *o;
2424 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2426 struct mac_entry *e;
2428 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan,
2431 mac_learning_expire(o->ml, e);
2437 mac_learning_expire(ml, mac);
2442 static struct ofbundle *
2443 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2445 struct ofbundle *bundle;
2447 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2448 &ofproto->bundles) {
2449 if (bundle->aux == aux) {
2456 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
2457 * ones that are found to 'bundles'. */
2459 bundle_lookup_multiple(struct ofproto_dpif *ofproto,
2460 void **auxes, size_t n_auxes,
2461 struct hmapx *bundles)
2465 hmapx_init(bundles);
2466 for (i = 0; i < n_auxes; i++) {
2467 struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
2469 hmapx_add(bundles, bundle);
2475 bundle_update(struct ofbundle *bundle)
2477 struct ofport_dpif *port;
2479 bundle->floodable = true;
2480 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2481 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2482 || !stp_forward_in_state(port->stp_state)) {
2483 bundle->floodable = false;
2490 bundle_del_port(struct ofport_dpif *port)
2492 struct ofbundle *bundle = port->bundle;
2494 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2496 list_remove(&port->bundle_node);
2497 port->bundle = NULL;
2500 lacp_slave_unregister(bundle->lacp, port);
2503 bond_slave_unregister(bundle->bond, port);
2506 bundle_update(bundle);
2510 bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port,
2511 struct lacp_slave_settings *lacp)
2513 struct ofport_dpif *port;
2515 port = get_ofp_port(bundle->ofproto, ofp_port);
2520 if (port->bundle != bundle) {
2521 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2523 bundle_del_port(port);
2526 port->bundle = bundle;
2527 list_push_back(&bundle->ports, &port->bundle_node);
2528 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2529 || !stp_forward_in_state(port->stp_state)) {
2530 bundle->floodable = false;
2534 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2535 lacp_slave_register(bundle->lacp, port, lacp);
2542 bundle_destroy(struct ofbundle *bundle)
2544 struct ofproto_dpif *ofproto;
2545 struct ofport_dpif *port, *next_port;
2552 ofproto = bundle->ofproto;
2553 for (i = 0; i < MAX_MIRRORS; i++) {
2554 struct ofmirror *m = ofproto->mirrors[i];
2556 if (m->out == bundle) {
2558 } else if (hmapx_find_and_delete(&m->srcs, bundle)
2559 || hmapx_find_and_delete(&m->dsts, bundle)) {
2560 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2565 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2566 bundle_del_port(port);
2569 bundle_flush_macs(bundle, true);
2570 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2572 free(bundle->trunks);
2573 lacp_destroy(bundle->lacp);
2574 bond_destroy(bundle->bond);
2579 bundle_set(struct ofproto *ofproto_, void *aux,
2580 const struct ofproto_bundle_settings *s)
2582 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2583 bool need_flush = false;
2584 struct ofport_dpif *port;
2585 struct ofbundle *bundle;
2586 unsigned long *trunks;
2592 bundle_destroy(bundle_lookup(ofproto, aux));
2596 ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2597 ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2599 bundle = bundle_lookup(ofproto, aux);
2601 bundle = xmalloc(sizeof *bundle);
2603 bundle->ofproto = ofproto;
2604 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2605 hash_pointer(aux, 0));
2607 bundle->name = NULL;
2609 list_init(&bundle->ports);
2610 bundle->vlan_mode = PORT_VLAN_TRUNK;
2612 bundle->trunks = NULL;
2613 bundle->use_priority_tags = s->use_priority_tags;
2614 bundle->lacp = NULL;
2615 bundle->bond = NULL;
2617 bundle->floodable = true;
2619 bundle->src_mirrors = 0;
2620 bundle->dst_mirrors = 0;
2621 bundle->mirror_out = 0;
2624 if (!bundle->name || strcmp(s->name, bundle->name)) {
2626 bundle->name = xstrdup(s->name);
2631 if (!bundle->lacp) {
2632 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2633 bundle->lacp = lacp_create();
2635 lacp_configure(bundle->lacp, s->lacp);
2637 lacp_destroy(bundle->lacp);
2638 bundle->lacp = NULL;
2641 /* Update set of ports. */
2643 for (i = 0; i < s->n_slaves; i++) {
2644 if (!bundle_add_port(bundle, s->slaves[i],
2645 s->lacp ? &s->lacp_slaves[i] : NULL)) {
2649 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2650 struct ofport_dpif *next_port;
2652 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2653 for (i = 0; i < s->n_slaves; i++) {
2654 if (s->slaves[i] == port->up.ofp_port) {
2659 bundle_del_port(port);
2663 ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2665 if (list_is_empty(&bundle->ports)) {
2666 bundle_destroy(bundle);
2670 /* Set VLAN tagging mode */
2671 if (s->vlan_mode != bundle->vlan_mode
2672 || s->use_priority_tags != bundle->use_priority_tags) {
2673 bundle->vlan_mode = s->vlan_mode;
2674 bundle->use_priority_tags = s->use_priority_tags;
2679 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2680 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2682 if (vlan != bundle->vlan) {
2683 bundle->vlan = vlan;
2687 /* Get trunked VLANs. */
2688 switch (s->vlan_mode) {
2689 case PORT_VLAN_ACCESS:
2693 case PORT_VLAN_TRUNK:
2694 trunks = CONST_CAST(unsigned long *, s->trunks);
2697 case PORT_VLAN_NATIVE_UNTAGGED:
2698 case PORT_VLAN_NATIVE_TAGGED:
2699 if (vlan != 0 && (!s->trunks
2700 || !bitmap_is_set(s->trunks, vlan)
2701 || bitmap_is_set(s->trunks, 0))) {
2702 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2704 trunks = bitmap_clone(s->trunks, 4096);
2706 trunks = bitmap_allocate1(4096);
2708 bitmap_set1(trunks, vlan);
2709 bitmap_set0(trunks, 0);
2711 trunks = CONST_CAST(unsigned long *, s->trunks);
2718 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2719 free(bundle->trunks);
2720 if (trunks == s->trunks) {
2721 bundle->trunks = vlan_bitmap_clone(trunks);
2723 bundle->trunks = trunks;
2728 if (trunks != s->trunks) {
2733 if (!list_is_short(&bundle->ports)) {
2734 bundle->ofproto->has_bonded_bundles = true;
2736 if (bond_reconfigure(bundle->bond, s->bond)) {
2737 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2740 bundle->bond = bond_create(s->bond);
2741 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2744 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2745 bond_slave_register(bundle->bond, port, port->up.netdev);
2748 bond_destroy(bundle->bond);
2749 bundle->bond = NULL;
2752 /* If we changed something that would affect MAC learning, un-learn
2753 * everything on this port and force flow revalidation. */
2755 bundle_flush_macs(bundle, false);
2762 bundle_remove(struct ofport *port_)
2764 struct ofport_dpif *port = ofport_dpif_cast(port_);
2765 struct ofbundle *bundle = port->bundle;
2768 bundle_del_port(port);
2769 if (list_is_empty(&bundle->ports)) {
2770 bundle_destroy(bundle);
2771 } else if (list_is_short(&bundle->ports)) {
2772 bond_destroy(bundle->bond);
2773 bundle->bond = NULL;
2779 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2781 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2782 struct ofport_dpif *port = port_;
2783 uint8_t ea[ETH_ADDR_LEN];
2786 error = netdev_get_etheraddr(port->up.netdev, ea);
2788 struct ofpbuf packet;
2791 ofpbuf_init(&packet, 0);
2792 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2794 memcpy(packet_pdu, pdu, pdu_size);
2796 send_packet(port, &packet);
2797 ofpbuf_uninit(&packet);
2799 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2800 "%s (%s)", port->bundle->name,
2801 netdev_get_name(port->up.netdev), strerror(error));
2806 bundle_send_learning_packets(struct ofbundle *bundle)
2808 struct ofproto_dpif *ofproto = bundle->ofproto;
2809 int error, n_packets, n_errors;
2810 struct mac_entry *e;
2812 error = n_packets = n_errors = 0;
2813 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2814 if (e->port.p != bundle) {
2815 struct ofpbuf *learning_packet;
2816 struct ofport_dpif *port;
2820 /* The assignment to "port" is unnecessary but makes "grep"ing for
2821 * struct ofport_dpif more effective. */
2822 learning_packet = bond_compose_learning_packet(bundle->bond,
2826 ret = send_packet(port, learning_packet);
2827 ofpbuf_delete(learning_packet);
2837 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2838 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2839 "packets, last error was: %s",
2840 bundle->name, n_errors, n_packets, strerror(error));
2842 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2843 bundle->name, n_packets);
2848 bundle_run(struct ofbundle *bundle)
2851 lacp_run(bundle->lacp, send_pdu_cb);
2854 struct ofport_dpif *port;
2856 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2857 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2860 bond_run(bundle->bond, &bundle->ofproto->backer->revalidate_set,
2861 lacp_status(bundle->lacp));
2862 if (bond_should_send_learning_packets(bundle->bond)) {
2863 bundle_send_learning_packets(bundle);
2869 bundle_wait(struct ofbundle *bundle)
2872 lacp_wait(bundle->lacp);
2875 bond_wait(bundle->bond);
2882 mirror_scan(struct ofproto_dpif *ofproto)
2886 for (idx = 0; idx < MAX_MIRRORS; idx++) {
2887 if (!ofproto->mirrors[idx]) {
2894 static struct ofmirror *
2895 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
2899 for (i = 0; i < MAX_MIRRORS; i++) {
2900 struct ofmirror *mirror = ofproto->mirrors[i];
2901 if (mirror && mirror->aux == aux) {
2909 /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */
2911 mirror_update_dups(struct ofproto_dpif *ofproto)
2915 for (i = 0; i < MAX_MIRRORS; i++) {
2916 struct ofmirror *m = ofproto->mirrors[i];
2919 m->dup_mirrors = MIRROR_MASK_C(1) << i;
2923 for (i = 0; i < MAX_MIRRORS; i++) {
2924 struct ofmirror *m1 = ofproto->mirrors[i];
2931 for (j = i + 1; j < MAX_MIRRORS; j++) {
2932 struct ofmirror *m2 = ofproto->mirrors[j];
2934 if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) {
2935 m1->dup_mirrors |= MIRROR_MASK_C(1) << j;
2936 m2->dup_mirrors |= m1->dup_mirrors;
2943 mirror_set(struct ofproto *ofproto_, void *aux,
2944 const struct ofproto_mirror_settings *s)
2946 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2947 mirror_mask_t mirror_bit;
2948 struct ofbundle *bundle;
2949 struct ofmirror *mirror;
2950 struct ofbundle *out;
2951 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
2952 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
2955 mirror = mirror_lookup(ofproto, aux);
2957 mirror_destroy(mirror);
2963 idx = mirror_scan(ofproto);
2965 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
2967 ofproto->up.name, MAX_MIRRORS, s->name);
2971 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
2972 mirror->ofproto = ofproto;
2975 mirror->out_vlan = -1;
2976 mirror->name = NULL;
2979 if (!mirror->name || strcmp(s->name, mirror->name)) {
2981 mirror->name = xstrdup(s->name);
2984 /* Get the new configuration. */
2985 if (s->out_bundle) {
2986 out = bundle_lookup(ofproto, s->out_bundle);
2988 mirror_destroy(mirror);
2994 out_vlan = s->out_vlan;
2996 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
2997 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
2999 /* If the configuration has not changed, do nothing. */
3000 if (hmapx_equals(&srcs, &mirror->srcs)
3001 && hmapx_equals(&dsts, &mirror->dsts)
3002 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
3003 && mirror->out == out
3004 && mirror->out_vlan == out_vlan)
3006 hmapx_destroy(&srcs);
3007 hmapx_destroy(&dsts);
3011 hmapx_swap(&srcs, &mirror->srcs);
3012 hmapx_destroy(&srcs);
3014 hmapx_swap(&dsts, &mirror->dsts);
3015 hmapx_destroy(&dsts);
3017 free(mirror->vlans);
3018 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
3021 mirror->out_vlan = out_vlan;
3023 /* Update bundles. */
3024 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
3025 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
3026 if (hmapx_contains(&mirror->srcs, bundle)) {
3027 bundle->src_mirrors |= mirror_bit;
3029 bundle->src_mirrors &= ~mirror_bit;
3032 if (hmapx_contains(&mirror->dsts, bundle)) {
3033 bundle->dst_mirrors |= mirror_bit;
3035 bundle->dst_mirrors &= ~mirror_bit;
3038 if (mirror->out == bundle) {
3039 bundle->mirror_out |= mirror_bit;
3041 bundle->mirror_out &= ~mirror_bit;
3045 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3046 ofproto->has_mirrors = true;
3047 mac_learning_flush(ofproto->ml,
3048 &ofproto->backer->revalidate_set);
3049 mirror_update_dups(ofproto);
3055 mirror_destroy(struct ofmirror *mirror)
3057 struct ofproto_dpif *ofproto;
3058 mirror_mask_t mirror_bit;
3059 struct ofbundle *bundle;
3066 ofproto = mirror->ofproto;
3067 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3068 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
3070 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
3071 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
3072 bundle->src_mirrors &= ~mirror_bit;
3073 bundle->dst_mirrors &= ~mirror_bit;
3074 bundle->mirror_out &= ~mirror_bit;
3077 hmapx_destroy(&mirror->srcs);
3078 hmapx_destroy(&mirror->dsts);
3079 free(mirror->vlans);
3081 ofproto->mirrors[mirror->idx] = NULL;
3085 mirror_update_dups(ofproto);
3087 ofproto->has_mirrors = false;
3088 for (i = 0; i < MAX_MIRRORS; i++) {
3089 if (ofproto->mirrors[i]) {
3090 ofproto->has_mirrors = true;
3097 mirror_get_stats(struct ofproto *ofproto_, void *aux,
3098 uint64_t *packets, uint64_t *bytes)
3100 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3101 struct ofmirror *mirror = mirror_lookup(ofproto, aux);
3104 *packets = *bytes = UINT64_MAX;
3110 *packets = mirror->packet_count;
3111 *bytes = mirror->byte_count;
3117 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
3119 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3120 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
3121 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
3127 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
3129 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3130 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
3131 return bundle && bundle->mirror_out != 0;
3135 forward_bpdu_changed(struct ofproto *ofproto_)
3137 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3138 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3142 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
3145 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3146 mac_learning_set_idle_time(ofproto->ml, idle_time);
3147 mac_learning_set_max_entries(ofproto->ml, max_entries);
3152 static struct ofport_dpif *
3153 get_ofp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
3155 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
3156 return ofport ? ofport_dpif_cast(ofport) : NULL;
3159 static struct ofport_dpif *
3160 get_odp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
3162 struct ofport_dpif *port = odp_port_to_ofport(ofproto->backer, odp_port);
3163 return port && &ofproto->up == port->up.ofproto ? port : NULL;
3167 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
3168 struct ofproto_port *ofproto_port,
3169 struct dpif_port *dpif_port)
3171 ofproto_port->name = dpif_port->name;
3172 ofproto_port->type = dpif_port->type;
3173 ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
3176 static struct ofport_dpif *
3177 ofport_get_peer(const struct ofport_dpif *ofport_dpif)
3179 const struct ofproto_dpif *ofproto;
3182 peer = netdev_vport_patch_peer(ofport_dpif->up.netdev);
3187 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
3188 struct ofport *ofport;
3190 ofport = shash_find_data(&ofproto->up.port_by_name, peer);
3191 if (ofport && ofport->ofproto->ofproto_class == &ofproto_dpif_class) {
3192 return ofport_dpif_cast(ofport);
3199 port_run_fast(struct ofport_dpif *ofport)
3201 if (ofport->cfm && cfm_should_send_ccm(ofport->cfm)) {
3202 struct ofpbuf packet;
3204 ofpbuf_init(&packet, 0);
3205 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
3206 send_packet(ofport, &packet);
3207 ofpbuf_uninit(&packet);
3212 port_run(struct ofport_dpif *ofport)
3214 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
3215 bool carrier_changed = carrier_seq != ofport->carrier_seq;
3216 bool enable = netdev_get_carrier(ofport->up.netdev);
3218 ofport->carrier_seq = carrier_seq;
3220 port_run_fast(ofport);
3222 if (ofport->tnl_port
3223 && tnl_port_reconfigure(&ofport->up, ofport->odp_port,
3224 &ofport->tnl_port)) {
3225 ofproto_dpif_cast(ofport->up.ofproto)->backer->need_revalidate = true;
3229 int cfm_opup = cfm_get_opup(ofport->cfm);
3231 cfm_run(ofport->cfm);
3232 enable = enable && !cfm_get_fault(ofport->cfm);
3234 if (cfm_opup >= 0) {
3235 enable = enable && cfm_opup;
3239 if (ofport->bundle) {
3240 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
3241 if (carrier_changed) {
3242 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
3246 if (ofport->may_enable != enable) {
3247 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3249 if (ofproto->has_bundle_action) {
3250 ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
3254 ofport->may_enable = enable;
3258 port_wait(struct ofport_dpif *ofport)
3261 cfm_wait(ofport->cfm);
3266 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
3267 struct ofproto_port *ofproto_port)
3269 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3270 struct dpif_port dpif_port;
3273 if (sset_contains(&ofproto->ghost_ports, devname)) {
3274 const char *type = netdev_get_type_from_name(devname);
3276 /* We may be called before ofproto->up.port_by_name is populated with
3277 * the appropriate ofport. For this reason, we must get the name and
3278 * type from the netdev layer directly. */
3280 const struct ofport *ofport;
3282 ofport = shash_find_data(&ofproto->up.port_by_name, devname);
3283 ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
3284 ofproto_port->name = xstrdup(devname);
3285 ofproto_port->type = xstrdup(type);
3291 if (!sset_contains(&ofproto->ports, devname)) {
3294 error = dpif_port_query_by_name(ofproto->backer->dpif,
3295 devname, &dpif_port);
3297 ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
3303 port_add(struct ofproto *ofproto_, struct netdev *netdev)
3305 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3306 const char *dp_port_name = netdev_vport_get_dpif_port(netdev);
3307 const char *devname = netdev_get_name(netdev);
3309 if (netdev_vport_is_patch(netdev)) {
3310 sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
3314 if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
3315 uint32_t port_no = UINT32_MAX;
3318 error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
3322 if (netdev_get_tunnel_config(netdev)) {
3323 simap_put(&ofproto->backer->tnl_backers, dp_port_name, port_no);
3327 if (netdev_get_tunnel_config(netdev)) {
3328 sset_add(&ofproto->ghost_ports, devname);
3330 sset_add(&ofproto->ports, devname);
3336 port_del(struct ofproto *ofproto_, uint16_t ofp_port)
3338 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3339 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
3346 sset_find_and_delete(&ofproto->ghost_ports,
3347 netdev_get_name(ofport->up.netdev));
3348 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3349 if (!ofport->tnl_port) {
3350 error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
3352 /* The caller is going to close ofport->up.netdev. If this is a
3353 * bonded port, then the bond is using that netdev, so remove it
3354 * from the bond. The client will need to reconfigure everything
3355 * after deleting ports, so then the slave will get re-added. */
3356 bundle_remove(&ofport->up);
3363 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
3365 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3370 error = netdev_get_stats(ofport->up.netdev, stats);
3372 if (!error && ofport_->ofp_port == OFPP_LOCAL) {
3373 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3375 /* ofproto->stats.tx_packets represents packets that we created
3376 * internally and sent to some port (e.g. packets sent with
3377 * send_packet()). Account for them as if they had come from
3378 * OFPP_LOCAL and got forwarded. */
3380 if (stats->rx_packets != UINT64_MAX) {
3381 stats->rx_packets += ofproto->stats.tx_packets;
3384 if (stats->rx_bytes != UINT64_MAX) {
3385 stats->rx_bytes += ofproto->stats.tx_bytes;
3388 /* ofproto->stats.rx_packets represents packets that were received on
3389 * some port and we processed internally and dropped (e.g. STP).
3390 * Account for them as if they had been forwarded to OFPP_LOCAL. */
3392 if (stats->tx_packets != UINT64_MAX) {
3393 stats->tx_packets += ofproto->stats.rx_packets;
3396 if (stats->tx_bytes != UINT64_MAX) {
3397 stats->tx_bytes += ofproto->stats.rx_bytes;
3404 /* Account packets for LOCAL port. */
3406 ofproto_update_local_port_stats(const struct ofproto *ofproto_,
3407 size_t tx_size, size_t rx_size)
3409 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3412 ofproto->stats.rx_packets++;
3413 ofproto->stats.rx_bytes += rx_size;
3416 ofproto->stats.tx_packets++;
3417 ofproto->stats.tx_bytes += tx_size;
3421 struct port_dump_state {
3426 struct ofproto_port port;
3431 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
3433 *statep = xzalloc(sizeof(struct port_dump_state));
3438 port_dump_next(const struct ofproto *ofproto_, void *state_,
3439 struct ofproto_port *port)
3441 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3442 struct port_dump_state *state = state_;
3443 const struct sset *sset;
3444 struct sset_node *node;
3446 if (state->has_port) {
3447 ofproto_port_destroy(&state->port);
3448 state->has_port = false;
3450 sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
3451 while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
3454 error = port_query_by_name(ofproto_, node->name, &state->port);
3456 *port = state->port;
3457 state->has_port = true;
3459 } else if (error != ENODEV) {
3464 if (!state->ghost) {
3465 state->ghost = true;
3468 return port_dump_next(ofproto_, state_, port);
3475 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3477 struct port_dump_state *state = state_;
3479 if (state->has_port) {
3480 ofproto_port_destroy(&state->port);
3487 port_poll(const struct ofproto *ofproto_, char **devnamep)
3489 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3491 if (ofproto->port_poll_errno) {
3492 int error = ofproto->port_poll_errno;
3493 ofproto->port_poll_errno = 0;
3497 if (sset_is_empty(&ofproto->port_poll_set)) {
3501 *devnamep = sset_pop(&ofproto->port_poll_set);
3506 port_poll_wait(const struct ofproto *ofproto_)
3508 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3509 dpif_port_poll_wait(ofproto->backer->dpif);
3513 port_is_lacp_current(const struct ofport *ofport_)
3515 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3516 return (ofport->bundle && ofport->bundle->lacp
3517 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3521 /* Upcall handling. */
3523 /* Flow miss batching.
3525 * Some dpifs implement operations faster when you hand them off in a batch.
3526 * To allow batching, "struct flow_miss" queues the dpif-related work needed
3527 * for a given flow. Each "struct flow_miss" corresponds to sending one or
3528 * more packets, plus possibly installing the flow in the dpif.
3530 * So far we only batch the operations that affect flow setup time the most.
