]> git.proxmox.com Git - mirror_ovs.git/blob - ofproto/ofproto-dpif.c
ofproto-dpif: Store relevant fields for wildcarding in facet.
[mirror_ovs.git] / ofproto / ofproto-dpif.c
1 /*
2 * Copyright (c) 2009, 2010, 2011, 2012, 2013 Nicira, Inc.
3 *
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:
7 *
8 * http://www.apache.org/licenses/LICENSE-2.0
9 *
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.
15 */
16
17 #include <config.h>
18
19 #include "ofproto/ofproto-provider.h"
20
21 #include <errno.h>
22
23 #include "bfd.h"
24 #include "bond.h"
25 #include "bundle.h"
26 #include "byte-order.h"
27 #include "connmgr.h"
28 #include "coverage.h"
29 #include "cfm.h"
30 #include "dpif.h"
31 #include "dynamic-string.h"
32 #include "fail-open.h"
33 #include "hmapx.h"
34 #include "lacp.h"
35 #include "learn.h"
36 #include "mac-learning.h"
37 #include "meta-flow.h"
38 #include "multipath.h"
39 #include "netdev-vport.h"
40 #include "netdev.h"
41 #include "netlink.h"
42 #include "nx-match.h"
43 #include "odp-util.h"
44 #include "odp-execute.h"
45 #include "ofp-util.h"
46 #include "ofpbuf.h"
47 #include "ofp-actions.h"
48 #include "ofp-parse.h"
49 #include "ofp-print.h"
50 #include "ofproto-dpif-governor.h"
51 #include "ofproto-dpif-ipfix.h"
52 #include "ofproto-dpif-sflow.h"
53 #include "poll-loop.h"
54 #include "simap.h"
55 #include "smap.h"
56 #include "timer.h"
57 #include "tunnel.h"
58 #include "unaligned.h"
59 #include "unixctl.h"
60 #include "vlan-bitmap.h"
61 #include "vlog.h"
62
63 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
64
65 COVERAGE_DEFINE(ofproto_dpif_expired);
66 COVERAGE_DEFINE(ofproto_dpif_xlate);
67 COVERAGE_DEFINE(facet_changed_rule);
68 COVERAGE_DEFINE(facet_revalidate);
69 COVERAGE_DEFINE(facet_unexpected);
70 COVERAGE_DEFINE(facet_suppress);
71
72 /* Maximum depth of flow table recursion (due to resubmit actions) in a
73 * flow translation. */
74 #define MAX_RESUBMIT_RECURSION 64
75
76 /* Number of implemented OpenFlow tables. */
77 enum { N_TABLES = 255 };
78 enum { TBL_INTERNAL = N_TABLES - 1 }; /* Used for internal hidden rules. */
79 BUILD_ASSERT_DECL(N_TABLES >= 2 && N_TABLES <= 255);
80
81 struct ofport_dpif;
82 struct ofproto_dpif;
83 struct flow_miss;
84 struct facet;
85
86 struct rule_dpif {
87 struct rule up;
88
89 /* These statistics:
90 *
91 * - Do include packets and bytes from facets that have been deleted or
92 * whose own statistics have been folded into the rule.
93 *
94 * - Do include packets and bytes sent "by hand" that were accounted to
95 * the rule without any facet being involved (this is a rare corner
96 * case in rule_execute()).
97 *
98 * - Do not include packet or bytes that can be obtained from any facet's
99 * packet_count or byte_count member or that can be obtained from the
100 * datapath by, e.g., dpif_flow_get() for any subfacet.
101 */
102 uint64_t packet_count; /* Number of packets received. */
103 uint64_t byte_count; /* Number of bytes received. */
104
105 tag_type tag; /* Caches rule_calculate_tag() result. */
106
107 struct list facets; /* List of "struct facet"s. */
108 };
109
110 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
111 {
112 return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
113 }
114
115 static struct rule_dpif *rule_dpif_lookup(struct ofproto_dpif *,
116 const struct flow *,
117 struct flow_wildcards *wc);
118 static struct rule_dpif *rule_dpif_lookup__(struct ofproto_dpif *,
119 const struct flow *,
120 struct flow_wildcards *wc,
121 uint8_t table);
122 static struct rule_dpif *rule_dpif_miss_rule(struct ofproto_dpif *ofproto,
123 const struct flow *flow);
124
125 static void rule_get_stats(struct rule *, uint64_t *packets, uint64_t *bytes);
126 static void rule_credit_stats(struct rule_dpif *,
127 const struct dpif_flow_stats *);
128 static tag_type rule_calculate_tag(const struct flow *,
129 const struct minimask *, uint32_t basis);
130 static void rule_invalidate(const struct rule_dpif *);
131
132 #define MAX_MIRRORS 32
133 typedef uint32_t mirror_mask_t;
134 #define MIRROR_MASK_C(X) UINT32_C(X)
135 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
136 struct ofmirror {
137 struct ofproto_dpif *ofproto; /* Owning ofproto. */
138 size_t idx; /* In ofproto's "mirrors" array. */
139 void *aux; /* Key supplied by ofproto's client. */
140 char *name; /* Identifier for log messages. */
141
142 /* Selection criteria. */
143 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
144 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
145 unsigned long *vlans; /* Bitmap of chosen VLANs, NULL selects all. */
146
147 /* Output (exactly one of out == NULL and out_vlan == -1 is true). */
148 struct ofbundle *out; /* Output port or NULL. */
149 int out_vlan; /* Output VLAN or -1. */
150 mirror_mask_t dup_mirrors; /* Bitmap of mirrors with the same output. */
151
152 /* Counters. */
153 int64_t packet_count; /* Number of packets sent. */
154 int64_t byte_count; /* Number of bytes sent. */
155 };
156
157 static void mirror_destroy(struct ofmirror *);
158 static void update_mirror_stats(struct ofproto_dpif *ofproto,
159 mirror_mask_t mirrors,
160 uint64_t packets, uint64_t bytes);
161
162 struct ofbundle {
163 struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
164 struct ofproto_dpif *ofproto; /* Owning ofproto. */
165 void *aux; /* Key supplied by ofproto's client. */
166 char *name; /* Identifier for log messages. */
167
168 /* Configuration. */
169 struct list ports; /* Contains "struct ofport"s. */
170 enum port_vlan_mode vlan_mode; /* VLAN mode */
171 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
172 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
173 * NULL if all VLANs are trunked. */
174 struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */
175 struct bond *bond; /* Nonnull iff more than one port. */
176 bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */
177
178 /* Status. */
179 bool floodable; /* True if no port has OFPUTIL_PC_NO_FLOOD set. */
180
181 /* Port mirroring info. */
182 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
183 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
184 mirror_mask_t mirror_out; /* Mirrors that output to this bundle. */
185 };
186
187 static void bundle_remove(struct ofport *);
188 static void bundle_update(struct ofbundle *);
189 static void bundle_destroy(struct ofbundle *);
190 static void bundle_del_port(struct ofport_dpif *);
191 static void bundle_run(struct ofbundle *);
192 static void bundle_wait(struct ofbundle *);
193 static struct ofbundle *lookup_input_bundle(const struct ofproto_dpif *,
194 uint16_t in_port, bool warn,
195 struct ofport_dpif **in_ofportp);
196
197 /* A controller may use OFPP_NONE as the ingress port to indicate that
198 * it did not arrive on a "real" port. 'ofpp_none_bundle' exists for
199 * when an input bundle is needed for validation (e.g., mirroring or
200 * OFPP_NORMAL processing). It is not connected to an 'ofproto' or have
201 * any 'port' structs, so care must be taken when dealing with it. */
202 static struct ofbundle ofpp_none_bundle = {
203 .name = "OFPP_NONE",
204 .vlan_mode = PORT_VLAN_TRUNK
205 };
206
207 static void stp_run(struct ofproto_dpif *ofproto);
208 static void stp_wait(struct ofproto_dpif *ofproto);
209 static int set_stp_port(struct ofport *,
210 const struct ofproto_port_stp_settings *);
211
212 static bool ofbundle_includes_vlan(const struct ofbundle *, uint16_t vlan);
213
214 struct xlate_ctx;
215
216 /* Initial values of fields of the packet that may be changed during
217 * flow processing and needed later. */
218 struct initial_vals {
219 /* This is the value of vlan_tci in the packet as actually received from
220 * dpif. This is the same as the facet's flow.vlan_tci unless the packet
221 * was received via a VLAN splinter. In that case, this value is 0
222 * (because the packet as actually received from the dpif had no 802.1Q
223 * tag) but the facet's flow.vlan_tci is set to the VLAN that the splinter
224 * represents.
225 *
226 * This member should be removed when the VLAN splinters feature is no
227 * longer needed. */
228 ovs_be16 vlan_tci;
229 };
230
231 struct xlate_out {
232 /* Wildcards relevant in translation. Any fields that were used to
233 * calculate the action must be set for caching and kernel
234 * wildcarding to work. For example, if the flow lookup involved
235 * performing the "normal" action on IPv4 and ARP packets, 'wc'
236 * would have the 'in_port' (always set), 'dl_type' (flow match),
237 * 'vlan_tci' (normal action), and 'dl_dst' (normal action) fields
238 * set. */
239 struct flow_wildcards wc;
240
241 tag_type tags; /* Tags associated with actions. */
242 enum slow_path_reason slow; /* 0 if fast path may be used. */
243 bool has_learn; /* Actions include NXAST_LEARN? */
244 bool has_normal; /* Actions output to OFPP_NORMAL? */
245 bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */
246 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
247 mirror_mask_t mirrors; /* Bitmap of associated mirrors. */
248
249 uint64_t odp_actions_stub[256 / 8];
250 struct ofpbuf odp_actions;
251 };
252
253 struct xlate_in {
254 struct ofproto_dpif *ofproto;
255
256 /* Flow to which the OpenFlow actions apply. xlate_actions() will modify
257 * this flow when actions change header fields. */
258 struct flow flow;
259
260 struct initial_vals initial_vals;
261
262 /* The packet corresponding to 'flow', or a null pointer if we are
263 * revalidating without a packet to refer to. */
264 const struct ofpbuf *packet;
265
266 /* Should OFPP_NORMAL update the MAC learning table? Should "learn"
267 * actions update the flow table?
268 *
269 * We want to update these tables if we are actually processing a packet,
270 * or if we are accounting for packets that the datapath has processed, but
271 * not if we are just revalidating. */
272 bool may_learn;
273
274 /* The rule initiating translation or NULL. */
275 struct rule_dpif *rule;
276
277 /* The actions to translate. If 'rule' is not NULL, these may be NULL. */
278 const struct ofpact *ofpacts;
279 size_t ofpacts_len;
280
281 /* Union of the set of TCP flags seen so far in this flow. (Used only by
282 * NXAST_FIN_TIMEOUT. Set to zero to avoid updating updating rules'
283 * timeouts.) */
284 uint8_t tcp_flags;
285
286 /* If nonnull, flow translation calls this function just before executing a
287 * resubmit or OFPP_TABLE action. In addition, disables logging of traces
288 * when the recursion depth is exceeded.
289 *
290 * 'rule' is the rule being submitted into. It will be null if the
291 * resubmit or OFPP_TABLE action didn't find a matching rule.
292 *
293 * This is normally null so the client has to set it manually after
294 * calling xlate_in_init(). */
295 void (*resubmit_hook)(struct xlate_ctx *, struct rule_dpif *rule);
296
297 /* If nonnull, flow translation calls this function to report some
298 * significant decision, e.g. to explain why OFPP_NORMAL translation
299 * dropped a packet. */
300 void (*report_hook)(struct xlate_ctx *, const char *s);
301
302 /* If nonnull, flow translation credits the specified statistics to each
303 * rule reached through a resubmit or OFPP_TABLE action.
304 *
305 * This is normally null so the client has to set it manually after
306 * calling xlate_in_init(). */
307 const struct dpif_flow_stats *resubmit_stats;
308 };
309
310 /* Context used by xlate_actions() and its callees. */
311 struct xlate_ctx {
312 struct xlate_in *xin;
313 struct xlate_out *xout;
314
315 struct ofproto_dpif *ofproto;
316
317 /* Flow at the last commit. */
318 struct flow base_flow;
319
320 /* Tunnel IP destination address as received. This is stored separately
321 * as the base_flow.tunnel is cleared on init to reflect the datapath
322 * behavior. Used to make sure not to send tunneled output to ourselves,
323 * which might lead to an infinite loop. This could happen easily
324 * if a tunnel is marked as 'ip_remote=flow', and the flow does not
325 * actually set the tun_dst field. */
326 ovs_be32 orig_tunnel_ip_dst;
327
328 /* Stack for the push and pop actions. Each stack element is of type
329 * "union mf_subvalue". */
330 union mf_subvalue init_stack[1024 / sizeof(union mf_subvalue)];
331 struct ofpbuf stack;
332
333 /* The rule that we are currently translating, or NULL. */
334 struct rule_dpif *rule;
335
336 int recurse; /* Recursion level, via xlate_table_action. */
337 bool max_resubmit_trigger; /* Recursed too deeply during translation. */
338 uint32_t orig_skb_priority; /* Priority when packet arrived. */
339 uint8_t table_id; /* OpenFlow table ID where flow was found. */
340 uint32_t sflow_n_outputs; /* Number of output ports. */
341 uint32_t sflow_odp_port; /* Output port for composing sFlow action. */
342 uint16_t user_cookie_offset;/* Used for user_action_cookie fixup. */
343 bool exit; /* No further actions should be processed. */
344 };
345
346 static void xlate_in_init(struct xlate_in *, struct ofproto_dpif *,
347 const struct flow *, const struct initial_vals *,
348 struct rule_dpif *, uint8_t tcp_flags,
349 const struct ofpbuf *);
350
351 static void xlate_out_uninit(struct xlate_out *);
352
353 static void xlate_actions(struct xlate_in *, struct xlate_out *);
354
355 static void xlate_actions_for_side_effects(struct xlate_in *);
356
357 static void xlate_table_action(struct xlate_ctx *, uint16_t in_port,
358 uint8_t table_id, bool may_packet_in);
359
360 static size_t put_userspace_action(const struct ofproto_dpif *,
361 struct ofpbuf *odp_actions,
362 const struct flow *,
363 const union user_action_cookie *,
364 const size_t);
365
366 static void compose_slow_path(const struct ofproto_dpif *, const struct flow *,
367 enum slow_path_reason,
368 uint64_t *stub, size_t stub_size,
369 const struct nlattr **actionsp,
370 size_t *actions_lenp);
371
372 static void xlate_report(struct xlate_ctx *ctx, const char *s);
373
374 static void xlate_out_copy(struct xlate_out *dst, const struct xlate_out *src);
375
376 /* A subfacet (see "struct subfacet" below) has three possible installation
377 * states:
378 *
379 * - SF_NOT_INSTALLED: Not installed in the datapath. This will only be the
380 * case just after the subfacet is created, just before the subfacet is
381 * destroyed, or if the datapath returns an error when we try to install a
382 * subfacet.
383 *
384 * - SF_FAST_PATH: The subfacet's actions are installed in the datapath.
385 *
386 * - SF_SLOW_PATH: An action that sends every packet for the subfacet through
387 * ofproto_dpif is installed in the datapath.
388 */
389 enum subfacet_path {
390 SF_NOT_INSTALLED, /* No datapath flow for this subfacet. */
391 SF_FAST_PATH, /* Full actions are installed. */
392 SF_SLOW_PATH, /* Send-to-userspace action is installed. */
393 };
394
395 /* A dpif flow and actions associated with a facet.
396 *
397 * See also the large comment on struct facet. */
398 struct subfacet {
399 /* Owners. */
400 struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */
401 struct list list_node; /* In struct facet's 'facets' list. */
402 struct facet *facet; /* Owning facet. */
403 struct dpif_backer *backer; /* Owning backer. */
404
405 enum odp_key_fitness key_fitness;
406 struct nlattr *key;
407 int key_len;
408
409 long long int used; /* Time last used; time created if not used. */
410 long long int created; /* Time created. */
411
412 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
413 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
414
415 enum subfacet_path path; /* Installed in datapath? */
416 };
417
418 #define SUBFACET_DESTROY_MAX_BATCH 50
419
420 static struct subfacet *subfacet_create(struct facet *, struct flow_miss *miss,
421 long long int now);
422 static struct subfacet *subfacet_find(struct dpif_backer *,
423 const struct nlattr *key, size_t key_len,
424 uint32_t key_hash);
425 static void subfacet_destroy(struct subfacet *);
426 static void subfacet_destroy__(struct subfacet *);
427 static void subfacet_destroy_batch(struct dpif_backer *,
428 struct subfacet **, int n);
429 static void subfacet_reset_dp_stats(struct subfacet *,
430 struct dpif_flow_stats *);
431 static void subfacet_update_stats(struct subfacet *,
432 const struct dpif_flow_stats *);
433 static int subfacet_install(struct subfacet *,
434 const struct ofpbuf *odp_actions,
435 struct dpif_flow_stats *);
436 static void subfacet_uninstall(struct subfacet *);
437
438 /* A unique, non-overlapping instantiation of an OpenFlow flow.
439 *
440 * A facet associates a "struct flow", which represents the Open vSwitch
441 * userspace idea of an exact-match flow, with one or more subfacets.
442 * While the facet is created based on an exact-match flow, it is stored
443 * within the ofproto based on the wildcards that could be expressed
444 * based on the flow table and other configuration. (See the 'wc'
445 * description in "struct xlate_out" for more details.)
446 *
447 * Each subfacet tracks the datapath's idea of the flow equivalent to
448 * the facet. When the kernel module (or other dpif implementation) and
449 * Open vSwitch userspace agree on the definition of a flow key, there
450 * is exactly one subfacet per facet. If the dpif implementation
451 * supports more-specific flow matching than userspace, however, a facet
452 * can have more than one subfacet. Examples include the dpif
453 * implementation not supporting the same wildcards as userspace or some
454 * distinction in flow that userspace simply doesn't understand.
455 *
456 * Flow expiration works in terms of subfacets, so a facet must have at
457 * least one subfacet or it will never expire, leaking memory. */
458 struct facet {
459 /* Owners. */
460 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
461 struct list list_node; /* In owning rule's 'facets' list. */
462 struct rule_dpif *rule; /* Owning rule. */
463
464 /* Owned data. */
465 struct list subfacets;
466 long long int used; /* Time last used; time created if not used. */
467
468 /* Key. */
469 struct flow flow; /* Flow of the creating subfacet. */
470 struct cls_rule cr; /* In 'ofproto_dpif's facets classifier. */
471
472 /* These statistics:
473 *
474 * - Do include packets and bytes sent "by hand", e.g. with
475 * dpif_execute().
476 *
477 * - Do include packets and bytes that were obtained from the datapath
478 * when a subfacet's statistics were reset (e.g. dpif_flow_put() with
479 * DPIF_FP_ZERO_STATS).
480 *
481 * - Do not include packets or bytes that can be obtained from the
482 * datapath for any existing subfacet.
483 */
484 uint64_t packet_count; /* Number of packets received. */
485 uint64_t byte_count; /* Number of bytes received. */
486
487 /* Resubmit statistics. */
488 uint64_t prev_packet_count; /* Number of packets from last stats push. */
489 uint64_t prev_byte_count; /* Number of bytes from last stats push. */
490 long long int prev_used; /* Used time from last stats push. */
491
492 /* Accounting. */
493 uint64_t accounted_bytes; /* Bytes processed by facet_account(). */
494 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
495 uint8_t tcp_flags; /* TCP flags seen for this 'rule'. */
496
497 struct xlate_out xout;
498
499 /* Initial values of the packet that may be needed later. */
500 struct initial_vals initial_vals;
501
502 /* Storage for a single subfacet, to reduce malloc() time and space
503 * overhead. (A facet always has at least one subfacet and in the common
504 * case has exactly one subfacet. However, 'one_subfacet' may not
505 * always be valid, since it could have been removed after newer
506 * subfacets were pushed onto the 'subfacets' list.) */
507 struct subfacet one_subfacet;
508
509 long long int learn_rl; /* Rate limiter for facet_learn(). */
510 };
511
512 static struct facet *facet_create(const struct flow_miss *, struct rule_dpif *,
513 struct xlate_out *,
514 struct dpif_flow_stats *);
515 static void facet_remove(struct facet *);
516 static void facet_free(struct facet *);
517
518 static struct facet *facet_find(struct ofproto_dpif *, const struct flow *);
519 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
520 const struct flow *);
521 static bool facet_revalidate(struct facet *);
522 static bool facet_check_consistency(struct facet *);
523
524 static void facet_flush_stats(struct facet *);
525
526 static void facet_reset_counters(struct facet *);
527 static void facet_push_stats(struct facet *, bool may_learn);
528 static void facet_learn(struct facet *);
529 static void facet_account(struct facet *);
530 static void push_all_stats(void);
531
532 static bool facet_is_controller_flow(struct facet *);
533
534 struct ofport_dpif {
535 struct hmap_node odp_port_node; /* In dpif_backer's "odp_to_ofport_map". */
536 struct ofport up;
537
538 uint32_t odp_port;
539 struct ofbundle *bundle; /* Bundle that contains this port, if any. */
540 struct list bundle_node; /* In struct ofbundle's "ports" list. */
541 struct cfm *cfm; /* Connectivity Fault Management, if any. */
542 struct bfd *bfd; /* BFD, if any. */
543 tag_type tag; /* Tag associated with this port. */
544 bool may_enable; /* May be enabled in bonds. */
545 long long int carrier_seq; /* Carrier status changes. */
546 struct tnl_port *tnl_port; /* Tunnel handle, or null. */
547
548 /* Spanning tree. */
549 struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */
550 enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */
551 long long int stp_state_entered;
552
553 struct hmap priorities; /* Map of attached 'priority_to_dscp's. */
554
555 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
556 *
557 * This is deprecated. It is only for compatibility with broken device
558 * drivers in old versions of Linux that do not properly support VLANs when
559 * VLAN devices are not used. When broken device drivers are no longer in
560 * widespread use, we will delete these interfaces. */
561 uint16_t realdev_ofp_port;
562 int vlandev_vid;
563 };
564
565 /* Node in 'ofport_dpif''s 'priorities' map. Used to maintain a map from
566 * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
567 * traffic egressing the 'ofport' with that priority should be marked with. */
568 struct priority_to_dscp {
569 struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'priorities' map. */
570 uint32_t priority; /* Priority of this queue (see struct flow). */
571
572 uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */
573 };
574
575 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
576 *
577 * This is deprecated. It is only for compatibility with broken device drivers
578 * in old versions of Linux that do not properly support VLANs when VLAN
579 * devices are not used. When broken device drivers are no longer in
580 * widespread use, we will delete these interfaces. */
581 struct vlan_splinter {
582 struct hmap_node realdev_vid_node;
583 struct hmap_node vlandev_node;
584 uint16_t realdev_ofp_port;
585 uint16_t vlandev_ofp_port;
586 int vid;
587 };
588
589 static uint16_t vsp_realdev_to_vlandev(const struct ofproto_dpif *,
590 uint16_t realdev_ofp_port,
591 ovs_be16 vlan_tci);
592 static bool vsp_adjust_flow(const struct ofproto_dpif *, struct flow *);
593 static void vsp_remove(struct ofport_dpif *);
594 static void vsp_add(struct ofport_dpif *, uint16_t realdev_ofp_port, int vid);
595
596 static uint32_t ofp_port_to_odp_port(const struct ofproto_dpif *,
597 uint16_t ofp_port);
598 static uint16_t odp_port_to_ofp_port(const struct ofproto_dpif *,
599 uint32_t odp_port);
600
601 static struct ofport_dpif *
602 ofport_dpif_cast(const struct ofport *ofport)
603 {
604 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
605 }
606
607 static void port_run(struct ofport_dpif *);
608 static void port_run_fast(struct ofport_dpif *);
609 static void port_wait(struct ofport_dpif *);
610 static int set_bfd(struct ofport *, const struct smap *);
611 static int set_cfm(struct ofport *, const struct cfm_settings *);
612 static void ofport_clear_priorities(struct ofport_dpif *);
613 static void run_fast_rl(void);
614
615 struct dpif_completion {
616 struct list list_node;
617 struct ofoperation *op;
618 };
619
620 /* Extra information about a classifier table.
621 * Currently used just for optimized flow revalidation. */
622 struct table_dpif {
623 /* If either of these is nonnull, then this table has a form that allows
624 * flows to be tagged to avoid revalidating most flows for the most common
625 * kinds of flow table changes. */
626 struct cls_table *catchall_table; /* Table that wildcards all fields. */
627 struct cls_table *other_table; /* Table with any other wildcard set. */
628 uint32_t basis; /* Keeps each table's tags separate. */
629 };
630
631 /* Reasons that we might need to revalidate every facet, and corresponding
632 * coverage counters.
633 *
634 * A value of 0 means that there is no need to revalidate.
635 *
636 * It would be nice to have some cleaner way to integrate with coverage
637 * counters, but with only a few reasons I guess this is good enough for
638 * now. */
639 enum revalidate_reason {
640 REV_RECONFIGURE = 1, /* Switch configuration changed. */
641 REV_STP, /* Spanning tree protocol port status change. */
642 REV_PORT_TOGGLED, /* Port enabled or disabled by CFM, LACP, ...*/
643 REV_FLOW_TABLE, /* Flow table changed. */
644 REV_INCONSISTENCY /* Facet self-check failed. */
645 };
646 COVERAGE_DEFINE(rev_reconfigure);
647 COVERAGE_DEFINE(rev_stp);
648 COVERAGE_DEFINE(rev_port_toggled);
649 COVERAGE_DEFINE(rev_flow_table);
650 COVERAGE_DEFINE(rev_inconsistency);
651
652 /* Drop keys are odp flow keys which have drop flows installed in the kernel.
653 * These are datapath flows which have no associated ofproto, if they did we
654 * would use facets. */
655 struct drop_key {
656 struct hmap_node hmap_node;
657 struct nlattr *key;
658 size_t key_len;
659 };
660
661 struct avg_subfacet_rates {
662 double add_rate; /* Moving average of new flows created per minute. */
663 double del_rate; /* Moving average of flows deleted per minute. */
664 };
665
666 /* All datapaths of a given type share a single dpif backer instance. */
667 struct dpif_backer {
668 char *type;
669 int refcount;
670 struct dpif *dpif;
671 struct timer next_expiration;
672 struct hmap odp_to_ofport_map; /* ODP port to ofport mapping. */
673
674 struct simap tnl_backers; /* Set of dpif ports backing tunnels. */
675
676 /* Facet revalidation flags applying to facets which use this backer. */
677 enum revalidate_reason need_revalidate; /* Revalidate every facet. */
678 struct tag_set revalidate_set; /* Revalidate only matching facets. */
679
680 struct hmap drop_keys; /* Set of dropped odp keys. */
681 bool recv_set_enable; /* Enables or disables receiving packets. */
682
683 struct hmap subfacets;
684 struct governor *governor;
685
686 /* Subfacet statistics.
687 *
688 * These keep track of the total number of subfacets added and deleted and
689 * flow life span. They are useful for computing the flow rates stats
690 * exposed via "ovs-appctl dpif/show". The goal is to learn about
691 * traffic patterns in ways that we can use later to improve Open vSwitch
692 * performance in new situations. */
693 long long int created; /* Time when it is created. */
694 unsigned max_n_subfacet; /* Maximum number of flows */
695 unsigned avg_n_subfacet; /* Average number of flows. */
696 long long int avg_subfacet_life; /* Average life span of subfacets. */
697
698 /* The average number of subfacets... */
699 struct avg_subfacet_rates hourly; /* ...over the last hour. */
700 struct avg_subfacet_rates daily; /* ...over the last day. */
701 struct avg_subfacet_rates lifetime; /* ...over the switch lifetime. */
702 long long int last_minute; /* Last time 'hourly' was updated. */
703
704 /* Number of subfacets added or deleted since 'last_minute'. */
705 unsigned subfacet_add_count;
706 unsigned subfacet_del_count;
707
708 /* Number of subfacets added or deleted from 'created' to 'last_minute.' */
709 unsigned long long int total_subfacet_add_count;
710 unsigned long long int total_subfacet_del_count;
711 };
712
713 /* All existing ofproto_backer instances, indexed by ofproto->up.type. */
714 static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers);
715
716 static void drop_key_clear(struct dpif_backer *);
717 static struct ofport_dpif *
718 odp_port_to_ofport(const struct dpif_backer *, uint32_t odp_port);
719 static void update_moving_averages(struct dpif_backer *backer);
720
721 struct ofproto_dpif {
722 struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
723 struct ofproto up;
724 struct dpif_backer *backer;
725
726 /* Special OpenFlow rules. */
727 struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */
728 struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */
729 struct rule_dpif *drop_frags_rule; /* Used in OFPC_FRAG_DROP mode. */
730
731 /* Bridging. */
732 struct netflow *netflow;
733 struct dpif_sflow *sflow;
734 struct dpif_ipfix *ipfix;
735 struct hmap bundles; /* Contains "struct ofbundle"s. */
736 struct mac_learning *ml;
737 struct ofmirror *mirrors[MAX_MIRRORS];
738 bool has_mirrors;
739 bool has_bonded_bundles;
740
741 /* Facets. */
742 struct classifier facets; /* Contains 'struct facet's. */
743 long long int consistency_rl;
744
745 /* Revalidation. */
746 struct table_dpif tables[N_TABLES];
747
748 /* Support for debugging async flow mods. */
749 struct list completions;
750
751 bool has_bundle_action; /* True when the first bundle action appears. */
752 struct netdev_stats stats; /* To account packets generated and consumed in
753 * userspace. */
754
755 /* Spanning tree. */
756 struct stp *stp;
757 long long int stp_last_tick;
758
759 /* VLAN splinters. */
760 struct hmap realdev_vid_map; /* (realdev,vid) -> vlandev. */
761 struct hmap vlandev_map; /* vlandev -> (realdev,vid). */
762
763 /* Ports. */
764 struct sset ports; /* Set of standard port names. */
765 struct sset ghost_ports; /* Ports with no datapath port. */
766 struct sset port_poll_set; /* Queued names for port_poll() reply. */
767 int port_poll_errno; /* Last errno for port_poll() reply. */
768
769 /* Per ofproto's dpif stats. */
770 uint64_t n_hit;
771 uint64_t n_missed;
772 };
773
774 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
775 * for debugging the asynchronous flow_mod implementation.) */
776 static bool clogged;
777
778 /* All existing ofproto_dpif instances, indexed by ->up.name. */
779 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
780
781 static void ofproto_dpif_unixctl_init(void);
782
783 static struct ofproto_dpif *
784 ofproto_dpif_cast(const struct ofproto *ofproto)
785 {
786 ovs_assert(ofproto->ofproto_class == &ofproto_dpif_class);
787 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
788 }
789
790 static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *,
791 uint16_t ofp_port);
792 static struct ofport_dpif *get_odp_port(const struct ofproto_dpif *,
793 uint32_t odp_port);
794 static void ofproto_trace(struct ofproto_dpif *, const struct flow *,
795 const struct ofpbuf *,
796 const struct initial_vals *, struct ds *);
797
798 /* Packet processing. */
799 static void update_learning_table(struct ofproto_dpif *, const struct flow *,
800 struct flow_wildcards *, int vlan,
801 struct ofbundle *);
802 /* Upcalls. */
803 #define FLOW_MISS_MAX_BATCH 50
804 static int handle_upcalls(struct dpif_backer *, unsigned int max_batch);
805
806 /* Flow expiration. */
807 static int expire(struct dpif_backer *);
808
809 /* NetFlow. */
810 static void send_netflow_active_timeouts(struct ofproto_dpif *);
811
812 /* Utilities. */
813 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
814 static size_t compose_sflow_action(const struct ofproto_dpif *,
815 struct ofpbuf *odp_actions,
816 const struct flow *, uint32_t odp_port);
817 static void compose_ipfix_action(const struct ofproto_dpif *,
818 struct ofpbuf *odp_actions,
819 const struct flow *);
820 static void add_mirror_actions(struct xlate_ctx *ctx,
821 const struct flow *flow);
822 /* Global variables. */
823 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
824
825 /* Initial mappings of port to bridge mappings. */
826 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
827 \f
828 /* Factory functions. */
829
830 static void
831 init(const struct shash *iface_hints)
832 {
833 struct shash_node *node;
834
835 /* Make a local copy, since we don't own 'iface_hints' elements. */
836 SHASH_FOR_EACH(node, iface_hints) {
837 const struct iface_hint *orig_hint = node->data;
838 struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
839
840 new_hint->br_name = xstrdup(orig_hint->br_name);
841 new_hint->br_type = xstrdup(orig_hint->br_type);
842 new_hint->ofp_port = orig_hint->ofp_port;
843
844 shash_add(&init_ofp_ports, node->name, new_hint);
845 }
846 }
847
848 static void
849 enumerate_types(struct sset *types)
850 {
851 dp_enumerate_types(types);
852 }
853
854 static int
855 enumerate_names(const char *type, struct sset *names)
856 {
857 struct ofproto_dpif *ofproto;
858
859 sset_clear(names);
860 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
861 if (strcmp(type, ofproto->up.type)) {
862 continue;
863 }
864 sset_add(names, ofproto->up.name);
865 }
866
867 return 0;
868 }
869
870 static int
871 del(const char *type, const char *name)
872 {
873 struct dpif *dpif;
874 int error;
875
876 error = dpif_open(name, type, &dpif);
877 if (!error) {
878 error = dpif_delete(dpif);
879 dpif_close(dpif);
880 }
881 return error;
882 }
883 \f
884 static const char *
885 port_open_type(const char *datapath_type, const char *port_type)
886 {
887 return dpif_port_open_type(datapath_type, port_type);
888 }
889
890 /* Type functions. */
891
892 static struct ofproto_dpif *
893 lookup_ofproto_dpif_by_port_name(const char *name)
894 {
895 struct ofproto_dpif *ofproto;
896
897 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
898 if (sset_contains(&ofproto->ports, name)) {
899 return ofproto;
900 }
901 }
902
903 return NULL;
904 }
905
906 static int
907 type_run(const char *type)
908 {
909 static long long int push_timer = LLONG_MIN;
910 struct dpif_backer *backer;
911 char *devname;
912 int error;
913
914 backer = shash_find_data(&all_dpif_backers, type);
915 if (!backer) {
916 /* This is not necessarily a problem, since backers are only
917 * created on demand. */
918 return 0;
919 }
920
921 dpif_run(backer->dpif);
922
923 /* The most natural place to push facet statistics is when they're pulled
924 * from the datapath. However, when there are many flows in the datapath,
925 * this expensive operation can occur so frequently, that it reduces our
926 * ability to quickly set up flows. To reduce the cost, we push statistics
927 * here instead. */
928 if (time_msec() > push_timer) {
929 push_timer = time_msec() + 2000;
930 push_all_stats();
931 }
932
933 /* If vswitchd started with other_config:flow_restore_wait set as "true",
934 * and the configuration has now changed to "false", enable receiving
935 * packets from the datapath. */
936 if (!backer->recv_set_enable && !ofproto_get_flow_restore_wait()) {
937 backer->recv_set_enable = true;
938
939 error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
940 if (error) {
941 VLOG_ERR("Failed to enable receiving packets in dpif.");
942 return error;
943 }
944 dpif_flow_flush(backer->dpif);
945 backer->need_revalidate = REV_RECONFIGURE;
946 }
947
948 if (backer->need_revalidate
949 || !tag_set_is_empty(&backer->revalidate_set)) {
950 struct tag_set revalidate_set = backer->revalidate_set;
951 bool need_revalidate = backer->need_revalidate;
952 struct ofproto_dpif *ofproto;
953 struct simap_node *node;
954 struct simap tmp_backers;
955
956 /* Handle tunnel garbage collection. */
957 simap_init(&tmp_backers);
958 simap_swap(&backer->tnl_backers, &tmp_backers);
959
960 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
961 struct ofport_dpif *iter;
962
963 if (backer != ofproto->backer) {
964 continue;
965 }
966
967 HMAP_FOR_EACH (iter, up.hmap_node, &ofproto->up.ports) {
968 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
969 const char *dp_port;
970
971 if (!iter->tnl_port) {
972 continue;
973 }
974
975 dp_port = netdev_vport_get_dpif_port(iter->up.netdev,
976 namebuf, sizeof namebuf);
977 node = simap_find(&tmp_backers, dp_port);
978 if (node) {
979 simap_put(&backer->tnl_backers, dp_port, node->data);
980 simap_delete(&tmp_backers, node);
981 node = simap_find(&backer->tnl_backers, dp_port);
982 } else {
983 node = simap_find(&backer->tnl_backers, dp_port);
984 if (!node) {
985 uint32_t odp_port = UINT32_MAX;
986
987 if (!dpif_port_add(backer->dpif, iter->up.netdev,
988 &odp_port)) {
989 simap_put(&backer->tnl_backers, dp_port, odp_port);
990 node = simap_find(&backer->tnl_backers, dp_port);
991 }
992 }
993 }
994
995 iter->odp_port = node ? node->data : OVSP_NONE;
996 if (tnl_port_reconfigure(&iter->up, iter->odp_port,
997 &iter->tnl_port)) {
998 backer->need_revalidate = REV_RECONFIGURE;
999 }
1000 }
1001 }
1002
1003 SIMAP_FOR_EACH (node, &tmp_backers) {
1004 dpif_port_del(backer->dpif, node->data);
1005 }
1006 simap_destroy(&tmp_backers);
1007
1008 switch (backer->need_revalidate) {
1009 case REV_RECONFIGURE: COVERAGE_INC(rev_reconfigure); break;
1010 case REV_STP: COVERAGE_INC(rev_stp); break;
1011 case REV_PORT_TOGGLED: COVERAGE_INC(rev_port_toggled); break;
1012 case REV_FLOW_TABLE: COVERAGE_INC(rev_flow_table); break;
1013 case REV_INCONSISTENCY: COVERAGE_INC(rev_inconsistency); break;
1014 }
1015
1016 if (backer->need_revalidate) {
1017 /* Clear the drop_keys in case we should now be accepting some
1018 * formerly dropped flows. */
1019 drop_key_clear(backer);
1020 }
1021
1022 /* Clear the revalidation flags. */
1023 tag_set_init(&backer->revalidate_set);
1024 backer->need_revalidate = 0;
1025
1026 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1027 struct facet *facet, *next;
1028 struct cls_cursor cursor;
1029
1030 if (ofproto->backer != backer) {
1031 continue;
1032 }
1033
1034 cls_cursor_init(&cursor, &ofproto->facets, NULL);
1035 CLS_CURSOR_FOR_EACH_SAFE (facet, next, cr, &cursor) {
1036 if (need_revalidate
1037 || tag_set_intersects(&revalidate_set, facet->xout.tags)) {
1038 facet_revalidate(facet);
1039 run_fast_rl();
1040 }
1041 }
1042 }
1043 }
1044
1045 if (!backer->recv_set_enable) {
1046 /* Wake up before a max of 1000ms. */
1047 timer_set_duration(&backer->next_expiration, 1000);
1048 } else if (timer_expired(&backer->next_expiration)) {
1049 int delay = expire(backer);
1050 timer_set_duration(&backer->next_expiration, delay);
1051 }
1052
1053 /* Check for port changes in the dpif. */
1054 while ((error = dpif_port_poll(backer->dpif, &devname)) == 0) {
1055 struct ofproto_dpif *ofproto;
1056 struct dpif_port port;
1057
1058 /* Don't report on the datapath's device. */
1059 if (!strcmp(devname, dpif_base_name(backer->dpif))) {
1060 goto next;
1061 }
1062
1063 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
1064 &all_ofproto_dpifs) {
1065 if (simap_contains(&ofproto->backer->tnl_backers, devname)) {
1066 goto next;
1067 }
1068 }
1069
1070 ofproto = lookup_ofproto_dpif_by_port_name(devname);
1071 if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
1072 /* The port was removed. If we know the datapath,
1073 * report it through poll_set(). If we don't, it may be
1074 * notifying us of a removal we initiated, so ignore it.
