]>
Commit | Line | Data |
---|---|---|
1 | /* | |
2 | * Copyright (c) 2009, 2010, 2011, 2012 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 "autopath.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.h" | |
40 | #include "netlink.h" | |
41 | #include "nx-match.h" | |
42 | #include "odp-util.h" | |
43 | #include "ofp-util.h" | |
44 | #include "ofpbuf.h" | |
45 | #include "ofp-actions.h" | |
46 | #include "ofp-parse.h" | |
47 | #include "ofp-print.h" | |
48 | #include "ofproto-dpif-governor.h" | |
49 | #include "ofproto-dpif-sflow.h" | |
50 | #include "poll-loop.h" | |
51 | #include "simap.h" | |
52 | #include "timer.h" | |
53 | #include "unaligned.h" | |
54 | #include "unixctl.h" | |
55 | #include "vlan-bitmap.h" | |
56 | #include "vlog.h" | |
57 | ||
58 | VLOG_DEFINE_THIS_MODULE(ofproto_dpif); | |
59 | ||
60 | COVERAGE_DEFINE(ofproto_dpif_expired); | |
61 | COVERAGE_DEFINE(ofproto_dpif_xlate); | |
62 | COVERAGE_DEFINE(facet_changed_rule); | |
63 | COVERAGE_DEFINE(facet_revalidate); | |
64 | COVERAGE_DEFINE(facet_unexpected); | |
65 | COVERAGE_DEFINE(facet_suppress); | |
66 | ||
67 | /* Maximum depth of flow table recursion (due to resubmit actions) in a | |
68 | * flow translation. */ | |
69 | #define MAX_RESUBMIT_RECURSION 32 | |
70 | ||
71 | /* Number of implemented OpenFlow tables. */ | |
72 | enum { N_TABLES = 255 }; | |
73 | enum { TBL_INTERNAL = N_TABLES - 1 }; /* Used for internal hidden rules. */ | |
74 | BUILD_ASSERT_DECL(N_TABLES >= 2 && N_TABLES <= 255); | |
75 | ||
76 | struct ofport_dpif; | |
77 | struct ofproto_dpif; | |
78 | ||
79 | struct rule_dpif { | |
80 | struct rule up; | |
81 | ||
82 | /* These statistics: | |
83 | * | |
84 | * - Do include packets and bytes from facets that have been deleted or | |
85 | * whose own statistics have been folded into the rule. | |
86 | * | |
87 | * - Do include packets and bytes sent "by hand" that were accounted to | |
88 | * the rule without any facet being involved (this is a rare corner | |
89 | * case in rule_execute()). | |
90 | * | |
91 | * - Do not include packet or bytes that can be obtained from any facet's | |
92 | * packet_count or byte_count member or that can be obtained from the | |
93 | * datapath by, e.g., dpif_flow_get() for any subfacet. | |
94 | */ | |
95 | uint64_t packet_count; /* Number of packets received. */ | |
96 | uint64_t byte_count; /* Number of bytes received. */ | |
97 | ||
98 | tag_type tag; /* Caches rule_calculate_tag() result. */ | |
99 | ||
100 | struct list facets; /* List of "struct facet"s. */ | |
101 | }; | |
102 | ||
103 | static struct rule_dpif *rule_dpif_cast(const struct rule *rule) | |
104 | { | |
105 | return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL; | |
106 | } | |
107 | ||
108 | static struct rule_dpif *rule_dpif_lookup(struct ofproto_dpif *, | |
109 | const struct flow *); | |
110 | static struct rule_dpif *rule_dpif_lookup__(struct ofproto_dpif *, | |
111 | const struct flow *, | |
112 | uint8_t table); | |
113 | ||
114 | static void rule_credit_stats(struct rule_dpif *, | |
115 | const struct dpif_flow_stats *); | |
116 | static void flow_push_stats(struct rule_dpif *, const struct flow *, | |
117 | const struct dpif_flow_stats *); | |
118 | static tag_type rule_calculate_tag(const struct flow *, | |
119 | const struct flow_wildcards *, | |
120 | uint32_t basis); | |
121 | static void rule_invalidate(const struct rule_dpif *); | |
122 | ||
123 | #define MAX_MIRRORS 32 | |
124 | typedef uint32_t mirror_mask_t; | |
125 | #define MIRROR_MASK_C(X) UINT32_C(X) | |
126 | BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS); | |
127 | struct ofmirror { | |
128 | struct ofproto_dpif *ofproto; /* Owning ofproto. */ | |
129 | size_t idx; /* In ofproto's "mirrors" array. */ | |
130 | void *aux; /* Key supplied by ofproto's client. */ | |
131 | char *name; /* Identifier for log messages. */ | |
132 | ||
133 | /* Selection criteria. */ | |
134 | struct hmapx srcs; /* Contains "struct ofbundle *"s. */ | |
135 | struct hmapx dsts; /* Contains "struct ofbundle *"s. */ | |
136 | unsigned long *vlans; /* Bitmap of chosen VLANs, NULL selects all. */ | |
137 | ||
138 | /* Output (exactly one of out == NULL and out_vlan == -1 is true). */ | |
139 | struct ofbundle *out; /* Output port or NULL. */ | |
140 | int out_vlan; /* Output VLAN or -1. */ | |
141 | mirror_mask_t dup_mirrors; /* Bitmap of mirrors with the same output. */ | |
142 | ||
143 | /* Counters. */ | |
144 | int64_t packet_count; /* Number of packets sent. */ | |
145 | int64_t byte_count; /* Number of bytes sent. */ | |
146 | }; | |
147 | ||
148 | static void mirror_destroy(struct ofmirror *); | |
149 | static void update_mirror_stats(struct ofproto_dpif *ofproto, | |
150 | mirror_mask_t mirrors, | |
151 | uint64_t packets, uint64_t bytes); | |
152 | ||
153 | struct ofbundle { | |
154 | struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */ | |
155 | struct ofproto_dpif *ofproto; /* Owning ofproto. */ | |
156 | void *aux; /* Key supplied by ofproto's client. */ | |
157 | char *name; /* Identifier for log messages. */ | |
158 | ||
159 | /* Configuration. */ | |
160 | struct list ports; /* Contains "struct ofport"s. */ | |
161 | enum port_vlan_mode vlan_mode; /* VLAN mode */ | |
162 | int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */ | |
163 | unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1. | |
164 | * NULL if all VLANs are trunked. */ | |
165 | struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */ | |
166 | struct bond *bond; /* Nonnull iff more than one port. */ | |
167 | bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */ | |
168 | ||
169 | /* Status. */ | |
170 | bool floodable; /* True if no port has OFPUTIL_PC_NO_FLOOD set. */ | |
171 | ||
172 | /* Port mirroring info. */ | |
173 | mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */ | |
174 | mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */ | |
175 | mirror_mask_t mirror_out; /* Mirrors that output to this bundle. */ | |
176 | }; | |
177 | ||
178 | static void bundle_remove(struct ofport *); | |
179 | static void bundle_update(struct ofbundle *); | |
180 | static void bundle_destroy(struct ofbundle *); | |
181 | static void bundle_del_port(struct ofport_dpif *); | |
182 | static void bundle_run(struct ofbundle *); | |
183 | static void bundle_wait(struct ofbundle *); | |
184 | static struct ofbundle *lookup_input_bundle(const struct ofproto_dpif *, | |
185 | uint16_t in_port, bool warn, | |
186 | struct ofport_dpif **in_ofportp); | |
187 | ||
188 | /* A controller may use OFPP_NONE as the ingress port to indicate that | |
189 | * it did not arrive on a "real" port. 'ofpp_none_bundle' exists for | |
190 | * when an input bundle is needed for validation (e.g., mirroring or | |
191 | * OFPP_NORMAL processing). It is not connected to an 'ofproto' or have | |
192 | * any 'port' structs, so care must be taken when dealing with it. */ | |
193 | static struct ofbundle ofpp_none_bundle = { | |
194 | .name = "OFPP_NONE", | |
195 | .vlan_mode = PORT_VLAN_TRUNK | |
196 | }; | |
197 | ||
198 | static void stp_run(struct ofproto_dpif *ofproto); | |
199 | static void stp_wait(struct ofproto_dpif *ofproto); | |
200 | static int set_stp_port(struct ofport *, | |
201 | const struct ofproto_port_stp_settings *); | |
202 | ||
203 | static bool ofbundle_includes_vlan(const struct ofbundle *, uint16_t vlan); | |
204 | ||
205 | struct action_xlate_ctx { | |
206 | /* action_xlate_ctx_init() initializes these members. */ | |
207 | ||
208 | /* The ofproto. */ | |
209 | struct ofproto_dpif *ofproto; | |
210 | ||
211 | /* Flow to which the OpenFlow actions apply. xlate_actions() will modify | |
212 | * this flow when actions change header fields. */ | |
213 | struct flow flow; | |
214 | ||
215 | /* The packet corresponding to 'flow', or a null pointer if we are | |
216 | * revalidating without a packet to refer to. */ | |
217 | const struct ofpbuf *packet; | |
218 | ||
219 | /* Should OFPP_NORMAL update the MAC learning table? Should "learn" | |
220 | * actions update the flow table? | |
221 | * | |
222 | * We want to update these tables if we are actually processing a packet, | |
223 | * or if we are accounting for packets that the datapath has processed, but | |
224 | * not if we are just revalidating. */ | |
225 | bool may_learn; | |
226 | ||
227 | /* The rule that we are currently translating, or NULL. */ | |
228 | struct rule_dpif *rule; | |
229 | ||
230 | /* Union of the set of TCP flags seen so far in this flow. (Used only by | |
231 | * NXAST_FIN_TIMEOUT. Set to zero to avoid updating updating rules' | |
232 | * timeouts.) */ | |
233 | uint8_t tcp_flags; | |
234 | ||
235 | /* If nonnull, flow translation calls this function just before executing a | |
236 | * resubmit or OFPP_TABLE action. In addition, disables logging of traces | |
237 | * when the recursion depth is exceeded. | |
238 | * | |
239 | * 'rule' is the rule being submitted into. It will be null if the | |
240 | * resubmit or OFPP_TABLE action didn't find a matching rule. | |
241 | * | |
242 | * This is normally null so the client has to set it manually after | |
243 | * calling action_xlate_ctx_init(). */ | |
244 | void (*resubmit_hook)(struct action_xlate_ctx *, struct rule_dpif *rule); | |
245 | ||
246 | /* If nonnull, flow translation calls this function to report some | |
247 | * significant decision, e.g. to explain why OFPP_NORMAL translation | |
248 | * dropped a packet. */ | |
249 | void (*report_hook)(struct action_xlate_ctx *, const char *s); | |
250 | ||
251 | /* If nonnull, flow translation credits the specified statistics to each | |
252 | * rule reached through a resubmit or OFPP_TABLE action. | |
253 | * | |
254 | * This is normally null so the client has to set it manually after | |
255 | * calling action_xlate_ctx_init(). */ | |
256 | const struct dpif_flow_stats *resubmit_stats; | |
257 | ||
258 | /* xlate_actions() initializes and uses these members. The client might want | |
259 | * to look at them after it returns. */ | |
260 | ||
261 | struct ofpbuf *odp_actions; /* Datapath actions. */ | |
262 | tag_type tags; /* Tags associated with actions. */ | |
263 | enum slow_path_reason slow; /* 0 if fast path may be used. */ | |
264 | bool has_learn; /* Actions include NXAST_LEARN? */ | |
265 | bool has_normal; /* Actions output to OFPP_NORMAL? */ | |
266 | bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */ | |
267 | uint16_t nf_output_iface; /* Output interface index for NetFlow. */ | |
268 | mirror_mask_t mirrors; /* Bitmap of associated mirrors. */ | |
269 | ||
270 | /* xlate_actions() initializes and uses these members, but the client has no | |
271 | * reason to look at them. */ | |
272 | ||
273 | int recurse; /* Recursion level, via xlate_table_action. */ | |
274 | bool max_resubmit_trigger; /* Recursed too deeply during translation. */ | |
275 | struct flow base_flow; /* Flow at the last commit. */ | |
276 | uint32_t orig_skb_priority; /* Priority when packet arrived. */ | |
277 | uint8_t table_id; /* OpenFlow table ID where flow was found. */ | |
278 | uint32_t sflow_n_outputs; /* Number of output ports. */ | |
279 | uint16_t sflow_odp_port; /* Output port for composing sFlow action. */ | |
280 | uint16_t user_cookie_offset;/* Used for user_action_cookie fixup. */ | |
281 | bool exit; /* No further actions should be processed. */ | |
282 | struct flow orig_flow; /* Copy of original flow. */ | |
283 | }; | |
284 | ||
285 | static void action_xlate_ctx_init(struct action_xlate_ctx *, | |
286 | struct ofproto_dpif *, const struct flow *, | |
287 | ovs_be16 initial_tci, struct rule_dpif *, | |
288 | uint8_t tcp_flags, const struct ofpbuf *); | |
289 | static void xlate_actions(struct action_xlate_ctx *, | |
290 | const struct ofpact *ofpacts, size_t ofpacts_len, | |
291 | struct ofpbuf *odp_actions); | |
292 | static void xlate_actions_for_side_effects(struct action_xlate_ctx *, | |
293 | const struct ofpact *ofpacts, | |
294 | size_t ofpacts_len); | |
295 | ||
296 | static size_t put_userspace_action(const struct ofproto_dpif *, | |
297 | struct ofpbuf *odp_actions, | |
298 | const struct flow *, | |
299 | const union user_action_cookie *); | |
300 | ||
301 | static void compose_slow_path(const struct ofproto_dpif *, const struct flow *, | |
302 | enum slow_path_reason, | |
303 | uint64_t *stub, size_t stub_size, | |
304 | const struct nlattr **actionsp, | |
305 | size_t *actions_lenp); | |
306 | ||
307 | static void xlate_report(struct action_xlate_ctx *ctx, const char *s); | |
308 | ||
309 | /* A subfacet (see "struct subfacet" below) has three possible installation | |
310 | * states: | |
311 | * | |
312 | * - SF_NOT_INSTALLED: Not installed in the datapath. This will only be the | |
313 | * case just after the subfacet is created, just before the subfacet is | |
314 | * destroyed, or if the datapath returns an error when we try to install a | |
315 | * subfacet. | |
316 | * | |
317 | * - SF_FAST_PATH: The subfacet's actions are installed in the datapath. | |
318 | * | |
319 | * - SF_SLOW_PATH: An action that sends every packet for the subfacet through | |
320 | * ofproto_dpif is installed in the datapath. | |
321 | */ | |
322 | enum subfacet_path { | |
323 | SF_NOT_INSTALLED, /* No datapath flow for this subfacet. */ | |
324 | SF_FAST_PATH, /* Full actions are installed. */ | |
325 | SF_SLOW_PATH, /* Send-to-userspace action is installed. */ | |
326 | }; | |
327 | ||
328 | static const char *subfacet_path_to_string(enum subfacet_path); | |
329 | ||
330 | /* A dpif flow and actions associated with a facet. | |
331 | * | |
332 | * See also the large comment on struct facet. */ | |
333 | struct subfacet { | |
334 | /* Owners. */ | |
335 | struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */ | |
336 | struct list list_node; /* In struct facet's 'facets' list. */ | |
337 | struct facet *facet; /* Owning facet. */ | |
338 | ||
339 | /* Key. | |
340 | * | |
341 | * To save memory in the common case, 'key' is NULL if 'key_fitness' is | |
342 | * ODP_FIT_PERFECT, that is, odp_flow_key_from_flow() can accurately | |
343 | * regenerate the ODP flow key from ->facet->flow. */ | |
344 | enum odp_key_fitness key_fitness; | |
345 | struct nlattr *key; | |
346 | int key_len; | |
347 | ||
348 | long long int used; /* Time last used; time created if not used. */ | |
349 | ||
350 | uint64_t dp_packet_count; /* Last known packet count in the datapath. */ | |
351 | uint64_t dp_byte_count; /* Last known byte count in the datapath. */ | |
352 | ||
353 | /* Datapath actions. | |
354 | * | |
355 | * These should be essentially identical for every subfacet in a facet, but | |
356 | * may differ in trivial ways due to VLAN splinters. */ | |
357 | size_t actions_len; /* Number of bytes in actions[]. */ | |
358 | struct nlattr *actions; /* Datapath actions. */ | |
359 | ||
360 | enum slow_path_reason slow; /* 0 if fast path may be used. */ | |
361 | enum subfacet_path path; /* Installed in datapath? */ | |
362 | ||
363 | /* This value is normally the same as ->facet->flow.vlan_tci. Only VLAN | |
364 | * splinters can cause it to differ. This value should be removed when | |
365 | * the VLAN splinters feature is no longer needed. */ | |
366 | ovs_be16 initial_tci; /* Initial VLAN TCI value. */ | |
367 | }; | |
368 | ||
369 | static struct subfacet *subfacet_create(struct facet *, enum odp_key_fitness, | |
370 | const struct nlattr *key, | |
371 | size_t key_len, ovs_be16 initial_tci); | |
372 | static struct subfacet *subfacet_find(struct ofproto_dpif *, | |
373 | const struct nlattr *key, size_t key_len); | |
374 | static void subfacet_destroy(struct subfacet *); | |
375 | static void subfacet_destroy__(struct subfacet *); | |
376 | static void subfacet_get_key(struct subfacet *, struct odputil_keybuf *, | |
377 | struct ofpbuf *key); | |
378 | static void subfacet_reset_dp_stats(struct subfacet *, | |
379 | struct dpif_flow_stats *); | |
380 | static void subfacet_update_time(struct subfacet *, long long int used); | |
381 | static void subfacet_update_stats(struct subfacet *, | |
382 | const struct dpif_flow_stats *); | |
383 | static void subfacet_make_actions(struct subfacet *, | |
384 | const struct ofpbuf *packet, | |
385 | struct ofpbuf *odp_actions); | |
386 | static int subfacet_install(struct subfacet *, | |
387 | const struct nlattr *actions, size_t actions_len, | |
388 | struct dpif_flow_stats *, enum slow_path_reason); | |
389 | static void subfacet_uninstall(struct subfacet *); | |
390 | ||
391 | static enum subfacet_path subfacet_want_path(enum slow_path_reason); | |
392 | ||
393 | /* An exact-match instantiation of an OpenFlow flow. | |
394 | * | |
395 | * A facet associates a "struct flow", which represents the Open vSwitch | |
396 | * userspace idea of an exact-match flow, with one or more subfacets. Each | |
397 | * subfacet tracks the datapath's idea of the exact-match flow equivalent to | |
398 | * the facet. When the kernel module (or other dpif implementation) and Open | |
399 | * vSwitch userspace agree on the definition of a flow key, there is exactly | |
400 | * one subfacet per facet. If the dpif implementation supports more-specific | |
401 | * flow matching than userspace, however, a facet can have more than one | |
402 | * subfacet, each of which corresponds to some distinction in flow that | |
403 | * userspace simply doesn't understand. | |
404 | * | |
405 | * Flow expiration works in terms of subfacets, so a facet must have at least | |
406 | * one subfacet or it will never expire, leaking memory. */ | |
407 | struct facet { | |
408 | /* Owners. */ | |
409 | struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */ | |
410 | struct list list_node; /* In owning rule's 'facets' list. */ | |
411 | struct rule_dpif *rule; /* Owning rule. */ | |
412 | ||
413 | /* Owned data. */ | |
414 | struct list subfacets; | |
415 | long long int used; /* Time last used; time created if not used. */ | |
416 | ||
417 | /* Key. */ | |
418 | struct flow flow; | |
419 | ||
420 | /* These statistics: | |
421 | * | |
422 | * - Do include packets and bytes sent "by hand", e.g. with | |
423 | * dpif_execute(). | |
424 | * | |
425 | * - Do include packets and bytes that were obtained from the datapath | |
426 | * when a subfacet's statistics were reset (e.g. dpif_flow_put() with | |
427 | * DPIF_FP_ZERO_STATS). | |
428 | * | |
429 | * - Do not include packets or bytes that can be obtained from the | |
430 | * datapath for any existing subfacet. | |
431 | */ | |
432 | uint64_t packet_count; /* Number of packets received. */ | |
433 | uint64_t byte_count; /* Number of bytes received. */ | |
434 | ||
435 | /* Resubmit statistics. */ | |
436 | uint64_t prev_packet_count; /* Number of packets from last stats push. */ | |
437 | uint64_t prev_byte_count; /* Number of bytes from last stats push. */ | |
438 | long long int prev_used; /* Used time from last stats push. */ | |
439 | ||
440 | /* Accounting. */ | |
441 | uint64_t accounted_bytes; /* Bytes processed by facet_account(). */ | |
442 | struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */ | |
443 | uint8_t tcp_flags; /* TCP flags seen for this 'rule'. */ | |
444 | ||
445 | /* Properties of datapath actions. | |
446 | * | |
447 | * Every subfacet has its own actions because actions can differ slightly | |
448 | * between splintered and non-splintered subfacets due to the VLAN tag | |
449 | * being initially different (present vs. absent). All of them have these | |
450 | * properties in common so we just store one copy of them here. */ | |
451 | bool has_learn; /* Actions include NXAST_LEARN? */ | |
452 | bool has_normal; /* Actions output to OFPP_NORMAL? */ | |
453 | bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */ | |
454 | tag_type tags; /* Tags that would require revalidation. */ | |
455 | mirror_mask_t mirrors; /* Bitmap of dependent mirrors. */ | |
456 | ||
457 | /* Storage for a single subfacet, to reduce malloc() time and space | |
458 | * overhead. (A facet always has at least one subfacet and in the common | |
459 | * case has exactly one subfacet.) */ | |
460 | struct subfacet one_subfacet; | |
461 | }; | |
462 | ||
463 | static struct facet *facet_create(struct rule_dpif *, | |
464 | const struct flow *, uint32_t hash); | |
465 | static void facet_remove(struct facet *); | |
466 | static void facet_free(struct facet *); | |
467 | ||
468 | static struct facet *facet_find(struct ofproto_dpif *, | |
469 | const struct flow *, uint32_t hash); | |
470 | static struct facet *facet_lookup_valid(struct ofproto_dpif *, | |
471 | const struct flow *, uint32_t hash); | |
472 | static void facet_revalidate(struct facet *); | |
473 | static bool facet_check_consistency(struct facet *); | |
474 | ||
475 | static void facet_flush_stats(struct facet *); | |
476 | ||
477 | static void facet_update_time(struct facet *, long long int used); | |
478 | static void facet_reset_counters(struct facet *); | |
479 | static void facet_push_stats(struct facet *); | |
480 | static void facet_learn(struct facet *); | |
481 | static void facet_account(struct facet *); | |
482 | ||
483 | static bool facet_is_controller_flow(struct facet *); | |
484 | ||
485 | struct ofport_dpif { | |
486 | struct ofport up; | |
487 | ||
488 | uint32_t odp_port; | |
489 | struct ofbundle *bundle; /* Bundle that contains this port, if any. */ | |
490 | struct list bundle_node; /* In struct ofbundle's "ports" list. */ | |
491 | struct cfm *cfm; /* Connectivity Fault Management, if any. */ | |
492 | tag_type tag; /* Tag associated with this port. */ | |
493 | uint32_t bond_stable_id; /* stable_id to use as bond slave, or 0. */ | |
494 | bool may_enable; /* May be enabled in bonds. */ | |
495 | long long int carrier_seq; /* Carrier status changes. */ | |
496 | ||
497 | /* Spanning tree. */ | |
498 | struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */ | |
499 | enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */ | |
500 | long long int stp_state_entered; | |
501 | ||
502 | struct hmap priorities; /* Map of attached 'priority_to_dscp's. */ | |
503 | ||
504 | /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.) | |
505 | * | |
506 | * This is deprecated. It is only for compatibility with broken device | |
507 | * drivers in old versions of Linux that do not properly support VLANs when | |
508 | * VLAN devices are not used. When broken device drivers are no longer in | |
509 | * widespread use, we will delete these interfaces. */ | |
510 | uint16_t realdev_ofp_port; | |
511 | int vlandev_vid; | |
512 | }; | |
513 | ||
514 | /* Node in 'ofport_dpif''s 'priorities' map. Used to maintain a map from | |
515 | * 'priority' (the datapath's term for QoS queue) to the dscp bits which all | |
516 | * traffic egressing the 'ofport' with that priority should be marked with. */ | |
517 | struct priority_to_dscp { | |
518 | struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'priorities' map. */ | |
519 | uint32_t priority; /* Priority of this queue (see struct flow). */ | |
520 | ||
521 | uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */ | |
522 | }; | |
523 | ||
524 | /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.) | |
525 | * | |
526 | * This is deprecated. It is only for compatibility with broken device drivers | |
527 | * in old versions of Linux that do not properly support VLANs when VLAN | |
528 | * devices are not used. When broken device drivers are no longer in | |
529 | * widespread use, we will delete these interfaces. */ | |
530 | struct vlan_splinter { | |
531 | struct hmap_node realdev_vid_node; | |
532 | struct hmap_node vlandev_node; | |
533 | uint16_t realdev_ofp_port; | |
534 | uint16_t vlandev_ofp_port; | |
535 | int vid; | |
536 | }; | |
537 | ||
538 | static uint32_t vsp_realdev_to_vlandev(const struct ofproto_dpif *, | |
539 | uint32_t realdev, ovs_be16 vlan_tci); | |
540 | static bool vsp_adjust_flow(const struct ofproto_dpif *, struct flow *); | |
541 | static void vsp_remove(struct ofport_dpif *); | |
542 | static void vsp_add(struct ofport_dpif *, uint16_t realdev_ofp_port, int vid); | |
543 | ||
544 | static struct ofport_dpif * | |
545 | ofport_dpif_cast(const struct ofport *ofport) | |
546 | { | |
547 | assert(ofport->ofproto->ofproto_class == &ofproto_dpif_class); | |
548 | return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL; | |
549 | } | |
550 | ||
551 | static void port_run(struct ofport_dpif *); | |
552 | static void port_run_fast(struct ofport_dpif *); | |
553 | static void port_wait(struct ofport_dpif *); | |
554 | static int set_cfm(struct ofport *, const struct cfm_settings *); | |
555 | static void ofport_clear_priorities(struct ofport_dpif *); | |
556 | ||
557 | struct dpif_completion { | |
558 | struct list list_node; | |
559 | struct ofoperation *op; | |
560 | }; | |
561 | ||
562 | /* Extra information about a classifier table. | |
563 | * Currently used just for optimized flow revalidation. */ | |
564 | struct table_dpif { | |
565 | /* If either of these is nonnull, then this table has a form that allows | |
566 | * flows to be tagged to avoid revalidating most flows for the most common | |
567 | * kinds of flow table changes. */ | |