3531 * It's possible to batch more than that, but the benefit might be minimal. */
3533 struct hmap_node hmap_node;
3534 struct ofproto_dpif *ofproto;
3536 enum odp_key_fitness key_fitness;
3537 const struct nlattr *key;
3539 struct initial_vals initial_vals;
3540 struct list packets;
3541 enum dpif_upcall_type upcall_type;
3542 uint32_t odp_in_port;
3545 struct flow_miss_op {
3546 struct dpif_op dpif_op;
3548 uint64_t slow_stub[128 / 8]; /* Buffer for compose_slow_path() */
3549 struct xlate_out xout;
3550 bool xout_garbage; /* 'xout' needs to be uninitialized? */
3553 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
3554 * OpenFlow controller as necessary according to their individual
3555 * configurations. */
3557 send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet,
3558 const struct flow *flow)
3560 struct ofputil_packet_in pin;
3562 pin.packet = packet->data;
3563 pin.packet_len = packet->size;
3564 pin.reason = OFPR_NO_MATCH;
3565 pin.controller_id = 0;
3570 pin.send_len = 0; /* not used for flow table misses */
3572 flow_get_metadata(flow, &pin.fmd);
3574 connmgr_send_packet_in(ofproto->up.connmgr, &pin);
3577 static enum slow_path_reason
3578 process_special(struct xlate_ctx *ctx, const struct flow *flow,
3579 const struct ofport_dpif *ofport, const struct ofpbuf *packet)
3581 struct ofproto_dpif *ofproto = ctx->ofproto;
3582 struct flow_wildcards *wc = &ctx->xout->wc;
3586 } else if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow, wc)) {
3588 cfm_process_heartbeat(ofport->cfm, packet);
3591 } else if (ofport->bundle && ofport->bundle->lacp
3592 && flow->dl_type == htons(ETH_TYPE_LACP)) {
3594 lacp_process_packet(ofport->bundle->lacp, ofport, packet);
3597 } else if (ofproto->stp && stp_should_process_flow(flow, wc)) {
3599 stp_process_packet(ofport, packet);
3607 static struct flow_miss *
3608 flow_miss_find(struct hmap *todo, const struct ofproto_dpif *ofproto,
3609 const struct flow *flow, uint32_t hash)
3611 struct flow_miss *miss;
3613 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
3614 if (miss->ofproto == ofproto && flow_equal(&miss->flow, flow)) {
3622 /* Partially Initializes 'op' as an "execute" operation for 'miss' and
3623 * 'packet'. The caller must initialize op->actions and op->actions_len. If
3624 * 'miss' is associated with a subfacet the caller must also initialize the
3625 * returned op->subfacet, and if anything needs to be freed after processing
3626 * the op, the caller must initialize op->garbage also. */
3628 init_flow_miss_execute_op(struct flow_miss *miss, struct ofpbuf *packet,
3629 struct flow_miss_op *op)
3631 if (miss->flow.vlan_tci != miss->initial_vals.vlan_tci) {
3632 /* This packet was received on a VLAN splinter port. We
3633 * added a VLAN to the packet to make the packet resemble
3634 * the flow, but the actions were composed assuming that
3635 * the packet contained no VLAN. So, we must remove the
3636 * VLAN header from the packet before trying to execute the
3638 eth_pop_vlan(packet);
3641 op->xout_garbage = false;
3642 op->dpif_op.type = DPIF_OP_EXECUTE;
3643 op->dpif_op.u.execute.key = miss->key;
3644 op->dpif_op.u.execute.key_len = miss->key_len;
3645 op->dpif_op.u.execute.packet = packet;
3648 /* Helper for handle_flow_miss_without_facet() and
3649 * handle_flow_miss_with_facet(). */
3651 handle_flow_miss_common(struct rule_dpif *rule,
3652 struct ofpbuf *packet, const struct flow *flow)
3654 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3656 if (rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
3658 * Extra-special case for fail-open mode.
3660 * We are in fail-open mode and the packet matched the fail-open
3661 * rule, but we are connected to a controller too. We should send
3662 * the packet up to the controller in the hope that it will try to
3663 * set up a flow and thereby allow us to exit fail-open.
3665 * See the top-level comment in fail-open.c for more information.
3667 send_packet_in_miss(ofproto, packet, flow);
3671 /* Figures out whether a flow that missed in 'ofproto', whose details are in
3672 * 'miss' masked by 'wc', is likely to be worth tracking in detail in userspace
3673 * and (usually) installing a datapath flow. The answer is usually "yes" (a
3674 * return value of true). However, for short flows the cost of bookkeeping is
3675 * much higher than the benefits, so when the datapath holds a large number of
3676 * flows we impose some heuristics to decide which flows are likely to be worth
3679 flow_miss_should_make_facet(struct flow_miss *miss, struct flow_wildcards *wc)
3681 struct ofproto_dpif *ofproto = miss->ofproto;
3684 if (!ofproto->governor) {
3687 n_subfacets = hmap_count(&ofproto->subfacets);
3688 if (n_subfacets * 2 <= ofproto->up.flow_eviction_threshold) {
3692 ofproto->governor = governor_create(ofproto->up.name);
3695 hash = flow_hash_in_wildcards(&miss->flow, wc, 0);
3696 return governor_should_install_flow(ofproto->governor, hash,
3697 list_size(&miss->packets));
3700 /* Handles 'miss' without creating a facet or subfacet or creating any datapath
3701 * flow. 'miss->flow' must have matched 'rule' and been xlated into 'xout'.
3702 * May add an "execute" operation to 'ops' and increment '*n_ops'. */
3704 handle_flow_miss_without_facet(struct rule_dpif *rule, struct xlate_out *xout,
3705 struct flow_miss *miss,
3706 struct flow_miss_op *ops, size_t *n_ops)
3708 struct ofpbuf *packet;
3710 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3712 COVERAGE_INC(facet_suppress);
3714 handle_flow_miss_common(rule, packet, &miss->flow);
3717 struct xlate_in xin;
3719 xlate_in_init(&xin, miss->ofproto, &miss->flow,
3720 &miss->initial_vals, rule, 0, packet);
3721 xlate_actions_for_side_effects(&xin);
3724 if (xout->odp_actions.size) {
3725 struct flow_miss_op *op = &ops[*n_ops];
3726 struct dpif_execute *execute = &op->dpif_op.u.execute;
3728 init_flow_miss_execute_op(miss, packet, op);
3729 xlate_out_copy(&op->xout, xout);
3730 execute->actions = op->xout.odp_actions.data;
3731 execute->actions_len = op->xout.odp_actions.size;
3732 op->xout_garbage = true;
3739 /* Handles 'miss', which matches 'facet'. May add any required datapath
3740 * operations to 'ops', incrementing '*n_ops' for each new op.
3742 * All of the packets in 'miss' are considered to have arrived at time 'now'.
3743 * This is really important only for new facets: if we just called time_msec()
3744 * here, then the new subfacet or its packets could look (occasionally) as
3745 * though it was used some time after the facet was used. That can make a
3746 * one-packet flow look like it has a nonzero duration, which looks odd in
3747 * e.g. NetFlow statistics.
3749 * If non-null, 'stats' will be folded into 'facet'. */
3751 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
3752 long long int now, struct dpif_flow_stats *stats,
3753 struct flow_miss_op *ops, size_t *n_ops)
3755 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3756 enum subfacet_path want_path;
3757 struct subfacet *subfacet;
3758 struct ofpbuf *packet;
3760 subfacet = subfacet_create(facet, miss, now);
3761 want_path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
3763 subfacet_update_stats(subfacet, stats);
3766 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3767 struct flow_miss_op *op = &ops[*n_ops];
3769 handle_flow_miss_common(facet->rule, packet, &miss->flow);
3771 if (want_path != SF_FAST_PATH) {
3772 struct xlate_in xin;
3774 xlate_in_init(&xin, ofproto, &miss->flow, &miss->initial_vals,
3775 facet->rule, 0, packet);
3776 xlate_actions_for_side_effects(&xin);
3779 if (facet->xout.odp_actions.size) {
3780 struct dpif_execute *execute = &op->dpif_op.u.execute;
3782 init_flow_miss_execute_op(miss, packet, op);
3783 execute->actions = facet->xout.odp_actions.data,
3784 execute->actions_len = facet->xout.odp_actions.size;
3789 if (miss->upcall_type == DPIF_UC_MISS || subfacet->path != want_path) {
3790 struct flow_miss_op *op = &ops[(*n_ops)++];
3791 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
3793 subfacet->path = want_path;
3795 op->xout_garbage = false;
3796 op->dpif_op.type = DPIF_OP_FLOW_PUT;
3797 put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
3798 put->key = miss->key;
3799 put->key_len = miss->key_len;
3800 if (want_path == SF_FAST_PATH) {
3801 put->actions = facet->xout.odp_actions.data;
3802 put->actions_len = facet->xout.odp_actions.size;
3804 compose_slow_path(ofproto, &miss->flow, facet->xout.slow,
3805 op->slow_stub, sizeof op->slow_stub,
3806 &put->actions, &put->actions_len);
3812 /* Handles flow miss 'miss'. May add any required datapath operations
3813 * to 'ops', incrementing '*n_ops' for each new op. */
3815 handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops,
3818 struct ofproto_dpif *ofproto = miss->ofproto;
3819 struct dpif_flow_stats stats__;
3820 struct dpif_flow_stats *stats = &stats__;
3821 struct ofpbuf *packet;
3822 struct facet *facet;
3826 memset(stats, 0, sizeof *stats);
3828 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3829 stats->tcp_flags |= packet_get_tcp_flags(packet, &miss->flow);
3830 stats->n_bytes += packet->size;
3834 facet = facet_lookup_valid(ofproto, &miss->flow);
3836 struct flow_wildcards wc;
3837 struct rule_dpif *rule;
3838 struct xlate_out xout;
3839 struct xlate_in xin;
3841 flow_wildcards_init_catchall(&wc);
3842 rule = rule_dpif_lookup(ofproto, &miss->flow, &wc);
3843 rule_credit_stats(rule, stats);
3845 xlate_in_init(&xin, ofproto, &miss->flow, &miss->initial_vals, rule,
3846 stats->tcp_flags, NULL);
3847 xin.resubmit_stats = stats;
3848 xin.may_learn = true;
3849 xlate_actions(&xin, &xout);
3850 flow_wildcards_or(&xout.wc, &xout.wc, &wc);
3852 /* There does not exist a bijection between 'struct flow' and datapath
3853 * flow keys with fitness ODP_FIT_TO_LITTLE. This breaks a fundamental
3854 * assumption used throughout the facet and subfacet handling code.
3855 * Since we have to handle these misses in userspace anyway, we simply
3856 * skip facet creation, avoiding the problem altogether. */
3857 if (miss->key_fitness == ODP_FIT_TOO_LITTLE
3858 || !flow_miss_should_make_facet(miss, &xout.wc)) {
3859 handle_flow_miss_without_facet(rule, &xout, miss, ops, n_ops);
3863 facet = facet_create(miss, rule, &xout, stats);
3866 handle_flow_miss_with_facet(miss, facet, now, stats, ops, n_ops);
3869 static struct drop_key *
3870 drop_key_lookup(const struct dpif_backer *backer, const struct nlattr *key,
3873 struct drop_key *drop_key;
3875 HMAP_FOR_EACH_WITH_HASH (drop_key, hmap_node, hash_bytes(key, key_len, 0),
3876 &backer->drop_keys) {
3877 if (drop_key->key_len == key_len
3878 && !memcmp(drop_key->key, key, key_len)) {
3886 drop_key_clear(struct dpif_backer *backer)
3888 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3889 struct drop_key *drop_key, *next;
3891 HMAP_FOR_EACH_SAFE (drop_key, next, hmap_node, &backer->drop_keys) {
3894 error = dpif_flow_del(backer->dpif, drop_key->key, drop_key->key_len,
3896 if (error && !VLOG_DROP_WARN(&rl)) {
3897 struct ds ds = DS_EMPTY_INITIALIZER;
3898 odp_flow_key_format(drop_key->key, drop_key->key_len, &ds);
3899 VLOG_WARN("Failed to delete drop key (%s) (%s)", strerror(error),
3904 hmap_remove(&backer->drop_keys, &drop_key->hmap_node);
3905 free(drop_key->key);
3910 /* Given a datpath, packet, and flow metadata ('backer', 'packet', and 'key'
3911 * respectively), populates 'flow' with the result of odp_flow_key_to_flow().
3912 * Optionally, if nonnull, populates 'fitnessp' with the fitness of 'flow' as
3913 * returned by odp_flow_key_to_flow(). Also, optionally populates 'ofproto'
3914 * with the ofproto_dpif, and 'odp_in_port' with the datapath in_port, that
3915 * 'packet' ingressed.
3917 * If 'ofproto' is nonnull, requires 'flow''s in_port to exist. Otherwise sets
3918 * 'flow''s in_port to OFPP_NONE.
3920 * This function does post-processing on data returned from
3921 * odp_flow_key_to_flow() to help make VLAN splinters transparent to the rest
3922 * of the upcall processing logic. In particular, if the extracted in_port is
3923 * a VLAN splinter port, it replaces flow->in_port by the "real" port, sets
3924 * flow->vlan_tci correctly for the VLAN of the VLAN splinter port, and pushes
3925 * a VLAN header onto 'packet' (if it is nonnull).
3927 * Optionally, if 'initial_vals' is nonnull, sets 'initial_vals->vlan_tci'
3928 * to the VLAN TCI with which the packet was really received, that is, the
3929 * actual VLAN TCI extracted by odp_flow_key_to_flow(). (This differs from
3930 * the value returned in flow->vlan_tci only for packets received on
3931 * VLAN splinters.) Also, if received on an IP tunnel, sets
3932 * 'initial_vals->tunnel_ip_tos' to the tunnel's IP TOS.
3934 * Similarly, this function also includes some logic to help with tunnels. It
3935 * may modify 'flow' as necessary to make the tunneling implementation
3936 * transparent to the upcall processing logic.
3938 * Returns 0 if successful, ENODEV if the parsed flow has no associated ofport,
3939 * or some other positive errno if there are other problems. */
3941 ofproto_receive(const struct dpif_backer *backer, struct ofpbuf *packet,
3942 const struct nlattr *key, size_t key_len,
3943 struct flow *flow, enum odp_key_fitness *fitnessp,
3944 struct ofproto_dpif **ofproto, uint32_t *odp_in_port,
3945 struct initial_vals *initial_vals)
3947 const struct ofport_dpif *port;
3948 enum odp_key_fitness fitness;
3951 fitness = odp_flow_key_to_flow(key, key_len, flow);
3952 if (fitness == ODP_FIT_ERROR) {
3958 initial_vals->vlan_tci = flow->vlan_tci;
3959 initial_vals->tunnel_ip_tos = flow->tunnel.ip_tos;
3963 *odp_in_port = flow->in_port;
3966 if (tnl_port_should_receive(flow)) {
3967 const struct ofport *ofport = tnl_port_receive(flow);
3969 flow->in_port = OFPP_NONE;
3972 port = ofport_dpif_cast(ofport);
3974 /* We can't reproduce 'key' from 'flow'. */
3975 fitness = fitness == ODP_FIT_PERFECT ? ODP_FIT_TOO_MUCH : fitness;
3977 /* XXX: Since the tunnel module is not scoped per backer, it's
3978 * theoretically possible that we'll receive an ofport belonging to an
3979 * entirely different datapath. In practice, this can't happen because
3980 * no platforms has two separate datapaths which each support
3982 ovs_assert(ofproto_dpif_cast(port->up.ofproto)->backer == backer);
3984 port = odp_port_to_ofport(backer, flow->in_port);
3986 flow->in_port = OFPP_NONE;
3990 flow->in_port = port->up.ofp_port;
3991 if (vsp_adjust_flow(ofproto_dpif_cast(port->up.ofproto), flow)) {
3993 /* Make the packet resemble the flow, so that it gets sent to
3994 * an OpenFlow controller properly, so that it looks correct
3995 * for sFlow, and so that flow_extract() will get the correct
3996 * vlan_tci if it is called on 'packet'.
3998 * The allocated space inside 'packet' probably also contains
3999 * 'key', that is, both 'packet' and 'key' are probably part of
4000 * a struct dpif_upcall (see the large comment on that
4001 * structure definition), so pushing data on 'packet' is in
4002 * general not a good idea since it could overwrite 'key' or
4003 * free it as a side effect. However, it's OK in this special
4004 * case because we know that 'packet' is inside a Netlink
4005 * attribute: pushing 4 bytes will just overwrite the 4-byte
4006 * "struct nlattr", which is fine since we don't need that
4007 * header anymore. */
4008 eth_push_vlan(packet, flow->vlan_tci);
4010 /* We can't reproduce 'key' from 'flow'. */
4011 fitness = fitness == ODP_FIT_PERFECT ? ODP_FIT_TOO_MUCH : fitness;
4017 *ofproto = ofproto_dpif_cast(port->up.ofproto);
4022 *fitnessp = fitness;
4028 handle_miss_upcalls(struct dpif_backer *backer, struct dpif_upcall *upcalls,
4031 struct dpif_upcall *upcall;
4032 struct flow_miss *miss;
4033 struct flow_miss misses[FLOW_MISS_MAX_BATCH];
4034 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
4035 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
4045 /* Construct the to-do list.
4047 * This just amounts to extracting the flow from each packet and sticking
4048 * the packets that have the same flow in the same "flow_miss" structure so
4049 * that we can process them together. */
4052 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
4053 struct flow_miss *miss = &misses[n_misses];
4054 struct flow_miss *existing_miss;
4055 struct ofproto_dpif *ofproto;
4056 uint32_t odp_in_port;
4061 error = ofproto_receive(backer, upcall->packet, upcall->key,
4062 upcall->key_len, &flow, &miss->key_fitness,
4063 &ofproto, &odp_in_port, &miss->initial_vals);
4064 if (error == ENODEV) {
4065 struct drop_key *drop_key;
4067 /* Received packet on port for which we couldn't associate
4068 * an ofproto. This can happen if a port is removed while
4069 * traffic is being received. Print a rate-limited message
4070 * in case it happens frequently. Install a drop flow so
4071 * that future packets of the flow are inexpensively dropped
4073 VLOG_INFO_RL(&rl, "received packet on unassociated port %"PRIu32,
4076 drop_key = drop_key_lookup(backer, upcall->key, upcall->key_len);
4078 drop_key = xmalloc(sizeof *drop_key);
4079 drop_key->key = xmemdup(upcall->key, upcall->key_len);
4080 drop_key->key_len = upcall->key_len;
4082 hmap_insert(&backer->drop_keys, &drop_key->hmap_node,
4083 hash_bytes(drop_key->key, drop_key->key_len, 0));
4084 dpif_flow_put(backer->dpif, DPIF_FP_CREATE | DPIF_FP_MODIFY,
4085 drop_key->key, drop_key->key_len, NULL, 0, NULL);
4093 ofproto->n_missed++;
4094 flow_extract(upcall->packet, flow.skb_priority, flow.skb_mark,
4095 &flow.tunnel, flow.in_port, &miss->flow);
4097 /* Add other packets to a to-do list. */
4098 hash = flow_hash(&miss->flow, 0);
4099 existing_miss = flow_miss_find(&todo, ofproto, &miss->flow, hash);
4100 if (!existing_miss) {
4101 hmap_insert(&todo, &miss->hmap_node, hash);
4102 miss->ofproto = ofproto;
4103 miss->key = upcall->key;
4104 miss->key_len = upcall->key_len;
4105 miss->upcall_type = upcall->type;
4106 miss->odp_in_port = odp_in_port;
4107 list_init(&miss->packets);
4111 miss = existing_miss;
4113 list_push_back(&miss->packets, &upcall->packet->list_node);
4116 /* Process each element in the to-do list, constructing the set of
4117 * operations to batch. */
4119 HMAP_FOR_EACH (miss, hmap_node, &todo) {
4120 handle_flow_miss(miss, flow_miss_ops, &n_ops);
4122 ovs_assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
4124 /* Execute batch. */
4125 for (i = 0; i < n_ops; i++) {
4126 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
4128 dpif_operate(backer->dpif, dpif_ops, n_ops);
4131 for (i = 0; i < n_ops; i++) {
4132 if (flow_miss_ops[i].xout_garbage) {
4133 xlate_out_uninit(&flow_miss_ops[i].xout);
4136 hmap_destroy(&todo);
4139 static enum { SFLOW_UPCALL, MISS_UPCALL, BAD_UPCALL, FLOW_SAMPLE_UPCALL,
4141 classify_upcall(const struct dpif_upcall *upcall)
4143 size_t userdata_len;
4144 union user_action_cookie cookie;
4146 /* First look at the upcall type. */
4147 switch (upcall->type) {
4148 case DPIF_UC_ACTION:
4154 case DPIF_N_UC_TYPES:
4156 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
4160 /* "action" upcalls need a closer look. */
4161 if (!upcall->userdata) {
4162 VLOG_WARN_RL(&rl, "action upcall missing cookie");
4165 userdata_len = nl_attr_get_size(upcall->userdata);
4166 if (userdata_len < sizeof cookie.type
4167 || userdata_len > sizeof cookie) {
4168 VLOG_WARN_RL(&rl, "action upcall cookie has unexpected size %zu",
4172 memset(&cookie, 0, sizeof cookie);
4173 memcpy(&cookie, nl_attr_get(upcall->userdata), userdata_len);
4174 if (userdata_len == sizeof cookie.sflow
4175 && cookie.type == USER_ACTION_COOKIE_SFLOW) {
4176 return SFLOW_UPCALL;
4177 } else if (userdata_len == sizeof cookie.slow_path
4178 && cookie.type == USER_ACTION_COOKIE_SLOW_PATH) {
4180 } else if (userdata_len == sizeof cookie.flow_sample
4181 && cookie.type == USER_ACTION_COOKIE_FLOW_SAMPLE) {
4182 return FLOW_SAMPLE_UPCALL;
4183 } else if (userdata_len == sizeof cookie.ipfix
4184 && cookie.type == USER_ACTION_COOKIE_IPFIX) {
4185 return IPFIX_UPCALL;
4187 VLOG_WARN_RL(&rl, "invalid user cookie of type %"PRIu16
4188 " and size %zu", cookie.type, userdata_len);
4194 handle_sflow_upcall(struct dpif_backer *backer,
4195 const struct dpif_upcall *upcall)
4197 struct ofproto_dpif *ofproto;
4198 union user_action_cookie cookie;
4200 uint32_t odp_in_port;
4202 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
4203 &flow, NULL, &ofproto, &odp_in_port, NULL)
4204 || !ofproto->sflow) {
4208 memset(&cookie, 0, sizeof cookie);
4209 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.sflow);
4210 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
4211 odp_in_port, &cookie);
4215 handle_flow_sample_upcall(struct dpif_backer *backer,
4216 const struct dpif_upcall *upcall)
4218 struct ofproto_dpif *ofproto;
4219 union user_action_cookie cookie;
4222 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
4223 &flow, NULL, &ofproto, NULL, NULL)
4224 || !ofproto->ipfix) {
4228 memset(&cookie, 0, sizeof cookie);
4229 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.flow_sample);
4231 /* The flow reflects exactly the contents of the packet. Sample
4232 * the packet using it. */
4233 dpif_ipfix_flow_sample(ofproto->ipfix, upcall->packet, &flow,
4234 cookie.flow_sample.collector_set_id,
4235 cookie.flow_sample.probability,
4236 cookie.flow_sample.obs_domain_id,
4237 cookie.flow_sample.obs_point_id);
4241 handle_ipfix_upcall(struct dpif_backer *backer,
4242 const struct dpif_upcall *upcall)
4244 struct ofproto_dpif *ofproto;
4247 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
4248 &flow, NULL, &ofproto, NULL, NULL)
4249 || !ofproto->ipfix) {
4253 /* The flow reflects exactly the contents of the packet. Sample
4254 * the packet using it. */
4255 dpif_ipfix_bridge_sample(ofproto->ipfix, upcall->packet, &flow);
4259 handle_upcalls(struct dpif_backer *backer, unsigned int max_batch)
4261 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
4262 struct ofpbuf miss_bufs[FLOW_MISS_MAX_BATCH];
4263 uint64_t miss_buf_stubs[FLOW_MISS_MAX_BATCH][4096 / 8];
4268 ovs_assert(max_batch <= FLOW_MISS_MAX_BATCH);
4271 for (n_processed = 0; n_processed < max_batch; n_processed++) {
4272 struct dpif_upcall *upcall = &misses[n_misses];
4273 struct ofpbuf *buf = &miss_bufs[n_misses];
4276 ofpbuf_use_stub(buf, miss_buf_stubs[n_misses],
4277 sizeof miss_buf_stubs[n_misses]);
4278 error = dpif_recv(backer->dpif, upcall, buf);
4284 switch (classify_upcall(upcall)) {
4286 /* Handle it later. */
4291 handle_sflow_upcall(backer, upcall);
4295 case FLOW_SAMPLE_UPCALL:
4296 handle_flow_sample_upcall(backer, upcall);
4301 handle_ipfix_upcall(backer, upcall);
4311 /* Handle deferred MISS_UPCALL processing. */
4312 handle_miss_upcalls(backer, misses, n_misses);
4313 for (i = 0; i < n_misses; i++) {
4314 ofpbuf_uninit(&miss_bufs[i]);
4320 /* Flow expiration. */
4322 static int subfacet_max_idle(const struct ofproto_dpif *);
4323 static void update_stats(struct dpif_backer *);
4324 static void rule_expire(struct rule_dpif *);
4325 static void expire_subfacets(struct ofproto_dpif *, int dp_max_idle);
4327 /* This function is called periodically by run(). Its job is to collect
4328 * updates for the flows that have been installed into the datapath, most
4329 * importantly when they last were used, and then use that information to
4330 * expire flows that have not been used recently.