1075 * If there's a pending ENOBUFS, let it stand, since
1076 * everything will be reevaluated. */
1077 if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
1078 sset_add(&ofproto->port_poll_set, devname);
1079 ofproto->port_poll_errno = 0;
1080 }
1081 } else if (!ofproto) {
1082 /* The port was added, but we don't know with which
1083 * ofproto we should associate it. Delete it. */
1084 dpif_port_del(backer->dpif, port.port_no);
1085 }
1086 dpif_port_destroy(&port);
1087
1088 next:
1089 free(devname);
1090 }
1091
1092 if (error != EAGAIN) {
1093 struct ofproto_dpif *ofproto;
1094
1095 /* There was some sort of error, so propagate it to all
1096 * ofprotos that use this backer. */
1097 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
1098 &all_ofproto_dpifs) {
1099 if (ofproto->backer == backer) {
1100 sset_clear(&ofproto->port_poll_set);
1101 ofproto->port_poll_errno = error;
1102 }
1103 }
1104 }
1105
1106 if (backer->governor) {
1107 size_t n_subfacets;
1108
1109 governor_run(backer->governor);
1110
1111 /* If the governor has shrunk to its minimum size and the number of
1112 * subfacets has dwindled, then drop the governor entirely.
1113 *
1114 * For hysteresis, the number of subfacets to drop the governor is
1115 * smaller than the number needed to trigger its creation. */
1116 n_subfacets = hmap_count(&backer->subfacets);
1117 if (n_subfacets * 4 < flow_eviction_threshold
1118 && governor_is_idle(backer->governor)) {
1119 governor_destroy(backer->governor);
1120 backer->governor = NULL;
1121 }
1122 }
1123
1124 return 0;
1125 }
1126
1127 static int
1128 dpif_backer_run_fast(struct dpif_backer *backer, int max_batch)
1129 {
1130 unsigned int work;
1131
1132 /* If recv_set_enable is false, we should not handle upcalls. */
1133 if (!backer->recv_set_enable) {
1134 return 0;
1135 }
1136
1137 /* Handle one or more batches of upcalls, until there's nothing left to do
1138 * or until we do a fixed total amount of work.
1139 *
1140 * We do work in batches because it can be much cheaper to set up a number
1141 * of flows and fire off their patches all at once. We do multiple batches
1142 * because in some cases handling a packet can cause another packet to be
1143 * queued almost immediately as part of the return flow. Both
1144 * optimizations can make major improvements on some benchmarks and
1145 * presumably for real traffic as well. */
1146 work = 0;
1147 while (work < max_batch) {
1148 int retval = handle_upcalls(backer, max_batch - work);
1149 if (retval <= 0) {
1150 return -retval;
1151 }
1152 work += retval;
1153 }
1154
1155 return 0;
1156 }
1157
1158 static int
1159 type_run_fast(const char *type)
1160 {
1161 struct dpif_backer *backer;
1162
1163 backer = shash_find_data(&all_dpif_backers, type);
1164 if (!backer) {
1165 /* This is not necessarily a problem, since backers are only
1166 * created on demand. */
1167 return 0;
1168 }
1169
1170 return dpif_backer_run_fast(backer, FLOW_MISS_MAX_BATCH);
1171 }
1172
1173 static void
1174 run_fast_rl(void)
1175 {
1176 static long long int port_rl = LLONG_MIN;
1177 static unsigned int backer_rl = 0;
1178
1179 if (time_msec() >= port_rl) {
1180 struct ofproto_dpif *ofproto;
1181 struct ofport_dpif *ofport;
1182
1183 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1184
1185 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1186 port_run_fast(ofport);
1187 }
1188 }
1189 port_rl = time_msec() + 200;
1190 }
1191
1192 /* XXX: We have to be careful not to do too much work in this function. If
1193 * we call dpif_backer_run_fast() too often, or with too large a batch,
1194 * performance improves signifcantly, but at a cost. It's possible for the
1195 * number of flows in the datapath to increase without bound, and for poll
1196 * loops to take 10s of seconds. The correct solution to this problem,
1197 * long term, is to separate flow miss handling into it's own thread so it
1198 * isn't affected by revalidations, and expirations. Until then, this is
1199 * the best we can do. */
1200 if (++backer_rl >= 10) {
1201 struct shash_node *node;
1202
1203 backer_rl = 0;
1204 SHASH_FOR_EACH (node, &all_dpif_backers) {
1205 dpif_backer_run_fast(node->data, 1);
1206 }
1207 }
1208 }
1209
1210 static void
1211 type_wait(const char *type)
1212 {
1213 struct dpif_backer *backer;
1214
1215 backer = shash_find_data(&all_dpif_backers, type);
1216 if (!backer) {
1217 /* This is not necessarily a problem, since backers are only
1218 * created on demand. */
1219 return;
1220 }
1221
1222 if (backer->governor) {
1223 governor_wait(backer->governor);
1224 }
1225
1226 timer_wait(&backer->next_expiration);
1227 }
1228 \f
1229 /* Basic life-cycle. */
1230
1231 static int add_internal_flows(struct ofproto_dpif *);
1232
1233 static struct ofproto *
1234 alloc(void)
1235 {
1236 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
1237 return &ofproto->up;
1238 }
1239
1240 static void
1241 dealloc(struct ofproto *ofproto_)
1242 {
1243 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1244 free(ofproto);
1245 }
1246
1247 static void
1248 close_dpif_backer(struct dpif_backer *backer)
1249 {
1250 struct shash_node *node;
1251
1252 ovs_assert(backer->refcount > 0);
1253
1254 if (--backer->refcount) {
1255 return;
1256 }
1257
1258 drop_key_clear(backer);
1259 hmap_destroy(&backer->drop_keys);
1260
1261 simap_destroy(&backer->tnl_backers);
1262 hmap_destroy(&backer->odp_to_ofport_map);
1263 node = shash_find(&all_dpif_backers, backer->type);
1264 free(backer->type);
1265 shash_delete(&all_dpif_backers, node);
1266 dpif_close(backer->dpif);
1267
1268 ovs_assert(hmap_is_empty(&backer->subfacets));
1269 hmap_destroy(&backer->subfacets);
1270 governor_destroy(backer->governor);
1271
1272 free(backer);
1273 }
1274
1275 /* Datapath port slated for removal from datapath. */
1276 struct odp_garbage {
1277 struct list list_node;
1278 uint32_t odp_port;
1279 };
1280
1281 static int
1282 open_dpif_backer(const char *type, struct dpif_backer **backerp)
1283 {
1284 struct dpif_backer *backer;
1285 struct dpif_port_dump port_dump;
1286 struct dpif_port port;
1287 struct shash_node *node;
1288 struct list garbage_list;
1289 struct odp_garbage *garbage, *next;
1290 struct sset names;
1291 char *backer_name;
1292 const char *name;
1293 int error;
1294
1295 backer = shash_find_data(&all_dpif_backers, type);
1296 if (backer) {
1297 backer->refcount++;
1298 *backerp = backer;
1299 return 0;
1300 }
1301
1302 backer_name = xasprintf("ovs-%s", type);
1303
1304 /* Remove any existing datapaths, since we assume we're the only
1305 * userspace controlling the datapath. */
1306 sset_init(&names);
1307 dp_enumerate_names(type, &names);
1308 SSET_FOR_EACH(name, &names) {
1309 struct dpif *old_dpif;
1310
1311 /* Don't remove our backer if it exists. */
1312 if (!strcmp(name, backer_name)) {
1313 continue;
1314 }
1315
1316 if (dpif_open(name, type, &old_dpif)) {
1317 VLOG_WARN("couldn't open old datapath %s to remove it", name);
1318 } else {
1319 dpif_delete(old_dpif);
1320 dpif_close(old_dpif);
1321 }
1322 }
1323 sset_destroy(&names);
1324
1325 backer = xmalloc(sizeof *backer);
1326
1327 error = dpif_create_and_open(backer_name, type, &backer->dpif);
1328 free(backer_name);
1329 if (error) {
1330 VLOG_ERR("failed to open datapath of type %s: %s", type,
1331 strerror(error));
1332 free(backer);
1333 return error;
1334 }
1335
1336 backer->type = xstrdup(type);
1337 backer->governor = NULL;
1338 backer->refcount = 1;
1339 hmap_init(&backer->odp_to_ofport_map);
1340 hmap_init(&backer->drop_keys);
1341 hmap_init(&backer->subfacets);
1342 timer_set_duration(&backer->next_expiration, 1000);
1343 backer->need_revalidate = 0;
1344 simap_init(&backer->tnl_backers);
1345 tag_set_init(&backer->revalidate_set);
1346 backer->recv_set_enable = !ofproto_get_flow_restore_wait();
1347 *backerp = backer;
1348
1349 if (backer->recv_set_enable) {
1350 dpif_flow_flush(backer->dpif);
1351 }
1352
1353 /* Loop through the ports already on the datapath and remove any
1354 * that we don't need anymore. */
1355 list_init(&garbage_list);
1356 dpif_port_dump_start(&port_dump, backer->dpif);
1357 while (dpif_port_dump_next(&port_dump, &port)) {
1358 node = shash_find(&init_ofp_ports, port.name);
1359 if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
1360 garbage = xmalloc(sizeof *garbage);
1361 garbage->odp_port = port.port_no;
1362 list_push_front(&garbage_list, &garbage->list_node);
1363 }
1364 }
1365 dpif_port_dump_done(&port_dump);
1366
1367 LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) {
1368 dpif_port_del(backer->dpif, garbage->odp_port);
1369 list_remove(&garbage->list_node);
1370 free(garbage);
1371 }
1372
1373 shash_add(&all_dpif_backers, type, backer);
1374
1375 error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
1376 if (error) {
1377 VLOG_ERR("failed to listen on datapath of type %s: %s",
1378 type, strerror(error));
1379 close_dpif_backer(backer);
1380 return error;
1381 }
1382
1383 backer->max_n_subfacet = 0;
1384 backer->created = time_msec();
1385 backer->last_minute = backer->created;
1386 memset(&backer->hourly, 0, sizeof backer->hourly);
1387 memset(&backer->daily, 0, sizeof backer->daily);
1388 memset(&backer->lifetime, 0, sizeof backer->lifetime);
1389 backer->subfacet_add_count = 0;
1390 backer->subfacet_del_count = 0;
1391 backer->total_subfacet_add_count = 0;
1392 backer->total_subfacet_del_count = 0;
1393 backer->avg_n_subfacet = 0;
1394 backer->avg_subfacet_life = 0;
1395
1396 return error;
1397 }
1398
1399 static int
1400 construct(struct ofproto *ofproto_)
1401 {
1402 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1403 struct shash_node *node, *next;
1404 int max_ports;
1405 int error;
1406 int i;
1407
1408 error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1409 if (error) {
1410 return error;
1411 }
1412
1413 max_ports = dpif_get_max_ports(ofproto->backer->dpif);
1414 ofproto_init_max_ports(ofproto_, MIN(max_ports, OFPP_MAX));
1415
1416 ofproto->netflow = NULL;
1417 ofproto->sflow = NULL;
1418 ofproto->ipfix = NULL;
1419 ofproto->stp = NULL;
1420 hmap_init(&ofproto->bundles);
1421 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1422 for (i = 0; i < MAX_MIRRORS; i++) {
1423 ofproto->mirrors[i] = NULL;
1424 }
1425 ofproto->has_bonded_bundles = false;
1426
1427 classifier_init(&ofproto->facets);
1428 ofproto->consistency_rl = LLONG_MIN;
1429
1430 for (i = 0; i < N_TABLES; i++) {
1431 struct table_dpif *table = &ofproto->tables[i];
1432
1433 table->catchall_table = NULL;
1434 table->other_table = NULL;
1435 table->basis = random_uint32();
1436 }
1437
1438 list_init(&ofproto->completions);
1439
1440 ofproto_dpif_unixctl_init();
1441
1442 ofproto->has_mirrors = false;
1443 ofproto->has_bundle_action = false;
1444
1445 hmap_init(&ofproto->vlandev_map);
1446 hmap_init(&ofproto->realdev_vid_map);
1447
1448 sset_init(&ofproto->ports);
1449 sset_init(&ofproto->ghost_ports);
1450 sset_init(&ofproto->port_poll_set);
1451 ofproto->port_poll_errno = 0;
1452
1453 SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1454 struct iface_hint *iface_hint = node->data;
1455
1456 if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1457 /* Check if the datapath already has this port. */
1458 if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1459 sset_add(&ofproto->ports, node->name);
1460 }
1461
1462 free(iface_hint->br_name);
1463 free(iface_hint->br_type);
1464 free(iface_hint);
1465 shash_delete(&init_ofp_ports, node);
1466 }
1467 }
1468
1469 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1470 hash_string(ofproto->up.name, 0));
1471 memset(&ofproto->stats, 0, sizeof ofproto->stats);
1472
1473 ofproto_init_tables(ofproto_, N_TABLES);
1474 error = add_internal_flows(ofproto);
1475 ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1476
1477 ofproto->n_hit = 0;
1478 ofproto->n_missed = 0;
1479
1480 return error;
1481 }
1482
1483 static int
1484 add_internal_flow(struct ofproto_dpif *ofproto, int id,
1485 const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1486 {
1487 struct ofputil_flow_mod fm;
1488 int error;
1489
1490 match_init_catchall(&fm.match);
1491 fm.priority = 0;
1492 match_set_reg(&fm.match, 0, id);
1493 fm.new_cookie = htonll(0);
1494 fm.cookie = htonll(0);
1495 fm.cookie_mask = htonll(0);
1496 fm.table_id = TBL_INTERNAL;
1497 fm.command = OFPFC_ADD;
1498 fm.idle_timeout = 0;
1499 fm.hard_timeout = 0;
1500 fm.buffer_id = 0;
1501 fm.out_port = 0;
1502 fm.flags = 0;
1503 fm.ofpacts = ofpacts->data;
1504 fm.ofpacts_len = ofpacts->size;
1505
1506 error = ofproto_flow_mod(&ofproto->up, &fm);
1507 if (error) {
1508 VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
1509 id, ofperr_to_string(error));
1510 return error;
1511 }
1512
1513 *rulep = rule_dpif_lookup__(ofproto, &fm.match.flow, NULL, TBL_INTERNAL);
1514 ovs_assert(*rulep != NULL);
1515
1516 return 0;
1517 }
1518
1519 static int
1520 add_internal_flows(struct ofproto_dpif *ofproto)
1521 {
1522 struct ofpact_controller *controller;
1523 uint64_t ofpacts_stub[128 / 8];
1524 struct ofpbuf ofpacts;
1525 int error;
1526 int id;
1527
1528 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1529 id = 1;
1530
1531 controller = ofpact_put_CONTROLLER(&ofpacts);
1532 controller->max_len = UINT16_MAX;
1533 controller->controller_id = 0;
1534 controller->reason = OFPR_NO_MATCH;
1535 ofpact_pad(&ofpacts);
1536
1537 error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule);
1538 if (error) {
1539 return error;
1540 }
1541
1542 ofpbuf_clear(&ofpacts);
1543 error = add_internal_flow(ofproto, id++, &ofpacts,
1544 &ofproto->no_packet_in_rule);
1545 if (error) {
1546 return error;
1547 }
1548
1549 error = add_internal_flow(ofproto, id++, &ofpacts,
1550 &ofproto->drop_frags_rule);
1551 return error;
1552 }
1553
1554 static void
1555 complete_operations(struct ofproto_dpif *ofproto)
1556 {
1557 struct dpif_completion *c, *next;
1558
1559 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
1560 ofoperation_complete(c->op, 0);
1561 list_remove(&c->list_node);
1562 free(c);
1563 }
1564 }
1565
1566 static void
1567 destruct(struct ofproto *ofproto_)
1568 {
1569 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1570 struct rule_dpif *rule, *next_rule;
1571 struct oftable *table;
1572 int i;
1573
1574 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1575 complete_operations(ofproto);
1576
1577 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1578 struct cls_cursor cursor;
1579
1580 cls_cursor_init(&cursor, &table->cls, NULL);
1581 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1582 ofproto_rule_destroy(&rule->up);
1583 }
1584 }
1585
1586 for (i = 0; i < MAX_MIRRORS; i++) {
1587 mirror_destroy(ofproto->mirrors[i]);
1588 }
1589
1590 netflow_destroy(ofproto->netflow);
1591 dpif_sflow_destroy(ofproto->sflow);
1592 hmap_destroy(&ofproto->bundles);
1593 mac_learning_destroy(ofproto->ml);
1594
1595 classifier_destroy(&ofproto->facets);
1596
1597 hmap_destroy(&ofproto->vlandev_map);
1598 hmap_destroy(&ofproto->realdev_vid_map);
1599
1600 sset_destroy(&ofproto->ports);
1601 sset_destroy(&ofproto->ghost_ports);
1602 sset_destroy(&ofproto->port_poll_set);
1603
1604 close_dpif_backer(ofproto->backer);
1605 }
1606
1607 static int
1608 run_fast(struct ofproto *ofproto_)
1609 {
1610 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1611 struct ofport_dpif *ofport;
1612
1613 /* Do not perform any periodic activity required by 'ofproto' while
1614 * waiting for flow restore to complete. */
1615 if (ofproto_get_flow_restore_wait()) {
1616 return 0;
1617 }
1618
1619 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1620 port_run_fast(ofport);
1621 }
1622
1623 return 0;
1624 }
1625
1626 static int
1627 run(struct ofproto *ofproto_)
1628 {
1629 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1630 struct ofport_dpif *ofport;
1631 struct ofbundle *bundle;
1632 int error;
1633
1634 if (!clogged) {
1635 complete_operations(ofproto);
1636 }
1637
1638 /* Do not perform any periodic activity below required by 'ofproto' while
1639 * waiting for flow restore to complete. */
1640 if (ofproto_get_flow_restore_wait()) {
1641 return 0;
1642 }
1643
1644 error = run_fast(ofproto_);
1645 if (error) {
1646 return error;
1647 }
1648
1649 if (ofproto->netflow) {
1650 if (netflow_run(ofproto->netflow)) {
1651 send_netflow_active_timeouts(ofproto);
1652 }
1653 }
1654 if (ofproto->sflow) {
1655 dpif_sflow_run(ofproto->sflow);
1656 }
1657
1658 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1659 port_run(ofport);
1660 }
1661 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1662 bundle_run(bundle);
1663 }
1664
1665 stp_run(ofproto);
1666 mac_learning_run(ofproto->ml, &ofproto->backer->revalidate_set);
1667
1668 /* Check the consistency of a random facet, to aid debugging. */
1669 if (time_msec() >= ofproto->consistency_rl
1670 && !classifier_is_empty(&ofproto->facets)
1671 && !ofproto->backer->need_revalidate) {
1672 struct cls_table *table;
1673 struct cls_rule *cr;
1674 struct facet *facet;
1675
1676 ofproto->consistency_rl = time_msec() + 250;
1677
1678 table = CONTAINER_OF(hmap_random_node(&ofproto->facets.tables),
1679 struct cls_table, hmap_node);
1680 cr = CONTAINER_OF(hmap_random_node(&table->rules), struct cls_rule,
1681 hmap_node);
1682 facet = CONTAINER_OF(cr, struct facet, cr);
1683
1684 if (!tag_set_intersects(&ofproto->backer->revalidate_set,
1685 facet->xout.tags)) {
1686 if (!facet_check_consistency(facet)) {
1687 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
1688 }
1689 }
1690 }
1691
1692 return 0;
1693 }
1694
1695 static void
1696 wait(struct ofproto *ofproto_)
1697 {
1698 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1699 struct ofport_dpif *ofport;
1700 struct ofbundle *bundle;
1701
1702 if (!clogged && !list_is_empty(&ofproto->completions)) {
1703 poll_immediate_wake();
1704 }
1705
1706 if (ofproto_get_flow_restore_wait()) {
1707 return;
1708 }
1709
1710 dpif_wait(ofproto->backer->dpif);
1711 dpif_recv_wait(ofproto->backer->dpif);
1712 if (ofproto->sflow) {
1713 dpif_sflow_wait(ofproto->sflow);
1714 }
1715 if (!tag_set_is_empty(&ofproto->backer->revalidate_set)) {
1716 poll_immediate_wake();
1717 }
1718 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1719 port_wait(ofport);
1720 }
1721 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1722 bundle_wait(bundle);
1723 }
1724 if (ofproto->netflow) {
1725 netflow_wait(ofproto->netflow);
1726 }
1727 mac_learning_wait(ofproto->ml);
1728 stp_wait(ofproto);
1729 if (ofproto->backer->need_revalidate) {
1730 /* Shouldn't happen, but if it does just go around again. */
1731 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1732 poll_immediate_wake();
1733 }
1734 }
1735
1736 static void
1737 get_memory_usage(const struct ofproto *ofproto_, struct simap *usage)
1738 {
1739 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1740 struct cls_cursor cursor;
1741 size_t n_subfacets = 0;
1742 struct facet *facet;
1743
1744 simap_increase(usage, "facets", classifier_count(&ofproto->facets));
1745
1746 cls_cursor_init(&cursor, &ofproto->facets, NULL);
1747 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
1748 n_subfacets += list_size(&facet->subfacets);
1749 }
1750 simap_increase(usage, "subfacets", n_subfacets);
1751 }
1752
1753 static void
1754 flush(struct ofproto *ofproto_)
1755 {
1756 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1757 struct subfacet *subfacet, *next_subfacet;
1758 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
1759 int n_batch;
1760
1761 n_batch = 0;
1762 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
1763 &ofproto->backer->subfacets) {
1764 if (ofproto_dpif_cast(subfacet->facet->rule->up.ofproto) != ofproto) {
1765 continue;
1766 }
1767
1768 if (subfacet->path != SF_NOT_INSTALLED) {
1769 batch[n_batch++] = subfacet;
1770 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
1771 subfacet_destroy_batch(ofproto->backer, batch, n_batch);
1772 n_batch = 0;
1773 }
1774 } else {
1775 subfacet_destroy(subfacet);
1776 }
1777 }
1778
1779 if (n_batch > 0) {
1780 subfacet_destroy_batch(ofproto->backer, batch, n_batch);
1781 }
1782 }
1783
1784 static void
1785 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1786 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1787 {
1788 *arp_match_ip = true;
1789 *actions = (OFPUTIL_A_OUTPUT |
1790 OFPUTIL_A_SET_VLAN_VID |
1791 OFPUTIL_A_SET_VLAN_PCP |
1792 OFPUTIL_A_STRIP_VLAN |
1793 OFPUTIL_A_SET_DL_SRC |
1794 OFPUTIL_A_SET_DL_DST |
1795 OFPUTIL_A_SET_NW_SRC |
1796 OFPUTIL_A_SET_NW_DST |
1797 OFPUTIL_A_SET_NW_TOS |
1798 OFPUTIL_A_SET_TP_SRC |
1799 OFPUTIL_A_SET_TP_DST |
1800 OFPUTIL_A_ENQUEUE);
1801 }
1802
1803 static void
1804 get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots)
1805 {
1806 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1807 struct dpif_dp_stats s;
1808 uint64_t n_miss, n_no_pkt_in, n_bytes, n_dropped_frags;
1809 uint64_t n_lookup;
1810
1811 strcpy(ots->name, "classifier");
1812
1813 dpif_get_dp_stats(ofproto->backer->dpif, &s);
1814 rule_get_stats(&ofproto->miss_rule->up, &n_miss, &n_bytes);
1815 rule_get_stats(&ofproto->no_packet_in_rule->up, &n_no_pkt_in, &n_bytes);
1816 rule_get_stats(&ofproto->drop_frags_rule->up, &n_dropped_frags, &n_bytes);
1817
1818 n_lookup = s.n_hit + s.n_missed - n_dropped_frags;
1819 ots->lookup_count = htonll(n_lookup);
1820 ots->matched_count = htonll(n_lookup - n_miss - n_no_pkt_in);
1821 }
1822
1823 static struct ofport *
1824 port_alloc(void)
1825 {
1826 struct ofport_dpif *port = xmalloc(sizeof *port);
1827 return &port->up;
1828 }
1829
1830 static void
1831 port_dealloc(struct ofport *port_)
1832 {
1833 struct ofport_dpif *port = ofport_dpif_cast(port_);
1834 free(port);
1835 }
1836
1837 static int
1838 port_construct(struct ofport *port_)
1839 {
1840 struct ofport_dpif *port = ofport_dpif_cast(port_);
1841 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1842 const struct netdev *netdev = port->up.netdev;
1843 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1844 struct dpif_port dpif_port;
1845 int error;
1846
1847 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1848 port->bundle = NULL;
1849 port->cfm = NULL;
1850 port->bfd = NULL;
1851 port->tag = tag_create_random();
1852 port->may_enable = true;
1853 port->stp_port = NULL;
1854 port->stp_state = STP_DISABLED;
1855 port->tnl_port = NULL;
1856 hmap_init(&port->priorities);
1857 port->realdev_ofp_port = 0;
1858 port->vlandev_vid = 0;
1859 port->carrier_seq = netdev_get_carrier_resets(netdev);
1860
1861 if (netdev_vport_is_patch(netdev)) {
1862 /* By bailing out here, we don't submit the port to the sFlow module
1863 * to be considered for counter polling export. This is correct
1864 * because the patch port represents an interface that sFlow considers
1865 * to be "internal" to the switch as a whole, and therefore not an
1866 * candidate for counter polling. */
1867 port->odp_port = OVSP_NONE;
1868 return 0;
1869 }
1870
1871 error = dpif_port_query_by_name(ofproto->backer->dpif,
1872 netdev_vport_get_dpif_port(netdev, namebuf,
1873 sizeof namebuf),
1874 &dpif_port);
1875 if (error) {
1876 return error;
1877 }
1878
1879 port->odp_port = dpif_port.port_no;
1880
1881 if (netdev_get_tunnel_config(netdev)) {
1882 port->tnl_port = tnl_port_add(&port->up, port->odp_port);
1883 } else {
1884 /* Sanity-check that a mapping doesn't already exist. This
1885 * shouldn't happen for non-tunnel ports. */
1886 if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1887 VLOG_ERR("port %s already has an OpenFlow port number",
1888 dpif_port.name);
1889 dpif_port_destroy(&dpif_port);
1890 return EBUSY;
1891 }
1892
1893 hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1894 hash_int(port->odp_port, 0));
1895 }
1896 dpif_port_destroy(&dpif_port);
1897
1898 if (ofproto->sflow) {
1899 dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1900 }
1901
1902 return 0;
1903 }
1904
1905 static void
1906 port_destruct(struct ofport *port_)
1907 {
1908 struct ofport_dpif *port = ofport_dpif_cast(port_);
1909 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1910 const char *devname = netdev_get_name(port->up.netdev);
1911 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1912 const char *dp_port_name;
1913
1914 dp_port_name = netdev_vport_get_dpif_port(port->up.netdev, namebuf,
1915 sizeof namebuf);
1916 if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1917 /* The underlying device is still there, so delete it. This
1918 * happens when the ofproto is being destroyed, since the caller
1919 * assumes that removal of attached ports will happen as part of
1920 * destruction. */
1921 if (!port->tnl_port) {
1922 dpif_port_del(ofproto->backer->dpif, port->odp_port);
1923 }
1924 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1925 }
1926
1927 if (port->odp_port != OVSP_NONE && !port->tnl_port) {
1928 hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1929 }
1930
1931 tnl_port_del(port->tnl_port);
1932 sset_find_and_delete(&ofproto->ports, devname);
1933 sset_find_and_delete(&ofproto->ghost_ports, devname);
1934 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1935 bundle_remove(port_);
1936 set_cfm(port_, NULL);
1937 set_bfd(port_, NULL);
1938 if (ofproto->sflow) {
1939 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1940 }
1941
1942 ofport_clear_priorities(port);
1943 hmap_destroy(&port->priorities);
1944 }
1945
1946 static void
1947 port_modified(struct ofport *port_)
1948 {
1949 struct ofport_dpif *port = ofport_dpif_cast(port_);
1950
1951 if (port->bundle && port->bundle->bond) {
1952 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1953 }
1954
1955 if (port->cfm) {
1956 cfm_set_netdev(port->cfm, port->up.netdev);
1957 }
1958 }
1959
1960 static void
1961 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1962 {
1963 struct ofport_dpif *port = ofport_dpif_cast(port_);
1964 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1965 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1966
1967 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1968 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1969 OFPUTIL_PC_NO_PACKET_IN)) {
1970 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1971
1972 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1973 bundle_update(port->bundle);
1974 }
1975 }
1976 }
1977
1978 static int
1979 set_sflow(struct ofproto *ofproto_,
1980 const struct ofproto_sflow_options *sflow_options)
1981 {
1982 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1983 struct dpif_sflow *ds = ofproto->sflow;
1984
1985 if (sflow_options) {
1986 if (!ds) {
1987 struct ofport_dpif *ofport;
1988
1989 ds = ofproto->sflow = dpif_sflow_create();
1990 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1991 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1992 }
1993 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1994 }
1995 dpif_sflow_set_options(ds, sflow_options);
1996 } else {
1997 if (ds) {
1998 dpif_sflow_destroy(ds);
1999 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2000 ofproto->sflow = NULL;
2001 }
2002 }
2003 return 0;
2004 }
2005
2006 static int
2007 set_ipfix(
2008 struct ofproto *ofproto_,
2009 const struct ofproto_ipfix_bridge_exporter_options *bridge_exporter_options,
2010 const struct ofproto_ipfix_flow_exporter_options *flow_exporters_options,
2011 size_t n_flow_exporters_options)
2012 {
2013 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2014 struct dpif_ipfix *di = ofproto->ipfix;
2015
2016 if (bridge_exporter_options || flow_exporters_options) {
2017 if (!di) {
2018 di = ofproto->ipfix = dpif_ipfix_create();
2019 }
2020 dpif_ipfix_set_options(
2021 di, bridge_exporter_options, flow_exporters_options,
2022 n_flow_exporters_options);
2023 } else {
2024 if (di) {
2025 dpif_ipfix_destroy(di);
2026 ofproto->ipfix = NULL;
2027 }
2028 }
2029 return 0;
2030 }
2031
2032 static int
2033 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
2034 {
2035 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2036 int error;
2037
2038 if (!s) {
2039 error = 0;
2040 } else {
2041 if (!ofport->cfm) {
2042 struct ofproto_dpif *ofproto;
2043
2044 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2045 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2046 ofport->cfm = cfm_create(ofport->up.netdev);
2047 }
2048
2049 if (cfm_configure(ofport->cfm, s)) {
2050 return 0;
2051 }
2052
2053 error = EINVAL;
2054 }
2055 cfm_destroy(ofport->cfm);
2056 ofport->cfm = NULL;
2057 return error;
2058 }
2059
2060 static bool
2061 get_cfm_status(const struct ofport *ofport_,
2062 struct ofproto_cfm_status *status)
2063 {
2064 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2065
2066 if (ofport->cfm) {
2067 status->faults = cfm_get_fault(ofport->cfm);
2068 status->remote_opstate = cfm_get_opup(ofport->cfm);
2069 status->health = cfm_get_health(ofport->cfm);
2070 cfm_get_remote_mpids(ofport->cfm, &status->rmps, &status->n_rmps);
2071 return true;
2072 } else {
2073 return false;
2074 }
2075 }
2076
2077 static int
2078 set_bfd(struct ofport *ofport_, const struct smap *cfg)
2079 {
2080 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
2081 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2082 struct bfd *old;
2083
2084 old = ofport->bfd;
2085 ofport->bfd = bfd_configure(old, netdev_get_name(ofport->up.netdev), cfg);
2086 if (ofport->bfd != old) {
2087 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2088 }
2089
2090 return 0;
2091 }
2092
2093 static int
2094 get_bfd_status(struct ofport *ofport_, struct smap *smap)
2095 {
2096 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2097
2098 if (ofport->bfd) {
2099 bfd_get_status(ofport->bfd, smap);
2100 return 0;
2101 } else {
2102 return ENOENT;
2103 }
2104 }
2105 \f
2106 /* Spanning Tree. */
2107
2108 static void
2109 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
2110 {
2111 struct ofproto_dpif *ofproto = ofproto_;
2112 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
2113 struct ofport_dpif *ofport;
2114
2115 ofport = stp_port_get_aux(sp);
2116 if (!ofport) {
2117 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
2118 ofproto->up.name, port_num);
2119 } else {
2120 struct eth_header *eth = pkt->l2;
2121
2122 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
2123 if (eth_addr_is_zero(eth->eth_src)) {
2124 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
2125 "with unknown MAC", ofproto->up.name, port_num);
2126 } else {
2127 send_packet(ofport, pkt);
2128 }
2129 }
2130 ofpbuf_delete(pkt);
2131 }
2132
2133 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
2134 static int
2135 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
2136 {
2137 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2138
2139 /* Only revalidate flows if the configuration changed. */
2140 if (!s != !ofproto->stp) {
2141 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2142 }
2143
2144 if (s) {
2145 if (!ofproto->stp) {
2146 ofproto->stp = stp_create(ofproto_->name, s->system_id,
2147 send_bpdu_cb, ofproto);
2148 ofproto->stp_last_tick = time_msec();
2149 }
2150
2151 stp_set_bridge_id(ofproto->stp, s->system_id);
2152 stp_set_bridge_priority(ofproto->stp, s->priority);
2153 stp_set_hello_time(ofproto->stp, s->hello_time);
2154 stp_set_max_age(ofproto->stp, s->max_age);
2155 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
2156 } else {
2157 struct ofport *ofport;
2158
2159 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
2160 set_stp_port(ofport, NULL);
2161 }
2162
2163 stp_destroy(ofproto->stp);
2164 ofproto->stp = NULL;
2165 }
2166
2167 return 0;
2168 }
2169
2170 static int
2171 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
2172 {
2173 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2174
2175 if (ofproto->stp) {
2176 s->enabled = true;
2177 s->bridge_id = stp_get_bridge_id(ofproto->stp);
2178 s->designated_root = stp_get_designated_root(ofproto->stp);
2179 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
2180 } else {
2181 s->enabled = false;
2182 }
2183
2184 return 0;
2185 }
2186
2187 static void
2188 update_stp_port_state(struct ofport_dpif *ofport)
2189 {
2190 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2191 enum stp_state state;
2192
2193 /* Figure out new state. */
2194 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
2195 : STP_DISABLED;
2196
2197 /* Update state. */
2198 if (ofport->stp_state != state) {
2199 enum ofputil_port_state of_state;
2200 bool fwd_change;
2201
2202 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
2203 netdev_get_name(ofport->up.netdev),
2204 stp_state_name(ofport->stp_state),
2205 stp_state_name(state));
2206 if (stp_learn_in_state(ofport->stp_state)
2207 != stp_learn_in_state(state)) {
2208 /* xxx Learning action flows should also be flushed. */
2209 mac_learning_flush(ofproto->ml,
2210 &ofproto->backer->revalidate_set);
2211 }
2212 fwd_change = stp_forward_in_state(ofport->stp_state)
2213 != stp_forward_in_state(state);
2214
2215 ofproto->backer->need_revalidate = REV_STP;
2216 ofport->stp_state = state;
2217 ofport->stp_state_entered = time_msec();
2218
2219 if (fwd_change && ofport->bundle) {
2220 bundle_update(ofport->bundle);
2221 }
2222
2223 /* Update the STP state bits in the OpenFlow port description. */
2224 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
2225 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
2226 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
2227 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
2228 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
2229 : 0);
2230 ofproto_port_set_state(&ofport->up, of_state);
2231 }
2232 }
2233
2234 /* Configures STP on 'ofport_' using the settings defined in 's'. The
2235 * caller is responsible for assigning STP port numbers and ensuring
2236 * there are no duplicates. */
2237 static int
2238 set_stp_port(struct ofport *ofport_,
2239 const struct ofproto_port_stp_settings *s)
2240 {
2241 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2242 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2243 struct stp_port *sp = ofport->stp_port;
2244
2245 if (!s || !s->enable) {
2246 if (sp) {
2247 ofport->stp_port = NULL;
2248 stp_port_disable(sp);
2249 update_stp_port_state(ofport);
2250 }
2251 return 0;
2252 } else if (sp && stp_port_no(sp) != s->port_num
2253 && ofport == stp_port_get_aux(sp)) {
2254 /* The port-id changed, so disable the old one if it's not
2255 * already in use by another port. */
2256 stp_port_disable(sp);
2257 }
2258
2259 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
2260 stp_port_enable(sp);
2261
2262 stp_port_set_aux(sp, ofport);
2263 stp_port_set_priority(sp, s->priority);
2264 stp_port_set_path_cost(sp, s->path_cost);
2265
2266 update_stp_port_state(ofport);
2267
2268 return 0;
2269 }
2270
2271 static int
2272 get_stp_port_status(struct ofport *ofport_,
2273 struct ofproto_port_stp_status *s)
2274 {
2275 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2276 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2277 struct stp_port *sp = ofport->stp_port;
2278
2279 if (!ofproto->stp || !sp) {
2280 s->enabled = false;
2281 return 0;
2282 }
2283
2284 s->enabled = true;
2285 s->port_id = stp_port_get_id(sp);
2286 s->state = stp_port_get_state(sp);
2287 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
2288 s->role = stp_port_get_role(sp);
2289 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
2290
2291 return 0;
2292 }
2293
2294 static void
2295 stp_run(struct ofproto_dpif *ofproto)
2296 {
2297 if (ofproto->stp) {
2298 long long int now = time_msec();
2299 long long int elapsed = now - ofproto->stp_last_tick;
2300 struct stp_port *sp;
2301
2302 if (elapsed > 0) {
2303 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
2304 ofproto->stp_last_tick = now;
2305 }
2306 while (stp_get_changed_port(ofproto->stp, &sp)) {
2307 struct ofport_dpif *ofport = stp_port_get_aux(sp);
2308
2309 if (ofport) {
2310 update_stp_port_state(ofport);
2311 }
2312 }
2313
2314 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
2315 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
2316 }
2317 }
2318 }
2319
2320 static void
2321 stp_wait(struct ofproto_dpif *ofproto)
2322 {
2323 if (ofproto->stp) {
2324 poll_timer_wait(1000);
2325 }
2326 }
2327
2328 /* Returns true if STP should process 'flow'. */
2329 static bool
2330 stp_should_process_flow(const struct flow *flow)
2331 {
2332 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
2333 }
2334
2335 static void
2336 stp_process_packet(const struct ofport_dpif *ofport,
2337 const struct ofpbuf *packet)
2338 {
2339 struct ofpbuf payload = *packet;
2340 struct eth_header *eth = payload.data;
2341 struct stp_port *sp = ofport->stp_port;
2342
2343 /* Sink packets on ports that have STP disabled when the bridge has
2344 * STP enabled. */
2345 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
2346 return;
2347 }
2348
2349 /* Trim off padding on payload. */
2350 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
2351 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
2352 }
2353
2354 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
2355 stp_received_bpdu(sp, payload.data, payload.size);
2356 }
2357 }
2358 \f
2359 static struct priority_to_dscp *
2360 get_priority(const struct ofport_dpif *ofport, uint32_t priority)
2361 {
2362 struct priority_to_dscp *pdscp;
2363 uint32_t hash;
2364
2365 hash = hash_int(priority, 0);
2366 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) {
2367 if (pdscp->priority == priority) {
2368 return pdscp;
2369 }
2370 }
2371 return NULL;
2372 }
2373
2374 static void
2375 ofport_clear_priorities(struct ofport_dpif *ofport)
2376 {
2377 struct priority_to_dscp *pdscp, *next;
2378
2379 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) {
2380 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
2381 free(pdscp);
2382 }
2383 }
2384
2385 static int
2386 set_queues(struct ofport *ofport_,
2387 const struct ofproto_port_queue *qdscp_list,
2388 size_t n_qdscp)
2389 {
2390 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2391 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2392 struct hmap new = HMAP_INITIALIZER(&new);
2393 size_t i;
2394
2395 for (i = 0; i < n_qdscp; i++) {
2396 struct priority_to_dscp *pdscp;
2397 uint32_t priority;
2398 uint8_t dscp;
2399
2400 dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
2401 if (dpif_queue_to_priority(ofproto->backer->dpif, qdscp_list[i].queue,
2402 &priority)) {
2403 continue;
2404 }
2405
2406 pdscp = get_priority(ofport, priority);
2407 if (pdscp) {
2408 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
2409 } else {
2410 pdscp = xmalloc(sizeof *pdscp);
2411 pdscp->priority = priority;
2412 pdscp->dscp = dscp;
2413 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2414 }
2415
2416 if (pdscp->dscp != dscp) {
2417 pdscp->dscp = dscp;
2418 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2419 }
2420
2421 hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0));
2422 }
2423
2424 if (!hmap_is_empty(&ofport->priorities)) {
2425 ofport_clear_priorities(ofport);
2426 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2427 }
2428
2429 hmap_swap(&new, &ofport->priorities);
2430 hmap_destroy(&new);
2431
2432 return 0;
2433 }
2434 \f
2435 /* Bundles. */
2436
2437 /* Expires all MAC learning entries associated with 'bundle' and forces its
2438 * ofproto to revalidate every flow.