568 | struct cls_table *catchall_table; /* Table that wildcards all fields. */ | |
569 | struct cls_table *other_table; /* Table with any other wildcard set. */ | |
570 | uint32_t basis; /* Keeps each table's tags separate. */ | |
571 | }; | |
572 | ||
573 | /* Reasons that we might need to revalidate every facet, and corresponding | |
574 | * coverage counters. | |
575 | * | |
576 | * A value of 0 means that there is no need to revalidate. | |
577 | * | |
578 | * It would be nice to have some cleaner way to integrate with coverage | |
579 | * counters, but with only a few reasons I guess this is good enough for | |
580 | * now. */ | |
581 | enum revalidate_reason { | |
582 | REV_RECONFIGURE = 1, /* Switch configuration changed. */ | |
583 | REV_STP, /* Spanning tree protocol port status change. */ | |
584 | REV_PORT_TOGGLED, /* Port enabled or disabled by CFM, LACP, ...*/ | |
585 | REV_FLOW_TABLE, /* Flow table changed. */ | |
586 | REV_INCONSISTENCY /* Facet self-check failed. */ | |
587 | }; | |
588 | COVERAGE_DEFINE(rev_reconfigure); | |
589 | COVERAGE_DEFINE(rev_stp); | |
590 | COVERAGE_DEFINE(rev_port_toggled); | |
591 | COVERAGE_DEFINE(rev_flow_table); | |
592 | COVERAGE_DEFINE(rev_inconsistency); | |
593 | ||
594 | struct ofproto_dpif { | |
595 | struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */ | |
596 | struct ofproto up; | |
597 | struct dpif *dpif; | |
598 | int max_ports; | |
599 | ||
600 | /* Special OpenFlow rules. */ | |
601 | struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */ | |
602 | struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */ | |
603 | ||
604 | /* Statistics. */ | |
605 | uint64_t n_matches; | |
606 | ||
607 | /* Bridging. */ | |
608 | struct netflow *netflow; | |
609 | struct dpif_sflow *sflow; | |
610 | struct hmap bundles; /* Contains "struct ofbundle"s. */ | |
611 | struct mac_learning *ml; | |
612 | struct ofmirror *mirrors[MAX_MIRRORS]; | |
613 | bool has_mirrors; | |
614 | bool has_bonded_bundles; | |
615 | ||
616 | /* Expiration. */ | |
617 | struct timer next_expiration; | |
618 | ||
619 | /* Facets. */ | |
620 | struct hmap facets; | |
621 | struct hmap subfacets; | |
622 | struct governor *governor; | |
623 | ||
624 | /* Revalidation. */ | |
625 | struct table_dpif tables[N_TABLES]; | |
626 | enum revalidate_reason need_revalidate; | |
627 | struct tag_set revalidate_set; | |
628 | ||
629 | /* Support for debugging async flow mods. */ | |
630 | struct list completions; | |
631 | ||
632 | bool has_bundle_action; /* True when the first bundle action appears. */ | |
633 | struct netdev_stats stats; /* To account packets generated and consumed in | |
634 | * userspace. */ | |
635 | ||
636 | /* Spanning tree. */ | |
637 | struct stp *stp; | |
638 | long long int stp_last_tick; | |
639 | ||
640 | /* VLAN splinters. */ | |
641 | struct hmap realdev_vid_map; /* (realdev,vid) -> vlandev. */ | |
642 | struct hmap vlandev_map; /* vlandev -> (realdev,vid). */ | |
643 | }; | |
644 | ||
645 | /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only | |
646 | * for debugging the asynchronous flow_mod implementation.) */ | |
647 | static bool clogged; | |
648 | ||
649 | /* All existing ofproto_dpif instances, indexed by ->up.name. */ | |
650 | static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs); | |
651 | ||
652 | static void ofproto_dpif_unixctl_init(void); | |
653 | ||
654 | static struct ofproto_dpif * | |
655 | ofproto_dpif_cast(const struct ofproto *ofproto) | |
656 | { | |
657 | assert(ofproto->ofproto_class == &ofproto_dpif_class); | |
658 | return CONTAINER_OF(ofproto, struct ofproto_dpif, up); | |
659 | } | |
660 | ||
661 | static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *, | |
662 | uint16_t ofp_port); | |
663 | static struct ofport_dpif *get_odp_port(const struct ofproto_dpif *, | |
664 | uint32_t odp_port); | |
665 | static void ofproto_trace(struct ofproto_dpif *, const struct flow *, | |
666 | const struct ofpbuf *, ovs_be16 initial_tci, | |
667 | struct ds *); | |
668 | ||
669 | /* Packet processing. */ | |
670 | static void update_learning_table(struct ofproto_dpif *, | |
671 | const struct flow *, int vlan, | |
672 | struct ofbundle *); | |
673 | /* Upcalls. */ | |
674 | #define FLOW_MISS_MAX_BATCH 50 | |
675 | static int handle_upcalls(struct ofproto_dpif *, unsigned int max_batch); | |
676 | ||
677 | /* Flow expiration. */ | |
678 | static int expire(struct ofproto_dpif *); | |
679 | ||
680 | /* NetFlow. */ | |
681 | static void send_netflow_active_timeouts(struct ofproto_dpif *); | |
682 | ||
683 | /* Utilities. */ | |
684 | static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet); | |
685 | static size_t compose_sflow_action(const struct ofproto_dpif *, | |
686 | struct ofpbuf *odp_actions, | |
687 | const struct flow *, uint32_t odp_port); | |
688 | static void add_mirror_actions(struct action_xlate_ctx *ctx, | |
689 | const struct flow *flow); | |
690 | /* Global variables. */ | |
691 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
692 | \f | |
693 | /* Factory functions. */ | |
694 | ||
695 | static void | |
696 | enumerate_types(struct sset *types) | |
697 | { | |
698 | dp_enumerate_types(types); | |
699 | } | |
700 | ||
701 | static int | |
702 | enumerate_names(const char *type, struct sset *names) | |
703 | { | |
704 | return dp_enumerate_names(type, names); | |
705 | } | |
706 | ||
707 | static int | |
708 | del(const char *type, const char *name) | |
709 | { | |
710 | struct dpif *dpif; | |
711 | int error; | |
712 | ||
713 | error = dpif_open(name, type, &dpif); | |
714 | if (!error) { | |
715 | error = dpif_delete(dpif); | |
716 | dpif_close(dpif); | |
717 | } | |
718 | return error; | |
719 | } | |
720 | \f | |
721 | /* Basic life-cycle. */ | |
722 | ||
723 | static int add_internal_flows(struct ofproto_dpif *); | |
724 | ||
725 | static struct ofproto * | |
726 | alloc(void) | |
727 | { | |
728 | struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto); | |
729 | return &ofproto->up; | |
730 | } | |
731 | ||
732 | static void | |
733 | dealloc(struct ofproto *ofproto_) | |
734 | { | |
735 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
736 | free(ofproto); | |
737 | } | |
738 | ||
739 | static int | |
740 | construct(struct ofproto *ofproto_) | |
741 | { | |
742 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
743 | const char *name = ofproto->up.name; | |
744 | int error; | |
745 | int i; | |
746 | ||
747 | error = dpif_create_and_open(name, ofproto->up.type, &ofproto->dpif); | |
748 | if (error) { | |
749 | VLOG_ERR("failed to open datapath %s: %s", name, strerror(error)); | |
750 | return error; | |
751 | } | |
752 | ||
753 | ofproto->max_ports = dpif_get_max_ports(ofproto->dpif); | |
754 | ofproto->n_matches = 0; | |
755 | ||
756 | dpif_flow_flush(ofproto->dpif); | |
757 | dpif_recv_purge(ofproto->dpif); | |
758 | ||
759 | error = dpif_recv_set(ofproto->dpif, true); | |
760 | if (error) { | |
761 | VLOG_ERR("failed to listen on datapath %s: %s", name, strerror(error)); | |
762 | dpif_close(ofproto->dpif); | |
763 | return error; | |
764 | } | |
765 | ||
766 | ofproto->netflow = NULL; | |
767 | ofproto->sflow = NULL; | |
768 | ofproto->stp = NULL; | |
769 | hmap_init(&ofproto->bundles); | |
770 | ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME); | |
771 | for (i = 0; i < MAX_MIRRORS; i++) { | |
772 | ofproto->mirrors[i] = NULL; | |
773 | } | |
774 | ofproto->has_bonded_bundles = false; | |
775 | ||
776 | timer_set_duration(&ofproto->next_expiration, 1000); | |
777 | ||
778 | hmap_init(&ofproto->facets); | |
779 | hmap_init(&ofproto->subfacets); | |
780 | ofproto->governor = NULL; | |
781 | ||
782 | for (i = 0; i < N_TABLES; i++) { | |
783 | struct table_dpif *table = &ofproto->tables[i]; | |
784 | ||
785 | table->catchall_table = NULL; | |
786 | table->other_table = NULL; | |
787 | table->basis = random_uint32(); | |
788 | } | |
789 | ofproto->need_revalidate = 0; | |
790 | tag_set_init(&ofproto->revalidate_set); | |
791 | ||
792 | list_init(&ofproto->completions); | |
793 | ||
794 | ofproto_dpif_unixctl_init(); | |
795 | ||
796 | ofproto->has_mirrors = false; | |
797 | ofproto->has_bundle_action = false; | |
798 | ||
799 | hmap_init(&ofproto->vlandev_map); | |
800 | hmap_init(&ofproto->realdev_vid_map); | |
801 | ||
802 | hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node, | |
803 | hash_string(ofproto->up.name, 0)); | |
804 | memset(&ofproto->stats, 0, sizeof ofproto->stats); | |
805 | ||
806 | ofproto_init_tables(ofproto_, N_TABLES); | |
807 | error = add_internal_flows(ofproto); | |
808 | ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY; | |
809 | ||
810 | return error; | |
811 | } | |
812 | ||
813 | static int | |
814 | add_internal_flow(struct ofproto_dpif *ofproto, int id, | |
815 | const struct ofpbuf *ofpacts, struct rule_dpif **rulep) | |
816 | { | |
817 | struct ofputil_flow_mod fm; | |
818 | int error; | |
819 | ||
820 | cls_rule_init_catchall(&fm.cr, 0); | |
821 | cls_rule_set_reg(&fm.cr, 0, id); | |
822 | fm.new_cookie = htonll(0); | |
823 | fm.cookie = htonll(0); | |
824 | fm.cookie_mask = htonll(0); | |
825 | fm.table_id = TBL_INTERNAL; | |
826 | fm.command = OFPFC_ADD; | |
827 | fm.idle_timeout = 0; | |
828 | fm.hard_timeout = 0; | |
829 | fm.buffer_id = 0; | |
830 | fm.out_port = 0; | |
831 | fm.flags = 0; | |
832 | fm.ofpacts = ofpacts->data; | |
833 | fm.ofpacts_len = ofpacts->size; | |
834 | ||
835 | error = ofproto_flow_mod(&ofproto->up, &fm); | |
836 | if (error) { | |
837 | VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)", | |
838 | id, ofperr_to_string(error)); | |
839 | return error; | |
840 | } | |
841 | ||
842 | *rulep = rule_dpif_lookup__(ofproto, &fm.cr.flow, TBL_INTERNAL); | |
843 | assert(*rulep != NULL); | |
844 | ||
845 | return 0; | |
846 | } | |
847 | ||
848 | static int | |
849 | add_internal_flows(struct ofproto_dpif *ofproto) | |
850 | { | |
851 | struct ofpact_controller *controller; | |
852 | uint64_t ofpacts_stub[128 / 8]; | |
853 | struct ofpbuf ofpacts; | |
854 | int error; | |
855 | int id; | |
856 | ||
857 | ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub); | |
858 | id = 1; | |
859 | ||
860 | controller = ofpact_put_CONTROLLER(&ofpacts); | |
861 | controller->max_len = UINT16_MAX; | |
862 | controller->controller_id = 0; | |
863 | controller->reason = OFPR_NO_MATCH; | |
864 | ofpact_pad(&ofpacts); | |
865 | ||
866 | error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule); | |
867 | if (error) { | |
868 | return error; | |
869 | } | |
870 | ||
871 | ofpbuf_clear(&ofpacts); | |
872 | error = add_internal_flow(ofproto, id++, &ofpacts, | |
873 | &ofproto->no_packet_in_rule); | |
874 | return error; | |
875 | } | |
876 | ||
877 | static void | |
878 | complete_operations(struct ofproto_dpif *ofproto) | |
879 | { | |
880 | struct dpif_completion *c, *next; | |
881 | ||
882 | LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) { | |
883 | ofoperation_complete(c->op, 0); | |
884 | list_remove(&c->list_node); | |
885 | free(c); | |
886 | } | |
887 | } | |
888 | ||
889 | static void | |
890 | destruct(struct ofproto *ofproto_) | |
891 | { | |
892 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
893 | struct rule_dpif *rule, *next_rule; | |
894 | struct oftable *table; | |
895 | int i; | |
896 | ||
897 | hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node); | |
898 | complete_operations(ofproto); | |
899 | ||
900 | OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) { | |
901 | struct cls_cursor cursor; | |
902 | ||
903 | cls_cursor_init(&cursor, &table->cls, NULL); | |
904 | CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) { | |
905 | ofproto_rule_destroy(&rule->up); | |
906 | } | |
907 | } | |
908 | ||
909 | for (i = 0; i < MAX_MIRRORS; i++) { | |
910 | mirror_destroy(ofproto->mirrors[i]); | |
911 | } | |
912 | ||
913 | netflow_destroy(ofproto->netflow); | |
914 | dpif_sflow_destroy(ofproto->sflow); | |
915 | hmap_destroy(&ofproto->bundles); | |
916 | mac_learning_destroy(ofproto->ml); | |
917 | ||
918 | hmap_destroy(&ofproto->facets); | |
919 | hmap_destroy(&ofproto->subfacets); | |
920 | governor_destroy(ofproto->governor); | |
921 | ||
922 | hmap_destroy(&ofproto->vlandev_map); | |
923 | hmap_destroy(&ofproto->realdev_vid_map); | |
924 | ||
925 | dpif_close(ofproto->dpif); | |
926 | } | |
927 | ||
928 | static int | |
929 | run_fast(struct ofproto *ofproto_) | |
930 | { | |
931 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
932 | struct ofport_dpif *ofport; | |
933 | unsigned int work; | |
934 | ||
935 | HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) { | |
936 | port_run_fast(ofport); | |
937 | } | |
938 | ||
939 | /* Handle one or more batches of upcalls, until there's nothing left to do | |
940 | * or until we do a fixed total amount of work. | |
941 | * | |
942 | * We do work in batches because it can be much cheaper to set up a number | |
943 | * of flows and fire off their patches all at once. We do multiple batches | |
944 | * because in some cases handling a packet can cause another packet to be | |
945 | * queued almost immediately as part of the return flow. Both | |
946 | * optimizations can make major improvements on some benchmarks and | |
947 | * presumably for real traffic as well. */ | |
948 | work = 0; | |
949 | while (work < FLOW_MISS_MAX_BATCH) { | |
950 | int retval = handle_upcalls(ofproto, FLOW_MISS_MAX_BATCH - work); | |
951 | if (retval <= 0) { | |
952 | return -retval; | |
953 | } | |
954 | work += retval; | |
955 | } | |
956 | return 0; | |
957 | } | |
958 | ||
959 | static int | |
960 | run(struct ofproto *ofproto_) | |
961 | { | |
962 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
963 | struct ofport_dpif *ofport; | |
964 | struct ofbundle *bundle; | |
965 | int error; | |
966 | ||
967 | if (!clogged) { | |
968 | complete_operations(ofproto); | |
969 | } | |
970 | dpif_run(ofproto->dpif); | |
971 | ||
972 | error = run_fast(ofproto_); | |
973 | if (error) { | |
974 | return error; | |
975 | } | |
976 | ||
977 | if (timer_expired(&ofproto->next_expiration)) { | |
978 | int delay = expire(ofproto); | |
979 | timer_set_duration(&ofproto->next_expiration, delay); | |
980 | } | |
981 | ||
982 | if (ofproto->netflow) { | |
983 | if (netflow_run(ofproto->netflow)) { | |
984 | send_netflow_active_timeouts(ofproto); | |
985 | } | |
986 | } | |
987 | if (ofproto->sflow) { | |
988 | dpif_sflow_run(ofproto->sflow); | |
989 | } | |
990 | ||
991 | HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) { | |
992 | port_run(ofport); | |
993 | } | |
994 | HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { | |
995 | bundle_run(bundle); | |
996 | } | |
997 | ||
998 | stp_run(ofproto); | |
999 | mac_learning_run(ofproto->ml, &ofproto->revalidate_set); | |
1000 | ||
1001 | /* Now revalidate if there's anything to do. */ | |
1002 | if (ofproto->need_revalidate | |
1003 | || !tag_set_is_empty(&ofproto->revalidate_set)) { | |
1004 | struct tag_set revalidate_set = ofproto->revalidate_set; | |
1005 | bool revalidate_all = ofproto->need_revalidate; | |
1006 | struct facet *facet; | |
1007 | ||
1008 | switch (ofproto->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 | /* Clear the revalidation flags. */ | |
1017 | tag_set_init(&ofproto->revalidate_set); | |
1018 | ofproto->need_revalidate = 0; | |
1019 | ||
1020 | HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) { | |
1021 | if (revalidate_all | |
1022 | || tag_set_intersects(&revalidate_set, facet->tags)) { | |
1023 | facet_revalidate(facet); | |
1024 | } | |
1025 | } | |
1026 | } | |
1027 | ||
1028 | /* Check the consistency of a random facet, to aid debugging. */ | |
1029 | if (!hmap_is_empty(&ofproto->facets) && !ofproto->need_revalidate) { | |
1030 | struct facet *facet; | |
1031 | ||
1032 | facet = CONTAINER_OF(hmap_random_node(&ofproto->facets), | |
1033 | struct facet, hmap_node); | |
1034 | if (!tag_set_intersects(&ofproto->revalidate_set, facet->tags)) { | |
1035 | if (!facet_check_consistency(facet)) { | |
1036 | ofproto->need_revalidate = REV_INCONSISTENCY; | |
1037 | } | |
1038 | } | |
1039 | } | |
1040 | ||
1041 | if (ofproto->governor) { | |
1042 | size_t n_subfacets; | |
1043 | ||
1044 | governor_run(ofproto->governor); | |
1045 | ||
1046 | /* If the governor has shrunk to its minimum size and the number of | |
1047 | * subfacets has dwindled, then drop the governor entirely. | |
1048 | * | |
1049 | * For hysteresis, the number of subfacets to drop the governor is | |
1050 | * smaller than the number needed to trigger its creation. */ | |
1051 | n_subfacets = hmap_count(&ofproto->subfacets); | |
1052 | if (n_subfacets * 4 < ofproto->up.flow_eviction_threshold | |
1053 | && governor_is_idle(ofproto->governor)) { | |
1054 | governor_destroy(ofproto->governor); | |
1055 | ofproto->governor = NULL; | |
1056 | } | |
1057 | } | |
1058 | ||
1059 | return 0; | |
1060 | } | |
1061 | ||
1062 | static void | |
1063 | wait(struct ofproto *ofproto_) | |
1064 | { | |
1065 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1066 | struct ofport_dpif *ofport; | |
1067 | struct ofbundle *bundle; | |
1068 | ||
1069 | if (!clogged && !list_is_empty(&ofproto->completions)) { | |
1070 | poll_immediate_wake(); | |
1071 | } | |
1072 | ||
1073 | dpif_wait(ofproto->dpif); | |
1074 | dpif_recv_wait(ofproto->dpif); | |
1075 | if (ofproto->sflow) { | |
1076 | dpif_sflow_wait(ofproto->sflow); | |
1077 | } | |
1078 | if (!tag_set_is_empty(&ofproto->revalidate_set)) { | |
1079 | poll_immediate_wake(); | |
1080 | } | |
1081 | HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) { | |
1082 | port_wait(ofport); | |
1083 | } | |
1084 | HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { | |
1085 | bundle_wait(bundle); | |
1086 | } | |
1087 | if (ofproto->netflow) { | |
1088 | netflow_wait(ofproto->netflow); | |
1089 | } | |
1090 | mac_learning_wait(ofproto->ml); | |
1091 | stp_wait(ofproto); | |
1092 | if (ofproto->need_revalidate) { | |
1093 | /* Shouldn't happen, but if it does just go around again. */ | |
1094 | VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()"); | |
1095 | poll_immediate_wake(); | |
1096 | } else { | |
1097 | timer_wait(&ofproto->next_expiration); | |
1098 | } | |
1099 | if (ofproto->governor) { | |
1100 | governor_wait(ofproto->governor); | |
1101 | } | |
1102 | } | |
1103 | ||
1104 | static void | |
1105 | get_memory_usage(const struct ofproto *ofproto_, struct simap *usage) | |
1106 | { | |
1107 | const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1108 | ||
1109 | simap_increase(usage, "facets", hmap_count(&ofproto->facets)); | |
1110 | simap_increase(usage, "subfacets", hmap_count(&ofproto->subfacets)); | |
1111 | } | |
1112 | ||
1113 | static void | |
1114 | flush(struct ofproto *ofproto_) | |
1115 | { | |
1116 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1117 | struct facet *facet, *next_facet; | |
1118 | ||
1119 | HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) { | |
1120 | /* Mark the facet as not installed so that facet_remove() doesn't | |
1121 | * bother trying to uninstall it. There is no point in uninstalling it | |
1122 | * individually since we are about to blow away all the facets with | |
1123 | * dpif_flow_flush(). */ | |
1124 | struct subfacet *subfacet; | |
1125 | ||
1126 | LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) { | |
1127 | subfacet->path = SF_NOT_INSTALLED; | |
1128 | subfacet->dp_packet_count = 0; | |
1129 | subfacet->dp_byte_count = 0; | |
1130 | } | |
1131 | facet_remove(facet); | |
1132 | } | |
1133 | dpif_flow_flush(ofproto->dpif); | |
1134 | } | |
1135 | ||
1136 | static void | |
1137 | get_features(struct ofproto *ofproto_ OVS_UNUSED, | |
1138 | bool *arp_match_ip, enum ofputil_action_bitmap *actions) | |
1139 | { | |
1140 | *arp_match_ip = true; | |
1141 | *actions = (OFPUTIL_A_OUTPUT | | |
1142 | OFPUTIL_A_SET_VLAN_VID | | |
1143 | OFPUTIL_A_SET_VLAN_PCP | | |
1144 | OFPUTIL_A_STRIP_VLAN | | |
1145 | OFPUTIL_A_SET_DL_SRC | | |
1146 | OFPUTIL_A_SET_DL_DST | | |
1147 | OFPUTIL_A_SET_NW_SRC | | |
1148 | OFPUTIL_A_SET_NW_DST | | |
1149 | OFPUTIL_A_SET_NW_TOS | | |
1150 | OFPUTIL_A_SET_TP_SRC | | |
1151 | OFPUTIL_A_SET_TP_DST | | |
1152 | OFPUTIL_A_ENQUEUE); | |
1153 | } | |
1154 | ||
1155 | static void | |
1156 | get_tables(struct ofproto *ofproto_, struct ofp10_table_stats *ots) | |
1157 | { | |
1158 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1159 | struct dpif_dp_stats s; | |
1160 | ||
1161 | strcpy(ots->name, "classifier"); | |
1162 | ||
1163 | dpif_get_dp_stats(ofproto->dpif, &s); | |
1164 | put_32aligned_be64(&ots->lookup_count, htonll(s.n_hit + s.n_missed)); | |
1165 | put_32aligned_be64(&ots->matched_count, | |
1166 | htonll(s.n_hit + ofproto->n_matches)); | |
1167 | } | |
1168 | ||
1169 | static struct ofport * | |
1170 | port_alloc(void) | |
1171 | { | |
1172 | struct ofport_dpif *port = xmalloc(sizeof *port); | |
1173 | return &port->up; | |
1174 | } | |
1175 | ||
1176 | static void | |
1177 | port_dealloc(struct ofport *port_) | |
1178 | { | |
1179 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
1180 | free(port); | |
1181 | } | |
1182 | ||
1183 | static int | |
1184 | port_construct(struct ofport *port_) | |
1185 | { | |
1186 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
1187 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto); | |
1188 | ||
1189 | ofproto->need_revalidate = REV_RECONFIGURE; | |
1190 | port->odp_port = ofp_port_to_odp_port(port->up.ofp_port); | |
1191 | port->bundle = NULL; | |
1192 | port->cfm = NULL; | |
1193 | port->tag = tag_create_random(); | |
1194 | port->may_enable = true; | |
1195 | port->stp_port = NULL; | |
1196 | port->stp_state = STP_DISABLED; | |
1197 | hmap_init(&port->priorities); | |
1198 | port->realdev_ofp_port = 0; | |
1199 | port->vlandev_vid = 0; | |
1200 | port->carrier_seq = netdev_get_carrier_resets(port->up.netdev); | |
1201 | ||
1202 | if (ofproto->sflow) { | |
1203 | dpif_sflow_add_port(ofproto->sflow, port_); | |
1204 | } | |
1205 | ||
1206 | return 0; | |
1207 | } | |
1208 | ||
1209 | static void | |
1210 | port_destruct(struct ofport *port_) | |
1211 | { | |
1212 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
1213 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto); | |
1214 | ||
1215 | ofproto->need_revalidate = REV_RECONFIGURE; | |
1216 | bundle_remove(port_); | |
1217 | set_cfm(port_, NULL); | |
1218 | if (ofproto->sflow) { | |
1219 | dpif_sflow_del_port(ofproto->sflow, port->odp_port); | |
1220 | } | |
1221 | ||
1222 | ofport_clear_priorities(port); | |
1223 | hmap_destroy(&port->priorities); | |
1224 | } | |
1225 | ||
1226 | static void | |
1227 | port_modified(struct ofport *port_) | |
1228 | { | |
1229 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
1230 | ||
1231 | if (port->bundle && port->bundle->bond) { | |
1232 | bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev); | |
1233 | } | |
1234 | } | |
1235 | ||
1236 | static void | |
1237 | port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config) | |
1238 | { | |
1239 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
1240 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto); | |
1241 | enum ofputil_port_config changed = old_config ^ port->up.pp.config; | |
1242 | ||
1243 | if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP | | |
1244 | OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD | | |
1245 | OFPUTIL_PC_NO_PACKET_IN)) { | |
1246 | ofproto->need_revalidate = REV_RECONFIGURE; | |
1247 | ||
1248 | if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) { | |
1249 | bundle_update(port->bundle); | |
1250 | } | |
1251 | } | |
1252 | } | |
1253 | ||
1254 | static int | |
1255 | set_sflow(struct ofproto *ofproto_, | |
1256 | const struct ofproto_sflow_options *sflow_options) | |
1257 | { | |
1258 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1259 | struct dpif_sflow *ds = ofproto->sflow; | |
1260 | ||
1261 | if (sflow_options) { | |
1262 | if (!ds) { | |
1263 | struct ofport_dpif *ofport; | |
1264 | ||
1265 | ds = ofproto->sflow = dpif_sflow_create(ofproto->dpif); | |
1266 | HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) { | |
1267 | dpif_sflow_add_port(ds, &ofport->up); | |
1268 | } | |
1269 | ofproto->need_revalidate = REV_RECONFIGURE; | |
1270 | } | |
1271 | dpif_sflow_set_options(ds, sflow_options); | |
1272 | } else { | |
1273 | if (ds) { | |
1274 | dpif_sflow_destroy(ds); | |
1275 | ofproto->need_revalidate = REV_RECONFIGURE; | |
1276 | ofproto->sflow = NULL; | |
1277 | } | |
1278 | } | |
1279 | return 0; | |
1280 | } | |
1281 | ||
1282 | static int | |
1283 | set_cfm(struct ofport *ofport_, const struct cfm_settings *s) | |
1284 | { | |
1285 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
1286 | int error; | |
1287 | ||
1288 | if (!s) { | |
1289 | error = 0; | |
1290 | } else { | |
1291 | if (!ofport->cfm) { | |
1292 | struct ofproto_dpif *ofproto; | |
1293 | ||
1294 | ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
1295 | ofproto->need_revalidate = REV_RECONFIGURE; | |
1296 | ofport->cfm = cfm_create(netdev_get_name(ofport->up.netdev)); | |
1297 | } | |
1298 | ||
1299 | if (cfm_configure(ofport->cfm, s)) { | |
1300 | return 0; | |
1301 | } | |
1302 | ||
1303 | error = EINVAL; | |