4332 * Returns the number of milliseconds after which it should be called again. */
4334 expire(struct dpif_backer *backer)
4336 struct ofproto_dpif *ofproto;
4337 int max_idle = INT32_MAX;
4339 /* Periodically clear out the drop keys in an effort to keep them
4340 * relatively few. */
4341 drop_key_clear(backer);
4343 /* Update stats for each flow in the backer. */
4344 update_stats(backer);
4346 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4347 struct rule *rule, *next_rule;
4350 if (ofproto->backer != backer) {
4354 /* Keep track of the max number of flows per ofproto_dpif. */
4355 update_max_subfacet_count(ofproto);
4357 /* Expire subfacets that have been idle too long. */
4358 dp_max_idle = subfacet_max_idle(ofproto);
4359 expire_subfacets(ofproto, dp_max_idle);
4361 max_idle = MIN(max_idle, dp_max_idle);
4363 /* Expire OpenFlow flows whose idle_timeout or hard_timeout
4365 LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
4366 &ofproto->up.expirable) {
4367 rule_expire(rule_dpif_cast(rule));
4370 /* All outstanding data in existing flows has been accounted, so it's a
4371 * good time to do bond rebalancing. */
4372 if (ofproto->has_bonded_bundles) {
4373 struct ofbundle *bundle;
4375 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
4377 bond_rebalance(bundle->bond, &backer->revalidate_set);
4383 return MIN(max_idle, 1000);
4386 /* Updates flow table statistics given that the datapath just reported 'stats'
4387 * as 'subfacet''s statistics. */
4389 update_subfacet_stats(struct subfacet *subfacet,
4390 const struct dpif_flow_stats *stats)
4392 struct facet *facet = subfacet->facet;
4393 struct dpif_flow_stats diff;
4395 diff.tcp_flags = stats->tcp_flags;
4396 diff.used = stats->used;
4398 if (stats->n_packets >= subfacet->dp_packet_count) {
4399 diff.n_packets = stats->n_packets - subfacet->dp_packet_count;
4401 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
4405 if (stats->n_bytes >= subfacet->dp_byte_count) {
4406 diff.n_bytes = stats->n_bytes - subfacet->dp_byte_count;
4408 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
4412 subfacet->dp_packet_count = stats->n_packets;
4413 subfacet->dp_byte_count = stats->n_bytes;
4414 subfacet_update_stats(subfacet, &diff);
4416 if (facet->accounted_bytes < facet->byte_count) {
4418 facet_account(facet);
4419 facet->accounted_bytes = facet->byte_count;
4423 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
4424 * about, or a flow that shouldn't be installed but was anyway. Delete it. */
4426 delete_unexpected_flow(struct ofproto_dpif *ofproto,
4427 const struct nlattr *key, size_t key_len)
4429 if (!VLOG_DROP_WARN(&rl)) {
4433 odp_flow_key_format(key, key_len, &s);
4434 VLOG_WARN("unexpected flow on %s: %s", ofproto->up.name, ds_cstr(&s));
4438 COVERAGE_INC(facet_unexpected);
4439 dpif_flow_del(ofproto->backer->dpif, key, key_len, NULL);
4442 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
4444 * This function also pushes statistics updates to rules which each facet
4445 * resubmits into. Generally these statistics will be accurate. However, if a
4446 * facet changes the rule it resubmits into at some time in between
4447 * update_stats() runs, it is possible that statistics accrued to the
4448 * old rule will be incorrectly attributed to the new rule. This could be
4449 * avoided by calling update_stats() whenever rules are created or
4450 * deleted. However, the performance impact of making so many calls to the
4451 * datapath do not justify the benefit of having perfectly accurate statistics.
4453 * In addition, this function maintains per ofproto flow hit counts. The patch
4454 * port is not treated specially. e.g. A packet ingress from br0 patched into
4455 * br1 will increase the hit count of br0 by 1, however, does not affect
4456 * the hit or miss counts of br1.
4459 update_stats(struct dpif_backer *backer)
4461 const struct dpif_flow_stats *stats;
4462 struct dpif_flow_dump dump;
4463 const struct nlattr *key;
4464 struct ofproto_dpif *ofproto;
4467 dpif_flow_dump_start(&dump, backer->dpif);
4468 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
4470 struct subfacet *subfacet;
4473 if (ofproto_receive(backer, NULL, key, key_len, &flow, NULL, &ofproto,
4478 ofproto->total_subfacet_count += hmap_count(&ofproto->subfacets);
4479 ofproto->n_update_stats++;
4481 key_hash = odp_flow_key_hash(key, key_len);
4482 subfacet = subfacet_find(ofproto, key, key_len, key_hash);
4483 switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
4485 /* Update ofproto_dpif's hit count. */
4486 if (stats->n_packets > subfacet->dp_packet_count) {
4487 uint64_t delta = stats->n_packets - subfacet->dp_packet_count;
4488 dpif_stats_update_hit_count(ofproto, delta);
4491 update_subfacet_stats(subfacet, stats);
4495 /* Stats are updated per-packet. */
4498 case SF_NOT_INSTALLED:
4500 delete_unexpected_flow(ofproto, key, key_len);
4505 dpif_flow_dump_done(&dump);
4507 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4508 update_moving_averages(ofproto);
4513 /* Calculates and returns the number of milliseconds of idle time after which
4514 * subfacets should expire from the datapath. When a subfacet expires, we fold
4515 * its statistics into its facet, and when a facet's last subfacet expires, we
4516 * fold its statistic into its rule. */
4518 subfacet_max_idle(const struct ofproto_dpif *ofproto)
4521 * Idle time histogram.
4523 * Most of the time a switch has a relatively small number of subfacets.
4524 * When this is the case we might as well keep statistics for all of them
4525 * in userspace and to cache them in the kernel datapath for performance as
4528 * As the number of subfacets increases, the memory required to maintain
4529 * statistics about them in userspace and in the kernel becomes
4530 * significant. However, with a large number of subfacets it is likely
4531 * that only a few of them are "heavy hitters" that consume a large amount
4532 * of bandwidth. At this point, only heavy hitters are worth caching in
4533 * the kernel and maintaining in userspaces; other subfacets we can
4536 * The technique used to compute the idle time is to build a histogram with
4537 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
4538 * that is installed in the kernel gets dropped in the appropriate bucket.
4539 * After the histogram has been built, we compute the cutoff so that only
4540 * the most-recently-used 1% of subfacets (but at least
4541 * ofproto->up.flow_eviction_threshold flows) are kept cached. At least
4542 * the most-recently-used bucket of subfacets is kept, so actually an
4543 * arbitrary number of subfacets can be kept in any given expiration run
4544 * (though the next run will delete most of those unless they receive
4547 * This requires a second pass through the subfacets, in addition to the
4548 * pass made by update_stats(), because the former function never looks at
4549 * uninstallable subfacets.
4551 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
4552 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
4553 int buckets[N_BUCKETS] = { 0 };
4554 int total, subtotal, bucket;
4555 struct subfacet *subfacet;
4559 total = hmap_count(&ofproto->subfacets);
4560 if (total <= ofproto->up.flow_eviction_threshold) {
4561 return N_BUCKETS * BUCKET_WIDTH;
4564 /* Build histogram. */
4566 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
4567 long long int idle = now - subfacet->used;
4568 int bucket = (idle <= 0 ? 0
4569 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
4570 : (unsigned int) idle / BUCKET_WIDTH);
4574 /* Find the first bucket whose flows should be expired. */
4575 subtotal = bucket = 0;
4577 subtotal += buckets[bucket++];
4578 } while (bucket < N_BUCKETS &&
4579 subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100));
4581 if (VLOG_IS_DBG_ENABLED()) {
4585 ds_put_cstr(&s, "keep");
4586 for (i = 0; i < N_BUCKETS; i++) {
4588 ds_put_cstr(&s, ", drop");
4591 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
4594 VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s));
4598 return bucket * BUCKET_WIDTH;
4602 expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle)
4604 /* Cutoff time for most flows. */
4605 long long int normal_cutoff = time_msec() - dp_max_idle;
4607 /* We really want to keep flows for special protocols around, so use a more
4608 * conservative cutoff. */
4609 long long int special_cutoff = time_msec() - 10000;
4611 struct subfacet *subfacet, *next_subfacet;
4612 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
4616 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
4617 &ofproto->subfacets) {
4618 long long int cutoff;
4620 cutoff = (subfacet->facet->xout.slow & (SLOW_CFM | SLOW_LACP
4624 if (subfacet->used < cutoff) {
4625 if (subfacet->path != SF_NOT_INSTALLED) {
4626 batch[n_batch++] = subfacet;
4627 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
4628 subfacet_destroy_batch(ofproto, batch, n_batch);
4632 subfacet_destroy(subfacet);
4638 subfacet_destroy_batch(ofproto, batch, n_batch);
4642 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
4643 * then delete it entirely. */
4645 rule_expire(struct rule_dpif *rule)
4647 struct facet *facet, *next_facet;
4651 if (rule->up.pending) {
4652 /* We'll have to expire it later. */
4656 /* Has 'rule' expired? */
4658 if (rule->up.hard_timeout
4659 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
4660 reason = OFPRR_HARD_TIMEOUT;
4661 } else if (rule->up.idle_timeout
4662 && now > rule->up.used + rule->up.idle_timeout * 1000) {
4663 reason = OFPRR_IDLE_TIMEOUT;
4668 COVERAGE_INC(ofproto_dpif_expired);
4670 /* Update stats. (This is a no-op if the rule expired due to an idle
4671 * timeout, because that only happens when the rule has no facets left.) */
4672 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4673 facet_remove(facet);
4676 /* Get rid of the rule. */
4677 ofproto_rule_expire(&rule->up, reason);
4682 /* Creates and returns a new facet based on 'miss'.
4684 * The caller must already have determined that no facet with an identical
4685 * 'miss->flow' exists in 'miss->ofproto'.
4687 * 'rule' and 'xout' must have been created based on 'miss'.
4689 * 'facet'' statistics are initialized based on 'stats'.
4691 * The facet will initially have no subfacets. The caller should create (at
4692 * least) one subfacet with subfacet_create(). */
4693 static struct facet *
4694 facet_create(const struct flow_miss *miss, struct rule_dpif *rule,
4695 struct xlate_out *xout, struct dpif_flow_stats *stats)
4697 struct ofproto_dpif *ofproto = miss->ofproto;
4698 struct facet *facet;
4701 facet = xzalloc(sizeof *facet);
4702 facet->packet_count = facet->prev_packet_count = stats->n_packets;
4703 facet->byte_count = facet->prev_byte_count = stats->n_bytes;
4704 facet->tcp_flags = stats->tcp_flags;
4705 facet->used = stats->used;
4706 facet->flow = miss->flow;
4707 facet->initial_vals = miss->initial_vals;
4708 facet->learn_rl = time_msec() + 500;
4711 list_push_back(&facet->rule->facets, &facet->list_node);
4712 list_init(&facet->subfacets);
4713 netflow_flow_init(&facet->nf_flow);
4714 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
4716 xlate_out_copy(&facet->xout, xout);
4718 match_init(&match, &facet->flow, &facet->xout.wc);
4719 cls_rule_init(&facet->cr, &match, OFP_DEFAULT_PRIORITY);
4720 classifier_insert(&ofproto->facets, &facet->cr);
4722 facet->nf_flow.output_iface = facet->xout.nf_output_iface;
4728 facet_free(struct facet *facet)
4731 xlate_out_uninit(&facet->xout);
4736 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
4737 * 'packet', which arrived on 'in_port'. */
4739 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
4740 const struct nlattr *odp_actions, size_t actions_len,
4741 struct ofpbuf *packet)
4743 struct odputil_keybuf keybuf;
4747 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4748 odp_flow_key_from_flow(&key, flow,
4749 ofp_port_to_odp_port(ofproto, flow->in_port));
4751 error = dpif_execute(ofproto->backer->dpif, key.data, key.size,
4752 odp_actions, actions_len, packet);
4756 /* Remove 'facet' from its ofproto and free up the associated memory:
4758 * - If 'facet' was installed in the datapath, uninstalls it and updates its
4759 * rule's statistics, via subfacet_uninstall().
4761 * - Removes 'facet' from its rule and from ofproto->facets.
4764 facet_remove(struct facet *facet)
4766 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4767 struct subfacet *subfacet, *next_subfacet;
4769 ovs_assert(!list_is_empty(&facet->subfacets));
4771 /* First uninstall all of the subfacets to get final statistics. */
4772 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4773 subfacet_uninstall(subfacet);
4776 /* Flush the final stats to the rule.
4778 * This might require us to have at least one subfacet around so that we
4779 * can use its actions for accounting in facet_account(), which is why we
4780 * have uninstalled but not yet destroyed the subfacets. */
4781 facet_flush_stats(facet);
4783 /* Now we're really all done so destroy everything. */
4784 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
4785 &facet->subfacets) {
4786 subfacet_destroy__(subfacet);
4788 classifier_remove(&ofproto->facets, &facet->cr);
4789 cls_rule_destroy(&facet->cr);
4790 list_remove(&facet->list_node);
4794 /* Feed information from 'facet' back into the learning table to keep it in
4795 * sync with what is actually flowing through the datapath. */
4797 facet_learn(struct facet *facet)
4799 long long int now = time_msec();
4801 if (!facet->xout.has_fin_timeout && now < facet->learn_rl) {
4805 facet->learn_rl = now + 500;
4807 if (!facet->xout.has_learn
4808 && !facet->xout.has_normal
4809 && (!facet->xout.has_fin_timeout
4810 || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
4814 facet_push_stats(facet, true);
4818 facet_account(struct facet *facet)
4820 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4821 const struct nlattr *a;
4826 if (!facet->xout.has_normal || !ofproto->has_bonded_bundles) {
4829 n_bytes = facet->byte_count - facet->accounted_bytes;
4831 /* This loop feeds byte counters to bond_account() for rebalancing to use
4832 * as a basis. We also need to track the actual VLAN on which the packet
4833 * is going to be sent to ensure that it matches the one passed to
4834 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
4837 * We use the actions from an arbitrary subfacet because they should all
4838 * be equally valid for our purpose. */
4839 vlan_tci = facet->flow.vlan_tci;
4840 NL_ATTR_FOR_EACH_UNSAFE (a, left, facet->xout.odp_actions.data,
4841 facet->xout.odp_actions.size) {
4842 const struct ovs_action_push_vlan *vlan;
4843 struct ofport_dpif *port;
4845 switch (nl_attr_type(a)) {
4846 case OVS_ACTION_ATTR_OUTPUT:
4847 port = get_odp_port(ofproto, nl_attr_get_u32(a));
4848 if (port && port->bundle && port->bundle->bond) {
4849 bond_account(port->bundle->bond, &facet->flow,
4850 vlan_tci_to_vid(vlan_tci), n_bytes);
4854 case OVS_ACTION_ATTR_POP_VLAN:
4855 vlan_tci = htons(0);
4858 case OVS_ACTION_ATTR_PUSH_VLAN:
4859 vlan = nl_attr_get(a);
4860 vlan_tci = vlan->vlan_tci;
4866 /* Returns true if the only action for 'facet' is to send to the controller.
4867 * (We don't report NetFlow expiration messages for such facets because they
4868 * are just part of the control logic for the network, not real traffic). */
4870 facet_is_controller_flow(struct facet *facet)
4873 const struct rule *rule = &facet->rule->up;
4874 const struct ofpact *ofpacts = rule->ofpacts;
4875 size_t ofpacts_len = rule->ofpacts_len;
4877 if (ofpacts_len > 0 &&
4878 ofpacts->type == OFPACT_CONTROLLER &&
4879 ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len)) {
4886 /* Folds all of 'facet''s statistics into its rule. Also updates the
4887 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
4888 * 'facet''s statistics in the datapath should have been zeroed and folded into
4889 * its packet and byte counts before this function is called. */
4891 facet_flush_stats(struct facet *facet)
4893 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4894 struct subfacet *subfacet;
4896 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4897 ovs_assert(!subfacet->dp_byte_count);
4898 ovs_assert(!subfacet->dp_packet_count);
4901 facet_push_stats(facet, false);
4902 if (facet->accounted_bytes < facet->byte_count) {
4903 facet_account(facet);
4904 facet->accounted_bytes = facet->byte_count;
4907 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
4908 struct ofexpired expired;
4909 expired.flow = facet->flow;
4910 expired.packet_count = facet->packet_count;
4911 expired.byte_count = facet->byte_count;
4912 expired.used = facet->used;
4913 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4916 /* Reset counters to prevent double counting if 'facet' ever gets
4918 facet_reset_counters(facet);
4920 netflow_flow_clear(&facet->nf_flow);
4921 facet->tcp_flags = 0;
4924 /* Searches 'ofproto''s table of facets for one which would be responsible for
4925 * 'flow'. Returns it if found, otherwise a null pointer.
4927 * The returned facet might need revalidation; use facet_lookup_valid()
4928 * instead if that is important. */
4929 static struct facet *
4930 facet_find(struct ofproto_dpif *ofproto, const struct flow *flow)
4932 struct cls_rule *cr = classifier_lookup(&ofproto->facets, flow, NULL);
4933 return cr ? CONTAINER_OF(cr, struct facet, cr) : NULL;
4936 /* Searches 'ofproto''s table of facets for one capable that covers
4937 * 'flow'. Returns it if found, otherwise a null pointer.
4939 * The returned facet is guaranteed to be valid. */
4940 static struct facet *
4941 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow)
4943 struct facet *facet;
4945 facet = facet_find(ofproto, flow);
4947 && (ofproto->backer->need_revalidate
4948 || tag_set_intersects(&ofproto->backer->revalidate_set,
4950 && !facet_revalidate(facet)) {
4958 facet_check_consistency(struct facet *facet)
4960 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
4962 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4964 struct xlate_out xout;
4965 struct xlate_in xin;
4967 struct rule_dpif *rule;
4970 /* Check the rule for consistency. */
4971 rule = rule_dpif_lookup(ofproto, &facet->flow, NULL);
4972 if (rule != facet->rule) {
4973 if (!VLOG_DROP_WARN(&rl)) {
4974 struct ds s = DS_EMPTY_INITIALIZER;
4976 flow_format(&s, &facet->flow);
4977 ds_put_format(&s, ": facet associated with wrong rule (was "
4978 "table=%"PRIu8",", facet->rule->up.table_id);
4979 cls_rule_format(&facet->rule->up.cr, &s);
4980 ds_put_format(&s, ") (should have been table=%"PRIu8",",
4982 cls_rule_format(&rule->up.cr, &s);
4983 ds_put_cstr(&s, ")\n");
4990 /* Check the datapath actions for consistency. */
4991 xlate_in_init(&xin, ofproto, &facet->flow, &facet->initial_vals, rule,
4993 xlate_actions(&xin, &xout);
4995 ok = ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)
4996 && facet->xout.slow == xout.slow;
4997 if (!ok && !VLOG_DROP_WARN(&rl)) {
4998 struct ds s = DS_EMPTY_INITIALIZER;
5000 flow_format(&s, &facet->flow);
5001 ds_put_cstr(&s, ": inconsistency in facet");
5003 if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
5004 ds_put_cstr(&s, " (actions were: ");
5005 format_odp_actions(&s, facet->xout.odp_actions.data,
5006 facet->xout.odp_actions.size);
5007 ds_put_cstr(&s, ") (correct actions: ");
5008 format_odp_actions(&s, xout.odp_actions.data,
5009 xout.odp_actions.size);
5010 ds_put_cstr(&s, ")");
5013 if (facet->xout.slow != xout.slow) {
5014 ds_put_format(&s, " slow path incorrect. should be %d", xout.slow);
5019 xlate_out_uninit(&xout);
5024 /* Re-searches the classifier for 'facet':
5026 * - If the rule found is different from 'facet''s current rule, moves
5027 * 'facet' to the new rule and recompiles its actions.
5029 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
5030 * where it is and recompiles its actions anyway.
5032 * - If any of 'facet''s subfacets correspond to a new flow according to
5033 * ofproto_receive(), 'facet' is removed.
5035 * Returns true if 'facet' is still valid. False if 'facet' was removed. */
5037 facet_revalidate(struct facet *facet)
5039 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5040 struct rule_dpif *new_rule;
5041 struct subfacet *subfacet;
5042 struct flow_wildcards wc;
5043 struct xlate_out xout;
5044 struct xlate_in xin;
5046 COVERAGE_INC(facet_revalidate);
5048 /* Check that child subfacets still correspond to this facet. Tunnel
5049 * configuration changes could cause a subfacet's OpenFlow in_port to
5051 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
5052 struct ofproto_dpif *recv_ofproto;
5053 struct flow recv_flow;
5056 error = ofproto_receive(ofproto->backer, NULL, subfacet->key,
5057 subfacet->key_len, &recv_flow, NULL,
5058 &recv_ofproto, NULL, NULL);
5060 || recv_ofproto != ofproto
5061 || facet != facet_find(ofproto, &recv_flow)) {
5062 facet_remove(facet);
5067 flow_wildcards_init_catchall(&wc);
5068 new_rule = rule_dpif_lookup(ofproto, &facet->flow, &wc);
5070 /* Calculate new datapath actions.
5072 * We do not modify any 'facet' state yet, because we might need to, e.g.,
5073 * emit a NetFlow expiration and, if so, we need to have the old state
5074 * around to properly compose it. */
5075 xlate_in_init(&xin, ofproto, &facet->flow, &facet->initial_vals, new_rule,
5077 xlate_actions(&xin, &xout);
5078 flow_wildcards_or(&xout.wc, &xout.wc, &wc);
5080 /* A facet's slow path reason should only change under dramatic
5081 * circumstances. Rather than try to update everything, it's simpler to
5082 * remove the facet and start over.