2439 *
2440 * Normally MAC learning entries are removed only from the ofproto associated
2441 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2442 * are removed from every ofproto. When patch ports and SLB bonds are in use
2443 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2444 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2445 * with the host from which it migrated. */
2446 static void
2447 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2448 {
2449 struct ofproto_dpif *ofproto = bundle->ofproto;
2450 struct mac_learning *ml = ofproto->ml;
2451 struct mac_entry *mac, *next_mac;
2452
2453 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2454 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2455 if (mac->port.p == bundle) {
2456 if (all_ofprotos) {
2457 struct ofproto_dpif *o;
2458
2459 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2460 if (o != ofproto) {
2461 struct mac_entry *e;
2462
2463 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan,
2464 NULL);
2465 if (e) {
2466 mac_learning_expire(o->ml, e);
2467 }
2468 }
2469 }
2470 }
2471
2472 mac_learning_expire(ml, mac);
2473 }
2474 }
2475 }
2476
2477 static struct ofbundle *
2478 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2479 {
2480 struct ofbundle *bundle;
2481
2482 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2483 &ofproto->bundles) {
2484 if (bundle->aux == aux) {
2485 return bundle;
2486 }
2487 }
2488 return NULL;
2489 }
2490
2491 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
2492 * ones that are found to 'bundles'. */
2493 static void
2494 bundle_lookup_multiple(struct ofproto_dpif *ofproto,
2495 void **auxes, size_t n_auxes,
2496 struct hmapx *bundles)
2497 {
2498 size_t i;
2499
2500 hmapx_init(bundles);
2501 for (i = 0; i < n_auxes; i++) {
2502 struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
2503 if (bundle) {
2504 hmapx_add(bundles, bundle);
2505 }
2506 }
2507 }
2508
2509 static void
2510 bundle_update(struct ofbundle *bundle)
2511 {
2512 struct ofport_dpif *port;
2513
2514 bundle->floodable = true;
2515 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2516 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2517 || !stp_forward_in_state(port->stp_state)) {
2518 bundle->floodable = false;
2519 break;
2520 }
2521 }
2522 }
2523
2524 static void
2525 bundle_del_port(struct ofport_dpif *port)
2526 {
2527 struct ofbundle *bundle = port->bundle;
2528
2529 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2530
2531 list_remove(&port->bundle_node);
2532 port->bundle = NULL;
2533
2534 if (bundle->lacp) {
2535 lacp_slave_unregister(bundle->lacp, port);
2536 }
2537 if (bundle->bond) {
2538 bond_slave_unregister(bundle->bond, port);
2539 }
2540
2541 bundle_update(bundle);
2542 }
2543
2544 static bool
2545 bundle_add_port(struct ofbundle *bundle, uint16_t ofp_port,
2546 struct lacp_slave_settings *lacp)
2547 {
2548 struct ofport_dpif *port;
2549
2550 port = get_ofp_port(bundle->ofproto, ofp_port);
2551 if (!port) {
2552 return false;
2553 }
2554
2555 if (port->bundle != bundle) {
2556 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2557 if (port->bundle) {
2558 bundle_del_port(port);
2559 }
2560
2561 port->bundle = bundle;
2562 list_push_back(&bundle->ports, &port->bundle_node);
2563 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2564 || !stp_forward_in_state(port->stp_state)) {
2565 bundle->floodable = false;
2566 }
2567 }
2568 if (lacp) {
2569 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2570 lacp_slave_register(bundle->lacp, port, lacp);
2571 }
2572
2573 return true;
2574 }
2575
2576 static void
2577 bundle_destroy(struct ofbundle *bundle)
2578 {
2579 struct ofproto_dpif *ofproto;
2580 struct ofport_dpif *port, *next_port;
2581 int i;
2582
2583 if (!bundle) {
2584 return;
2585 }
2586
2587 ofproto = bundle->ofproto;
2588 for (i = 0; i < MAX_MIRRORS; i++) {
2589 struct ofmirror *m = ofproto->mirrors[i];
2590 if (m) {
2591 if (m->out == bundle) {
2592 mirror_destroy(m);
2593 } else if (hmapx_find_and_delete(&m->srcs, bundle)
2594 || hmapx_find_and_delete(&m->dsts, bundle)) {
2595 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2596 }
2597 }
2598 }
2599
2600 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2601 bundle_del_port(port);
2602 }
2603
2604 bundle_flush_macs(bundle, true);
2605 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2606 free(bundle->name);
2607 free(bundle->trunks);
2608 lacp_destroy(bundle->lacp);
2609 bond_destroy(bundle->bond);
2610 free(bundle);
2611 }
2612
2613 static int
2614 bundle_set(struct ofproto *ofproto_, void *aux,
2615 const struct ofproto_bundle_settings *s)
2616 {
2617 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2618 bool need_flush = false;
2619 struct ofport_dpif *port;
2620 struct ofbundle *bundle;
2621 unsigned long *trunks;
2622 int vlan;
2623 size_t i;
2624 bool ok;
2625
2626 if (!s) {
2627 bundle_destroy(bundle_lookup(ofproto, aux));
2628 return 0;
2629 }
2630
2631 ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2632 ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2633
2634 bundle = bundle_lookup(ofproto, aux);
2635 if (!bundle) {
2636 bundle = xmalloc(sizeof *bundle);
2637
2638 bundle->ofproto = ofproto;
2639 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2640 hash_pointer(aux, 0));
2641 bundle->aux = aux;
2642 bundle->name = NULL;
2643
2644 list_init(&bundle->ports);
2645 bundle->vlan_mode = PORT_VLAN_TRUNK;
2646 bundle->vlan = -1;
2647 bundle->trunks = NULL;
2648 bundle->use_priority_tags = s->use_priority_tags;
2649 bundle->lacp = NULL;
2650 bundle->bond = NULL;
2651
2652 bundle->floodable = true;
2653
2654 bundle->src_mirrors = 0;
2655 bundle->dst_mirrors = 0;
2656 bundle->mirror_out = 0;
2657 }
2658
2659 if (!bundle->name || strcmp(s->name, bundle->name)) {
2660 free(bundle->name);
2661 bundle->name = xstrdup(s->name);
2662 }
2663
2664 /* LACP. */
2665 if (s->lacp) {
2666 if (!bundle->lacp) {
2667 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2668 bundle->lacp = lacp_create();
2669 }
2670 lacp_configure(bundle->lacp, s->lacp);
2671 } else {
2672 lacp_destroy(bundle->lacp);
2673 bundle->lacp = NULL;
2674 }
2675
2676 /* Update set of ports. */
2677 ok = true;
2678 for (i = 0; i < s->n_slaves; i++) {
2679 if (!bundle_add_port(bundle, s->slaves[i],
2680 s->lacp ? &s->lacp_slaves[i] : NULL)) {
2681 ok = false;
2682 }
2683 }
2684 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2685 struct ofport_dpif *next_port;
2686
2687 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2688 for (i = 0; i < s->n_slaves; i++) {
2689 if (s->slaves[i] == port->up.ofp_port) {
2690 goto found;
2691 }
2692 }
2693
2694 bundle_del_port(port);
2695 found: ;
2696 }
2697 }
2698 ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2699
2700 if (list_is_empty(&bundle->ports)) {
2701 bundle_destroy(bundle);
2702 return EINVAL;
2703 }
2704
2705 /* Set VLAN tagging mode */
2706 if (s->vlan_mode != bundle->vlan_mode
2707 || s->use_priority_tags != bundle->use_priority_tags) {
2708 bundle->vlan_mode = s->vlan_mode;
2709 bundle->use_priority_tags = s->use_priority_tags;
2710 need_flush = true;
2711 }
2712
2713 /* Set VLAN tag. */
2714 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2715 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2716 : 0);
2717 if (vlan != bundle->vlan) {
2718 bundle->vlan = vlan;
2719 need_flush = true;
2720 }
2721
2722 /* Get trunked VLANs. */
2723 switch (s->vlan_mode) {
2724 case PORT_VLAN_ACCESS:
2725 trunks = NULL;
2726 break;
2727
2728 case PORT_VLAN_TRUNK:
2729 trunks = CONST_CAST(unsigned long *, s->trunks);
2730 break;
2731
2732 case PORT_VLAN_NATIVE_UNTAGGED:
2733 case PORT_VLAN_NATIVE_TAGGED:
2734 if (vlan != 0 && (!s->trunks
2735 || !bitmap_is_set(s->trunks, vlan)
2736 || bitmap_is_set(s->trunks, 0))) {
2737 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2738 if (s->trunks) {
2739 trunks = bitmap_clone(s->trunks, 4096);
2740 } else {
2741 trunks = bitmap_allocate1(4096);
2742 }
2743 bitmap_set1(trunks, vlan);
2744 bitmap_set0(trunks, 0);
2745 } else {
2746 trunks = CONST_CAST(unsigned long *, s->trunks);
2747 }
2748 break;
2749
2750 default:
2751 NOT_REACHED();
2752 }
2753 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2754 free(bundle->trunks);
2755 if (trunks == s->trunks) {
2756 bundle->trunks = vlan_bitmap_clone(trunks);
2757 } else {
2758 bundle->trunks = trunks;
2759 trunks = NULL;
2760 }
2761 need_flush = true;
2762 }
2763 if (trunks != s->trunks) {
2764 free(trunks);
2765 }
2766
2767 /* Bonding. */
2768 if (!list_is_short(&bundle->ports)) {
2769 bundle->ofproto->has_bonded_bundles = true;
2770 if (bundle->bond) {
2771 if (bond_reconfigure(bundle->bond, s->bond)) {
2772 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2773 }
2774 } else {
2775 bundle->bond = bond_create(s->bond);
2776 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2777 }
2778
2779 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2780 bond_slave_register(bundle->bond, port, port->up.netdev);
2781 }
2782 } else {
2783 bond_destroy(bundle->bond);
2784 bundle->bond = NULL;
2785 }
2786
2787 /* If we changed something that would affect MAC learning, un-learn
2788 * everything on this port and force flow revalidation. */
2789 if (need_flush) {
2790 bundle_flush_macs(bundle, false);
2791 }
2792
2793 return 0;
2794 }
2795
2796 static void
2797 bundle_remove(struct ofport *port_)
2798 {
2799 struct ofport_dpif *port = ofport_dpif_cast(port_);
2800 struct ofbundle *bundle = port->bundle;
2801
2802 if (bundle) {
2803 bundle_del_port(port);
2804 if (list_is_empty(&bundle->ports)) {
2805 bundle_destroy(bundle);
2806 } else if (list_is_short(&bundle->ports)) {
2807 bond_destroy(bundle->bond);
2808 bundle->bond = NULL;
2809 }
2810 }
2811 }
2812
2813 static void
2814 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2815 {
2816 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2817 struct ofport_dpif *port = port_;
2818 uint8_t ea[ETH_ADDR_LEN];
2819 int error;
2820
2821 error = netdev_get_etheraddr(port->up.netdev, ea);
2822 if (!error) {
2823 struct ofpbuf packet;
2824 void *packet_pdu;
2825
2826 ofpbuf_init(&packet, 0);
2827 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2828 pdu_size);
2829 memcpy(packet_pdu, pdu, pdu_size);
2830
2831 send_packet(port, &packet);
2832 ofpbuf_uninit(&packet);
2833 } else {
2834 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2835 "%s (%s)", port->bundle->name,
2836 netdev_get_name(port->up.netdev), strerror(error));
2837 }
2838 }
2839
2840 static void
2841 bundle_send_learning_packets(struct ofbundle *bundle)
2842 {
2843 struct ofproto_dpif *ofproto = bundle->ofproto;
2844 int error, n_packets, n_errors;
2845 struct mac_entry *e;
2846
2847 error = n_packets = n_errors = 0;
2848 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2849 if (e->port.p != bundle) {
2850 struct ofpbuf *learning_packet;
2851 struct ofport_dpif *port;
2852 void *port_void;
2853 int ret;
2854
2855 /* The assignment to "port" is unnecessary but makes "grep"ing for
2856 * struct ofport_dpif more effective. */
2857 learning_packet = bond_compose_learning_packet(bundle->bond,
2858 e->mac, e->vlan,
2859 &port_void);
2860 port = port_void;
2861 ret = send_packet(port, learning_packet);
2862 ofpbuf_delete(learning_packet);
2863 if (ret) {
2864 error = ret;
2865 n_errors++;
2866 }
2867 n_packets++;
2868 }
2869 }
2870
2871 if (n_errors) {
2872 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2873 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2874 "packets, last error was: %s",
2875 bundle->name, n_errors, n_packets, strerror(error));
2876 } else {
2877 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2878 bundle->name, n_packets);
2879 }
2880 }
2881
2882 static void
2883 bundle_run(struct ofbundle *bundle)
2884 {
2885 if (bundle->lacp) {
2886 lacp_run(bundle->lacp, send_pdu_cb);
2887 }
2888 if (bundle->bond) {
2889 struct ofport_dpif *port;
2890
2891 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2892 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2893 }
2894
2895 bond_run(bundle->bond, &bundle->ofproto->backer->revalidate_set,
2896 lacp_status(bundle->lacp));
2897 if (bond_should_send_learning_packets(bundle->bond)) {
2898 bundle_send_learning_packets(bundle);
2899 }
2900 }
2901 }
2902
2903 static void
2904 bundle_wait(struct ofbundle *bundle)
2905 {
2906 if (bundle->lacp) {
2907 lacp_wait(bundle->lacp);
2908 }
2909 if (bundle->bond) {
2910 bond_wait(bundle->bond);
2911 }
2912 }
2913 \f
2914 /* Mirrors. */
2915
2916 static int
2917 mirror_scan(struct ofproto_dpif *ofproto)
2918 {
2919 int idx;
2920
2921 for (idx = 0; idx < MAX_MIRRORS; idx++) {
2922 if (!ofproto->mirrors[idx]) {
2923 return idx;
2924 }
2925 }
2926 return -1;
2927 }
2928
2929 static struct ofmirror *
2930 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
2931 {
2932 int i;
2933
2934 for (i = 0; i < MAX_MIRRORS; i++) {
2935 struct ofmirror *mirror = ofproto->mirrors[i];
2936 if (mirror && mirror->aux == aux) {
2937 return mirror;
2938 }
2939 }
2940
2941 return NULL;
2942 }
2943
2944 /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */
2945 static void
2946 mirror_update_dups(struct ofproto_dpif *ofproto)
2947 {
2948 int i;
2949
2950 for (i = 0; i < MAX_MIRRORS; i++) {
2951 struct ofmirror *m = ofproto->mirrors[i];
2952
2953 if (m) {
2954 m->dup_mirrors = MIRROR_MASK_C(1) << i;
2955 }
2956 }
2957
2958 for (i = 0; i < MAX_MIRRORS; i++) {
2959 struct ofmirror *m1 = ofproto->mirrors[i];
2960 int j;
2961
2962 if (!m1) {
2963 continue;
2964 }
2965
2966 for (j = i + 1; j < MAX_MIRRORS; j++) {
2967 struct ofmirror *m2 = ofproto->mirrors[j];
2968
2969 if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) {
2970 m1->dup_mirrors |= MIRROR_MASK_C(1) << j;
2971 m2->dup_mirrors |= m1->dup_mirrors;
2972 }
2973 }
2974 }
2975 }
2976
2977 static int
2978 mirror_set(struct ofproto *ofproto_, void *aux,
2979 const struct ofproto_mirror_settings *s)
2980 {
2981 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2982 mirror_mask_t mirror_bit;
2983 struct ofbundle *bundle;
2984 struct ofmirror *mirror;
2985 struct ofbundle *out;
2986 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
2987 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
2988 int out_vlan;
2989
2990 mirror = mirror_lookup(ofproto, aux);
2991 if (!s) {
2992 mirror_destroy(mirror);
2993 return 0;
2994 }
2995 if (!mirror) {
2996 int idx;
2997
2998 idx = mirror_scan(ofproto);
2999 if (idx < 0) {
3000 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
3001 "cannot create %s",
3002 ofproto->up.name, MAX_MIRRORS, s->name);
3003 return EFBIG;
3004 }
3005
3006 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
3007 mirror->ofproto = ofproto;
3008 mirror->idx = idx;
3009 mirror->aux = aux;
3010 mirror->out_vlan = -1;
3011 mirror->name = NULL;
3012 }
3013
3014 if (!mirror->name || strcmp(s->name, mirror->name)) {
3015 free(mirror->name);
3016 mirror->name = xstrdup(s->name);
3017 }
3018
3019 /* Get the new configuration. */
3020 if (s->out_bundle) {
3021 out = bundle_lookup(ofproto, s->out_bundle);
3022 if (!out) {
3023 mirror_destroy(mirror);
3024 return EINVAL;
3025 }
3026 out_vlan = -1;
3027 } else {
3028 out = NULL;
3029 out_vlan = s->out_vlan;
3030 }
3031 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
3032 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
3033
3034 /* If the configuration has not changed, do nothing. */
3035 if (hmapx_equals(&srcs, &mirror->srcs)
3036 && hmapx_equals(&dsts, &mirror->dsts)
3037 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
3038 && mirror->out == out
3039 && mirror->out_vlan == out_vlan)
3040 {
3041 hmapx_destroy(&srcs);
3042 hmapx_destroy(&dsts);
3043 return 0;
3044 }
3045
3046 hmapx_swap(&srcs, &mirror->srcs);
3047 hmapx_destroy(&srcs);
3048
3049 hmapx_swap(&dsts, &mirror->dsts);
3050 hmapx_destroy(&dsts);
3051
3052 free(mirror->vlans);
3053 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
3054
3055 mirror->out = out;
3056 mirror->out_vlan = out_vlan;
3057
3058 /* Update bundles. */
3059 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
3060 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
3061 if (hmapx_contains(&mirror->srcs, bundle)) {
3062 bundle->src_mirrors |= mirror_bit;
3063 } else {
3064 bundle->src_mirrors &= ~mirror_bit;
3065 }
3066
3067 if (hmapx_contains(&mirror->dsts, bundle)) {
3068 bundle->dst_mirrors |= mirror_bit;
3069 } else {
3070 bundle->dst_mirrors &= ~mirror_bit;
3071 }
3072
3073 if (mirror->out == bundle) {
3074 bundle->mirror_out |= mirror_bit;
3075 } else {
3076 bundle->mirror_out &= ~mirror_bit;
3077 }
3078 }
3079
3080 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3081 ofproto->has_mirrors = true;
3082 mac_learning_flush(ofproto->ml,
3083 &ofproto->backer->revalidate_set);
3084 mirror_update_dups(ofproto);
3085
3086 return 0;
3087 }
3088
3089 static void
3090 mirror_destroy(struct ofmirror *mirror)
3091 {
3092 struct ofproto_dpif *ofproto;
3093 mirror_mask_t mirror_bit;
3094 struct ofbundle *bundle;
3095 int i;
3096
3097 if (!mirror) {
3098 return;
3099 }
3100
3101 ofproto = mirror->ofproto;
3102 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3103 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
3104
3105 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
3106 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
3107 bundle->src_mirrors &= ~mirror_bit;
3108 bundle->dst_mirrors &= ~mirror_bit;
3109 bundle->mirror_out &= ~mirror_bit;
3110 }
3111
3112 hmapx_destroy(&mirror->srcs);
3113 hmapx_destroy(&mirror->dsts);
3114 free(mirror->vlans);
3115
3116 ofproto->mirrors[mirror->idx] = NULL;
3117 free(mirror->name);
3118 free(mirror);
3119
3120 mirror_update_dups(ofproto);
3121
3122 ofproto->has_mirrors = false;
3123 for (i = 0; i < MAX_MIRRORS; i++) {
3124 if (ofproto->mirrors[i]) {
3125 ofproto->has_mirrors = true;
3126 break;
3127 }
3128 }
3129 }
3130
3131 static int
3132 mirror_get_stats(struct ofproto *ofproto_, void *aux,
3133 uint64_t *packets, uint64_t *bytes)
3134 {
3135 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3136 struct ofmirror *mirror = mirror_lookup(ofproto, aux);
3137
3138 if (!mirror) {
3139 *packets = *bytes = UINT64_MAX;
3140 return 0;
3141 }
3142
3143 push_all_stats();
3144
3145 *packets = mirror->packet_count;
3146 *bytes = mirror->byte_count;
3147
3148 return 0;
3149 }
3150
3151 static int
3152 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
3153 {
3154 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3155 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
3156 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
3157 }
3158 return 0;
3159 }
3160
3161 static bool
3162 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
3163 {
3164 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3165 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
3166 return bundle && bundle->mirror_out != 0;
3167 }
3168
3169 static void
3170 forward_bpdu_changed(struct ofproto *ofproto_)
3171 {
3172 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3173 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3174 }
3175
3176 static void
3177 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
3178 size_t max_entries)
3179 {
3180 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3181 mac_learning_set_idle_time(ofproto->ml, idle_time);
3182 mac_learning_set_max_entries(ofproto->ml, max_entries);
3183 }
3184 \f
3185 /* Ports. */
3186
3187 static struct ofport_dpif *
3188 get_ofp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
3189 {
3190 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
3191 return ofport ? ofport_dpif_cast(ofport) : NULL;
3192 }
3193
3194 static struct ofport_dpif *
3195 get_odp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
3196 {
3197 struct ofport_dpif *port = odp_port_to_ofport(ofproto->backer, odp_port);
3198 return port && &ofproto->up == port->up.ofproto ? port : NULL;
3199 }
3200
3201 static void
3202 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
3203 struct ofproto_port *ofproto_port,
3204 struct dpif_port *dpif_port)
3205 {
3206 ofproto_port->name = dpif_port->name;
3207 ofproto_port->type = dpif_port->type;
3208 ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
3209 }
3210
3211 static struct ofport_dpif *
3212 ofport_get_peer(const struct ofport_dpif *ofport_dpif)
3213 {
3214 const struct ofproto_dpif *ofproto;
3215 const char *peer;
3216
3217 peer = netdev_vport_patch_peer(ofport_dpif->up.netdev);
3218 if (!peer) {
3219 return NULL;
3220 }
3221
3222 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
3223 struct ofport *ofport;
3224
3225 ofport = shash_find_data(&ofproto->up.port_by_name, peer);
3226 if (ofport && ofport->ofproto->ofproto_class == &ofproto_dpif_class) {
3227 return ofport_dpif_cast(ofport);
3228 }
3229 }
3230 return NULL;
3231 }
3232
3233 static void
3234 port_run_fast(struct ofport_dpif *ofport)
3235 {
3236 if (ofport->cfm && cfm_should_send_ccm(ofport->cfm)) {
3237 struct ofpbuf packet;
3238
3239 ofpbuf_init(&packet, 0);
3240 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
3241 send_packet(ofport, &packet);
3242 ofpbuf_uninit(&packet);
3243 }
3244
3245 if (ofport->bfd && bfd_should_send_packet(ofport->bfd)) {
3246 struct ofpbuf packet;
3247
3248 ofpbuf_init(&packet, 0);
3249 bfd_put_packet(ofport->bfd, &packet, ofport->up.pp.hw_addr);
3250 send_packet(ofport, &packet);
3251 ofpbuf_uninit(&packet);
3252 }
3253 }
3254
3255 static void
3256 port_run(struct ofport_dpif *ofport)
3257 {
3258 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
3259 bool carrier_changed = carrier_seq != ofport->carrier_seq;
3260 bool enable = netdev_get_carrier(ofport->up.netdev);
3261
3262 ofport->carrier_seq = carrier_seq;
3263
3264 port_run_fast(ofport);
3265
3266 if (ofport->tnl_port
3267 && tnl_port_reconfigure(&ofport->up, ofport->odp_port,
3268 &ofport->tnl_port)) {
3269 ofproto_dpif_cast(ofport->up.ofproto)->backer->need_revalidate = true;
3270 }
3271
3272 if (ofport->cfm) {
3273 int cfm_opup = cfm_get_opup(ofport->cfm);
3274
3275 cfm_run(ofport->cfm);
3276 enable = enable && !cfm_get_fault(ofport->cfm);
3277
3278 if (cfm_opup >= 0) {
3279 enable = enable && cfm_opup;
3280 }
3281 }
3282
3283 if (ofport->bfd) {
3284 bfd_run(ofport->bfd);
3285 enable = enable && bfd_forwarding(ofport->bfd);
3286 }
3287
3288 if (ofport->bundle) {
3289 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
3290 if (carrier_changed) {
3291 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
3292 }
3293 }
3294
3295 if (ofport->may_enable != enable) {
3296 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3297
3298 if (ofproto->has_bundle_action) {
3299 ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
3300 }
3301 }
3302
3303 ofport->may_enable = enable;
3304 }
3305
3306 static void
3307 port_wait(struct ofport_dpif *ofport)
3308 {
3309 if (ofport->cfm) {
3310 cfm_wait(ofport->cfm);
3311 }
3312
3313 if (ofport->bfd) {
3314 bfd_wait(ofport->bfd);
3315 }
3316 }
3317
3318 static int
3319 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
3320 struct ofproto_port *ofproto_port)
3321 {
3322 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3323 struct dpif_port dpif_port;
3324 int error;
3325
3326 if (sset_contains(&ofproto->ghost_ports, devname)) {
3327 const char *type = netdev_get_type_from_name(devname);
3328
3329 /* We may be called before ofproto->up.port_by_name is populated with
3330 * the appropriate ofport. For this reason, we must get the name and
3331 * type from the netdev layer directly. */
3332 if (type) {
3333 const struct ofport *ofport;
3334
3335 ofport = shash_find_data(&ofproto->up.port_by_name, devname);
3336 ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
3337 ofproto_port->name = xstrdup(devname);
3338 ofproto_port->type = xstrdup(type);
3339 return 0;
3340 }
3341 return ENODEV;
3342 }
3343
3344 if (!sset_contains(&ofproto->ports, devname)) {
3345 return ENODEV;
3346 }
3347 error = dpif_port_query_by_name(ofproto->backer->dpif,
3348 devname, &dpif_port);
3349 if (!error) {
3350 ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
3351 }
3352 return error;
3353 }
3354
3355 static int
3356 port_add(struct ofproto *ofproto_, struct netdev *netdev)
3357 {
3358 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3359 const char *devname = netdev_get_name(netdev);
3360 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
3361 const char *dp_port_name;
3362
3363 if (netdev_vport_is_patch(netdev)) {
3364 sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
3365 return 0;
3366 }
3367
3368 dp_port_name = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
3369 if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
3370 uint32_t port_no = UINT32_MAX;
3371 int error;
3372
3373 error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
3374 if (error) {
3375 return error;
3376 }
3377 if (netdev_get_tunnel_config(netdev)) {
3378 simap_put(&ofproto->backer->tnl_backers, dp_port_name, port_no);
3379 }
3380 }
3381
3382 if (netdev_get_tunnel_config(netdev)) {
3383 sset_add(&ofproto->ghost_ports, devname);
3384 } else {
3385 sset_add(&ofproto->ports, devname);
3386 }
3387 return 0;
3388 }
3389
3390 static int
3391 port_del(struct ofproto *ofproto_, uint16_t ofp_port)
3392 {
3393 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3394 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
3395 int error = 0;
3396
3397 if (!ofport) {
3398 return 0;
3399 }
3400
3401 sset_find_and_delete(&ofproto->ghost_ports,
3402 netdev_get_name(ofport->up.netdev));
3403 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3404 if (!ofport->tnl_port) {
3405 error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
3406 if (!error) {
3407 /* The caller is going to close ofport->up.netdev. If this is a
3408 * bonded port, then the bond is using that netdev, so remove it
3409 * from the bond. The client will need to reconfigure everything
3410 * after deleting ports, so then the slave will get re-added. */
3411 bundle_remove(&ofport->up);
3412 }
3413 }
3414 return error;
3415 }
3416
3417 static int
3418 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
3419 {
3420 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3421 int error;
3422
3423 push_all_stats();
3424
3425 error = netdev_get_stats(ofport->up.netdev, stats);
3426
3427 if (!error && ofport_->ofp_port == OFPP_LOCAL) {
3428 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3429
3430 /* ofproto->stats.tx_packets represents packets that we created
3431 * internally and sent to some port (e.g. packets sent with
3432 * send_packet()). Account for them as if they had come from
3433 * OFPP_LOCAL and got forwarded. */
3434
3435 if (stats->rx_packets != UINT64_MAX) {
3436 stats->rx_packets += ofproto->stats.tx_packets;
3437 }
3438
3439 if (stats->rx_bytes != UINT64_MAX) {
3440 stats->rx_bytes += ofproto->stats.tx_bytes;
3441 }
3442
3443 /* ofproto->stats.rx_packets represents packets that were received on
3444 * some port and we processed internally and dropped (e.g. STP).
3445 * Account for them as if they had been forwarded to OFPP_LOCAL. */
3446
3447 if (stats->tx_packets != UINT64_MAX) {
3448 stats->tx_packets += ofproto->stats.rx_packets;
3449 }
3450
3451 if (stats->tx_bytes != UINT64_MAX) {
3452 stats->tx_bytes += ofproto->stats.rx_bytes;
3453 }
3454 }
3455
3456 return error;
3457 }
3458
3459 struct port_dump_state {
3460 uint32_t bucket;
3461 uint32_t offset;
3462 bool ghost;
3463
3464 struct ofproto_port port;
3465 bool has_port;
3466 };
3467
3468 static int
3469 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
3470 {
3471 *statep = xzalloc(sizeof(struct port_dump_state));
3472 return 0;
3473 }
3474
3475 static int
3476 port_dump_next(const struct ofproto *ofproto_, void *state_,
3477 struct ofproto_port *port)
3478 {
3479 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3480 struct port_dump_state *state = state_;
3481 const struct sset *sset;
3482 struct sset_node *node;
3483
3484 if (state->has_port) {
3485 ofproto_port_destroy(&state->port);
3486 state->has_port = false;
3487 }
3488 sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
3489 while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
3490 int error;
3491
3492 error = port_query_by_name(ofproto_, node->name, &state->port);
3493 if (!error) {
3494 *port = state->port;
3495 state->has_port = true;
3496 return 0;
3497 } else if (error != ENODEV) {
3498 return error;
3499 }
3500 }
3501
3502 if (!state->ghost) {
3503 state->ghost = true;
3504 state->bucket = 0;
3505 state->offset = 0;
3506 return port_dump_next(ofproto_, state_, port);
3507 }
3508
3509 return EOF;
3510 }
3511
3512 static int
3513 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3514 {
3515 struct port_dump_state *state = state_;
3516
3517 if (state->has_port) {
3518 ofproto_port_destroy(&state->port);
3519 }
3520 free(state);
3521 return 0;
3522 }
3523
3524 static int
3525 port_poll(const struct ofproto *ofproto_, char **devnamep)
3526 {
3527 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3528
3529 if (ofproto->port_poll_errno) {
3530 int error = ofproto->port_poll_errno;
3531 ofproto->port_poll_errno = 0;
3532 return error;
3533 }
3534
3535 if (sset_is_empty(&ofproto->port_poll_set)) {
3536 return EAGAIN;
3537 }
3538
3539 *devnamep = sset_pop(&ofproto->port_poll_set);
3540 return 0;
3541 }
3542
3543 static void
3544 port_poll_wait(const struct ofproto *ofproto_)
3545 {
3546 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3547 dpif_port_poll_wait(ofproto->backer->dpif);
3548 }
3549
3550 static int
3551 port_is_lacp_current(const struct ofport *ofport_)
3552 {
3553 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3554 return (ofport->bundle && ofport->bundle->lacp
3555 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3556 : -1);
3557 }
3558 \f
3559 /* Upcall handling. */
3560
3561 /* Flow miss batching.
3562 *
3563 * Some dpifs implement operations faster when you hand them off in a batch.
3564 * To allow batching, "struct flow_miss" queues the dpif-related work needed
3565 * for a given flow. Each "struct flow_miss" corresponds to sending one or
3566 * more packets, plus possibly installing the flow in the dpif.
3567 *
3568 * So far we only batch the operations that affect flow setup time the most.
3569 * It's possible to batch more than that, but the benefit might be minimal. */
3570 struct flow_miss {
3571 struct hmap_node hmap_node;
3572 struct ofproto_dpif *ofproto;
3573 struct flow flow;
3574 enum odp_key_fitness key_fitness;
3575 const struct nlattr *key;
3576 size_t key_len;
3577 struct initial_vals initial_vals;
3578 struct list packets;
3579 enum dpif_upcall_type upcall_type;
3580 };
3581
3582 struct flow_miss_op {
3583 struct dpif_op dpif_op;
3584
3585 uint64_t slow_stub[128 / 8]; /* Buffer for compose_slow_path() */
3586 struct xlate_out xout;
3587 bool xout_garbage; /* 'xout' needs to be uninitialized? */
3588 };
3589
3590 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
3591 * OpenFlow controller as necessary according to their individual
3592 * configurations. */
3593 static void
3594 send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet,
3595 const struct flow *flow)
3596 {
3597 struct ofputil_packet_in pin;
3598
3599 pin.packet = packet->data;
3600 pin.packet_len = packet->size;
3601 pin.reason = OFPR_NO_MATCH;
3602 pin.controller_id = 0;
3603
3604 pin.table_id = 0;
3605 pin.cookie = 0;
3606
3607 pin.send_len = 0; /* not used for flow table misses */
3608
3609 flow_get_metadata(flow, &pin.fmd);
3610
3611 connmgr_send_packet_in(ofproto->up.connmgr, &pin);
3612 }
3613
3614 static enum slow_path_reason
3615 process_special(struct ofproto_dpif *ofproto, const struct flow *flow,
3616 const struct ofport_dpif *ofport, const struct ofpbuf *packet)
3617 {
3618 if (!ofport) {
3619 return 0;
3620 } else if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) {
3621 if (packet) {
3622 cfm_process_heartbeat(ofport->cfm, packet);
3623 }
3624 return SLOW_CFM;
3625 } else if (ofport->bfd && bfd_should_process_flow(flow)) {
3626 if (packet) {
3627 bfd_process_packet(ofport->bfd, flow, packet);
3628 }
3629 return SLOW_BFD;
3630 } else if (ofport->bundle && ofport->bundle->lacp
3631 && flow->dl_type == htons(ETH_TYPE_LACP)) {
3632 if (packet) {
3633 lacp_process_packet(ofport->bundle->lacp, ofport, packet);
3634 }
3635 return SLOW_LACP;
3636 } else if (ofproto->stp && stp_should_process_flow(flow)) {
3637 if (packet) {
3638 stp_process_packet(ofport, packet);
3639 }
3640 return SLOW_STP;
3641 } else {
3642 return 0;
3643 }
3644 }
3645
3646 static struct flow_miss *
3647 flow_miss_find(struct hmap *todo, const struct ofproto_dpif *ofproto,
3648 const struct flow *flow, uint32_t hash)
3649 {
3650 struct flow_miss *miss;
3651
3652 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
3653 if (miss->ofproto == ofproto && flow_equal(&miss->flow, flow)) {
3654 return miss;
3655 }
3656 }
3657
3658 return NULL;
3659 }
3660
3661 /* Partially Initializes 'op' as an "execute" operation for 'miss' and
3662 * 'packet'. The caller must initialize op->actions and op->actions_len. If
3663 * 'miss' is associated with a subfacet the caller must also initialize the
3664 * returned op->subfacet, and if anything needs to be freed after processing
3665 * the op, the caller must initialize op->garbage also. */
3666 static void
3667 init_flow_miss_execute_op(struct flow_miss *miss, struct ofpbuf *packet,
3668 struct flow_miss_op *op)
3669 {
3670 if (miss->flow.vlan_tci != miss->initial_vals.vlan_tci) {
3671 /* This packet was received on a VLAN splinter port. We
3672 * added a VLAN to the packet to make the packet resemble
3673 * the flow, but the actions were composed assuming that
3674 * the packet contained no VLAN. So, we must remove the
3675 * VLAN header from the packet before trying to execute the
3676 * actions. */
3677 eth_pop_vlan(packet);
3678 }
3679
3680 op->xout_garbage = false;
3681 op->dpif_op.type = DPIF_OP_EXECUTE;
3682 op->dpif_op.u.execute.key = miss->key;
3683 op->dpif_op.u.execute.key_len = miss->key_len;
3684 op->dpif_op.u.execute.packet = packet;
3685 }
3686
3687 /* Helper for handle_flow_miss_without_facet() and
3688 * handle_flow_miss_with_facet(). */
3689 static void
3690 handle_flow_miss_common(struct rule_dpif *rule,
3691 struct ofpbuf *packet, const struct flow *flow)
3692 {
3693 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3694
3695 if (rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
3696 /*
3697 * Extra-special case for fail-open mode.
3698 *
3699 * We are in fail-open mode and the packet matched the fail-open
3700 * rule, but we are connected to a controller too. We should send
3701 * the packet up to the controller in the hope that it will try to
3702 * set up a flow and thereby allow us to exit fail-open.
3703 *
3704 * See the top-level comment in fail-open.c for more information.