1304 | } | |
1305 | cfm_destroy(ofport->cfm); | |
1306 | ofport->cfm = NULL; | |
1307 | return error; | |
1308 | } | |
1309 | ||
1310 | static int | |
1311 | get_cfm_fault(const struct ofport *ofport_) | |
1312 | { | |
1313 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
1314 | ||
1315 | return ofport->cfm ? cfm_get_fault(ofport->cfm) : -1; | |
1316 | } | |
1317 | ||
1318 | static int | |
1319 | get_cfm_opup(const struct ofport *ofport_) | |
1320 | { | |
1321 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
1322 | ||
1323 | return ofport->cfm ? cfm_get_opup(ofport->cfm) : -1; | |
1324 | } | |
1325 | ||
1326 | static int | |
1327 | get_cfm_remote_mpids(const struct ofport *ofport_, const uint64_t **rmps, | |
1328 | size_t *n_rmps) | |
1329 | { | |
1330 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
1331 | ||
1332 | if (ofport->cfm) { | |
1333 | cfm_get_remote_mpids(ofport->cfm, rmps, n_rmps); | |
1334 | return 0; | |
1335 | } else { | |
1336 | return -1; | |
1337 | } | |
1338 | } | |
1339 | ||
1340 | static int | |
1341 | get_cfm_health(const struct ofport *ofport_) | |
1342 | { | |
1343 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
1344 | ||
1345 | return ofport->cfm ? cfm_get_health(ofport->cfm) : -1; | |
1346 | } | |
1347 | \f | |
1348 | /* Spanning Tree. */ | |
1349 | ||
1350 | static void | |
1351 | send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_) | |
1352 | { | |
1353 | struct ofproto_dpif *ofproto = ofproto_; | |
1354 | struct stp_port *sp = stp_get_port(ofproto->stp, port_num); | |
1355 | struct ofport_dpif *ofport; | |
1356 | ||
1357 | ofport = stp_port_get_aux(sp); | |
1358 | if (!ofport) { | |
1359 | VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d", | |
1360 | ofproto->up.name, port_num); | |
1361 | } else { | |
1362 | struct eth_header *eth = pkt->l2; | |
1363 | ||
1364 | netdev_get_etheraddr(ofport->up.netdev, eth->eth_src); | |
1365 | if (eth_addr_is_zero(eth->eth_src)) { | |
1366 | VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d " | |
1367 | "with unknown MAC", ofproto->up.name, port_num); | |
1368 | } else { | |
1369 | send_packet(ofport, pkt); | |
1370 | } | |
1371 | } | |
1372 | ofpbuf_delete(pkt); | |
1373 | } | |
1374 | ||
1375 | /* Configures STP on 'ofproto_' using the settings defined in 's'. */ | |
1376 | static int | |
1377 | set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s) | |
1378 | { | |
1379 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1380 | ||
1381 | /* Only revalidate flows if the configuration changed. */ | |
1382 | if (!s != !ofproto->stp) { | |
1383 | ofproto->need_revalidate = REV_RECONFIGURE; | |
1384 | } | |
1385 | ||
1386 | if (s) { | |
1387 | if (!ofproto->stp) { | |
1388 | ofproto->stp = stp_create(ofproto_->name, s->system_id, | |
1389 | send_bpdu_cb, ofproto); | |
1390 | ofproto->stp_last_tick = time_msec(); | |
1391 | } | |
1392 | ||
1393 | stp_set_bridge_id(ofproto->stp, s->system_id); | |
1394 | stp_set_bridge_priority(ofproto->stp, s->priority); | |
1395 | stp_set_hello_time(ofproto->stp, s->hello_time); | |
1396 | stp_set_max_age(ofproto->stp, s->max_age); | |
1397 | stp_set_forward_delay(ofproto->stp, s->fwd_delay); | |
1398 | } else { | |
1399 | struct ofport *ofport; | |
1400 | ||
1401 | HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) { | |
1402 | set_stp_port(ofport, NULL); | |
1403 | } | |
1404 | ||
1405 | stp_destroy(ofproto->stp); | |
1406 | ofproto->stp = NULL; | |
1407 | } | |
1408 | ||
1409 | return 0; | |
1410 | } | |
1411 | ||
1412 | static int | |
1413 | get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s) | |
1414 | { | |
1415 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1416 | ||
1417 | if (ofproto->stp) { | |
1418 | s->enabled = true; | |
1419 | s->bridge_id = stp_get_bridge_id(ofproto->stp); | |
1420 | s->designated_root = stp_get_designated_root(ofproto->stp); | |
1421 | s->root_path_cost = stp_get_root_path_cost(ofproto->stp); | |
1422 | } else { | |
1423 | s->enabled = false; | |
1424 | } | |
1425 | ||
1426 | return 0; | |
1427 | } | |
1428 | ||
1429 | static void | |
1430 | update_stp_port_state(struct ofport_dpif *ofport) | |
1431 | { | |
1432 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
1433 | enum stp_state state; | |
1434 | ||
1435 | /* Figure out new state. */ | |
1436 | state = ofport->stp_port ? stp_port_get_state(ofport->stp_port) | |
1437 | : STP_DISABLED; | |
1438 | ||
1439 | /* Update state. */ | |
1440 | if (ofport->stp_state != state) { | |
1441 | enum ofputil_port_state of_state; | |
1442 | bool fwd_change; | |
1443 | ||
1444 | VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s", | |
1445 | netdev_get_name(ofport->up.netdev), | |
1446 | stp_state_name(ofport->stp_state), | |
1447 | stp_state_name(state)); | |
1448 | if (stp_learn_in_state(ofport->stp_state) | |
1449 | != stp_learn_in_state(state)) { | |
1450 | /* xxx Learning action flows should also be flushed. */ | |
1451 | mac_learning_flush(ofproto->ml, &ofproto->revalidate_set); | |
1452 | } | |
1453 | fwd_change = stp_forward_in_state(ofport->stp_state) | |
1454 | != stp_forward_in_state(state); | |
1455 | ||
1456 | ofproto->need_revalidate = REV_STP; | |
1457 | ofport->stp_state = state; | |
1458 | ofport->stp_state_entered = time_msec(); | |
1459 | ||
1460 | if (fwd_change && ofport->bundle) { | |
1461 | bundle_update(ofport->bundle); | |
1462 | } | |
1463 | ||
1464 | /* Update the STP state bits in the OpenFlow port description. */ | |
1465 | of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK; | |
1466 | of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN | |
1467 | : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN | |
1468 | : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD | |
1469 | : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK | |
1470 | : 0); | |
1471 | ofproto_port_set_state(&ofport->up, of_state); | |
1472 | } | |
1473 | } | |
1474 | ||
1475 | /* Configures STP on 'ofport_' using the settings defined in 's'. The | |
1476 | * caller is responsible for assigning STP port numbers and ensuring | |
1477 | * there are no duplicates. */ | |
1478 | static int | |
1479 | set_stp_port(struct ofport *ofport_, | |
1480 | const struct ofproto_port_stp_settings *s) | |
1481 | { | |
1482 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
1483 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
1484 | struct stp_port *sp = ofport->stp_port; | |
1485 | ||
1486 | if (!s || !s->enable) { | |
1487 | if (sp) { | |
1488 | ofport->stp_port = NULL; | |
1489 | stp_port_disable(sp); | |
1490 | update_stp_port_state(ofport); | |
1491 | } | |
1492 | return 0; | |
1493 | } else if (sp && stp_port_no(sp) != s->port_num | |
1494 | && ofport == stp_port_get_aux(sp)) { | |
1495 | /* The port-id changed, so disable the old one if it's not | |
1496 | * already in use by another port. */ | |
1497 | stp_port_disable(sp); | |
1498 | } | |
1499 | ||
1500 | sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num); | |
1501 | stp_port_enable(sp); | |
1502 | ||
1503 | stp_port_set_aux(sp, ofport); | |
1504 | stp_port_set_priority(sp, s->priority); | |
1505 | stp_port_set_path_cost(sp, s->path_cost); | |
1506 | ||
1507 | update_stp_port_state(ofport); | |
1508 | ||
1509 | return 0; | |
1510 | } | |
1511 | ||
1512 | static int | |
1513 | get_stp_port_status(struct ofport *ofport_, | |
1514 | struct ofproto_port_stp_status *s) | |
1515 | { | |
1516 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
1517 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
1518 | struct stp_port *sp = ofport->stp_port; | |
1519 | ||
1520 | if (!ofproto->stp || !sp) { | |
1521 | s->enabled = false; | |
1522 | return 0; | |
1523 | } | |
1524 | ||
1525 | s->enabled = true; | |
1526 | s->port_id = stp_port_get_id(sp); | |
1527 | s->state = stp_port_get_state(sp); | |
1528 | s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000; | |
1529 | s->role = stp_port_get_role(sp); | |
1530 | stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count); | |
1531 | ||
1532 | return 0; | |
1533 | } | |
1534 | ||
1535 | static void | |
1536 | stp_run(struct ofproto_dpif *ofproto) | |
1537 | { | |
1538 | if (ofproto->stp) { | |
1539 | long long int now = time_msec(); | |
1540 | long long int elapsed = now - ofproto->stp_last_tick; | |
1541 | struct stp_port *sp; | |
1542 | ||
1543 | if (elapsed > 0) { | |
1544 | stp_tick(ofproto->stp, MIN(INT_MAX, elapsed)); | |
1545 | ofproto->stp_last_tick = now; | |
1546 | } | |
1547 | while (stp_get_changed_port(ofproto->stp, &sp)) { | |
1548 | struct ofport_dpif *ofport = stp_port_get_aux(sp); | |
1549 | ||
1550 | if (ofport) { | |
1551 | update_stp_port_state(ofport); | |
1552 | } | |
1553 | } | |
1554 | ||
1555 | if (stp_check_and_reset_fdb_flush(ofproto->stp)) { | |
1556 | mac_learning_flush(ofproto->ml, &ofproto->revalidate_set); | |
1557 | } | |
1558 | } | |
1559 | } | |
1560 | ||
1561 | static void | |
1562 | stp_wait(struct ofproto_dpif *ofproto) | |
1563 | { | |
1564 | if (ofproto->stp) { | |
1565 | poll_timer_wait(1000); | |
1566 | } | |
1567 | } | |
1568 | ||
1569 | /* Returns true if STP should process 'flow'. */ | |
1570 | static bool | |
1571 | stp_should_process_flow(const struct flow *flow) | |
1572 | { | |
1573 | return eth_addr_equals(flow->dl_dst, eth_addr_stp); | |
1574 | } | |
1575 | ||
1576 | static void | |
1577 | stp_process_packet(const struct ofport_dpif *ofport, | |
1578 | const struct ofpbuf *packet) | |
1579 | { | |
1580 | struct ofpbuf payload = *packet; | |
1581 | struct eth_header *eth = payload.data; | |
1582 | struct stp_port *sp = ofport->stp_port; | |
1583 | ||
1584 | /* Sink packets on ports that have STP disabled when the bridge has | |
1585 | * STP enabled. */ | |
1586 | if (!sp || stp_port_get_state(sp) == STP_DISABLED) { | |
1587 | return; | |
1588 | } | |
1589 | ||
1590 | /* Trim off padding on payload. */ | |
1591 | if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) { | |
1592 | payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN; | |
1593 | } | |
1594 | ||
1595 | if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) { | |
1596 | stp_received_bpdu(sp, payload.data, payload.size); | |
1597 | } | |
1598 | } | |
1599 | \f | |
1600 | static struct priority_to_dscp * | |
1601 | get_priority(const struct ofport_dpif *ofport, uint32_t priority) | |
1602 | { | |
1603 | struct priority_to_dscp *pdscp; | |
1604 | uint32_t hash; | |
1605 | ||
1606 | hash = hash_int(priority, 0); | |
1607 | HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) { | |
1608 | if (pdscp->priority == priority) { | |
1609 | return pdscp; | |
1610 | } | |
1611 | } | |
1612 | return NULL; | |
1613 | } | |
1614 | ||
1615 | static void | |
1616 | ofport_clear_priorities(struct ofport_dpif *ofport) | |
1617 | { | |
1618 | struct priority_to_dscp *pdscp, *next; | |
1619 | ||
1620 | HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) { | |
1621 | hmap_remove(&ofport->priorities, &pdscp->hmap_node); | |
1622 | free(pdscp); | |
1623 | } | |
1624 | } | |
1625 | ||
1626 | static int | |
1627 | set_queues(struct ofport *ofport_, | |
1628 | const struct ofproto_port_queue *qdscp_list, | |
1629 | size_t n_qdscp) | |
1630 | { | |
1631 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
1632 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
1633 | struct hmap new = HMAP_INITIALIZER(&new); | |
1634 | size_t i; | |
1635 | ||
1636 | for (i = 0; i < n_qdscp; i++) { | |
1637 | struct priority_to_dscp *pdscp; | |
1638 | uint32_t priority; | |
1639 | uint8_t dscp; | |
1640 | ||
1641 | dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK; | |
1642 | if (dpif_queue_to_priority(ofproto->dpif, qdscp_list[i].queue, | |
1643 | &priority)) { | |
1644 | continue; | |
1645 | } | |
1646 | ||
1647 | pdscp = get_priority(ofport, priority); | |
1648 | if (pdscp) { | |
1649 | hmap_remove(&ofport->priorities, &pdscp->hmap_node); | |
1650 | } else { | |
1651 | pdscp = xmalloc(sizeof *pdscp); | |
1652 | pdscp->priority = priority; | |
1653 | pdscp->dscp = dscp; | |
1654 | ofproto->need_revalidate = REV_RECONFIGURE; | |
1655 | } | |
1656 | ||
1657 | if (pdscp->dscp != dscp) { | |
1658 | pdscp->dscp = dscp; | |
1659 | ofproto->need_revalidate = REV_RECONFIGURE; | |
1660 | } | |
1661 | ||
1662 | hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0)); | |
1663 | } | |
1664 | ||
1665 | if (!hmap_is_empty(&ofport->priorities)) { | |
1666 | ofport_clear_priorities(ofport); | |
1667 | ofproto->need_revalidate = REV_RECONFIGURE; | |
1668 | } | |
1669 | ||
1670 | hmap_swap(&new, &ofport->priorities); | |
1671 | hmap_destroy(&new); | |
1672 | ||
1673 | return 0; | |
1674 | } | |
1675 | \f | |
1676 | /* Bundles. */ | |
1677 | ||
1678 | /* Expires all MAC learning entries associated with 'bundle' and forces its | |
1679 | * ofproto to revalidate every flow. | |
1680 | * | |
1681 | * Normally MAC learning entries are removed only from the ofproto associated | |
1682 | * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries | |
1683 | * are removed from every ofproto. When patch ports and SLB bonds are in use | |
1684 | * and a VM migration happens and the gratuitous ARPs are somehow lost, this | |
1685 | * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate | |
1686 | * with the host from which it migrated. */ | |
1687 | static void | |
1688 | bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos) | |
1689 | { | |
1690 | struct ofproto_dpif *ofproto = bundle->ofproto; | |
1691 | struct mac_learning *ml = ofproto->ml; | |
1692 | struct mac_entry *mac, *next_mac; | |
1693 | ||
1694 | ofproto->need_revalidate = REV_RECONFIGURE; | |
1695 | LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) { | |
1696 | if (mac->port.p == bundle) { | |
1697 | if (all_ofprotos) { | |
1698 | struct ofproto_dpif *o; | |
1699 | ||
1700 | HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
1701 | if (o != ofproto) { | |
1702 | struct mac_entry *e; | |
1703 | ||
1704 | e = mac_learning_lookup(o->ml, mac->mac, mac->vlan, | |
1705 | NULL); | |
1706 | if (e) { | |
1707 | tag_set_add(&o->revalidate_set, e->tag); | |
1708 | mac_learning_expire(o->ml, e); | |
1709 | } | |
1710 | } | |
1711 | } | |
1712 | } | |
1713 | ||
1714 | mac_learning_expire(ml, mac); | |
1715 | } | |
1716 | } | |
1717 | } | |
1718 | ||
1719 | static struct ofbundle * | |
1720 | bundle_lookup(const struct ofproto_dpif *ofproto, void *aux) | |
1721 | { | |
1722 | struct ofbundle *bundle; | |
1723 | ||
1724 | HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0), | |
1725 | &ofproto->bundles) { | |
1726 | if (bundle->aux == aux) { | |
1727 | return bundle; | |
1728 | } | |
1729 | } | |
1730 | return NULL; | |
1731 | } | |
1732 | ||
1733 | /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the | |
1734 | * ones that are found to 'bundles'. */ | |
1735 | static void | |
1736 | bundle_lookup_multiple(struct ofproto_dpif *ofproto, | |
1737 | void **auxes, size_t n_auxes, | |
1738 | struct hmapx *bundles) | |
1739 | { | |
1740 | size_t i; | |
1741 | ||
1742 | hmapx_init(bundles); | |
1743 | for (i = 0; i < n_auxes; i++) { | |
1744 | struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]); | |
1745 | if (bundle) { | |
1746 | hmapx_add(bundles, bundle); | |
1747 | } | |
1748 | } | |
1749 | } | |
1750 | ||
1751 | static void | |
1752 | bundle_update(struct ofbundle *bundle) | |
1753 | { | |
1754 | struct ofport_dpif *port; | |
1755 | ||
1756 | bundle->floodable = true; | |
1757 | LIST_FOR_EACH (port, bundle_node, &bundle->ports) { | |
1758 | if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD | |
1759 | || !stp_forward_in_state(port->stp_state)) { | |
1760 | bundle->floodable = false; | |
1761 | break; | |
1762 | } | |
1763 | } | |
1764 | } | |
1765 | ||
1766 | static void | |
1767 | bundle_del_port(struct ofport_dpif *port) | |
1768 | { | |
1769 | struct ofbundle *bundle = port->bundle; | |
1770 | ||
1771 | bundle->ofproto->need_revalidate = REV_RECONFIGURE; | |
1772 | ||
1773 | list_remove(&port->bundle_node); | |
1774 | port->bundle = NULL; | |
1775 | ||
1776 | if (bundle->lacp) { | |
1777 | lacp_slave_unregister(bundle->lacp, port); | |
1778 | } | |
1779 | if (bundle->bond) { | |
1780 | bond_slave_unregister(bundle->bond, port); | |
1781 | } | |
1782 | ||
1783 | bundle_update(bundle); | |
1784 | } | |
1785 | ||
1786 | static bool | |
1787 | bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port, | |
1788 | struct lacp_slave_settings *lacp, | |
1789 | uint32_t bond_stable_id) | |
1790 | { | |
1791 | struct ofport_dpif *port; | |
1792 | ||
1793 | port = get_ofp_port(bundle->ofproto, ofp_port); | |
1794 | if (!port) { | |
1795 | return false; | |
1796 | } | |
1797 | ||
1798 | if (port->bundle != bundle) { | |
1799 | bundle->ofproto->need_revalidate = REV_RECONFIGURE; | |
1800 | if (port->bundle) { | |
1801 | bundle_del_port(port); | |
1802 | } | |
1803 | ||
1804 | port->bundle = bundle; | |
1805 | list_push_back(&bundle->ports, &port->bundle_node); | |
1806 | if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD | |
1807 | || !stp_forward_in_state(port->stp_state)) { | |
1808 | bundle->floodable = false; | |
1809 | } | |
1810 | } | |
1811 | if (lacp) { | |
1812 | port->bundle->ofproto->need_revalidate = REV_RECONFIGURE; | |
1813 | lacp_slave_register(bundle->lacp, port, lacp); | |
1814 | } | |
1815 | ||
1816 | port->bond_stable_id = bond_stable_id; | |
1817 | ||
1818 | return true; | |
1819 | } | |
1820 | ||
1821 | static void | |
1822 | bundle_destroy(struct ofbundle *bundle) | |
1823 | { | |
1824 | struct ofproto_dpif *ofproto; | |
1825 | struct ofport_dpif *port, *next_port; | |
1826 | int i; | |
1827 | ||
1828 | if (!bundle) { | |
1829 | return; | |
1830 | } | |
1831 | ||
1832 | ofproto = bundle->ofproto; | |
1833 | for (i = 0; i < MAX_MIRRORS; i++) { | |
1834 | struct ofmirror *m = ofproto->mirrors[i]; | |
1835 | if (m) { | |
1836 | if (m->out == bundle) { | |
1837 | mirror_destroy(m); | |
1838 | } else if (hmapx_find_and_delete(&m->srcs, bundle) | |
1839 | || hmapx_find_and_delete(&m->dsts, bundle)) { | |
1840 | ofproto->need_revalidate = REV_RECONFIGURE; | |
1841 | } | |
1842 | } | |
1843 | } | |
1844 | ||
1845 | LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) { | |
1846 | bundle_del_port(port); | |
1847 | } | |
1848 | ||
1849 | bundle_flush_macs(bundle, true); | |
1850 | hmap_remove(&ofproto->bundles, &bundle->hmap_node); | |
1851 | free(bundle->name); | |
1852 | free(bundle->trunks); | |
1853 | lacp_destroy(bundle->lacp); | |
1854 | bond_destroy(bundle->bond); | |
1855 | free(bundle); | |
1856 | } | |
1857 | ||
1858 | static int | |
1859 | bundle_set(struct ofproto *ofproto_, void *aux, | |
1860 | const struct ofproto_bundle_settings *s) | |
1861 | { | |
1862 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1863 | bool need_flush = false; | |
1864 | struct ofport_dpif *port; | |
1865 | struct ofbundle *bundle; | |
1866 | unsigned long *trunks; | |
1867 | int vlan; | |
1868 | size_t i; | |
1869 | bool ok; | |
1870 | ||
1871 | if (!s) { | |
1872 | bundle_destroy(bundle_lookup(ofproto, aux)); | |
1873 | return 0; | |
1874 | } | |
1875 | ||
1876 | assert(s->n_slaves == 1 || s->bond != NULL); | |
1877 | assert((s->lacp != NULL) == (s->lacp_slaves != NULL)); | |
1878 | ||
1879 | bundle = bundle_lookup(ofproto, aux); | |
1880 | if (!bundle) { | |
1881 | bundle = xmalloc(sizeof *bundle); | |
1882 | ||
1883 | bundle->ofproto = ofproto; | |
1884 | hmap_insert(&ofproto->bundles, &bundle->hmap_node, | |
1885 | hash_pointer(aux, 0)); | |
1886 | bundle->aux = aux; | |
1887 | bundle->name = NULL; | |
1888 | ||
1889 | list_init(&bundle->ports); | |
1890 | bundle->vlan_mode = PORT_VLAN_TRUNK; | |
1891 | bundle->vlan = -1; | |
1892 | bundle->trunks = NULL; | |
1893 | bundle->use_priority_tags = s->use_priority_tags; | |
1894 | bundle->lacp = NULL; | |
1895 | bundle->bond = NULL; | |
1896 | ||
1897 | bundle->floodable = true; | |
1898 | ||
1899 | bundle->src_mirrors = 0; | |
1900 | bundle->dst_mirrors = 0; | |
1901 | bundle->mirror_out = 0; | |
1902 | } | |
1903 | ||
1904 | if (!bundle->name || strcmp(s->name, bundle->name)) { | |
1905 | free(bundle->name); | |
1906 | bundle->name = xstrdup(s->name); | |
1907 | } | |
1908 | ||
1909 | /* LACP. */ | |
1910 | if (s->lacp) { | |
1911 | if (!bundle->lacp) { | |
1912 | ofproto->need_revalidate = REV_RECONFIGURE; | |
1913 | bundle->lacp = lacp_create(); | |
1914 | } | |
1915 | lacp_configure(bundle->lacp, s->lacp); | |
1916 | } else { | |
1917 | lacp_destroy(bundle->lacp); | |
1918 | bundle->lacp = NULL; | |
1919 | } | |
1920 | ||
1921 | /* Update set of ports. */ | |
1922 | ok = true; | |
1923 | for (i = 0; i < s->n_slaves; i++) { | |
1924 | if (!bundle_add_port(bundle, s->slaves[i], | |
1925 | s->lacp ? &s->lacp_slaves[i] : NULL, | |
1926 | s->bond_stable_ids ? s->bond_stable_ids[i] : 0)) { | |
1927 | ok = false; | |
1928 | } | |
1929 | } | |
1930 | if (!ok || list_size(&bundle->ports) != s->n_slaves) { | |
1931 | struct ofport_dpif *next_port; | |
1932 | ||
1933 | LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) { | |
1934 | for (i = 0; i < s->n_slaves; i++) { | |
1935 | if (s->slaves[i] == port->up.ofp_port) { | |
1936 | goto found; | |
1937 | } | |
1938 | } | |
1939 | ||
1940 | bundle_del_port(port); | |
1941 | found: ; | |
1942 | } | |
1943 | } | |
1944 | assert(list_size(&bundle->ports) <= s->n_slaves); | |
1945 | ||
1946 | if (list_is_empty(&bundle->ports)) { | |
1947 | bundle_destroy(bundle); | |
1948 | return EINVAL; | |
1949 | } | |
1950 | ||
1951 | /* Set VLAN tagging mode */ | |
1952 | if (s->vlan_mode != bundle->vlan_mode | |
1953 | || s->use_priority_tags != bundle->use_priority_tags) { | |
1954 | bundle->vlan_mode = s->vlan_mode; | |
1955 | bundle->use_priority_tags = s->use_priority_tags; | |
1956 | need_flush = true; | |
1957 | } | |
1958 | ||
1959 | /* Set VLAN tag. */ | |
1960 | vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1 | |
1961 | : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan | |
1962 | : 0); | |
1963 | if (vlan != bundle->vlan) { | |
1964 | bundle->vlan = vlan; | |
1965 | need_flush = true; | |
1966 | } | |
1967 | ||
1968 | /* Get trunked VLANs. */ | |
1969 | switch (s->vlan_mode) { | |
1970 | case PORT_VLAN_ACCESS: | |
1971 | trunks = NULL; | |
1972 | break; | |
1973 | ||
1974 | case PORT_VLAN_TRUNK: | |
1975 | trunks = CONST_CAST(unsigned long *, s->trunks); | |
1976 | break; | |
1977 | ||
1978 | case PORT_VLAN_NATIVE_UNTAGGED: | |
1979 | case PORT_VLAN_NATIVE_TAGGED: | |
1980 | if (vlan != 0 && (!s->trunks | |
1981 | || !bitmap_is_set(s->trunks, vlan) | |
1982 | || bitmap_is_set(s->trunks, 0))) { | |
1983 | /* Force trunking the native VLAN and prohibit trunking VLAN 0. */ | |
1984 | if (s->trunks) { | |
1985 | trunks = bitmap_clone(s->trunks, 4096); | |
1986 | } else { | |
1987 | trunks = bitmap_allocate1(4096); | |
1988 | } | |
1989 | bitmap_set1(trunks, vlan); | |
1990 | bitmap_set0(trunks, 0); | |
1991 | } else { | |
1992 | trunks = CONST_CAST(unsigned long *, s->trunks); | |
1993 | } | |
1994 | break; | |
1995 | ||
1996 | default: | |
1997 | NOT_REACHED(); | |
1998 | } | |
1999 | if (!vlan_bitmap_equal(trunks, bundle->trunks)) { | |
2000 | free(bundle->trunks); | |
2001 | if (trunks == s->trunks) { | |
2002 | bundle->trunks = vlan_bitmap_clone(trunks); | |
2003 | } else { | |
2004 | bundle->trunks = trunks; | |
2005 | trunks = NULL; | |
2006 | } | |
2007 | need_flush = true; | |
2008 | } | |
2009 | if (trunks != s->trunks) { | |
2010 | free(trunks); | |
2011 | } | |
2012 | ||
2013 | /* Bonding. */ | |
2014 | if (!list_is_short(&bundle->ports)) { | |
2015 | bundle->ofproto->has_bonded_bundles = true; | |
2016 | if (bundle->bond) { | |
2017 | if (bond_reconfigure(bundle->bond, s->bond)) { | |
2018 | ofproto->need_revalidate = REV_RECONFIGURE; | |
2019 | } | |
2020 | } else { | |
2021 | bundle->bond = bond_create(s->bond); | |
2022 | ofproto->need_revalidate = REV_RECONFIGURE; | |
2023 | } | |
2024 | ||
2025 | LIST_FOR_EACH (port, bundle_node, &bundle->ports) { | |
2026 | bond_slave_register(bundle->bond, port, port->bond_stable_id, | |
2027 | port->up.netdev); | |
2028 | } | |
2029 | } else { | |
2030 | bond_destroy(bundle->bond); | |
2031 | bundle->bond = NULL; | |
2032 | } | |
2033 | ||
2034 | /* If we changed something that would affect MAC learning, un-learn | |
2035 | * everything on this port and force flow revalidation. */ | |
2036 | if (need_flush) { | |
2037 | bundle_flush_macs(bundle, false); | |
2038 | } | |
2039 | ||
2040 | return 0; | |
2041 | } | |
2042 | ||
2043 | static void | |
2044 | bundle_remove(struct ofport *port_) | |
2045 | { | |
2046 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
2047 | struct ofbundle *bundle = port->bundle; | |
2048 | ||
2049 | if (bundle) { | |
2050 | bundle_del_port(port); | |
2051 | if (list_is_empty(&bundle->ports)) { | |
2052 | bundle_destroy(bundle); | |
2053 | } else if (list_is_short(&bundle->ports)) { | |
2054 | bond_destroy(bundle->bond); | |
2055 | bundle->bond = NULL; | |
2056 | } | |
2057 | } | |
2058 | } | |
2059 | ||
2060 | static void | |
2061 | send_pdu_cb(void *port_, const void *pdu, size_t pdu_size) | |
2062 | { | |
2063 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10); | |
2064 | struct ofport_dpif *port = port_; | |