5084 * More importantly, if a facet's wildcards change, it will be relatively
5085 * difficult to figure out if its subfacets still belong to it, and if not
5086 * which facet they may belong to. Again, to avoid the complexity, we
5087 * simply give up instead. */
5088 if (facet->xout.slow != xout.slow
5089 || memcmp(&facet->xout.wc, &xout.wc, sizeof xout.wc)) {
5090 facet_remove(facet);
5091 xlate_out_uninit(&xout);
5095 if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
5096 LIST_FOR_EACH(subfacet, list_node, &facet->subfacets) {
5097 if (subfacet->path == SF_FAST_PATH) {
5098 struct dpif_flow_stats stats;
5100 subfacet_install(subfacet, &xout.odp_actions, &stats);
5101 subfacet_update_stats(subfacet, &stats);
5105 facet_flush_stats(facet);
5107 ofpbuf_clear(&facet->xout.odp_actions);
5108 ofpbuf_put(&facet->xout.odp_actions, xout.odp_actions.data,
5109 xout.odp_actions.size);
5112 /* Update 'facet' now that we've taken care of all the old state. */
5113 facet->xout.tags = xout.tags;
5114 facet->xout.slow = xout.slow;
5115 facet->xout.has_learn = xout.has_learn;
5116 facet->xout.has_normal = xout.has_normal;
5117 facet->xout.has_fin_timeout = xout.has_fin_timeout;
5118 facet->xout.nf_output_iface = xout.nf_output_iface;
5119 facet->xout.mirrors = xout.mirrors;
5120 facet->nf_flow.output_iface = facet->xout.nf_output_iface;
5122 if (facet->rule != new_rule) {
5123 COVERAGE_INC(facet_changed_rule);
5124 list_remove(&facet->list_node);
5125 list_push_back(&new_rule->facets, &facet->list_node);
5126 facet->rule = new_rule;
5127 facet->used = new_rule->up.created;
5128 facet->prev_used = facet->used;
5131 xlate_out_uninit(&xout);
5136 facet_reset_counters(struct facet *facet)
5138 facet->packet_count = 0;
5139 facet->byte_count = 0;
5140 facet->prev_packet_count = 0;
5141 facet->prev_byte_count = 0;
5142 facet->accounted_bytes = 0;
5146 facet_push_stats(struct facet *facet, bool may_learn)
5148 struct dpif_flow_stats stats;
5150 ovs_assert(facet->packet_count >= facet->prev_packet_count);
5151 ovs_assert(facet->byte_count >= facet->prev_byte_count);
5152 ovs_assert(facet->used >= facet->prev_used);
5154 stats.n_packets = facet->packet_count - facet->prev_packet_count;
5155 stats.n_bytes = facet->byte_count - facet->prev_byte_count;
5156 stats.used = facet->used;
5157 stats.tcp_flags = facet->tcp_flags;
5159 if (may_learn || stats.n_packets || facet->used > facet->prev_used) {
5160 struct ofproto_dpif *ofproto =
5161 ofproto_dpif_cast(facet->rule->up.ofproto);
5163 struct ofport_dpif *in_port;
5164 struct xlate_in xin;
5166 facet->prev_packet_count = facet->packet_count;
5167 facet->prev_byte_count = facet->byte_count;
5168 facet->prev_used = facet->used;
5170 in_port = get_ofp_port(ofproto, facet->flow.in_port);
5171 if (in_port && in_port->tnl_port) {
5172 netdev_vport_inc_rx(in_port->up.netdev, &stats);
5175 rule_credit_stats(facet->rule, &stats);
5176 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow,
5178 netflow_flow_update_flags(&facet->nf_flow, facet->tcp_flags);
5179 update_mirror_stats(ofproto, facet->xout.mirrors, stats.n_packets,
5182 xlate_in_init(&xin, ofproto, &facet->flow, &facet->initial_vals,
5183 facet->rule, stats.tcp_flags, NULL);
5184 xin.resubmit_stats = &stats;
5185 xin.may_learn = may_learn;
5186 xlate_actions_for_side_effects(&xin);
5191 push_all_stats__(bool run_fast)
5193 static long long int rl = LLONG_MIN;
5194 struct ofproto_dpif *ofproto;
5196 if (time_msec() < rl) {
5200 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5201 struct cls_cursor cursor;
5202 struct facet *facet;
5204 cls_cursor_init(&cursor, &ofproto->facets, NULL);
5205 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5206 facet_push_stats(facet, false);
5213 rl = time_msec() + 100;
5217 push_all_stats(void)
5219 push_all_stats__(true);
5223 rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats)
5225 rule->packet_count += stats->n_packets;
5226 rule->byte_count += stats->n_bytes;
5227 ofproto_rule_update_used(&rule->up, stats->used);
5232 static struct subfacet *
5233 subfacet_find(struct ofproto_dpif *ofproto,
5234 const struct nlattr *key, size_t key_len, uint32_t key_hash)
5236 struct subfacet *subfacet;
5238 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
5239 &ofproto->subfacets) {
5240 if (subfacet->key_len == key_len
5241 && !memcmp(key, subfacet->key, key_len)) {
5249 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
5250 * 'key_fitness', 'key', and 'key_len' members in 'miss'. Returns the
5251 * existing subfacet if there is one, otherwise creates and returns a
5253 static struct subfacet *
5254 subfacet_create(struct facet *facet, struct flow_miss *miss,
5257 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5258 enum odp_key_fitness key_fitness = miss->key_fitness;
5259 const struct nlattr *key = miss->key;
5260 size_t key_len = miss->key_len;
5262 struct subfacet *subfacet;
5264 key_hash = odp_flow_key_hash(key, key_len);
5266 if (list_is_empty(&facet->subfacets)) {
5267 subfacet = &facet->one_subfacet;
5269 subfacet = subfacet_find(ofproto, key, key_len, key_hash);
5271 if (subfacet->facet == facet) {
5275 /* This shouldn't happen. */
5276 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
5277 subfacet_destroy(subfacet);
5280 subfacet = xmalloc(sizeof *subfacet);
5283 hmap_insert(&ofproto->subfacets, &subfacet->hmap_node, key_hash);
5284 list_push_back(&facet->subfacets, &subfacet->list_node);
5285 subfacet->facet = facet;
5286 subfacet->key_fitness = key_fitness;
5287 subfacet->key = xmemdup(key, key_len);
5288 subfacet->key_len = key_len;
5289 subfacet->used = now;
5290 subfacet->created = now;
5291 subfacet->dp_packet_count = 0;
5292 subfacet->dp_byte_count = 0;
5293 subfacet->path = SF_NOT_INSTALLED;
5294 subfacet->odp_in_port = miss->odp_in_port;
5296 ofproto->subfacet_add_count++;
5300 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
5301 * its facet within 'ofproto', and frees it. */
5303 subfacet_destroy__(struct subfacet *subfacet)
5305 struct facet *facet = subfacet->facet;
5306 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5308 /* Update ofproto stats before uninstall the subfacet. */
5309 ofproto->subfacet_del_count++;
5310 ofproto->total_subfacet_life_span += (time_msec() - subfacet->created);
5312 subfacet_uninstall(subfacet);
5313 hmap_remove(&ofproto->subfacets, &subfacet->hmap_node);
5314 list_remove(&subfacet->list_node);
5315 free(subfacet->key);
5316 if (subfacet != &facet->one_subfacet) {
5321 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
5322 * last remaining subfacet in its facet destroys the facet too. */
5324 subfacet_destroy(struct subfacet *subfacet)
5326 struct facet *facet = subfacet->facet;
5328 if (list_is_singleton(&facet->subfacets)) {
5329 /* facet_remove() needs at least one subfacet (it will remove it). */
5330 facet_remove(facet);
5332 subfacet_destroy__(subfacet);
5337 subfacet_destroy_batch(struct ofproto_dpif *ofproto,
5338 struct subfacet **subfacets, int n)
5340 struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH];
5341 struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH];
5342 struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH];
5345 for (i = 0; i < n; i++) {
5346 ops[i].type = DPIF_OP_FLOW_DEL;
5347 ops[i].u.flow_del.key = subfacets[i]->key;
5348 ops[i].u.flow_del.key_len = subfacets[i]->key_len;
5349 ops[i].u.flow_del.stats = &stats[i];
5353 dpif_operate(ofproto->backer->dpif, opsp, n);
5354 for (i = 0; i < n; i++) {
5355 subfacet_reset_dp_stats(subfacets[i], &stats[i]);
5356 subfacets[i]->path = SF_NOT_INSTALLED;
5357 subfacet_destroy(subfacets[i]);
5362 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
5363 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
5364 * in the datapath will be zeroed and 'stats' will be updated with traffic new
5365 * since 'subfacet' was last updated.
5367 * Returns 0 if successful, otherwise a positive errno value. */
5369 subfacet_install(struct subfacet *subfacet, const struct ofpbuf *odp_actions,
5370 struct dpif_flow_stats *stats)
5372 struct facet *facet = subfacet->facet;
5373 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5374 enum subfacet_path path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
5375 const struct nlattr *actions = odp_actions->data;
5376 size_t actions_len = odp_actions->size;
5378 uint64_t slow_path_stub[128 / 8];
5379 enum dpif_flow_put_flags flags;
5382 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
5384 flags |= DPIF_FP_ZERO_STATS;
5387 if (path == SF_SLOW_PATH) {
5388 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
5389 slow_path_stub, sizeof slow_path_stub,
5390 &actions, &actions_len);
5393 ret = dpif_flow_put(ofproto->backer->dpif, flags, subfacet->key,
5394 subfacet->key_len, actions, actions_len, stats);
5397 subfacet_reset_dp_stats(subfacet, stats);
5401 subfacet->path = path;
5406 /* If 'subfacet' is installed in the datapath, uninstalls it. */
5408 subfacet_uninstall(struct subfacet *subfacet)
5410 if (subfacet->path != SF_NOT_INSTALLED) {
5411 struct rule_dpif *rule = subfacet->facet->rule;
5412 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5413 struct dpif_flow_stats stats;
5416 error = dpif_flow_del(ofproto->backer->dpif, subfacet->key,
5417 subfacet->key_len, &stats);
5418 subfacet_reset_dp_stats(subfacet, &stats);
5420 subfacet_update_stats(subfacet, &stats);
5422 subfacet->path = SF_NOT_INSTALLED;
5424 ovs_assert(subfacet->dp_packet_count == 0);
5425 ovs_assert(subfacet->dp_byte_count == 0);
5429 /* Resets 'subfacet''s datapath statistics counters. This should be called
5430 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
5431 * non-null, it should contain the statistics returned by dpif when 'subfacet'
5432 * was reset in the datapath. 'stats' will be modified to include only
5433 * statistics new since 'subfacet' was last updated. */
5435 subfacet_reset_dp_stats(struct subfacet *subfacet,
5436 struct dpif_flow_stats *stats)
5439 && subfacet->dp_packet_count <= stats->n_packets
5440 && subfacet->dp_byte_count <= stats->n_bytes) {
5441 stats->n_packets -= subfacet->dp_packet_count;
5442 stats->n_bytes -= subfacet->dp_byte_count;
5445 subfacet->dp_packet_count = 0;
5446 subfacet->dp_byte_count = 0;
5449 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
5451 * Because of the meaning of a subfacet's counters, it only makes sense to do
5452 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
5453 * represents a packet that was sent by hand or if it represents statistics
5454 * that have been cleared out of the datapath. */
5456 subfacet_update_stats(struct subfacet *subfacet,
5457 const struct dpif_flow_stats *stats)
5459 if (stats->n_packets || stats->used > subfacet->used) {
5460 struct facet *facet = subfacet->facet;
5462 subfacet->used = MAX(subfacet->used, stats->used);
5463 facet->used = MAX(facet->used, stats->used);
5464 facet->packet_count += stats->n_packets;
5465 facet->byte_count += stats->n_bytes;
5466 facet->tcp_flags |= stats->tcp_flags;
5472 /* Lookup 'flow' in 'ofproto''s classifier. If 'wc' is non-null, sets
5473 * the fields that were relevant as part of the lookup. */
5474 static struct rule_dpif *
5475 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow,
5476 struct flow_wildcards *wc)
5478 struct rule_dpif *rule;
5480 rule = rule_dpif_lookup__(ofproto, flow, wc, 0);
5485 return rule_dpif_miss_rule(ofproto, flow);
5488 static struct rule_dpif *
5489 rule_dpif_lookup__(struct ofproto_dpif *ofproto, const struct flow *flow,
5490 struct flow_wildcards *wc, uint8_t table_id)
5492 struct cls_rule *cls_rule;
5493 struct classifier *cls;
5496 if (table_id >= N_TABLES) {
5500 cls = &ofproto->up.tables[table_id].cls;
5501 frag = (flow->nw_frag & FLOW_NW_FRAG_ANY) != 0;
5502 if (frag && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
5503 /* We must pretend that transport ports are unavailable. */
5504 struct flow ofpc_normal_flow = *flow;
5505 ofpc_normal_flow.tp_src = htons(0);
5506 ofpc_normal_flow.tp_dst = htons(0);
5507 cls_rule = classifier_lookup(cls, &ofpc_normal_flow, wc);
5508 } else if (frag && ofproto->up.frag_handling == OFPC_FRAG_DROP) {
5509 cls_rule = &ofproto->drop_frags_rule->up.cr;
5511 flow_wildcards_init_exact(wc);
5514 cls_rule = classifier_lookup(cls, flow, wc);
5516 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
5519 static struct rule_dpif *
5520 rule_dpif_miss_rule(struct ofproto_dpif *ofproto, const struct flow *flow)
5522 struct ofport_dpif *port;
5524 port = get_ofp_port(ofproto, flow->in_port);
5526 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, flow->in_port);
5527 return ofproto->miss_rule;
5530 if (port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN) {
5531 return ofproto->no_packet_in_rule;
5533 return ofproto->miss_rule;
5537 complete_operation(struct rule_dpif *rule)
5539 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5541 rule_invalidate(rule);
5543 struct dpif_completion *c = xmalloc(sizeof *c);
5544 c->op = rule->up.pending;
5545 list_push_back(&ofproto->completions, &c->list_node);
5547 ofoperation_complete(rule->up.pending, 0);
5551 static struct rule *
5554 struct rule_dpif *rule = xmalloc(sizeof *rule);
5559 rule_dealloc(struct rule *rule_)
5561 struct rule_dpif *rule = rule_dpif_cast(rule_);
5566 rule_construct(struct rule *rule_)
5568 struct rule_dpif *rule = rule_dpif_cast(rule_);
5569 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5570 struct rule_dpif *victim;
5573 rule->packet_count = 0;
5574 rule->byte_count = 0;
5576 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
5577 if (victim && !list_is_empty(&victim->facets)) {
5578 struct facet *facet;
5580 rule->facets = victim->facets;
5581 list_moved(&rule->facets);
5582 LIST_FOR_EACH (facet, list_node, &rule->facets) {
5583 /* XXX: We're only clearing our local counters here. It's possible
5584 * that quite a few packets are unaccounted for in the datapath
5585 * statistics. These will be accounted to the new rule instead of
5586 * cleared as required. This could be fixed by clearing out the
5587 * datapath statistics for this facet, but currently it doesn't
5589 facet_reset_counters(facet);
5593 /* Must avoid list_moved() in this case. */
5594 list_init(&rule->facets);
5597 table_id = rule->up.table_id;
5599 rule->tag = victim->tag;
5600 } else if (table_id == 0) {
5605 miniflow_expand(&rule->up.cr.match.flow, &flow);
5606 rule->tag = rule_calculate_tag(&flow, &rule->up.cr.match.mask,
5607 ofproto->tables[table_id].basis);
5610 complete_operation(rule);
5615 rule_destruct(struct rule *rule_)
5617 struct rule_dpif *rule = rule_dpif_cast(rule_);
5618 struct facet *facet, *next_facet;
5620 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
5621 facet_revalidate(facet);
5624 complete_operation(rule);
5628 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
5630 struct rule_dpif *rule = rule_dpif_cast(rule_);
5632 /* push_all_stats() can handle flow misses which, when using the learn
5633 * action, can cause rules to be added and deleted. This can corrupt our
5634 * caller's datastructures which assume that rule_get_stats() doesn't have
5635 * an impact on the flow table. To be safe, we disable miss handling. */
5636 push_all_stats__(false);
5638 /* Start from historical data for 'rule' itself that are no longer tracked
5639 * in facets. This counts, for example, facets that have expired. */
5640 *packets = rule->packet_count;
5641 *bytes = rule->byte_count;
5645 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
5646 struct ofpbuf *packet)
5648 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5649 struct initial_vals initial_vals;
5650 struct dpif_flow_stats stats;
5651 struct xlate_out xout;
5652 struct xlate_in xin;
5654 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
5655 rule_credit_stats(rule, &stats);
5657 initial_vals.vlan_tci = flow->vlan_tci;
5658 initial_vals.tunnel_ip_tos = flow->tunnel.ip_tos;
5659 xlate_in_init(&xin, ofproto, flow, &initial_vals, rule, stats.tcp_flags,
5661 xin.resubmit_stats = &stats;
5662 xlate_actions(&xin, &xout);
5664 execute_odp_actions(ofproto, flow, xout.odp_actions.data,
5665 xout.odp_actions.size, packet);
5667 xlate_out_uninit(&xout);
5671 rule_execute(struct rule *rule, const struct flow *flow,
5672 struct ofpbuf *packet)
5674 rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
5675 ofpbuf_delete(packet);
5680 rule_modify_actions(struct rule *rule_)
5682 struct rule_dpif *rule = rule_dpif_cast(rule_);
5684 complete_operation(rule);
5687 /* Sends 'packet' out 'ofport'.
5688 * May modify 'packet'.
5689 * Returns 0 if successful, otherwise a positive errno value. */
5691 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
5693 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
5694 uint64_t odp_actions_stub[1024 / 8];
5695 struct ofpbuf key, odp_actions;
5696 struct odputil_keybuf keybuf;
5701 flow_extract(packet, 0, 0, NULL, OFPP_LOCAL, &flow);
5702 if (netdev_vport_is_patch(ofport->up.netdev)) {
5703 struct ofproto_dpif *peer_ofproto;
5704 struct dpif_flow_stats stats;
5705 struct ofport_dpif *peer;
5706 struct rule_dpif *rule;
5708 peer = ofport_get_peer(ofport);
5713 dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
5714 netdev_vport_inc_tx(ofport->up.netdev, &stats);
5715 netdev_vport_inc_rx(peer->up.netdev, &stats);
5717 flow.in_port = peer->up.ofp_port;
5718 peer_ofproto = ofproto_dpif_cast(peer->up.ofproto);
5719 rule = rule_dpif_lookup(peer_ofproto, &flow, NULL);
5720 rule_dpif_execute(rule, &flow, packet);
5725 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5727 if (ofport->tnl_port) {
5728 struct flow_wildcards wc;
5729 struct dpif_flow_stats stats;
5731 odp_port = tnl_port_send(ofport->tnl_port, &flow, &wc);
5732 if (odp_port == OVSP_NONE) {
5736 dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
5737 netdev_vport_inc_tx(ofport->up.netdev, &stats);
5738 odp_put_tunnel_action(&flow.tunnel, &odp_actions);
5739 odp_put_skb_mark_action(flow.skb_mark, &odp_actions);
5741 odp_port = vsp_realdev_to_vlandev(ofproto, ofport->odp_port,
5743 if (odp_port != ofport->odp_port) {
5744 eth_pop_vlan(packet);
5745 flow.vlan_tci = htons(0);
5749 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5750 odp_flow_key_from_flow(&key, &flow,
5751 ofp_port_to_odp_port(ofproto, flow.in_port));
5753 compose_sflow_action(ofproto, &odp_actions, &flow, odp_port);
5754 compose_ipfix_action(ofproto, &odp_actions, &flow);
5756 nl_msg_put_u32(&odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port);
5757 error = dpif_execute(ofproto->backer->dpif,
5759 odp_actions.data, odp_actions.size,
5761 ofpbuf_uninit(&odp_actions);
5764 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)",
5765 ofproto->up.name, odp_port, strerror(error));
5767 ofproto_update_local_port_stats(ofport->up.ofproto, packet->size, 0);
5771 /* OpenFlow to datapath action translation. */
5773 static bool may_receive(const struct ofport_dpif *, struct xlate_ctx *);
5774 static void do_xlate_actions(const struct ofpact *, size_t ofpacts_len,
5775 struct xlate_ctx *);
5776 static void xlate_normal(struct xlate_ctx *);
5778 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
5779 * The action will state 'slow' as the reason that the action is in the slow
5780 * path. (This is purely informational: it allows a human viewing "ovs-dpctl
5781 * dump-flows" output to see why a flow is in the slow path.)
5783 * The 'stub_size' bytes in 'stub' will be used to store the action.
5784 * 'stub_size' must be large enough for the action.
5786 * The action and its size will be stored in '*actionsp' and '*actions_lenp',
5789 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
5790 enum slow_path_reason slow,
5791 uint64_t *stub, size_t stub_size,
5792 const struct nlattr **actionsp, size_t *actions_lenp)
5794 union user_action_cookie cookie;
5797 cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
5798 cookie.slow_path.unused = 0;
5799 cookie.slow_path.reason = slow;
5801 ofpbuf_use_stack(&buf, stub, stub_size);
5802 if (slow & (SLOW_CFM | SLOW_LACP | SLOW_STP)) {
5803 uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif, UINT32_MAX);
5804 odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
5806 put_userspace_action(ofproto, &buf, flow, &cookie,
5807 sizeof cookie.slow_path);
5809 *actionsp = buf.data;
5810 *actions_lenp = buf.size;
5814 put_userspace_action(const struct ofproto_dpif *ofproto,
5815 struct ofpbuf *odp_actions,
5816 const struct flow *flow,
5817 const union user_action_cookie *cookie,
5818 const size_t cookie_size)
5822 pid = dpif_port_get_pid(ofproto->backer->dpif,
5823 ofp_port_to_odp_port(ofproto, flow->in_port));
5825 return odp_put_userspace_action(pid, cookie, cookie_size, odp_actions);
5828 /* Compose SAMPLE action for sFlow or IPFIX. The given probability is
5829 * the number of packets out of UINT32_MAX to sample. The given
5830 * cookie is passed back in the callback for each sampled packet.
5833 compose_sample_action(const struct ofproto_dpif *ofproto,
5834 struct ofpbuf *odp_actions,
5835 const struct flow *flow,
5836 const uint32_t probability,
5837 const union user_action_cookie *cookie,
5838 const size_t cookie_size)
5840 size_t sample_offset, actions_offset;
5843 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
5845 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
5847 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
5848 cookie_offset = put_userspace_action(ofproto, odp_actions, flow, cookie,
5851 nl_msg_end_nested(odp_actions, actions_offset);
5852 nl_msg_end_nested(odp_actions, sample_offset);
5853 return cookie_offset;
5857 compose_sflow_cookie(const struct ofproto_dpif *ofproto,
5858 ovs_be16 vlan_tci, uint32_t odp_port,
5859 unsigned int n_outputs, union user_action_cookie *cookie)
5863 cookie->type = USER_ACTION_COOKIE_SFLOW;
5864 cookie->sflow.vlan_tci = vlan_tci;
5866 /* See http://www.sflow.org/sflow_version_5.txt (search for "Input/output
5867 * port information") for the interpretation of cookie->output. */
5868 switch (n_outputs) {
5870 /* 0x40000000 | 256 means "packet dropped for unknown reason". */
5871 cookie->sflow.output = 0x40000000 | 256;
5875 ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port);
5877 cookie->sflow.output = ifindex;
5882 /* 0x80000000 means "multiple output ports. */
5883 cookie->sflow.output = 0x80000000 | n_outputs;
5888 /* Compose SAMPLE action for sFlow bridge sampling. */
5890 compose_sflow_action(const struct ofproto_dpif *ofproto,
5891 struct ofpbuf *odp_actions,
5892 const struct flow *flow,
5895 uint32_t probability;
5896 union user_action_cookie cookie;
5898 if (!ofproto->sflow || flow->in_port == OFPP_NONE) {
5902 probability = dpif_sflow_get_probability(ofproto->sflow);
5903 compose_sflow_cookie(ofproto, htons(0), odp_port,
5904 odp_port == OVSP_NONE ? 0 : 1, &cookie);
5906 return compose_sample_action(ofproto, odp_actions, flow, probability,
5907 &cookie, sizeof cookie.sflow);
5911 compose_flow_sample_cookie(uint16_t probability, uint32_t collector_set_id,
5912 uint32_t obs_domain_id, uint32_t obs_point_id,
5913 union user_action_cookie *cookie)
5915 cookie->type = USER_ACTION_COOKIE_FLOW_SAMPLE;
5916 cookie->flow_sample.probability = probability;
5917 cookie->flow_sample.collector_set_id = collector_set_id;
5918 cookie->flow_sample.obs_domain_id = obs_domain_id;
5919 cookie->flow_sample.obs_point_id = obs_point_id;
5923 compose_ipfix_cookie(union user_action_cookie *cookie)
5925 cookie->type = USER_ACTION_COOKIE_IPFIX;
5928 /* Compose SAMPLE action for IPFIX bridge sampling. */
5930 compose_ipfix_action(const struct ofproto_dpif *ofproto,
5931 struct ofpbuf *odp_actions,
5932 const struct flow *flow)
5934 uint32_t probability;
5935 union user_action_cookie cookie;
5937 if (!ofproto->ipfix || flow->in_port == OFPP_NONE) {
5941 probability = dpif_ipfix_get_bridge_exporter_probability(ofproto->ipfix);
5942 compose_ipfix_cookie(&cookie);
5944 compose_sample_action(ofproto, odp_actions, flow, probability,
5945 &cookie, sizeof cookie.ipfix);
5948 /* SAMPLE action for sFlow must be first action in any given list of
5949 * actions. At this point we do not have all information required to
5950 * build it. So try to build sample action as complete as possible. */
5952 add_sflow_action(struct xlate_ctx *ctx)
5954 ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto,
5955 &ctx->xout->odp_actions,
5956 &ctx->xin->flow, OVSP_NONE);
5957 ctx->sflow_odp_port = 0;
5958 ctx->sflow_n_outputs = 0;
5961 /* SAMPLE action for IPFIX must be 1st or 2nd action in any given list
5962 * of actions, eventually after the SAMPLE action for sFlow. */
5964 add_ipfix_action(struct xlate_ctx *ctx)
5966 compose_ipfix_action(ctx->ofproto, &ctx->xout->odp_actions,
5970 /* Fix SAMPLE action according to data collected while composing ODP actions.