3705 */
3706 send_packet_in_miss(ofproto, packet, flow);
3707 }
3708 }
3709
3710 /* Figures out whether a flow that missed in 'ofproto', whose details are in
3711 * 'miss' masked by 'wc', is likely to be worth tracking in detail in userspace
3712 * and (usually) installing a datapath flow. The answer is usually "yes" (a
3713 * return value of true). However, for short flows the cost of bookkeeping is
3714 * much higher than the benefits, so when the datapath holds a large number of
3715 * flows we impose some heuristics to decide which flows are likely to be worth
3716 * tracking. */
3717 static bool
3718 flow_miss_should_make_facet(struct flow_miss *miss, struct flow_wildcards *wc)
3719 {
3720 struct dpif_backer *backer = miss->ofproto->backer;
3721 uint32_t hash;
3722
3723 if (!backer->governor) {
3724 size_t n_subfacets;
3725
3726 n_subfacets = hmap_count(&backer->subfacets);
3727 if (n_subfacets * 2 <= flow_eviction_threshold) {
3728 return true;
3729 }
3730
3731 backer->governor = governor_create();
3732 }
3733
3734 hash = flow_hash_in_wildcards(&miss->flow, wc, 0);
3735 return governor_should_install_flow(backer->governor, hash,
3736 list_size(&miss->packets));
3737 }
3738
3739 /* Handles 'miss' without creating a facet or subfacet or creating any datapath
3740 * flow. 'miss->flow' must have matched 'rule' and been xlated into 'xout'.
3741 * May add an "execute" operation to 'ops' and increment '*n_ops'. */
3742 static void
3743 handle_flow_miss_without_facet(struct rule_dpif *rule, struct xlate_out *xout,
3744 struct flow_miss *miss,
3745 struct flow_miss_op *ops, size_t *n_ops)
3746 {
3747 struct ofpbuf *packet;
3748
3749 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3750
3751 COVERAGE_INC(facet_suppress);
3752
3753 handle_flow_miss_common(rule, packet, &miss->flow);
3754
3755 if (xout->slow) {
3756 struct xlate_in xin;
3757
3758 xlate_in_init(&xin, miss->ofproto, &miss->flow,
3759 &miss->initial_vals, rule, 0, packet);
3760 xlate_actions_for_side_effects(&xin);
3761 }
3762
3763 if (xout->odp_actions.size) {
3764 struct flow_miss_op *op = &ops[*n_ops];
3765 struct dpif_execute *execute = &op->dpif_op.u.execute;
3766
3767 init_flow_miss_execute_op(miss, packet, op);
3768 xlate_out_copy(&op->xout, xout);
3769 execute->actions = op->xout.odp_actions.data;
3770 execute->actions_len = op->xout.odp_actions.size;
3771 op->xout_garbage = true;
3772
3773 (*n_ops)++;
3774 }
3775 }
3776 }
3777
3778 /* Handles 'miss', which matches 'facet'. May add any required datapath
3779 * operations to 'ops', incrementing '*n_ops' for each new op.
3780 *
3781 * All of the packets in 'miss' are considered to have arrived at time 'now'.
3782 * This is really important only for new facets: if we just called time_msec()
3783 * here, then the new subfacet or its packets could look (occasionally) as
3784 * though it was used some time after the facet was used. That can make a
3785 * one-packet flow look like it has a nonzero duration, which looks odd in
3786 * e.g. NetFlow statistics.
3787 *
3788 * If non-null, 'stats' will be folded into 'facet'. */
3789 static void
3790 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
3791 long long int now, struct dpif_flow_stats *stats,
3792 struct flow_miss_op *ops, size_t *n_ops)
3793 {
3794 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3795 enum subfacet_path want_path;
3796 struct subfacet *subfacet;
3797 struct ofpbuf *packet;
3798
3799 subfacet = subfacet_create(facet, miss, now);
3800 want_path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
3801 if (stats) {
3802 subfacet_update_stats(subfacet, stats);
3803 }
3804
3805 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3806 struct flow_miss_op *op = &ops[*n_ops];
3807
3808 handle_flow_miss_common(facet->rule, packet, &miss->flow);
3809
3810 if (want_path != SF_FAST_PATH) {
3811 struct xlate_in xin;
3812
3813 xlate_in_init(&xin, ofproto, &facet->flow, &facet->initial_vals,
3814 facet->rule, 0, packet);
3815 xlate_actions_for_side_effects(&xin);
3816 }
3817
3818 if (facet->xout.odp_actions.size) {
3819 struct dpif_execute *execute = &op->dpif_op.u.execute;
3820
3821 init_flow_miss_execute_op(miss, packet, op);
3822 execute->actions = facet->xout.odp_actions.data,
3823 execute->actions_len = facet->xout.odp_actions.size;
3824 (*n_ops)++;
3825 }
3826 }
3827
3828 if (miss->upcall_type == DPIF_UC_MISS || subfacet->path != want_path) {
3829 struct flow_miss_op *op = &ops[(*n_ops)++];
3830 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
3831
3832 subfacet->path = want_path;
3833
3834 op->xout_garbage = false;
3835 op->dpif_op.type = DPIF_OP_FLOW_PUT;
3836 put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
3837 put->key = miss->key;
3838 put->key_len = miss->key_len;
3839 if (want_path == SF_FAST_PATH) {
3840 put->actions = facet->xout.odp_actions.data;
3841 put->actions_len = facet->xout.odp_actions.size;
3842 } else {
3843 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
3844 op->slow_stub, sizeof op->slow_stub,
3845 &put->actions, &put->actions_len);
3846 }
3847 put->stats = NULL;
3848 }
3849 }
3850
3851 /* Handles flow miss 'miss'. May add any required datapath operations
3852 * to 'ops', incrementing '*n_ops' for each new op. */
3853 static void
3854 handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops,
3855 size_t *n_ops)
3856 {
3857 struct ofproto_dpif *ofproto = miss->ofproto;
3858 struct dpif_flow_stats stats__;
3859 struct dpif_flow_stats *stats = &stats__;
3860 struct ofpbuf *packet;
3861 struct facet *facet;
3862 long long int now;
3863
3864 now = time_msec();
3865 memset(stats, 0, sizeof *stats);
3866 stats->used = now;
3867 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3868 stats->tcp_flags |= packet_get_tcp_flags(packet, &miss->flow);
3869 stats->n_bytes += packet->size;
3870 stats->n_packets++;
3871 }
3872
3873 facet = facet_lookup_valid(ofproto, &miss->flow);
3874 if (!facet) {
3875 struct flow_wildcards wc;
3876 struct rule_dpif *rule;
3877 struct xlate_out xout;
3878 struct xlate_in xin;
3879
3880 flow_wildcards_init_catchall(&wc);
3881 rule = rule_dpif_lookup(ofproto, &miss->flow, &wc);
3882 rule_credit_stats(rule, stats);
3883
3884 xlate_in_init(&xin, ofproto, &miss->flow, &miss->initial_vals, rule,
3885 stats->tcp_flags, NULL);
3886 xin.resubmit_stats = stats;
3887 xin.may_learn = true;
3888 xlate_actions(&xin, &xout);
3889 flow_wildcards_or(&xout.wc, &xout.wc, &wc);
3890
3891 /* There does not exist a bijection between 'struct flow' and datapath
3892 * flow keys with fitness ODP_FIT_TO_LITTLE. This breaks a fundamental
3893 * assumption used throughout the facet and subfacet handling code.
3894 * Since we have to handle these misses in userspace anyway, we simply
3895 * skip facet creation, avoiding the problem altogether. */
3896 if (miss->key_fitness == ODP_FIT_TOO_LITTLE
3897 || !flow_miss_should_make_facet(miss, &xout.wc)) {
3898 handle_flow_miss_without_facet(rule, &xout, miss, ops, n_ops);
3899 return;
3900 }
3901
3902 facet = facet_create(miss, rule, &xout, stats);
3903 stats = NULL;
3904 }
3905 handle_flow_miss_with_facet(miss, facet, now, stats, ops, n_ops);
3906 }
3907
3908 static struct drop_key *
3909 drop_key_lookup(const struct dpif_backer *backer, const struct nlattr *key,
3910 size_t key_len)
3911 {
3912 struct drop_key *drop_key;
3913
3914 HMAP_FOR_EACH_WITH_HASH (drop_key, hmap_node, hash_bytes(key, key_len, 0),
3915 &backer->drop_keys) {
3916 if (drop_key->key_len == key_len
3917 && !memcmp(drop_key->key, key, key_len)) {
3918 return drop_key;
3919 }
3920 }
3921 return NULL;
3922 }
3923
3924 static void
3925 drop_key_clear(struct dpif_backer *backer)
3926 {
3927 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3928 struct drop_key *drop_key, *next;
3929
3930 HMAP_FOR_EACH_SAFE (drop_key, next, hmap_node, &backer->drop_keys) {
3931 int error;
3932
3933 error = dpif_flow_del(backer->dpif, drop_key->key, drop_key->key_len,
3934 NULL);
3935 if (error && !VLOG_DROP_WARN(&rl)) {
3936 struct ds ds = DS_EMPTY_INITIALIZER;
3937 odp_flow_key_format(drop_key->key, drop_key->key_len, &ds);
3938 VLOG_WARN("Failed to delete drop key (%s) (%s)", strerror(error),
3939 ds_cstr(&ds));
3940 ds_destroy(&ds);
3941 }
3942
3943 hmap_remove(&backer->drop_keys, &drop_key->hmap_node);
3944 free(drop_key->key);
3945 free(drop_key);
3946 }
3947 }
3948
3949 /* Given a datpath, packet, and flow metadata ('backer', 'packet', and 'key'
3950 * respectively), populates 'flow' with the result of odp_flow_key_to_flow().
3951 * Optionally, if nonnull, populates 'fitnessp' with the fitness of 'flow' as
3952 * returned by odp_flow_key_to_flow(). Also, optionally populates 'ofproto'
3953 * with the ofproto_dpif, and 'odp_in_port' with the datapath in_port, that
3954 * 'packet' ingressed.
3955 *
3956 * If 'ofproto' is nonnull, requires 'flow''s in_port to exist. Otherwise sets
3957 * 'flow''s in_port to OFPP_NONE.
3958 *
3959 * This function does post-processing on data returned from
3960 * odp_flow_key_to_flow() to help make VLAN splinters transparent to the rest
3961 * of the upcall processing logic. In particular, if the extracted in_port is
3962 * a VLAN splinter port, it replaces flow->in_port by the "real" port, sets
3963 * flow->vlan_tci correctly for the VLAN of the VLAN splinter port, and pushes
3964 * a VLAN header onto 'packet' (if it is nonnull).
3965 *
3966 * Optionally, if 'initial_vals' is nonnull, sets 'initial_vals->vlan_tci'
3967 * to the VLAN TCI with which the packet was really received, that is, the
3968 * actual VLAN TCI extracted by odp_flow_key_to_flow(). (This differs from
3969 * the value returned in flow->vlan_tci only for packets received on
3970 * VLAN splinters.)
3971 *
3972 * Similarly, this function also includes some logic to help with tunnels. It
3973 * may modify 'flow' as necessary to make the tunneling implementation
3974 * transparent to the upcall processing logic.
3975 *
3976 * Returns 0 if successful, ENODEV if the parsed flow has no associated ofport,
3977 * or some other positive errno if there are other problems. */
3978 static int
3979 ofproto_receive(const struct dpif_backer *backer, struct ofpbuf *packet,
3980 const struct nlattr *key, size_t key_len,
3981 struct flow *flow, enum odp_key_fitness *fitnessp,
3982 struct ofproto_dpif **ofproto, uint32_t *odp_in_port,
3983 struct initial_vals *initial_vals)
3984 {
3985 const struct ofport_dpif *port;
3986 enum odp_key_fitness fitness;
3987 int error = ENODEV;
3988
3989 fitness = odp_flow_key_to_flow(key, key_len, flow);
3990 if (fitness == ODP_FIT_ERROR) {
3991 error = EINVAL;
3992 goto exit;
3993 }
3994
3995 if (initial_vals) {
3996 initial_vals->vlan_tci = flow->vlan_tci;
3997 }
3998
3999 if (odp_in_port) {
4000 *odp_in_port = flow->in_port;
4001 }
4002
4003 port = (tnl_port_should_receive(flow)
4004 ? ofport_dpif_cast(tnl_port_receive(flow))
4005 : odp_port_to_ofport(backer, flow->in_port));
4006 flow->in_port = port ? port->up.ofp_port : OFPP_NONE;
4007 if (!port) {
4008 goto exit;
4009 }
4010
4011 /* XXX: Since the tunnel module is not scoped per backer, for a tunnel port
4012 * it's theoretically possible that we'll receive an ofport belonging to an
4013 * entirely different datapath. In practice, this can't happen because no
4014 * platforms has two separate datapaths which each support tunneling. */
4015 ovs_assert(ofproto_dpif_cast(port->up.ofproto)->backer == backer);
4016
4017 if (vsp_adjust_flow(ofproto_dpif_cast(port->up.ofproto), flow)) {
4018 if (packet) {
4019 /* Make the packet resemble the flow, so that it gets sent to
4020 * an OpenFlow controller properly, so that it looks correct
4021 * for sFlow, and so that flow_extract() will get the correct
4022 * vlan_tci if it is called on 'packet'.
4023 *
4024 * The allocated space inside 'packet' probably also contains
4025 * 'key', that is, both 'packet' and 'key' are probably part of
4026 * a struct dpif_upcall (see the large comment on that
4027 * structure definition), so pushing data on 'packet' is in
4028 * general not a good idea since it could overwrite 'key' or
4029 * free it as a side effect. However, it's OK in this special
4030 * case because we know that 'packet' is inside a Netlink
4031 * attribute: pushing 4 bytes will just overwrite the 4-byte
4032 * "struct nlattr", which is fine since we don't need that
4033 * header anymore. */
4034 eth_push_vlan(packet, flow->vlan_tci);
4035 }
4036 /* We can't reproduce 'key' from 'flow'. */
4037 fitness = fitness == ODP_FIT_PERFECT ? ODP_FIT_TOO_MUCH : fitness;
4038 }
4039 error = 0;
4040
4041 if (ofproto) {
4042 *ofproto = ofproto_dpif_cast(port->up.ofproto);
4043 }
4044
4045 exit:
4046 if (fitnessp) {
4047 *fitnessp = fitness;
4048 }
4049 return error;
4050 }
4051
4052 static void
4053 handle_miss_upcalls(struct dpif_backer *backer, struct dpif_upcall *upcalls,
4054 size_t n_upcalls)
4055 {
4056 struct dpif_upcall *upcall;
4057 struct flow_miss *miss;
4058 struct flow_miss misses[FLOW_MISS_MAX_BATCH];
4059 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
4060 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
4061 struct hmap todo;
4062 int n_misses;
4063 size_t n_ops;
4064 size_t i;
4065
4066 if (!n_upcalls) {
4067 return;
4068 }
4069
4070 /* Construct the to-do list.
4071 *
4072 * This just amounts to extracting the flow from each packet and sticking
4073 * the packets that have the same flow in the same "flow_miss" structure so
4074 * that we can process them together. */
4075 hmap_init(&todo);
4076 n_misses = 0;
4077 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
4078 struct flow_miss *miss = &misses[n_misses];
4079 struct flow_miss *existing_miss;
4080 struct ofproto_dpif *ofproto;
4081 uint32_t odp_in_port;
4082 struct flow flow;
4083 uint32_t hash;
4084 int error;
4085
4086 error = ofproto_receive(backer, upcall->packet, upcall->key,
4087 upcall->key_len, &flow, &miss->key_fitness,
4088 &ofproto, &odp_in_port, &miss->initial_vals);
4089 if (error == ENODEV) {
4090 struct drop_key *drop_key;
4091
4092 /* Received packet on datapath port for which we couldn't
4093 * associate an ofproto. This can happen if a port is removed
4094 * while traffic is being received. Print a rate-limited message
4095 * in case it happens frequently. Install a drop flow so
4096 * that future packets of the flow are inexpensively dropped
4097 * in the kernel. */
4098 VLOG_INFO_RL(&rl, "received packet on unassociated datapath port "
4099 "%"PRIu32, odp_in_port);
4100
4101 drop_key = drop_key_lookup(backer, upcall->key, upcall->key_len);
4102 if (!drop_key) {
4103 drop_key = xmalloc(sizeof *drop_key);
4104 drop_key->key = xmemdup(upcall->key, upcall->key_len);
4105 drop_key->key_len = upcall->key_len;
4106
4107 hmap_insert(&backer->drop_keys, &drop_key->hmap_node,
4108 hash_bytes(drop_key->key, drop_key->key_len, 0));
4109 dpif_flow_put(backer->dpif, DPIF_FP_CREATE | DPIF_FP_MODIFY,
4110 drop_key->key, drop_key->key_len, NULL, 0, NULL);
4111 }
4112 continue;
4113 }
4114 if (error) {
4115 continue;
4116 }
4117
4118 ofproto->n_missed++;
4119 flow_extract(upcall->packet, flow.skb_priority, flow.skb_mark,
4120 &flow.tunnel, flow.in_port, &miss->flow);
4121
4122 /* Add other packets to a to-do list. */
4123 hash = flow_hash(&miss->flow, 0);
4124 existing_miss = flow_miss_find(&todo, ofproto, &miss->flow, hash);
4125 if (!existing_miss) {
4126 hmap_insert(&todo, &miss->hmap_node, hash);
4127 miss->ofproto = ofproto;
4128 miss->key = upcall->key;
4129 miss->key_len = upcall->key_len;
4130 miss->upcall_type = upcall->type;
4131 list_init(&miss->packets);
4132
4133 n_misses++;
4134 } else {
4135 miss = existing_miss;
4136 }
4137 list_push_back(&miss->packets, &upcall->packet->list_node);
4138 }
4139
4140 /* Process each element in the to-do list, constructing the set of
4141 * operations to batch. */
4142 n_ops = 0;
4143 HMAP_FOR_EACH (miss, hmap_node, &todo) {
4144 handle_flow_miss(miss, flow_miss_ops, &n_ops);
4145 }
4146 ovs_assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
4147
4148 /* Execute batch. */
4149 for (i = 0; i < n_ops; i++) {
4150 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
4151 }
4152 dpif_operate(backer->dpif, dpif_ops, n_ops);
4153
4154 /* Free memory. */
4155 for (i = 0; i < n_ops; i++) {
4156 if (flow_miss_ops[i].xout_garbage) {
4157 xlate_out_uninit(&flow_miss_ops[i].xout);
4158 }
4159 }
4160 hmap_destroy(&todo);
4161 }
4162
4163 static enum { SFLOW_UPCALL, MISS_UPCALL, BAD_UPCALL, FLOW_SAMPLE_UPCALL,
4164 IPFIX_UPCALL }
4165 classify_upcall(const struct dpif_upcall *upcall)
4166 {
4167 size_t userdata_len;
4168 union user_action_cookie cookie;
4169
4170 /* First look at the upcall type. */
4171 switch (upcall->type) {
4172 case DPIF_UC_ACTION:
4173 break;
4174
4175 case DPIF_UC_MISS:
4176 return MISS_UPCALL;
4177
4178 case DPIF_N_UC_TYPES:
4179 default:
4180 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
4181 return BAD_UPCALL;
4182 }
4183
4184 /* "action" upcalls need a closer look. */
4185 if (!upcall->userdata) {
4186 VLOG_WARN_RL(&rl, "action upcall missing cookie");
4187 return BAD_UPCALL;
4188 }
4189 userdata_len = nl_attr_get_size(upcall->userdata);
4190 if (userdata_len < sizeof cookie.type
4191 || userdata_len > sizeof cookie) {
4192 VLOG_WARN_RL(&rl, "action upcall cookie has unexpected size %zu",
4193 userdata_len);
4194 return BAD_UPCALL;
4195 }
4196 memset(&cookie, 0, sizeof cookie);
4197 memcpy(&cookie, nl_attr_get(upcall->userdata), userdata_len);
4198 if (userdata_len == sizeof cookie.sflow
4199 && cookie.type == USER_ACTION_COOKIE_SFLOW) {
4200 return SFLOW_UPCALL;
4201 } else if (userdata_len == sizeof cookie.slow_path
4202 && cookie.type == USER_ACTION_COOKIE_SLOW_PATH) {
4203 return MISS_UPCALL;
4204 } else if (userdata_len == sizeof cookie.flow_sample
4205 && cookie.type == USER_ACTION_COOKIE_FLOW_SAMPLE) {
4206 return FLOW_SAMPLE_UPCALL;
4207 } else if (userdata_len == sizeof cookie.ipfix
4208 && cookie.type == USER_ACTION_COOKIE_IPFIX) {
4209 return IPFIX_UPCALL;
4210 } else {
4211 VLOG_WARN_RL(&rl, "invalid user cookie of type %"PRIu16
4212 " and size %zu", cookie.type, userdata_len);
4213 return BAD_UPCALL;
4214 }
4215 }
4216
4217 static void
4218 handle_sflow_upcall(struct dpif_backer *backer,
4219 const struct dpif_upcall *upcall)
4220 {
4221 struct ofproto_dpif *ofproto;
4222 union user_action_cookie cookie;
4223 struct flow flow;
4224 uint32_t odp_in_port;
4225
4226 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
4227 &flow, NULL, &ofproto, &odp_in_port, NULL)
4228 || !ofproto->sflow) {
4229 return;
4230 }
4231
4232 memset(&cookie, 0, sizeof cookie);
4233 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.sflow);
4234 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
4235 odp_in_port, &cookie);
4236 }
4237
4238 static void
4239 handle_flow_sample_upcall(struct dpif_backer *backer,
4240 const struct dpif_upcall *upcall)
4241 {
4242 struct ofproto_dpif *ofproto;
4243 union user_action_cookie cookie;
4244 struct flow flow;
4245
4246 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
4247 &flow, NULL, &ofproto, NULL, NULL)
4248 || !ofproto->ipfix) {
4249 return;
4250 }
4251
4252 memset(&cookie, 0, sizeof cookie);
4253 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.flow_sample);
4254
4255 /* The flow reflects exactly the contents of the packet. Sample
4256 * the packet using it. */
4257 dpif_ipfix_flow_sample(ofproto->ipfix, upcall->packet, &flow,
4258 cookie.flow_sample.collector_set_id,
4259 cookie.flow_sample.probability,
4260 cookie.flow_sample.obs_domain_id,
4261 cookie.flow_sample.obs_point_id);
4262 }
4263
4264 static void
4265 handle_ipfix_upcall(struct dpif_backer *backer,
4266 const struct dpif_upcall *upcall)
4267 {
4268 struct ofproto_dpif *ofproto;
4269 struct flow flow;
4270
4271 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
4272 &flow, NULL, &ofproto, NULL, NULL)
4273 || !ofproto->ipfix) {
4274 return;
4275 }
4276
4277 /* The flow reflects exactly the contents of the packet. Sample
4278 * the packet using it. */
4279 dpif_ipfix_bridge_sample(ofproto->ipfix, upcall->packet, &flow);
4280 }
4281
4282 static int
4283 handle_upcalls(struct dpif_backer *backer, unsigned int max_batch)
4284 {
4285 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
4286 struct ofpbuf miss_bufs[FLOW_MISS_MAX_BATCH];
4287 uint64_t miss_buf_stubs[FLOW_MISS_MAX_BATCH][4096 / 8];
4288 int n_processed;
4289 int n_misses;
4290 int i;
4291
4292 ovs_assert(max_batch <= FLOW_MISS_MAX_BATCH);
4293
4294 n_misses = 0;
4295 for (n_processed = 0; n_processed < max_batch; n_processed++) {
4296 struct dpif_upcall *upcall = &misses[n_misses];
4297 struct ofpbuf *buf = &miss_bufs[n_misses];
4298 int error;
4299
4300 ofpbuf_use_stub(buf, miss_buf_stubs[n_misses],
4301 sizeof miss_buf_stubs[n_misses]);
4302 error = dpif_recv(backer->dpif, upcall, buf);
4303 if (error) {
4304 ofpbuf_uninit(buf);
4305 break;
4306 }
4307
4308 switch (classify_upcall(upcall)) {
4309 case MISS_UPCALL:
4310 /* Handle it later. */
4311 n_misses++;
4312 break;
4313
4314 case SFLOW_UPCALL:
4315 handle_sflow_upcall(backer, upcall);
4316 ofpbuf_uninit(buf);
4317 break;
4318
4319 case FLOW_SAMPLE_UPCALL:
4320 handle_flow_sample_upcall(backer, upcall);
4321 ofpbuf_uninit(buf);
4322 break;
4323
4324 case IPFIX_UPCALL:
4325 handle_ipfix_upcall(backer, upcall);
4326 ofpbuf_uninit(buf);
4327 break;
4328
4329 case BAD_UPCALL:
4330 ofpbuf_uninit(buf);
4331 break;
4332 }
4333 }
4334
4335 /* Handle deferred MISS_UPCALL processing. */
4336 handle_miss_upcalls(backer, misses, n_misses);
4337 for (i = 0; i < n_misses; i++) {
4338 ofpbuf_uninit(&miss_bufs[i]);
4339 }
4340
4341 return n_processed;
4342 }
4343 \f
4344 /* Flow expiration. */
4345
4346 static int subfacet_max_idle(const struct dpif_backer *);
4347 static void update_stats(struct dpif_backer *);
4348 static void rule_expire(struct rule_dpif *);
4349 static void expire_subfacets(struct dpif_backer *, int dp_max_idle);
4350
4351 /* This function is called periodically by run(). Its job is to collect
4352 * updates for the flows that have been installed into the datapath, most
4353 * importantly when they last were used, and then use that information to
4354 * expire flows that have not been used recently.
4355 *
4356 * Returns the number of milliseconds after which it should be called again. */
4357 static int
4358 expire(struct dpif_backer *backer)
4359 {
4360 struct ofproto_dpif *ofproto;
4361 size_t n_subfacets;
4362 int max_idle;
4363
4364 /* Periodically clear out the drop keys in an effort to keep them
4365 * relatively few. */
4366 drop_key_clear(backer);
4367
4368 /* Update stats for each flow in the backer. */
4369 update_stats(backer);
4370
4371 n_subfacets = hmap_count(&backer->subfacets);
4372 if (n_subfacets) {
4373 struct subfacet *subfacet;
4374 long long int total, now;
4375
4376 total = 0;
4377 now = time_msec();
4378 HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) {
4379 total += now - subfacet->created;
4380 }
4381 backer->avg_subfacet_life += total / n_subfacets;
4382 }
4383 backer->avg_subfacet_life /= 2;
4384
4385 backer->avg_n_subfacet += n_subfacets;
4386 backer->avg_n_subfacet /= 2;
4387
4388 backer->max_n_subfacet = MAX(backer->max_n_subfacet, n_subfacets);
4389
4390 max_idle = subfacet_max_idle(backer);
4391 expire_subfacets(backer, max_idle);
4392
4393 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4394 struct rule *rule, *next_rule;
4395
4396 if (ofproto->backer != backer) {
4397 continue;
4398 }
4399
4400 /* Expire OpenFlow flows whose idle_timeout or hard_timeout
4401 * has passed. */
4402 LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
4403 &ofproto->up.expirable) {
4404 rule_expire(rule_dpif_cast(rule));
4405 }
4406
4407 /* All outstanding data in existing flows has been accounted, so it's a
4408 * good time to do bond rebalancing. */
4409 if (ofproto->has_bonded_bundles) {
4410 struct ofbundle *bundle;
4411
4412 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
4413 if (bundle->bond) {
4414 bond_rebalance(bundle->bond, &backer->revalidate_set);
4415 }
4416 }
4417 }
4418 }
4419
4420 return MIN(max_idle, 1000);
4421 }
4422
4423 /* Updates flow table statistics given that the datapath just reported 'stats'
4424 * as 'subfacet''s statistics. */
4425 static void
4426 update_subfacet_stats(struct subfacet *subfacet,
4427 const struct dpif_flow_stats *stats)
4428 {
4429 struct facet *facet = subfacet->facet;
4430 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4431 struct dpif_flow_stats diff;
4432
4433 diff.tcp_flags = stats->tcp_flags;
4434 diff.used = stats->used;
4435
4436 if (stats->n_packets >= subfacet->dp_packet_count) {
4437 diff.n_packets = stats->n_packets - subfacet->dp_packet_count;
4438 } else {
4439 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
4440 diff.n_packets = 0;
4441 }
4442
4443 if (stats->n_bytes >= subfacet->dp_byte_count) {
4444 diff.n_bytes = stats->n_bytes - subfacet->dp_byte_count;
4445 } else {
4446 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
4447 diff.n_bytes = 0;
4448 }
4449
4450 ofproto->n_hit += diff.n_packets;
4451 subfacet->dp_packet_count = stats->n_packets;
4452 subfacet->dp_byte_count = stats->n_bytes;
4453 subfacet_update_stats(subfacet, &diff);
4454
4455 if (facet->accounted_bytes < facet->byte_count) {
4456 facet_learn(facet);
4457 facet_account(facet);
4458 facet->accounted_bytes = facet->byte_count;
4459 }
4460 }
4461
4462 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
4463 * about, or a flow that shouldn't be installed but was anyway. Delete it. */
4464 static void
4465 delete_unexpected_flow(struct dpif_backer *backer,
4466 const struct nlattr *key, size_t key_len)
4467 {
4468 if (!VLOG_DROP_WARN(&rl)) {
4469 struct ds s;
4470
4471 ds_init(&s);
4472 odp_flow_key_format(key, key_len, &s);
4473 VLOG_WARN("unexpected flow: %s", ds_cstr(&s));
4474 ds_destroy(&s);
4475 }
4476
4477 COVERAGE_INC(facet_unexpected);
4478 dpif_flow_del(backer->dpif, key, key_len, NULL);
4479 }
4480
4481 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
4482 *
4483 * This function also pushes statistics updates to rules which each facet
4484 * resubmits into. Generally these statistics will be accurate. However, if a
4485 * facet changes the rule it resubmits into at some time in between
4486 * update_stats() runs, it is possible that statistics accrued to the
4487 * old rule will be incorrectly attributed to the new rule. This could be
4488 * avoided by calling update_stats() whenever rules are created or
4489 * deleted. However, the performance impact of making so many calls to the
4490 * datapath do not justify the benefit of having perfectly accurate statistics.
4491 *
4492 * In addition, this function maintains per ofproto flow hit counts. The patch
4493 * port is not treated specially. e.g. A packet ingress from br0 patched into
4494 * br1 will increase the hit count of br0 by 1, however, does not affect
4495 * the hit or miss counts of br1.
4496 */
4497 static void
4498 update_stats(struct dpif_backer *backer)
4499 {
4500 const struct dpif_flow_stats *stats;
4501 struct dpif_flow_dump dump;
4502 const struct nlattr *key;
4503 size_t key_len;
4504
4505 dpif_flow_dump_start(&dump, backer->dpif);
4506 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
4507 struct subfacet *subfacet;
4508 uint32_t key_hash;
4509
4510 key_hash = odp_flow_key_hash(key, key_len);
4511 subfacet = subfacet_find(backer, key, key_len, key_hash);
4512 switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
4513 case SF_FAST_PATH:
4514 update_subfacet_stats(subfacet, stats);
4515 break;
4516
4517 case SF_SLOW_PATH:
4518 /* Stats are updated per-packet. */
4519 break;
4520
4521 case SF_NOT_INSTALLED:
4522 default:
4523 delete_unexpected_flow(backer, key, key_len);
4524 break;
4525 }
4526 run_fast_rl();
4527 }
4528 dpif_flow_dump_done(&dump);
4529
4530 update_moving_averages(backer);
4531 }
4532
4533 /* Calculates and returns the number of milliseconds of idle time after which
4534 * subfacets should expire from the datapath. When a subfacet expires, we fold
4535 * its statistics into its facet, and when a facet's last subfacet expires, we
4536 * fold its statistic into its rule. */
4537 static int
4538 subfacet_max_idle(const struct dpif_backer *backer)
4539 {
4540 /*
4541 * Idle time histogram.
4542 *
4543 * Most of the time a switch has a relatively small number of subfacets.
4544 * When this is the case we might as well keep statistics for all of them
4545 * in userspace and to cache them in the kernel datapath for performance as
4546 * well.
4547 *
4548 * As the number of subfacets increases, the memory required to maintain
4549 * statistics about them in userspace and in the kernel becomes
4550 * significant. However, with a large number of subfacets it is likely
4551 * that only a few of them are "heavy hitters" that consume a large amount
4552 * of bandwidth. At this point, only heavy hitters are worth caching in
4553 * the kernel and maintaining in userspaces; other subfacets we can
4554 * discard.
4555 *
4556 * The technique used to compute the idle time is to build a histogram with
4557 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
4558 * that is installed in the kernel gets dropped in the appropriate bucket.
4559 * After the histogram has been built, we compute the cutoff so that only
4560 * the most-recently-used 1% of subfacets (but at least
4561 * flow_eviction_threshold flows) are kept cached. At least
4562 * the most-recently-used bucket of subfacets is kept, so actually an
4563 * arbitrary number of subfacets can be kept in any given expiration run
4564 * (though the next run will delete most of those unless they receive
4565 * additional data).
4566 *
4567 * This requires a second pass through the subfacets, in addition to the
4568 * pass made by update_stats(), because the former function never looks at
4569 * uninstallable subfacets.
4570 */
4571 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
4572 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
4573 int buckets[N_BUCKETS] = { 0 };
4574 int total, subtotal, bucket;
4575 struct subfacet *subfacet;
4576 long long int now;
4577 int i;
4578
4579 total = hmap_count(&backer->subfacets);
4580 if (total <= flow_eviction_threshold) {
4581 return N_BUCKETS * BUCKET_WIDTH;
4582 }
4583
4584 /* Build histogram. */
4585 now = time_msec();
4586 HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) {
4587 long long int idle = now - subfacet->used;
4588 int bucket = (idle <= 0 ? 0
4589 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
4590 : (unsigned int) idle / BUCKET_WIDTH);
4591 buckets[bucket]++;
4592 }
4593
4594 /* Find the first bucket whose flows should be expired. */
4595 subtotal = bucket = 0;
4596 do {
4597 subtotal += buckets[bucket++];
4598 } while (bucket < N_BUCKETS &&
4599 subtotal < MAX(flow_eviction_threshold, total / 100));
4600
4601 if (VLOG_IS_DBG_ENABLED()) {
4602 struct ds s;
4603
4604 ds_init(&s);
4605 ds_put_cstr(&s, "keep");
4606 for (i = 0; i < N_BUCKETS; i++) {
4607 if (i == bucket) {
4608 ds_put_cstr(&s, ", drop");
4609 }
4610 if (buckets[i]) {
4611 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
4612 }
4613 }
4614 VLOG_INFO("%s (msec:count)", ds_cstr(&s));
4615 ds_destroy(&s);
4616 }
4617
4618 return bucket * BUCKET_WIDTH;
4619 }
4620
4621 static void
4622 expire_subfacets(struct dpif_backer *backer, int dp_max_idle)
4623 {
4624 /* Cutoff time for most flows. */
4625 long long int normal_cutoff = time_msec() - dp_max_idle;
4626
4627 /* We really want to keep flows for special protocols around, so use a more
4628 * conservative cutoff. */
4629 long long int special_cutoff = time_msec() - 10000;
4630
4631 struct subfacet *subfacet, *next_subfacet;
4632 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
4633 int n_batch;
4634
4635 n_batch = 0;
4636 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
4637 &backer->subfacets) {
4638 long long int cutoff;
4639
4640 cutoff = (subfacet->facet->xout.slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP
4641 | SLOW_STP)
4642 ? special_cutoff
4643 : normal_cutoff);
4644 if (subfacet->used < cutoff) {
4645 if (subfacet->path != SF_NOT_INSTALLED) {
4646 batch[n_batch++] = subfacet;
4647 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
4648 subfacet_destroy_batch(backer, batch, n_batch);
4649 n_batch = 0;
4650 }
4651 } else {
4652 subfacet_destroy(subfacet);
4653 }
4654 }
4655 }
4656
4657 if (n_batch > 0) {
4658 subfacet_destroy_batch(backer, batch, n_batch);
4659 }
4660 }
4661
4662 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
4663 * then delete it entirely. */
4664 static void
4665 rule_expire(struct rule_dpif *rule)
4666 {
4667 struct facet *facet, *next_facet;
4668 long long int now;
4669 uint8_t reason;
4670
4671 if (rule->up.pending) {
4672 /* We'll have to expire it later. */
4673 return;
4674 }
4675
4676 /* Has 'rule' expired? */
4677 now = time_msec();
4678 if (rule->up.hard_timeout
4679 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
4680 reason = OFPRR_HARD_TIMEOUT;
4681 } else if (rule->up.idle_timeout
4682 && now > rule->up.used + rule->up.idle_timeout * 1000) {
4683 reason = OFPRR_IDLE_TIMEOUT;
4684 } else {
4685 return;
4686 }
4687
4688 COVERAGE_INC(ofproto_dpif_expired);
4689
4690 /* Update stats. (This is a no-op if the rule expired due to an idle
4691 * timeout, because that only happens when the rule has no facets left.) */
4692 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4693 facet_remove(facet);
4694 }
4695
4696 /* Get rid of the rule. */
4697 ofproto_rule_expire(&rule->up, reason);
4698 }
4699 \f
4700 /* Facets. */
4701
4702 /* Creates and returns a new facet based on 'miss'.
4703 *
4704 * The caller must already have determined that no facet with an identical
4705 * 'miss->flow' exists in 'miss->ofproto'.
4706 *
4707 * 'rule' and 'xout' must have been created based on 'miss'.
4708 *
4709 * 'facet'' statistics are initialized based on 'stats'.
4710 *
4711 * The facet will initially have no subfacets. The caller should create (at
4712 * least) one subfacet with subfacet_create(). */
4713 static struct facet *
4714 facet_create(const struct flow_miss *miss, struct rule_dpif *rule,
4715 struct xlate_out *xout, struct dpif_flow_stats *stats)
4716 {
4717 struct ofproto_dpif *ofproto = miss->ofproto;
4718 struct facet *facet;
4719 struct match match;
4720
4721 facet = xzalloc(sizeof *facet);
4722 facet->packet_count = facet->prev_packet_count = stats->n_packets;
4723 facet->byte_count = facet->prev_byte_count = stats->n_bytes;
4724 facet->tcp_flags = stats->tcp_flags;
4725 facet->used = stats->used;
4726 facet->flow = miss->flow;
4727 facet->initial_vals = miss->initial_vals;
4728 facet->learn_rl = time_msec() + 500;
4729 facet->rule = rule;
4730
4731 list_push_back(&facet->rule->facets, &facet->list_node);
4732 list_init(&facet->subfacets);
4733 netflow_flow_init(&facet->nf_flow);
4734 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
4735
4736 xlate_out_copy(&facet->xout, xout);
4737
4738 match_init(&match, &facet->flow, &facet->xout.wc);
4739 cls_rule_init(&facet->cr, &match, OFP_DEFAULT_PRIORITY);
4740 classifier_insert(&ofproto->facets, &facet->cr);
4741
4742 facet->nf_flow.output_iface = facet->xout.nf_output_iface;
4743
4744 return facet;
4745 }
4746
4747 static void
4748 facet_free(struct facet *facet)
4749 {
4750 if (facet) {
4751 xlate_out_uninit(&facet->xout);
4752 free(facet);
4753 }
4754 }
4755
4756 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
4757 * 'packet', which arrived on 'in_port'. */
4758 static bool
4759 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
4760 const struct nlattr *odp_actions, size_t actions_len,
4761 struct ofpbuf *packet)
4762 {
4763 struct odputil_keybuf keybuf;
4764 struct ofpbuf key;
4765 int error;
4766
4767 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4768 odp_flow_key_from_flow(&key, flow,
4769 ofp_port_to_odp_port(ofproto, flow->in_port));
4770
4771 error = dpif_execute(ofproto->backer->dpif, key.data, key.size,
4772 odp_actions, actions_len, packet);
4773 return !error;
4774 }
4775
4776 /* Remove 'facet' from its ofproto and free up the associated memory:
4777 *
4778 * - If 'facet' was installed in the datapath, uninstalls it and updates its
4779 * rule's statistics, via subfacet_uninstall().