2065 | uint8_t ea[ETH_ADDR_LEN]; | |
2066 | int error; | |
2067 | ||
2068 | error = netdev_get_etheraddr(port->up.netdev, ea); | |
2069 | if (!error) { | |
2070 | struct ofpbuf packet; | |
2071 | void *packet_pdu; | |
2072 | ||
2073 | ofpbuf_init(&packet, 0); | |
2074 | packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP, | |
2075 | pdu_size); | |
2076 | memcpy(packet_pdu, pdu, pdu_size); | |
2077 | ||
2078 | send_packet(port, &packet); | |
2079 | ofpbuf_uninit(&packet); | |
2080 | } else { | |
2081 | VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface " | |
2082 | "%s (%s)", port->bundle->name, | |
2083 | netdev_get_name(port->up.netdev), strerror(error)); | |
2084 | } | |
2085 | } | |
2086 | ||
2087 | static void | |
2088 | bundle_send_learning_packets(struct ofbundle *bundle) | |
2089 | { | |
2090 | struct ofproto_dpif *ofproto = bundle->ofproto; | |
2091 | int error, n_packets, n_errors; | |
2092 | struct mac_entry *e; | |
2093 | ||
2094 | error = n_packets = n_errors = 0; | |
2095 | LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) { | |
2096 | if (e->port.p != bundle) { | |
2097 | struct ofpbuf *learning_packet; | |
2098 | struct ofport_dpif *port; | |
2099 | void *port_void; | |
2100 | int ret; | |
2101 | ||
2102 | /* The assignment to "port" is unnecessary but makes "grep"ing for | |
2103 | * struct ofport_dpif more effective. */ | |
2104 | learning_packet = bond_compose_learning_packet(bundle->bond, | |
2105 | e->mac, e->vlan, | |
2106 | &port_void); | |
2107 | port = port_void; | |
2108 | ret = send_packet(port, learning_packet); | |
2109 | ofpbuf_delete(learning_packet); | |
2110 | if (ret) { | |
2111 | error = ret; | |
2112 | n_errors++; | |
2113 | } | |
2114 | n_packets++; | |
2115 | } | |
2116 | } | |
2117 | ||
2118 | if (n_errors) { | |
2119 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
2120 | VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning " | |
2121 | "packets, last error was: %s", | |
2122 | bundle->name, n_errors, n_packets, strerror(error)); | |
2123 | } else { | |
2124 | VLOG_DBG("bond %s: sent %d gratuitous learning packets", | |
2125 | bundle->name, n_packets); | |
2126 | } | |
2127 | } | |
2128 | ||
2129 | static void | |
2130 | bundle_run(struct ofbundle *bundle) | |
2131 | { | |
2132 | if (bundle->lacp) { | |
2133 | lacp_run(bundle->lacp, send_pdu_cb); | |
2134 | } | |
2135 | if (bundle->bond) { | |
2136 | struct ofport_dpif *port; | |
2137 | ||
2138 | LIST_FOR_EACH (port, bundle_node, &bundle->ports) { | |
2139 | bond_slave_set_may_enable(bundle->bond, port, port->may_enable); | |
2140 | } | |
2141 | ||
2142 | bond_run(bundle->bond, &bundle->ofproto->revalidate_set, | |
2143 | lacp_status(bundle->lacp)); | |
2144 | if (bond_should_send_learning_packets(bundle->bond)) { | |
2145 | bundle_send_learning_packets(bundle); | |
2146 | } | |
2147 | } | |
2148 | } | |
2149 | ||
2150 | static void | |
2151 | bundle_wait(struct ofbundle *bundle) | |
2152 | { | |
2153 | if (bundle->lacp) { | |
2154 | lacp_wait(bundle->lacp); | |
2155 | } | |
2156 | if (bundle->bond) { | |
2157 | bond_wait(bundle->bond); | |
2158 | } | |
2159 | } | |
2160 | \f | |
2161 | /* Mirrors. */ | |
2162 | ||
2163 | static int | |
2164 | mirror_scan(struct ofproto_dpif *ofproto) | |
2165 | { | |
2166 | int idx; | |
2167 | ||
2168 | for (idx = 0; idx < MAX_MIRRORS; idx++) { | |
2169 | if (!ofproto->mirrors[idx]) { | |
2170 | return idx; | |
2171 | } | |
2172 | } | |
2173 | return -1; | |
2174 | } | |
2175 | ||
2176 | static struct ofmirror * | |
2177 | mirror_lookup(struct ofproto_dpif *ofproto, void *aux) | |
2178 | { | |
2179 | int i; | |
2180 | ||
2181 | for (i = 0; i < MAX_MIRRORS; i++) { | |
2182 | struct ofmirror *mirror = ofproto->mirrors[i]; | |
2183 | if (mirror && mirror->aux == aux) { | |
2184 | return mirror; | |
2185 | } | |
2186 | } | |
2187 | ||
2188 | return NULL; | |
2189 | } | |
2190 | ||
2191 | /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */ | |
2192 | static void | |
2193 | mirror_update_dups(struct ofproto_dpif *ofproto) | |
2194 | { | |
2195 | int i; | |
2196 | ||
2197 | for (i = 0; i < MAX_MIRRORS; i++) { | |
2198 | struct ofmirror *m = ofproto->mirrors[i]; | |
2199 | ||
2200 | if (m) { | |
2201 | m->dup_mirrors = MIRROR_MASK_C(1) << i; | |
2202 | } | |
2203 | } | |
2204 | ||
2205 | for (i = 0; i < MAX_MIRRORS; i++) { | |
2206 | struct ofmirror *m1 = ofproto->mirrors[i]; | |
2207 | int j; | |
2208 | ||
2209 | if (!m1) { | |
2210 | continue; | |
2211 | } | |
2212 | ||
2213 | for (j = i + 1; j < MAX_MIRRORS; j++) { | |
2214 | struct ofmirror *m2 = ofproto->mirrors[j]; | |
2215 | ||
2216 | if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) { | |
2217 | m1->dup_mirrors |= MIRROR_MASK_C(1) << j; | |
2218 | m2->dup_mirrors |= m1->dup_mirrors; | |
2219 | } | |
2220 | } | |
2221 | } | |
2222 | } | |
2223 | ||
2224 | static int | |
2225 | mirror_set(struct ofproto *ofproto_, void *aux, | |
2226 | const struct ofproto_mirror_settings *s) | |
2227 | { | |
2228 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2229 | mirror_mask_t mirror_bit; | |
2230 | struct ofbundle *bundle; | |
2231 | struct ofmirror *mirror; | |
2232 | struct ofbundle *out; | |
2233 | struct hmapx srcs; /* Contains "struct ofbundle *"s. */ | |
2234 | struct hmapx dsts; /* Contains "struct ofbundle *"s. */ | |
2235 | int out_vlan; | |
2236 | ||
2237 | mirror = mirror_lookup(ofproto, aux); | |
2238 | if (!s) { | |
2239 | mirror_destroy(mirror); | |
2240 | return 0; | |
2241 | } | |
2242 | if (!mirror) { | |
2243 | int idx; | |
2244 | ||
2245 | idx = mirror_scan(ofproto); | |
2246 | if (idx < 0) { | |
2247 | VLOG_WARN("bridge %s: maximum of %d port mirrors reached, " | |
2248 | "cannot create %s", | |
2249 | ofproto->up.name, MAX_MIRRORS, s->name); | |
2250 | return EFBIG; | |
2251 | } | |
2252 | ||
2253 | mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror); | |
2254 | mirror->ofproto = ofproto; | |
2255 | mirror->idx = idx; | |
2256 | mirror->aux = aux; | |
2257 | mirror->out_vlan = -1; | |
2258 | mirror->name = NULL; | |
2259 | } | |
2260 | ||
2261 | if (!mirror->name || strcmp(s->name, mirror->name)) { | |
2262 | free(mirror->name); | |
2263 | mirror->name = xstrdup(s->name); | |
2264 | } | |
2265 | ||
2266 | /* Get the new configuration. */ | |
2267 | if (s->out_bundle) { | |
2268 | out = bundle_lookup(ofproto, s->out_bundle); | |
2269 | if (!out) { | |
2270 | mirror_destroy(mirror); | |
2271 | return EINVAL; | |
2272 | } | |
2273 | out_vlan = -1; | |
2274 | } else { | |
2275 | out = NULL; | |
2276 | out_vlan = s->out_vlan; | |
2277 | } | |
2278 | bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs); | |
2279 | bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts); | |
2280 | ||
2281 | /* If the configuration has not changed, do nothing. */ | |
2282 | if (hmapx_equals(&srcs, &mirror->srcs) | |
2283 | && hmapx_equals(&dsts, &mirror->dsts) | |
2284 | && vlan_bitmap_equal(mirror->vlans, s->src_vlans) | |
2285 | && mirror->out == out | |
2286 | && mirror->out_vlan == out_vlan) | |
2287 | { | |
2288 | hmapx_destroy(&srcs); | |
2289 | hmapx_destroy(&dsts); | |
2290 | return 0; | |
2291 | } | |
2292 | ||
2293 | hmapx_swap(&srcs, &mirror->srcs); | |
2294 | hmapx_destroy(&srcs); | |
2295 | ||
2296 | hmapx_swap(&dsts, &mirror->dsts); | |
2297 | hmapx_destroy(&dsts); | |
2298 | ||
2299 | free(mirror->vlans); | |
2300 | mirror->vlans = vlan_bitmap_clone(s->src_vlans); | |
2301 | ||
2302 | mirror->out = out; | |
2303 | mirror->out_vlan = out_vlan; | |
2304 | ||
2305 | /* Update bundles. */ | |
2306 | mirror_bit = MIRROR_MASK_C(1) << mirror->idx; | |
2307 | HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) { | |
2308 | if (hmapx_contains(&mirror->srcs, bundle)) { | |
2309 | bundle->src_mirrors |= mirror_bit; | |
2310 | } else { | |
2311 | bundle->src_mirrors &= ~mirror_bit; | |
2312 | } | |
2313 | ||
2314 | if (hmapx_contains(&mirror->dsts, bundle)) { | |
2315 | bundle->dst_mirrors |= mirror_bit; | |
2316 | } else { | |
2317 | bundle->dst_mirrors &= ~mirror_bit; | |
2318 | } | |
2319 | ||
2320 | if (mirror->out == bundle) { | |
2321 | bundle->mirror_out |= mirror_bit; | |
2322 | } else { | |
2323 | bundle->mirror_out &= ~mirror_bit; | |
2324 | } | |
2325 | } | |
2326 | ||
2327 | ofproto->need_revalidate = REV_RECONFIGURE; | |
2328 | ofproto->has_mirrors = true; | |
2329 | mac_learning_flush(ofproto->ml, &ofproto->revalidate_set); | |
2330 | mirror_update_dups(ofproto); | |
2331 | ||
2332 | return 0; | |
2333 | } | |
2334 | ||
2335 | static void | |
2336 | mirror_destroy(struct ofmirror *mirror) | |
2337 | { | |
2338 | struct ofproto_dpif *ofproto; | |
2339 | mirror_mask_t mirror_bit; | |
2340 | struct ofbundle *bundle; | |
2341 | int i; | |
2342 | ||
2343 | if (!mirror) { | |
2344 | return; | |
2345 | } | |
2346 | ||
2347 | ofproto = mirror->ofproto; | |
2348 | ofproto->need_revalidate = REV_RECONFIGURE; | |
2349 | mac_learning_flush(ofproto->ml, &ofproto->revalidate_set); | |
2350 | ||
2351 | mirror_bit = MIRROR_MASK_C(1) << mirror->idx; | |
2352 | HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { | |
2353 | bundle->src_mirrors &= ~mirror_bit; | |
2354 | bundle->dst_mirrors &= ~mirror_bit; | |
2355 | bundle->mirror_out &= ~mirror_bit; | |
2356 | } | |
2357 | ||
2358 | hmapx_destroy(&mirror->srcs); | |
2359 | hmapx_destroy(&mirror->dsts); | |
2360 | free(mirror->vlans); | |
2361 | ||
2362 | ofproto->mirrors[mirror->idx] = NULL; | |
2363 | free(mirror->name); | |
2364 | free(mirror); | |
2365 | ||
2366 | mirror_update_dups(ofproto); | |
2367 | ||
2368 | ofproto->has_mirrors = false; | |
2369 | for (i = 0; i < MAX_MIRRORS; i++) { | |
2370 | if (ofproto->mirrors[i]) { | |
2371 | ofproto->has_mirrors = true; | |
2372 | break; | |
2373 | } | |
2374 | } | |
2375 | } | |
2376 | ||
2377 | static int | |
2378 | mirror_get_stats(struct ofproto *ofproto_, void *aux, | |
2379 | uint64_t *packets, uint64_t *bytes) | |
2380 | { | |
2381 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2382 | struct ofmirror *mirror = mirror_lookup(ofproto, aux); | |
2383 | ||
2384 | if (!mirror) { | |
2385 | *packets = *bytes = UINT64_MAX; | |
2386 | return 0; | |
2387 | } | |
2388 | ||
2389 | *packets = mirror->packet_count; | |
2390 | *bytes = mirror->byte_count; | |
2391 | ||
2392 | return 0; | |
2393 | } | |
2394 | ||
2395 | static int | |
2396 | set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans) | |
2397 | { | |
2398 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2399 | if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) { | |
2400 | mac_learning_flush(ofproto->ml, &ofproto->revalidate_set); | |
2401 | } | |
2402 | return 0; | |
2403 | } | |
2404 | ||
2405 | static bool | |
2406 | is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux) | |
2407 | { | |
2408 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2409 | struct ofbundle *bundle = bundle_lookup(ofproto, aux); | |
2410 | return bundle && bundle->mirror_out != 0; | |
2411 | } | |
2412 | ||
2413 | static void | |
2414 | forward_bpdu_changed(struct ofproto *ofproto_) | |
2415 | { | |
2416 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2417 | ofproto->need_revalidate = REV_RECONFIGURE; | |
2418 | } | |
2419 | ||
2420 | static void | |
2421 | set_mac_idle_time(struct ofproto *ofproto_, unsigned int idle_time) | |
2422 | { | |
2423 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2424 | mac_learning_set_idle_time(ofproto->ml, idle_time); | |
2425 | } | |
2426 | \f | |
2427 | /* Ports. */ | |
2428 | ||
2429 | static struct ofport_dpif * | |
2430 | get_ofp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port) | |
2431 | { | |
2432 | struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port); | |
2433 | return ofport ? ofport_dpif_cast(ofport) : NULL; | |
2434 | } | |
2435 | ||
2436 | static struct ofport_dpif * | |
2437 | get_odp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port) | |
2438 | { | |
2439 | return get_ofp_port(ofproto, odp_port_to_ofp_port(odp_port)); | |
2440 | } | |
2441 | ||
2442 | static void | |
2443 | ofproto_port_from_dpif_port(struct ofproto_port *ofproto_port, | |
2444 | struct dpif_port *dpif_port) | |
2445 | { | |
2446 | ofproto_port->name = dpif_port->name; | |
2447 | ofproto_port->type = dpif_port->type; | |
2448 | ofproto_port->ofp_port = odp_port_to_ofp_port(dpif_port->port_no); | |
2449 | } | |
2450 | ||
2451 | static void | |
2452 | port_run_fast(struct ofport_dpif *ofport) | |
2453 | { | |
2454 | if (ofport->cfm && cfm_should_send_ccm(ofport->cfm)) { | |
2455 | struct ofpbuf packet; | |
2456 | ||
2457 | ofpbuf_init(&packet, 0); | |
2458 | cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr); | |
2459 | send_packet(ofport, &packet); | |
2460 | ofpbuf_uninit(&packet); | |
2461 | } | |
2462 | } | |
2463 | ||
2464 | static void | |
2465 | port_run(struct ofport_dpif *ofport) | |
2466 | { | |
2467 | long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev); | |
2468 | bool carrier_changed = carrier_seq != ofport->carrier_seq; | |
2469 | bool enable = netdev_get_carrier(ofport->up.netdev); | |
2470 | ||
2471 | ofport->carrier_seq = carrier_seq; | |
2472 | ||
2473 | port_run_fast(ofport); | |
2474 | if (ofport->cfm) { | |
2475 | int cfm_opup = cfm_get_opup(ofport->cfm); | |
2476 | ||
2477 | cfm_run(ofport->cfm); | |
2478 | enable = enable && !cfm_get_fault(ofport->cfm); | |
2479 | ||
2480 | if (cfm_opup >= 0) { | |
2481 | enable = enable && cfm_opup; | |
2482 | } | |
2483 | } | |
2484 | ||
2485 | if (ofport->bundle) { | |
2486 | enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport); | |
2487 | if (carrier_changed) { | |
2488 | lacp_slave_carrier_changed(ofport->bundle->lacp, ofport); | |
2489 | } | |
2490 | } | |
2491 | ||
2492 | if (ofport->may_enable != enable) { | |
2493 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
2494 | ||
2495 | if (ofproto->has_bundle_action) { | |
2496 | ofproto->need_revalidate = REV_PORT_TOGGLED; | |
2497 | } | |
2498 | } | |
2499 | ||
2500 | ofport->may_enable = enable; | |
2501 | } | |
2502 | ||
2503 | static void | |
2504 | port_wait(struct ofport_dpif *ofport) | |
2505 | { | |
2506 | if (ofport->cfm) { | |
2507 | cfm_wait(ofport->cfm); | |
2508 | } | |
2509 | } | |
2510 | ||
2511 | static int | |
2512 | port_query_by_name(const struct ofproto *ofproto_, const char *devname, | |
2513 | struct ofproto_port *ofproto_port) | |
2514 | { | |
2515 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2516 | struct dpif_port dpif_port; | |
2517 | int error; | |
2518 | ||
2519 | error = dpif_port_query_by_name(ofproto->dpif, devname, &dpif_port); | |
2520 | if (!error) { | |
2521 | ofproto_port_from_dpif_port(ofproto_port, &dpif_port); | |
2522 | } | |
2523 | return error; | |
2524 | } | |
2525 | ||
2526 | static int | |
2527 | port_add(struct ofproto *ofproto_, struct netdev *netdev, uint16_t *ofp_portp) | |
2528 | { | |
2529 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2530 | uint16_t odp_port = UINT16_MAX; | |
2531 | int error; | |
2532 | ||
2533 | error = dpif_port_add(ofproto->dpif, netdev, &odp_port); | |
2534 | if (!error) { | |
2535 | *ofp_portp = odp_port_to_ofp_port(odp_port); | |
2536 | } | |
2537 | return error; | |
2538 | } | |
2539 | ||
2540 | static int | |
2541 | port_del(struct ofproto *ofproto_, uint16_t ofp_port) | |
2542 | { | |
2543 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2544 | int error; | |
2545 | ||
2546 | error = dpif_port_del(ofproto->dpif, ofp_port_to_odp_port(ofp_port)); | |
2547 | if (!error) { | |
2548 | struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port); | |
2549 | if (ofport) { | |
2550 | /* The caller is going to close ofport->up.netdev. If this is a | |
2551 | * bonded port, then the bond is using that netdev, so remove it | |
2552 | * from the bond. The client will need to reconfigure everything | |
2553 | * after deleting ports, so then the slave will get re-added. */ | |
2554 | bundle_remove(&ofport->up); | |
2555 | } | |
2556 | } | |
2557 | return error; | |
2558 | } | |
2559 | ||
2560 | static int | |
2561 | port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats) | |
2562 | { | |
2563 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
2564 | int error; | |
2565 | ||
2566 | error = netdev_get_stats(ofport->up.netdev, stats); | |
2567 | ||
2568 | if (!error && ofport->odp_port == OVSP_LOCAL) { | |
2569 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
2570 | ||
2571 | /* ofproto->stats.tx_packets represents packets that we created | |
2572 | * internally and sent to some port (e.g. packets sent with | |
2573 | * send_packet()). Account for them as if they had come from | |
2574 | * OFPP_LOCAL and got forwarded. */ | |
2575 | ||
2576 | if (stats->rx_packets != UINT64_MAX) { | |
2577 | stats->rx_packets += ofproto->stats.tx_packets; | |
2578 | } | |
2579 | ||
2580 | if (stats->rx_bytes != UINT64_MAX) { | |
2581 | stats->rx_bytes += ofproto->stats.tx_bytes; | |
2582 | } | |
2583 | ||
2584 | /* ofproto->stats.rx_packets represents packets that were received on | |
2585 | * some port and we processed internally and dropped (e.g. STP). | |
2586 | * Account fro them as if they had been forwarded to OFPP_LOCAL. */ | |
2587 | ||
2588 | if (stats->tx_packets != UINT64_MAX) { | |
2589 | stats->tx_packets += ofproto->stats.rx_packets; | |
2590 | } | |
2591 | ||
2592 | if (stats->tx_bytes != UINT64_MAX) { | |
2593 | stats->tx_bytes += ofproto->stats.rx_bytes; | |
2594 | } | |
2595 | } | |
2596 | ||
2597 | return error; | |
2598 | } | |
2599 | ||
2600 | /* Account packets for LOCAL port. */ | |
2601 | static void | |
2602 | ofproto_update_local_port_stats(const struct ofproto *ofproto_, | |
2603 | size_t tx_size, size_t rx_size) | |
2604 | { | |
2605 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2606 | ||
2607 | if (rx_size) { | |
2608 | ofproto->stats.rx_packets++; | |
2609 | ofproto->stats.rx_bytes += rx_size; | |
2610 | } | |
2611 | if (tx_size) { | |
2612 | ofproto->stats.tx_packets++; | |
2613 | ofproto->stats.tx_bytes += tx_size; | |
2614 | } | |
2615 | } | |
2616 | ||
2617 | struct port_dump_state { | |
2618 | struct dpif_port_dump dump; | |
2619 | bool done; | |
2620 | }; | |
2621 | ||
2622 | static int | |
2623 | port_dump_start(const struct ofproto *ofproto_, void **statep) | |
2624 | { | |
2625 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2626 | struct port_dump_state *state; | |
2627 | ||
2628 | *statep = state = xmalloc(sizeof *state); | |
2629 | dpif_port_dump_start(&state->dump, ofproto->dpif); | |
2630 | state->done = false; | |
2631 | return 0; | |
2632 | } | |
2633 | ||
2634 | static int | |
2635 | port_dump_next(const struct ofproto *ofproto_ OVS_UNUSED, void *state_, | |
2636 | struct ofproto_port *port) | |
2637 | { | |
2638 | struct port_dump_state *state = state_; | |
2639 | struct dpif_port dpif_port; | |
2640 | ||
2641 | if (dpif_port_dump_next(&state->dump, &dpif_port)) { | |
2642 | ofproto_port_from_dpif_port(port, &dpif_port); | |
2643 | return 0; | |
2644 | } else { | |
2645 | int error = dpif_port_dump_done(&state->dump); | |
2646 | state->done = true; | |
2647 | return error ? error : EOF; | |
2648 | } | |
2649 | } | |
2650 | ||
2651 | static int | |
2652 | port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_) | |
2653 | { | |
2654 | struct port_dump_state *state = state_; | |
2655 | ||
2656 | if (!state->done) { | |
2657 | dpif_port_dump_done(&state->dump); | |
2658 | } | |
2659 | free(state); | |
2660 | return 0; | |
2661 | } | |
2662 | ||
2663 | static int | |
2664 | port_poll(const struct ofproto *ofproto_, char **devnamep) | |
2665 | { | |
2666 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2667 | return dpif_port_poll(ofproto->dpif, devnamep); | |
2668 | } | |
2669 | ||
2670 | static void | |
2671 | port_poll_wait(const struct ofproto *ofproto_) | |
2672 | { | |
2673 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2674 | dpif_port_poll_wait(ofproto->dpif); | |
2675 | } | |
2676 | ||
2677 | static int | |
2678 | port_is_lacp_current(const struct ofport *ofport_) | |
2679 | { | |
2680 | const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
2681 | return (ofport->bundle && ofport->bundle->lacp | |
2682 | ? lacp_slave_is_current(ofport->bundle->lacp, ofport) | |
2683 | : -1); | |
2684 | } | |
2685 | \f | |
2686 | /* Upcall handling. */ | |
2687 | ||
2688 | /* Flow miss batching. | |
2689 | * | |
2690 | * Some dpifs implement operations faster when you hand them off in a batch. | |
2691 | * To allow batching, "struct flow_miss" queues the dpif-related work needed | |
2692 | * for a given flow. Each "struct flow_miss" corresponds to sending one or | |
2693 | * more packets, plus possibly installing the flow in the dpif. | |
2694 | * | |
2695 | * So far we only batch the operations that affect flow setup time the most. | |
2696 | * It's possible to batch more than that, but the benefit might be minimal. */ | |
2697 | struct flow_miss { | |
2698 | struct hmap_node hmap_node; | |
2699 | struct flow flow; | |
2700 | enum odp_key_fitness key_fitness; | |
2701 | const struct nlattr *key; | |
2702 | size_t key_len; | |
2703 | ovs_be16 initial_tci; | |
2704 | struct list packets; | |
2705 | enum dpif_upcall_type upcall_type; | |
2706 | }; | |
2707 | ||
2708 | struct flow_miss_op { | |
2709 | struct dpif_op dpif_op; | |
2710 | struct subfacet *subfacet; /* Subfacet */ | |
2711 | void *garbage; /* Pointer to pass to free(), NULL if none. */ | |
2712 | uint64_t stub[1024 / 8]; /* Temporary buffer. */ | |
2713 | }; | |
2714 | ||
2715 | /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each | |
2716 | * OpenFlow controller as necessary according to their individual | |
2717 | * configurations. */ | |
2718 | static void | |
2719 | send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet, | |
2720 | const struct flow *flow) | |
2721 | { | |
2722 | struct ofputil_packet_in pin; | |
2723 | ||
2724 | pin.packet = packet->data; | |
2725 | pin.packet_len = packet->size; | |
2726 | pin.reason = OFPR_NO_MATCH; | |
2727 | pin.controller_id = 0; | |
2728 | ||
2729 | pin.table_id = 0; | |
2730 | pin.cookie = 0; | |
2731 | ||
2732 | pin.send_len = 0; /* not used for flow table misses */ | |
2733 | ||
2734 | flow_get_metadata(flow, &pin.fmd); | |
2735 | ||
2736 | /* Registers aren't meaningful on a miss. */ | |
2737 | memset(pin.fmd.reg_masks, 0, sizeof pin.fmd.reg_masks); | |
2738 | ||
2739 | connmgr_send_packet_in(ofproto->up.connmgr, &pin); | |
2740 | } | |
2741 | ||
2742 | static enum slow_path_reason | |
2743 | process_special(struct ofproto_dpif *ofproto, const struct flow *flow, | |
2744 | const struct ofpbuf *packet) | |
2745 | { | |
2746 | struct ofport_dpif *ofport = get_ofp_port(ofproto, flow->in_port); | |
2747 | ||
2748 | if (!ofport) { | |
2749 | return 0; | |
2750 | } | |
2751 | ||
2752 | if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) { | |
2753 | if (packet) { | |
2754 | cfm_process_heartbeat(ofport->cfm, packet); | |
2755 | } | |
2756 | return SLOW_CFM; | |
2757 | } else if (ofport->bundle && ofport->bundle->lacp | |
2758 | && flow->dl_type == htons(ETH_TYPE_LACP)) { | |
2759 | if (packet) { | |
2760 | lacp_process_packet(ofport->bundle->lacp, ofport, packet); | |
2761 | } | |
2762 | return SLOW_LACP; | |
2763 | } else if (ofproto->stp && stp_should_process_flow(flow)) { | |
2764 | if (packet) { | |
2765 | stp_process_packet(ofport, packet); | |
2766 | } | |
2767 | return SLOW_STP; | |
2768 | } | |
2769 | return 0; | |
2770 | } | |
2771 | ||
2772 | static struct flow_miss * | |
2773 | flow_miss_find(struct hmap *todo, const struct flow *flow, uint32_t hash) | |
2774 | { | |
2775 | struct flow_miss *miss; | |
2776 | ||
2777 | HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) { | |
2778 | if (flow_equal(&miss->flow, flow)) { | |
2779 | return miss; | |
2780 | } | |
2781 | } | |
2782 | ||
2783 | return NULL; | |
2784 | } | |
2785 | ||
2786 | /* Partially Initializes 'op' as an "execute" operation for 'miss' and | |
2787 | * 'packet'. The caller must initialize op->actions and op->actions_len. If | |
2788 | * 'miss' is associated with a subfacet the caller must also initialize the | |
2789 | * returned op->subfacet, and if anything needs to be freed after processing | |
2790 | * the op, the caller must initialize op->garbage also. */ | |
2791 | static void | |
2792 | init_flow_miss_execute_op(struct flow_miss *miss, struct ofpbuf *packet, | |
2793 | struct flow_miss_op *op) | |
2794 | { | |
2795 | if (miss->flow.vlan_tci != miss->initial_tci) { | |
2796 | /* This packet was received on a VLAN splinter port. We | |
2797 | * added a VLAN to the packet to make the packet resemble | |
2798 | * the flow, but the actions were composed assuming that | |
2799 | * the packet contained no VLAN. So, we must remove the | |
2800 | * VLAN header from the packet before trying to execute the | |
2801 | * actions. */ | |
2802 | eth_pop_vlan(packet); | |
2803 | } | |
2804 | ||
2805 | op->subfacet = NULL; | |
2806 | op->garbage = NULL; | |
2807 | op->dpif_op.type = DPIF_OP_EXECUTE; | |
2808 | op->dpif_op.u.execute.key = miss->key; | |
2809 | op->dpif_op.u.execute.key_len = miss->key_len; | |
2810 | op->dpif_op.u.execute.packet = packet; | |
2811 | } | |
2812 | ||