5971 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
5972 * USERSPACE action's user-cookie which is required for sflow. */
5974 fix_sflow_action(struct xlate_ctx *ctx)
5976 const struct flow *base = &ctx->base_flow;
5977 union user_action_cookie *cookie;
5979 if (!ctx->user_cookie_offset) {
5983 cookie = ofpbuf_at(&ctx->xout->odp_actions, ctx->user_cookie_offset,
5984 sizeof cookie->sflow);
5985 ovs_assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
5987 compose_sflow_cookie(ctx->ofproto, base->vlan_tci,
5988 ctx->sflow_odp_port, ctx->sflow_n_outputs, cookie);
5992 compose_output_action__(struct xlate_ctx *ctx, uint16_t ofp_port,
5995 const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
5996 ovs_be16 flow_vlan_tci;
5997 uint32_t flow_skb_mark;
5998 uint8_t flow_nw_tos;
5999 struct priority_to_dscp *pdscp;
6000 uint32_t out_port, odp_port;
6002 /* If 'struct flow' gets additional metadata, we'll need to zero it out
6003 * before traversing a patch port. */
6004 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 20);
6007 xlate_report(ctx, "Nonexistent output port");
6009 } else if (ofport->up.pp.config & OFPUTIL_PC_NO_FWD) {
6010 xlate_report(ctx, "OFPPC_NO_FWD set, skipping output");
6012 } else if (check_stp && !stp_forward_in_state(ofport->stp_state)) {
6013 xlate_report(ctx, "STP not in forwarding state, skipping output");
6017 if (netdev_vport_is_patch(ofport->up.netdev)) {
6018 struct ofport_dpif *peer = ofport_get_peer(ofport);
6019 struct flow old_flow = ctx->xin->flow;
6020 const struct ofproto_dpif *peer_ofproto;
6021 enum slow_path_reason special;
6022 struct ofport_dpif *in_port;
6025 xlate_report(ctx, "Nonexistent patch port peer");
6029 peer_ofproto = ofproto_dpif_cast(peer->up.ofproto);
6030 if (peer_ofproto->backer != ctx->ofproto->backer) {
6031 xlate_report(ctx, "Patch port peer on a different datapath");
6035 ctx->ofproto = ofproto_dpif_cast(peer->up.ofproto);
6036 ctx->xin->flow.in_port = peer->up.ofp_port;
6037 ctx->xin->flow.metadata = htonll(0);
6038 memset(&ctx->xin->flow.tunnel, 0, sizeof ctx->xin->flow.tunnel);
6039 memset(ctx->xin->flow.regs, 0, sizeof ctx->xin->flow.regs);
6041 in_port = get_ofp_port(ctx->ofproto, ctx->xin->flow.in_port);
6042 special = process_special(ctx, &ctx->xin->flow, in_port,
6045 ctx->xout->slow = special;
6046 } else if (!in_port || may_receive(in_port, ctx)) {
6047 if (!in_port || stp_forward_in_state(in_port->stp_state)) {
6048 xlate_table_action(ctx, ctx->xin->flow.in_port, 0, true);
6050 /* Forwarding is disabled by STP. Let OFPP_NORMAL and the
6051 * learning action look at the packet, then drop it. */
6052 struct flow old_base_flow = ctx->base_flow;
6053 size_t old_size = ctx->xout->odp_actions.size;
6054 xlate_table_action(ctx, ctx->xin->flow.in_port, 0, true);
6055 ctx->base_flow = old_base_flow;
6056 ctx->xout->odp_actions.size = old_size;
6060 ctx->xin->flow = old_flow;
6061 ctx->ofproto = ofproto_dpif_cast(ofport->up.ofproto);
6063 if (ctx->xin->resubmit_stats) {
6064 netdev_vport_inc_tx(ofport->up.netdev, ctx->xin->resubmit_stats);
6065 netdev_vport_inc_rx(peer->up.netdev, ctx->xin->resubmit_stats);
6071 flow_vlan_tci = ctx->xin->flow.vlan_tci;
6072 flow_skb_mark = ctx->xin->flow.skb_mark;
6073 flow_nw_tos = ctx->xin->flow.nw_tos;
6075 pdscp = get_priority(ofport, ctx->xin->flow.skb_priority);
6077 ctx->xin->flow.nw_tos &= ~IP_DSCP_MASK;
6078 ctx->xin->flow.nw_tos |= pdscp->dscp;
6081 if (ofport->tnl_port) {
6082 /* Save tunnel metadata so that changes made due to
6083 * the Logical (tunnel) Port are not visible for any further
6084 * matches, while explicit set actions on tunnel metadata are.
6086 struct flow_tnl flow_tnl = ctx->xin->flow.tunnel;
6087 odp_port = tnl_port_send(ofport->tnl_port, &ctx->xin->flow,
6089 if (odp_port == OVSP_NONE) {
6090 xlate_report(ctx, "Tunneling decided against output");
6091 goto out; /* restore flow_nw_tos */
6094 if (ctx->xin->resubmit_stats) {
6095 netdev_vport_inc_tx(ofport->up.netdev, ctx->xin->resubmit_stats);
6097 out_port = odp_port;
6098 commit_odp_tunnel_action(&ctx->xin->flow, &ctx->base_flow,
6099 &ctx->xout->odp_actions);
6100 ctx->xin->flow.tunnel = flow_tnl; /* Restore tunnel metadata */
6102 uint16_t vlandev_port;
6103 odp_port = ofport->odp_port;
6104 vlandev_port = vsp_realdev_to_vlandev(ctx->ofproto, ofp_port,
6105 ctx->xin->flow.vlan_tci);
6106 if (vlandev_port == ofp_port) {
6107 out_port = odp_port;
6109 out_port = ofp_port_to_odp_port(ctx->ofproto, vlandev_port);
6110 ctx->xin->flow.vlan_tci = htons(0);
6112 ctx->xin->flow.skb_mark &= ~IPSEC_MARK;
6114 commit_odp_actions(&ctx->xin->flow, &ctx->base_flow,
6115 &ctx->xout->odp_actions);
6116 nl_msg_put_u32(&ctx->xout->odp_actions, OVS_ACTION_ATTR_OUTPUT, out_port);
6118 ctx->sflow_odp_port = odp_port;
6119 ctx->sflow_n_outputs++;
6120 ctx->xout->nf_output_iface = ofp_port;
6123 ctx->xin->flow.vlan_tci = flow_vlan_tci;
6124 ctx->xin->flow.skb_mark = flow_skb_mark;
6126 ctx->xin->flow.nw_tos = flow_nw_tos;
6130 compose_output_action(struct xlate_ctx *ctx, uint16_t ofp_port)
6132 compose_output_action__(ctx, ofp_port, true);
6136 tag_the_flow(struct xlate_ctx *ctx, struct rule_dpif *rule)
6138 struct ofproto_dpif *ofproto = ctx->ofproto;
6139 uint8_t table_id = ctx->table_id;
6141 if (table_id > 0 && table_id < N_TABLES) {
6142 struct table_dpif *table = &ofproto->tables[table_id];
6143 if (table->other_table) {
6144 ctx->xout->tags |= (rule && rule->tag
6146 : rule_calculate_tag(&ctx->xin->flow,
6147 &table->other_table->mask,
6153 /* Common rule processing in one place to avoid duplicating code. */
6154 static struct rule_dpif *
6155 ctx_rule_hooks(struct xlate_ctx *ctx, struct rule_dpif *rule,
6158 if (ctx->xin->resubmit_hook) {
6159 ctx->xin->resubmit_hook(ctx, rule);
6161 if (rule == NULL && may_packet_in) {
6163 * check if table configuration flags
6164 * OFPTC_TABLE_MISS_CONTROLLER, default.
6165 * OFPTC_TABLE_MISS_CONTINUE,
6166 * OFPTC_TABLE_MISS_DROP
6167 * When OF1.0, OFPTC_TABLE_MISS_CONTINUE is used. What to do?
6169 rule = rule_dpif_miss_rule(ctx->ofproto, &ctx->xin->flow);
6171 if (rule && ctx->xin->resubmit_stats) {
6172 rule_credit_stats(rule, ctx->xin->resubmit_stats);
6178 xlate_table_action(struct xlate_ctx *ctx,
6179 uint16_t in_port, uint8_t table_id, bool may_packet_in)
6181 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
6182 struct rule_dpif *rule;
6183 uint16_t old_in_port = ctx->xin->flow.in_port;
6184 uint8_t old_table_id = ctx->table_id;
6186 ctx->table_id = table_id;
6188 /* Look up a flow with 'in_port' as the input port. */
6189 ctx->xin->flow.in_port = in_port;
6190 rule = rule_dpif_lookup__(ctx->ofproto, &ctx->xin->flow,
6191 &ctx->xout->wc, table_id);
6193 tag_the_flow(ctx, rule);
6195 /* Restore the original input port. Otherwise OFPP_NORMAL and
6196 * OFPP_IN_PORT will have surprising behavior. */
6197 ctx->xin->flow.in_port = old_in_port;
6199 rule = ctx_rule_hooks(ctx, rule, may_packet_in);
6202 struct rule_dpif *old_rule = ctx->rule;
6206 do_xlate_actions(rule->up.ofpacts, rule->up.ofpacts_len, ctx);
6207 ctx->rule = old_rule;
6211 ctx->table_id = old_table_id;
6213 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
6215 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
6216 MAX_RESUBMIT_RECURSION);
6217 ctx->max_resubmit_trigger = true;
6222 xlate_ofpact_resubmit(struct xlate_ctx *ctx,
6223 const struct ofpact_resubmit *resubmit)
6228 in_port = resubmit->in_port;
6229 if (in_port == OFPP_IN_PORT) {
6230 in_port = ctx->xin->flow.in_port;
6233 table_id = resubmit->table_id;
6234 if (table_id == 255) {
6235 table_id = ctx->table_id;
6238 xlate_table_action(ctx, in_port, table_id, false);
6242 flood_packets(struct xlate_ctx *ctx, bool all)
6244 struct ofport_dpif *ofport;
6246 HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) {
6247 uint16_t ofp_port = ofport->up.ofp_port;
6249 if (ofp_port == ctx->xin->flow.in_port) {
6254 compose_output_action__(ctx, ofp_port, false);
6255 } else if (!(ofport->up.pp.config & OFPUTIL_PC_NO_FLOOD)) {
6256 compose_output_action(ctx, ofp_port);
6260 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
6264 execute_controller_action(struct xlate_ctx *ctx, int len,
6265 enum ofp_packet_in_reason reason,
6266 uint16_t controller_id)
6268 struct ofputil_packet_in pin;
6269 struct ofpbuf *packet;
6271 ovs_assert(!ctx->xout->slow || ctx->xout->slow == SLOW_CONTROLLER);
6272 ctx->xout->slow = SLOW_CONTROLLER;
6273 if (!ctx->xin->packet) {
6277 packet = ofpbuf_clone(ctx->xin->packet);
6279 if (packet->l2 && packet->l3) {
6280 struct eth_header *eh;
6281 uint16_t mpls_depth;
6283 eth_pop_vlan(packet);
6286 memcpy(eh->eth_src, ctx->xin->flow.dl_src, sizeof eh->eth_src);
6287 memcpy(eh->eth_dst, ctx->xin->flow.dl_dst, sizeof eh->eth_dst);
6289 if (ctx->xin->flow.vlan_tci & htons(VLAN_CFI)) {
6290 eth_push_vlan(packet, ctx->xin->flow.vlan_tci);
6293 mpls_depth = eth_mpls_depth(packet);
6295 if (mpls_depth < ctx->xin->flow.mpls_depth) {
6296 push_mpls(packet, ctx->xin->flow.dl_type, ctx->xin->flow.mpls_lse);
6297 } else if (mpls_depth > ctx->xin->flow.mpls_depth) {
6298 pop_mpls(packet, ctx->xin->flow.dl_type);
6299 } else if (mpls_depth) {
6300 set_mpls_lse(packet, ctx->xin->flow.mpls_lse);
6304 if (ctx->xin->flow.dl_type == htons(ETH_TYPE_IP)) {
6305 packet_set_ipv4(packet, ctx->xin->flow.nw_src,
6306 ctx->xin->flow.nw_dst, ctx->xin->flow.nw_tos,
6307 ctx->xin->flow.nw_ttl);
6311 if (ctx->xin->flow.nw_proto == IPPROTO_TCP) {
6312 packet_set_tcp_port(packet, ctx->xin->flow.tp_src,
6313 ctx->xin->flow.tp_dst);
6314 } else if (ctx->xin->flow.nw_proto == IPPROTO_UDP) {
6315 packet_set_udp_port(packet, ctx->xin->flow.tp_src,
6316 ctx->xin->flow.tp_dst);
6322 pin.packet = packet->data;
6323 pin.packet_len = packet->size;
6324 pin.reason = reason;
6325 pin.controller_id = controller_id;
6326 pin.table_id = ctx->table_id;
6327 pin.cookie = ctx->rule ? ctx->rule->up.flow_cookie : 0;
6330 flow_get_metadata(&ctx->xin->flow, &pin.fmd);
6332 connmgr_send_packet_in(ctx->ofproto->up.connmgr, &pin);
6333 ofpbuf_delete(packet);
6337 execute_mpls_push_action(struct xlate_ctx *ctx, ovs_be16 eth_type)
6339 ovs_assert(eth_type_mpls(eth_type));
6341 memset(&ctx->xout->wc.masks.dl_type, 0xff,
6342 sizeof ctx->xout->wc.masks.dl_type);
6343 memset(&ctx->xout->wc.masks.mpls_lse, 0xff,
6344 sizeof ctx->xout->wc.masks.mpls_lse);
6345 memset(&ctx->xout->wc.masks.mpls_depth, 0xff,
6346 sizeof ctx->xout->wc.masks.mpls_depth);
6348 if (ctx->base_flow.mpls_depth) {
6349 ctx->xin->flow.mpls_lse &= ~htonl(MPLS_BOS_MASK);
6350 ctx->xin->flow.mpls_depth++;
6355 if (ctx->xin->flow.dl_type == htons(ETH_TYPE_IPV6)) {
6356 label = htonl(0x2); /* IPV6 Explicit Null. */
6358 label = htonl(0x0); /* IPV4 Explicit Null. */
6360 tc = (ctx->xin->flow.nw_tos & IP_DSCP_MASK) >> 2;
6361 ttl = ctx->xin->flow.nw_ttl ? ctx->xin->flow.nw_ttl : 0x40;
6362 ctx->xin->flow.mpls_lse = set_mpls_lse_values(ttl, tc, 1, label);
6363 ctx->xin->flow.mpls_depth = 1;
6365 ctx->xin->flow.dl_type = eth_type;
6369 execute_mpls_pop_action(struct xlate_ctx *ctx, ovs_be16 eth_type)
6371 ovs_assert(eth_type_mpls(ctx->xin->flow.dl_type));
6372 ovs_assert(!eth_type_mpls(eth_type));
6374 memset(&ctx->xout->wc.masks.dl_type, 0xff,
6375 sizeof ctx->xout->wc.masks.dl_type);
6376 memset(&ctx->xout->wc.masks.mpls_lse, 0xff,
6377 sizeof ctx->xout->wc.masks.mpls_lse);
6378 memset(&ctx->xout->wc.masks.mpls_depth, 0xff,
6379 sizeof ctx->xout->wc.masks.mpls_depth);
6381 if (ctx->xin->flow.mpls_depth) {
6382 ctx->xin->flow.mpls_depth--;
6383 ctx->xin->flow.mpls_lse = htonl(0);
6384 if (!ctx->xin->flow.mpls_depth) {
6385 ctx->xin->flow.dl_type = eth_type;
6391 compose_dec_ttl(struct xlate_ctx *ctx, struct ofpact_cnt_ids *ids)
6393 if (ctx->xin->flow.dl_type != htons(ETH_TYPE_IP) &&
6394 ctx->xin->flow.dl_type != htons(ETH_TYPE_IPV6)) {
6398 if (ctx->xin->flow.nw_ttl > 1) {
6399 ctx->xin->flow.nw_ttl--;
6404 for (i = 0; i < ids->n_controllers; i++) {
6405 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL,
6409 /* Stop processing for current table. */
6415 execute_set_mpls_ttl_action(struct xlate_ctx *ctx, uint8_t ttl)
6417 if (!eth_type_mpls(ctx->xin->flow.dl_type)) {
6421 set_mpls_lse_ttl(&ctx->xin->flow.mpls_lse, ttl);
6426 execute_dec_mpls_ttl_action(struct xlate_ctx *ctx)
6428 uint8_t ttl = mpls_lse_to_ttl(ctx->xin->flow.mpls_lse);
6430 if (!eth_type_mpls(ctx->xin->flow.dl_type)) {
6436 set_mpls_lse_ttl(&ctx->xin->flow.mpls_lse, ttl);
6439 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL, 0);
6441 /* Stop processing for current table. */
6447 xlate_output_action(struct xlate_ctx *ctx,
6448 uint16_t port, uint16_t max_len, bool may_packet_in)
6450 uint16_t prev_nf_output_iface = ctx->xout->nf_output_iface;
6452 ctx->xout->nf_output_iface = NF_OUT_DROP;
6456 compose_output_action(ctx, ctx->xin->flow.in_port);
6459 xlate_table_action(ctx, ctx->xin->flow.in_port, 0, may_packet_in);
6465 flood_packets(ctx, false);
6468 flood_packets(ctx, true);
6470 case OFPP_CONTROLLER:
6471 execute_controller_action(ctx, max_len, OFPR_ACTION, 0);
6477 if (port != ctx->xin->flow.in_port) {
6478 compose_output_action(ctx, port);
6480 xlate_report(ctx, "skipping output to input port");
6485 if (prev_nf_output_iface == NF_OUT_FLOOD) {
6486 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
6487 } else if (ctx->xout->nf_output_iface == NF_OUT_DROP) {
6488 ctx->xout->nf_output_iface = prev_nf_output_iface;
6489 } else if (prev_nf_output_iface != NF_OUT_DROP &&
6490 ctx->xout->nf_output_iface != NF_OUT_FLOOD) {
6491 ctx->xout->nf_output_iface = NF_OUT_MULTI;
6496 xlate_output_reg_action(struct xlate_ctx *ctx,
6497 const struct ofpact_output_reg *or)
6499 uint64_t port = mf_get_subfield(&or->src, &ctx->xin->flow);
6500 if (port <= UINT16_MAX) {
6501 union mf_subvalue value;
6503 memset(&value, 0xff, sizeof value);
6504 mf_write_subfield_flow(&or->src, &value, &ctx->xout->wc.masks);
6505 xlate_output_action(ctx, port, or->max_len, false);
6510 xlate_enqueue_action(struct xlate_ctx *ctx,
6511 const struct ofpact_enqueue *enqueue)
6513 uint16_t ofp_port = enqueue->port;
6514 uint32_t queue_id = enqueue->queue;
6515 uint32_t flow_priority, priority;
6518 /* Translate queue to priority. */
6519 error = dpif_queue_to_priority(ctx->ofproto->backer->dpif,
6520 queue_id, &priority);
6522 /* Fall back to ordinary output action. */
6523 xlate_output_action(ctx, enqueue->port, 0, false);
6527 /* Check output port. */
6528 if (ofp_port == OFPP_IN_PORT) {
6529 ofp_port = ctx->xin->flow.in_port;
6530 } else if (ofp_port == ctx->xin->flow.in_port) {
6534 /* Add datapath actions. */
6535 flow_priority = ctx->xin->flow.skb_priority;
6536 ctx->xin->flow.skb_priority = priority;
6537 compose_output_action(ctx, ofp_port);
6538 ctx->xin->flow.skb_priority = flow_priority;
6540 /* Update NetFlow output port. */
6541 if (ctx->xout->nf_output_iface == NF_OUT_DROP) {
6542 ctx->xout->nf_output_iface = ofp_port;
6543 } else if (ctx->xout->nf_output_iface != NF_OUT_FLOOD) {
6544 ctx->xout->nf_output_iface = NF_OUT_MULTI;
6549 xlate_set_queue_action(struct xlate_ctx *ctx, uint32_t queue_id)
6551 uint32_t skb_priority;
6553 if (!dpif_queue_to_priority(ctx->ofproto->backer->dpif,
6554 queue_id, &skb_priority)) {
6555 ctx->xin->flow.skb_priority = skb_priority;
6557 /* Couldn't translate queue to a priority. Nothing to do. A warning
6558 * has already been logged. */
6563 slave_enabled_cb(uint16_t ofp_port, void *ofproto_)
6565 struct ofproto_dpif *ofproto = ofproto_;
6566 struct ofport_dpif *port;
6576 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
6579 port = get_ofp_port(ofproto, ofp_port);
6580 return port ? port->may_enable : false;
6585 xlate_bundle_action(struct xlate_ctx *ctx,
6586 const struct ofpact_bundle *bundle)
6590 port = bundle_execute(bundle, &ctx->xin->flow, &ctx->xout->wc,
6591 slave_enabled_cb, ctx->ofproto);
6592 if (bundle->dst.field) {
6593 nxm_reg_load(&bundle->dst, port, &ctx->xin->flow);
6595 xlate_output_action(ctx, port, 0, false);
6600 xlate_learn_action(struct xlate_ctx *ctx,
6601 const struct ofpact_learn *learn)
6603 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
6604 struct ofputil_flow_mod fm;
6605 uint64_t ofpacts_stub[1024 / 8];
6606 struct ofpbuf ofpacts;
6609 ctx->xout->has_learn = true;
6611 learn_mask(learn, &ctx->xout->wc);
6613 if (!ctx->xin->may_learn) {
6617 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
6618 learn_execute(learn, &ctx->xin->flow, &fm, &ofpacts);
6620 error = ofproto_flow_mod(&ctx->ofproto->up, &fm);
6621 if (error && !VLOG_DROP_WARN(&rl)) {
6622 VLOG_WARN("learning action failed to modify flow table (%s)",
6623 ofperr_get_name(error));
6626 ofpbuf_uninit(&ofpacts);
6629 /* Reduces '*timeout' to no more than 'max'. A value of zero in either case
6630 * means "infinite". */
6632 reduce_timeout(uint16_t max, uint16_t *timeout)
6634 if (max && (!*timeout || *timeout > max)) {
6640 xlate_fin_timeout(struct xlate_ctx *ctx,
6641 const struct ofpact_fin_timeout *oft)
6643 if (ctx->xin->tcp_flags & (TCP_FIN | TCP_RST) && ctx->rule) {
6644 struct rule_dpif *rule = ctx->rule;
6646 reduce_timeout(oft->fin_idle_timeout, &rule->up.idle_timeout);
6647 reduce_timeout(oft->fin_hard_timeout, &rule->up.hard_timeout);