4780 *
4781 * - Removes 'facet' from its rule and from ofproto->facets.
4782 */
4783 static void
4784 facet_remove(struct facet *facet)
4785 {
4786 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4787 struct subfacet *subfacet, *next_subfacet;
4788
4789 ovs_assert(!list_is_empty(&facet->subfacets));
4790
4791 /* First uninstall all of the subfacets to get final statistics. */
4792 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4793 subfacet_uninstall(subfacet);
4794 }
4795
4796 /* Flush the final stats to the rule.
4797 *
4798 * This might require us to have at least one subfacet around so that we
4799 * can use its actions for accounting in facet_account(), which is why we
4800 * have uninstalled but not yet destroyed the subfacets. */
4801 facet_flush_stats(facet);
4802
4803 /* Now we're really all done so destroy everything. */
4804 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
4805 &facet->subfacets) {
4806 subfacet_destroy__(subfacet);
4807 }
4808 classifier_remove(&ofproto->facets, &facet->cr);
4809 cls_rule_destroy(&facet->cr);
4810 list_remove(&facet->list_node);
4811 facet_free(facet);
4812 }
4813
4814 /* Feed information from 'facet' back into the learning table to keep it in
4815 * sync with what is actually flowing through the datapath. */
4816 static void
4817 facet_learn(struct facet *facet)
4818 {
4819 long long int now = time_msec();
4820
4821 if (!facet->xout.has_fin_timeout && now < facet->learn_rl) {
4822 return;
4823 }
4824
4825 facet->learn_rl = now + 500;
4826
4827 if (!facet->xout.has_learn
4828 && !facet->xout.has_normal
4829 && (!facet->xout.has_fin_timeout
4830 || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
4831 return;
4832 }
4833
4834 facet_push_stats(facet, true);
4835 }
4836
4837 static void
4838 facet_account(struct facet *facet)
4839 {
4840 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4841 const struct nlattr *a;
4842 unsigned int left;
4843 ovs_be16 vlan_tci;
4844 uint64_t n_bytes;
4845
4846 if (!facet->xout.has_normal || !ofproto->has_bonded_bundles) {
4847 return;
4848 }
4849 n_bytes = facet->byte_count - facet->accounted_bytes;
4850
4851 /* This loop feeds byte counters to bond_account() for rebalancing to use
4852 * as a basis. We also need to track the actual VLAN on which the packet
4853 * is going to be sent to ensure that it matches the one passed to
4854 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
4855 * hash bucket.)
4856 *
4857 * We use the actions from an arbitrary subfacet because they should all
4858 * be equally valid for our purpose. */
4859 vlan_tci = facet->flow.vlan_tci;
4860 NL_ATTR_FOR_EACH_UNSAFE (a, left, facet->xout.odp_actions.data,
4861 facet->xout.odp_actions.size) {
4862 const struct ovs_action_push_vlan *vlan;
4863 struct ofport_dpif *port;
4864
4865 switch (nl_attr_type(a)) {
4866 case OVS_ACTION_ATTR_OUTPUT:
4867 port = get_odp_port(ofproto, nl_attr_get_u32(a));
4868 if (port && port->bundle && port->bundle->bond) {
4869 bond_account(port->bundle->bond, &facet->flow,
4870 vlan_tci_to_vid(vlan_tci), n_bytes);
4871 }
4872 break;
4873
4874 case OVS_ACTION_ATTR_POP_VLAN:
4875 vlan_tci = htons(0);
4876 break;
4877
4878 case OVS_ACTION_ATTR_PUSH_VLAN:
4879 vlan = nl_attr_get(a);
4880 vlan_tci = vlan->vlan_tci;
4881 break;
4882 }
4883 }
4884 }
4885
4886 /* Returns true if the only action for 'facet' is to send to the controller.
4887 * (We don't report NetFlow expiration messages for such facets because they
4888 * are just part of the control logic for the network, not real traffic). */
4889 static bool
4890 facet_is_controller_flow(struct facet *facet)
4891 {
4892 if (facet) {
4893 const struct rule *rule = &facet->rule->up;
4894 const struct ofpact *ofpacts = rule->ofpacts;
4895 size_t ofpacts_len = rule->ofpacts_len;
4896
4897 if (ofpacts_len > 0 &&
4898 ofpacts->type == OFPACT_CONTROLLER &&
4899 ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len)) {
4900 return true;
4901 }
4902 }
4903 return false;
4904 }
4905
4906 /* Folds all of 'facet''s statistics into its rule. Also updates the
4907 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
4908 * 'facet''s statistics in the datapath should have been zeroed and folded into
4909 * its packet and byte counts before this function is called. */
4910 static void
4911 facet_flush_stats(struct facet *facet)
4912 {
4913 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4914 struct subfacet *subfacet;
4915
4916 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4917 ovs_assert(!subfacet->dp_byte_count);
4918 ovs_assert(!subfacet->dp_packet_count);
4919 }
4920
4921 facet_push_stats(facet, false);
4922 if (facet->accounted_bytes < facet->byte_count) {
4923 facet_account(facet);
4924 facet->accounted_bytes = facet->byte_count;
4925 }
4926
4927 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
4928 struct ofexpired expired;
4929 expired.flow = facet->flow;
4930 expired.packet_count = facet->packet_count;
4931 expired.byte_count = facet->byte_count;
4932 expired.used = facet->used;
4933 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4934 }
4935
4936 /* Reset counters to prevent double counting if 'facet' ever gets
4937 * reinstalled. */
4938 facet_reset_counters(facet);
4939
4940 netflow_flow_clear(&facet->nf_flow);
4941 facet->tcp_flags = 0;
4942 }
4943
4944 /* Searches 'ofproto''s table of facets for one which would be responsible for
4945 * 'flow'. Returns it if found, otherwise a null pointer.
4946 *
4947 * The returned facet might need revalidation; use facet_lookup_valid()
4948 * instead if that is important. */
4949 static struct facet *
4950 facet_find(struct ofproto_dpif *ofproto, const struct flow *flow)
4951 {
4952 struct cls_rule *cr = classifier_lookup(&ofproto->facets, flow, NULL);
4953 return cr ? CONTAINER_OF(cr, struct facet, cr) : NULL;
4954 }
4955
4956 /* Searches 'ofproto''s table of facets for one capable that covers
4957 * 'flow'. Returns it if found, otherwise a null pointer.
4958 *
4959 * The returned facet is guaranteed to be valid. */
4960 static struct facet *
4961 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow)
4962 {
4963 struct facet *facet;
4964
4965 facet = facet_find(ofproto, flow);
4966 if (facet
4967 && (ofproto->backer->need_revalidate
4968 || tag_set_intersects(&ofproto->backer->revalidate_set,
4969 facet->xout.tags))
4970 && !facet_revalidate(facet)) {
4971 return NULL;
4972 }
4973
4974 return facet;
4975 }
4976
4977 static bool
4978 facet_check_consistency(struct facet *facet)
4979 {
4980 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
4981
4982 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4983
4984 struct xlate_out xout;
4985 struct xlate_in xin;
4986
4987 struct rule_dpif *rule;
4988 bool ok;
4989
4990 /* Check the rule for consistency. */
4991 rule = rule_dpif_lookup(ofproto, &facet->flow, NULL);
4992 if (rule != facet->rule) {
4993 if (!VLOG_DROP_WARN(&rl)) {
4994 struct ds s = DS_EMPTY_INITIALIZER;
4995
4996 flow_format(&s, &facet->flow);
4997 ds_put_format(&s, ": facet associated with wrong rule (was "
4998 "table=%"PRIu8",", facet->rule->up.table_id);
4999 cls_rule_format(&facet->rule->up.cr, &s);
5000 ds_put_format(&s, ") (should have been table=%"PRIu8",",
5001 rule->up.table_id);
5002 cls_rule_format(&rule->up.cr, &s);
5003 ds_put_char(&s, ')');
5004
5005 VLOG_WARN("%s", ds_cstr(&s));
5006 ds_destroy(&s);
5007 }
5008 return false;
5009 }
5010
5011 /* Check the datapath actions for consistency. */
5012 xlate_in_init(&xin, ofproto, &facet->flow, &facet->initial_vals, rule,
5013 0, NULL);
5014 xlate_actions(&xin, &xout);
5015
5016 ok = ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)
5017 && facet->xout.slow == xout.slow;
5018 if (!ok && !VLOG_DROP_WARN(&rl)) {
5019 struct ds s = DS_EMPTY_INITIALIZER;
5020
5021 flow_format(&s, &facet->flow);
5022 ds_put_cstr(&s, ": inconsistency in facet");
5023
5024 if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
5025 ds_put_cstr(&s, " (actions were: ");
5026 format_odp_actions(&s, facet->xout.odp_actions.data,
5027 facet->xout.odp_actions.size);
5028 ds_put_cstr(&s, ") (correct actions: ");
5029 format_odp_actions(&s, xout.odp_actions.data,
5030 xout.odp_actions.size);
5031 ds_put_char(&s, ')');
5032 }
5033
5034 if (facet->xout.slow != xout.slow) {
5035 ds_put_format(&s, " slow path incorrect. should be %d", xout.slow);
5036 }
5037
5038 VLOG_WARN("%s", ds_cstr(&s));
5039 ds_destroy(&s);
5040 }
5041 xlate_out_uninit(&xout);
5042
5043 return ok;
5044 }
5045
5046 /* Re-searches the classifier for 'facet':
5047 *
5048 * - If the rule found is different from 'facet''s current rule, moves
5049 * 'facet' to the new rule and recompiles its actions.
5050 *
5051 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
5052 * where it is and recompiles its actions anyway.
5053 *
5054 * - If any of 'facet''s subfacets correspond to a new flow according to
5055 * ofproto_receive(), 'facet' is removed.
5056 *
5057 * Returns true if 'facet' is still valid. False if 'facet' was removed. */
5058 static bool
5059 facet_revalidate(struct facet *facet)
5060 {
5061 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5062 struct rule_dpif *new_rule;
5063 struct subfacet *subfacet;
5064 struct flow_wildcards wc;
5065 struct xlate_out xout;
5066 struct xlate_in xin;
5067
5068 COVERAGE_INC(facet_revalidate);
5069
5070 /* Check that child subfacets still correspond to this facet. Tunnel
5071 * configuration changes could cause a subfacet's OpenFlow in_port to
5072 * change. */
5073 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
5074 struct ofproto_dpif *recv_ofproto;
5075 struct flow recv_flow;
5076 int error;
5077
5078 error = ofproto_receive(ofproto->backer, NULL, subfacet->key,
5079 subfacet->key_len, &recv_flow, NULL,
5080 &recv_ofproto, NULL, NULL);
5081 if (error
5082 || recv_ofproto != ofproto
5083 || facet != facet_find(ofproto, &recv_flow)) {
5084 facet_remove(facet);
5085 return false;
5086 }
5087 }
5088
5089 flow_wildcards_init_catchall(&wc);
5090 new_rule = rule_dpif_lookup(ofproto, &facet->flow, &wc);
5091
5092 /* Calculate new datapath actions.
5093 *
5094 * We do not modify any 'facet' state yet, because we might need to, e.g.,
5095 * emit a NetFlow expiration and, if so, we need to have the old state
5096 * around to properly compose it. */
5097 xlate_in_init(&xin, ofproto, &facet->flow, &facet->initial_vals, new_rule,
5098 0, NULL);
5099 xlate_actions(&xin, &xout);
5100 flow_wildcards_or(&xout.wc, &xout.wc, &wc);
5101
5102 /* A facet's slow path reason should only change under dramatic
5103 * circumstances. Rather than try to update everything, it's simpler to
5104 * remove the facet and start over.
5105 *
5106 * More importantly, if a facet's wildcards change, it will be relatively
5107 * difficult to figure out if its subfacets still belong to it, and if not
5108 * which facet they may belong to. Again, to avoid the complexity, we
5109 * simply give up instead. */
5110 if (facet->xout.slow != xout.slow
5111 || memcmp(&facet->xout.wc, &xout.wc, sizeof xout.wc)) {
5112 facet_remove(facet);
5113 xlate_out_uninit(&xout);
5114 return false;
5115 }
5116
5117 if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
5118 LIST_FOR_EACH(subfacet, list_node, &facet->subfacets) {
5119 if (subfacet->path == SF_FAST_PATH) {
5120 struct dpif_flow_stats stats;
5121
5122 subfacet_install(subfacet, &xout.odp_actions, &stats);
5123 subfacet_update_stats(subfacet, &stats);
5124 }
5125 }
5126
5127 facet_flush_stats(facet);
5128
5129 ofpbuf_clear(&facet->xout.odp_actions);
5130 ofpbuf_put(&facet->xout.odp_actions, xout.odp_actions.data,
5131 xout.odp_actions.size);
5132 }
5133
5134 /* Update 'facet' now that we've taken care of all the old state. */
5135 facet->xout.tags = xout.tags;
5136 facet->xout.slow = xout.slow;
5137 facet->xout.has_learn = xout.has_learn;
5138 facet->xout.has_normal = xout.has_normal;
5139 facet->xout.has_fin_timeout = xout.has_fin_timeout;
5140 facet->xout.nf_output_iface = xout.nf_output_iface;
5141 facet->xout.mirrors = xout.mirrors;
5142 facet->nf_flow.output_iface = facet->xout.nf_output_iface;
5143
5144 if (facet->rule != new_rule) {
5145 COVERAGE_INC(facet_changed_rule);
5146 list_remove(&facet->list_node);
5147 list_push_back(&new_rule->facets, &facet->list_node);
5148 facet->rule = new_rule;
5149 facet->used = new_rule->up.created;
5150 facet->prev_used = facet->used;
5151 }
5152
5153 xlate_out_uninit(&xout);
5154 return true;
5155 }
5156
5157 static void
5158 facet_reset_counters(struct facet *facet)
5159 {
5160 facet->packet_count = 0;
5161 facet->byte_count = 0;
5162 facet->prev_packet_count = 0;
5163 facet->prev_byte_count = 0;
5164 facet->accounted_bytes = 0;
5165 }
5166
5167 static void
5168 facet_push_stats(struct facet *facet, bool may_learn)
5169 {
5170 struct dpif_flow_stats stats;
5171
5172 ovs_assert(facet->packet_count >= facet->prev_packet_count);
5173 ovs_assert(facet->byte_count >= facet->prev_byte_count);
5174 ovs_assert(facet->used >= facet->prev_used);
5175
5176 stats.n_packets = facet->packet_count - facet->prev_packet_count;
5177 stats.n_bytes = facet->byte_count - facet->prev_byte_count;
5178 stats.used = facet->used;
5179 stats.tcp_flags = facet->tcp_flags;
5180
5181 if (may_learn || stats.n_packets || facet->used > facet->prev_used) {
5182 struct ofproto_dpif *ofproto =
5183 ofproto_dpif_cast(facet->rule->up.ofproto);
5184
5185 struct ofport_dpif *in_port;
5186 struct xlate_in xin;
5187
5188 facet->prev_packet_count = facet->packet_count;
5189 facet->prev_byte_count = facet->byte_count;
5190 facet->prev_used = facet->used;
5191
5192 in_port = get_ofp_port(ofproto, facet->flow.in_port);
5193 if (in_port && in_port->tnl_port) {
5194 netdev_vport_inc_rx(in_port->up.netdev, &stats);
5195 }
5196
5197 rule_credit_stats(facet->rule, &stats);
5198 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow,
5199 facet->used);
5200 netflow_flow_update_flags(&facet->nf_flow, facet->tcp_flags);
5201 update_mirror_stats(ofproto, facet->xout.mirrors, stats.n_packets,
5202 stats.n_bytes);
5203
5204 xlate_in_init(&xin, ofproto, &facet->flow, &facet->initial_vals,
5205 facet->rule, stats.tcp_flags, NULL);
5206 xin.resubmit_stats = &stats;
5207 xin.may_learn = may_learn;
5208 xlate_actions_for_side_effects(&xin);
5209 }
5210 }
5211
5212 static void
5213 push_all_stats__(bool run_fast)
5214 {
5215 static long long int rl = LLONG_MIN;
5216 struct ofproto_dpif *ofproto;
5217
5218 if (time_msec() < rl) {
5219 return;
5220 }
5221
5222 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5223 struct cls_cursor cursor;
5224 struct facet *facet;
5225
5226 cls_cursor_init(&cursor, &ofproto->facets, NULL);
5227 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5228 facet_push_stats(facet, false);
5229 if (run_fast) {
5230 run_fast_rl();
5231 }
5232 }
5233 }
5234
5235 rl = time_msec() + 100;
5236 }
5237
5238 static void
5239 push_all_stats(void)
5240 {
5241 push_all_stats__(true);
5242 }
5243
5244 static void
5245 rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats)
5246 {
5247 rule->packet_count += stats->n_packets;
5248 rule->byte_count += stats->n_bytes;
5249 ofproto_rule_update_used(&rule->up, stats->used);
5250 }
5251 \f
5252 /* Subfacets. */
5253
5254 static struct subfacet *
5255 subfacet_find(struct dpif_backer *backer, const struct nlattr *key,
5256 size_t key_len, uint32_t key_hash)
5257 {
5258 struct subfacet *subfacet;
5259
5260 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
5261 &backer->subfacets) {
5262 if (subfacet->key_len == key_len
5263 && !memcmp(key, subfacet->key, key_len)) {
5264 return subfacet;
5265 }
5266 }
5267
5268 return NULL;
5269 }
5270
5271 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
5272 * 'key_fitness', 'key', and 'key_len' members in 'miss'. Returns the
5273 * existing subfacet if there is one, otherwise creates and returns a
5274 * new subfacet. */
5275 static struct subfacet *
5276 subfacet_create(struct facet *facet, struct flow_miss *miss,
5277 long long int now)
5278 {
5279 struct dpif_backer *backer = miss->ofproto->backer;
5280 enum odp_key_fitness key_fitness = miss->key_fitness;
5281 const struct nlattr *key = miss->key;
5282 size_t key_len = miss->key_len;
5283 uint32_t key_hash;
5284 struct subfacet *subfacet;
5285
5286 key_hash = odp_flow_key_hash(key, key_len);
5287
5288 if (list_is_empty(&facet->subfacets)) {
5289 subfacet = &facet->one_subfacet;
5290 } else {
5291 subfacet = subfacet_find(backer, key, key_len, key_hash);
5292 if (subfacet) {
5293 if (subfacet->facet == facet) {
5294 return subfacet;
5295 }
5296
5297 /* This shouldn't happen. */
5298 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
5299 subfacet_destroy(subfacet);
5300 }
5301
5302 subfacet = xmalloc(sizeof *subfacet);
5303 }
5304
5305 hmap_insert(&backer->subfacets, &subfacet->hmap_node, key_hash);
5306 list_push_back(&facet->subfacets, &subfacet->list_node);
5307 subfacet->facet = facet;
5308 subfacet->key_fitness = key_fitness;
5309 subfacet->key = xmemdup(key, key_len);
5310 subfacet->key_len = key_len;
5311 subfacet->used = now;
5312 subfacet->created = now;
5313 subfacet->dp_packet_count = 0;
5314 subfacet->dp_byte_count = 0;
5315 subfacet->path = SF_NOT_INSTALLED;
5316 subfacet->backer = backer;
5317
5318 backer->subfacet_add_count++;
5319 return subfacet;
5320 }
5321
5322 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
5323 * its facet within 'ofproto', and frees it. */
5324 static void
5325 subfacet_destroy__(struct subfacet *subfacet)
5326 {
5327 struct facet *facet = subfacet->facet;
5328 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5329
5330 /* Update ofproto stats before uninstall the subfacet. */
5331 ofproto->backer->subfacet_del_count++;
5332
5333 subfacet_uninstall(subfacet);
5334 hmap_remove(&subfacet->backer->subfacets, &subfacet->hmap_node);
5335 list_remove(&subfacet->list_node);
5336 free(subfacet->key);
5337 if (subfacet != &facet->one_subfacet) {
5338 free(subfacet);
5339 }
5340 }
5341
5342 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
5343 * last remaining subfacet in its facet destroys the facet too. */
5344 static void
5345 subfacet_destroy(struct subfacet *subfacet)
5346 {
5347 struct facet *facet = subfacet->facet;
5348
5349 if (list_is_singleton(&facet->subfacets)) {
5350 /* facet_remove() needs at least one subfacet (it will remove it). */
5351 facet_remove(facet);
5352 } else {
5353 subfacet_destroy__(subfacet);
5354 }
5355 }
5356
5357 static void
5358 subfacet_destroy_batch(struct dpif_backer *backer,
5359 struct subfacet **subfacets, int n)
5360 {
5361 struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH];
5362 struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH];
5363 struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH];
5364 int i;
5365
5366 for (i = 0; i < n; i++) {
5367 ops[i].type = DPIF_OP_FLOW_DEL;
5368 ops[i].u.flow_del.key = subfacets[i]->key;
5369 ops[i].u.flow_del.key_len = subfacets[i]->key_len;
5370 ops[i].u.flow_del.stats = &stats[i];
5371 opsp[i] = &ops[i];
5372 }
5373
5374 dpif_operate(backer->dpif, opsp, n);
5375 for (i = 0; i < n; i++) {
5376 subfacet_reset_dp_stats(subfacets[i], &stats[i]);
5377 subfacets[i]->path = SF_NOT_INSTALLED;
5378 subfacet_destroy(subfacets[i]);
5379 run_fast_rl();
5380 }
5381 }
5382
5383 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
5384 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
5385 * in the datapath will be zeroed and 'stats' will be updated with traffic new
5386 * since 'subfacet' was last updated.
5387 *
5388 * Returns 0 if successful, otherwise a positive errno value. */
5389 static int
5390 subfacet_install(struct subfacet *subfacet, const struct ofpbuf *odp_actions,
5391 struct dpif_flow_stats *stats)
5392 {
5393 struct facet *facet = subfacet->facet;
5394 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5395 enum subfacet_path path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
5396 const struct nlattr *actions = odp_actions->data;
5397 size_t actions_len = odp_actions->size;
5398
5399 uint64_t slow_path_stub[128 / 8];
5400 enum dpif_flow_put_flags flags;
5401 int ret;
5402
5403 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
5404 if (stats) {
5405 flags |= DPIF_FP_ZERO_STATS;
5406 }
5407
5408 if (path == SF_SLOW_PATH) {
5409 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
5410 slow_path_stub, sizeof slow_path_stub,
5411 &actions, &actions_len);
5412 }
5413
5414 ret = dpif_flow_put(subfacet->backer->dpif, flags, subfacet->key,
5415 subfacet->key_len, actions, actions_len, stats);
5416
5417 if (stats) {
5418 subfacet_reset_dp_stats(subfacet, stats);
5419 }
5420
5421 if (!ret) {
5422 subfacet->path = path;
5423 }
5424 return ret;
5425 }
5426
5427 /* If 'subfacet' is installed in the datapath, uninstalls it. */
5428 static void
5429 subfacet_uninstall(struct subfacet *subfacet)
5430 {
5431 if (subfacet->path != SF_NOT_INSTALLED) {
5432 struct rule_dpif *rule = subfacet->facet->rule;
5433 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5434 struct dpif_flow_stats stats;
5435 int error;
5436
5437 error = dpif_flow_del(ofproto->backer->dpif, subfacet->key,
5438 subfacet->key_len, &stats);
5439 subfacet_reset_dp_stats(subfacet, &stats);
5440 if (!error) {
5441 subfacet_update_stats(subfacet, &stats);
5442 }
5443 subfacet->path = SF_NOT_INSTALLED;
5444 } else {
5445 ovs_assert(subfacet->dp_packet_count == 0);
5446 ovs_assert(subfacet->dp_byte_count == 0);
5447 }
5448 }
5449
5450 /* Resets 'subfacet''s datapath statistics counters. This should be called
5451 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
5452 * non-null, it should contain the statistics returned by dpif when 'subfacet'
5453 * was reset in the datapath. 'stats' will be modified to include only
5454 * statistics new since 'subfacet' was last updated. */
5455 static void
5456 subfacet_reset_dp_stats(struct subfacet *subfacet,
5457 struct dpif_flow_stats *stats)
5458 {
5459 if (stats
5460 && subfacet->dp_packet_count <= stats->n_packets
5461 && subfacet->dp_byte_count <= stats->n_bytes) {
5462 stats->n_packets -= subfacet->dp_packet_count;
5463 stats->n_bytes -= subfacet->dp_byte_count;
5464 }
5465
5466 subfacet->dp_packet_count = 0;
5467 subfacet->dp_byte_count = 0;
5468 }
5469
5470 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
5471 *
5472 * Because of the meaning of a subfacet's counters, it only makes sense to do
5473 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
5474 * represents a packet that was sent by hand or if it represents statistics
5475 * that have been cleared out of the datapath. */
5476 static void
5477 subfacet_update_stats(struct subfacet *subfacet,
5478 const struct dpif_flow_stats *stats)
5479 {
5480 if (stats->n_packets || stats->used > subfacet->used) {
5481 struct facet *facet = subfacet->facet;
5482
5483 subfacet->used = MAX(subfacet->used, stats->used);
5484 facet->used = MAX(facet->used, stats->used);
5485 facet->packet_count += stats->n_packets;
5486 facet->byte_count += stats->n_bytes;
5487 facet->tcp_flags |= stats->tcp_flags;
5488 }
5489 }
5490 \f
5491 /* Rules. */
5492
5493 /* Lookup 'flow' in 'ofproto''s classifier. If 'wc' is non-null, sets
5494 * the fields that were relevant as part of the lookup. */
5495 static struct rule_dpif *
5496 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow,
5497 struct flow_wildcards *wc)
5498 {
5499 struct rule_dpif *rule;
5500
5501 rule = rule_dpif_lookup__(ofproto, flow, wc, 0);
5502 if (rule) {
5503 return rule;
5504 }
5505
5506 return rule_dpif_miss_rule(ofproto, flow);
5507 }
5508
5509 static struct rule_dpif *
5510 rule_dpif_lookup__(struct ofproto_dpif *ofproto, const struct flow *flow,
5511 struct flow_wildcards *wc, uint8_t table_id)
5512 {
5513 struct cls_rule *cls_rule;
5514 struct classifier *cls;
5515 bool frag;
5516
5517 if (table_id >= N_TABLES) {
5518 return NULL;
5519 }
5520
5521 cls = &ofproto->up.tables[table_id].cls;
5522 frag = (flow->nw_frag & FLOW_NW_FRAG_ANY) != 0;
5523 if (frag && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
5524 /* We must pretend that transport ports are unavailable. */
5525 struct flow ofpc_normal_flow = *flow;
5526 ofpc_normal_flow.tp_src = htons(0);
5527 ofpc_normal_flow.tp_dst = htons(0);
5528 cls_rule = classifier_lookup(cls, &ofpc_normal_flow, wc);
5529 } else if (frag && ofproto->up.frag_handling == OFPC_FRAG_DROP) {
5530 cls_rule = &ofproto->drop_frags_rule->up.cr;
5531 if (wc) {
5532 flow_wildcards_init_exact(wc);
5533 }
5534 } else {
5535 cls_rule = classifier_lookup(cls, flow, wc);
5536 }
5537 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
5538 }
5539
5540 static struct rule_dpif *
5541 rule_dpif_miss_rule(struct ofproto_dpif *ofproto, const struct flow *flow)
5542 {
5543 struct ofport_dpif *port;
5544
5545 port = get_ofp_port(ofproto, flow->in_port);
5546 if (!port) {
5547 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, flow->in_port);
5548 return ofproto->miss_rule;
5549 }
5550
5551 if (port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN) {
5552 return ofproto->no_packet_in_rule;
5553 }
5554 return ofproto->miss_rule;
5555 }
5556
5557 static void
5558 complete_operation(struct rule_dpif *rule)
5559 {
5560 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5561
5562 rule_invalidate(rule);
5563 if (clogged) {
5564 struct dpif_completion *c = xmalloc(sizeof *c);
5565 c->op = rule->up.pending;
5566 list_push_back(&ofproto->completions, &c->list_node);
5567 } else {
5568 ofoperation_complete(rule->up.pending, 0);
5569 }
5570 }
5571
5572 static struct rule *
5573 rule_alloc(void)
5574 {
5575 struct rule_dpif *rule = xmalloc(sizeof *rule);
5576 return &rule->up;
5577 }
5578
5579 static void
5580 rule_dealloc(struct rule *rule_)
5581 {
5582 struct rule_dpif *rule = rule_dpif_cast(rule_);
5583 free(rule);
5584 }
5585
5586 static enum ofperr
5587 rule_construct(struct rule *rule_)
5588 {
5589 struct rule_dpif *rule = rule_dpif_cast(rule_);
5590 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5591 struct rule_dpif *victim;
5592 uint8_t table_id;
5593
5594 rule->packet_count = 0;
5595 rule->byte_count = 0;
5596
5597 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
5598 if (victim && !list_is_empty(&victim->facets)) {
5599 struct facet *facet;
5600
5601 rule->facets = victim->facets;
5602 list_moved(&rule->facets);
5603 LIST_FOR_EACH (facet, list_node, &rule->facets) {
5604 /* XXX: We're only clearing our local counters here. It's possible
5605 * that quite a few packets are unaccounted for in the datapath
5606 * statistics. These will be accounted to the new rule instead of
5607 * cleared as required. This could be fixed by clearing out the
5608 * datapath statistics for this facet, but currently it doesn't
5609 * seem worth it. */
5610 facet_reset_counters(facet);
5611 facet->rule = rule;
5612 }
5613 } else {
5614 /* Must avoid list_moved() in this case. */
5615 list_init(&rule->facets);
5616 }
5617
5618 table_id = rule->up.table_id;
5619 if (victim) {
5620 rule->tag = victim->tag;
5621 } else if (table_id == 0) {
5622 rule->tag = 0;
5623 } else {
5624 struct flow flow;
5625
5626 miniflow_expand(&rule->up.cr.match.flow, &flow);
5627 rule->tag = rule_calculate_tag(&flow, &rule->up.cr.match.mask,
5628 ofproto->tables[table_id].basis);
5629 }
5630
5631 complete_operation(rule);
5632 return 0;
5633 }
5634
5635 static void
5636 rule_destruct(struct rule *rule_)
5637 {
5638 struct rule_dpif *rule = rule_dpif_cast(rule_);
5639 struct facet *facet, *next_facet;
5640
5641 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
5642 facet_revalidate(facet);
5643 }
5644
5645 complete_operation(rule);
5646 }
5647
5648 static void
5649 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
5650 {
5651 struct rule_dpif *rule = rule_dpif_cast(rule_);
5652
5653 /* push_all_stats() can handle flow misses which, when using the learn
5654 * action, can cause rules to be added and deleted. This can corrupt our
5655 * caller's datastructures which assume that rule_get_stats() doesn't have
5656 * an impact on the flow table. To be safe, we disable miss handling. */
5657 push_all_stats__(false);
5658
5659 /* Start from historical data for 'rule' itself that are no longer tracked
5660 * in facets. This counts, for example, facets that have expired. */
5661 *packets = rule->packet_count;
5662 *bytes = rule->byte_count;
5663 }
5664
5665 static void
5666 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
5667 struct ofpbuf *packet)
5668 {
5669 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5670 struct initial_vals initial_vals;
5671 struct dpif_flow_stats stats;
5672 struct xlate_out xout;
5673 struct xlate_in xin;
5674
5675 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
5676 rule_credit_stats(rule, &stats);
5677
5678 initial_vals.vlan_tci = flow->vlan_tci;
5679 xlate_in_init(&xin, ofproto, flow, &initial_vals, rule, stats.tcp_flags,
5680 packet);
5681 xin.resubmit_stats = &stats;
5682 xlate_actions(&xin, &xout);
5683
5684 execute_odp_actions(ofproto, flow, xout.odp_actions.data,
5685 xout.odp_actions.size, packet);
5686
5687 xlate_out_uninit(&xout);
5688 }
5689
5690 static enum ofperr
5691 rule_execute(struct rule *rule, const struct flow *flow,
5692 struct ofpbuf *packet)
5693 {
5694 rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
5695 ofpbuf_delete(packet);
5696 return 0;
5697 }
5698
5699 static void
5700 rule_modify_actions(struct rule *rule_)
5701 {
5702 struct rule_dpif *rule = rule_dpif_cast(rule_);
5703
5704 complete_operation(rule);
5705 }
5706 \f
5707 /* Sends 'packet' out 'ofport'.
5708 * May modify 'packet'.
5709 * Returns 0 if successful, otherwise a positive errno value. */
5710 static int
5711 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
5712 {
5713 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
5714 uint64_t odp_actions_stub[1024 / 8];
5715 struct ofpbuf key, odp_actions;
5716 struct dpif_flow_stats stats;
5717 struct odputil_keybuf keybuf;
5718 struct ofpact_output output;
5719 struct xlate_out xout;
5720 struct xlate_in xin;
5721 struct flow flow;
5722 int error;
5723
5724 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5725 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5726
5727 /* Use OFPP_NONE as the in_port to avoid special packet processing. */
5728 flow_extract(packet, 0, 0, NULL, OFPP_NONE, &flow);
5729 odp_flow_key_from_flow(&key, &flow, ofp_port_to_odp_port(ofproto,
5730 OFPP_LOCAL));
5731 dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
5732
5733 ofpact_init(&output.ofpact, OFPACT_OUTPUT, sizeof output);
5734 output.port = ofport->up.ofp_port;
5735 output.max_len = 0;
5736
5737 xlate_in_init(&xin, ofproto, &flow, NULL, NULL, 0, packet);
5738 xin.ofpacts_len = sizeof output;
5739 xin.ofpacts = &output.ofpact;
5740 xin.resubmit_stats = &stats;
5741 xlate_actions(&xin, &xout);
5742
5743 error = dpif_execute(ofproto->backer->dpif,
5744 key.data, key.size,
5745 xout.odp_actions.data, xout.odp_actions.size,
5746 packet);
5747 xlate_out_uninit(&xout);
5748
5749 if (error) {
5750 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %s (%s)",
5751 ofproto->up.name, netdev_get_name(ofport->up.netdev),
5752 strerror(error));
5753 }
5754
5755 ofproto->stats.tx_packets++;
5756 ofproto->stats.tx_bytes += packet->size;
5757 return error;
5758 }
5759 \f
5760 /* OpenFlow to datapath action translation. */
5761
5762 static bool may_receive(const struct ofport_dpif *, struct xlate_ctx *);
5763 static void do_xlate_actions(const struct ofpact *, size_t ofpacts_len,
5764 struct xlate_ctx *);
5765 static void xlate_normal(struct xlate_ctx *);
5766
5767 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
5768 * The action will state 'slow' as the reason that the action is in the slow
5769 * path. (This is purely informational: it allows a human viewing "ovs-dpctl
5770 * dump-flows" output to see why a flow is in the slow path.)
5771 *
5772 * The 'stub_size' bytes in 'stub' will be used to store the action.
5773 * 'stub_size' must be large enough for the action.
5774 *
5775 * The action and its size will be stored in '*actionsp' and '*actions_lenp',
5776 * respectively. */
5777 static void
5778 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
5779 enum slow_path_reason slow,
5780 uint64_t *stub, size_t stub_size,
5781 const struct nlattr **actionsp, size_t *actions_lenp)
5782 {
5783 union user_action_cookie cookie;
5784 struct ofpbuf buf;
5785
5786 cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
5787 cookie.slow_path.unused = 0;
5788 cookie.slow_path.reason = slow;
5789
5790 ofpbuf_use_stack(&buf, stub, stub_size);
5791 if (slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP | SLOW_STP)) {
5792 uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif, UINT32_MAX);
5793 odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
5794 } else {
5795 put_userspace_action(ofproto, &buf, flow, &cookie,
5796 sizeof cookie.slow_path);
5797 }
5798 *actionsp = buf.data;
5799 *actions_lenp = buf.size;
5800 }
5801
5802 static size_t
5803 put_userspace_action(const struct ofproto_dpif *ofproto,
5804 struct ofpbuf *odp_actions,
5805 const struct flow *flow,
5806 const union user_action_cookie *cookie,
5807 const size_t cookie_size)
5808 {
5809 uint32_t pid;
5810
5811 pid = dpif_port_get_pid(ofproto->backer->dpif,
5812 ofp_port_to_odp_port(ofproto, flow->in_port));
5813
5814 return odp_put_userspace_action(pid, cookie, cookie_size, odp_actions);
5815 }
5816
5817 /* Compose SAMPLE action for sFlow or IPFIX. The given probability is
5818 * the number of packets out of UINT32_MAX to sample. The given
5819 * cookie is passed back in the callback for each sampled packet.