2813 | /* Helper for handle_flow_miss_without_facet() and | |
2814 | * handle_flow_miss_with_facet(). */ | |
2815 | static void | |
2816 | handle_flow_miss_common(struct rule_dpif *rule, | |
2817 | struct ofpbuf *packet, const struct flow *flow) | |
2818 | { | |
2819 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
2820 | ||
2821 | ofproto->n_matches++; | |
2822 | ||
2823 | if (rule->up.cr.priority == FAIL_OPEN_PRIORITY) { | |
2824 | /* | |
2825 | * Extra-special case for fail-open mode. | |
2826 | * | |
2827 | * We are in fail-open mode and the packet matched the fail-open | |
2828 | * rule, but we are connected to a controller too. We should send | |
2829 | * the packet up to the controller in the hope that it will try to | |
2830 | * set up a flow and thereby allow us to exit fail-open. | |
2831 | * | |
2832 | * See the top-level comment in fail-open.c for more information. | |
2833 | */ | |
2834 | send_packet_in_miss(ofproto, packet, flow); | |
2835 | } | |
2836 | } | |
2837 | ||
2838 | /* Figures out whether a flow that missed in 'ofproto', whose details are in | |
2839 | * 'miss', is likely to be worth tracking in detail in userspace and (usually) | |
2840 | * installing a datapath flow. The answer is usually "yes" (a return value of | |
2841 | * true). However, for short flows the cost of bookkeeping is much higher than | |
2842 | * the benefits, so when the datapath holds a large number of flows we impose | |
2843 | * some heuristics to decide which flows are likely to be worth tracking. */ | |
2844 | static bool | |
2845 | flow_miss_should_make_facet(struct ofproto_dpif *ofproto, | |
2846 | struct flow_miss *miss, uint32_t hash) | |
2847 | { | |
2848 | if (!ofproto->governor) { | |
2849 | size_t n_subfacets; | |
2850 | ||
2851 | n_subfacets = hmap_count(&ofproto->subfacets); | |
2852 | if (n_subfacets * 2 <= ofproto->up.flow_eviction_threshold) { | |
2853 | return true; | |
2854 | } | |
2855 | ||
2856 | ofproto->governor = governor_create(ofproto->up.name); | |
2857 | } | |
2858 | ||
2859 | return governor_should_install_flow(ofproto->governor, hash, | |
2860 | list_size(&miss->packets)); | |
2861 | } | |
2862 | ||
2863 | /* Handles 'miss', which matches 'rule', without creating a facet or subfacet | |
2864 | * or creating any datapath flow. May add an "execute" operation to 'ops' and | |
2865 | * increment '*n_ops'. */ | |
2866 | static void | |
2867 | handle_flow_miss_without_facet(struct flow_miss *miss, | |
2868 | struct rule_dpif *rule, | |
2869 | struct flow_miss_op *ops, size_t *n_ops) | |
2870 | { | |
2871 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
2872 | struct action_xlate_ctx ctx; | |
2873 | struct ofpbuf *packet; | |
2874 | ||
2875 | LIST_FOR_EACH (packet, list_node, &miss->packets) { | |
2876 | struct flow_miss_op *op = &ops[*n_ops]; | |
2877 | struct dpif_flow_stats stats; | |
2878 | struct ofpbuf odp_actions; | |
2879 | ||
2880 | COVERAGE_INC(facet_suppress); | |
2881 | ||
2882 | ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub); | |
2883 | ||
2884 | dpif_flow_stats_extract(&miss->flow, packet, &stats); | |
2885 | rule_credit_stats(rule, &stats); | |
2886 | ||
2887 | action_xlate_ctx_init(&ctx, ofproto, &miss->flow, miss->initial_tci, | |
2888 | rule, 0, packet); | |
2889 | ctx.resubmit_stats = &stats; | |
2890 | xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, | |
2891 | &odp_actions); | |
2892 | ||
2893 | if (odp_actions.size) { | |
2894 | struct dpif_execute *execute = &op->dpif_op.u.execute; | |
2895 | ||
2896 | init_flow_miss_execute_op(miss, packet, op); | |
2897 | execute->actions = odp_actions.data; | |
2898 | execute->actions_len = odp_actions.size; | |
2899 | op->garbage = ofpbuf_get_uninit_pointer(&odp_actions); | |
2900 | ||
2901 | (*n_ops)++; | |
2902 | } else { | |
2903 | ofpbuf_uninit(&odp_actions); | |
2904 | } | |
2905 | } | |
2906 | } | |
2907 | ||
2908 | /* Handles 'miss', which matches 'facet'. May add any required datapath | |
2909 | * operations to 'ops', incrementing '*n_ops' for each new op. */ | |
2910 | static void | |
2911 | handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet, | |
2912 | struct flow_miss_op *ops, size_t *n_ops) | |
2913 | { | |
2914 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); | |
2915 | enum subfacet_path want_path; | |
2916 | struct subfacet *subfacet; | |
2917 | struct ofpbuf *packet; | |
2918 | ||
2919 | subfacet = subfacet_create(facet, | |
2920 | miss->key_fitness, miss->key, miss->key_len, | |
2921 | miss->initial_tci); | |
2922 | ||
2923 | LIST_FOR_EACH (packet, list_node, &miss->packets) { | |
2924 | struct flow_miss_op *op = &ops[*n_ops]; | |
2925 | struct dpif_flow_stats stats; | |
2926 | struct ofpbuf odp_actions; | |
2927 | ||
2928 | handle_flow_miss_common(facet->rule, packet, &miss->flow); | |
2929 | ||
2930 | ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub); | |
2931 | if (!subfacet->actions || subfacet->slow) { | |
2932 | subfacet_make_actions(subfacet, packet, &odp_actions); | |
2933 | } | |
2934 | ||
2935 | dpif_flow_stats_extract(&facet->flow, packet, &stats); | |
2936 | subfacet_update_stats(subfacet, &stats); | |
2937 | ||
2938 | if (subfacet->actions_len) { | |
2939 | struct dpif_execute *execute = &op->dpif_op.u.execute; | |
2940 | ||
2941 | init_flow_miss_execute_op(miss, packet, op); | |
2942 | op->subfacet = subfacet; | |
2943 | if (!subfacet->slow) { | |
2944 | execute->actions = subfacet->actions; | |
2945 | execute->actions_len = subfacet->actions_len; | |
2946 | ofpbuf_uninit(&odp_actions); | |
2947 | } else { | |
2948 | execute->actions = odp_actions.data; | |
2949 | execute->actions_len = odp_actions.size; | |
2950 | op->garbage = ofpbuf_get_uninit_pointer(&odp_actions); | |
2951 | } | |
2952 | ||
2953 | (*n_ops)++; | |
2954 | } else { | |
2955 | ofpbuf_uninit(&odp_actions); | |
2956 | } | |
2957 | } | |
2958 | ||
2959 | want_path = subfacet_want_path(subfacet->slow); | |
2960 | if (miss->upcall_type == DPIF_UC_MISS || subfacet->path != want_path) { | |
2961 | struct flow_miss_op *op = &ops[(*n_ops)++]; | |
2962 | struct dpif_flow_put *put = &op->dpif_op.u.flow_put; | |
2963 | ||
2964 | op->subfacet = subfacet; | |
2965 | op->garbage = NULL; | |
2966 | op->dpif_op.type = DPIF_OP_FLOW_PUT; | |
2967 | put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY; | |
2968 | put->key = miss->key; | |
2969 | put->key_len = miss->key_len; | |
2970 | if (want_path == SF_FAST_PATH) { | |
2971 | put->actions = subfacet->actions; | |
2972 | put->actions_len = subfacet->actions_len; | |
2973 | } else { | |
2974 | compose_slow_path(ofproto, &facet->flow, subfacet->slow, | |
2975 | op->stub, sizeof op->stub, | |
2976 | &put->actions, &put->actions_len); | |
2977 | } | |
2978 | put->stats = NULL; | |
2979 | } | |
2980 | } | |
2981 | ||
2982 | /* Handles flow miss 'miss' on 'ofproto'. May add any required datapath | |
2983 | * operations to 'ops', incrementing '*n_ops' for each new op. */ | |
2984 | static void | |
2985 | handle_flow_miss(struct ofproto_dpif *ofproto, struct flow_miss *miss, | |
2986 | struct flow_miss_op *ops, size_t *n_ops) | |
2987 | { | |
2988 | struct facet *facet; | |
2989 | uint32_t hash; | |
2990 | ||
2991 | /* The caller must ensure that miss->hmap_node.hash contains | |
2992 | * flow_hash(miss->flow, 0). */ | |
2993 | hash = miss->hmap_node.hash; | |
2994 | ||
2995 | facet = facet_lookup_valid(ofproto, &miss->flow, hash); | |
2996 | if (!facet) { | |
2997 | struct rule_dpif *rule = rule_dpif_lookup(ofproto, &miss->flow); | |
2998 | ||
2999 | if (!flow_miss_should_make_facet(ofproto, miss, hash)) { | |
3000 | handle_flow_miss_without_facet(miss, rule, ops, n_ops); | |
3001 | return; | |
3002 | } | |
3003 | ||
3004 | facet = facet_create(rule, &miss->flow, hash); | |
3005 | } | |
3006 | handle_flow_miss_with_facet(miss, facet, ops, n_ops); | |
3007 | } | |
3008 | ||
3009 | /* Like odp_flow_key_to_flow(), this function converts the 'key_len' bytes of | |
3010 | * OVS_KEY_ATTR_* attributes in 'key' to a flow structure in 'flow' and returns | |
3011 | * an ODP_FIT_* value that indicates how well 'key' fits our expectations for | |
3012 | * what a flow key should contain. | |
3013 | * | |
3014 | * This function also includes some logic to help make VLAN splinters | |
3015 | * transparent to the rest of the upcall processing logic. In particular, if | |
3016 | * the extracted in_port is a VLAN splinter port, it replaces flow->in_port by | |
3017 | * the "real" port, sets flow->vlan_tci correctly for the VLAN of the VLAN | |
3018 | * splinter port, and pushes a VLAN header onto 'packet' (if it is nonnull). | |
3019 | * | |
3020 | * Sets '*initial_tci' to the VLAN TCI with which the packet was really | |
3021 | * received, that is, the actual VLAN TCI extracted by odp_flow_key_to_flow(). | |
3022 | * (This differs from the value returned in flow->vlan_tci only for packets | |
3023 | * received on VLAN splinters.) | |
3024 | */ | |
3025 | static enum odp_key_fitness | |
3026 | ofproto_dpif_extract_flow_key(const struct ofproto_dpif *ofproto, | |
3027 | const struct nlattr *key, size_t key_len, | |
3028 | struct flow *flow, ovs_be16 *initial_tci, | |
3029 | struct ofpbuf *packet) | |
3030 | { | |
3031 | enum odp_key_fitness fitness; | |
3032 | ||
3033 | fitness = odp_flow_key_to_flow(key, key_len, flow); | |
3034 | if (fitness == ODP_FIT_ERROR) { | |
3035 | return fitness; | |
3036 | } | |
3037 | *initial_tci = flow->vlan_tci; | |
3038 | ||
3039 | if (vsp_adjust_flow(ofproto, flow)) { | |
3040 | if (packet) { | |
3041 | /* Make the packet resemble the flow, so that it gets sent to an | |
3042 | * OpenFlow controller properly, so that it looks correct for | |
3043 | * sFlow, and so that flow_extract() will get the correct vlan_tci | |
3044 | * if it is called on 'packet'. | |
3045 | * | |
3046 | * The allocated space inside 'packet' probably also contains | |
3047 | * 'key', that is, both 'packet' and 'key' are probably part of a | |
3048 | * struct dpif_upcall (see the large comment on that structure | |
3049 | * definition), so pushing data on 'packet' is in general not a | |
3050 | * good idea since it could overwrite 'key' or free it as a side | |
3051 | * effect. However, it's OK in this special case because we know | |
3052 | * that 'packet' is inside a Netlink attribute: pushing 4 bytes | |
3053 | * will just overwrite the 4-byte "struct nlattr", which is fine | |
3054 | * since we don't need that header anymore. */ | |
3055 | eth_push_vlan(packet, flow->vlan_tci); | |
3056 | } | |
3057 | ||
3058 | /* Let the caller know that we can't reproduce 'key' from 'flow'. */ | |
3059 | if (fitness == ODP_FIT_PERFECT) { | |
3060 | fitness = ODP_FIT_TOO_MUCH; | |
3061 | } | |
3062 | } | |
3063 | ||
3064 | return fitness; | |
3065 | } | |
3066 | ||
3067 | static void | |
3068 | handle_miss_upcalls(struct ofproto_dpif *ofproto, struct dpif_upcall *upcalls, | |
3069 | size_t n_upcalls) | |
3070 | { | |
3071 | struct dpif_upcall *upcall; | |
3072 | struct flow_miss *miss; | |
3073 | struct flow_miss misses[FLOW_MISS_MAX_BATCH]; | |
3074 | struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2]; | |
3075 | struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2]; | |
3076 | struct hmap todo; | |
3077 | int n_misses; | |
3078 | size_t n_ops; | |
3079 | size_t i; | |
3080 | ||
3081 | if (!n_upcalls) { | |
3082 | return; | |
3083 | } | |
3084 | ||
3085 | /* Construct the to-do list. | |
3086 | * | |
3087 | * This just amounts to extracting the flow from each packet and sticking | |
3088 | * the packets that have the same flow in the same "flow_miss" structure so | |
3089 | * that we can process them together. */ | |
3090 | hmap_init(&todo); | |
3091 | n_misses = 0; | |
3092 | for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) { | |
3093 | struct flow_miss *miss = &misses[n_misses]; | |
3094 | struct flow_miss *existing_miss; | |
3095 | uint32_t hash; | |
3096 | ||
3097 | /* Obtain metadata and check userspace/kernel agreement on flow match, | |
3098 | * then set 'flow''s header pointers. */ | |
3099 | miss->key_fitness = ofproto_dpif_extract_flow_key( | |
3100 | ofproto, upcall->key, upcall->key_len, | |
3101 | &miss->flow, &miss->initial_tci, upcall->packet); | |
3102 | if (miss->key_fitness == ODP_FIT_ERROR) { | |
3103 | continue; | |
3104 | } | |
3105 | flow_extract(upcall->packet, miss->flow.skb_priority, | |
3106 | miss->flow.tun_id, miss->flow.in_port, &miss->flow); | |
3107 | ||
3108 | /* Add other packets to a to-do list. */ | |
3109 | hash = flow_hash(&miss->flow, 0); | |
3110 | existing_miss = flow_miss_find(&todo, &miss->flow, hash); | |
3111 | if (!existing_miss) { | |
3112 | hmap_insert(&todo, &miss->hmap_node, hash); | |
3113 | miss->key = upcall->key; | |
3114 | miss->key_len = upcall->key_len; | |
3115 | miss->upcall_type = upcall->type; | |
3116 | list_init(&miss->packets); | |
3117 | ||
3118 | n_misses++; | |
3119 | } else { | |
3120 | miss = existing_miss; | |
3121 | } | |
3122 | list_push_back(&miss->packets, &upcall->packet->list_node); | |
3123 | } | |
3124 | ||
3125 | /* Process each element in the to-do list, constructing the set of | |
3126 | * operations to batch. */ | |
3127 | n_ops = 0; | |
3128 | HMAP_FOR_EACH (miss, hmap_node, &todo) { | |
3129 | handle_flow_miss(ofproto, miss, flow_miss_ops, &n_ops); | |
3130 | } | |
3131 | assert(n_ops <= ARRAY_SIZE(flow_miss_ops)); | |
3132 | ||
3133 | /* Execute batch. */ | |
3134 | for (i = 0; i < n_ops; i++) { | |
3135 | dpif_ops[i] = &flow_miss_ops[i].dpif_op; | |
3136 | } | |
3137 | dpif_operate(ofproto->dpif, dpif_ops, n_ops); | |
3138 | ||
3139 | /* Free memory and update facets. */ | |
3140 | for (i = 0; i < n_ops; i++) { | |
3141 | struct flow_miss_op *op = &flow_miss_ops[i]; | |
3142 | ||
3143 | switch (op->dpif_op.type) { | |
3144 | case DPIF_OP_EXECUTE: | |
3145 | break; | |
3146 | ||
3147 | case DPIF_OP_FLOW_PUT: | |
3148 | if (!op->dpif_op.error) { | |
3149 | op->subfacet->path = subfacet_want_path(op->subfacet->slow); | |
3150 | } | |
3151 | break; | |
3152 | ||
3153 | case DPIF_OP_FLOW_DEL: | |
3154 | NOT_REACHED(); | |
3155 | } | |
3156 | ||
3157 | free(op->garbage); | |
3158 | } | |
3159 | hmap_destroy(&todo); | |
3160 | } | |
3161 | ||
3162 | static enum { SFLOW_UPCALL, MISS_UPCALL, BAD_UPCALL } | |
3163 | classify_upcall(const struct dpif_upcall *upcall) | |
3164 | { | |
3165 | union user_action_cookie cookie; | |
3166 | ||
3167 | /* First look at the upcall type. */ | |
3168 | switch (upcall->type) { | |
3169 | case DPIF_UC_ACTION: | |
3170 | break; | |
3171 | ||
3172 | case DPIF_UC_MISS: | |
3173 | return MISS_UPCALL; | |
3174 | ||
3175 | case DPIF_N_UC_TYPES: | |
3176 | default: | |
3177 | VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type); | |
3178 | return BAD_UPCALL; | |
3179 | } | |
3180 | ||
3181 | /* "action" upcalls need a closer look. */ | |
3182 | memcpy(&cookie, &upcall->userdata, sizeof(cookie)); | |
3183 | switch (cookie.type) { | |
3184 | case USER_ACTION_COOKIE_SFLOW: | |
3185 | return SFLOW_UPCALL; | |
3186 | ||
3187 | case USER_ACTION_COOKIE_SLOW_PATH: | |
3188 | return MISS_UPCALL; | |
3189 | ||
3190 | case USER_ACTION_COOKIE_UNSPEC: | |
3191 | default: | |
3192 | VLOG_WARN_RL(&rl, "invalid user cookie : 0x%"PRIx64, upcall->userdata); | |
3193 | return BAD_UPCALL; | |
3194 | } | |
3195 | } | |
3196 | ||
3197 | static void | |
3198 | handle_sflow_upcall(struct ofproto_dpif *ofproto, | |
3199 | const struct dpif_upcall *upcall) | |
3200 | { | |
3201 | union user_action_cookie cookie; | |
3202 | enum odp_key_fitness fitness; | |
3203 | ovs_be16 initial_tci; | |
3204 | struct flow flow; | |
3205 | ||
3206 | fitness = ofproto_dpif_extract_flow_key(ofproto, upcall->key, | |
3207 | upcall->key_len, &flow, | |
3208 | &initial_tci, upcall->packet); | |
3209 | if (fitness == ODP_FIT_ERROR) { | |
3210 | return; | |
3211 | } | |
3212 | ||
3213 | memcpy(&cookie, &upcall->userdata, sizeof(cookie)); | |
3214 | dpif_sflow_received(ofproto->sflow, upcall->packet, &flow, &cookie); | |
3215 | } | |
3216 | ||
3217 | static int | |
3218 | handle_upcalls(struct ofproto_dpif *ofproto, unsigned int max_batch) | |
3219 | { | |
3220 | struct dpif_upcall misses[FLOW_MISS_MAX_BATCH]; | |
3221 | struct ofpbuf miss_bufs[FLOW_MISS_MAX_BATCH]; | |
3222 | uint64_t miss_buf_stubs[FLOW_MISS_MAX_BATCH][4096 / 8]; | |
3223 | int n_processed; | |
3224 | int n_misses; | |
3225 | int i; | |
3226 | ||
3227 | assert(max_batch <= FLOW_MISS_MAX_BATCH); | |
3228 | ||
3229 | n_misses = 0; | |
3230 | for (n_processed = 0; n_processed < max_batch; n_processed++) { | |
3231 | struct dpif_upcall *upcall = &misses[n_misses]; | |
3232 | struct ofpbuf *buf = &miss_bufs[n_misses]; | |
3233 | int error; | |
3234 | ||
3235 | ofpbuf_use_stub(buf, miss_buf_stubs[n_misses], | |
3236 | sizeof miss_buf_stubs[n_misses]); | |
3237 | error = dpif_recv(ofproto->dpif, upcall, buf); | |
3238 | if (error) { | |
3239 | ofpbuf_uninit(buf); | |
3240 | break; | |
3241 | } | |
3242 | ||
3243 | switch (classify_upcall(upcall)) { | |
3244 | case MISS_UPCALL: | |
3245 | /* Handle it later. */ | |
3246 | n_misses++; | |
3247 | break; | |
3248 | ||
3249 | case SFLOW_UPCALL: | |
3250 | if (ofproto->sflow) { | |
3251 | handle_sflow_upcall(ofproto, upcall); | |
3252 | } | |
3253 | ofpbuf_uninit(buf); | |
3254 | break; | |
3255 | ||
3256 | case BAD_UPCALL: | |
3257 | ofpbuf_uninit(buf); | |
3258 | break; | |
3259 | } | |
3260 | } | |
3261 | ||
3262 | /* Handle deferred MISS_UPCALL processing. */ | |
3263 | handle_miss_upcalls(ofproto, misses, n_misses); | |
3264 | for (i = 0; i < n_misses; i++) { | |
3265 | ofpbuf_uninit(&miss_bufs[i]); | |
3266 | } | |
3267 | ||
3268 | return n_processed; | |
3269 | } | |
3270 | \f | |
3271 | /* Flow expiration. */ | |
3272 | ||
3273 | static int subfacet_max_idle(const struct ofproto_dpif *); | |
3274 | static void update_stats(struct ofproto_dpif *); | |
3275 | static void rule_expire(struct rule_dpif *); | |
3276 | static void expire_subfacets(struct ofproto_dpif *, int dp_max_idle); | |
3277 | ||
3278 | /* This function is called periodically by run(). Its job is to collect | |
3279 | * updates for the flows that have been installed into the datapath, most | |
3280 | * importantly when they last were used, and then use that information to | |
3281 | * expire flows that have not been used recently. | |
3282 | * | |
3283 | * Returns the number of milliseconds after which it should be called again. */ | |
3284 | static int | |
3285 | expire(struct ofproto_dpif *ofproto) | |
3286 | { | |
3287 | struct rule_dpif *rule, *next_rule; | |
3288 | struct oftable *table; | |
3289 | int dp_max_idle; | |
3290 | ||
3291 | /* Update stats for each flow in the datapath. */ | |
3292 | update_stats(ofproto); | |
3293 | ||
3294 | /* Expire subfacets that have been idle too long. */ | |
3295 | dp_max_idle = subfacet_max_idle(ofproto); | |
3296 | expire_subfacets(ofproto, dp_max_idle); | |
3297 | ||
3298 | /* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */ | |
3299 | OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) { | |
3300 | struct cls_cursor cursor; | |
3301 | ||
3302 | cls_cursor_init(&cursor, &table->cls, NULL); | |
3303 | CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) { | |
3304 | rule_expire(rule); | |
3305 | } | |
3306 | } | |
3307 | ||
3308 | /* All outstanding data in existing flows has been accounted, so it's a | |
3309 | * good time to do bond rebalancing. */ | |
3310 | if (ofproto->has_bonded_bundles) { | |
3311 | struct ofbundle *bundle; | |
3312 | ||
3313 | HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { | |
3314 | if (bundle->bond) { | |
3315 | bond_rebalance(bundle->bond, &ofproto->revalidate_set); | |
3316 | } | |
3317 | } | |
3318 | } | |
3319 | ||
3320 | return MIN(dp_max_idle, 1000); | |
3321 | } | |
3322 | ||
3323 | /* Updates flow table statistics given that the datapath just reported 'stats' | |
3324 | * as 'subfacet''s statistics. */ | |
3325 | static void | |
3326 | update_subfacet_stats(struct subfacet *subfacet, | |
3327 | const struct dpif_flow_stats *stats) | |
3328 | { | |
3329 | struct facet *facet = subfacet->facet; | |
3330 | ||
3331 | if (stats->n_packets >= subfacet->dp_packet_count) { | |
3332 | uint64_t extra = stats->n_packets - subfacet->dp_packet_count; | |
3333 | facet->packet_count += extra; | |
3334 | } else { | |
3335 | VLOG_WARN_RL(&rl, "unexpected packet count from the datapath"); | |
3336 | } | |
3337 | ||
3338 | if (stats->n_bytes >= subfacet->dp_byte_count) { | |
3339 | facet->byte_count += stats->n_bytes - subfacet->dp_byte_count; | |
3340 | } else { | |
3341 | VLOG_WARN_RL(&rl, "unexpected byte count from datapath"); | |
3342 | } | |
3343 | ||
3344 | subfacet->dp_packet_count = stats->n_packets; | |
3345 | subfacet->dp_byte_count = stats->n_bytes; | |
3346 | ||
3347 | facet->tcp_flags |= stats->tcp_flags; | |
3348 | ||
3349 | subfacet_update_time(subfacet, stats->used); | |
3350 | if (facet->accounted_bytes < facet->byte_count) { | |
3351 | facet_learn(facet); | |
3352 | facet_account(facet); | |
3353 | facet->accounted_bytes = facet->byte_count; | |
3354 | } | |
3355 | facet_push_stats(facet); | |
3356 | } | |
3357 | ||
3358 | /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing | |
3359 | * about, or a flow that shouldn't be installed but was anyway. Delete it. */ | |
3360 | static void | |
3361 | delete_unexpected_flow(struct dpif *dpif, | |
3362 | const struct nlattr *key, size_t key_len) | |
3363 | { | |
3364 | if (!VLOG_DROP_WARN(&rl)) { | |
3365 | struct ds s; | |
3366 | ||
3367 | ds_init(&s); | |
3368 | odp_flow_key_format(key, key_len, &s); | |
3369 | VLOG_WARN("unexpected flow from datapath %s", ds_cstr(&s)); | |
3370 | ds_destroy(&s); | |
3371 | } | |
3372 | ||
3373 | COVERAGE_INC(facet_unexpected); | |
3374 | dpif_flow_del(dpif, key, key_len, NULL); | |
3375 | } | |
3376 | ||
3377 | /* Update 'packet_count', 'byte_count', and 'used' members of installed facets. | |
3378 | * | |
3379 | * This function also pushes statistics updates to rules which each facet | |
3380 | * resubmits into. Generally these statistics will be accurate. However, if a | |
3381 | * facet changes the rule it resubmits into at some time in between | |
3382 | * update_stats() runs, it is possible that statistics accrued to the | |
3383 | * old rule will be incorrectly attributed to the new rule. This could be | |
3384 | * avoided by calling update_stats() whenever rules are created or | |
3385 | * deleted. However, the performance impact of making so many calls to the | |
3386 | * datapath do not justify the benefit of having perfectly accurate statistics. | |
3387 | */ | |
3388 | static void | |
3389 | update_stats(struct ofproto_dpif *p) | |
3390 | { | |
3391 | const struct dpif_flow_stats *stats; | |
3392 | struct dpif_flow_dump dump; | |
3393 | const struct nlattr *key; | |
3394 | size_t key_len; | |
3395 | ||
3396 | dpif_flow_dump_start(&dump, p->dpif); | |
3397 | while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) { | |
3398 | struct subfacet *subfacet; | |
3399 | ||
3400 | subfacet = subfacet_find(p, key, key_len); | |
3401 | switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) { | |
3402 | case SF_FAST_PATH: | |
3403 | update_subfacet_stats(subfacet, stats); | |
3404 | break; | |
3405 | ||
3406 | case SF_SLOW_PATH: | |
3407 | /* Stats are updated per-packet. */ | |
3408 | break; | |
3409 | ||
3410 | case SF_NOT_INSTALLED: | |
3411 | default: | |
3412 | delete_unexpected_flow(p->dpif, key, key_len); | |
3413 | break; | |
3414 | } | |
3415 | } | |
3416 | dpif_flow_dump_done(&dump); | |
3417 | } | |
3418 | ||
3419 | /* Calculates and returns the number of milliseconds of idle time after which | |
3420 | * subfacets should expire from the datapath. When a subfacet expires, we fold | |
3421 | * its statistics into its facet, and when a facet's last subfacet expires, we | |
3422 | * fold its statistic into its rule. */ | |
3423 | static int | |
3424 | subfacet_max_idle(const struct ofproto_dpif *ofproto) | |
3425 | { | |
3426 | /* | |
3427 | * Idle time histogram. | |
3428 | * | |
3429 | * Most of the time a switch has a relatively small number of subfacets. | |
3430 | * When this is the case we might as well keep statistics for all of them | |
3431 | * in userspace and to cache them in the kernel datapath for performance as | |
3432 | * well. | |
3433 | * | |
3434 | * As the number of subfacets increases, the memory required to maintain | |
3435 | * statistics about them in userspace and in the kernel becomes | |
3436 | * significant. However, with a large number of subfacets it is likely | |
3437 | * that only a few of them are "heavy hitters" that consume a large amount | |
3438 | * of bandwidth. At this point, only heavy hitters are worth caching in | |
3439 | * the kernel and maintaining in userspaces; other subfacets we can | |
3440 | * discard. | |
3441 | * | |
3442 | * The technique used to compute the idle time is to build a histogram with | |
3443 | * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet | |
3444 | * that is installed in the kernel gets dropped in the appropriate bucket. | |
3445 | * After the histogram has been built, we compute the cutoff so that only | |