6652 xlate_sample_action(struct xlate_ctx *ctx,
6653 const struct ofpact_sample *os)
6655 union user_action_cookie cookie;
6656 /* Scale the probability from 16-bit to 32-bit while representing
6657 * the same percentage. */
6658 uint32_t probability = (os->probability << 16) | os->probability;
6660 commit_odp_actions(&ctx->xin->flow, &ctx->base_flow,
6661 &ctx->xout->odp_actions);
6663 compose_flow_sample_cookie(os->probability, os->collector_set_id,
6664 os->obs_domain_id, os->obs_point_id, &cookie);
6665 compose_sample_action(ctx->ofproto, &ctx->xout->odp_actions, &ctx->xin->flow,
6666 probability, &cookie, sizeof cookie.flow_sample);
6670 may_receive(const struct ofport_dpif *port, struct xlate_ctx *ctx)
6672 if (port->up.pp.config & (eth_addr_equals(ctx->xin->flow.dl_dst,
6674 ? OFPUTIL_PC_NO_RECV_STP
6675 : OFPUTIL_PC_NO_RECV)) {
6679 /* Only drop packets here if both forwarding and learning are
6680 * disabled. If just learning is enabled, we need to have
6681 * OFPP_NORMAL and the learning action have a look at the packet
6682 * before we can drop it. */
6683 if (!stp_forward_in_state(port->stp_state)
6684 && !stp_learn_in_state(port->stp_state)) {
6692 tunnel_ecn_ok(struct xlate_ctx *ctx)
6694 if (is_ip_any(&ctx->base_flow)
6695 && (ctx->base_flow.tunnel.ip_tos & IP_ECN_MASK) == IP_ECN_CE) {
6696 if ((ctx->base_flow.nw_tos & IP_ECN_MASK) == IP_ECN_NOT_ECT) {
6697 VLOG_WARN_RL(&rl, "dropping tunnel packet marked ECN CE"
6698 " but is not ECN capable");
6701 /* Set the ECN CE value in the tunneled packet. */
6702 ctx->xin->flow.nw_tos |= IP_ECN_CE;
6710 do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len,
6711 struct xlate_ctx *ctx)
6713 bool was_evictable = true;
6714 const struct ofpact *a;
6717 /* Don't let the rule we're working on get evicted underneath us. */
6718 was_evictable = ctx->rule->up.evictable;
6719 ctx->rule->up.evictable = false;
6722 do_xlate_actions_again:
6723 OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) {
6724 struct ofpact_controller *controller;
6725 const struct ofpact_metadata *metadata;
6733 xlate_output_action(ctx, ofpact_get_OUTPUT(a)->port,
6734 ofpact_get_OUTPUT(a)->max_len, true);
6737 case OFPACT_CONTROLLER:
6738 controller = ofpact_get_CONTROLLER(a);
6739 execute_controller_action(ctx, controller->max_len,
6741 controller->controller_id);
6744 case OFPACT_ENQUEUE:
6745 xlate_enqueue_action(ctx, ofpact_get_ENQUEUE(a));
6748 case OFPACT_SET_VLAN_VID:
6749 ctx->xin->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
6750 ctx->xin->flow.vlan_tci |=
6751 (htons(ofpact_get_SET_VLAN_VID(a)->vlan_vid)
6755 case OFPACT_SET_VLAN_PCP:
6756 ctx->xin->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
6757 ctx->xin->flow.vlan_tci |=
6758 htons((ofpact_get_SET_VLAN_PCP(a)->vlan_pcp << VLAN_PCP_SHIFT)
6762 case OFPACT_STRIP_VLAN:
6763 ctx->xin->flow.vlan_tci = htons(0);
6766 case OFPACT_PUSH_VLAN:
6767 /* XXX 802.1AD(QinQ) */
6768 ctx->xin->flow.vlan_tci = htons(VLAN_CFI);
6771 case OFPACT_SET_ETH_SRC:
6772 memcpy(ctx->xin->flow.dl_src, ofpact_get_SET_ETH_SRC(a)->mac,
6776 case OFPACT_SET_ETH_DST:
6777 memcpy(ctx->xin->flow.dl_dst, ofpact_get_SET_ETH_DST(a)->mac,
6781 case OFPACT_SET_IPV4_SRC:
6782 memset(&ctx->xout->wc.masks.dl_type, 0xff,
6783 sizeof ctx->xout->wc.masks.dl_type);
6784 if (ctx->xin->flow.dl_type == htons(ETH_TYPE_IP)) {
6785 ctx->xin->flow.nw_src = ofpact_get_SET_IPV4_SRC(a)->ipv4;
6789 case OFPACT_SET_IPV4_DST:
6790 memset(&ctx->xout->wc.masks.dl_type, 0xff,
6791 sizeof ctx->xout->wc.masks.dl_type);
6792 if (ctx->xin->flow.dl_type == htons(ETH_TYPE_IP)) {
6793 ctx->xin->flow.nw_dst = ofpact_get_SET_IPV4_DST(a)->ipv4;
6797 case OFPACT_SET_IPV4_DSCP:
6798 /* OpenFlow 1.0 only supports IPv4. */
6799 memset(&ctx->xout->wc.masks.dl_type, 0xff,
6800 sizeof ctx->xout->wc.masks.dl_type);
6801 if (ctx->xin->flow.dl_type == htons(ETH_TYPE_IP)) {
6802 ctx->xin->flow.nw_tos &= ~IP_DSCP_MASK;
6803 ctx->xin->flow.nw_tos |= ofpact_get_SET_IPV4_DSCP(a)->dscp;
6807 case OFPACT_SET_L4_SRC_PORT:
6808 memset(&ctx->xout->wc.masks.dl_type, 0xff,
6809 sizeof ctx->xout->wc.masks.dl_type);
6810 memset(&ctx->xout->wc.masks.nw_proto, 0xff,
6811 sizeof ctx->xout->wc.masks.nw_proto);
6812 if (is_ip_any(&ctx->xin->flow)) {
6813 ctx->xin->flow.tp_src =
6814 htons(ofpact_get_SET_L4_SRC_PORT(a)->port);
6818 case OFPACT_SET_L4_DST_PORT:
6819 memset(&ctx->xout->wc.masks.dl_type, 0xff,
6820 sizeof ctx->xout->wc.masks.dl_type);
6821 memset(&ctx->xout->wc.masks.nw_proto, 0xff,
6822 sizeof ctx->xout->wc.masks.nw_proto);
6823 if (is_ip_any(&ctx->xin->flow)) {
6824 ctx->xin->flow.tp_dst =
6825 htons(ofpact_get_SET_L4_DST_PORT(a)->port);
6829 case OFPACT_RESUBMIT:
6830 xlate_ofpact_resubmit(ctx, ofpact_get_RESUBMIT(a));
6833 case OFPACT_SET_TUNNEL:
6834 ctx->xin->flow.tunnel.tun_id =
6835 htonll(ofpact_get_SET_TUNNEL(a)->tun_id);
6838 case OFPACT_SET_QUEUE:
6839 xlate_set_queue_action(ctx, ofpact_get_SET_QUEUE(a)->queue_id);
6842 case OFPACT_POP_QUEUE:
6843 memset(&ctx->xout->wc.masks.skb_priority, 0xff,
6844 sizeof ctx->xout->wc.masks.skb_priority);
6846 ctx->xin->flow.skb_priority = ctx->orig_skb_priority;
6849 case OFPACT_REG_MOVE:
6850 nxm_execute_reg_move(ofpact_get_REG_MOVE(a), &ctx->xin->flow,
6854 case OFPACT_REG_LOAD:
6855 nxm_execute_reg_load(ofpact_get_REG_LOAD(a), &ctx->xin->flow);
6858 case OFPACT_STACK_PUSH:
6859 nxm_execute_stack_push(ofpact_get_STACK_PUSH(a), &ctx->xin->flow,
6860 &ctx->xout->wc, &ctx->stack);
6863 case OFPACT_STACK_POP:
6864 nxm_execute_stack_pop(ofpact_get_STACK_POP(a), &ctx->xin->flow,
6868 case OFPACT_PUSH_MPLS:
6869 execute_mpls_push_action(ctx, ofpact_get_PUSH_MPLS(a)->ethertype);
6872 case OFPACT_POP_MPLS:
6873 execute_mpls_pop_action(ctx, ofpact_get_POP_MPLS(a)->ethertype);
6876 case OFPACT_SET_MPLS_TTL:
6877 if (execute_set_mpls_ttl_action(ctx,
6878 ofpact_get_SET_MPLS_TTL(a)->ttl)) {
6883 case OFPACT_DEC_MPLS_TTL:
6884 if (execute_dec_mpls_ttl_action(ctx)) {
6889 case OFPACT_DEC_TTL:
6890 memset(&ctx->xout->wc.masks.dl_type, 0xff,
6891 sizeof ctx->xout->wc.masks.dl_type);
6892 if (compose_dec_ttl(ctx, ofpact_get_DEC_TTL(a))) {
6898 /* Nothing to do. */
6901 case OFPACT_MULTIPATH:
6902 multipath_execute(ofpact_get_MULTIPATH(a), &ctx->xin->flow,
6907 ctx->ofproto->has_bundle_action = true;
6908 xlate_bundle_action(ctx, ofpact_get_BUNDLE(a));
6911 case OFPACT_OUTPUT_REG:
6912 xlate_output_reg_action(ctx, ofpact_get_OUTPUT_REG(a));
6916 xlate_learn_action(ctx, ofpact_get_LEARN(a));
6923 case OFPACT_FIN_TIMEOUT:
6924 memset(&ctx->xout->wc.masks.dl_type, 0xff,
6925 sizeof ctx->xout->wc.masks.dl_type);
6926 memset(&ctx->xout->wc.masks.nw_proto, 0xff,
6927 sizeof ctx->xout->wc.masks.nw_proto);
6928 ctx->xout->has_fin_timeout = true;
6929 xlate_fin_timeout(ctx, ofpact_get_FIN_TIMEOUT(a));
6932 case OFPACT_CLEAR_ACTIONS:
6934 * Nothing to do because writa-actions is not supported for now.
6935 * When writa-actions is supported, clear-actions also must
6936 * be supported at the same time.
6940 case OFPACT_WRITE_METADATA:
6941 metadata = ofpact_get_WRITE_METADATA(a);
6942 ctx->xin->flow.metadata &= ~metadata->mask;
6943 ctx->xin->flow.metadata |= metadata->metadata & metadata->mask;
6946 case OFPACT_GOTO_TABLE: {
6947 /* It is assumed that goto-table is the last action. */
6948 struct ofpact_goto_table *ogt = ofpact_get_GOTO_TABLE(a);
6949 struct rule_dpif *rule;
6951 ovs_assert(ctx->table_id < ogt->table_id);
6953 ctx->table_id = ogt->table_id;
6955 /* Look up a flow from the new table. */
6956 rule = rule_dpif_lookup__(ctx->ofproto, &ctx->xin->flow,
6957 &ctx->xout->wc, ctx->table_id);
6959 tag_the_flow(ctx, rule);
6961 rule = ctx_rule_hooks(ctx, rule, true);
6965 ctx->rule->up.evictable = was_evictable;
6968 was_evictable = rule->up.evictable;
6969 rule->up.evictable = false;
6971 /* Tail recursion removal. */
6972 ofpacts = rule->up.ofpacts;
6973 ofpacts_len = rule->up.ofpacts_len;
6974 goto do_xlate_actions_again;
6980 xlate_sample_action(ctx, ofpact_get_SAMPLE(a));
6987 ctx->rule->up.evictable = was_evictable;
6992 xlate_in_init(struct xlate_in *xin, struct ofproto_dpif *ofproto,
6993 const struct flow *flow,
6994 const struct initial_vals *initial_vals,
6995 struct rule_dpif *rule, uint8_t tcp_flags,
6996 const struct ofpbuf *packet)
6998 xin->ofproto = ofproto;
7000 xin->packet = packet;
7001 xin->may_learn = packet != NULL;
7003 xin->ofpacts = NULL;
7004 xin->ofpacts_len = 0;
7005 xin->tcp_flags = tcp_flags;
7006 xin->resubmit_hook = NULL;
7007 xin->report_hook = NULL;
7008 xin->resubmit_stats = NULL;
7011 xin->initial_vals = *initial_vals;
7013 xin->initial_vals.vlan_tci = xin->flow.vlan_tci;
7014 xin->initial_vals.tunnel_ip_tos = xin->flow.tunnel.ip_tos;
7019 xlate_out_uninit(struct xlate_out *xout)
7022 ofpbuf_uninit(&xout->odp_actions);
7026 /* Translates the 'ofpacts_len' bytes of "struct ofpacts" starting at 'ofpacts'
7027 * into datapath actions in 'odp_actions', using 'ctx'. */
7029 xlate_actions(struct xlate_in *xin, struct xlate_out *xout)
7031 /* Normally false. Set to true if we ever hit MAX_RESUBMIT_RECURSION, so
7032 * that in the future we always keep a copy of the original flow for
7033 * tracing purposes. */
7034 static bool hit_resubmit_limit;
7036 enum slow_path_reason special;
7037 const struct ofpact *ofpacts;
7038 struct ofport_dpif *in_port;
7039 struct flow orig_flow;
7040 struct xlate_ctx ctx;
7043 COVERAGE_INC(ofproto_dpif_xlate);
7045 /* Flow initialization rules:
7046 * - 'base_flow' must match the kernel's view of the packet at the
7047 * time that action processing starts. 'flow' represents any
7048 * transformations we wish to make through actions.
7049 * - By default 'base_flow' and 'flow' are the same since the input
7050 * packet matches the output before any actions are applied.
7051 * - When using VLAN splinters, 'base_flow''s VLAN is set to the value
7052 * of the received packet as seen by the kernel. If we later output
7053 * to another device without any modifications this will cause us to
7054 * insert a new tag since the original one was stripped off by the
7056 * - Tunnel 'flow' is largely cleared when transitioning between
7057 * the input and output stages since it does not make sense to output
7058 * a packet with the exact headers that it was received with (i.e.
7059 * the destination IP is us). The one exception is the tun_id, which
7060 * is preserved to allow use in later resubmit lookups and loads into
7062 * - Tunnel 'base_flow' is completely cleared since that is what the
7063 * kernel does. If we wish to maintain the original values an action
7064 * needs to be generated. */
7069 ctx.ofproto = xin->ofproto;
7070 ctx.rule = xin->rule;
7072 ctx.base_flow = ctx.xin->flow;
7073 memset(&ctx.base_flow.tunnel, 0, sizeof ctx.base_flow.tunnel);
7074 ctx.base_flow.vlan_tci = xin->initial_vals.vlan_tci;
7075 ctx.base_flow.tunnel.ip_tos = xin->initial_vals.tunnel_ip_tos;
7077 flow_wildcards_init_catchall(&ctx.xout->wc);
7078 memset(&ctx.xout->wc.masks.in_port, 0xff,
7079 sizeof ctx.xout->wc.masks.in_port);
7081 if (tnl_port_should_receive(&ctx.xin->flow)) {
7082 memset(&ctx.xout->wc.masks.tunnel, 0xff,
7083 sizeof ctx.xout->wc.masks.tunnel);
7086 /* Disable most wildcarding for NetFlow. */
7087 if (xin->ofproto->netflow) {
7088 memset(&ctx.xout->wc.masks.dl_src, 0xff,
7089 sizeof ctx.xout->wc.masks.dl_src);
7090 memset(&ctx.xout->wc.masks.dl_dst, 0xff,
7091 sizeof ctx.xout->wc.masks.dl_dst);
7092 memset(&ctx.xout->wc.masks.dl_type, 0xff,
7093 sizeof ctx.xout->wc.masks.dl_type);
7094 memset(&ctx.xout->wc.masks.vlan_tci, 0xff,
7095 sizeof ctx.xout->wc.masks.vlan_tci);
7096 memset(&ctx.xout->wc.masks.nw_proto, 0xff,
7097 sizeof ctx.xout->wc.masks.nw_proto);
7098 memset(&ctx.xout->wc.masks.nw_src, 0xff,
7099 sizeof ctx.xout->wc.masks.nw_src);
7100 memset(&ctx.xout->wc.masks.nw_dst, 0xff,
7101 sizeof ctx.xout->wc.masks.nw_dst);
7102 memset(&ctx.xout->wc.masks.tp_src, 0xff,
7103 sizeof ctx.xout->wc.masks.tp_src);
7104 memset(&ctx.xout->wc.masks.tp_dst, 0xff,
7105 sizeof ctx.xout->wc.masks.tp_dst);
7110 ctx.xout->has_learn = false;
7111 ctx.xout->has_normal = false;
7112 ctx.xout->has_fin_timeout = false;
7113 ctx.xout->nf_output_iface = NF_OUT_DROP;
7114 ctx.xout->mirrors = 0;
7116 ofpbuf_use_stub(&ctx.xout->odp_actions, ctx.xout->odp_actions_stub,
7117 sizeof ctx.xout->odp_actions_stub);
7118 ofpbuf_reserve(&ctx.xout->odp_actions, NL_A_U32_SIZE);
7121 ctx.max_resubmit_trigger = false;
7122 ctx.orig_skb_priority = ctx.xin->flow.skb_priority;
7127 ofpacts = xin->ofpacts;
7128 ofpacts_len = xin->ofpacts_len;
7129 } else if (xin->rule) {
7130 ofpacts = xin->rule->up.ofpacts;
7131 ofpacts_len = xin->rule->up.ofpacts_len;
7136 ofpbuf_use_stub(&ctx.stack, ctx.init_stack, sizeof ctx.init_stack);
7138 if (ctx.ofproto->has_mirrors || hit_resubmit_limit) {
7139 /* Do this conditionally because the copy is expensive enough that it
7140 * shows up in profiles. */
7141 orig_flow = ctx.xin->flow;
7144 if (ctx.xin->flow.nw_frag & FLOW_NW_FRAG_ANY) {
7145 switch (ctx.ofproto->up.frag_handling) {
7146 case OFPC_FRAG_NORMAL:
7147 /* We must pretend that transport ports are unavailable. */
7148 ctx.xin->flow.tp_src = ctx.base_flow.tp_src = htons(0);
7149 ctx.xin->flow.tp_dst = ctx.base_flow.tp_dst = htons(0);
7152 case OFPC_FRAG_DROP:
7155 case OFPC_FRAG_REASM:
7158 case OFPC_FRAG_NX_MATCH:
7159 /* Nothing to do. */
7162 case OFPC_INVALID_TTL_TO_CONTROLLER:
7167 in_port = get_ofp_port(ctx.ofproto, ctx.xin->flow.in_port);
7168 special = process_special(&ctx, &ctx.xin->flow, in_port,
7171 ctx.xout->slow = special;
7173 static struct vlog_rate_limit trace_rl = VLOG_RATE_LIMIT_INIT(1, 1);
7174 struct initial_vals initial_vals;
7175 size_t sample_actions_len;
7176 uint32_t local_odp_port;
7178 initial_vals.vlan_tci = ctx.base_flow.vlan_tci;
7179 initial_vals.tunnel_ip_tos = ctx.base_flow.tunnel.ip_tos;
7181 add_sflow_action(&ctx);
7182 add_ipfix_action(&ctx);
7183 sample_actions_len = ctx.xout->odp_actions.size;
7185 if (tunnel_ecn_ok(&ctx) && (!in_port || may_receive(in_port, &ctx))) {
7186 do_xlate_actions(ofpacts, ofpacts_len, &ctx);
7188 /* We've let OFPP_NORMAL and the learning action look at the
7189 * packet, so drop it now if forwarding is disabled. */
7190 if (in_port && !stp_forward_in_state(in_port->stp_state)) {
7191 ctx.xout->odp_actions.size = sample_actions_len;
7195 if (ctx.max_resubmit_trigger && !ctx.xin->resubmit_hook) {
7196 if (!hit_resubmit_limit) {
7197 /* We didn't record the original flow. Make sure we do from
7199 hit_resubmit_limit = true;
7200 } else if (!VLOG_DROP_ERR(&trace_rl)) {
7201 struct ds ds = DS_EMPTY_INITIALIZER;
7203 ofproto_trace(ctx.ofproto, &orig_flow, ctx.xin->packet,
7204 &initial_vals, &ds);
7205 VLOG_ERR("Trace triggered by excessive resubmit "
7206 "recursion:\n%s", ds_cstr(&ds));
7211 local_odp_port = ofp_port_to_odp_port(ctx.ofproto, OFPP_LOCAL);
7212 if (!connmgr_must_output_local(ctx.ofproto->up.connmgr, &ctx.xin->flow,
7214 ctx.xout->odp_actions.data,
7215 ctx.xout->odp_actions.size)) {
7216 compose_output_action(&ctx, OFPP_LOCAL);
7218 if (ctx.ofproto->has_mirrors) {
7219 add_mirror_actions(&ctx, &orig_flow);
7221 fix_sflow_action(&ctx);
7224 ofpbuf_uninit(&ctx.stack);
7226 /* Clear the metadata and register wildcard masks, because we won't
7227 * use non-header fields as part of the cache. */
7228 memset(&ctx.xout->wc.masks.metadata, 0,
7229 sizeof ctx.xout->wc.masks.metadata);
7230 memset(&ctx.xout->wc.masks.regs, 0, sizeof ctx.xout->wc.masks.regs);
7233 /* Translates the 'ofpacts_len' bytes of "struct ofpact"s starting at 'ofpacts'
7234 * into datapath actions, using 'ctx', and discards the datapath actions. */
7236 xlate_actions_for_side_effects(struct xlate_in *xin)
7238 struct xlate_out xout;
7240 xlate_actions(xin, &xout);
7241 xlate_out_uninit(&xout);
7245 xlate_report(struct xlate_ctx *ctx, const char *s)
7247 if (ctx->xin->report_hook) {
7248 ctx->xin->report_hook(ctx, s);
7253 xlate_out_copy(struct xlate_out *dst, const struct xlate_out *src)
7256 dst->tags = src->tags;
7257 dst->slow = src->slow;
7258 dst->has_learn = src->has_learn;
7259 dst->has_normal = src->has_normal;
7260 dst->has_fin_timeout = src->has_fin_timeout;
7261 dst->nf_output_iface = src->nf_output_iface;
7262 dst->mirrors = src->mirrors;
7264 ofpbuf_use_stub(&dst->odp_actions, dst->odp_actions_stub,
7265 sizeof dst->odp_actions_stub);
7266 ofpbuf_put(&dst->odp_actions, src->odp_actions.data,
7267 src->odp_actions.size);
7270 /* OFPP_NORMAL implementation. */
7272 static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *);
7274 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
7275 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle',
7276 * the bundle on which the packet was received, returns the VLAN to which the
7279 * Both 'vid' and the return value are in the range 0...4095. */
7281 input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid)
7283 switch (in_bundle->vlan_mode) {
7284 case PORT_VLAN_ACCESS:
7285 return in_bundle->vlan;
7288 case PORT_VLAN_TRUNK:
7291 case PORT_VLAN_NATIVE_UNTAGGED:
7292 case PORT_VLAN_NATIVE_TAGGED:
7293 return vid ? vid : in_bundle->vlan;
7300 /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'.