5820 */
5821 static size_t
5822 compose_sample_action(const struct ofproto_dpif *ofproto,
5823 struct ofpbuf *odp_actions,
5824 const struct flow *flow,
5825 const uint32_t probability,
5826 const union user_action_cookie *cookie,
5827 const size_t cookie_size)
5828 {
5829 size_t sample_offset, actions_offset;
5830 int cookie_offset;
5831
5832 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
5833
5834 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
5835
5836 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
5837 cookie_offset = put_userspace_action(ofproto, odp_actions, flow, cookie,
5838 cookie_size);
5839
5840 nl_msg_end_nested(odp_actions, actions_offset);
5841 nl_msg_end_nested(odp_actions, sample_offset);
5842 return cookie_offset;
5843 }
5844
5845 static void
5846 compose_sflow_cookie(const struct ofproto_dpif *ofproto,
5847 ovs_be16 vlan_tci, uint32_t odp_port,
5848 unsigned int n_outputs, union user_action_cookie *cookie)
5849 {
5850 int ifindex;
5851
5852 cookie->type = USER_ACTION_COOKIE_SFLOW;
5853 cookie->sflow.vlan_tci = vlan_tci;
5854
5855 /* See http://www.sflow.org/sflow_version_5.txt (search for "Input/output
5856 * port information") for the interpretation of cookie->output. */
5857 switch (n_outputs) {
5858 case 0:
5859 /* 0x40000000 | 256 means "packet dropped for unknown reason". */
5860 cookie->sflow.output = 0x40000000 | 256;
5861 break;
5862
5863 case 1:
5864 ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port);
5865 if (ifindex) {
5866 cookie->sflow.output = ifindex;
5867 break;
5868 }
5869 /* Fall through. */
5870 default:
5871 /* 0x80000000 means "multiple output ports. */
5872 cookie->sflow.output = 0x80000000 | n_outputs;
5873 break;
5874 }
5875 }
5876
5877 /* Compose SAMPLE action for sFlow bridge sampling. */
5878 static size_t
5879 compose_sflow_action(const struct ofproto_dpif *ofproto,
5880 struct ofpbuf *odp_actions,
5881 const struct flow *flow,
5882 uint32_t odp_port)
5883 {
5884 uint32_t probability;
5885 union user_action_cookie cookie;
5886
5887 if (!ofproto->sflow || flow->in_port == OFPP_NONE) {
5888 return 0;
5889 }
5890
5891 probability = dpif_sflow_get_probability(ofproto->sflow);
5892 compose_sflow_cookie(ofproto, htons(0), odp_port,
5893 odp_port == OVSP_NONE ? 0 : 1, &cookie);
5894
5895 return compose_sample_action(ofproto, odp_actions, flow, probability,
5896 &cookie, sizeof cookie.sflow);
5897 }
5898
5899 static void
5900 compose_flow_sample_cookie(uint16_t probability, uint32_t collector_set_id,
5901 uint32_t obs_domain_id, uint32_t obs_point_id,
5902 union user_action_cookie *cookie)
5903 {
5904 cookie->type = USER_ACTION_COOKIE_FLOW_SAMPLE;
5905 cookie->flow_sample.probability = probability;
5906 cookie->flow_sample.collector_set_id = collector_set_id;
5907 cookie->flow_sample.obs_domain_id = obs_domain_id;
5908 cookie->flow_sample.obs_point_id = obs_point_id;
5909 }
5910
5911 static void
5912 compose_ipfix_cookie(union user_action_cookie *cookie)
5913 {
5914 cookie->type = USER_ACTION_COOKIE_IPFIX;
5915 }
5916
5917 /* Compose SAMPLE action for IPFIX bridge sampling. */
5918 static void
5919 compose_ipfix_action(const struct ofproto_dpif *ofproto,
5920 struct ofpbuf *odp_actions,
5921 const struct flow *flow)
5922 {
5923 uint32_t probability;
5924 union user_action_cookie cookie;
5925
5926 if (!ofproto->ipfix || flow->in_port == OFPP_NONE) {
5927 return;
5928 }
5929
5930 probability = dpif_ipfix_get_bridge_exporter_probability(ofproto->ipfix);
5931 compose_ipfix_cookie(&cookie);
5932
5933 compose_sample_action(ofproto, odp_actions, flow, probability,
5934 &cookie, sizeof cookie.ipfix);
5935 }
5936
5937 /* SAMPLE action for sFlow must be first action in any given list of
5938 * actions. At this point we do not have all information required to
5939 * build it. So try to build sample action as complete as possible. */
5940 static void
5941 add_sflow_action(struct xlate_ctx *ctx)
5942 {
5943 ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto,
5944 &ctx->xout->odp_actions,
5945 &ctx->xin->flow, OVSP_NONE);
5946 ctx->sflow_odp_port = 0;
5947 ctx->sflow_n_outputs = 0;
5948 }
5949
5950 /* SAMPLE action for IPFIX must be 1st or 2nd action in any given list
5951 * of actions, eventually after the SAMPLE action for sFlow. */
5952 static void
5953 add_ipfix_action(struct xlate_ctx *ctx)
5954 {
5955 compose_ipfix_action(ctx->ofproto, &ctx->xout->odp_actions,
5956 &ctx->xin->flow);
5957 }
5958
5959 /* Fix SAMPLE action according to data collected while composing ODP actions.
5960 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
5961 * USERSPACE action's user-cookie which is required for sflow. */
5962 static void
5963 fix_sflow_action(struct xlate_ctx *ctx)
5964 {
5965 const struct flow *base = &ctx->base_flow;
5966 union user_action_cookie *cookie;
5967
5968 if (!ctx->user_cookie_offset) {
5969 return;
5970 }
5971
5972 cookie = ofpbuf_at(&ctx->xout->odp_actions, ctx->user_cookie_offset,
5973 sizeof cookie->sflow);
5974 ovs_assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
5975
5976 compose_sflow_cookie(ctx->ofproto, base->vlan_tci,
5977 ctx->sflow_odp_port, ctx->sflow_n_outputs, cookie);
5978 }
5979
5980 static void
5981 compose_output_action__(struct xlate_ctx *ctx, uint16_t ofp_port,
5982 bool check_stp)
5983 {
5984 const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
5985 ovs_be16 flow_vlan_tci;
5986 uint32_t flow_skb_mark;
5987 uint8_t flow_nw_tos;
5988 struct priority_to_dscp *pdscp;
5989 uint32_t out_port, odp_port;
5990
5991 /* If 'struct flow' gets additional metadata, we'll need to zero it out
5992 * before traversing a patch port. */
5993 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 20);
5994
5995 if (!ofport) {
5996 xlate_report(ctx, "Nonexistent output port");
5997 return;
5998 } else if (ofport->up.pp.config & OFPUTIL_PC_NO_FWD) {
5999 xlate_report(ctx, "OFPPC_NO_FWD set, skipping output");
6000 return;
6001 } else if (check_stp && !stp_forward_in_state(ofport->stp_state)) {
6002 xlate_report(ctx, "STP not in forwarding state, skipping output");
6003 return;
6004 }
6005
6006 if (netdev_vport_is_patch(ofport->up.netdev)) {
6007 struct ofport_dpif *peer = ofport_get_peer(ofport);
6008 struct flow old_flow = ctx->xin->flow;
6009 const struct ofproto_dpif *peer_ofproto;
6010 enum slow_path_reason special;
6011 struct ofport_dpif *in_port;
6012
6013 if (!peer) {
6014 xlate_report(ctx, "Nonexistent patch port peer");
6015 return;
6016 }
6017
6018 peer_ofproto = ofproto_dpif_cast(peer->up.ofproto);
6019 if (peer_ofproto->backer != ctx->ofproto->backer) {
6020 xlate_report(ctx, "Patch port peer on a different datapath");
6021 return;
6022 }
6023
6024 ctx->ofproto = ofproto_dpif_cast(peer->up.ofproto);
6025 ctx->xin->flow.in_port = peer->up.ofp_port;
6026 ctx->xin->flow.metadata = htonll(0);
6027 memset(&ctx->xin->flow.tunnel, 0, sizeof ctx->xin->flow.tunnel);
6028 memset(ctx->xin->flow.regs, 0, sizeof ctx->xin->flow.regs);
6029
6030 in_port = get_ofp_port(ctx->ofproto, ctx->xin->flow.in_port);
6031 special = process_special(ctx->ofproto, &ctx->xin->flow, in_port,
6032 ctx->xin->packet);
6033 if (special) {
6034 ctx->xout->slow = special;
6035 } else if (!in_port || may_receive(in_port, ctx)) {
6036 if (!in_port || stp_forward_in_state(in_port->stp_state)) {
6037 xlate_table_action(ctx, ctx->xin->flow.in_port, 0, true);
6038 } else {
6039 /* Forwarding is disabled by STP. Let OFPP_NORMAL and the
6040 * learning action look at the packet, then drop it. */
6041 struct flow old_base_flow = ctx->base_flow;
6042 size_t old_size = ctx->xout->odp_actions.size;
6043 xlate_table_action(ctx, ctx->xin->flow.in_port, 0, true);
6044 ctx->base_flow = old_base_flow;
6045 ctx->xout->odp_actions.size = old_size;
6046 }
6047 }
6048
6049 ctx->xin->flow = old_flow;
6050 ctx->ofproto = ofproto_dpif_cast(ofport->up.ofproto);
6051
6052 if (ctx->xin->resubmit_stats) {
6053 netdev_vport_inc_tx(ofport->up.netdev, ctx->xin->resubmit_stats);
6054 netdev_vport_inc_rx(peer->up.netdev, ctx->xin->resubmit_stats);
6055 }
6056
6057 return;
6058 }
6059
6060 flow_vlan_tci = ctx->xin->flow.vlan_tci;
6061 flow_skb_mark = ctx->xin->flow.skb_mark;
6062 flow_nw_tos = ctx->xin->flow.nw_tos;
6063
6064 pdscp = get_priority(ofport, ctx->xin->flow.skb_priority);
6065 if (pdscp) {
6066 ctx->xin->flow.nw_tos &= ~IP_DSCP_MASK;
6067 ctx->xin->flow.nw_tos |= pdscp->dscp;
6068 }
6069
6070 if (ofport->tnl_port) {
6071 /* Save tunnel metadata so that changes made due to
6072 * the Logical (tunnel) Port are not visible for any further
6073 * matches, while explicit set actions on tunnel metadata are.
6074 */
6075 struct flow_tnl flow_tnl = ctx->xin->flow.tunnel;
6076 odp_port = tnl_port_send(ofport->tnl_port, &ctx->xin->flow);
6077 if (odp_port == OVSP_NONE) {
6078 xlate_report(ctx, "Tunneling decided against output");
6079 goto out; /* restore flow_nw_tos */
6080 }
6081 if (ctx->xin->flow.tunnel.ip_dst == ctx->orig_tunnel_ip_dst) {
6082 xlate_report(ctx, "Not tunneling to our own address");
6083 goto out; /* restore flow_nw_tos */
6084 }
6085 if (ctx->xin->resubmit_stats) {
6086 netdev_vport_inc_tx(ofport->up.netdev, ctx->xin->resubmit_stats);
6087 }
6088 out_port = odp_port;
6089 commit_odp_tunnel_action(&ctx->xin->flow, &ctx->base_flow,
6090 &ctx->xout->odp_actions);
6091 ctx->xin->flow.tunnel = flow_tnl; /* Restore tunnel metadata */
6092 } else {
6093 uint16_t vlandev_port;
6094 odp_port = ofport->odp_port;
6095 vlandev_port = vsp_realdev_to_vlandev(ctx->ofproto, ofp_port,
6096 ctx->xin->flow.vlan_tci);
6097 if (vlandev_port == ofp_port) {
6098 out_port = odp_port;
6099 } else {
6100 out_port = ofp_port_to_odp_port(ctx->ofproto, vlandev_port);
6101 ctx->xin->flow.vlan_tci = htons(0);
6102 }
6103 ctx->xin->flow.skb_mark &= ~IPSEC_MARK;
6104 }
6105 commit_odp_actions(&ctx->xin->flow, &ctx->base_flow,
6106 &ctx->xout->odp_actions);
6107 nl_msg_put_u32(&ctx->xout->odp_actions, OVS_ACTION_ATTR_OUTPUT, out_port);
6108
6109 ctx->sflow_odp_port = odp_port;
6110 ctx->sflow_n_outputs++;
6111 ctx->xout->nf_output_iface = ofp_port;
6112
6113 /* Restore flow */
6114 ctx->xin->flow.vlan_tci = flow_vlan_tci;
6115 ctx->xin->flow.skb_mark = flow_skb_mark;
6116 out:
6117 ctx->xin->flow.nw_tos = flow_nw_tos;
6118 }
6119
6120 static void
6121 compose_output_action(struct xlate_ctx *ctx, uint16_t ofp_port)
6122 {
6123 compose_output_action__(ctx, ofp_port, true);
6124 }
6125
6126 static void
6127 tag_the_flow(struct xlate_ctx *ctx, struct rule_dpif *rule)
6128 {
6129 struct ofproto_dpif *ofproto = ctx->ofproto;
6130 uint8_t table_id = ctx->table_id;
6131
6132 if (table_id > 0 && table_id < N_TABLES) {
6133 struct table_dpif *table = &ofproto->tables[table_id];
6134 if (table->other_table) {
6135 ctx->xout->tags |= (rule && rule->tag
6136 ? rule->tag
6137 : rule_calculate_tag(&ctx->xin->flow,
6138 &table->other_table->mask,
6139 table->basis));
6140 }
6141 }
6142 }
6143
6144 /* Common rule processing in one place to avoid duplicating code. */
6145 static struct rule_dpif *
6146 ctx_rule_hooks(struct xlate_ctx *ctx, struct rule_dpif *rule,
6147 bool may_packet_in)
6148 {
6149 if (ctx->xin->resubmit_hook) {
6150 ctx->xin->resubmit_hook(ctx, rule);
6151 }
6152 if (rule == NULL && may_packet_in) {
6153 /* XXX
6154 * check if table configuration flags
6155 * OFPTC_TABLE_MISS_CONTROLLER, default.
6156 * OFPTC_TABLE_MISS_CONTINUE,
6157 * OFPTC_TABLE_MISS_DROP
6158 * When OF1.0, OFPTC_TABLE_MISS_CONTINUE is used. What to do?
6159 */
6160 rule = rule_dpif_miss_rule(ctx->ofproto, &ctx->xin->flow);
6161 }
6162 if (rule && ctx->xin->resubmit_stats) {
6163 rule_credit_stats(rule, ctx->xin->resubmit_stats);
6164 }
6165 return rule;
6166 }
6167
6168 static void
6169 xlate_table_action(struct xlate_ctx *ctx,
6170 uint16_t in_port, uint8_t table_id, bool may_packet_in)
6171 {
6172 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
6173 struct rule_dpif *rule;
6174 uint16_t old_in_port = ctx->xin->flow.in_port;
6175 uint8_t old_table_id = ctx->table_id;
6176
6177 ctx->table_id = table_id;
6178
6179 /* Look up a flow with 'in_port' as the input port. */
6180 ctx->xin->flow.in_port = in_port;
6181 rule = rule_dpif_lookup__(ctx->ofproto, &ctx->xin->flow,
6182 &ctx->xout->wc, table_id);
6183
6184 tag_the_flow(ctx, rule);
6185
6186 /* Restore the original input port. Otherwise OFPP_NORMAL and
6187 * OFPP_IN_PORT will have surprising behavior. */
6188 ctx->xin->flow.in_port = old_in_port;
6189
6190 rule = ctx_rule_hooks(ctx, rule, may_packet_in);
6191
6192 if (rule) {
6193 struct rule_dpif *old_rule = ctx->rule;
6194
6195 ctx->recurse++;
6196 ctx->rule = rule;
6197 do_xlate_actions(rule->up.ofpacts, rule->up.ofpacts_len, ctx);
6198 ctx->rule = old_rule;
6199 ctx->recurse--;
6200 }
6201
6202 ctx->table_id = old_table_id;
6203 } else {
6204 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
6205
6206 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
6207 MAX_RESUBMIT_RECURSION);
6208 ctx->max_resubmit_trigger = true;
6209 }
6210 }
6211
6212 static void
6213 xlate_ofpact_resubmit(struct xlate_ctx *ctx,
6214 const struct ofpact_resubmit *resubmit)
6215 {
6216 uint16_t in_port;
6217 uint8_t table_id;
6218
6219 in_port = resubmit->in_port;
6220 if (in_port == OFPP_IN_PORT) {
6221 in_port = ctx->xin->flow.in_port;
6222 }
6223
6224 table_id = resubmit->table_id;
6225 if (table_id == 255) {
6226 table_id = ctx->table_id;
6227 }
6228
6229 xlate_table_action(ctx, in_port, table_id, false);
6230 }
6231
6232 static void
6233 flood_packets(struct xlate_ctx *ctx, bool all)
6234 {
6235 struct ofport_dpif *ofport;
6236
6237 HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) {
6238 uint16_t ofp_port = ofport->up.ofp_port;
6239
6240 if (ofp_port == ctx->xin->flow.in_port) {
6241 continue;
6242 }
6243
6244 if (all) {
6245 compose_output_action__(ctx, ofp_port, false);
6246 } else if (!(ofport->up.pp.config & OFPUTIL_PC_NO_FLOOD)) {
6247 compose_output_action(ctx, ofp_port);
6248 }
6249 }
6250
6251 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
6252 }
6253
6254 static void
6255 execute_controller_action(struct xlate_ctx *ctx, int len,
6256 enum ofp_packet_in_reason reason,
6257 uint16_t controller_id)
6258 {
6259 struct ofputil_packet_in pin;
6260 struct ofpbuf *packet;
6261 struct flow key;
6262
6263 ovs_assert(!ctx->xout->slow || ctx->xout->slow == SLOW_CONTROLLER);
6264 ctx->xout->slow = SLOW_CONTROLLER;
6265 if (!ctx->xin->packet) {
6266 return;
6267 }
6268
6269 packet = ofpbuf_clone(ctx->xin->packet);
6270
6271 key.skb_priority = 0;
6272 key.skb_mark = 0;
6273 memset(&key.tunnel, 0, sizeof key.tunnel);
6274
6275 commit_odp_actions(&ctx->xin->flow, &ctx->base_flow,
6276 &ctx->xout->odp_actions);
6277
6278 odp_execute_actions(NULL, packet, &key, ctx->xout->odp_actions.data,
6279 ctx->xout->odp_actions.size, NULL, NULL);
6280
6281 pin.packet = packet->data;
6282 pin.packet_len = packet->size;
6283 pin.reason = reason;
6284 pin.controller_id = controller_id;
6285 pin.table_id = ctx->table_id;
6286 pin.cookie = ctx->rule ? ctx->rule->up.flow_cookie : 0;
6287
6288 pin.send_len = len;
6289 flow_get_metadata(&ctx->xin->flow, &pin.fmd);
6290
6291 connmgr_send_packet_in(ctx->ofproto->up.connmgr, &pin);
6292 ofpbuf_delete(packet);
6293 }
6294
6295 static void
6296 execute_mpls_push_action(struct xlate_ctx *ctx, ovs_be16 eth_type)
6297 {
6298 ovs_assert(eth_type_mpls(eth_type));
6299
6300 memset(&ctx->xout->wc.masks.dl_type, 0xff,
6301 sizeof ctx->xout->wc.masks.dl_type);
6302 memset(&ctx->xout->wc.masks.mpls_lse, 0xff,
6303 sizeof ctx->xout->wc.masks.mpls_lse);
6304 memset(&ctx->xout->wc.masks.mpls_depth, 0xff,
6305 sizeof ctx->xout->wc.masks.mpls_depth);
6306
6307 if (ctx->base_flow.mpls_depth) {
6308 ctx->xin->flow.mpls_lse &= ~htonl(MPLS_BOS_MASK);
6309 ctx->xin->flow.mpls_depth++;
6310 } else {
6311 ovs_be32 label;
6312 uint8_t tc, ttl;
6313
6314 if (ctx->xin->flow.dl_type == htons(ETH_TYPE_IPV6)) {
6315 label = htonl(0x2); /* IPV6 Explicit Null. */
6316 } else {
6317 label = htonl(0x0); /* IPV4 Explicit Null. */
6318 }
6319 tc = (ctx->xin->flow.nw_tos & IP_DSCP_MASK) >> 2;
6320 ttl = ctx->xin->flow.nw_ttl ? ctx->xin->flow.nw_ttl : 0x40;
6321 ctx->xin->flow.mpls_lse = set_mpls_lse_values(ttl, tc, 1, label);
6322 ctx->xin->flow.mpls_depth = 1;
6323 }
6324 ctx->xin->flow.dl_type = eth_type;
6325 }
6326
6327 static void
6328 execute_mpls_pop_action(struct xlate_ctx *ctx, ovs_be16 eth_type)
6329 {
6330 ovs_assert(eth_type_mpls(ctx->xin->flow.dl_type));
6331 ovs_assert(!eth_type_mpls(eth_type));
6332
6333 memset(&ctx->xout->wc.masks.dl_type, 0xff,
6334 sizeof ctx->xout->wc.masks.dl_type);
6335 memset(&ctx->xout->wc.masks.mpls_lse, 0xff,
6336 sizeof ctx->xout->wc.masks.mpls_lse);
6337 memset(&ctx->xout->wc.masks.mpls_depth, 0xff,
6338 sizeof ctx->xout->wc.masks.mpls_depth);
6339
6340 if (ctx->xin->flow.mpls_depth) {
6341 ctx->xin->flow.mpls_depth--;
6342 ctx->xin->flow.mpls_lse = htonl(0);
6343 if (!ctx->xin->flow.mpls_depth) {
6344 ctx->xin->flow.dl_type = eth_type;
6345 }
6346 }
6347 }
6348
6349 static bool
6350 compose_dec_ttl(struct xlate_ctx *ctx, struct ofpact_cnt_ids *ids)
6351 {
6352 if (ctx->xin->flow.dl_type != htons(ETH_TYPE_IP) &&
6353 ctx->xin->flow.dl_type != htons(ETH_TYPE_IPV6)) {
6354 return false;
6355 }
6356
6357 if (ctx->xin->flow.nw_ttl > 1) {
6358 ctx->xin->flow.nw_ttl--;
6359 return false;
6360 } else {
6361 size_t i;
6362
6363 for (i = 0; i < ids->n_controllers; i++) {
6364 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL,
6365 ids->cnt_ids[i]);
6366 }
6367
6368 /* Stop processing for current table. */
6369 return true;
6370 }
6371 }
6372
6373 static bool
6374 execute_set_mpls_ttl_action(struct xlate_ctx *ctx, uint8_t ttl)
6375 {
6376 if (!eth_type_mpls(ctx->xin->flow.dl_type)) {
6377 return true;
6378 }
6379
6380 set_mpls_lse_ttl(&ctx->xin->flow.mpls_lse, ttl);
6381 return false;
6382 }
6383
6384 static bool
6385 execute_dec_mpls_ttl_action(struct xlate_ctx *ctx)
6386 {
6387 uint8_t ttl = mpls_lse_to_ttl(ctx->xin->flow.mpls_lse);
6388
6389 if (!eth_type_mpls(ctx->xin->flow.dl_type)) {
6390 return false;
6391 }
6392
6393 if (ttl > 1) {
6394 ttl--;
6395 set_mpls_lse_ttl(&ctx->xin->flow.mpls_lse, ttl);
6396 return false;
6397 } else {
6398 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL, 0);
6399
6400 /* Stop processing for current table. */
6401 return true;
6402 }
6403 }
6404
6405 static void
6406 xlate_output_action(struct xlate_ctx *ctx,
6407 uint16_t port, uint16_t max_len, bool may_packet_in)
6408 {
6409 uint16_t prev_nf_output_iface = ctx->xout->nf_output_iface;
6410
6411 ctx->xout->nf_output_iface = NF_OUT_DROP;
6412
6413 switch (port) {
6414 case OFPP_IN_PORT:
6415 compose_output_action(ctx, ctx->xin->flow.in_port);
6416 break;
6417 case OFPP_TABLE:
6418 xlate_table_action(ctx, ctx->xin->flow.in_port, 0, may_packet_in);
6419 break;
6420 case OFPP_NORMAL:
6421 xlate_normal(ctx);
6422 break;
6423 case OFPP_FLOOD:
6424 flood_packets(ctx, false);
6425 break;
6426 case OFPP_ALL:
6427 flood_packets(ctx, true);
6428 break;
6429 case OFPP_CONTROLLER:
6430 execute_controller_action(ctx, max_len, OFPR_ACTION, 0);
6431 break;
6432 case OFPP_NONE:
6433 break;
6434 case OFPP_LOCAL:
6435 default:
6436 if (port != ctx->xin->flow.in_port) {
6437 compose_output_action(ctx, port);
6438 } else {
6439 xlate_report(ctx, "skipping output to input port");
6440 }
6441 break;
6442 }
6443
6444 if (prev_nf_output_iface == NF_OUT_FLOOD) {
6445 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
6446 } else if (ctx->xout->nf_output_iface == NF_OUT_DROP) {
6447 ctx->xout->nf_output_iface = prev_nf_output_iface;
6448 } else if (prev_nf_output_iface != NF_OUT_DROP &&
6449 ctx->xout->nf_output_iface != NF_OUT_FLOOD) {
6450 ctx->xout->nf_output_iface = NF_OUT_MULTI;
6451 }
6452 }
6453
6454 static void
6455 xlate_output_reg_action(struct xlate_ctx *ctx,
6456 const struct ofpact_output_reg *or)
6457 {
6458 uint64_t port = mf_get_subfield(&or->src, &ctx->xin->flow);
6459 if (port <= UINT16_MAX) {
6460 union mf_subvalue value;
6461
6462 memset(&value, 0xff, sizeof value);
6463 mf_write_subfield_flow(&or->src, &value, &ctx->xout->wc.masks);
6464 xlate_output_action(ctx, port, or->max_len, false);
6465 }
6466 }
6467
6468 static void
6469 xlate_enqueue_action(struct xlate_ctx *ctx,
6470 const struct ofpact_enqueue *enqueue)
6471 {
6472 uint16_t ofp_port = enqueue->port;
6473 uint32_t queue_id = enqueue->queue;
6474 uint32_t flow_priority, priority;
6475 int error;
6476
6477 /* Translate queue to priority. */
6478 error = dpif_queue_to_priority(ctx->ofproto->backer->dpif,
6479 queue_id, &priority);
6480 if (error) {
6481 /* Fall back to ordinary output action. */
6482 xlate_output_action(ctx, enqueue->port, 0, false);
6483 return;
6484 }
6485
6486 /* Check output port. */
6487 if (ofp_port == OFPP_IN_PORT) {
6488 ofp_port = ctx->xin->flow.in_port;
6489 } else if (ofp_port == ctx->xin->flow.in_port) {
6490 return;
6491 }
6492
6493 /* Add datapath actions. */
6494 flow_priority = ctx->xin->flow.skb_priority;
6495 ctx->xin->flow.skb_priority = priority;
6496 compose_output_action(ctx, ofp_port);
6497 ctx->xin->flow.skb_priority = flow_priority;
6498
6499 /* Update NetFlow output port. */
6500 if (ctx->xout->nf_output_iface == NF_OUT_DROP) {
6501 ctx->xout->nf_output_iface = ofp_port;
6502 } else if (ctx->xout->nf_output_iface != NF_OUT_FLOOD) {
6503 ctx->xout->nf_output_iface = NF_OUT_MULTI;
6504 }
6505 }
6506
6507 static void
6508 xlate_set_queue_action(struct xlate_ctx *ctx, uint32_t queue_id)
6509 {
6510 uint32_t skb_priority;
6511
6512 if (!dpif_queue_to_priority(ctx->ofproto->backer->dpif,
6513 queue_id, &skb_priority)) {
6514 ctx->xin->flow.skb_priority = skb_priority;
6515 } else {
6516 /* Couldn't translate queue to a priority. Nothing to do. A warning
6517 * has already been logged. */
6518 }
6519 }
6520
6521 static bool
6522 slave_enabled_cb(uint16_t ofp_port, void *ofproto_)
6523 {
6524 struct ofproto_dpif *ofproto = ofproto_;
6525 struct ofport_dpif *port;
6526
6527 switch (ofp_port) {
6528 case OFPP_IN_PORT:
6529 case OFPP_TABLE:
6530 case OFPP_NORMAL:
6531 case OFPP_FLOOD:
6532 case OFPP_ALL:
6533 case OFPP_NONE:
6534 return true;
6535 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
6536 return false;
6537 default:
6538 port = get_ofp_port(ofproto, ofp_port);
6539 return port ? port->may_enable : false;
6540 }
6541 }
6542
6543 static void
6544 xlate_bundle_action(struct xlate_ctx *ctx,
6545 const struct ofpact_bundle *bundle)
6546 {
6547 uint16_t port;
6548
6549 port = bundle_execute(bundle, &ctx->xin->flow, &ctx->xout->wc,
6550 slave_enabled_cb, ctx->ofproto);
6551 if (bundle->dst.field) {
6552 nxm_reg_load(&bundle->dst, port, &ctx->xin->flow);
6553 } else {
6554 xlate_output_action(ctx, port, 0, false);
6555 }
6556 }
6557
6558 static void
6559 xlate_learn_action(struct xlate_ctx *ctx,
6560 const struct ofpact_learn *learn)
6561 {
6562 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
6563 struct ofputil_flow_mod fm;
6564 uint64_t ofpacts_stub[1024 / 8];
6565 struct ofpbuf ofpacts;
6566 int error;
6567
6568 ctx->xout->has_learn = true;
6569
6570 learn_mask(learn, &ctx->xout->wc);
6571
6572 if (!ctx->xin->may_learn) {
6573 return;
6574 }
6575
6576 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
6577 learn_execute(learn, &ctx->xin->flow, &fm, &ofpacts);
6578
6579 error = ofproto_flow_mod(&ctx->ofproto->up, &fm);
6580 if (error && !VLOG_DROP_WARN(&rl)) {
6581 VLOG_WARN("learning action failed to modify flow table (%s)",
6582 ofperr_get_name(error));
6583 }
6584
6585 ofpbuf_uninit(&ofpacts);
6586 }
6587
6588 /* Reduces '*timeout' to no more than 'max'. A value of zero in either case
6589 * means "infinite". */
6590 static void
6591 reduce_timeout(uint16_t max, uint16_t *timeout)
6592 {
6593 if (max && (!*timeout || *timeout > max)) {
6594 *timeout = max;
6595 }
6596 }
6597
6598 static void
6599 xlate_fin_timeout(struct xlate_ctx *ctx,
6600 const struct ofpact_fin_timeout *oft)
6601 {
6602 if (ctx->xin->tcp_flags & (TCP_FIN | TCP_RST) && ctx->rule) {
6603 struct rule_dpif *rule = ctx->rule;
6604
6605 reduce_timeout(oft->fin_idle_timeout, &rule->up.idle_timeout);
6606 reduce_timeout(oft->fin_hard_timeout, &rule->up.hard_timeout);
6607 }
6608 }
6609
6610 static void
6611 xlate_sample_action(struct xlate_ctx *ctx,
6612 const struct ofpact_sample *os)
6613 {
6614 union user_action_cookie cookie;
6615 /* Scale the probability from 16-bit to 32-bit while representing
6616 * the same percentage. */
6617 uint32_t probability = (os->probability << 16) | os->probability;
6618
6619 commit_odp_actions(&ctx->xin->flow, &ctx->base_flow,
6620 &ctx->xout->odp_actions);
6621
6622 compose_flow_sample_cookie(os->probability, os->collector_set_id,
6623 os->obs_domain_id, os->obs_point_id, &cookie);
6624 compose_sample_action(ctx->ofproto, &ctx->xout->odp_actions, &ctx->xin->flow,
6625 probability, &cookie, sizeof cookie.flow_sample);
6626 }
6627
6628 static bool
6629 may_receive(const struct ofport_dpif *port, struct xlate_ctx *ctx)
6630 {
6631 if (port->up.pp.config & (eth_addr_equals(ctx->xin->flow.dl_dst,
6632 eth_addr_stp)
6633 ? OFPUTIL_PC_NO_RECV_STP
6634 : OFPUTIL_PC_NO_RECV)) {
6635 return false;
6636 }
6637
6638 /* Only drop packets here if both forwarding and learning are
6639 * disabled. If just learning is enabled, we need to have
6640 * OFPP_NORMAL and the learning action have a look at the packet
6641 * before we can drop it. */
6642 if (!stp_forward_in_state(port->stp_state)
6643 && !stp_learn_in_state(port->stp_state)) {
6644 return false;
6645 }
6646
6647 return true;
6648 }
6649
6650 static bool
6651 tunnel_ecn_ok(struct xlate_ctx *ctx)
6652 {
6653 if (is_ip_any(&ctx->base_flow)
6654 && (ctx->xin->flow.tunnel.ip_tos & IP_ECN_MASK) == IP_ECN_CE) {
6655 if ((ctx->base_flow.nw_tos & IP_ECN_MASK) == IP_ECN_NOT_ECT) {
6656 VLOG_WARN_RL(&rl, "dropping tunnel packet marked ECN CE"
6657 " but is not ECN capable");
6658 return false;
6659 } else {
6660 /* Set the ECN CE value in the tunneled packet. */
6661 ctx->xin->flow.nw_tos |= IP_ECN_CE;
6662 }
6663 }
6664
6665 return true;
6666 }
6667
6668 static void
6669 do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len,
6670 struct xlate_ctx *ctx)
6671 {
6672 bool was_evictable = true;
6673 const struct ofpact *a;
6674
6675 if (ctx->rule) {
6676 /* Don't let the rule we're working on get evicted underneath us. */
6677 was_evictable = ctx->rule->up.evictable;
6678 ctx->rule->up.evictable = false;
6679 }
6680
6681 do_xlate_actions_again:
6682 OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) {
6683 struct ofpact_controller *controller;
6684 const struct ofpact_metadata *metadata;
6685
6686 if (ctx->exit) {
6687 break;
6688 }
6689
6690 switch (a->type) {
6691 case OFPACT_OUTPUT:
6692 xlate_output_action(ctx, ofpact_get_OUTPUT(a)->port,
6693 ofpact_get_OUTPUT(a)->max_len, true);
6694 break;
6695
6696 case OFPACT_CONTROLLER:
6697 controller = ofpact_get_CONTROLLER(a);
6698 execute_controller_action(ctx, controller->max_len,
6699 controller->reason,
6700 controller->controller_id);
6701 break;
6702
6703 case OFPACT_ENQUEUE:
6704 xlate_enqueue_action(ctx, ofpact_get_ENQUEUE(a));
6705 break;
6706
6707 case OFPACT_SET_VLAN_VID:
6708 ctx->xin->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
6709 ctx->xin->flow.vlan_tci |=
6710 (htons(ofpact_get_SET_VLAN_VID(a)->vlan_vid)
6711 | htons(VLAN_CFI));
6712 break;
6713
6714 case OFPACT_SET_VLAN_PCP:
6715 ctx->xin->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
6716 ctx->xin->flow.vlan_tci |=
6717 htons((ofpact_get_SET_VLAN_PCP(a)->vlan_pcp << VLAN_PCP_SHIFT)
6718 | VLAN_CFI);
6719 break;
6720
6721 case OFPACT_STRIP_VLAN:
6722 ctx->xin->flow.vlan_tci = htons(0);
6723 break;
6724
6725 case OFPACT_PUSH_VLAN:
6726 /* XXX 802.1AD(QinQ) */
6727 ctx->xin->flow.vlan_tci = htons(VLAN_CFI);
6728 break;
6729
6730 case OFPACT_SET_ETH_SRC:
6731 memcpy(ctx->xin->flow.dl_src, ofpact_get_SET_ETH_SRC(a)->mac,
6732 ETH_ADDR_LEN);
6733 break;
6734
6735 case OFPACT_SET_ETH_DST:
6736 memcpy(ctx->xin->flow.dl_dst, ofpact_get_SET_ETH_DST(a)->mac,
6737 ETH_ADDR_LEN);
6738 break;
6739
6740 case OFPACT_SET_IPV4_SRC:
6741 memset(&ctx->xout->wc.masks.dl_type, 0xff,
6742 sizeof ctx->xout->wc.masks.dl_type);
6743 if (ctx->xin->flow.dl_type == htons(ETH_TYPE_IP)) {
6744 ctx->xin->flow.nw_src = ofpact_get_SET_IPV4_SRC(a)->ipv4;
6745 }
6746 break;
6747
6748 case OFPACT_SET_IPV4_DST:
6749 memset(&ctx->xout->wc.masks.dl_type, 0xff,
6750 sizeof ctx->xout->wc.masks.dl_type);
6751 if (ctx->xin->flow.dl_type == htons(ETH_TYPE_IP)) {
6752 ctx->xin->flow.nw_dst = ofpact_get_SET_IPV4_DST(a)->ipv4;
6753 }
6754 break;
6755
6756 case OFPACT_SET_IPV4_DSCP:
6757 /* OpenFlow 1.0 only supports IPv4. */
6758 memset(&ctx->xout->wc.masks.dl_type, 0xff,
6759 sizeof ctx->xout->wc.masks.dl_type);
6760 if (ctx->xin->flow.dl_type == htons(ETH_TYPE_IP)) {
6761 ctx->xin->flow.nw_tos &= ~IP_DSCP_MASK;
6762 ctx->xin->flow.nw_tos |= ofpact_get_SET_IPV4_DSCP(a)->dscp;
6763 }
6764 break;
6765
6766 case OFPACT_SET_L4_SRC_PORT:
6767 memset(&ctx->xout->wc.masks.dl_type, 0xff,
6768 sizeof ctx->xout->wc.masks.dl_type);
6769 memset(&ctx->xout->wc.masks.nw_proto, 0xff,
6770 sizeof ctx->xout->wc.masks.nw_proto);
6771 if (is_ip_any(&ctx->xin->flow)) {
6772 ctx->xin->flow.tp_src =
6773 htons(ofpact_get_SET_L4_SRC_PORT(a)->port);
6774 }
6775 break;
6776
6777 case OFPACT_SET_L4_DST_PORT:
6778 memset(&ctx->xout->wc.masks.dl_type, 0xff,
6779 sizeof ctx->xout->wc.masks.dl_type);
6780 memset(&ctx->xout->wc.masks.nw_proto, 0xff,
6781 sizeof ctx->xout->wc.masks.nw_proto);
6782 if (is_ip_any(&ctx->xin->flow)) {
6783 ctx->xin->flow.tp_dst =
6784 htons(ofpact_get_SET_L4_DST_PORT(a)->port);
6785 }
6786 break;
6787
6788 case OFPACT_RESUBMIT:
6789 xlate_ofpact_resubmit(ctx, ofpact_get_RESUBMIT(a));
6790 break;
6791
6792 case OFPACT_SET_TUNNEL:
6793 ctx->xin->flow.tunnel.tun_id =
6794 htonll(ofpact_get_SET_TUNNEL(a)->tun_id);
6795 break;
6796
6797 case OFPACT_SET_QUEUE:
6798 xlate_set_queue_action(ctx, ofpact_get_SET_QUEUE(a)->queue_id);
6799 break;
6800
6801 case OFPACT_POP_QUEUE:
6802 memset(&ctx->xout->wc.masks.skb_priority, 0xff,
6803 sizeof ctx->xout->wc.masks.skb_priority);
6804
6805 ctx->xin->flow.skb_priority = ctx->orig_skb_priority;
6806 break;
6807
6808 case OFPACT_REG_MOVE:
6809 nxm_execute_reg_move(ofpact_get_REG_MOVE(a), &ctx->xin->flow,
6810 &ctx->xout->wc);
6811 break;
6812
6813 case OFPACT_REG_LOAD:
6814 nxm_execute_reg_load(ofpact_get_REG_LOAD(a), &ctx->xin->flow);
6815 break;
6816
6817 case OFPACT_STACK_PUSH:
6818 nxm_execute_stack_push(ofpact_get_STACK_PUSH(a), &ctx->xin->flow,
6819 &ctx->xout->wc, &ctx->stack);
6820 break;
6821
6822 case OFPACT_STACK_POP:
6823 nxm_execute_stack_pop(ofpact_get_STACK_POP(a), &ctx->xin->flow,
6824 &ctx->stack);
6825 break;
6826
6827 case OFPACT_PUSH_MPLS:
6828 execute_mpls_push_action(ctx, ofpact_get_PUSH_MPLS(a)->ethertype);
6829 break;
6830
6831 case OFPACT_POP_MPLS:
6832 execute_mpls_pop_action(ctx, ofpact_get_POP_MPLS(a)->ethertype);
6833 break;
6834
6835 case OFPACT_SET_MPLS_TTL:
6836 if (execute_set_mpls_ttl_action(ctx,
6837 ofpact_get_SET_MPLS_TTL(a)->ttl)) {
6838 goto out;
6839 }
6840 break;
6841
6842 case OFPACT_DEC_MPLS_TTL:
6843 if (execute_dec_mpls_ttl_action(ctx)) {
6844 goto out;
6845 }
6846 break;
6847
6848 case OFPACT_DEC_TTL:
6849 memset(&ctx->xout->wc.masks.dl_type, 0xff,
6850 sizeof ctx->xout->wc.masks.dl_type);
6851 if (compose_dec_ttl(ctx, ofpact_get_DEC_TTL(a))) {
6852 goto out;
6853 }
6854 break;
6855
6856 case OFPACT_NOTE:
6857 /* Nothing to do. */
6858 break;
6859
6860 case OFPACT_MULTIPATH:
6861 multipath_execute(ofpact_get_MULTIPATH(a), &ctx->xin->flow,
6862 &ctx->xout->wc);
6863 break;
6864
6865 case OFPACT_BUNDLE:
6866 ctx->ofproto->has_bundle_action = true;
6867 xlate_bundle_action(ctx, ofpact_get_BUNDLE(a));
6868 break;
6869
6870 case OFPACT_OUTPUT_REG:
6871 xlate_output_reg_action(ctx, ofpact_get_OUTPUT_REG(a));
6872 break;
6873
6874 case OFPACT_LEARN:
6875 xlate_learn_action(ctx, ofpact_get_LEARN(a));
6876 break;
6877
6878 case OFPACT_EXIT:
6879 ctx->exit = true;
6880 break;
6881
6882 case OFPACT_FIN_TIMEOUT:
6883 memset(&ctx->xout->wc.masks.dl_type, 0xff,
6884 sizeof ctx->xout->wc.masks.dl_type);
6885 memset(&ctx->xout->wc.masks.nw_proto, 0xff,
6886 sizeof ctx->xout->wc.masks.nw_proto);
6887 ctx->xout->has_fin_timeout = true;
6888 xlate_fin_timeout(ctx, ofpact_get_FIN_TIMEOUT(a));
6889 break;
6890
6891 case OFPACT_CLEAR_ACTIONS:
6892 /* XXX
6893 * Nothing to do because writa-actions is not supported for now.