3446 | * the most-recently-used 1% of subfacets (but at least | |
3447 | * ofproto->up.flow_eviction_threshold flows) are kept cached. At least | |
3448 | * the most-recently-used bucket of subfacets is kept, so actually an | |
3449 | * arbitrary number of subfacets can be kept in any given expiration run | |
3450 | * (though the next run will delete most of those unless they receive | |
3451 | * additional data). | |
3452 | * | |
3453 | * This requires a second pass through the subfacets, in addition to the | |
3454 | * pass made by update_stats(), because the former function never looks at | |
3455 | * uninstallable subfacets. | |
3456 | */ | |
3457 | enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) }; | |
3458 | enum { N_BUCKETS = 5000 / BUCKET_WIDTH }; | |
3459 | int buckets[N_BUCKETS] = { 0 }; | |
3460 | int total, subtotal, bucket; | |
3461 | struct subfacet *subfacet; | |
3462 | long long int now; | |
3463 | int i; | |
3464 | ||
3465 | total = hmap_count(&ofproto->subfacets); | |
3466 | if (total <= ofproto->up.flow_eviction_threshold) { | |
3467 | return N_BUCKETS * BUCKET_WIDTH; | |
3468 | } | |
3469 | ||
3470 | /* Build histogram. */ | |
3471 | now = time_msec(); | |
3472 | HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) { | |
3473 | long long int idle = now - subfacet->used; | |
3474 | int bucket = (idle <= 0 ? 0 | |
3475 | : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1 | |
3476 | : (unsigned int) idle / BUCKET_WIDTH); | |
3477 | buckets[bucket]++; | |
3478 | } | |
3479 | ||
3480 | /* Find the first bucket whose flows should be expired. */ | |
3481 | subtotal = bucket = 0; | |
3482 | do { | |
3483 | subtotal += buckets[bucket++]; | |
3484 | } while (bucket < N_BUCKETS && | |
3485 | subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100)); | |
3486 | ||
3487 | if (VLOG_IS_DBG_ENABLED()) { | |
3488 | struct ds s; | |
3489 | ||
3490 | ds_init(&s); | |
3491 | ds_put_cstr(&s, "keep"); | |
3492 | for (i = 0; i < N_BUCKETS; i++) { | |
3493 | if (i == bucket) { | |
3494 | ds_put_cstr(&s, ", drop"); | |
3495 | } | |
3496 | if (buckets[i]) { | |
3497 | ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]); | |
3498 | } | |
3499 | } | |
3500 | VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s)); | |
3501 | ds_destroy(&s); | |
3502 | } | |
3503 | ||
3504 | return bucket * BUCKET_WIDTH; | |
3505 | } | |
3506 | ||
3507 | enum { EXPIRE_MAX_BATCH = 50 }; | |
3508 | ||
3509 | static void | |
3510 | expire_batch(struct ofproto_dpif *ofproto, struct subfacet **subfacets, int n) | |
3511 | { | |
3512 | struct odputil_keybuf keybufs[EXPIRE_MAX_BATCH]; | |
3513 | struct dpif_op ops[EXPIRE_MAX_BATCH]; | |
3514 | struct dpif_op *opsp[EXPIRE_MAX_BATCH]; | |
3515 | struct ofpbuf keys[EXPIRE_MAX_BATCH]; | |
3516 | struct dpif_flow_stats stats[EXPIRE_MAX_BATCH]; | |
3517 | int i; | |
3518 | ||
3519 | for (i = 0; i < n; i++) { | |
3520 | ops[i].type = DPIF_OP_FLOW_DEL; | |
3521 | subfacet_get_key(subfacets[i], &keybufs[i], &keys[i]); | |
3522 | ops[i].u.flow_del.key = keys[i].data; | |
3523 | ops[i].u.flow_del.key_len = keys[i].size; | |
3524 | ops[i].u.flow_del.stats = &stats[i]; | |
3525 | opsp[i] = &ops[i]; | |
3526 | } | |
3527 | ||
3528 | dpif_operate(ofproto->dpif, opsp, n); | |
3529 | for (i = 0; i < n; i++) { | |
3530 | subfacet_reset_dp_stats(subfacets[i], &stats[i]); | |
3531 | subfacets[i]->path = SF_NOT_INSTALLED; | |
3532 | subfacet_destroy(subfacets[i]); | |
3533 | } | |
3534 | } | |
3535 | ||
3536 | static void | |
3537 | expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle) | |
3538 | { | |
3539 | /* Cutoff time for most flows. */ | |
3540 | long long int normal_cutoff = time_msec() - dp_max_idle; | |
3541 | ||
3542 | /* We really want to keep flows for special protocols around, so use a more | |
3543 | * conservative cutoff. */ | |
3544 | long long int special_cutoff = time_msec() - 10000; | |
3545 | ||
3546 | struct subfacet *subfacet, *next_subfacet; | |
3547 | struct subfacet *batch[EXPIRE_MAX_BATCH]; | |
3548 | int n_batch; | |
3549 | ||
3550 | n_batch = 0; | |
3551 | HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node, | |
3552 | &ofproto->subfacets) { | |
3553 | long long int cutoff; | |
3554 | ||
3555 | cutoff = (subfacet->slow & (SLOW_CFM | SLOW_LACP | SLOW_STP) | |
3556 | ? special_cutoff | |
3557 | : normal_cutoff); | |
3558 | if (subfacet->used < cutoff) { | |
3559 | if (subfacet->path != SF_NOT_INSTALLED) { | |
3560 | batch[n_batch++] = subfacet; | |
3561 | if (n_batch >= EXPIRE_MAX_BATCH) { | |
3562 | expire_batch(ofproto, batch, n_batch); | |
3563 | n_batch = 0; | |
3564 | } | |
3565 | } else { | |
3566 | subfacet_destroy(subfacet); | |
3567 | } | |
3568 | } | |
3569 | } | |
3570 | ||
3571 | if (n_batch > 0) { | |
3572 | expire_batch(ofproto, batch, n_batch); | |
3573 | } | |
3574 | } | |
3575 | ||
3576 | /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules, | |
3577 | * then delete it entirely. */ | |
3578 | static void | |
3579 | rule_expire(struct rule_dpif *rule) | |
3580 | { | |
3581 | struct facet *facet, *next_facet; | |
3582 | long long int now; | |
3583 | uint8_t reason; | |
3584 | ||
3585 | if (rule->up.pending) { | |
3586 | /* We'll have to expire it later. */ | |
3587 | return; | |
3588 | } | |
3589 | ||
3590 | /* Has 'rule' expired? */ | |
3591 | now = time_msec(); | |
3592 | if (rule->up.hard_timeout | |
3593 | && now > rule->up.modified + rule->up.hard_timeout * 1000) { | |
3594 | reason = OFPRR_HARD_TIMEOUT; | |
3595 | } else if (rule->up.idle_timeout | |
3596 | && now > rule->up.used + rule->up.idle_timeout * 1000) { | |
3597 | reason = OFPRR_IDLE_TIMEOUT; | |
3598 | } else { | |
3599 | return; | |
3600 | } | |
3601 | ||
3602 | COVERAGE_INC(ofproto_dpif_expired); | |
3603 | ||
3604 | /* Update stats. (This is a no-op if the rule expired due to an idle | |
3605 | * timeout, because that only happens when the rule has no facets left.) */ | |
3606 | LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) { | |
3607 | facet_remove(facet); | |
3608 | } | |
3609 | ||
3610 | /* Get rid of the rule. */ | |
3611 | ofproto_rule_expire(&rule->up, reason); | |
3612 | } | |
3613 | \f | |
3614 | /* Facets. */ | |
3615 | ||
3616 | /* Creates and returns a new facet owned by 'rule', given a 'flow'. | |
3617 | * | |
3618 | * The caller must already have determined that no facet with an identical | |
3619 | * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in | |
3620 | * the ofproto's classifier table. | |
3621 | * | |
3622 | * 'hash' must be the return value of flow_hash(flow, 0). | |
3623 | * | |
3624 | * The facet will initially have no subfacets. The caller should create (at | |
3625 | * least) one subfacet with subfacet_create(). */ | |
3626 | static struct facet * | |
3627 | facet_create(struct rule_dpif *rule, const struct flow *flow, uint32_t hash) | |
3628 | { | |
3629 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
3630 | struct facet *facet; | |
3631 | ||
3632 | facet = xzalloc(sizeof *facet); | |
3633 | facet->used = time_msec(); | |
3634 | hmap_insert(&ofproto->facets, &facet->hmap_node, hash); | |
3635 | list_push_back(&rule->facets, &facet->list_node); | |
3636 | facet->rule = rule; | |
3637 | facet->flow = *flow; | |
3638 | list_init(&facet->subfacets); | |
3639 | netflow_flow_init(&facet->nf_flow); | |
3640 | netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used); | |
3641 | ||
3642 | return facet; | |
3643 | } | |
3644 | ||
3645 | static void | |
3646 | facet_free(struct facet *facet) | |
3647 | { | |
3648 | free(facet); | |
3649 | } | |
3650 | ||
3651 | /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on | |
3652 | * 'packet', which arrived on 'in_port'. | |
3653 | * | |
3654 | * Takes ownership of 'packet'. */ | |
3655 | static bool | |
3656 | execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow, | |
3657 | const struct nlattr *odp_actions, size_t actions_len, | |
3658 | struct ofpbuf *packet) | |
3659 | { | |
3660 | struct odputil_keybuf keybuf; | |
3661 | struct ofpbuf key; | |
3662 | int error; | |
3663 | ||
3664 | ofpbuf_use_stack(&key, &keybuf, sizeof keybuf); | |
3665 | odp_flow_key_from_flow(&key, flow); | |
3666 | ||
3667 | error = dpif_execute(ofproto->dpif, key.data, key.size, | |
3668 | odp_actions, actions_len, packet); | |
3669 | ||
3670 | ofpbuf_delete(packet); | |
3671 | return !error; | |
3672 | } | |
3673 | ||
3674 | /* Remove 'facet' from 'ofproto' and free up the associated memory: | |
3675 | * | |
3676 | * - If 'facet' was installed in the datapath, uninstalls it and updates its | |
3677 | * rule's statistics, via subfacet_uninstall(). | |
3678 | * | |
3679 | * - Removes 'facet' from its rule and from ofproto->facets. | |
3680 | */ | |
3681 | static void | |
3682 | facet_remove(struct facet *facet) | |
3683 | { | |
3684 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); | |
3685 | struct subfacet *subfacet, *next_subfacet; | |
3686 | ||
3687 | assert(!list_is_empty(&facet->subfacets)); | |
3688 | ||
3689 | /* First uninstall all of the subfacets to get final statistics. */ | |
3690 | LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) { | |
3691 | subfacet_uninstall(subfacet); | |
3692 | } | |
3693 | ||
3694 | /* Flush the final stats to the rule. | |
3695 | * | |
3696 | * This might require us to have at least one subfacet around so that we | |
3697 | * can use its actions for accounting in facet_account(), which is why we | |
3698 | * have uninstalled but not yet destroyed the subfacets. */ | |
3699 | facet_flush_stats(facet); | |
3700 | ||
3701 | /* Now we're really all done so destroy everything. */ | |
3702 | LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node, | |
3703 | &facet->subfacets) { | |
3704 | subfacet_destroy__(subfacet); | |
3705 | } | |
3706 | hmap_remove(&ofproto->facets, &facet->hmap_node); | |
3707 | list_remove(&facet->list_node); | |
3708 | facet_free(facet); | |
3709 | } | |
3710 | ||
3711 | /* Feed information from 'facet' back into the learning table to keep it in | |
3712 | * sync with what is actually flowing through the datapath. */ | |
3713 | static void | |
3714 | facet_learn(struct facet *facet) | |
3715 | { | |
3716 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); | |
3717 | struct action_xlate_ctx ctx; | |
3718 | ||
3719 | if (!facet->has_learn | |
3720 | && !facet->has_normal | |
3721 | && (!facet->has_fin_timeout | |
3722 | || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) { | |
3723 | return; | |
3724 | } | |
3725 | ||
3726 | action_xlate_ctx_init(&ctx, ofproto, &facet->flow, | |
3727 | facet->flow.vlan_tci, | |
3728 | facet->rule, facet->tcp_flags, NULL); | |
3729 | ctx.may_learn = true; | |
3730 | xlate_actions_for_side_effects(&ctx, facet->rule->up.ofpacts, | |
3731 | facet->rule->up.ofpacts_len); | |
3732 | } | |
3733 | ||
3734 | static void | |
3735 | facet_account(struct facet *facet) | |
3736 | { | |
3737 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); | |
3738 | struct subfacet *subfacet; | |
3739 | const struct nlattr *a; | |
3740 | unsigned int left; | |
3741 | ovs_be16 vlan_tci; | |
3742 | uint64_t n_bytes; | |
3743 | ||
3744 | if (!facet->has_normal || !ofproto->has_bonded_bundles) { | |
3745 | return; | |
3746 | } | |
3747 | n_bytes = facet->byte_count - facet->accounted_bytes; | |
3748 | ||
3749 | /* This loop feeds byte counters to bond_account() for rebalancing to use | |
3750 | * as a basis. We also need to track the actual VLAN on which the packet | |
3751 | * is going to be sent to ensure that it matches the one passed to | |
3752 | * bond_choose_output_slave(). (Otherwise, we will account to the wrong | |
3753 | * hash bucket.) | |
3754 | * | |
3755 | * We use the actions from an arbitrary subfacet because they should all | |
3756 | * be equally valid for our purpose. */ | |
3757 | subfacet = CONTAINER_OF(list_front(&facet->subfacets), | |
3758 | struct subfacet, list_node); | |
3759 | vlan_tci = facet->flow.vlan_tci; | |
3760 | NL_ATTR_FOR_EACH_UNSAFE (a, left, | |
3761 | subfacet->actions, subfacet->actions_len) { | |
3762 | const struct ovs_action_push_vlan *vlan; | |
3763 | struct ofport_dpif *port; | |
3764 | ||
3765 | switch (nl_attr_type(a)) { | |
3766 | case OVS_ACTION_ATTR_OUTPUT: | |
3767 | port = get_odp_port(ofproto, nl_attr_get_u32(a)); | |
3768 | if (port && port->bundle && port->bundle->bond) { | |
3769 | bond_account(port->bundle->bond, &facet->flow, | |
3770 | vlan_tci_to_vid(vlan_tci), n_bytes); | |
3771 | } | |
3772 | break; | |
3773 | ||
3774 | case OVS_ACTION_ATTR_POP_VLAN: | |
3775 | vlan_tci = htons(0); | |
3776 | break; | |
3777 | ||
3778 | case OVS_ACTION_ATTR_PUSH_VLAN: | |
3779 | vlan = nl_attr_get(a); | |
3780 | vlan_tci = vlan->vlan_tci; | |
3781 | break; | |
3782 | } | |
3783 | } | |
3784 | } | |
3785 | ||
3786 | /* Returns true if the only action for 'facet' is to send to the controller. | |
3787 | * (We don't report NetFlow expiration messages for such facets because they | |
3788 | * are just part of the control logic for the network, not real traffic). */ | |
3789 | static bool | |
3790 | facet_is_controller_flow(struct facet *facet) | |
3791 | { | |
3792 | if (facet) { | |
3793 | const struct rule *rule = &facet->rule->up; | |
3794 | const struct ofpact *ofpacts = rule->ofpacts; | |
3795 | size_t ofpacts_len = rule->ofpacts_len; | |
3796 | ||
3797 | if (ofpacts->type == OFPACT_CONTROLLER && | |
3798 | ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len)) { | |
3799 | return true; | |
3800 | } | |
3801 | } | |
3802 | return false; | |
3803 | } | |
3804 | ||
3805 | /* Folds all of 'facet''s statistics into its rule. Also updates the | |
3806 | * accounting ofhook and emits a NetFlow expiration if appropriate. All of | |
3807 | * 'facet''s statistics in the datapath should have been zeroed and folded into | |
3808 | * its packet and byte counts before this function is called. */ | |
3809 | static void | |
3810 | facet_flush_stats(struct facet *facet) | |
3811 | { | |
3812 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); | |
3813 | struct subfacet *subfacet; | |
3814 | ||
3815 | LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) { | |
3816 | assert(!subfacet->dp_byte_count); | |
3817 | assert(!subfacet->dp_packet_count); | |
3818 | } | |
3819 | ||
3820 | facet_push_stats(facet); | |
3821 | if (facet->accounted_bytes < facet->byte_count) { | |
3822 | facet_account(facet); | |
3823 | facet->accounted_bytes = facet->byte_count; | |
3824 | } | |
3825 | ||
3826 | if (ofproto->netflow && !facet_is_controller_flow(facet)) { | |
3827 | struct ofexpired expired; | |
3828 | expired.flow = facet->flow; | |
3829 | expired.packet_count = facet->packet_count; | |
3830 | expired.byte_count = facet->byte_count; | |
3831 | expired.used = facet->used; | |
3832 | netflow_expire(ofproto->netflow, &facet->nf_flow, &expired); | |
3833 | } | |
3834 | ||
3835 | facet->rule->packet_count += facet->packet_count; | |
3836 | facet->rule->byte_count += facet->byte_count; | |
3837 | ||
3838 | /* Reset counters to prevent double counting if 'facet' ever gets | |
3839 | * reinstalled. */ | |
3840 | facet_reset_counters(facet); | |
3841 | ||
3842 | netflow_flow_clear(&facet->nf_flow); | |
3843 | facet->tcp_flags = 0; | |
3844 | } | |
3845 | ||
3846 | /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'. | |
3847 | * Returns it if found, otherwise a null pointer. | |
3848 | * | |
3849 | * 'hash' must be the return value of flow_hash(flow, 0). | |
3850 | * | |
3851 | * The returned facet might need revalidation; use facet_lookup_valid() | |
3852 | * instead if that is important. */ | |
3853 | static struct facet * | |
3854 | facet_find(struct ofproto_dpif *ofproto, | |
3855 | const struct flow *flow, uint32_t hash) | |
3856 | { | |
3857 | struct facet *facet; | |
3858 | ||
3859 | HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, hash, &ofproto->facets) { | |
3860 | if (flow_equal(flow, &facet->flow)) { | |
3861 | return facet; | |
3862 | } | |
3863 | } | |
3864 | ||
3865 | return NULL; | |
3866 | } | |
3867 | ||
3868 | /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'. | |
3869 | * Returns it if found, otherwise a null pointer. | |
3870 | * | |
3871 | * 'hash' must be the return value of flow_hash(flow, 0). | |
3872 | * | |
3873 | * The returned facet is guaranteed to be valid. */ | |
3874 | static struct facet * | |
3875 | facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow, | |
3876 | uint32_t hash) | |
3877 | { | |
3878 | struct facet *facet; | |
3879 | ||
3880 | facet = facet_find(ofproto, flow, hash); | |
3881 | if (facet | |
3882 | && (ofproto->need_revalidate | |
3883 | || tag_set_intersects(&ofproto->revalidate_set, facet->tags))) { | |
3884 | facet_revalidate(facet); | |
3885 | } | |
3886 | ||
3887 | return facet; | |
3888 | } | |
3889 | ||
3890 | static const char * | |
3891 | subfacet_path_to_string(enum subfacet_path path) | |
3892 | { | |
3893 | switch (path) { | |
3894 | case SF_NOT_INSTALLED: | |
3895 | return "not installed"; | |
3896 | case SF_FAST_PATH: | |
3897 | return "in fast path"; | |
3898 | case SF_SLOW_PATH: | |
3899 | return "in slow path"; | |
3900 | default: | |
3901 | return "<error>"; | |
3902 | } | |
3903 | } | |
3904 | ||
3905 | /* Returns the path in which a subfacet should be installed if its 'slow' | |
3906 | * member has the specified value. */ | |
3907 | static enum subfacet_path | |
3908 | subfacet_want_path(enum slow_path_reason slow) | |
3909 | { | |
3910 | return slow ? SF_SLOW_PATH : SF_FAST_PATH; | |
3911 | } | |
3912 | ||
3913 | /* Returns true if 'subfacet' needs to have its datapath flow updated, | |
3914 | * supposing that its actions have been recalculated as 'want_actions' and that | |
3915 | * 'slow' is nonzero iff 'subfacet' should be in the slow path. */ | |
3916 | static bool | |
3917 | subfacet_should_install(struct subfacet *subfacet, enum slow_path_reason slow, | |
3918 | const struct ofpbuf *want_actions) | |
3919 | { | |
3920 | enum subfacet_path want_path = subfacet_want_path(slow); | |
3921 | return (want_path != subfacet->path | |
3922 | || (want_path == SF_FAST_PATH | |
3923 | && (subfacet->actions_len != want_actions->size | |
3924 | || memcmp(subfacet->actions, want_actions->data, | |
3925 | subfacet->actions_len)))); | |
3926 | } | |
3927 | ||
3928 | static bool | |
3929 | facet_check_consistency(struct facet *facet) | |
3930 | { | |
3931 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15); | |
3932 | ||
3933 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); | |
3934 | ||
3935 | uint64_t odp_actions_stub[1024 / 8]; | |
3936 | struct ofpbuf odp_actions; | |
3937 | ||
3938 | struct rule_dpif *rule; | |
3939 | struct subfacet *subfacet; | |
3940 | bool may_log = false; | |
3941 | bool ok; | |
3942 | ||
3943 | /* Check the rule for consistency. */ | |
3944 | rule = rule_dpif_lookup(ofproto, &facet->flow); | |
3945 | ok = rule == facet->rule; | |
3946 | if (!ok) { | |
3947 | may_log = !VLOG_DROP_WARN(&rl); | |
3948 | if (may_log) { | |
3949 | struct ds s; | |
3950 | ||
3951 | ds_init(&s); | |
3952 | flow_format(&s, &facet->flow); | |
3953 | ds_put_format(&s, ": facet associated with wrong rule (was " | |
3954 | "table=%"PRIu8",", facet->rule->up.table_id); | |
3955 | cls_rule_format(&facet->rule->up.cr, &s); | |
3956 | ds_put_format(&s, ") (should have been table=%"PRIu8",", | |
3957 | rule->up.table_id); | |
3958 | cls_rule_format(&rule->up.cr, &s); | |
3959 | ds_put_char(&s, ')'); | |
3960 | ||
3961 | VLOG_WARN("%s", ds_cstr(&s)); | |
3962 | ds_destroy(&s); | |
3963 | } | |
3964 | } | |
3965 | ||
3966 | /* Check the datapath actions for consistency. */ | |
3967 | ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub); | |
3968 | LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) { | |
3969 | enum subfacet_path want_path; | |
3970 | struct odputil_keybuf keybuf; | |
3971 | struct action_xlate_ctx ctx; | |
3972 | struct ofpbuf key; | |
3973 | struct ds s; | |
3974 | ||
3975 | action_xlate_ctx_init(&ctx, ofproto, &facet->flow, | |
3976 | subfacet->initial_tci, rule, 0, NULL); | |
3977 | xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, | |
3978 | &odp_actions); | |
3979 | ||
3980 | if (subfacet->path == SF_NOT_INSTALLED) { | |
3981 | /* This only happens if the datapath reported an error when we | |
3982 | * tried to install the flow. Don't flag another error here. */ | |
3983 | continue; | |
3984 | } | |
3985 | ||
3986 | want_path = subfacet_want_path(subfacet->slow); | |
3987 | if (want_path == SF_SLOW_PATH && subfacet->path == SF_SLOW_PATH) { | |
3988 | /* The actions for slow-path flows may legitimately vary from one | |
3989 | * packet to the next. We're done. */ | |
3990 | continue; | |
3991 | } | |
3992 | ||
3993 | if (!subfacet_should_install(subfacet, subfacet->slow, &odp_actions)) { | |
3994 | continue; | |
3995 | } | |
3996 | ||
3997 | /* Inconsistency! */ | |
3998 | if (ok) { | |
3999 | may_log = !VLOG_DROP_WARN(&rl); | |
4000 | ok = false; | |
4001 | } | |
4002 | if (!may_log) { | |
4003 | /* Rate-limited, skip reporting. */ | |
4004 | continue; | |
4005 | } | |
4006 | ||
4007 | ds_init(&s); | |
4008 | subfacet_get_key(subfacet, &keybuf, &key); | |
4009 | odp_flow_key_format(key.data, key.size, &s); | |
4010 | ||
4011 | ds_put_cstr(&s, ": inconsistency in subfacet"); | |
4012 | if (want_path != subfacet->path) { | |
4013 | enum odp_key_fitness fitness = subfacet->key_fitness; | |
4014 | ||
4015 | ds_put_format(&s, " (%s, fitness=%s)", | |
4016 | subfacet_path_to_string(subfacet->path), | |
4017 | odp_key_fitness_to_string(fitness)); | |
4018 | ds_put_format(&s, " (should have been %s)", | |
4019 | subfacet_path_to_string(want_path)); | |
4020 | } else if (want_path == SF_FAST_PATH) { | |
4021 | ds_put_cstr(&s, " (actions were: "); | |
4022 | format_odp_actions(&s, subfacet->actions, | |
4023 | subfacet->actions_len); | |
4024 | ds_put_cstr(&s, ") (correct actions: "); | |
4025 | format_odp_actions(&s, odp_actions.data, odp_actions.size); | |
4026 | ds_put_char(&s, ')'); | |
4027 | } else { | |
4028 | ds_put_cstr(&s, " (actions: "); | |
4029 | format_odp_actions(&s, subfacet->actions, | |
4030 | subfacet->actions_len); | |
4031 | ds_put_char(&s, ')'); | |
4032 | } | |
4033 | VLOG_WARN("%s", ds_cstr(&s)); | |
4034 | ds_destroy(&s); | |
4035 | } | |
4036 | ofpbuf_uninit(&odp_actions); | |
4037 | ||
4038 | return ok; | |
4039 | } | |
4040 | ||
4041 | /* Re-searches the classifier for 'facet': | |
4042 | * | |
4043 | * - If the rule found is different from 'facet''s current rule, moves | |
4044 | * 'facet' to the new rule and recompiles its actions. | |
4045 | * | |
4046 | * - If the rule found is the same as 'facet''s current rule, leaves 'facet' | |
4047 | * where it is and recompiles its actions anyway. */ | |
4048 | static void | |
4049 | facet_revalidate(struct facet *facet) | |
4050 | { | |
4051 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); | |
4052 | struct actions { | |
4053 | struct nlattr *odp_actions; | |
4054 | size_t actions_len; | |
4055 | }; | |
4056 | struct actions *new_actions; | |
4057 | ||
4058 | struct action_xlate_ctx ctx; | |
4059 | uint64_t odp_actions_stub[1024 / 8]; | |
4060 | struct ofpbuf odp_actions; | |
4061 | ||
4062 | struct rule_dpif *new_rule; | |
4063 | struct subfacet *subfacet; | |
4064 | int i; | |
4065 | ||
4066 | COVERAGE_INC(facet_revalidate); | |
4067 | ||
4068 | new_rule = rule_dpif_lookup(ofproto, &facet->flow); | |
4069 | ||
4070 | /* Calculate new datapath actions. | |
4071 | * | |
4072 | * We do not modify any 'facet' state yet, because we might need to, e.g., | |
4073 | * emit a NetFlow expiration and, if so, we need to have the old state | |
4074 | * around to properly compose it. */ | |
4075 | ||
4076 | /* If the datapath actions changed or the installability changed, | |
4077 | * then we need to talk to the datapath. */ | |
4078 | i = 0; | |
4079 | new_actions = NULL; | |
4080 | memset(&ctx, 0, sizeof ctx); | |
4081 | ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub); | |
4082 | LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) { | |
4083 | enum slow_path_reason slow; | |
4084 | ||
4085 | action_xlate_ctx_init(&ctx, ofproto, &facet->flow, | |
4086 | subfacet->initial_tci, new_rule, 0, NULL); | |
4087 | xlate_actions(&ctx, new_rule->up.ofpacts, new_rule->up.ofpacts_len, | |
4088 | &odp_actions); | |
4089 | ||
4090 | slow = (subfacet->slow & SLOW_MATCH) | ctx.slow; | |
4091 | if (subfacet_should_install(subfacet, slow, &odp_actions)) { | |
4092 | struct dpif_flow_stats stats; | |
4093 | ||
4094 | subfacet_install(subfacet, | |
4095 | odp_actions.data, odp_actions.size, &stats, slow); | |
4096 | subfacet_update_stats(subfacet, &stats); | |
4097 | ||
4098 | if (!new_actions) { | |
4099 | new_actions = xcalloc(list_size(&facet->subfacets), | |
4100 | sizeof *new_actions); | |
4101 | } | |
4102 | new_actions[i].odp_actions = xmemdup(odp_actions.data, | |
4103 | odp_actions.size); | |
4104 | new_actions[i].actions_len = odp_actions.size; | |
4105 | } | |
4106 | ||
4107 | i++; | |
4108 | } | |
4109 | ofpbuf_uninit(&odp_actions); | |
4110 | ||
4111 | if (new_actions) { | |
4112 | facet_flush_stats(facet); | |
4113 | } | |
4114 | ||
4115 | /* Update 'facet' now that we've taken care of all the old state. */ | |