7301 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
7304 * 'vid' should be the VID obtained from the 802.1Q header that was received as
7305 * part of a packet (specify 0 if there was no 802.1Q header), in the range
7308 input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn)
7310 /* Allow any VID on the OFPP_NONE port. */
7311 if (in_bundle == &ofpp_none_bundle) {
7315 switch (in_bundle->vlan_mode) {
7316 case PORT_VLAN_ACCESS:
7319 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7320 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
7321 "packet received on port %s configured as VLAN "
7322 "%"PRIu16" access port",
7323 in_bundle->ofproto->up.name, vid,
7324 in_bundle->name, in_bundle->vlan);
7330 case PORT_VLAN_NATIVE_UNTAGGED:
7331 case PORT_VLAN_NATIVE_TAGGED:
7333 /* Port must always carry its native VLAN. */
7337 case PORT_VLAN_TRUNK:
7338 if (!ofbundle_includes_vlan(in_bundle, vid)) {
7340 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7341 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet "
7342 "received on port %s not configured for trunking "
7344 in_bundle->ofproto->up.name, vid,
7345 in_bundle->name, vid);
7357 /* Given 'vlan', the VLAN that a packet belongs to, and
7358 * 'out_bundle', a bundle on which the packet is to be output, returns the VID
7359 * that should be included in the 802.1Q header. (If the return value is 0,
7360 * then the 802.1Q header should only be included in the packet if there is a
7363 * Both 'vlan' and the return value are in the range 0...4095. */
7365 output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan)
7367 switch (out_bundle->vlan_mode) {
7368 case PORT_VLAN_ACCESS:
7371 case PORT_VLAN_TRUNK:
7372 case PORT_VLAN_NATIVE_TAGGED:
7375 case PORT_VLAN_NATIVE_UNTAGGED:
7376 return vlan == out_bundle->vlan ? 0 : vlan;
7384 output_normal(struct xlate_ctx *ctx, const struct ofbundle *out_bundle,
7387 struct ofport_dpif *port;
7389 ovs_be16 tci, old_tci;
7391 vid = output_vlan_to_vid(out_bundle, vlan);
7392 if (!out_bundle->bond) {
7393 port = ofbundle_get_a_port(out_bundle);
7395 port = bond_choose_output_slave(out_bundle->bond, &ctx->xin->flow,
7396 &ctx->xout->wc, vid, &ctx->xout->tags);
7398 /* No slaves enabled, so drop packet. */
7403 old_tci = ctx->xin->flow.vlan_tci;
7405 if (tci || out_bundle->use_priority_tags) {
7406 tci |= ctx->xin->flow.vlan_tci & htons(VLAN_PCP_MASK);
7408 tci |= htons(VLAN_CFI);
7411 ctx->xin->flow.vlan_tci = tci;
7413 compose_output_action(ctx, port->up.ofp_port);
7414 ctx->xin->flow.vlan_tci = old_tci;
7418 mirror_mask_ffs(mirror_mask_t mask)
7420 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
7425 ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
7427 return (bundle->vlan_mode != PORT_VLAN_ACCESS
7428 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
7432 ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan)
7434 return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan);
7437 /* Returns an arbitrary interface within 'bundle'. */
7438 static struct ofport_dpif *
7439 ofbundle_get_a_port(const struct ofbundle *bundle)
7441 return CONTAINER_OF(list_front(&bundle->ports),
7442 struct ofport_dpif, bundle_node);
7446 vlan_is_mirrored(const struct ofmirror *m, int vlan)
7448 return !m->vlans || bitmap_is_set(m->vlans, vlan);
7452 add_mirror_actions(struct xlate_ctx *ctx, const struct flow *orig_flow)
7454 struct ofproto_dpif *ofproto = ctx->ofproto;
7455 mirror_mask_t mirrors;
7456 struct ofbundle *in_bundle;
7459 const struct nlattr *a;
7462 in_bundle = lookup_input_bundle(ctx->ofproto, orig_flow->in_port,
7463 ctx->xin->packet != NULL, NULL);
7467 mirrors = in_bundle->src_mirrors;
7469 /* Drop frames on bundles reserved for mirroring. */
7470 if (in_bundle->mirror_out) {
7471 if (ctx->xin->packet != NULL) {
7472 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7473 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
7474 "%s, which is reserved exclusively for mirroring",
7475 ctx->ofproto->up.name, in_bundle->name);
7481 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
7482 if (!input_vid_is_valid(vid, in_bundle, ctx->xin->packet != NULL)) {
7485 vlan = input_vid_to_vlan(in_bundle, vid);
7487 /* Look at the output ports to check for destination selections. */
7489 NL_ATTR_FOR_EACH (a, left, ctx->xout->odp_actions.data,
7490 ctx->xout->odp_actions.size) {
7491 enum ovs_action_attr type = nl_attr_type(a);
7492 struct ofport_dpif *ofport;
7494 if (type != OVS_ACTION_ATTR_OUTPUT) {
7498 ofport = get_odp_port(ofproto, nl_attr_get_u32(a));
7499 if (ofport && ofport->bundle) {
7500 mirrors |= ofport->bundle->dst_mirrors;
7508 /* Restore the original packet before adding the mirror actions. */
7509 ctx->xin->flow = *orig_flow;
7514 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
7517 ctx->xout->wc.masks.vlan_tci |= htons(VLAN_CFI | VLAN_VID_MASK);
7520 if (!vlan_is_mirrored(m, vlan)) {
7521 mirrors = zero_rightmost_1bit(mirrors);
7525 mirrors &= ~m->dup_mirrors;
7526 ctx->xout->mirrors |= m->dup_mirrors;
7528 output_normal(ctx, m->out, vlan);
7529 } else if (vlan != m->out_vlan
7530 && !eth_addr_is_reserved(orig_flow->dl_dst)) {
7531 struct ofbundle *bundle;
7533 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
7534 if (ofbundle_includes_vlan(bundle, m->out_vlan)
7535 && !bundle->mirror_out) {
7536 output_normal(ctx, bundle, m->out_vlan);
7544 update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors,
7545 uint64_t packets, uint64_t bytes)
7551 for (; mirrors; mirrors = zero_rightmost_1bit(mirrors)) {
7554 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
7557 /* In normal circumstances 'm' will not be NULL. However,
7558 * if mirrors are reconfigured, we can temporarily get out
7559 * of sync in facet_revalidate(). We could "correct" the
7560 * mirror list before reaching here, but doing that would
7561 * not properly account the traffic stats we've currently
7562 * accumulated for previous mirror configuration. */
7566 m->packet_count += packets;
7567 m->byte_count += bytes;
7571 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
7572 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
7573 * indicate this; newer upstream kernels use gratuitous ARP requests. */
7575 is_gratuitous_arp(const struct flow *flow, struct flow_wildcards *wc)
7577 if (flow->dl_type != htons(ETH_TYPE_ARP)) {
7581 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
7582 if (!eth_addr_is_broadcast(flow->dl_dst)) {
7586 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
7587 if (flow->nw_proto == ARP_OP_REPLY) {
7589 } else if (flow->nw_proto == ARP_OP_REQUEST) {
7590 memset(&wc->masks.nw_src, 0xff, sizeof wc->masks.nw_src);
7591 memset(&wc->masks.nw_dst, 0xff, sizeof wc->masks.nw_dst);
7593 return flow->nw_src == flow->nw_dst;
7600 update_learning_table(struct ofproto_dpif *ofproto,
7601 const struct flow *flow, struct flow_wildcards *wc,
7602 int vlan, struct ofbundle *in_bundle)
7604 struct mac_entry *mac;
7606 /* Don't learn the OFPP_NONE port. */
7607 if (in_bundle == &ofpp_none_bundle) {
7611 if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
7615 mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan);
7616 if (is_gratuitous_arp(flow, wc)) {
7617 /* We don't want to learn from gratuitous ARP packets that are
7618 * reflected back over bond slaves so we lock the learning table. */
7619 if (!in_bundle->bond) {
7620 mac_entry_set_grat_arp_lock(mac);
7621 } else if (mac_entry_is_grat_arp_locked(mac)) {
7626 if (mac_entry_is_new(mac) || mac->port.p != in_bundle) {
7627 /* The log messages here could actually be useful in debugging,
7628 * so keep the rate limit relatively high. */
7629 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
7630 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
7631 "on port %s in VLAN %d",
7632 ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src),
7633 in_bundle->name, vlan);
7635 mac->port.p = in_bundle;
7636 tag_set_add(&ofproto->backer->revalidate_set,
7637 mac_learning_changed(ofproto->ml, mac));
7641 static struct ofbundle *
7642 lookup_input_bundle(const struct ofproto_dpif *ofproto, uint16_t in_port,
7643 bool warn, struct ofport_dpif **in_ofportp)
7645 struct ofport_dpif *ofport;
7647 /* Find the port and bundle for the received packet. */
7648 ofport = get_ofp_port(ofproto, in_port);
7650 *in_ofportp = ofport;
7652 if (ofport && ofport->bundle) {
7653 return ofport->bundle;
7656 /* Special-case OFPP_NONE, which a controller may use as the ingress
7657 * port for traffic that it is sourcing. */
7658 if (in_port == OFPP_NONE) {
7659 return &ofpp_none_bundle;
7662 /* Odd. A few possible reasons here:
7664 * - We deleted a port but there are still a few packets queued up
7667 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
7668 * we don't know about.
7670 * - The ofproto client didn't configure the port as part of a bundle.
7671 * This is particularly likely to happen if a packet was received on the
7672 * port after it was created, but before the client had a chance to
7673 * configure its bundle.
7676 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7678 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
7679 "port %"PRIu16, ofproto->up.name, in_port);
7684 /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or
7685 * dropped. Returns true if they may be forwarded, false if they should be
7688 * 'in_port' must be the ofport_dpif that corresponds to flow->in_port.
7689 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
7691 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
7692 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
7693 * checked by input_vid_is_valid().
7695 * May also add tags to '*tags', although the current implementation only does
7696 * so in one special case.
7699 is_admissible(struct xlate_ctx *ctx, struct ofport_dpif *in_port,
7702 struct ofproto_dpif *ofproto = ctx->ofproto;
7703 struct flow *flow = &ctx->xin->flow;
7704 struct ofbundle *in_bundle = in_port->bundle;
7706 /* Drop frames for reserved multicast addresses
7707 * only if forward_bpdu option is absent. */
7708 if (!ofproto->up.forward_bpdu && eth_addr_is_reserved(flow->dl_dst)) {
7709 xlate_report(ctx, "packet has reserved destination MAC, dropping");
7713 if (in_bundle->bond) {
7714 struct mac_entry *mac;
7716 switch (bond_check_admissibility(in_bundle->bond, in_port,
7717 flow->dl_dst, &ctx->xout->tags)) {
7722 xlate_report(ctx, "bonding refused admissibility, dropping");
7725 case BV_DROP_IF_MOVED:
7726 mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL);
7727 if (mac && mac->port.p != in_bundle &&
7728 (!is_gratuitous_arp(flow, &ctx->xout->wc)
7729 || mac_entry_is_grat_arp_locked(mac))) {
7730 xlate_report(ctx, "SLB bond thinks this packet looped back, "
7742 xlate_normal(struct xlate_ctx *ctx)
7744 struct ofport_dpif *in_port;
7745 struct ofbundle *in_bundle;
7746 struct mac_entry *mac;
7750 ctx->xout->has_normal = true;
7752 /* Check the dl_type, since we may check for gratuituous ARP. */
7753 memset(&ctx->xout->wc.masks.dl_type, 0xff,
7754 sizeof ctx->xout->wc.masks.dl_type);
7756 memset(&ctx->xout->wc.masks.dl_src, 0xff,
7757 sizeof ctx->xout->wc.masks.dl_src);
7758 memset(&ctx->xout->wc.masks.dl_dst, 0xff,
7759 sizeof ctx->xout->wc.masks.dl_dst);
7760 memset(&ctx->xout->wc.masks.vlan_tci, 0xff,
7761 sizeof ctx->xout->wc.masks.vlan_tci);
7763 in_bundle = lookup_input_bundle(ctx->ofproto, ctx->xin->flow.in_port,
7764 ctx->xin->packet != NULL, &in_port);
7766 xlate_report(ctx, "no input bundle, dropping");
7770 /* Drop malformed frames. */
7771 if (ctx->xin->flow.dl_type == htons(ETH_TYPE_VLAN) &&
7772 !(ctx->xin->flow.vlan_tci & htons(VLAN_CFI))) {
7773 if (ctx->xin->packet != NULL) {
7774 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7775 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
7776 "VLAN tag received on port %s",
7777 ctx->ofproto->up.name, in_bundle->name);
7779 xlate_report(ctx, "partial VLAN tag, dropping");
7783 /* Drop frames on bundles reserved for mirroring. */
7784 if (in_bundle->mirror_out) {
7785 if (ctx->xin->packet != NULL) {
7786 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7787 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
7788 "%s, which is reserved exclusively for mirroring",
7789 ctx->ofproto->up.name, in_bundle->name);
7791 xlate_report(ctx, "input port is mirror output port, dropping");
7796 vid = vlan_tci_to_vid(ctx->xin->flow.vlan_tci);
7797 if (!input_vid_is_valid(vid, in_bundle, ctx->xin->packet != NULL)) {
7798 xlate_report(ctx, "disallowed VLAN VID for this input port, dropping");
7801 vlan = input_vid_to_vlan(in_bundle, vid);
7803 /* Check other admissibility requirements. */
7804 if (in_port && !is_admissible(ctx, in_port, vlan)) {
7808 /* Learn source MAC. */
7809 if (ctx->xin->may_learn) {
7810 update_learning_table(ctx->ofproto, &ctx->xin->flow, &ctx->xout->wc,
7814 /* Determine output bundle. */
7815 mac = mac_learning_lookup(ctx->ofproto->ml, ctx->xin->flow.dl_dst, vlan,
7818 if (mac->port.p != in_bundle) {
7819 xlate_report(ctx, "forwarding to learned port");
7820 output_normal(ctx, mac->port.p, vlan);
7822 xlate_report(ctx, "learned port is input port, dropping");
7825 struct ofbundle *bundle;
7827 xlate_report(ctx, "no learned MAC for destination, flooding");
7828 HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
7829 if (bundle != in_bundle
7830 && ofbundle_includes_vlan(bundle, vlan)
7831 && bundle->floodable
7832 && !bundle->mirror_out) {
7833 output_normal(ctx, bundle, vlan);
7836 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
7840 /* Optimized flow revalidation.
7842 * It's a difficult problem, in general, to tell which facets need to have
7843 * their actions recalculated whenever the OpenFlow flow table changes. We
7844 * don't try to solve that general problem: for most kinds of OpenFlow flow
7845 * table changes, we recalculate the actions for every facet. This is
7846 * relatively expensive, but it's good enough if the OpenFlow flow table
7847 * doesn't change very often.
7849 * However, we can expect one particular kind of OpenFlow flow table change to
7850 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
7851 * of CPU on revalidating every facet whenever MAC learning modifies the flow
7852 * table, we add a special case that applies to flow tables in which every rule
7853 * has the same form (that is, the same wildcards), except that the table is
7854 * also allowed to have a single "catch-all" flow that matches all packets. We
7855 * optimize this case by tagging all of the facets that resubmit into the table
7856 * and invalidating the same tag whenever a flow changes in that table. The
7857 * end result is that we revalidate just the facets that need it (and sometimes
7858 * a few more, but not all of the facets or even all of the facets that
7859 * resubmit to the table modified by MAC learning). */
7861 /* Calculates the tag to use for 'flow' and mask 'mask' when it is inserted
7862 * into an OpenFlow table with the given 'basis'. */
7864 rule_calculate_tag(const struct flow *flow, const struct minimask *mask,
7867 if (minimask_is_catchall(mask)) {
7870 uint32_t hash = flow_hash_in_minimask(flow, mask, secret);
7871 return tag_create_deterministic(hash);
7875 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
7876 * taggability of that table.
7878 * This function must be called after *each* change to a flow table. If you
7879 * skip calling it on some changes then the pointer comparisons at the end can
7880 * be invalid if you get unlucky. For example, if a flow removal causes a
7881 * cls_table to be destroyed and then a flow insertion causes a cls_table with
7882 * different wildcards to be created with the same address, then this function
7883 * will incorrectly skip revalidation. */
7885 table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
7887 struct table_dpif *table = &ofproto->tables[table_id];
7888 const struct oftable *oftable = &ofproto->up.tables[table_id];
7889 struct cls_table *catchall, *other;
7890 struct cls_table *t;
7892 catchall = other = NULL;
7894 switch (hmap_count(&oftable->cls.tables)) {
7896 /* We could tag this OpenFlow table but it would make the logic a
7897 * little harder and it's a corner case that doesn't seem worth it
7903 HMAP_FOR_EACH (t, hmap_node, &oftable->cls.tables) {
7904 if (cls_table_is_catchall(t)) {
7906 } else if (!other) {
7909 /* Indicate that we can't tag this by setting both tables to
7910 * NULL. (We know that 'catchall' is already NULL.) */
7917 /* Can't tag this table. */
7921 if (table->catchall_table != catchall || table->other_table != other) {
7922 table->catchall_table = catchall;
7923 table->other_table = other;
7924 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
7928 /* Given 'rule' that has changed in some way (either it is a rule being
7929 * inserted, a rule being deleted, or a rule whose actions are being
7930 * modified), marks facets for revalidation to ensure that packets will be
7931 * forwarded correctly according to the new state of the flow table.
7933 * This function must be called after *each* change to a flow table. See
7934 * the comment on table_update_taggable() for more information. */
7936 rule_invalidate(const struct rule_dpif *rule)
7938 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
7940 table_update_taggable(ofproto, rule->up.table_id);
7942 if (!ofproto->backer->need_revalidate) {
7943 struct table_dpif *table = &ofproto->tables[rule->up.table_id];
7945 if (table->other_table && rule->tag) {
7946 tag_set_add(&ofproto->backer->revalidate_set, rule->tag);
7948 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
7954 set_frag_handling(struct ofproto *ofproto_,
7955 enum ofp_config_flags frag_handling)
7957 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7958 if (frag_handling != OFPC_FRAG_REASM) {
7959 ofproto->backer->need_revalidate = REV_RECONFIGURE;
7967 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
7968 const struct flow *flow,
7969 const struct ofpact *ofpacts, size_t ofpacts_len)
7971 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7972 struct initial_vals initial_vals;
7973 struct odputil_keybuf keybuf;
7974 struct dpif_flow_stats stats;
7975 struct xlate_out xout;
7976 struct xlate_in xin;
7980 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
7981 odp_flow_key_from_flow(&key, flow,
7982 ofp_port_to_odp_port(ofproto, flow->in_port));
7984 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
7986 initial_vals.vlan_tci = flow->vlan_tci;
7987 initial_vals.tunnel_ip_tos = 0;
7988 xlate_in_init(&xin, ofproto, flow, &initial_vals, NULL, stats.tcp_flags,
7990 xin.resubmit_stats = &stats;
7991 xin.ofpacts_len = ofpacts_len;
7992 xin.ofpacts = ofpacts;
7994 xlate_actions(&xin, &xout);
7995 dpif_execute(ofproto->backer->dpif, key.data, key.size,
7996 xout.odp_actions.data, xout.odp_actions.size, packet);
7997 xlate_out_uninit(&xout);
8005 set_netflow(struct ofproto *ofproto_,
8006 const struct netflow_options *netflow_options)
8008 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
8010 if (netflow_options) {
8011 if (!ofproto->netflow) {
8012 ofproto->netflow = netflow_create();
8014 return netflow_set_options(ofproto->netflow, netflow_options);
8016 netflow_destroy(ofproto->netflow);
8017 ofproto->netflow = NULL;
8023 get_netflow_ids(const struct ofproto *ofproto_,
8024 uint8_t *engine_type, uint8_t *engine_id)
8026 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
8028 dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
8032 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
8034 if (!facet_is_controller_flow(facet) &&
8035 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
8036 struct subfacet *subfacet;
8037 struct ofexpired expired;
8039 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
8040 if (subfacet->path == SF_FAST_PATH) {
8041 struct dpif_flow_stats stats;
8043 subfacet_install(subfacet, &facet->xout.odp_actions,
8045 subfacet_update_stats(subfacet, &stats);
8049 expired.flow = facet->flow;
8050 expired.packet_count = facet->packet_count;
8051 expired.byte_count = facet->byte_count;
8052 expired.used = facet->used;
8053 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
8058 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
8060 struct cls_cursor cursor;
8061 struct facet *facet;
8063 cls_cursor_init(&cursor, &ofproto->facets, NULL);
8064 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
8065 send_active_timeout(ofproto, facet);
8069 static struct ofproto_dpif *
8070 ofproto_dpif_lookup(const char *name)
8072 struct ofproto_dpif *ofproto;
8074 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
8075 hash_string(name, 0), &all_ofproto_dpifs) {
8076 if (!strcmp(ofproto->up.name, name)) {
8084 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
8085 const char *argv[], void *aux OVS_UNUSED)
8087 struct ofproto_dpif *ofproto;
8090 ofproto = ofproto_dpif_lookup(argv[1]);
8092 unixctl_command_reply_error(conn, "no such bridge");
8095 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
8097 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
8098 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
8102 unixctl_command_reply(conn, "table successfully flushed");
8106 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
8107 const char *argv[], void *aux OVS_UNUSED)
8109 struct ds ds = DS_EMPTY_INITIALIZER;
8110 const struct ofproto_dpif *ofproto;
8111 const struct mac_entry *e;
8113 ofproto = ofproto_dpif_lookup(argv[1]);
8115 unixctl_command_reply_error(conn, "no such bridge");
8119 ds_put_cstr(&ds, " port VLAN MAC Age\n");
8120 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
8121 struct ofbundle *bundle = e->port.p;
8122 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
8123 ofbundle_get_a_port(bundle)->odp_port,
8124 e->vlan, ETH_ADDR_ARGS(e->mac),
8125 mac_entry_age(ofproto->ml, e));
8127 unixctl_command_reply(conn, ds_cstr(&ds));
8132 struct xlate_out xout;
8133 struct xlate_in xin;
8139 trace_format_rule(struct ds *result, uint8_t table_id, int level,
8140 const struct rule_dpif *rule)
8142 ds_put_char_multiple(result, '\t', level);
8144 ds_put_cstr(result, "No match\n");
8148 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
8149 table_id, ntohll(rule->up.flow_cookie));
8150 cls_rule_format(&rule->up.cr, result);
8151 ds_put_char(result, '\n');
8153 ds_put_char_multiple(result, '\t', level);
8154 ds_put_cstr(result, "OpenFlow ");
8155 ofpacts_format(rule->up.ofpacts, rule->up.ofpacts_len, result);
8156 ds_put_char(result, '\n');
8160 trace_format_flow(struct ds *result, int level, const char *title,
8161 struct trace_ctx *trace)
8163 ds_put_char_multiple(result, '\t', level);
8164 ds_put_format(result, "%s: ", title);
8165 if (flow_equal(&trace->xin.flow, &trace->flow)) {
8166 ds_put_cstr(result, "unchanged");
8168 flow_format(result, &trace->xin.flow);
8169 trace->flow = trace->xin.flow;
8171 ds_put_char(result, '\n');
8175 trace_format_regs(struct ds *result, int level, const char *title,
8176 struct trace_ctx *trace)
8180 ds_put_char_multiple(result, '\t', level);
8181 ds_put_format(result, "%s:", title);
8182 for (i = 0; i < FLOW_N_REGS; i++) {
8183 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
8185 ds_put_char(result, '\n');
8189 trace_format_odp(struct ds *result, int level, const char *title,
8190 struct trace_ctx *trace)
8192 struct ofpbuf *odp_actions = &trace->xout.odp_actions;
8194 ds_put_char_multiple(result, '\t', level);
8195 ds_put_format(result, "%s: ", title);
8196 format_odp_actions(result, odp_actions->data, odp_actions->size);
8197 ds_put_char(result, '\n');
8201 trace_resubmit(struct xlate_ctx *ctx, struct rule_dpif *rule)
8203 struct trace_ctx *trace = CONTAINER_OF(ctx->xin, struct trace_ctx, xin);
8204 struct ds *result = trace->result;
8206 ds_put_char(result, '\n');
8207 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
8208 trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace);
8209 trace_format_odp(result, ctx->recurse + 1, "Resubmitted odp", trace);
8210 trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule);
8214 trace_report(struct xlate_ctx *ctx, const char *s)
8216 struct trace_ctx *trace = CONTAINER_OF(ctx->xin, struct trace_ctx, xin);
8217 struct ds *result = trace->result;
8219 ds_put_char_multiple(result, '\t', ctx->recurse);
8220 ds_put_cstr(result, s);
8221 ds_put_char(result, '\n');
8225 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
8226 void *aux OVS_UNUSED)
8228 const char *dpname = argv[1];
8229 struct ofproto_dpif *ofproto;
8230 struct ofpbuf odp_key;
8231 struct ofpbuf *packet;
8232 struct initial_vals initial_vals;
8238 ofpbuf_init(&odp_key, 0);
8241 ofproto = ofproto_dpif_lookup(dpname);
8243 unixctl_command_reply_error(conn, "Unknown ofproto (use ofproto/list "
8247 if (argc == 3 || (argc == 4 && !strcmp(argv[3], "-generate"))) {
8248 /* ofproto/trace dpname flow [-generate] */
8249 const char *flow_s = argv[2];
8250 const char *generate_s = argv[3];
8252 /* Allow 'flow_s' to be either a datapath flow or an OpenFlow-like
8253 * flow. We guess which type it is based on whether 'flow_s' contains
8254 * an '(', since a datapath flow always contains '(') but an
8255 * OpenFlow-like flow should not (in fact it's allowed but I believe
8256 * that's not documented anywhere).
8258 * An alternative would be to try to parse 'flow_s' both ways, but then
8259 * it would be tricky giving a sensible error message. After all, do
8260 * you just say "syntax error" or do you present both error messages?