6894 * When writa-actions is supported, clear-actions also must
6895 * be supported at the same time.
6896 */
6897 break;
6898
6899 case OFPACT_WRITE_METADATA:
6900 metadata = ofpact_get_WRITE_METADATA(a);
6901 ctx->xin->flow.metadata &= ~metadata->mask;
6902 ctx->xin->flow.metadata |= metadata->metadata & metadata->mask;
6903 break;
6904
6905 case OFPACT_GOTO_TABLE: {
6906 /* It is assumed that goto-table is the last action. */
6907 struct ofpact_goto_table *ogt = ofpact_get_GOTO_TABLE(a);
6908 struct rule_dpif *rule;
6909
6910 ovs_assert(ctx->table_id < ogt->table_id);
6911
6912 ctx->table_id = ogt->table_id;
6913
6914 /* Look up a flow from the new table. */
6915 rule = rule_dpif_lookup__(ctx->ofproto, &ctx->xin->flow,
6916 &ctx->xout->wc, ctx->table_id);
6917
6918 tag_the_flow(ctx, rule);
6919
6920 rule = ctx_rule_hooks(ctx, rule, true);
6921
6922 if (rule) {
6923 if (ctx->rule) {
6924 ctx->rule->up.evictable = was_evictable;
6925 }
6926 ctx->rule = rule;
6927 was_evictable = rule->up.evictable;
6928 rule->up.evictable = false;
6929
6930 /* Tail recursion removal. */
6931 ofpacts = rule->up.ofpacts;
6932 ofpacts_len = rule->up.ofpacts_len;
6933 goto do_xlate_actions_again;
6934 }
6935 break;
6936 }
6937
6938 case OFPACT_SAMPLE:
6939 xlate_sample_action(ctx, ofpact_get_SAMPLE(a));
6940 break;
6941 }
6942 }
6943
6944 out:
6945 if (ctx->rule) {
6946 ctx->rule->up.evictable = was_evictable;
6947 }
6948 }
6949
6950 static void
6951 xlate_in_init(struct xlate_in *xin, struct ofproto_dpif *ofproto,
6952 const struct flow *flow,
6953 const struct initial_vals *initial_vals,
6954 struct rule_dpif *rule, uint8_t tcp_flags,
6955 const struct ofpbuf *packet)
6956 {
6957 xin->ofproto = ofproto;
6958 xin->flow = *flow;
6959 xin->packet = packet;
6960 xin->may_learn = packet != NULL;
6961 xin->rule = rule;
6962 xin->ofpacts = NULL;
6963 xin->ofpacts_len = 0;
6964 xin->tcp_flags = tcp_flags;
6965 xin->resubmit_hook = NULL;
6966 xin->report_hook = NULL;
6967 xin->resubmit_stats = NULL;
6968
6969 if (initial_vals) {
6970 xin->initial_vals = *initial_vals;
6971 } else {
6972 xin->initial_vals.vlan_tci = xin->flow.vlan_tci;
6973 }
6974 }
6975
6976 static void
6977 xlate_out_uninit(struct xlate_out *xout)
6978 {
6979 if (xout) {
6980 ofpbuf_uninit(&xout->odp_actions);
6981 }
6982 }
6983
6984 /* Translates the 'ofpacts_len' bytes of "struct ofpacts" starting at 'ofpacts'
6985 * into datapath actions in 'odp_actions', using 'ctx'. */
6986 static void
6987 xlate_actions(struct xlate_in *xin, struct xlate_out *xout)
6988 {
6989 /* Normally false. Set to true if we ever hit MAX_RESUBMIT_RECURSION, so
6990 * that in the future we always keep a copy of the original flow for
6991 * tracing purposes. */
6992 static bool hit_resubmit_limit;
6993
6994 enum slow_path_reason special;
6995 const struct ofpact *ofpacts;
6996 struct ofport_dpif *in_port;
6997 struct flow orig_flow;
6998 struct xlate_ctx ctx;
6999 size_t ofpacts_len;
7000
7001 COVERAGE_INC(ofproto_dpif_xlate);
7002
7003 /* Flow initialization rules:
7004 * - 'base_flow' must match the kernel's view of the packet at the
7005 * time that action processing starts. 'flow' represents any
7006 * transformations we wish to make through actions.
7007 * - By default 'base_flow' and 'flow' are the same since the input
7008 * packet matches the output before any actions are applied.
7009 * - When using VLAN splinters, 'base_flow''s VLAN is set to the value
7010 * of the received packet as seen by the kernel. If we later output
7011 * to another device without any modifications this will cause us to
7012 * insert a new tag since the original one was stripped off by the
7013 * VLAN device.
7014 * - Tunnel metadata as received is retained in 'flow'. This allows
7015 * tunnel metadata matching also in later tables.
7016 * Since a kernel action for setting the tunnel metadata will only be
7017 * generated with actual tunnel output, changing the tunnel metadata
7018 * values in 'flow' (such as tun_id) will only have effect with a later
7019 * tunnel output action.
7020 * - Tunnel 'base_flow' is completely cleared since that is what the
7021 * kernel does. If we wish to maintain the original values an action
7022 * needs to be generated. */
7023
7024 ctx.xin = xin;
7025 ctx.xout = xout;
7026
7027 ctx.ofproto = xin->ofproto;
7028 ctx.rule = xin->rule;
7029
7030 ctx.base_flow = ctx.xin->flow;
7031 ctx.base_flow.vlan_tci = xin->initial_vals.vlan_tci;
7032 memset(&ctx.base_flow.tunnel, 0, sizeof ctx.base_flow.tunnel);
7033 ctx.orig_tunnel_ip_dst = ctx.xin->flow.tunnel.ip_dst;
7034
7035 flow_wildcards_init_catchall(&ctx.xout->wc);
7036 memset(&ctx.xout->wc.masks.in_port, 0xff,
7037 sizeof ctx.xout->wc.masks.in_port);
7038
7039 if (tnl_port_should_receive(&ctx.xin->flow)) {
7040 memset(&ctx.xout->wc.masks.tunnel, 0xff,
7041 sizeof ctx.xout->wc.masks.tunnel);
7042 }
7043
7044 /* Disable most wildcarding for NetFlow. */
7045 if (xin->ofproto->netflow) {
7046 memset(&ctx.xout->wc.masks.dl_src, 0xff,
7047 sizeof ctx.xout->wc.masks.dl_src);
7048 memset(&ctx.xout->wc.masks.dl_dst, 0xff,
7049 sizeof ctx.xout->wc.masks.dl_dst);
7050 memset(&ctx.xout->wc.masks.dl_type, 0xff,
7051 sizeof ctx.xout->wc.masks.dl_type);
7052 memset(&ctx.xout->wc.masks.vlan_tci, 0xff,
7053 sizeof ctx.xout->wc.masks.vlan_tci);
7054 memset(&ctx.xout->wc.masks.nw_proto, 0xff,
7055 sizeof ctx.xout->wc.masks.nw_proto);
7056 memset(&ctx.xout->wc.masks.nw_src, 0xff,
7057 sizeof ctx.xout->wc.masks.nw_src);
7058 memset(&ctx.xout->wc.masks.nw_dst, 0xff,
7059 sizeof ctx.xout->wc.masks.nw_dst);
7060 memset(&ctx.xout->wc.masks.tp_src, 0xff,
7061 sizeof ctx.xout->wc.masks.tp_src);
7062 memset(&ctx.xout->wc.masks.tp_dst, 0xff,
7063 sizeof ctx.xout->wc.masks.tp_dst);
7064 }
7065
7066 ctx.xout->tags = 0;
7067 ctx.xout->slow = 0;
7068 ctx.xout->has_learn = false;
7069 ctx.xout->has_normal = false;
7070 ctx.xout->has_fin_timeout = false;
7071 ctx.xout->nf_output_iface = NF_OUT_DROP;
7072 ctx.xout->mirrors = 0;
7073
7074 ofpbuf_use_stub(&ctx.xout->odp_actions, ctx.xout->odp_actions_stub,
7075 sizeof ctx.xout->odp_actions_stub);
7076 ofpbuf_reserve(&ctx.xout->odp_actions, NL_A_U32_SIZE);
7077
7078 ctx.recurse = 0;
7079 ctx.max_resubmit_trigger = false;
7080 ctx.orig_skb_priority = ctx.xin->flow.skb_priority;
7081 ctx.table_id = 0;
7082 ctx.exit = false;
7083
7084 if (xin->ofpacts) {
7085 ofpacts = xin->ofpacts;
7086 ofpacts_len = xin->ofpacts_len;
7087 } else if (xin->rule) {
7088 ofpacts = xin->rule->up.ofpacts;
7089 ofpacts_len = xin->rule->up.ofpacts_len;
7090 } else {
7091 NOT_REACHED();
7092 }
7093
7094 ofpbuf_use_stub(&ctx.stack, ctx.init_stack, sizeof ctx.init_stack);
7095
7096 if (ctx.ofproto->has_mirrors || hit_resubmit_limit) {
7097 /* Do this conditionally because the copy is expensive enough that it
7098 * shows up in profiles. */
7099 orig_flow = ctx.xin->flow;
7100 }
7101
7102 if (ctx.xin->flow.nw_frag & FLOW_NW_FRAG_ANY) {
7103 switch (ctx.ofproto->up.frag_handling) {
7104 case OFPC_FRAG_NORMAL:
7105 /* We must pretend that transport ports are unavailable. */
7106 ctx.xin->flow.tp_src = ctx.base_flow.tp_src = htons(0);
7107 ctx.xin->flow.tp_dst = ctx.base_flow.tp_dst = htons(0);
7108 break;
7109
7110 case OFPC_FRAG_DROP:
7111 return;
7112
7113 case OFPC_FRAG_REASM:
7114 NOT_REACHED();
7115
7116 case OFPC_FRAG_NX_MATCH:
7117 /* Nothing to do. */
7118 break;
7119
7120 case OFPC_INVALID_TTL_TO_CONTROLLER:
7121 NOT_REACHED();
7122 }
7123 }
7124
7125 in_port = get_ofp_port(ctx.ofproto, ctx.xin->flow.in_port);
7126 special = process_special(ctx.ofproto, &ctx.xin->flow, in_port,
7127 ctx.xin->packet);
7128 if (special) {
7129 ctx.xout->slow = special;
7130 } else {
7131 static struct vlog_rate_limit trace_rl = VLOG_RATE_LIMIT_INIT(1, 1);
7132 struct initial_vals initial_vals;
7133 size_t sample_actions_len;
7134 uint32_t local_odp_port;
7135
7136 initial_vals.vlan_tci = ctx.base_flow.vlan_tci;
7137
7138 add_sflow_action(&ctx);
7139 add_ipfix_action(&ctx);
7140 sample_actions_len = ctx.xout->odp_actions.size;
7141
7142 if (tunnel_ecn_ok(&ctx) && (!in_port || may_receive(in_port, &ctx))) {
7143 do_xlate_actions(ofpacts, ofpacts_len, &ctx);
7144
7145 /* We've let OFPP_NORMAL and the learning action look at the
7146 * packet, so drop it now if forwarding is disabled. */
7147 if (in_port && !stp_forward_in_state(in_port->stp_state)) {
7148 ctx.xout->odp_actions.size = sample_actions_len;
7149 }
7150 }
7151
7152 if (ctx.max_resubmit_trigger && !ctx.xin->resubmit_hook) {
7153 if (!hit_resubmit_limit) {
7154 /* We didn't record the original flow. Make sure we do from
7155 * now on. */
7156 hit_resubmit_limit = true;
7157 } else if (!VLOG_DROP_ERR(&trace_rl)) {
7158 struct ds ds = DS_EMPTY_INITIALIZER;
7159
7160 ofproto_trace(ctx.ofproto, &orig_flow, ctx.xin->packet,
7161 &initial_vals, &ds);
7162 VLOG_ERR("Trace triggered by excessive resubmit "
7163 "recursion:\n%s", ds_cstr(&ds));
7164 ds_destroy(&ds);
7165 }
7166 }
7167
7168 local_odp_port = ofp_port_to_odp_port(ctx.ofproto, OFPP_LOCAL);
7169 if (!connmgr_must_output_local(ctx.ofproto->up.connmgr, &ctx.xin->flow,
7170 local_odp_port,
7171 ctx.xout->odp_actions.data,
7172 ctx.xout->odp_actions.size)) {
7173 compose_output_action(&ctx, OFPP_LOCAL);
7174 }
7175 if (ctx.ofproto->has_mirrors) {
7176 add_mirror_actions(&ctx, &orig_flow);
7177 }
7178 fix_sflow_action(&ctx);
7179 }
7180
7181 ofpbuf_uninit(&ctx.stack);
7182
7183 /* Clear the metadata and register wildcard masks, because we won't
7184 * use non-header fields as part of the cache. */
7185 memset(&ctx.xout->wc.masks.metadata, 0,
7186 sizeof ctx.xout->wc.masks.metadata);
7187 memset(&ctx.xout->wc.masks.regs, 0, sizeof ctx.xout->wc.masks.regs);
7188 }
7189
7190 /* Translates the 'ofpacts_len' bytes of "struct ofpact"s starting at 'ofpacts'
7191 * into datapath actions, using 'ctx', and discards the datapath actions. */
7192 static void
7193 xlate_actions_for_side_effects(struct xlate_in *xin)
7194 {
7195 struct xlate_out xout;
7196
7197 xlate_actions(xin, &xout);
7198 xlate_out_uninit(&xout);
7199 }
7200
7201 static void
7202 xlate_report(struct xlate_ctx *ctx, const char *s)
7203 {
7204 if (ctx->xin->report_hook) {
7205 ctx->xin->report_hook(ctx, s);
7206 }
7207 }
7208
7209 static void
7210 xlate_out_copy(struct xlate_out *dst, const struct xlate_out *src)
7211 {
7212 dst->wc = src->wc;
7213 dst->tags = src->tags;
7214 dst->slow = src->slow;
7215 dst->has_learn = src->has_learn;
7216 dst->has_normal = src->has_normal;
7217 dst->has_fin_timeout = src->has_fin_timeout;
7218 dst->nf_output_iface = src->nf_output_iface;
7219 dst->mirrors = src->mirrors;
7220
7221 ofpbuf_use_stub(&dst->odp_actions, dst->odp_actions_stub,
7222 sizeof dst->odp_actions_stub);
7223 ofpbuf_put(&dst->odp_actions, src->odp_actions.data,
7224 src->odp_actions.size);
7225 }
7226 \f
7227 /* OFPP_NORMAL implementation. */
7228
7229 static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *);
7230
7231 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
7232 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle',
7233 * the bundle on which the packet was received, returns the VLAN to which the
7234 * packet belongs.
7235 *
7236 * Both 'vid' and the return value are in the range 0...4095. */
7237 static uint16_t
7238 input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid)
7239 {
7240 switch (in_bundle->vlan_mode) {
7241 case PORT_VLAN_ACCESS:
7242 return in_bundle->vlan;
7243 break;
7244
7245 case PORT_VLAN_TRUNK:
7246 return vid;
7247
7248 case PORT_VLAN_NATIVE_UNTAGGED:
7249 case PORT_VLAN_NATIVE_TAGGED:
7250 return vid ? vid : in_bundle->vlan;
7251
7252 default:
7253 NOT_REACHED();
7254 }
7255 }
7256
7257 /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'.
7258 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
7259 * a warning.
7260 *
7261 * 'vid' should be the VID obtained from the 802.1Q header that was received as
7262 * part of a packet (specify 0 if there was no 802.1Q header), in the range
7263 * 0...4095. */
7264 static bool
7265 input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn)
7266 {
7267 /* Allow any VID on the OFPP_NONE port. */
7268 if (in_bundle == &ofpp_none_bundle) {
7269 return true;
7270 }
7271
7272 switch (in_bundle->vlan_mode) {
7273 case PORT_VLAN_ACCESS:
7274 if (vid) {
7275 if (warn) {
7276 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7277 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
7278 "packet received on port %s configured as VLAN "
7279 "%"PRIu16" access port",
7280 in_bundle->ofproto->up.name, vid,
7281 in_bundle->name, in_bundle->vlan);
7282 }
7283 return false;
7284 }
7285 return true;
7286
7287 case PORT_VLAN_NATIVE_UNTAGGED:
7288 case PORT_VLAN_NATIVE_TAGGED:
7289 if (!vid) {
7290 /* Port must always carry its native VLAN. */
7291 return true;
7292 }
7293 /* Fall through. */
7294 case PORT_VLAN_TRUNK:
7295 if (!ofbundle_includes_vlan(in_bundle, vid)) {
7296 if (warn) {
7297 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7298 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet "
7299 "received on port %s not configured for trunking "
7300 "VLAN %"PRIu16,
7301 in_bundle->ofproto->up.name, vid,
7302 in_bundle->name, vid);
7303 }
7304 return false;
7305 }
7306 return true;
7307
7308 default:
7309 NOT_REACHED();
7310 }
7311
7312 }
7313
7314 /* Given 'vlan', the VLAN that a packet belongs to, and
7315 * 'out_bundle', a bundle on which the packet is to be output, returns the VID
7316 * that should be included in the 802.1Q header. (If the return value is 0,
7317 * then the 802.1Q header should only be included in the packet if there is a
7318 * nonzero PCP.)
7319 *
7320 * Both 'vlan' and the return value are in the range 0...4095. */
7321 static uint16_t
7322 output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan)
7323 {
7324 switch (out_bundle->vlan_mode) {
7325 case PORT_VLAN_ACCESS:
7326 return 0;
7327
7328 case PORT_VLAN_TRUNK:
7329 case PORT_VLAN_NATIVE_TAGGED:
7330 return vlan;
7331
7332 case PORT_VLAN_NATIVE_UNTAGGED:
7333 return vlan == out_bundle->vlan ? 0 : vlan;
7334
7335 default:
7336 NOT_REACHED();
7337 }
7338 }
7339
7340 static void
7341 output_normal(struct xlate_ctx *ctx, const struct ofbundle *out_bundle,
7342 uint16_t vlan)
7343 {
7344 struct ofport_dpif *port;
7345 uint16_t vid;
7346 ovs_be16 tci, old_tci;
7347
7348 vid = output_vlan_to_vid(out_bundle, vlan);
7349 if (!out_bundle->bond) {
7350 port = ofbundle_get_a_port(out_bundle);
7351 } else {
7352 port = bond_choose_output_slave(out_bundle->bond, &ctx->xin->flow,
7353 &ctx->xout->wc, vid, &ctx->xout->tags);
7354 if (!port) {
7355 /* No slaves enabled, so drop packet. */
7356 return;
7357 }
7358 }
7359
7360 old_tci = ctx->xin->flow.vlan_tci;
7361 tci = htons(vid);
7362 if (tci || out_bundle->use_priority_tags) {
7363 tci |= ctx->xin->flow.vlan_tci & htons(VLAN_PCP_MASK);
7364 if (tci) {
7365 tci |= htons(VLAN_CFI);
7366 }
7367 }
7368 ctx->xin->flow.vlan_tci = tci;
7369
7370 compose_output_action(ctx, port->up.ofp_port);
7371 ctx->xin->flow.vlan_tci = old_tci;
7372 }
7373
7374 static int
7375 mirror_mask_ffs(mirror_mask_t mask)
7376 {
7377 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
7378 return ffs(mask);
7379 }
7380
7381 static bool
7382 ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
7383 {
7384 return (bundle->vlan_mode != PORT_VLAN_ACCESS
7385 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
7386 }
7387
7388 static bool
7389 ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan)
7390 {
7391 return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan);
7392 }
7393
7394 /* Returns an arbitrary interface within 'bundle'. */
7395 static struct ofport_dpif *
7396 ofbundle_get_a_port(const struct ofbundle *bundle)
7397 {
7398 return CONTAINER_OF(list_front(&bundle->ports),
7399 struct ofport_dpif, bundle_node);
7400 }
7401
7402 static bool
7403 vlan_is_mirrored(const struct ofmirror *m, int vlan)
7404 {
7405 return !m->vlans || bitmap_is_set(m->vlans, vlan);
7406 }
7407
7408 static void
7409 add_mirror_actions(struct xlate_ctx *ctx, const struct flow *orig_flow)
7410 {
7411 struct ofproto_dpif *ofproto = ctx->ofproto;
7412 mirror_mask_t mirrors;
7413 struct ofbundle *in_bundle;
7414 uint16_t vlan;
7415 uint16_t vid;
7416 const struct nlattr *a;
7417 size_t left;
7418
7419 in_bundle = lookup_input_bundle(ctx->ofproto, orig_flow->in_port,
7420 ctx->xin->packet != NULL, NULL);
7421 if (!in_bundle) {
7422 return;
7423 }
7424 mirrors = in_bundle->src_mirrors;
7425
7426 /* Drop frames on bundles reserved for mirroring. */
7427 if (in_bundle->mirror_out) {
7428 if (ctx->xin->packet != NULL) {
7429 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7430 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
7431 "%s, which is reserved exclusively for mirroring",
7432 ctx->ofproto->up.name, in_bundle->name);
7433 }
7434 return;
7435 }
7436
7437 /* Check VLAN. */
7438 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
7439 if (!input_vid_is_valid(vid, in_bundle, ctx->xin->packet != NULL)) {
7440 return;
7441 }
7442 vlan = input_vid_to_vlan(in_bundle, vid);
7443
7444 /* Look at the output ports to check for destination selections. */
7445
7446 NL_ATTR_FOR_EACH (a, left, ctx->xout->odp_actions.data,
7447 ctx->xout->odp_actions.size) {
7448 enum ovs_action_attr type = nl_attr_type(a);
7449 struct ofport_dpif *ofport;
7450
7451 if (type != OVS_ACTION_ATTR_OUTPUT) {
7452 continue;
7453 }
7454
7455 ofport = get_odp_port(ofproto, nl_attr_get_u32(a));
7456 if (ofport && ofport->bundle) {
7457 mirrors |= ofport->bundle->dst_mirrors;
7458 }
7459 }
7460
7461 if (!mirrors) {
7462 return;
7463 }
7464
7465 /* Restore the original packet before adding the mirror actions. */
7466 ctx->xin->flow = *orig_flow;
7467
7468 while (mirrors) {
7469 struct ofmirror *m;
7470
7471 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
7472
7473 if (m->vlans) {
7474 ctx->xout->wc.masks.vlan_tci |= htons(VLAN_CFI | VLAN_VID_MASK);
7475 }
7476
7477 if (!vlan_is_mirrored(m, vlan)) {
7478 mirrors = zero_rightmost_1bit(mirrors);
7479 continue;
7480 }
7481
7482 mirrors &= ~m->dup_mirrors;
7483 ctx->xout->mirrors |= m->dup_mirrors;
7484 if (m->out) {
7485 output_normal(ctx, m->out, vlan);
7486 } else if (vlan != m->out_vlan
7487 && !eth_addr_is_reserved(orig_flow->dl_dst)) {
7488 struct ofbundle *bundle;
7489
7490 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
7491 if (ofbundle_includes_vlan(bundle, m->out_vlan)
7492 && !bundle->mirror_out) {
7493 output_normal(ctx, bundle, m->out_vlan);
7494 }
7495 }
7496 }
7497 }
7498 }
7499
7500 static void
7501 update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors,
7502 uint64_t packets, uint64_t bytes)
7503 {
7504 if (!mirrors) {
7505 return;
7506 }
7507
7508 for (; mirrors; mirrors = zero_rightmost_1bit(mirrors)) {
7509 struct ofmirror *m;
7510
7511 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
7512
7513 if (!m) {
7514 /* In normal circumstances 'm' will not be NULL. However,
7515 * if mirrors are reconfigured, we can temporarily get out
7516 * of sync in facet_revalidate(). We could "correct" the
7517 * mirror list before reaching here, but doing that would
7518 * not properly account the traffic stats we've currently
7519 * accumulated for previous mirror configuration. */
7520 continue;
7521 }
7522
7523 m->packet_count += packets;
7524 m->byte_count += bytes;
7525 }
7526 }
7527
7528 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
7529 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
7530 * indicate this; newer upstream kernels use gratuitous ARP requests. */
7531 static bool
7532 is_gratuitous_arp(const struct flow *flow, struct flow_wildcards *wc)
7533 {
7534 memset(&wc->masks.dl_type, 0xff, sizeof wc->masks.dl_type);
7535 if (flow->dl_type != htons(ETH_TYPE_ARP)) {
7536 return false;
7537 }
7538
7539 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
7540 if (!eth_addr_is_broadcast(flow->dl_dst)) {
7541 return false;
7542 }
7543
7544 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
7545 if (flow->nw_proto == ARP_OP_REPLY) {
7546 return true;
7547 } else if (flow->nw_proto == ARP_OP_REQUEST) {
7548 memset(&wc->masks.nw_src, 0xff, sizeof wc->masks.nw_src);
7549 memset(&wc->masks.nw_dst, 0xff, sizeof wc->masks.nw_dst);
7550
7551 return flow->nw_src == flow->nw_dst;
7552 } else {
7553 return false;
7554 }
7555 }
7556
7557 static void
7558 update_learning_table(struct ofproto_dpif *ofproto,
7559 const struct flow *flow, struct flow_wildcards *wc,
7560 int vlan, struct ofbundle *in_bundle)
7561 {
7562 struct mac_entry *mac;
7563
7564 /* Don't learn the OFPP_NONE port. */
7565 if (in_bundle == &ofpp_none_bundle) {
7566 return;
7567 }
7568
7569 if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
7570 return;
7571 }
7572
7573 mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan);
7574 if (is_gratuitous_arp(flow, wc)) {
7575 /* We don't want to learn from gratuitous ARP packets that are
7576 * reflected back over bond slaves so we lock the learning table. */
7577 if (!in_bundle->bond) {
7578 mac_entry_set_grat_arp_lock(mac);
7579 } else if (mac_entry_is_grat_arp_locked(mac)) {
7580 return;
7581 }
7582 }
7583
7584 if (mac_entry_is_new(mac) || mac->port.p != in_bundle) {
7585 /* The log messages here could actually be useful in debugging,
7586 * so keep the rate limit relatively high. */
7587 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
7588 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
7589 "on port %s in VLAN %d",
7590 ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src),
7591 in_bundle->name, vlan);
7592
7593 mac->port.p = in_bundle;
7594 tag_set_add(&ofproto->backer->revalidate_set,
7595 mac_learning_changed(ofproto->ml, mac));
7596 }
7597 }
7598
7599 static struct ofbundle *
7600 lookup_input_bundle(const struct ofproto_dpif *ofproto, uint16_t in_port,
7601 bool warn, struct ofport_dpif **in_ofportp)
7602 {
7603 struct ofport_dpif *ofport;
7604
7605 /* Find the port and bundle for the received packet. */
7606 ofport = get_ofp_port(ofproto, in_port);
7607 if (in_ofportp) {
7608 *in_ofportp = ofport;
7609 }
7610 if (ofport && ofport->bundle) {
7611 return ofport->bundle;
7612 }
7613
7614 /* Special-case OFPP_NONE, which a controller may use as the ingress
7615 * port for traffic that it is sourcing. */
7616 if (in_port == OFPP_NONE) {
7617 return &ofpp_none_bundle;
7618 }
7619
7620 /* Odd. A few possible reasons here:
7621 *
7622 * - We deleted a port but there are still a few packets queued up
7623 * from it.
7624 *
7625 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
7626 * we don't know about.
7627 *
7628 * - The ofproto client didn't configure the port as part of a bundle.
7629 * This is particularly likely to happen if a packet was received on the
7630 * port after it was created, but before the client had a chance to
7631 * configure its bundle.
7632 */
7633 if (warn) {
7634 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7635
7636 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
7637 "port %"PRIu16, ofproto->up.name, in_port);
7638 }
7639 return NULL;
7640 }
7641
7642 /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or
7643 * dropped. Returns true if they may be forwarded, false if they should be
7644 * dropped.
7645 *
7646 * 'in_port' must be the ofport_dpif that corresponds to flow->in_port.
7647 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
7648 *
7649 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
7650 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
7651 * checked by input_vid_is_valid().
7652 *
7653 * May also add tags to '*tags', although the current implementation only does
7654 * so in one special case.
7655 */
7656 static bool
7657 is_admissible(struct xlate_ctx *ctx, struct ofport_dpif *in_port,
7658 uint16_t vlan)
7659 {
7660 struct ofproto_dpif *ofproto = ctx->ofproto;
7661 struct flow *flow = &ctx->xin->flow;
7662 struct ofbundle *in_bundle = in_port->bundle;
7663
7664 /* Drop frames for reserved multicast addresses
7665 * only if forward_bpdu option is absent. */
7666 if (!ofproto->up.forward_bpdu && eth_addr_is_reserved(flow->dl_dst)) {
7667 xlate_report(ctx, "packet has reserved destination MAC, dropping");
7668 return false;
7669 }
7670
7671 if (in_bundle->bond) {
7672 struct mac_entry *mac;
7673
7674 switch (bond_check_admissibility(in_bundle->bond, in_port,
7675 flow->dl_dst, &ctx->xout->tags)) {
7676 case BV_ACCEPT:
7677 break;
7678
7679 case BV_DROP:
7680 xlate_report(ctx, "bonding refused admissibility, dropping");
7681 return false;
7682
7683 case BV_DROP_IF_MOVED:
7684 mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL);
7685 if (mac && mac->port.p != in_bundle &&
7686 (!is_gratuitous_arp(flow, &ctx->xout->wc)
7687 || mac_entry_is_grat_arp_locked(mac))) {
7688 xlate_report(ctx, "SLB bond thinks this packet looped back, "
7689 "dropping");
7690 return false;
7691 }
7692 break;
7693 }
7694 }
7695
7696 return true;
7697 }
7698
7699 static void
7700 xlate_normal(struct xlate_ctx *ctx)
7701 {
7702 struct ofport_dpif *in_port;
7703 struct ofbundle *in_bundle;
7704 struct mac_entry *mac;
7705 uint16_t vlan;
7706 uint16_t vid;
7707
7708 ctx->xout->has_normal = true;
7709
7710 memset(&ctx->xout->wc.masks.dl_src, 0xff,
7711 sizeof ctx->xout->wc.masks.dl_src);
7712 memset(&ctx->xout->wc.masks.dl_dst, 0xff,
7713 sizeof ctx->xout->wc.masks.dl_dst);
7714 memset(&ctx->xout->wc.masks.vlan_tci, 0xff,
7715 sizeof ctx->xout->wc.masks.vlan_tci);
7716
7717 in_bundle = lookup_input_bundle(ctx->ofproto, ctx->xin->flow.in_port,
7718 ctx->xin->packet != NULL, &in_port);
7719 if (!in_bundle) {
7720 xlate_report(ctx, "no input bundle, dropping");
7721 return;
7722 }
7723
7724 /* Drop malformed frames. */
7725 if (ctx->xin->flow.dl_type == htons(ETH_TYPE_VLAN) &&
7726 !(ctx->xin->flow.vlan_tci & htons(VLAN_CFI))) {
7727 if (ctx->xin->packet != NULL) {
7728 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7729 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
7730 "VLAN tag received on port %s",
7731 ctx->ofproto->up.name, in_bundle->name);
7732 }
7733 xlate_report(ctx, "partial VLAN tag, dropping");
7734 return;
7735 }
7736
7737 /* Drop frames on bundles reserved for mirroring. */
7738 if (in_bundle->mirror_out) {
7739 if (ctx->xin->packet != NULL) {
7740 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7741 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
7742 "%s, which is reserved exclusively for mirroring",
7743 ctx->ofproto->up.name, in_bundle->name);
7744 }
7745 xlate_report(ctx, "input port is mirror output port, dropping");
7746 return;
7747 }
7748
7749 /* Check VLAN. */
7750 vid = vlan_tci_to_vid(ctx->xin->flow.vlan_tci);
7751 if (!input_vid_is_valid(vid, in_bundle, ctx->xin->packet != NULL)) {
7752 xlate_report(ctx, "disallowed VLAN VID for this input port, dropping");
7753 return;
7754 }
7755 vlan = input_vid_to_vlan(in_bundle, vid);
7756
7757 /* Check other admissibility requirements. */
7758 if (in_port && !is_admissible(ctx, in_port, vlan)) {
7759 return;
7760 }
7761
7762 /* Learn source MAC. */
7763 if (ctx->xin->may_learn) {
7764 update_learning_table(ctx->ofproto, &ctx->xin->flow, &ctx->xout->wc,
7765 vlan, in_bundle);
7766 }
7767
7768 /* Determine output bundle. */
7769 mac = mac_learning_lookup(ctx->ofproto->ml, ctx->xin->flow.dl_dst, vlan,
7770 &ctx->xout->tags);
7771 if (mac) {
7772 if (mac->port.p != in_bundle) {
7773 xlate_report(ctx, "forwarding to learned port");
7774 output_normal(ctx, mac->port.p, vlan);
7775 } else {
7776 xlate_report(ctx, "learned port is input port, dropping");
7777 }
7778 } else {
7779 struct ofbundle *bundle;
7780
7781 xlate_report(ctx, "no learned MAC for destination, flooding");
7782 HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
7783 if (bundle != in_bundle
7784 && ofbundle_includes_vlan(bundle, vlan)
7785 && bundle->floodable
7786 && !bundle->mirror_out) {
7787 output_normal(ctx, bundle, vlan);
7788 }
7789 }
7790 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
7791 }
7792 }
7793 \f
7794 /* Optimized flow revalidation.
7795 *
7796 * It's a difficult problem, in general, to tell which facets need to have
7797 * their actions recalculated whenever the OpenFlow flow table changes. We
7798 * don't try to solve that general problem: for most kinds of OpenFlow flow
7799 * table changes, we recalculate the actions for every facet. This is
7800 * relatively expensive, but it's good enough if the OpenFlow flow table
7801 * doesn't change very often.
7802 *
7803 * However, we can expect one particular kind of OpenFlow flow table change to
7804 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
7805 * of CPU on revalidating every facet whenever MAC learning modifies the flow
7806 * table, we add a special case that applies to flow tables in which every rule
7807 * has the same form (that is, the same wildcards), except that the table is
7808 * also allowed to have a single "catch-all" flow that matches all packets. We
7809 * optimize this case by tagging all of the facets that resubmit into the table
7810 * and invalidating the same tag whenever a flow changes in that table. The
7811 * end result is that we revalidate just the facets that need it (and sometimes
7812 * a few more, but not all of the facets or even all of the facets that
7813 * resubmit to the table modified by MAC learning). */
7814
7815 /* Calculates the tag to use for 'flow' and mask 'mask' when it is inserted
7816 * into an OpenFlow table with the given 'basis'. */
7817 static tag_type
7818 rule_calculate_tag(const struct flow *flow, const struct minimask *mask,
7819 uint32_t secret)
7820 {
7821 if (minimask_is_catchall(mask)) {
7822 return 0;
7823 } else {
7824 uint32_t hash = flow_hash_in_minimask(flow, mask, secret);
7825 return tag_create_deterministic(hash);
7826 }
7827 }
7828
7829 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
7830 * taggability of that table.
7831 *
7832 * This function must be called after *each* change to a flow table. If you
7833 * skip calling it on some changes then the pointer comparisons at the end can
7834 * be invalid if you get unlucky. For example, if a flow removal causes a
7835 * cls_table to be destroyed and then a flow insertion causes a cls_table with
7836 * different wildcards to be created with the same address, then this function
7837 * will incorrectly skip revalidation. */
7838 static void
7839 table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
7840 {
7841 struct table_dpif *table = &ofproto->tables[table_id];
7842 const struct oftable *oftable = &ofproto->up.tables[table_id];
7843 struct cls_table *catchall, *other;
7844 struct cls_table *t;
7845
7846 catchall = other = NULL;
7847
7848 switch (hmap_count(&oftable->cls.tables)) {
7849 case 0:
7850 /* We could tag this OpenFlow table but it would make the logic a
7851 * little harder and it's a corner case that doesn't seem worth it
7852 * yet. */
7853 break;
7854
7855 case 1:
7856 case 2:
7857 HMAP_FOR_EACH (t, hmap_node, &oftable->cls.tables) {
7858 if (cls_table_is_catchall(t)) {
7859 catchall = t;
7860 } else if (!other) {
7861 other = t;
7862 } else {
7863 /* Indicate that we can't tag this by setting both tables to
7864 * NULL. (We know that 'catchall' is already NULL.) */
7865 other = NULL;
7866 }
7867 }
7868 break;
7869
7870 default:
7871 /* Can't tag this table. */
7872 break;
7873 }
7874
7875 if (table->catchall_table != catchall || table->other_table != other) {
7876 table->catchall_table = catchall;
7877 table->other_table = other;
7878 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
7879 }
7880 }
7881
7882 /* Given 'rule' that has changed in some way (either it is a rule being
7883 * inserted, a rule being deleted, or a rule whose actions are being
7884 * modified), marks facets for revalidation to ensure that packets will be
7885 * forwarded correctly according to the new state of the flow table.