4116 | facet->tags = ctx.tags; | |
4117 | facet->nf_flow.output_iface = ctx.nf_output_iface; | |
4118 | facet->has_learn = ctx.has_learn; | |
4119 | facet->has_normal = ctx.has_normal; | |
4120 | facet->has_fin_timeout = ctx.has_fin_timeout; | |
4121 | facet->mirrors = ctx.mirrors; | |
4122 | ||
4123 | i = 0; | |
4124 | LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) { | |
4125 | subfacet->slow = (subfacet->slow & SLOW_MATCH) | ctx.slow; | |
4126 | ||
4127 | if (new_actions && new_actions[i].odp_actions) { | |
4128 | free(subfacet->actions); | |
4129 | subfacet->actions = new_actions[i].odp_actions; | |
4130 | subfacet->actions_len = new_actions[i].actions_len; | |
4131 | } | |
4132 | i++; | |
4133 | } | |
4134 | free(new_actions); | |
4135 | ||
4136 | if (facet->rule != new_rule) { | |
4137 | COVERAGE_INC(facet_changed_rule); | |
4138 | list_remove(&facet->list_node); | |
4139 | list_push_back(&new_rule->facets, &facet->list_node); | |
4140 | facet->rule = new_rule; | |
4141 | facet->used = new_rule->up.created; | |
4142 | facet->prev_used = facet->used; | |
4143 | } | |
4144 | } | |
4145 | ||
4146 | /* Updates 'facet''s used time. Caller is responsible for calling | |
4147 | * facet_push_stats() to update the flows which 'facet' resubmits into. */ | |
4148 | static void | |
4149 | facet_update_time(struct facet *facet, long long int used) | |
4150 | { | |
4151 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); | |
4152 | if (used > facet->used) { | |
4153 | facet->used = used; | |
4154 | ofproto_rule_update_used(&facet->rule->up, used); | |
4155 | netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used); | |
4156 | } | |
4157 | } | |
4158 | ||
4159 | static void | |
4160 | facet_reset_counters(struct facet *facet) | |
4161 | { | |
4162 | facet->packet_count = 0; | |
4163 | facet->byte_count = 0; | |
4164 | facet->prev_packet_count = 0; | |
4165 | facet->prev_byte_count = 0; | |
4166 | facet->accounted_bytes = 0; | |
4167 | } | |
4168 | ||
4169 | static void | |
4170 | facet_push_stats(struct facet *facet) | |
4171 | { | |
4172 | struct dpif_flow_stats stats; | |
4173 | ||
4174 | assert(facet->packet_count >= facet->prev_packet_count); | |
4175 | assert(facet->byte_count >= facet->prev_byte_count); | |
4176 | assert(facet->used >= facet->prev_used); | |
4177 | ||
4178 | stats.n_packets = facet->packet_count - facet->prev_packet_count; | |
4179 | stats.n_bytes = facet->byte_count - facet->prev_byte_count; | |
4180 | stats.used = facet->used; | |
4181 | stats.tcp_flags = 0; | |
4182 | ||
4183 | if (stats.n_packets || stats.n_bytes || facet->used > facet->prev_used) { | |
4184 | facet->prev_packet_count = facet->packet_count; | |
4185 | facet->prev_byte_count = facet->byte_count; | |
4186 | facet->prev_used = facet->used; | |
4187 | ||
4188 | flow_push_stats(facet->rule, &facet->flow, &stats); | |
4189 | ||
4190 | update_mirror_stats(ofproto_dpif_cast(facet->rule->up.ofproto), | |
4191 | facet->mirrors, stats.n_packets, stats.n_bytes); | |
4192 | } | |
4193 | } | |
4194 | ||
4195 | static void | |
4196 | rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats) | |
4197 | { | |
4198 | rule->packet_count += stats->n_packets; | |
4199 | rule->byte_count += stats->n_bytes; | |
4200 | ofproto_rule_update_used(&rule->up, stats->used); | |
4201 | } | |
4202 | ||
4203 | /* Pushes flow statistics to the rules which 'flow' resubmits into given | |
4204 | * 'rule''s actions and mirrors. */ | |
4205 | static void | |
4206 | flow_push_stats(struct rule_dpif *rule, | |
4207 | const struct flow *flow, const struct dpif_flow_stats *stats) | |
4208 | { | |
4209 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
4210 | struct action_xlate_ctx ctx; | |
4211 | ||
4212 | ofproto_rule_update_used(&rule->up, stats->used); | |
4213 | ||
4214 | action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci, rule, | |
4215 | 0, NULL); | |
4216 | ctx.resubmit_stats = stats; | |
4217 | xlate_actions_for_side_effects(&ctx, rule->up.ofpacts, | |
4218 | rule->up.ofpacts_len); | |
4219 | } | |
4220 | \f | |
4221 | /* Subfacets. */ | |
4222 | ||
4223 | static struct subfacet * | |
4224 | subfacet_find__(struct ofproto_dpif *ofproto, | |
4225 | const struct nlattr *key, size_t key_len, uint32_t key_hash, | |
4226 | const struct flow *flow) | |
4227 | { | |
4228 | struct subfacet *subfacet; | |
4229 | ||
4230 | HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash, | |
4231 | &ofproto->subfacets) { | |
4232 | if (subfacet->key | |
4233 | ? (subfacet->key_len == key_len | |
4234 | && !memcmp(key, subfacet->key, key_len)) | |
4235 | : flow_equal(flow, &subfacet->facet->flow)) { | |
4236 | return subfacet; | |
4237 | } | |
4238 | } | |
4239 | ||
4240 | return NULL; | |
4241 | } | |
4242 | ||
4243 | /* Searches 'facet' (within 'ofproto') for a subfacet with the specified | |
4244 | * 'key_fitness', 'key', and 'key_len'. Returns the existing subfacet if | |
4245 | * there is one, otherwise creates and returns a new subfacet. | |
4246 | * | |
4247 | * If the returned subfacet is new, then subfacet->actions will be NULL, in | |
4248 | * which case the caller must populate the actions with | |
4249 | * subfacet_make_actions(). */ | |
4250 | static struct subfacet * | |
4251 | subfacet_create(struct facet *facet, enum odp_key_fitness key_fitness, | |
4252 | const struct nlattr *key, size_t key_len, ovs_be16 initial_tci) | |
4253 | { | |
4254 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); | |
4255 | uint32_t key_hash = odp_flow_key_hash(key, key_len); | |
4256 | struct subfacet *subfacet; | |
4257 | ||
4258 | if (list_is_empty(&facet->subfacets)) { | |
4259 | subfacet = &facet->one_subfacet; | |
4260 | ||
4261 | /* This subfacet should conceptually be created, and have its first | |
4262 | * packet pass through, at the same time that its facet was created. | |
4263 | * If we called time_msec() here, then the subfacet could look | |
4264 | * (occasionally) as though it was used some time after the facet was | |
4265 | * used. That can make a one-packet flow look like it has a nonzero | |
4266 | * duration, which looks odd in e.g. NetFlow statistics. */ | |
4267 | subfacet->used = facet->used; | |
4268 | } else { | |
4269 | subfacet = subfacet_find__(ofproto, key, key_len, key_hash, | |
4270 | &facet->flow); | |
4271 | if (subfacet) { | |
4272 | if (subfacet->facet == facet) { | |
4273 | return subfacet; | |
4274 | } | |
4275 | ||
4276 | /* This shouldn't happen. */ | |
4277 | VLOG_ERR_RL(&rl, "subfacet with wrong facet"); | |
4278 | subfacet_destroy(subfacet); | |
4279 | } | |
4280 | ||
4281 | subfacet = xmalloc(sizeof *subfacet); | |
4282 | subfacet->used = time_msec(); | |
4283 | } | |
4284 | ||
4285 | hmap_insert(&ofproto->subfacets, &subfacet->hmap_node, key_hash); | |
4286 | list_push_back(&facet->subfacets, &subfacet->list_node); | |
4287 | subfacet->facet = facet; | |
4288 | subfacet->key_fitness = key_fitness; | |
4289 | if (key_fitness != ODP_FIT_PERFECT) { | |
4290 | subfacet->key = xmemdup(key, key_len); | |
4291 | subfacet->key_len = key_len; | |
4292 | } else { | |
4293 | subfacet->key = NULL; | |
4294 | subfacet->key_len = 0; | |
4295 | } | |
4296 | subfacet->dp_packet_count = 0; | |
4297 | subfacet->dp_byte_count = 0; | |
4298 | subfacet->actions_len = 0; | |
4299 | subfacet->actions = NULL; | |
4300 | subfacet->slow = (subfacet->key_fitness == ODP_FIT_TOO_LITTLE | |
4301 | ? SLOW_MATCH | |
4302 | : 0); | |
4303 | subfacet->path = SF_NOT_INSTALLED; | |
4304 | subfacet->initial_tci = initial_tci; | |
4305 | ||
4306 | return subfacet; | |
4307 | } | |
4308 | ||
4309 | /* Searches 'ofproto' for a subfacet with the given 'key', 'key_len', and | |
4310 | * 'flow'. Returns the subfacet if one exists, otherwise NULL. */ | |
4311 | static struct subfacet * | |
4312 | subfacet_find(struct ofproto_dpif *ofproto, | |
4313 | const struct nlattr *key, size_t key_len) | |
4314 | { | |
4315 | uint32_t key_hash = odp_flow_key_hash(key, key_len); | |
4316 | enum odp_key_fitness fitness; | |
4317 | struct flow flow; | |
4318 | ||
4319 | fitness = odp_flow_key_to_flow(key, key_len, &flow); | |
4320 | if (fitness == ODP_FIT_ERROR) { | |
4321 | return NULL; | |
4322 | } | |
4323 | ||
4324 | return subfacet_find__(ofproto, key, key_len, key_hash, &flow); | |
4325 | } | |
4326 | ||
4327 | /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from | |
4328 | * its facet within 'ofproto', and frees it. */ | |
4329 | static void | |
4330 | subfacet_destroy__(struct subfacet *subfacet) | |
4331 | { | |
4332 | struct facet *facet = subfacet->facet; | |
4333 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); | |
4334 | ||
4335 | subfacet_uninstall(subfacet); | |
4336 | hmap_remove(&ofproto->subfacets, &subfacet->hmap_node); | |
4337 | list_remove(&subfacet->list_node); | |
4338 | free(subfacet->key); | |
4339 | free(subfacet->actions); | |
4340 | if (subfacet != &facet->one_subfacet) { | |
4341 | free(subfacet); | |
4342 | } | |
4343 | } | |
4344 | ||
4345 | /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the | |
4346 | * last remaining subfacet in its facet destroys the facet too. */ | |
4347 | static void | |
4348 | subfacet_destroy(struct subfacet *subfacet) | |
4349 | { | |
4350 | struct facet *facet = subfacet->facet; | |
4351 | ||
4352 | if (list_is_singleton(&facet->subfacets)) { | |
4353 | /* facet_remove() needs at least one subfacet (it will remove it). */ | |
4354 | facet_remove(facet); | |
4355 | } else { | |
4356 | subfacet_destroy__(subfacet); | |
4357 | } | |
4358 | } | |
4359 | ||
4360 | /* Initializes 'key' with the sequence of OVS_KEY_ATTR_* Netlink attributes | |
4361 | * that can be used to refer to 'subfacet'. The caller must provide 'keybuf' | |
4362 | * for use as temporary storage. */ | |
4363 | static void | |
4364 | subfacet_get_key(struct subfacet *subfacet, struct odputil_keybuf *keybuf, | |
4365 | struct ofpbuf *key) | |
4366 | { | |
4367 | if (!subfacet->key) { | |
4368 | ofpbuf_use_stack(key, keybuf, sizeof *keybuf); | |
4369 | odp_flow_key_from_flow(key, &subfacet->facet->flow); | |
4370 | } else { | |
4371 | ofpbuf_use_const(key, subfacet->key, subfacet->key_len); | |
4372 | } | |
4373 | } | |
4374 | ||
4375 | /* Composes the datapath actions for 'subfacet' based on its rule's actions. | |
4376 | * Translates the actions into 'odp_actions', which the caller must have | |
4377 | * initialized and is responsible for uninitializing. */ | |
4378 | static void | |
4379 | subfacet_make_actions(struct subfacet *subfacet, const struct ofpbuf *packet, | |
4380 | struct ofpbuf *odp_actions) | |
4381 | { | |
4382 | struct facet *facet = subfacet->facet; | |
4383 | struct rule_dpif *rule = facet->rule; | |
4384 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
4385 | ||
4386 | struct action_xlate_ctx ctx; | |
4387 | ||
4388 | action_xlate_ctx_init(&ctx, ofproto, &facet->flow, subfacet->initial_tci, | |
4389 | rule, 0, packet); | |
4390 | xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, odp_actions); | |
4391 | facet->tags = ctx.tags; | |
4392 | facet->has_learn = ctx.has_learn; | |
4393 | facet->has_normal = ctx.has_normal; | |
4394 | facet->has_fin_timeout = ctx.has_fin_timeout; | |
4395 | facet->nf_flow.output_iface = ctx.nf_output_iface; | |
4396 | facet->mirrors = ctx.mirrors; | |
4397 | ||
4398 | subfacet->slow = (subfacet->slow & SLOW_MATCH) | ctx.slow; | |
4399 | if (subfacet->actions_len != odp_actions->size | |
4400 | || memcmp(subfacet->actions, odp_actions->data, odp_actions->size)) { | |
4401 | free(subfacet->actions); | |
4402 | subfacet->actions_len = odp_actions->size; | |
4403 | subfacet->actions = xmemdup(odp_actions->data, odp_actions->size); | |
4404 | } | |
4405 | } | |
4406 | ||
4407 | /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len' | |
4408 | * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters | |
4409 | * in the datapath will be zeroed and 'stats' will be updated with traffic new | |
4410 | * since 'subfacet' was last updated. | |
4411 | * | |
4412 | * Returns 0 if successful, otherwise a positive errno value. */ | |
4413 | static int | |
4414 | subfacet_install(struct subfacet *subfacet, | |
4415 | const struct nlattr *actions, size_t actions_len, | |
4416 | struct dpif_flow_stats *stats, | |
4417 | enum slow_path_reason slow) | |
4418 | { | |
4419 | struct facet *facet = subfacet->facet; | |
4420 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); | |
4421 | enum subfacet_path path = subfacet_want_path(slow); | |
4422 | uint64_t slow_path_stub[128 / 8]; | |
4423 | struct odputil_keybuf keybuf; | |
4424 | enum dpif_flow_put_flags flags; | |
4425 | struct ofpbuf key; | |
4426 | int ret; | |
4427 | ||
4428 | flags = DPIF_FP_CREATE | DPIF_FP_MODIFY; | |
4429 | if (stats) { | |
4430 | flags |= DPIF_FP_ZERO_STATS; | |
4431 | } | |
4432 | ||
4433 | if (path == SF_SLOW_PATH) { | |
4434 | compose_slow_path(ofproto, &facet->flow, slow, | |
4435 | slow_path_stub, sizeof slow_path_stub, | |
4436 | &actions, &actions_len); | |
4437 | } | |
4438 | ||
4439 | subfacet_get_key(subfacet, &keybuf, &key); | |
4440 | ret = dpif_flow_put(ofproto->dpif, flags, key.data, key.size, | |
4441 | actions, actions_len, stats); | |
4442 | ||
4443 | if (stats) { | |
4444 | subfacet_reset_dp_stats(subfacet, stats); | |
4445 | } | |
4446 | ||
4447 | if (!ret) { | |
4448 | subfacet->path = path; | |
4449 | } | |
4450 | return ret; | |
4451 | } | |
4452 | ||
4453 | static int | |
4454 | subfacet_reinstall(struct subfacet *subfacet, struct dpif_flow_stats *stats) | |
4455 | { | |
4456 | return subfacet_install(subfacet, subfacet->actions, subfacet->actions_len, | |
4457 | stats, subfacet->slow); | |
4458 | } | |
4459 | ||
4460 | /* If 'subfacet' is installed in the datapath, uninstalls it. */ | |
4461 | static void | |
4462 | subfacet_uninstall(struct subfacet *subfacet) | |
4463 | { | |
4464 | if (subfacet->path != SF_NOT_INSTALLED) { | |
4465 | struct rule_dpif *rule = subfacet->facet->rule; | |
4466 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
4467 | struct odputil_keybuf keybuf; | |
4468 | struct dpif_flow_stats stats; | |
4469 | struct ofpbuf key; | |
4470 | int error; | |
4471 | ||
4472 | subfacet_get_key(subfacet, &keybuf, &key); | |
4473 | error = dpif_flow_del(ofproto->dpif, key.data, key.size, &stats); | |
4474 | subfacet_reset_dp_stats(subfacet, &stats); | |
4475 | if (!error) { | |
4476 | subfacet_update_stats(subfacet, &stats); | |
4477 | } | |
4478 | subfacet->path = SF_NOT_INSTALLED; | |
4479 | } else { | |
4480 | assert(subfacet->dp_packet_count == 0); | |
4481 | assert(subfacet->dp_byte_count == 0); | |
4482 | } | |
4483 | } | |
4484 | ||
4485 | /* Resets 'subfacet''s datapath statistics counters. This should be called | |
4486 | * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is | |
4487 | * non-null, it should contain the statistics returned by dpif when 'subfacet' | |
4488 | * was reset in the datapath. 'stats' will be modified to include only | |
4489 | * statistics new since 'subfacet' was last updated. */ | |
4490 | static void | |
4491 | subfacet_reset_dp_stats(struct subfacet *subfacet, | |
4492 | struct dpif_flow_stats *stats) | |
4493 | { | |
4494 | if (stats | |
4495 | && subfacet->dp_packet_count <= stats->n_packets | |
4496 | && subfacet->dp_byte_count <= stats->n_bytes) { | |
4497 | stats->n_packets -= subfacet->dp_packet_count; | |
4498 | stats->n_bytes -= subfacet->dp_byte_count; | |
4499 | } | |
4500 | ||
4501 | subfacet->dp_packet_count = 0; | |
4502 | subfacet->dp_byte_count = 0; | |
4503 | } | |
4504 | ||
4505 | /* Updates 'subfacet''s used time. The caller is responsible for calling | |
4506 | * facet_push_stats() to update the flows which 'subfacet' resubmits into. */ | |
4507 | static void | |
4508 | subfacet_update_time(struct subfacet *subfacet, long long int used) | |
4509 | { | |
4510 | if (used > subfacet->used) { | |
4511 | subfacet->used = used; | |
4512 | facet_update_time(subfacet->facet, used); | |
4513 | } | |
4514 | } | |
4515 | ||
4516 | /* Folds the statistics from 'stats' into the counters in 'subfacet'. | |
4517 | * | |
4518 | * Because of the meaning of a subfacet's counters, it only makes sense to do | |
4519 | * this if 'stats' are not tracked in the datapath, that is, if 'stats' | |
4520 | * represents a packet that was sent by hand or if it represents statistics | |
4521 | * that have been cleared out of the datapath. */ | |
4522 | static void | |
4523 | subfacet_update_stats(struct subfacet *subfacet, | |
4524 | const struct dpif_flow_stats *stats) | |
4525 | { | |
4526 | if (stats->n_packets || stats->used > subfacet->used) { | |
4527 | struct facet *facet = subfacet->facet; | |
4528 | ||
4529 | subfacet_update_time(subfacet, stats->used); | |
4530 | facet->packet_count += stats->n_packets; | |
4531 | facet->byte_count += stats->n_bytes; | |
4532 | facet->tcp_flags |= stats->tcp_flags; | |
4533 | facet_push_stats(facet); | |
4534 | netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags); | |
4535 | } | |
4536 | } | |
4537 | \f | |
4538 | /* Rules. */ | |
4539 | ||
4540 | static struct rule_dpif * | |
4541 | rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow) | |
4542 | { | |
4543 | struct ofport_dpif *port; | |
4544 | struct rule_dpif *rule; | |
4545 | ||
4546 | rule = rule_dpif_lookup__(ofproto, flow, 0); | |
4547 | if (rule) { | |
4548 | return rule; | |
4549 | } | |
4550 | ||
4551 | port = get_ofp_port(ofproto, flow->in_port); | |
4552 | if (!port) { | |
4553 | VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, flow->in_port); | |
4554 | return ofproto->miss_rule; | |
4555 | } | |
4556 | ||
4557 | if (port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN) { | |
4558 | return ofproto->no_packet_in_rule; | |
4559 | } | |
4560 | return ofproto->miss_rule; | |
4561 | } | |
4562 | ||
4563 | static struct rule_dpif * | |
4564 | rule_dpif_lookup__(struct ofproto_dpif *ofproto, const struct flow *flow, | |
4565 | uint8_t table_id) | |
4566 | { | |
4567 | struct cls_rule *cls_rule; | |
4568 | struct classifier *cls; | |
4569 | ||
4570 | if (table_id >= N_TABLES) { | |
4571 | return NULL; | |
4572 | } | |
4573 | ||
4574 | cls = &ofproto->up.tables[table_id].cls; | |
4575 | if (flow->nw_frag & FLOW_NW_FRAG_ANY | |
4576 | && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) { | |
4577 | /* For OFPC_NORMAL frag_handling, we must pretend that transport ports | |
4578 | * are unavailable. */ | |
4579 | struct flow ofpc_normal_flow = *flow; | |
4580 | ofpc_normal_flow.tp_src = htons(0); | |
4581 | ofpc_normal_flow.tp_dst = htons(0); | |
4582 | cls_rule = classifier_lookup(cls, &ofpc_normal_flow); | |
4583 | } else { | |
4584 | cls_rule = classifier_lookup(cls, flow); | |
4585 | } | |
4586 | return rule_dpif_cast(rule_from_cls_rule(cls_rule)); | |
4587 | } | |
4588 | ||
4589 | static void | |
4590 | complete_operation(struct rule_dpif *rule) | |
4591 | { | |
4592 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
4593 | ||
4594 | rule_invalidate(rule); | |
4595 | if (clogged) { | |
4596 | struct dpif_completion *c = xmalloc(sizeof *c); | |
4597 | c->op = rule->up.pending; | |
4598 | list_push_back(&ofproto->completions, &c->list_node); | |
4599 | } else { | |
4600 | ofoperation_complete(rule->up.pending, 0); | |
4601 | } | |
4602 | } | |
4603 | ||
4604 | static struct rule * | |
4605 | rule_alloc(void) | |
4606 | { | |
4607 | struct rule_dpif *rule = xmalloc(sizeof *rule); | |
4608 | return &rule->up; | |
4609 | } | |
4610 | ||
4611 | static void | |
4612 | rule_dealloc(struct rule *rule_) | |
4613 | { | |
4614 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
4615 | free(rule); | |
4616 | } | |
4617 | ||
4618 | static enum ofperr | |
4619 | rule_construct(struct rule *rule_) | |
4620 | { | |
4621 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
4622 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
4623 | struct rule_dpif *victim; | |
4624 | uint8_t table_id; | |
4625 | enum ofperr error; | |
4626 | ||
4627 | error = ofpacts_check(rule->up.ofpacts, rule->up.ofpacts_len, | |
4628 | &rule->up.cr.flow, ofproto->max_ports); | |
4629 | if (error) { | |
4630 | return error; | |
4631 | } | |
4632 | ||
4633 | rule->packet_count = 0; | |
4634 | rule->byte_count = 0; | |
4635 | ||
4636 | victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending)); | |
4637 | if (victim && !list_is_empty(&victim->facets)) { | |
4638 | struct facet *facet; | |
4639 | ||
4640 | rule->facets = victim->facets; | |
4641 | list_moved(&rule->facets); | |
4642 | LIST_FOR_EACH (facet, list_node, &rule->facets) { | |
4643 | /* XXX: We're only clearing our local counters here. It's possible | |
4644 | * that quite a few packets are unaccounted for in the datapath | |
4645 | * statistics. These will be accounted to the new rule instead of | |
4646 | * cleared as required. This could be fixed by clearing out the | |
4647 | * datapath statistics for this facet, but currently it doesn't | |
4648 | * seem worth it. */ | |
4649 | facet_reset_counters(facet); | |
4650 | facet->rule = rule; | |
4651 | } | |
4652 | } else { | |
4653 | /* Must avoid list_moved() in this case. */ | |
4654 | list_init(&rule->facets); | |
4655 | } | |
4656 | ||
4657 | table_id = rule->up.table_id; | |
4658 | rule->tag = (victim ? victim->tag | |
4659 | : table_id == 0 ? 0 | |
4660 | : rule_calculate_tag(&rule->up.cr.flow, &rule->up.cr.wc, | |
4661 | ofproto->tables[table_id].basis)); | |
4662 | ||
4663 | complete_operation(rule); | |
4664 | return 0; | |
4665 | } | |
4666 | ||
4667 | static void | |
4668 | rule_destruct(struct rule *rule_) | |
4669 | { | |
4670 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
4671 | struct facet *facet, *next_facet; | |
4672 | ||
4673 | LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) { | |
4674 | facet_revalidate(facet); | |
4675 | } | |
4676 | ||
4677 | complete_operation(rule); | |
4678 | } | |
4679 | ||
4680 | static void | |
4681 | rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes) | |
4682 | { | |
4683 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
4684 | struct facet *facet; | |
4685 | ||
4686 | /* Start from historical data for 'rule' itself that are no longer tracked | |
4687 | * in facets. This counts, for example, facets that have expired. */ | |
4688 | *packets = rule->packet_count; | |
4689 | *bytes = rule->byte_count; | |
4690 | ||
4691 | /* Add any statistics that are tracked by facets. This includes | |
4692 | * statistical data recently updated by ofproto_update_stats() as well as | |
4693 | * stats for packets that were executed "by hand" via dpif_execute(). */ | |
4694 | LIST_FOR_EACH (facet, list_node, &rule->facets) { | |
4695 | *packets += facet->packet_count; | |
4696 | *bytes += facet->byte_count; | |
4697 | } | |
4698 | } | |
4699 | ||
4700 | static enum ofperr | |
4701 | rule_execute(struct rule *rule_, const struct flow *flow, | |
4702 | struct ofpbuf *packet) | |
4703 | { | |
4704 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
4705 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
4706 | ||
4707 | struct dpif_flow_stats stats; | |
4708 | ||
4709 | struct action_xlate_ctx ctx; | |
4710 | uint64_t odp_actions_stub[1024 / 8]; | |
4711 | struct ofpbuf odp_actions; | |
4712 | ||
4713 | dpif_flow_stats_extract(flow, packet, &stats); | |
4714 | rule_credit_stats(rule, &stats); | |
4715 | ||
4716 | ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub); | |
4717 | action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci, | |
4718 | rule, stats.tcp_flags, packet); | |
4719 | ctx.resubmit_stats = &stats; | |
4720 | xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, &odp_actions); | |
4721 | ||
4722 | execute_odp_actions(ofproto, flow, odp_actions.data, | |
4723 | odp_actions.size, packet); | |
4724 | ||
4725 | ofpbuf_uninit(&odp_actions); | |
4726 | ||
4727 | return 0; | |
4728 | } | |
4729 | ||
4730 | static void | |
4731 | rule_modify_actions(struct rule *rule_) | |
4732 | { | |
4733 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
4734 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
4735 | enum ofperr error; | |
4736 | ||
4737 | error = ofpacts_check(rule->up.ofpacts, rule->up.ofpacts_len, | |
4738 | &rule->up.cr.flow, ofproto->max_ports); | |
4739 | if (error) { | |
4740 | ofoperation_complete(rule->up.pending, error); | |
4741 | return; | |
4742 | } | |
4743 | ||
4744 | complete_operation(rule); | |
4745 | } | |
4746 | \f | |
4747 | /* Sends 'packet' out 'ofport'. | |
4748 | * May modify 'packet'. | |
4749 | * Returns 0 if successful, otherwise a positive errno value. */ | |
4750 | static int | |
4751 | send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet) | |
4752 | { | |
4753 | const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
4754 | struct ofpbuf key, odp_actions; | |
4755 | struct odputil_keybuf keybuf; | |
4756 | uint16_t odp_port; | |
4757 | struct flow flow; | |
4758 | int error; | |
4759 | ||
4760 | flow_extract(packet, 0, 0, 0, &flow); | |
4761 | odp_port = vsp_realdev_to_vlandev(ofproto, ofport->odp_port, | |
4762 | flow.vlan_tci); | |
4763 | if (odp_port != ofport->odp_port) { | |
4764 | eth_pop_vlan(packet); | |
4765 | flow.vlan_tci = htons(0); | |
4766 | } | |