8261 * Both choices seem lousy. */
8262 if (strchr(flow_s, '(')) {
8265 /* Convert string to datapath key. */
8266 ofpbuf_init(&odp_key, 0);
8267 error = odp_flow_key_from_string(flow_s, NULL, &odp_key);
8269 unixctl_command_reply_error(conn, "Bad flow syntax");
8273 /* The user might have specified the wrong ofproto but within the
8274 * same backer. That's OK, ofproto_receive() can find the right
8276 if (ofproto_receive(ofproto->backer, NULL, odp_key.data,
8277 odp_key.size, &flow, NULL, &ofproto, NULL,
8279 unixctl_command_reply_error(conn, "Invalid flow");
8282 ds_put_format(&result, "Bridge: %s\n", ofproto->up.name);
8286 error_s = parse_ofp_exact_flow(&flow, argv[2]);
8288 unixctl_command_reply_error(conn, error_s);
8293 initial_vals.vlan_tci = flow.vlan_tci;
8294 initial_vals.tunnel_ip_tos = flow.tunnel.ip_tos;
8297 /* Generate a packet, if requested. */
8299 packet = ofpbuf_new(0);
8300 flow_compose(packet, &flow);
8302 } else if (argc == 7) {
8303 /* ofproto/trace dpname priority tun_id in_port mark packet */
8304 const char *priority_s = argv[2];
8305 const char *tun_id_s = argv[3];
8306 const char *in_port_s = argv[4];
8307 const char *mark_s = argv[5];
8308 const char *packet_s = argv[6];
8309 uint32_t in_port = atoi(in_port_s);
8310 ovs_be64 tun_id = htonll(strtoull(tun_id_s, NULL, 0));
8311 uint32_t priority = atoi(priority_s);
8312 uint32_t mark = atoi(mark_s);
8315 msg = eth_from_hex(packet_s, &packet);
8317 unixctl_command_reply_error(conn, msg);
8321 ds_put_cstr(&result, "Packet: ");
8322 s = ofp_packet_to_string(packet->data, packet->size);
8323 ds_put_cstr(&result, s);
8326 flow_extract(packet, priority, mark, NULL, in_port, &flow);
8327 flow.tunnel.tun_id = tun_id;
8328 initial_vals.vlan_tci = flow.vlan_tci;
8329 initial_vals.tunnel_ip_tos = flow.tunnel.ip_tos;
8331 unixctl_command_reply_error(conn, "Bad command syntax");
8335 ofproto_trace(ofproto, &flow, packet, &initial_vals, &result);
8336 unixctl_command_reply(conn, ds_cstr(&result));
8339 ds_destroy(&result);
8340 ofpbuf_delete(packet);
8341 ofpbuf_uninit(&odp_key);
8345 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
8346 const struct ofpbuf *packet,
8347 const struct initial_vals *initial_vals, struct ds *ds)
8349 struct rule_dpif *rule;
8351 ds_put_cstr(ds, "Flow: ");
8352 flow_format(ds, flow);
8353 ds_put_char(ds, '\n');
8355 rule = rule_dpif_lookup(ofproto, flow, NULL);
8357 trace_format_rule(ds, 0, 0, rule);
8358 if (rule == ofproto->miss_rule) {
8359 ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
8360 } else if (rule == ofproto->no_packet_in_rule) {
8361 ds_put_cstr(ds, "\nNo match, packets dropped because "
8362 "OFPPC_NO_PACKET_IN is set on in_port.\n");
8363 } else if (rule == ofproto->drop_frags_rule) {
8364 ds_put_cstr(ds, "\nPackets dropped because they are IP fragments "
8365 "and the fragment handling mode is \"drop\".\n");
8369 uint64_t odp_actions_stub[1024 / 8];
8370 struct ofpbuf odp_actions;
8371 struct trace_ctx trace;
8375 tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
8378 ofpbuf_use_stub(&odp_actions,
8379 odp_actions_stub, sizeof odp_actions_stub);
8380 xlate_in_init(&trace.xin, ofproto, flow, initial_vals, rule, tcp_flags,
8382 trace.xin.resubmit_hook = trace_resubmit;
8383 trace.xin.report_hook = trace_report;
8385 xlate_actions(&trace.xin, &trace.xout);
8387 ds_put_char(ds, '\n');
8388 trace_format_flow(ds, 0, "Final flow", &trace);
8390 match_init(&match, flow, &trace.xout.wc);
8391 ds_put_cstr(ds, "Relevant fields: ");
8392 match_format(&match, ds, OFP_DEFAULT_PRIORITY);
8393 ds_put_char(ds, '\n');
8395 ds_put_cstr(ds, "Datapath actions: ");
8396 format_odp_actions(ds, trace.xout.odp_actions.data,
8397 trace.xout.odp_actions.size);
8399 if (trace.xout.slow) {
8400 ds_put_cstr(ds, "\nThis flow is handled by the userspace "
8401 "slow path because it:");
8402 switch (trace.xout.slow) {
8404 ds_put_cstr(ds, "\n\t- Consists of CFM packets.");
8407 ds_put_cstr(ds, "\n\t- Consists of LACP packets.");
8410 ds_put_cstr(ds, "\n\t- Consists of STP packets.");
8412 case SLOW_CONTROLLER:
8413 ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages "
8414 "to the OpenFlow controller.");
8421 xlate_out_uninit(&trace.xout);
8426 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
8427 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
8430 unixctl_command_reply(conn, NULL);
8434 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
8435 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
8438 unixctl_command_reply(conn, NULL);
8441 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
8442 * 'reply' describing the results. */
8444 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
8446 struct cls_cursor cursor;
8447 struct facet *facet;
8451 cls_cursor_init(&cursor, &ofproto->facets, NULL);
8452 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
8453 if (!facet_check_consistency(facet)) {
8458 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
8462 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
8463 ofproto->up.name, errors);
8465 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
8470 ofproto_dpif_self_check(struct unixctl_conn *conn,
8471 int argc, const char *argv[], void *aux OVS_UNUSED)
8473 struct ds reply = DS_EMPTY_INITIALIZER;
8474 struct ofproto_dpif *ofproto;
8477 ofproto = ofproto_dpif_lookup(argv[1]);
8479 unixctl_command_reply_error(conn, "Unknown ofproto (use "
8480 "ofproto/list for help)");
8483 ofproto_dpif_self_check__(ofproto, &reply);
8485 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
8486 ofproto_dpif_self_check__(ofproto, &reply);
8490 unixctl_command_reply(conn, ds_cstr(&reply));
8494 /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list
8495 * of the 'ofproto_shash' nodes. It is the responsibility of the caller
8496 * to destroy 'ofproto_shash' and free the returned value. */
8497 static const struct shash_node **
8498 get_ofprotos(struct shash *ofproto_shash)
8500 const struct ofproto_dpif *ofproto;
8502 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
8503 char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
8504 shash_add_nocopy(ofproto_shash, name, ofproto);
8507 return shash_sort(ofproto_shash);
8511 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
8512 const char *argv[] OVS_UNUSED,
8513 void *aux OVS_UNUSED)
8515 struct ds ds = DS_EMPTY_INITIALIZER;
8516 struct shash ofproto_shash;
8517 const struct shash_node **sorted_ofprotos;
8520 shash_init(&ofproto_shash);
8521 sorted_ofprotos = get_ofprotos(&ofproto_shash);
8522 for (i = 0; i < shash_count(&ofproto_shash); i++) {
8523 const struct shash_node *node = sorted_ofprotos[i];
8524 ds_put_format(&ds, "%s\n", node->name);
8527 shash_destroy(&ofproto_shash);
8528 free(sorted_ofprotos);
8530 unixctl_command_reply(conn, ds_cstr(&ds));
8535 show_dp_format(const struct ofproto_dpif *ofproto, struct ds *ds)
8537 const struct shash_node **ports;
8539 struct avg_subfacet_rates lifetime;
8540 unsigned long long int minutes;
8541 const int min_ms = 60 * 1000; /* milliseconds in one minute. */
8543 minutes = (time_msec() - ofproto->created) / min_ms;
8546 lifetime.add_rate = (double)ofproto->total_subfacet_add_count
8548 lifetime.del_rate = (double)ofproto->total_subfacet_del_count
8551 lifetime.add_rate = 0.0;
8552 lifetime.del_rate = 0.0;
8555 ds_put_format(ds, "%s (%s):\n", ofproto->up.name,
8556 dpif_name(ofproto->backer->dpif));
8558 "\tlookups: hit:%"PRIu64" missed:%"PRIu64"\n",
8559 ofproto->n_hit, ofproto->n_missed);
8560 ds_put_format(ds, "\tflows: cur: %zu, avg: %5.3f, max: %d,"
8561 " life span: %llu(ms)\n",
8562 hmap_count(&ofproto->subfacets),
8563 avg_subfacet_count(ofproto),
8564 ofproto->max_n_subfacet,
8565 avg_subfacet_life_span(ofproto));
8566 if (minutes >= 60) {
8567 show_dp_rates(ds, "\t\thourly avg:", &ofproto->hourly);
8569 if (minutes >= 60 * 24) {
8570 show_dp_rates(ds, "\t\tdaily avg:", &ofproto->daily);
8572 show_dp_rates(ds, "\t\toverall avg:", &lifetime);
8574 ports = shash_sort(&ofproto->up.port_by_name);
8575 for (i = 0; i < shash_count(&ofproto->up.port_by_name); i++) {
8576 const struct shash_node *node = ports[i];
8577 struct ofport *ofport = node->data;
8578 const char *name = netdev_get_name(ofport->netdev);
8579 const char *type = netdev_get_type(ofport->netdev);
8582 ds_put_format(ds, "\t%s %u/", name, ofport->ofp_port);
8584 odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
8585 if (odp_port != OVSP_NONE) {
8586 ds_put_format(ds, "%"PRIu32":", odp_port);
8588 ds_put_cstr(ds, "none:");
8591 if (strcmp(type, "system")) {
8592 struct netdev *netdev;
8595 ds_put_format(ds, " (%s", type);
8597 error = netdev_open(name, type, &netdev);
8602 error = netdev_get_config(netdev, &config);
8604 const struct smap_node **nodes;
8607 nodes = smap_sort(&config);
8608 for (i = 0; i < smap_count(&config); i++) {
8609 const struct smap_node *node = nodes[i];
8610 ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
8611 node->key, node->value);
8615 smap_destroy(&config);
8617 netdev_close(netdev);
8619 ds_put_char(ds, ')');
8621 ds_put_char(ds, '\n');
8627 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc,
8628 const char *argv[], void *aux OVS_UNUSED)
8630 struct ds ds = DS_EMPTY_INITIALIZER;
8631 const struct ofproto_dpif *ofproto;
8635 for (i = 1; i < argc; i++) {
8636 ofproto = ofproto_dpif_lookup(argv[i]);
8638 ds_put_format(&ds, "Unknown bridge %s (use dpif/dump-dps "
8639 "for help)", argv[i]);
8640 unixctl_command_reply_error(conn, ds_cstr(&ds));
8643 show_dp_format(ofproto, &ds);
8646 struct shash ofproto_shash;
8647 const struct shash_node **sorted_ofprotos;
8650 shash_init(&ofproto_shash);
8651 sorted_ofprotos = get_ofprotos(&ofproto_shash);
8652 for (i = 0; i < shash_count(&ofproto_shash); i++) {
8653 const struct shash_node *node = sorted_ofprotos[i];
8654 show_dp_format(node->data, &ds);
8657 shash_destroy(&ofproto_shash);
8658 free(sorted_ofprotos);
8661 unixctl_command_reply(conn, ds_cstr(&ds));
8665 /* Dump the megaflow (facet) cache. This is useful to check the
8666 * correctness of flow wildcarding, since the same mechanism is used for
8667 * both xlate caching and kernel wildcarding.
8669 * It's important to note that in the output the flow description uses
8670 * OpenFlow (OFP) ports, but the actions use datapath (ODP) ports.
8672 * This command is only needed for advanced debugging, so it's not
8673 * documented in the man page. */
8675 ofproto_unixctl_dpif_dump_megaflows(struct unixctl_conn *conn,
8676 int argc OVS_UNUSED, const char *argv[],
8677 void *aux OVS_UNUSED)
8679 struct ds ds = DS_EMPTY_INITIALIZER;
8680 const struct ofproto_dpif *ofproto;
8681 long long int now = time_msec();
8682 struct cls_cursor cursor;
8683 struct facet *facet;
8685 ofproto = ofproto_dpif_lookup(argv[1]);
8687 unixctl_command_reply_error(conn, "no such bridge");
8691 cls_cursor_init(&cursor, &ofproto->facets, NULL);
8692 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
8693 cls_rule_format(&facet->cr, &ds);
8694 ds_put_cstr(&ds, ", ");
8695 ds_put_format(&ds, "n_subfacets:%zu, ", list_size(&facet->subfacets));
8696 ds_put_format(&ds, "used:%.3fs, ", (now - facet->used) / 1000.0);
8697 ds_put_cstr(&ds, "Datapath actions: ");
8698 if (facet->xout.slow) {
8699 uint64_t slow_path_stub[128 / 8];
8700 const struct nlattr *actions;
8703 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
8704 slow_path_stub, sizeof slow_path_stub,
8705 &actions, &actions_len);
8706 format_odp_actions(&ds, actions, actions_len);
8708 format_odp_actions(&ds, facet->xout.odp_actions.data,
8709 facet->xout.odp_actions.size);
8711 ds_put_cstr(&ds, "\n");
8714 ds_chomp(&ds, '\n');
8715 unixctl_command_reply(conn, ds_cstr(&ds));
8720 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
8721 int argc OVS_UNUSED, const char *argv[],
8722 void *aux OVS_UNUSED)
8724 struct ds ds = DS_EMPTY_INITIALIZER;
8725 const struct ofproto_dpif *ofproto;
8726 struct subfacet *subfacet;
8728 ofproto = ofproto_dpif_lookup(argv[1]);
8730 unixctl_command_reply_error(conn, "no such bridge");
8734 update_stats(ofproto->backer);
8736 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
8737 struct facet *facet = subfacet->facet;
8739 odp_flow_key_format(subfacet->key, subfacet->key_len, &ds);
8741 ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:",
8742 subfacet->dp_packet_count, subfacet->dp_byte_count);
8743 if (subfacet->used) {
8744 ds_put_format(&ds, "%.3fs",
8745 (time_msec() - subfacet->used) / 1000.0);
8747 ds_put_format(&ds, "never");
8749 if (subfacet->facet->tcp_flags) {
8750 ds_put_cstr(&ds, ", flags:");
8751 packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags);
8754 ds_put_cstr(&ds, ", actions:");
8755 if (facet->xout.slow) {
8756 uint64_t slow_path_stub[128 / 8];
8757 const struct nlattr *actions;
8760 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
8761 slow_path_stub, sizeof slow_path_stub,
8762 &actions, &actions_len);
8763 format_odp_actions(&ds, actions, actions_len);
8765 format_odp_actions(&ds, facet->xout.odp_actions.data,
8766 facet->xout.odp_actions.size);
8768 ds_put_char(&ds, '\n');
8771 unixctl_command_reply(conn, ds_cstr(&ds));
8776 ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn,
8777 int argc OVS_UNUSED, const char *argv[],
8778 void *aux OVS_UNUSED)
8780 struct ds ds = DS_EMPTY_INITIALIZER;
8781 struct ofproto_dpif *ofproto;
8783 ofproto = ofproto_dpif_lookup(argv[1]);
8785 unixctl_command_reply_error(conn, "no such bridge");
8789 flush(&ofproto->up);
8791 unixctl_command_reply(conn, ds_cstr(&ds));
8796 ofproto_dpif_unixctl_init(void)
8798 static bool registered;
8804 unixctl_command_register(
8806 "bridge {priority tun_id in_port mark packet | odp_flow [-generate]}",
8807 2, 6, ofproto_unixctl_trace, NULL);
8808 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
8809 ofproto_unixctl_fdb_flush, NULL);
8810 unixctl_command_register("fdb/show", "bridge", 1, 1,
8811 ofproto_unixctl_fdb_show, NULL);
8812 unixctl_command_register("ofproto/clog", "", 0, 0,
8813 ofproto_dpif_clog, NULL);
8814 unixctl_command_register("ofproto/unclog", "", 0, 0,
8815 ofproto_dpif_unclog, NULL);
8816 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
8817 ofproto_dpif_self_check, NULL);
8818 unixctl_command_register("dpif/dump-dps", "", 0, 0,
8819 ofproto_unixctl_dpif_dump_dps, NULL);
8820 unixctl_command_register("dpif/show", "[bridge]", 0, INT_MAX,
8821 ofproto_unixctl_dpif_show, NULL);
8822 unixctl_command_register("dpif/dump-flows", "bridge", 1, 1,
8823 ofproto_unixctl_dpif_dump_flows, NULL);
8824 unixctl_command_register("dpif/del-flows", "bridge", 1, 1,
8825 ofproto_unixctl_dpif_del_flows, NULL);
8826 unixctl_command_register("dpif/dump-megaflows", "bridge", 1, 1,
8827 ofproto_unixctl_dpif_dump_megaflows, NULL);
8830 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
8832 * This is deprecated. It is only for compatibility with broken device drivers
8833 * in old versions of Linux that do not properly support VLANs when VLAN
8834 * devices are not used. When broken device drivers are no longer in
8835 * widespread use, we will delete these interfaces. */
8838 set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid)
8840 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
8841 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
8843 if (realdev_ofp_port == ofport->realdev_ofp_port
8844 && vid == ofport->vlandev_vid) {
8848 ofproto->backer->need_revalidate = REV_RECONFIGURE;
8850 if (ofport->realdev_ofp_port) {
8853 if (realdev_ofp_port && ofport->bundle) {
8854 /* vlandevs are enslaved to their realdevs, so they are not allowed to
8855 * themselves be part of a bundle. */
8856 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
8859 ofport->realdev_ofp_port = realdev_ofp_port;
8860 ofport->vlandev_vid = vid;
8862 if (realdev_ofp_port) {
8863 vsp_add(ofport, realdev_ofp_port, vid);
8870 hash_realdev_vid(uint16_t realdev_ofp_port, int vid)
8872 return hash_2words(realdev_ofp_port, vid);
8875 /* Returns the OFP port number of the Linux VLAN device that corresponds to
8876 * 'vlan_tci' on the network device with port number 'realdev_ofp_port' in
8877 * 'struct ofport_dpif'. For example, given 'realdev_ofp_port' of eth0 and
8878 * 'vlan_tci' 9, it would return the port number of eth0.9.
8880 * Unless VLAN splinters are enabled for port 'realdev_ofp_port', this
8881 * function just returns its 'realdev_ofp_port' argument. */
8883 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
8884 uint16_t realdev_ofp_port, ovs_be16 vlan_tci)
8886 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
8887 int vid = vlan_tci_to_vid(vlan_tci);
8888 const struct vlan_splinter *vsp;
8890 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
8891 hash_realdev_vid(realdev_ofp_port, vid),
8892 &ofproto->realdev_vid_map) {
8893 if (vsp->realdev_ofp_port == realdev_ofp_port
8894 && vsp->vid == vid) {
8895 return vsp->vlandev_ofp_port;
8899 return realdev_ofp_port;
8902 static struct vlan_splinter *
8903 vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port)
8905 struct vlan_splinter *vsp;
8907 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, hash_int(vlandev_ofp_port, 0),
8908 &ofproto->vlandev_map) {
8909 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
8917 /* Returns the OpenFlow port number of the "real" device underlying the Linux
8918 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
8919 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
8920 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
8921 * eth0 and store 9 in '*vid'.
8923 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
8924 * VLAN device. Unless VLAN splinters are enabled, this is what this function
8927 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
8928 uint16_t vlandev_ofp_port, int *vid)
8930 if (!hmap_is_empty(&ofproto->vlandev_map)) {
8931 const struct vlan_splinter *vsp;
8933 vsp = vlandev_find(ofproto, vlandev_ofp_port);
8938 return vsp->realdev_ofp_port;
8944 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
8945 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
8946 * 'flow->in_port' to the "real" device backing the VLAN device, sets
8947 * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is
8948 * always the case unless VLAN splinters are enabled), returns false without
8949 * making any changes. */
8951 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
8956 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port, &vid);
8961 /* Cause the flow to be processed as if it came in on the real device with
8962 * the VLAN device's VLAN ID. */
8963 flow->in_port = realdev;
8964 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
8969 vsp_remove(struct ofport_dpif *port)
8971 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
8972 struct vlan_splinter *vsp;
8974 vsp = vlandev_find(ofproto, port->up.ofp_port);
8976 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
8977 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
8980 port->realdev_ofp_port = 0;
8982 VLOG_ERR("missing vlan device record");
8987 vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid)
8989 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
8991 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
8992 && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid))
8993 == realdev_ofp_port)) {
8994 struct vlan_splinter *vsp;
8996 vsp = xmalloc(sizeof *vsp);
8997 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
8998 hash_int(port->up.ofp_port, 0));
8999 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
9000 hash_realdev_vid(realdev_ofp_port, vid));
9001 vsp->realdev_ofp_port = realdev_ofp_port;
9002 vsp->vlandev_ofp_port = port->up.ofp_port;
9005 port->realdev_ofp_port = realdev_ofp_port;
9007 VLOG_ERR("duplicate vlan device record");
9012 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
9014 const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
9015 return ofport ? ofport->odp_port : OVSP_NONE;
9018 static struct ofport_dpif *
9019 odp_port_to_ofport(const struct dpif_backer *backer, uint32_t odp_port)
9021 struct ofport_dpif *port;
9023 HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node,
9024 hash_int(odp_port, 0),
9025 &backer->odp_to_ofport_map) {
9026 if (port->odp_port == odp_port) {
9035 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
9037 struct ofport_dpif *port;
9039 port = odp_port_to_ofport(ofproto->backer, odp_port);
9040 if (port && &ofproto->up == port->up.ofproto) {
9041 return port->up.ofp_port;
9046 static unsigned long long int
9047 avg_subfacet_life_span(const struct ofproto_dpif *ofproto)
9049 unsigned long long int dc;
9050 unsigned long long int avg;
9052 dc = ofproto->total_subfacet_del_count + ofproto->subfacet_del_count;
9053 avg = dc ? ofproto->total_subfacet_life_span / dc : 0;
9059 avg_subfacet_count(const struct ofproto_dpif *ofproto)
9063 if (ofproto->n_update_stats) {
9064 avg_c = (double)ofproto->total_subfacet_count
9065 / ofproto->n_update_stats;
9072 show_dp_rates(struct ds *ds, const char *heading,
9073 const struct avg_subfacet_rates *rates)
9075 ds_put_format(ds, "%s add rate: %5.3f/min, del rate: %5.3f/min\n",
9076 heading, rates->add_rate, rates->del_rate);
9080 update_max_subfacet_count(struct ofproto_dpif *ofproto)
9082 ofproto->max_n_subfacet = MAX(ofproto->max_n_subfacet,
9083 hmap_count(&ofproto->subfacets));
9086 /* Compute exponentially weighted moving average, adding 'new' as the newest,
9087 * most heavily weighted element. 'base' designates the rate of decay: after
9088 * 'base' further updates, 'new''s weight in the EWMA decays to about 1/e
9091 exp_mavg(double *avg, int base, double new)
9093 *avg = (*avg * (base - 1) + new) / base;
9097 update_moving_averages(struct ofproto_dpif *ofproto)
9099 const int min_ms = 60 * 1000; /* milliseconds in one minute. */
9101 /* Update hourly averages on the minute boundaries. */
9102 if (time_msec() - ofproto->last_minute >= min_ms) {
9103 exp_mavg(&ofproto->hourly.add_rate, 60, ofproto->subfacet_add_count);
9104 exp_mavg(&ofproto->hourly.del_rate, 60, ofproto->subfacet_del_count);
9106 /* Update daily averages on the hour boundaries. */
9107 if ((ofproto->last_minute - ofproto->created) / min_ms % 60 == 59) {
9108 exp_mavg(&ofproto->daily.add_rate, 24, ofproto->hourly.add_rate);
9109 exp_mavg(&ofproto->daily.del_rate, 24, ofproto->hourly.del_rate);
9112 ofproto->total_subfacet_add_count += ofproto->subfacet_add_count;
9113 ofproto->total_subfacet_del_count += ofproto->subfacet_del_count;
9114 ofproto->subfacet_add_count = 0;
9115 ofproto->subfacet_del_count = 0;
9116 ofproto->last_minute += min_ms;
9121 dpif_stats_update_hit_count(struct ofproto_dpif *ofproto, uint64_t delta)
9123 ofproto->n_hit += delta;
9126 const struct ofproto_class ofproto_dpif_class = {
9161 port_is_lacp_current,
9162 NULL, /* rule_choose_table */
9169 rule_modify_actions,
9181 get_stp_port_status,
9188 is_mirror_output_bundle,
9189 forward_bpdu_changed,
9190 set_mac_table_config,