7886 *
7887 * This function must be called after *each* change to a flow table. See
7888 * the comment on table_update_taggable() for more information. */
7889 static void
7890 rule_invalidate(const struct rule_dpif *rule)
7891 {
7892 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
7893
7894 table_update_taggable(ofproto, rule->up.table_id);
7895
7896 if (!ofproto->backer->need_revalidate) {
7897 struct table_dpif *table = &ofproto->tables[rule->up.table_id];
7898
7899 if (table->other_table && rule->tag) {
7900 tag_set_add(&ofproto->backer->revalidate_set, rule->tag);
7901 } else {
7902 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
7903 }
7904 }
7905 }
7906 \f
7907 static bool
7908 set_frag_handling(struct ofproto *ofproto_,
7909 enum ofp_config_flags frag_handling)
7910 {
7911 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7912 if (frag_handling != OFPC_FRAG_REASM) {
7913 ofproto->backer->need_revalidate = REV_RECONFIGURE;
7914 return true;
7915 } else {
7916 return false;
7917 }
7918 }
7919
7920 static enum ofperr
7921 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
7922 const struct flow *flow,
7923 const struct ofpact *ofpacts, size_t ofpacts_len)
7924 {
7925 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7926 struct initial_vals initial_vals;
7927 struct odputil_keybuf keybuf;
7928 struct dpif_flow_stats stats;
7929 struct xlate_out xout;
7930 struct xlate_in xin;
7931 struct ofpbuf key;
7932
7933
7934 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
7935 odp_flow_key_from_flow(&key, flow,
7936 ofp_port_to_odp_port(ofproto, flow->in_port));
7937
7938 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
7939
7940 initial_vals.vlan_tci = flow->vlan_tci;
7941 xlate_in_init(&xin, ofproto, flow, &initial_vals, NULL, stats.tcp_flags,
7942 packet);
7943 xin.resubmit_stats = &stats;
7944 xin.ofpacts_len = ofpacts_len;
7945 xin.ofpacts = ofpacts;
7946
7947 xlate_actions(&xin, &xout);
7948 dpif_execute(ofproto->backer->dpif, key.data, key.size,
7949 xout.odp_actions.data, xout.odp_actions.size, packet);
7950 xlate_out_uninit(&xout);
7951
7952 return 0;
7953 }
7954 \f
7955 /* NetFlow. */
7956
7957 static int
7958 set_netflow(struct ofproto *ofproto_,
7959 const struct netflow_options *netflow_options)
7960 {
7961 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7962
7963 if (netflow_options) {
7964 if (!ofproto->netflow) {
7965 ofproto->netflow = netflow_create();
7966 }
7967 return netflow_set_options(ofproto->netflow, netflow_options);
7968 } else {
7969 netflow_destroy(ofproto->netflow);
7970 ofproto->netflow = NULL;
7971 return 0;
7972 }
7973 }
7974
7975 static void
7976 get_netflow_ids(const struct ofproto *ofproto_,
7977 uint8_t *engine_type, uint8_t *engine_id)
7978 {
7979 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7980
7981 dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
7982 }
7983
7984 static void
7985 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
7986 {
7987 if (!facet_is_controller_flow(facet) &&
7988 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
7989 struct subfacet *subfacet;
7990 struct ofexpired expired;
7991
7992 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
7993 if (subfacet->path == SF_FAST_PATH) {
7994 struct dpif_flow_stats stats;
7995
7996 subfacet_install(subfacet, &facet->xout.odp_actions,
7997 &stats);
7998 subfacet_update_stats(subfacet, &stats);
7999 }
8000 }
8001
8002 expired.flow = facet->flow;
8003 expired.packet_count = facet->packet_count;
8004 expired.byte_count = facet->byte_count;
8005 expired.used = facet->used;
8006 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
8007 }
8008 }
8009
8010 static void
8011 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
8012 {
8013 struct cls_cursor cursor;
8014 struct facet *facet;
8015
8016 cls_cursor_init(&cursor, &ofproto->facets, NULL);
8017 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
8018 send_active_timeout(ofproto, facet);
8019 }
8020 }
8021 \f
8022 static struct ofproto_dpif *
8023 ofproto_dpif_lookup(const char *name)
8024 {
8025 struct ofproto_dpif *ofproto;
8026
8027 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
8028 hash_string(name, 0), &all_ofproto_dpifs) {
8029 if (!strcmp(ofproto->up.name, name)) {
8030 return ofproto;
8031 }
8032 }
8033 return NULL;
8034 }
8035
8036 static void
8037 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
8038 const char *argv[], void *aux OVS_UNUSED)
8039 {
8040 struct ofproto_dpif *ofproto;
8041
8042 if (argc > 1) {
8043 ofproto = ofproto_dpif_lookup(argv[1]);
8044 if (!ofproto) {
8045 unixctl_command_reply_error(conn, "no such bridge");
8046 return;
8047 }
8048 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
8049 } else {
8050 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
8051 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
8052 }
8053 }
8054
8055 unixctl_command_reply(conn, "table successfully flushed");
8056 }
8057
8058 static void
8059 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
8060 const char *argv[], void *aux OVS_UNUSED)
8061 {
8062 struct ds ds = DS_EMPTY_INITIALIZER;
8063 const struct ofproto_dpif *ofproto;
8064 const struct mac_entry *e;
8065
8066 ofproto = ofproto_dpif_lookup(argv[1]);
8067 if (!ofproto) {
8068 unixctl_command_reply_error(conn, "no such bridge");
8069 return;
8070 }
8071
8072 ds_put_cstr(&ds, " port VLAN MAC Age\n");
8073 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
8074 struct ofbundle *bundle = e->port.p;
8075 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
8076 ofbundle_get_a_port(bundle)->odp_port,
8077 e->vlan, ETH_ADDR_ARGS(e->mac),
8078 mac_entry_age(ofproto->ml, e));
8079 }
8080 unixctl_command_reply(conn, ds_cstr(&ds));
8081 ds_destroy(&ds);
8082 }
8083
8084 struct trace_ctx {
8085 struct xlate_out xout;
8086 struct xlate_in xin;
8087 struct flow flow;
8088 struct ds *result;
8089 };
8090
8091 static void
8092 trace_format_rule(struct ds *result, uint8_t table_id, int level,
8093 const struct rule_dpif *rule)
8094 {
8095 ds_put_char_multiple(result, '\t', level);
8096 if (!rule) {
8097 ds_put_cstr(result, "No match\n");
8098 return;
8099 }
8100
8101 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
8102 table_id, ntohll(rule->up.flow_cookie));
8103 cls_rule_format(&rule->up.cr, result);
8104 ds_put_char(result, '\n');
8105
8106 ds_put_char_multiple(result, '\t', level);
8107 ds_put_cstr(result, "OpenFlow ");
8108 ofpacts_format(rule->up.ofpacts, rule->up.ofpacts_len, result);
8109 ds_put_char(result, '\n');
8110 }
8111
8112 static void
8113 trace_format_flow(struct ds *result, int level, const char *title,
8114 struct trace_ctx *trace)
8115 {
8116 ds_put_char_multiple(result, '\t', level);
8117 ds_put_format(result, "%s: ", title);
8118 if (flow_equal(&trace->xin.flow, &trace->flow)) {
8119 ds_put_cstr(result, "unchanged");
8120 } else {
8121 flow_format(result, &trace->xin.flow);
8122 trace->flow = trace->xin.flow;
8123 }
8124 ds_put_char(result, '\n');
8125 }
8126
8127 static void
8128 trace_format_regs(struct ds *result, int level, const char *title,
8129 struct trace_ctx *trace)
8130 {
8131 size_t i;
8132
8133 ds_put_char_multiple(result, '\t', level);
8134 ds_put_format(result, "%s:", title);
8135 for (i = 0; i < FLOW_N_REGS; i++) {
8136 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
8137 }
8138 ds_put_char(result, '\n');
8139 }
8140
8141 static void
8142 trace_format_odp(struct ds *result, int level, const char *title,
8143 struct trace_ctx *trace)
8144 {
8145 struct ofpbuf *odp_actions = &trace->xout.odp_actions;
8146
8147 ds_put_char_multiple(result, '\t', level);
8148 ds_put_format(result, "%s: ", title);
8149 format_odp_actions(result, odp_actions->data, odp_actions->size);
8150 ds_put_char(result, '\n');
8151 }
8152
8153 static void
8154 trace_resubmit(struct xlate_ctx *ctx, struct rule_dpif *rule)
8155 {
8156 struct trace_ctx *trace = CONTAINER_OF(ctx->xin, struct trace_ctx, xin);
8157 struct ds *result = trace->result;
8158
8159 ds_put_char(result, '\n');
8160 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
8161 trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace);
8162 trace_format_odp(result, ctx->recurse + 1, "Resubmitted odp", trace);
8163 trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule);
8164 }
8165
8166 static void
8167 trace_report(struct xlate_ctx *ctx, const char *s)
8168 {
8169 struct trace_ctx *trace = CONTAINER_OF(ctx->xin, struct trace_ctx, xin);
8170 struct ds *result = trace->result;
8171
8172 ds_put_char_multiple(result, '\t', ctx->recurse);
8173 ds_put_cstr(result, s);
8174 ds_put_char(result, '\n');
8175 }
8176
8177 static void
8178 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
8179 void *aux OVS_UNUSED)
8180 {
8181 const struct dpif_backer *backer;
8182 struct ofproto_dpif *ofproto;
8183 struct ofpbuf odp_key;
8184 struct ofpbuf *packet;
8185 struct initial_vals initial_vals;
8186 struct ds result;
8187 struct flow flow;
8188 char *s;
8189
8190 packet = NULL;
8191 backer = NULL;
8192 ds_init(&result);
8193 ofpbuf_init(&odp_key, 0);
8194
8195 /* Handle "-generate" or a hex string as the last argument. */
8196 if (!strcmp(argv[argc - 1], "-generate")) {
8197 packet = ofpbuf_new(0);
8198 argc--;
8199 } else {
8200 const char *error = eth_from_hex(argv[argc - 1], &packet);
8201 if (!error) {
8202 argc--;
8203 } else if (argc == 4) {
8204 /* The 3-argument form must end in "-generate' or a hex string. */
8205 unixctl_command_reply_error(conn, error);
8206 goto exit;
8207 }
8208 }
8209
8210 /* Parse the flow and determine whether a datapath or
8211 * bridge is specified. If function odp_flow_key_from_string()
8212 * returns 0, the flow is a odp_flow. If function
8213 * parse_ofp_exact_flow() returns 0, the flow is a br_flow. */
8214 if (!odp_flow_key_from_string(argv[argc - 1], NULL, &odp_key)) {
8215 /* If the odp_flow is the second argument,
8216 * the datapath name is the first argument. */
8217 if (argc == 3) {
8218 const char *dp_type;
8219 if (!strncmp(argv[1], "ovs-", 4)) {
8220 dp_type = argv[1] + 4;
8221 } else {
8222 dp_type = argv[1];
8223 }
8224 backer = shash_find_data(&all_dpif_backers, dp_type);
8225 if (!backer) {
8226 unixctl_command_reply_error(conn, "Cannot find datapath "
8227 "of this name");
8228 goto exit;
8229 }
8230 } else {
8231 /* No datapath name specified, so there should be only one
8232 * datapath. */
8233 struct shash_node *node;
8234 if (shash_count(&all_dpif_backers) != 1) {
8235 unixctl_command_reply_error(conn, "Must specify datapath "
8236 "name, there is more than one type of datapath");
8237 goto exit;
8238 }
8239 node = shash_first(&all_dpif_backers);
8240 backer = node->data;
8241 }
8242
8243 /* Extract the ofproto_dpif object from the ofproto_receive()
8244 * function. */
8245 if (ofproto_receive(backer, NULL, odp_key.data,
8246 odp_key.size, &flow, NULL, &ofproto, NULL,
8247 &initial_vals)) {
8248 unixctl_command_reply_error(conn, "Invalid datapath flow");
8249 goto exit;
8250 }
8251 ds_put_format(&result, "Bridge: %s\n", ofproto->up.name);
8252 } else if (!parse_ofp_exact_flow(&flow, argv[argc - 1])) {
8253 if (argc != 3) {
8254 unixctl_command_reply_error(conn, "Must specify bridge name");
8255 goto exit;
8256 }
8257
8258 ofproto = ofproto_dpif_lookup(argv[1]);
8259 if (!ofproto) {
8260 unixctl_command_reply_error(conn, "Unknown bridge name");
8261 goto exit;
8262 }
8263 initial_vals.vlan_tci = flow.vlan_tci;
8264 } else {
8265 unixctl_command_reply_error(conn, "Bad flow syntax");
8266 goto exit;
8267 }
8268
8269 /* Generate a packet, if requested. */
8270 if (packet) {
8271 if (!packet->size) {
8272 flow_compose(packet, &flow);
8273 } else {
8274 ds_put_cstr(&result, "Packet: ");
8275 s = ofp_packet_to_string(packet->data, packet->size);
8276 ds_put_cstr(&result, s);
8277 free(s);
8278
8279 /* Use the metadata from the flow and the packet argument
8280 * to reconstruct the flow. */
8281 flow_extract(packet, flow.skb_priority, flow.skb_mark, NULL,
8282 flow.in_port, &flow);
8283 initial_vals.vlan_tci = flow.vlan_tci;
8284 }
8285 }
8286
8287 ofproto_trace(ofproto, &flow, packet, &initial_vals, &result);
8288 unixctl_command_reply(conn, ds_cstr(&result));
8289
8290 exit:
8291 ds_destroy(&result);
8292 ofpbuf_delete(packet);
8293 ofpbuf_uninit(&odp_key);
8294 }
8295
8296 static void
8297 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
8298 const struct ofpbuf *packet,
8299 const struct initial_vals *initial_vals, struct ds *ds)
8300 {
8301 struct rule_dpif *rule;
8302
8303 ds_put_cstr(ds, "Flow: ");
8304 flow_format(ds, flow);
8305 ds_put_char(ds, '\n');
8306
8307 rule = rule_dpif_lookup(ofproto, flow, NULL);
8308
8309 trace_format_rule(ds, 0, 0, rule);
8310 if (rule == ofproto->miss_rule) {
8311 ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
8312 } else if (rule == ofproto->no_packet_in_rule) {
8313 ds_put_cstr(ds, "\nNo match, packets dropped because "
8314 "OFPPC_NO_PACKET_IN is set on in_port.\n");
8315 } else if (rule == ofproto->drop_frags_rule) {
8316 ds_put_cstr(ds, "\nPackets dropped because they are IP fragments "
8317 "and the fragment handling mode is \"drop\".\n");
8318 }
8319
8320 if (rule) {
8321 uint64_t odp_actions_stub[1024 / 8];
8322 struct ofpbuf odp_actions;
8323 struct trace_ctx trace;
8324 struct match match;
8325 uint8_t tcp_flags;
8326
8327 tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
8328 trace.result = ds;
8329 trace.flow = *flow;
8330 ofpbuf_use_stub(&odp_actions,
8331 odp_actions_stub, sizeof odp_actions_stub);
8332 xlate_in_init(&trace.xin, ofproto, flow, initial_vals, rule, tcp_flags,
8333 packet);
8334 trace.xin.resubmit_hook = trace_resubmit;
8335 trace.xin.report_hook = trace_report;
8336
8337 xlate_actions(&trace.xin, &trace.xout);
8338
8339 ds_put_char(ds, '\n');
8340 trace_format_flow(ds, 0, "Final flow", &trace);
8341
8342 match_init(&match, flow, &trace.xout.wc);
8343 ds_put_cstr(ds, "Relevant fields: ");
8344 match_format(&match, ds, OFP_DEFAULT_PRIORITY);
8345 ds_put_char(ds, '\n');
8346
8347 ds_put_cstr(ds, "Datapath actions: ");
8348 format_odp_actions(ds, trace.xout.odp_actions.data,
8349 trace.xout.odp_actions.size);
8350
8351 if (trace.xout.slow) {
8352 ds_put_cstr(ds, "\nThis flow is handled by the userspace "
8353 "slow path because it:");
8354 switch (trace.xout.slow) {
8355 case SLOW_CFM:
8356 ds_put_cstr(ds, "\n\t- Consists of CFM packets.");
8357 break;
8358 case SLOW_LACP:
8359 ds_put_cstr(ds, "\n\t- Consists of LACP packets.");
8360 break;
8361 case SLOW_STP:
8362 ds_put_cstr(ds, "\n\t- Consists of STP packets.");
8363 break;
8364 case SLOW_BFD:
8365 ds_put_cstr(ds, "\n\t- Consists of BFD packets.");
8366 break;
8367 case SLOW_CONTROLLER:
8368 ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages "
8369 "to the OpenFlow controller.");
8370 break;
8371 case __SLOW_MAX:
8372 NOT_REACHED();
8373 }
8374 }
8375
8376 xlate_out_uninit(&trace.xout);
8377 }
8378 }
8379
8380 static void
8381 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
8382 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
8383 {
8384 clogged = true;
8385 unixctl_command_reply(conn, NULL);
8386 }
8387
8388 static void
8389 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
8390 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
8391 {
8392 clogged = false;
8393 unixctl_command_reply(conn, NULL);
8394 }
8395
8396 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
8397 * 'reply' describing the results. */
8398 static void
8399 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
8400 {
8401 struct cls_cursor cursor;
8402 struct facet *facet;
8403 int errors;
8404
8405 errors = 0;
8406 cls_cursor_init(&cursor, &ofproto->facets, NULL);
8407 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
8408 if (!facet_check_consistency(facet)) {
8409 errors++;
8410 }
8411 }
8412 if (errors) {
8413 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
8414 }
8415
8416 if (errors) {
8417 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
8418 ofproto->up.name, errors);
8419 } else {
8420 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
8421 }
8422 }
8423
8424 static void
8425 ofproto_dpif_self_check(struct unixctl_conn *conn,
8426 int argc, const char *argv[], void *aux OVS_UNUSED)
8427 {
8428 struct ds reply = DS_EMPTY_INITIALIZER;
8429 struct ofproto_dpif *ofproto;
8430
8431 if (argc > 1) {
8432 ofproto = ofproto_dpif_lookup(argv[1]);
8433 if (!ofproto) {
8434 unixctl_command_reply_error(conn, "Unknown ofproto (use "
8435 "ofproto/list for help)");
8436 return;
8437 }
8438 ofproto_dpif_self_check__(ofproto, &reply);
8439 } else {
8440 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
8441 ofproto_dpif_self_check__(ofproto, &reply);
8442 }
8443 }
8444
8445 unixctl_command_reply(conn, ds_cstr(&reply));
8446 ds_destroy(&reply);
8447 }
8448
8449 /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list
8450 * of the 'ofproto_shash' nodes. It is the responsibility of the caller
8451 * to destroy 'ofproto_shash' and free the returned value. */
8452 static const struct shash_node **
8453 get_ofprotos(struct shash *ofproto_shash)
8454 {
8455 const struct ofproto_dpif *ofproto;
8456
8457 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
8458 char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
8459 shash_add_nocopy(ofproto_shash, name, ofproto);
8460 }
8461
8462 return shash_sort(ofproto_shash);
8463 }
8464
8465 static void
8466 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
8467 const char *argv[] OVS_UNUSED,
8468 void *aux OVS_UNUSED)
8469 {
8470 struct ds ds = DS_EMPTY_INITIALIZER;
8471 struct shash ofproto_shash;
8472 const struct shash_node **sorted_ofprotos;
8473 int i;
8474
8475 shash_init(&ofproto_shash);
8476 sorted_ofprotos = get_ofprotos(&ofproto_shash);
8477 for (i = 0; i < shash_count(&ofproto_shash); i++) {
8478 const struct shash_node *node = sorted_ofprotos[i];
8479 ds_put_format(&ds, "%s\n", node->name);
8480 }
8481
8482 shash_destroy(&ofproto_shash);
8483 free(sorted_ofprotos);
8484
8485 unixctl_command_reply(conn, ds_cstr(&ds));
8486 ds_destroy(&ds);
8487 }
8488
8489 static void
8490 show_dp_rates(struct ds *ds, const char *heading,
8491 const struct avg_subfacet_rates *rates)
8492 {
8493 ds_put_format(ds, "%s add rate: %5.3f/min, del rate: %5.3f/min\n",
8494 heading, rates->add_rate, rates->del_rate);
8495 }
8496
8497 static void
8498 dpif_show_backer(const struct dpif_backer *backer, struct ds *ds)
8499 {
8500 const struct shash_node **ofprotos;
8501 struct ofproto_dpif *ofproto;
8502 struct shash ofproto_shash;
8503 uint64_t n_hit, n_missed;
8504 long long int minutes;
8505 size_t i;
8506
8507 n_hit = n_missed = 0;
8508 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
8509 if (ofproto->backer == backer) {
8510 n_missed += ofproto->n_missed;
8511 n_hit += ofproto->n_hit;
8512 }
8513 }
8514
8515 ds_put_format(ds, "%s: hit:%"PRIu64" missed:%"PRIu64"\n",
8516 dpif_name(backer->dpif), n_hit, n_missed);
8517 ds_put_format(ds, "\tflows: cur: %zu, avg: %u, max: %u,"
8518 " life span: %lldms\n", hmap_count(&backer->subfacets),
8519 backer->avg_n_subfacet, backer->max_n_subfacet,
8520 backer->avg_subfacet_life);
8521
8522 minutes = (time_msec() - backer->created) / (1000 * 60);
8523 if (minutes >= 60) {
8524 show_dp_rates(ds, "\thourly avg:", &backer->hourly);
8525 }
8526 if (minutes >= 60 * 24) {
8527 show_dp_rates(ds, "\tdaily avg:", &backer->daily);
8528 }
8529 show_dp_rates(ds, "\toverall avg:", &backer->lifetime);
8530
8531 shash_init(&ofproto_shash);
8532 ofprotos = get_ofprotos(&ofproto_shash);
8533 for (i = 0; i < shash_count(&ofproto_shash); i++) {
8534 struct ofproto_dpif *ofproto = ofprotos[i]->data;
8535 const struct shash_node **ports;
8536 size_t j;
8537
8538 if (ofproto->backer != backer) {
8539 continue;
8540 }
8541
8542 ds_put_format(ds, "\t%s: hit:%"PRIu64" missed:%"PRIu64"\n",
8543 ofproto->up.name, ofproto->n_hit, ofproto->n_missed);
8544
8545 ports = shash_sort(&ofproto->up.port_by_name);
8546 for (j = 0; j < shash_count(&ofproto->up.port_by_name); j++) {
8547 const struct shash_node *node = ports[j];
8548 struct ofport *ofport = node->data;
8549 struct smap config;
8550 uint32_t odp_port;
8551
8552 ds_put_format(ds, "\t\t%s %u/", netdev_get_name(ofport->netdev),
8553 ofport->ofp_port);
8554
8555 odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
8556 if (odp_port != OVSP_NONE) {
8557 ds_put_format(ds, "%"PRIu32":", odp_port);
8558 } else {
8559 ds_put_cstr(ds, "none:");
8560 }
8561
8562 ds_put_format(ds, " (%s", netdev_get_type(ofport->netdev));
8563
8564 smap_init(&config);
8565 if (!netdev_get_config(ofport->netdev, &config)) {
8566 const struct smap_node **nodes;
8567 size_t i;
8568
8569 nodes = smap_sort(&config);
8570 for (i = 0; i < smap_count(&config); i++) {
8571 const struct smap_node *node = nodes[i];
8572 ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
8573 node->key, node->value);
8574 }
8575 free(nodes);
8576 }
8577 smap_destroy(&config);
8578
8579 ds_put_char(ds, ')');
8580 ds_put_char(ds, '\n');
8581 }
8582 free(ports);
8583 }
8584 shash_destroy(&ofproto_shash);
8585 free(ofprotos);
8586 }
8587
8588 static void
8589 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
8590 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
8591 {
8592 struct ds ds = DS_EMPTY_INITIALIZER;
8593 const struct shash_node **backers;
8594 int i;
8595
8596 backers = shash_sort(&all_dpif_backers);
8597 for (i = 0; i < shash_count(&all_dpif_backers); i++) {
8598 dpif_show_backer(backers[i]->data, &ds);
8599 }
8600 free(backers);
8601
8602 unixctl_command_reply(conn, ds_cstr(&ds));
8603 ds_destroy(&ds);
8604 }
8605
8606 /* Dump the megaflow (facet) cache. This is useful to check the
8607 * correctness of flow wildcarding, since the same mechanism is used for
8608 * both xlate caching and kernel wildcarding.
8609 *
8610 * It's important to note that in the output the flow description uses
8611 * OpenFlow (OFP) ports, but the actions use datapath (ODP) ports.
8612 *
8613 * This command is only needed for advanced debugging, so it's not
8614 * documented in the man page. */
8615 static void
8616 ofproto_unixctl_dpif_dump_megaflows(struct unixctl_conn *conn,
8617 int argc OVS_UNUSED, const char *argv[],
8618 void *aux OVS_UNUSED)
8619 {
8620 struct ds ds = DS_EMPTY_INITIALIZER;
8621 const struct ofproto_dpif *ofproto;
8622 long long int now = time_msec();
8623 struct cls_cursor cursor;
8624 struct facet *facet;
8625
8626 ofproto = ofproto_dpif_lookup(argv[1]);
8627 if (!ofproto) {
8628 unixctl_command_reply_error(conn, "no such bridge");
8629 return;
8630 }
8631
8632 cls_cursor_init(&cursor, &ofproto->facets, NULL);
8633 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
8634 cls_rule_format(&facet->cr, &ds);
8635 ds_put_cstr(&ds, ", ");
8636 ds_put_format(&ds, "n_subfacets:%"PRIu64", ",
8637 list_size(&facet->subfacets));
8638 ds_put_format(&ds, "used:%.3fs, ", (now - facet->used) / 1000.0);
8639 ds_put_cstr(&ds, "Datapath actions: ");
8640 format_odp_actions(&ds, facet->xout.odp_actions.data,
8641 facet->xout.odp_actions.size);
8642 ds_put_cstr(&ds, "\n");
8643 }
8644
8645 ds_chomp(&ds, '\n');
8646 unixctl_command_reply(conn, ds_cstr(&ds));
8647 ds_destroy(&ds);
8648 }
8649
8650 static void
8651 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
8652 int argc OVS_UNUSED, const char *argv[],
8653 void *aux OVS_UNUSED)
8654 {
8655 struct ds ds = DS_EMPTY_INITIALIZER;
8656 const struct ofproto_dpif *ofproto;
8657 struct subfacet *subfacet;
8658
8659 ofproto = ofproto_dpif_lookup(argv[1]);
8660 if (!ofproto) {
8661 unixctl_command_reply_error(conn, "no such bridge");
8662 return;
8663 }
8664
8665 update_stats(ofproto->backer);
8666
8667 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->backer->subfacets) {
8668 struct facet *facet = subfacet->facet;
8669
8670 if (ofproto_dpif_cast(facet->rule->up.ofproto) != ofproto) {
8671 continue;
8672 }
8673
8674 odp_flow_key_format(subfacet->key, subfacet->key_len, &ds);
8675
8676 ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:",
8677 subfacet->dp_packet_count, subfacet->dp_byte_count);
8678 if (subfacet->used) {
8679 ds_put_format(&ds, "%.3fs",
8680 (time_msec() - subfacet->used) / 1000.0);
8681 } else {
8682 ds_put_format(&ds, "never");
8683 }
8684 if (subfacet->facet->tcp_flags) {
8685 ds_put_cstr(&ds, ", flags:");
8686 packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags);
8687 }
8688
8689 ds_put_cstr(&ds, ", actions:");
8690 if (facet->xout.slow) {
8691 uint64_t slow_path_stub[128 / 8];
8692 const struct nlattr *actions;
8693 size_t actions_len;
8694
8695 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
8696 slow_path_stub, sizeof slow_path_stub,
8697 &actions, &actions_len);
8698 format_odp_actions(&ds, actions, actions_len);
8699 } else {
8700 format_odp_actions(&ds, facet->xout.odp_actions.data,
8701 facet->xout.odp_actions.size);
8702 }
8703 ds_put_char(&ds, '\n');
8704 }
8705
8706 unixctl_command_reply(conn, ds_cstr(&ds));
8707 ds_destroy(&ds);
8708 }
8709
8710 static void
8711 ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn,
8712 int argc OVS_UNUSED, const char *argv[],
8713 void *aux OVS_UNUSED)
8714 {
8715 struct ds ds = DS_EMPTY_INITIALIZER;
8716 struct ofproto_dpif *ofproto;
8717
8718 ofproto = ofproto_dpif_lookup(argv[1]);
8719 if (!ofproto) {
8720 unixctl_command_reply_error(conn, "no such bridge");
8721 return;
8722 }
8723
8724 flush(&ofproto->up);
8725
8726 unixctl_command_reply(conn, ds_cstr(&ds));
8727 ds_destroy(&ds);
8728 }
8729
8730 static void
8731 ofproto_dpif_unixctl_init(void)
8732 {
8733 static bool registered;
8734 if (registered) {
8735 return;
8736 }
8737 registered = true;
8738
8739 unixctl_command_register(
8740 "ofproto/trace",
8741 "[dp_name]|bridge odp_flow|br_flow [-generate|packet]",
8742 1, 3, ofproto_unixctl_trace, NULL);
8743 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
8744 ofproto_unixctl_fdb_flush, NULL);
8745 unixctl_command_register("fdb/show", "bridge", 1, 1,
8746 ofproto_unixctl_fdb_show, NULL);
8747 unixctl_command_register("ofproto/clog", "", 0, 0,
8748 ofproto_dpif_clog, NULL);
8749 unixctl_command_register("ofproto/unclog", "", 0, 0,
8750 ofproto_dpif_unclog, NULL);
8751 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
8752 ofproto_dpif_self_check, NULL);
8753 unixctl_command_register("dpif/dump-dps", "", 0, 0,
8754 ofproto_unixctl_dpif_dump_dps, NULL);
8755 unixctl_command_register("dpif/show", "", 0, 0, ofproto_unixctl_dpif_show,
8756 NULL);
8757 unixctl_command_register("dpif/dump-flows", "bridge", 1, 1,
8758 ofproto_unixctl_dpif_dump_flows, NULL);
8759 unixctl_command_register("dpif/del-flows", "bridge", 1, 1,
8760 ofproto_unixctl_dpif_del_flows, NULL);
8761 unixctl_command_register("dpif/dump-megaflows", "bridge", 1, 1,
8762 ofproto_unixctl_dpif_dump_megaflows, NULL);
8763 }
8764 \f
8765 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
8766 *
8767 * This is deprecated. It is only for compatibility with broken device drivers
8768 * in old versions of Linux that do not properly support VLANs when VLAN
8769 * devices are not used. When broken device drivers are no longer in
8770 * widespread use, we will delete these interfaces. */
8771
8772 static int
8773 set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid)
8774 {
8775 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
8776 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
8777
8778 if (realdev_ofp_port == ofport->realdev_ofp_port
8779 && vid == ofport->vlandev_vid) {
8780 return 0;
8781 }
8782
8783 ofproto->backer->need_revalidate = REV_RECONFIGURE;
8784
8785 if (ofport->realdev_ofp_port) {
8786 vsp_remove(ofport);
8787 }
8788 if (realdev_ofp_port && ofport->bundle) {
8789 /* vlandevs are enslaved to their realdevs, so they are not allowed to
8790 * themselves be part of a bundle. */
8791 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
8792 }
8793
8794 ofport->realdev_ofp_port = realdev_ofp_port;
8795 ofport->vlandev_vid = vid;
8796
8797 if (realdev_ofp_port) {
8798 vsp_add(ofport, realdev_ofp_port, vid);
8799 }
8800
8801 return 0;
8802 }
8803
8804 static uint32_t
8805 hash_realdev_vid(uint16_t realdev_ofp_port, int vid)
8806 {
8807 return hash_2words(realdev_ofp_port, vid);
8808 }
8809
8810 /* Returns the OFP port number of the Linux VLAN device that corresponds to
8811 * 'vlan_tci' on the network device with port number 'realdev_ofp_port' in
8812 * 'struct ofport_dpif'. For example, given 'realdev_ofp_port' of eth0 and
8813 * 'vlan_tci' 9, it would return the port number of eth0.9.
8814 *
8815 * Unless VLAN splinters are enabled for port 'realdev_ofp_port', this
8816 * function just returns its 'realdev_ofp_port' argument. */
8817 static uint16_t
8818 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
8819 uint16_t realdev_ofp_port, ovs_be16 vlan_tci)
8820 {
8821 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
8822 int vid = vlan_tci_to_vid(vlan_tci);
8823 const struct vlan_splinter *vsp;
8824
8825 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
8826 hash_realdev_vid(realdev_ofp_port, vid),
8827 &ofproto->realdev_vid_map) {
8828 if (vsp->realdev_ofp_port == realdev_ofp_port
8829 && vsp->vid == vid) {
8830 return vsp->vlandev_ofp_port;
8831 }
8832 }
8833 }
8834 return realdev_ofp_port;
8835 }
8836
8837 static struct vlan_splinter *
8838 vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port)
8839 {
8840 struct vlan_splinter *vsp;
8841
8842 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, hash_int(vlandev_ofp_port, 0),
8843 &ofproto->vlandev_map) {
8844 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
8845 return vsp;
8846 }
8847 }
8848
8849 return NULL;
8850 }
8851
8852 /* Returns the OpenFlow port number of the "real" device underlying the Linux
8853 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
8854 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
8855 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
8856 * eth0 and store 9 in '*vid'.
8857 *
8858 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
8859 * VLAN device. Unless VLAN splinters are enabled, this is what this function
8860 * always does.*/
8861 static uint16_t
8862 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
8863 uint16_t vlandev_ofp_port, int *vid)
8864 {
8865 if (!hmap_is_empty(&ofproto->vlandev_map)) {
8866 const struct vlan_splinter *vsp;
8867
8868 vsp = vlandev_find(ofproto, vlandev_ofp_port);
8869 if (vsp) {
8870 if (vid) {
8871 *vid = vsp->vid;
8872 }
8873 return vsp->realdev_ofp_port;
8874 }
8875 }
8876 return 0;
8877 }
8878
8879 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
8880 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
8881 * 'flow->in_port' to the "real" device backing the VLAN device, sets
8882 * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is
8883 * always the case unless VLAN splinters are enabled), returns false without
8884 * making any changes. */
8885 static bool
8886 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
8887 {
8888 uint16_t realdev;
8889 int vid;
8890
8891 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port, &vid);
8892 if (!realdev) {
8893 return false;
8894 }
8895
8896 /* Cause the flow to be processed as if it came in on the real device with
8897 * the VLAN device's VLAN ID. */
8898 flow->in_port = realdev;
8899 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
8900 return true;
8901 }
8902
8903 static void
8904 vsp_remove(struct ofport_dpif *port)
8905 {
8906 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
8907 struct vlan_splinter *vsp;
8908
8909 vsp = vlandev_find(ofproto, port->up.ofp_port);
8910 if (vsp) {
8911 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
8912 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
8913 free(vsp);
8914
8915 port->realdev_ofp_port = 0;
8916 } else {
8917 VLOG_ERR("missing vlan device record");
8918 }
8919 }
8920
8921 static void
8922 vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid)
8923 {
8924 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
8925
8926 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
8927 && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid))
8928 == realdev_ofp_port)) {
8929 struct vlan_splinter *vsp;
8930
8931 vsp = xmalloc(sizeof *vsp);
8932 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
8933 hash_int(port->up.ofp_port, 0));
8934 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
8935 hash_realdev_vid(realdev_ofp_port, vid));
8936 vsp->realdev_ofp_port = realdev_ofp_port;
8937 vsp->vlandev_ofp_port = port->up.ofp_port;
8938 vsp->vid = vid;
8939
8940 port->realdev_ofp_port = realdev_ofp_port;
8941 } else {
8942 VLOG_ERR("duplicate vlan device record");
8943 }
8944 }
8945
8946 static uint32_t
8947 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
8948 {
8949 const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
8950 return ofport ? ofport->odp_port : OVSP_NONE;
8951 }
8952
8953 static struct ofport_dpif *
8954 odp_port_to_ofport(const struct dpif_backer *backer, uint32_t odp_port)
8955 {
8956 struct ofport_dpif *port;
8957
8958 HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node,
8959 hash_int(odp_port, 0),
8960 &backer->odp_to_ofport_map) {
8961 if (port->odp_port == odp_port) {
8962 return port;
8963 }
8964 }
8965
8966 return NULL;
8967 }
8968
8969 static uint16_t
8970 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
8971 {
8972 struct ofport_dpif *port;
8973
8974 port = odp_port_to_ofport(ofproto->backer, odp_port);
8975 if (port && &ofproto->up == port->up.ofproto) {
8976 return port->up.ofp_port;
8977 } else {
8978 return OFPP_NONE;
8979 }
8980 }
8981
8982 /* Compute exponentially weighted moving average, adding 'new' as the newest,
8983 * most heavily weighted element. 'base' designates the rate of decay: after
8984 * 'base' further updates, 'new''s weight in the EWMA decays to about 1/e
8985 * (about .37). */
8986 static void
8987 exp_mavg(double *avg, int base, double new)
8988 {
8989 *avg = (*avg * (base - 1) + new) / base;
8990 }
8991
8992 static void
8993 update_moving_averages(struct dpif_backer *backer)
8994 {
8995 const int min_ms = 60 * 1000; /* milliseconds in one minute. */
8996 long long int minutes = (time_msec() - backer->created) / min_ms;
8997
8998 if (minutes > 0) {
8999 backer->lifetime.add_rate = (double) backer->total_subfacet_add_count
9000 / minutes;
9001 backer->lifetime.del_rate = (double) backer->total_subfacet_del_count
9002 / minutes;
9003 } else {
9004 backer->lifetime.add_rate = 0.0;
9005 backer->lifetime.del_rate = 0.0;
9006 }
9007
9008 /* Update hourly averages on the minute boundaries. */
9009 if (time_msec() - backer->last_minute >= min_ms) {
9010 exp_mavg(&backer->hourly.add_rate, 60, backer->subfacet_add_count);
9011 exp_mavg(&backer->hourly.del_rate, 60, backer->subfacet_del_count);
9012
9013 /* Update daily averages on the hour boundaries. */
9014 if ((backer->last_minute - backer->created) / min_ms % 60 == 59) {
9015 exp_mavg(&backer->daily.add_rate, 24, backer->hourly.add_rate);
9016 exp_mavg(&backer->daily.del_rate, 24, backer->hourly.del_rate);
9017 }
9018
9019 backer->total_subfacet_add_count += backer->subfacet_add_count;
9020 backer->total_subfacet_del_count += backer->subfacet_del_count;
9021 backer->subfacet_add_count = 0;
9022 backer->subfacet_del_count = 0;
9023 backer->last_minute += min_ms;
9024 }
9025 }
9026
9027 const struct ofproto_class ofproto_dpif_class = {
9028 init,
9029 enumerate_types,
9030 enumerate_names,
9031 del,
9032 port_open_type,
9033 type_run,
9034 type_run_fast,
9035 type_wait,
9036 alloc,
9037 construct,
9038 destruct,
9039 dealloc,
9040 run,
9041 run_fast,
9042 wait,
9043 get_memory_usage,
9044 flush,
9045 get_features,
9046 get_tables,
9047 port_alloc,
9048 port_construct,
9049 port_destruct,
9050 port_dealloc,
9051 port_modified,
9052 port_reconfigured,
9053 port_query_by_name,
9054 port_add,
9055 port_del,
9056 port_get_stats,
9057 port_dump_start,
9058 port_dump_next,
9059 port_dump_done,
9060 port_poll,
9061 port_poll_wait,
9062 port_is_lacp_current,
9063 NULL, /* rule_choose_table */
9064 rule_alloc,
9065 rule_construct,
9066 rule_destruct,
9067 rule_dealloc,
9068 rule_get_stats,
9069 rule_execute,
9070 rule_modify_actions,
9071 set_frag_handling,
9072 packet_out,
9073 set_netflow,
9074 get_netflow_ids,
9075 set_sflow,
9076 set_ipfix,
9077 set_cfm,
9078 get_cfm_status,
9079 set_bfd,
9080 get_bfd_status,
9081 set_stp,
9082 get_stp_status,
9083 set_stp_port,
9084 get_stp_port_status,
9085 set_queues,
9086 bundle_set,
9087 bundle_remove,
9088 mirror_set,
9089 mirror_get_stats,
9090 set_flood_vlans,
9091 is_mirror_output_bundle,
9092 forward_bpdu_changed,
9093 set_mac_table_config,
9094 set_realdev,
9095 };