4767 | ||
4768 | ofpbuf_use_stack(&key, &keybuf, sizeof keybuf); | |
4769 | odp_flow_key_from_flow(&key, &flow); | |
4770 | ||
4771 | ofpbuf_init(&odp_actions, 32); | |
4772 | compose_sflow_action(ofproto, &odp_actions, &flow, odp_port); | |
4773 | ||
4774 | nl_msg_put_u32(&odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port); | |
4775 | error = dpif_execute(ofproto->dpif, | |
4776 | key.data, key.size, | |
4777 | odp_actions.data, odp_actions.size, | |
4778 | packet); | |
4779 | ofpbuf_uninit(&odp_actions); | |
4780 | ||
4781 | if (error) { | |
4782 | VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)", | |
4783 | ofproto->up.name, odp_port, strerror(error)); | |
4784 | } | |
4785 | ofproto_update_local_port_stats(ofport->up.ofproto, packet->size, 0); | |
4786 | return error; | |
4787 | } | |
4788 | \f | |
4789 | /* OpenFlow to datapath action translation. */ | |
4790 | ||
4791 | static void do_xlate_actions(const struct ofpact *, size_t ofpacts_len, | |
4792 | struct action_xlate_ctx *); | |
4793 | static void xlate_normal(struct action_xlate_ctx *); | |
4794 | ||
4795 | /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'. | |
4796 | * The action will state 'slow' as the reason that the action is in the slow | |
4797 | * path. (This is purely informational: it allows a human viewing "ovs-dpctl | |
4798 | * dump-flows" output to see why a flow is in the slow path.) | |
4799 | * | |
4800 | * The 'stub_size' bytes in 'stub' will be used to store the action. | |
4801 | * 'stub_size' must be large enough for the action. | |
4802 | * | |
4803 | * The action and its size will be stored in '*actionsp' and '*actions_lenp', | |
4804 | * respectively. */ | |
4805 | static void | |
4806 | compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow, | |
4807 | enum slow_path_reason slow, | |
4808 | uint64_t *stub, size_t stub_size, | |
4809 | const struct nlattr **actionsp, size_t *actions_lenp) | |
4810 | { | |
4811 | union user_action_cookie cookie; | |
4812 | struct ofpbuf buf; | |
4813 | ||
4814 | cookie.type = USER_ACTION_COOKIE_SLOW_PATH; | |
4815 | cookie.slow_path.unused = 0; | |
4816 | cookie.slow_path.reason = slow; | |
4817 | ||
4818 | ofpbuf_use_stack(&buf, stub, stub_size); | |
4819 | if (slow & (SLOW_CFM | SLOW_LACP | SLOW_STP)) { | |
4820 | uint32_t pid = dpif_port_get_pid(ofproto->dpif, UINT16_MAX); | |
4821 | odp_put_userspace_action(pid, &cookie, &buf); | |
4822 | } else { | |
4823 | put_userspace_action(ofproto, &buf, flow, &cookie); | |
4824 | } | |
4825 | *actionsp = buf.data; | |
4826 | *actions_lenp = buf.size; | |
4827 | } | |
4828 | ||
4829 | static size_t | |
4830 | put_userspace_action(const struct ofproto_dpif *ofproto, | |
4831 | struct ofpbuf *odp_actions, | |
4832 | const struct flow *flow, | |
4833 | const union user_action_cookie *cookie) | |
4834 | { | |
4835 | uint32_t pid; | |
4836 | ||
4837 | pid = dpif_port_get_pid(ofproto->dpif, | |
4838 | ofp_port_to_odp_port(flow->in_port)); | |
4839 | ||
4840 | return odp_put_userspace_action(pid, cookie, odp_actions); | |
4841 | } | |
4842 | ||
4843 | static void | |
4844 | compose_sflow_cookie(const struct ofproto_dpif *ofproto, | |
4845 | ovs_be16 vlan_tci, uint32_t odp_port, | |
4846 | unsigned int n_outputs, union user_action_cookie *cookie) | |
4847 | { | |
4848 | int ifindex; | |
4849 | ||
4850 | cookie->type = USER_ACTION_COOKIE_SFLOW; | |
4851 | cookie->sflow.vlan_tci = vlan_tci; | |
4852 | ||
4853 | /* See http://www.sflow.org/sflow_version_5.txt (search for "Input/output | |
4854 | * port information") for the interpretation of cookie->output. */ | |
4855 | switch (n_outputs) { | |
4856 | case 0: | |
4857 | /* 0x40000000 | 256 means "packet dropped for unknown reason". */ | |
4858 | cookie->sflow.output = 0x40000000 | 256; | |
4859 | break; | |
4860 | ||
4861 | case 1: | |
4862 | ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port); | |
4863 | if (ifindex) { | |
4864 | cookie->sflow.output = ifindex; | |
4865 | break; | |
4866 | } | |
4867 | /* Fall through. */ | |
4868 | default: | |
4869 | /* 0x80000000 means "multiple output ports. */ | |
4870 | cookie->sflow.output = 0x80000000 | n_outputs; | |
4871 | break; | |
4872 | } | |
4873 | } | |
4874 | ||
4875 | /* Compose SAMPLE action for sFlow. */ | |
4876 | static size_t | |
4877 | compose_sflow_action(const struct ofproto_dpif *ofproto, | |
4878 | struct ofpbuf *odp_actions, | |
4879 | const struct flow *flow, | |
4880 | uint32_t odp_port) | |
4881 | { | |
4882 | uint32_t probability; | |
4883 | union user_action_cookie cookie; | |
4884 | size_t sample_offset, actions_offset; | |
4885 | int cookie_offset; | |
4886 | ||
4887 | if (!ofproto->sflow || flow->in_port == OFPP_NONE) { | |
4888 | return 0; | |
4889 | } | |
4890 | ||
4891 | sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE); | |
4892 | ||
4893 | /* Number of packets out of UINT_MAX to sample. */ | |
4894 | probability = dpif_sflow_get_probability(ofproto->sflow); | |
4895 | nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability); | |
4896 | ||
4897 | actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS); | |
4898 | compose_sflow_cookie(ofproto, htons(0), odp_port, | |
4899 | odp_port == OVSP_NONE ? 0 : 1, &cookie); | |
4900 | cookie_offset = put_userspace_action(ofproto, odp_actions, flow, &cookie); | |
4901 | ||
4902 | nl_msg_end_nested(odp_actions, actions_offset); | |
4903 | nl_msg_end_nested(odp_actions, sample_offset); | |
4904 | return cookie_offset; | |
4905 | } | |
4906 | ||
4907 | /* SAMPLE action must be first action in any given list of actions. | |
4908 | * At this point we do not have all information required to build it. So try to | |
4909 | * build sample action as complete as possible. */ | |
4910 | static void | |
4911 | add_sflow_action(struct action_xlate_ctx *ctx) | |
4912 | { | |
4913 | ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto, | |
4914 | ctx->odp_actions, | |
4915 | &ctx->flow, OVSP_NONE); | |
4916 | ctx->sflow_odp_port = 0; | |
4917 | ctx->sflow_n_outputs = 0; | |
4918 | } | |
4919 | ||
4920 | /* Fix SAMPLE action according to data collected while composing ODP actions. | |
4921 | * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested | |
4922 | * USERSPACE action's user-cookie which is required for sflow. */ | |
4923 | static void | |
4924 | fix_sflow_action(struct action_xlate_ctx *ctx) | |
4925 | { | |
4926 | const struct flow *base = &ctx->base_flow; | |
4927 | union user_action_cookie *cookie; | |
4928 | ||
4929 | if (!ctx->user_cookie_offset) { | |
4930 | return; | |
4931 | } | |
4932 | ||
4933 | cookie = ofpbuf_at(ctx->odp_actions, ctx->user_cookie_offset, | |
4934 | sizeof(*cookie)); | |
4935 | assert(cookie->type == USER_ACTION_COOKIE_SFLOW); | |
4936 | ||
4937 | compose_sflow_cookie(ctx->ofproto, base->vlan_tci, | |
4938 | ctx->sflow_odp_port, ctx->sflow_n_outputs, cookie); | |
4939 | } | |
4940 | ||
4941 | static void | |
4942 | compose_output_action__(struct action_xlate_ctx *ctx, uint16_t ofp_port, | |
4943 | bool check_stp) | |
4944 | { | |
4945 | const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port); | |
4946 | uint16_t odp_port = ofp_port_to_odp_port(ofp_port); | |
4947 | ovs_be16 flow_vlan_tci = ctx->flow.vlan_tci; | |
4948 | uint8_t flow_nw_tos = ctx->flow.nw_tos; | |
4949 | uint16_t out_port; | |
4950 | ||
4951 | if (ofport) { | |
4952 | struct priority_to_dscp *pdscp; | |
4953 | ||
4954 | if (ofport->up.pp.config & OFPUTIL_PC_NO_FWD | |
4955 | || (check_stp && !stp_forward_in_state(ofport->stp_state))) { | |
4956 | return; | |
4957 | } | |
4958 | ||
4959 | pdscp = get_priority(ofport, ctx->flow.skb_priority); | |
4960 | if (pdscp) { | |
4961 | ctx->flow.nw_tos &= ~IP_DSCP_MASK; | |
4962 | ctx->flow.nw_tos |= pdscp->dscp; | |
4963 | } | |
4964 | } else { | |
4965 | /* We may not have an ofport record for this port, but it doesn't hurt | |
4966 | * to allow forwarding to it anyhow. Maybe such a port will appear | |
4967 | * later and we're pre-populating the flow table. */ | |
4968 | } | |
4969 | ||
4970 | out_port = vsp_realdev_to_vlandev(ctx->ofproto, odp_port, | |
4971 | ctx->flow.vlan_tci); | |
4972 | if (out_port != odp_port) { | |
4973 | ctx->flow.vlan_tci = htons(0); | |
4974 | } | |
4975 | commit_odp_actions(&ctx->flow, &ctx->base_flow, ctx->odp_actions); | |
4976 | nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_OUTPUT, out_port); | |
4977 | ||
4978 | ctx->sflow_odp_port = odp_port; | |
4979 | ctx->sflow_n_outputs++; | |
4980 | ctx->nf_output_iface = ofp_port; | |
4981 | ctx->flow.vlan_tci = flow_vlan_tci; | |
4982 | ctx->flow.nw_tos = flow_nw_tos; | |
4983 | } | |
4984 | ||
4985 | static void | |
4986 | compose_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port) | |
4987 | { | |
4988 | compose_output_action__(ctx, ofp_port, true); | |
4989 | } | |
4990 | ||
4991 | static void | |
4992 | xlate_table_action(struct action_xlate_ctx *ctx, | |
4993 | uint16_t in_port, uint8_t table_id) | |
4994 | { | |
4995 | if (ctx->recurse < MAX_RESUBMIT_RECURSION) { | |
4996 | struct ofproto_dpif *ofproto = ctx->ofproto; | |
4997 | struct rule_dpif *rule; | |
4998 | uint16_t old_in_port; | |
4999 | uint8_t old_table_id; | |
5000 | ||
5001 | old_table_id = ctx->table_id; | |
5002 | ctx->table_id = table_id; | |
5003 | ||
5004 | /* Look up a flow with 'in_port' as the input port. */ | |
5005 | old_in_port = ctx->flow.in_port; | |
5006 | ctx->flow.in_port = in_port; | |
5007 | rule = rule_dpif_lookup__(ofproto, &ctx->flow, table_id); | |
5008 | ||
5009 | /* Tag the flow. */ | |
5010 | if (table_id > 0 && table_id < N_TABLES) { | |
5011 | struct table_dpif *table = &ofproto->tables[table_id]; | |
5012 | if (table->other_table) { | |
5013 | ctx->tags |= (rule && rule->tag | |
5014 | ? rule->tag | |
5015 | : rule_calculate_tag(&ctx->flow, | |
5016 | &table->other_table->wc, | |
5017 | table->basis)); | |
5018 | } | |
5019 | } | |
5020 | ||
5021 | /* Restore the original input port. Otherwise OFPP_NORMAL and | |
5022 | * OFPP_IN_PORT will have surprising behavior. */ | |
5023 | ctx->flow.in_port = old_in_port; | |
5024 | ||
5025 | if (ctx->resubmit_hook) { | |
5026 | ctx->resubmit_hook(ctx, rule); | |
5027 | } | |
5028 | ||
5029 | if (rule) { | |
5030 | struct rule_dpif *old_rule = ctx->rule; | |
5031 | ||
5032 | if (ctx->resubmit_stats) { | |
5033 | rule_credit_stats(rule, ctx->resubmit_stats); | |
5034 | } | |
5035 | ||
5036 | ctx->recurse++; | |
5037 | ctx->rule = rule; | |
5038 | do_xlate_actions(rule->up.ofpacts, rule->up.ofpacts_len, ctx); | |
5039 | ctx->rule = old_rule; | |
5040 | ctx->recurse--; | |
5041 | } | |
5042 | ||
5043 | ctx->table_id = old_table_id; | |
5044 | } else { | |
5045 | static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1); | |
5046 | ||
5047 | VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times", | |
5048 | MAX_RESUBMIT_RECURSION); | |
5049 | ctx->max_resubmit_trigger = true; | |
5050 | } | |
5051 | } | |
5052 | ||
5053 | static void | |
5054 | xlate_ofpact_resubmit(struct action_xlate_ctx *ctx, | |
5055 | const struct ofpact_resubmit *resubmit) | |
5056 | { | |
5057 | uint16_t in_port; | |
5058 | uint8_t table_id; | |
5059 | ||
5060 | in_port = resubmit->in_port; | |
5061 | if (in_port == OFPP_IN_PORT) { | |
5062 | in_port = ctx->flow.in_port; | |
5063 | } | |
5064 | ||
5065 | table_id = resubmit->table_id; | |
5066 | if (table_id == 255) { | |
5067 | table_id = ctx->table_id; | |
5068 | } | |
5069 | ||
5070 | xlate_table_action(ctx, in_port, table_id); | |
5071 | } | |
5072 | ||
5073 | static void | |
5074 | flood_packets(struct action_xlate_ctx *ctx, bool all) | |
5075 | { | |
5076 | struct ofport_dpif *ofport; | |
5077 | ||
5078 | HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) { | |
5079 | uint16_t ofp_port = ofport->up.ofp_port; | |
5080 | ||
5081 | if (ofp_port == ctx->flow.in_port) { | |
5082 | continue; | |
5083 | } | |
5084 | ||
5085 | if (all) { | |
5086 | compose_output_action__(ctx, ofp_port, false); | |
5087 | } else if (!(ofport->up.pp.config & OFPUTIL_PC_NO_FLOOD)) { | |
5088 | compose_output_action(ctx, ofp_port); | |
5089 | } | |
5090 | } | |
5091 | ||
5092 | ctx->nf_output_iface = NF_OUT_FLOOD; | |
5093 | } | |
5094 | ||
5095 | static void | |
5096 | execute_controller_action(struct action_xlate_ctx *ctx, int len, | |
5097 | enum ofp_packet_in_reason reason, | |
5098 | uint16_t controller_id) | |
5099 | { | |
5100 | struct ofputil_packet_in pin; | |
5101 | struct ofpbuf *packet; | |
5102 | ||
5103 | ctx->slow |= SLOW_CONTROLLER; | |
5104 | if (!ctx->packet) { | |
5105 | return; | |
5106 | } | |
5107 | ||
5108 | packet = ofpbuf_clone(ctx->packet); | |
5109 | ||
5110 | if (packet->l2 && packet->l3) { | |
5111 | struct eth_header *eh; | |
5112 | ||
5113 | eth_pop_vlan(packet); | |
5114 | eh = packet->l2; | |
5115 | ||
5116 | /* If the Ethernet type is less than ETH_TYPE_MIN, it's likely an 802.2 | |
5117 | * LLC frame. Calculating the Ethernet type of these frames is more | |
5118 | * trouble than seems appropriate for a simple assertion. */ | |
5119 | assert(ntohs(eh->eth_type) < ETH_TYPE_MIN | |
5120 | || eh->eth_type == ctx->flow.dl_type); | |
5121 | ||
5122 | memcpy(eh->eth_src, ctx->flow.dl_src, sizeof eh->eth_src); | |
5123 | memcpy(eh->eth_dst, ctx->flow.dl_dst, sizeof eh->eth_dst); | |
5124 | ||
5125 | if (ctx->flow.vlan_tci & htons(VLAN_CFI)) { | |
5126 | eth_push_vlan(packet, ctx->flow.vlan_tci); | |
5127 | } | |
5128 | ||
5129 | if (packet->l4) { | |
5130 | if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) { | |
5131 | packet_set_ipv4(packet, ctx->flow.nw_src, ctx->flow.nw_dst, | |
5132 | ctx->flow.nw_tos, ctx->flow.nw_ttl); | |
5133 | } | |
5134 | ||
5135 | if (packet->l7) { | |
5136 | if (ctx->flow.nw_proto == IPPROTO_TCP) { | |
5137 | packet_set_tcp_port(packet, ctx->flow.tp_src, | |
5138 | ctx->flow.tp_dst); | |
5139 | } else if (ctx->flow.nw_proto == IPPROTO_UDP) { | |
5140 | packet_set_udp_port(packet, ctx->flow.tp_src, | |
5141 | ctx->flow.tp_dst); | |
5142 | } | |
5143 | } | |
5144 | } | |
5145 | } | |
5146 | ||
5147 | pin.packet = packet->data; | |
5148 | pin.packet_len = packet->size; | |
5149 | pin.reason = reason; | |
5150 | pin.controller_id = controller_id; | |
5151 | pin.table_id = ctx->table_id; | |
5152 | pin.cookie = ctx->rule ? ctx->rule->up.flow_cookie : 0; | |
5153 | ||
5154 | pin.send_len = len; | |
5155 | flow_get_metadata(&ctx->flow, &pin.fmd); | |
5156 | ||
5157 | connmgr_send_packet_in(ctx->ofproto->up.connmgr, &pin); | |
5158 | ofpbuf_delete(packet); | |
5159 | } | |
5160 | ||
5161 | static bool | |
5162 | compose_dec_ttl(struct action_xlate_ctx *ctx) | |
5163 | { | |
5164 | if (ctx->flow.dl_type != htons(ETH_TYPE_IP) && | |
5165 | ctx->flow.dl_type != htons(ETH_TYPE_IPV6)) { | |
5166 | return false; | |
5167 | } | |
5168 | ||
5169 | if (ctx->flow.nw_ttl > 1) { | |
5170 | ctx->flow.nw_ttl--; | |
5171 | return false; | |
5172 | } else { | |
5173 | execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL, 0); | |
5174 | ||
5175 | /* Stop processing for current table. */ | |
5176 | return true; | |
5177 | } | |
5178 | } | |
5179 | ||
5180 | static void | |
5181 | xlate_output_action(struct action_xlate_ctx *ctx, | |
5182 | uint16_t port, uint16_t max_len) | |
5183 | { | |
5184 | uint16_t prev_nf_output_iface = ctx->nf_output_iface; | |
5185 | ||
5186 | ctx->nf_output_iface = NF_OUT_DROP; | |
5187 | ||
5188 | switch (port) { | |
5189 | case OFPP_IN_PORT: | |
5190 | compose_output_action(ctx, ctx->flow.in_port); | |
5191 | break; | |
5192 | case OFPP_TABLE: | |
5193 | xlate_table_action(ctx, ctx->flow.in_port, 0); | |
5194 | break; | |
5195 | case OFPP_NORMAL: | |
5196 | xlate_normal(ctx); | |
5197 | break; | |
5198 | case OFPP_FLOOD: | |
5199 | flood_packets(ctx, false); | |
5200 | break; | |
5201 | case OFPP_ALL: | |
5202 | flood_packets(ctx, true); | |
5203 | break; | |
5204 | case OFPP_CONTROLLER: | |
5205 | execute_controller_action(ctx, max_len, OFPR_ACTION, 0); | |
5206 | break; | |
5207 | case OFPP_NONE: | |
5208 | break; | |
5209 | case OFPP_LOCAL: | |
5210 | default: | |
5211 | if (port != ctx->flow.in_port) { | |
5212 | compose_output_action(ctx, port); | |
5213 | } | |
5214 | break; | |
5215 | } | |
5216 | ||
5217 | if (prev_nf_output_iface == NF_OUT_FLOOD) { | |
5218 | ctx->nf_output_iface = NF_OUT_FLOOD; | |
5219 | } else if (ctx->nf_output_iface == NF_OUT_DROP) { | |
5220 | ctx->nf_output_iface = prev_nf_output_iface; | |
5221 | } else if (prev_nf_output_iface != NF_OUT_DROP && | |
5222 | ctx->nf_output_iface != NF_OUT_FLOOD) { | |
5223 | ctx->nf_output_iface = NF_OUT_MULTI; | |
5224 | } | |
5225 | } | |
5226 | ||
5227 | static void | |
5228 | xlate_output_reg_action(struct action_xlate_ctx *ctx, | |
5229 | const struct ofpact_output_reg *or) | |
5230 | { | |
5231 | uint64_t port = mf_get_subfield(&or->src, &ctx->flow); | |
5232 | if (port <= UINT16_MAX) { | |
5233 | xlate_output_action(ctx, port, or->max_len); | |
5234 | } | |
5235 | } | |
5236 | ||
5237 | static void | |
5238 | xlate_enqueue_action(struct action_xlate_ctx *ctx, | |
5239 | const struct ofpact_enqueue *enqueue) | |
5240 | { | |
5241 | uint16_t ofp_port = enqueue->port; | |
5242 | uint32_t queue_id = enqueue->queue; | |
5243 | uint32_t flow_priority, priority; | |
5244 | int error; | |
5245 | ||
5246 | /* Translate queue to priority. */ | |
5247 | error = dpif_queue_to_priority(ctx->ofproto->dpif, queue_id, &priority); | |
5248 | if (error) { | |
5249 | /* Fall back to ordinary output action. */ | |
5250 | xlate_output_action(ctx, enqueue->port, 0); | |
5251 | return; | |
5252 | } | |
5253 | ||
5254 | /* Check output port. */ | |
5255 | if (ofp_port == OFPP_IN_PORT) { | |
5256 | ofp_port = ctx->flow.in_port; | |
5257 | } else if (ofp_port == ctx->flow.in_port) { | |
5258 | return; | |
5259 | } | |
5260 | ||
5261 | /* Add datapath actions. */ | |
5262 | flow_priority = ctx->flow.skb_priority; | |
5263 | ctx->flow.skb_priority = priority; | |
5264 | compose_output_action(ctx, ofp_port); | |
5265 | ctx->flow.skb_priority = flow_priority; | |
5266 | ||
5267 | /* Update NetFlow output port. */ | |
5268 | if (ctx->nf_output_iface == NF_OUT_DROP) { | |
5269 | ctx->nf_output_iface = ofp_port; | |
5270 | } else if (ctx->nf_output_iface != NF_OUT_FLOOD) { | |
5271 | ctx->nf_output_iface = NF_OUT_MULTI; | |
5272 | } | |
5273 | } | |
5274 | ||
5275 | static void | |
5276 | xlate_set_queue_action(struct action_xlate_ctx *ctx, uint32_t queue_id) | |
5277 | { | |
5278 | uint32_t skb_priority; | |
5279 | ||
5280 | if (!dpif_queue_to_priority(ctx->ofproto->dpif, queue_id, &skb_priority)) { | |
5281 | ctx->flow.skb_priority = skb_priority; | |
5282 | } else { | |
5283 | /* Couldn't translate queue to a priority. Nothing to do. A warning | |
5284 | * has already been logged. */ | |
5285 | } | |
5286 | } | |
5287 | ||
5288 | struct xlate_reg_state { | |
5289 | ovs_be16 vlan_tci; | |
5290 | ovs_be64 tun_id; | |
5291 | }; | |
5292 | ||
5293 | static void | |
5294 | xlate_autopath(struct action_xlate_ctx *ctx, | |
5295 | const struct ofpact_autopath *ap) | |
5296 | { | |
5297 | uint16_t ofp_port = ap->port; | |
5298 | struct ofport_dpif *port = get_ofp_port(ctx->ofproto, ofp_port); | |
5299 | ||
5300 | if (!port || !port->bundle) { | |
5301 | ofp_port = OFPP_NONE; | |
5302 | } else if (port->bundle->bond) { | |
5303 | /* Autopath does not support VLAN hashing. */ | |
5304 | struct ofport_dpif *slave = bond_choose_output_slave( | |
5305 | port->bundle->bond, &ctx->flow, 0, &ctx->tags); | |
5306 | if (slave) { | |
5307 | ofp_port = slave->up.ofp_port; | |
5308 | } | |
5309 | } | |
5310 | nxm_reg_load(&ap->dst, ofp_port, &ctx->flow); | |
5311 | } | |
5312 | ||
5313 | static bool | |
5314 | slave_enabled_cb(uint16_t ofp_port, void *ofproto_) | |
5315 | { | |
5316 | struct ofproto_dpif *ofproto = ofproto_; | |
5317 | struct ofport_dpif *port; | |
5318 | ||
5319 | switch (ofp_port) { | |
5320 | case OFPP_IN_PORT: | |
5321 | case OFPP_TABLE: | |
5322 | case OFPP_NORMAL: | |
5323 | case OFPP_FLOOD: | |
5324 | case OFPP_ALL: | |
5325 | case OFPP_NONE: | |
5326 | return true; | |
5327 | case OFPP_CONTROLLER: /* Not supported by the bundle action. */ | |
5328 | return false; | |
5329 | default: | |
5330 | port = get_ofp_port(ofproto, ofp_port); | |
5331 | return port ? port->may_enable : false; | |
5332 | } | |
5333 | } | |
5334 | ||
5335 | static void | |
5336 | xlate_bundle_action(struct action_xlate_ctx *ctx, | |
5337 | const struct ofpact_bundle *bundle) | |
5338 | { | |
5339 | uint16_t port; | |
5340 | ||
5341 | port = bundle_execute(bundle, &ctx->flow, slave_enabled_cb, ctx->ofproto); | |
5342 | if (bundle->dst.field) { | |
5343 | nxm_reg_load(&bundle->dst, port, &ctx->flow); | |
5344 | } else { | |
5345 | xlate_output_action(ctx, port, 0); | |
5346 | } | |
5347 | } | |
5348 | ||
5349 | static void | |
5350 | xlate_learn_action(struct action_xlate_ctx *ctx, | |
5351 | const struct ofpact_learn *learn) | |
5352 | { | |
5353 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1); | |
5354 | struct ofputil_flow_mod fm; | |
5355 | uint64_t ofpacts_stub[1024 / 8]; | |
5356 | struct ofpbuf ofpacts; | |
5357 | int error; | |
5358 | ||
5359 | ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub); | |
5360 | learn_execute(learn, &ctx->flow, &fm, &ofpacts); | |
5361 | ||
5362 | error = ofproto_flow_mod(&ctx->ofproto->up, &fm); | |
5363 | if (error && !VLOG_DROP_WARN(&rl)) { | |
5364 | VLOG_WARN("learning action failed to modify flow table (%s)", | |
5365 | ofperr_get_name(error)); | |
5366 | } | |
5367 | ||
5368 | ofpbuf_uninit(&ofpacts); | |
5369 | } | |
5370 | ||
5371 | /* Reduces '*timeout' to no more than 'max'. A value of zero in either case | |
5372 | * means "infinite". */ | |
5373 | static void | |
5374 | reduce_timeout(uint16_t max, uint16_t *timeout) | |
5375 | { | |
5376 | if (max && (!*timeout || *timeout > max)) { | |
5377 | *timeout = max; | |
5378 | } | |
5379 | } | |
5380 | ||
5381 | static void | |
5382 | xlate_fin_timeout(struct action_xlate_ctx *ctx, | |
5383 | const struct ofpact_fin_timeout *oft) | |
5384 | { | |
5385 | if (ctx->tcp_flags & (TCP_FIN | TCP_RST) && ctx->rule) { | |
5386 | struct rule_dpif *rule = ctx->rule; | |
5387 | ||
5388 | reduce_timeout(oft->fin_idle_timeout, &rule->up.idle_timeout); | |
5389 | reduce_timeout(oft->fin_hard_timeout, &rule->up.hard_timeout); | |
5390 | } | |
5391 | } | |
5392 | ||
5393 | static bool | |
5394 | may_receive(const struct ofport_dpif *port, struct action_xlate_ctx *ctx) | |
5395 | { | |
5396 | if (port->up.pp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp) | |
5397 | ? OFPUTIL_PC_NO_RECV_STP | |
5398 | : OFPUTIL_PC_NO_RECV)) { | |
5399 | return false; | |
5400 | } | |
5401 | ||
5402 | /* Only drop packets here if both forwarding and learning are | |
5403 | * disabled. If just learning is enabled, we need to have | |
5404 | * OFPP_NORMAL and the learning action have a look at the packet | |
5405 | * before we can drop it. */ | |
5406 | if (!stp_forward_in_state(port->stp_state) | |
5407 | && !stp_learn_in_state(port->stp_state)) { | |
5408 | return false; | |
5409 | } | |
5410 | ||
5411 | return true; | |
5412 | } | |
5413 | ||
5414 | static void | |
5415 | do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len, | |
5416 | struct action_xlate_ctx *ctx) | |
5417 | { | |
5418 | const struct ofport_dpif *port; | |
5419 | bool was_evictable = true; | |
5420 | const struct ofpact *a; | |
5421 | ||
5422 | port = get_ofp_port(ctx->ofproto, ctx->flow.in_port); | |
5423 | if (port && !may_receive(port, ctx)) { | |
5424 | /* Drop this flow. */ | |
5425 | return; | |
5426 | } | |
5427 | ||
5428 | if (ctx->rule) { | |
5429 | /* Don't let the rule we're working on get evicted underneath us. */ | |
5430 | was_evictable = ctx->rule->up.evictable; | |
5431 | ctx->rule->up.evictable = false; | |
5432 | } | |
5433 | OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) { | |
5434 | struct ofpact_controller *controller; | |
5435 | ||
5436 | if (ctx->exit) { | |
5437 | break; | |
5438 | } | |
5439 | ||
5440 | switch (a->type) { | |
5441 | case OFPACT_OUTPUT: | |
5442 | xlate_output_action(ctx, ofpact_get_OUTPUT(a)->port, | |
5443 | ofpact_get_OUTPUT(a)->max_len); | |
5444 | break; | |
5445 | ||
5446 | case OFPACT_CONTROLLER: | |
5447 | controller = ofpact_get_CONTROLLER(a); | |
5448 | execute_controller_action(ctx, controller->max_len, | |
5449 | controller->reason, | |
5450 | controller->controller_id); | |
5451 | break; | |
5452 | ||
5453 | case OFPACT_ENQUEUE: | |
5454 | xlate_enqueue_action(ctx, ofpact_get_ENQUEUE(a)); | |
5455 | break; | |
5456 | ||
5457 | case OFPACT_SET_VLAN_VID: | |
5458 | ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK); | |
5459 | ctx->flow.vlan_tci |= (htons(ofpact_get_SET_VLAN_VID(a)->vlan_vid) | |
5460 | | htons(VLAN_CFI)); | |
5461 | break; | |
5462 | ||
5463 | case OFPACT_SET_VLAN_PCP: | |
5464 | ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK); | |
5465 | ctx->flow.vlan_tci |= htons((ofpact_get_SET_VLAN_PCP(a)->vlan_pcp | |
5466 | << VLAN_PCP_SHIFT) | |
5467 | | VLAN_CFI); | |
5468 | break; | |
5469 | ||
5470 | case OFPACT_STRIP_VLAN: | |
5471 | ctx->flow.vlan_tci = htons(0); | |
5472 | break; | |
5473 | ||
5474 | case OFPACT_SET_ETH_SRC: | |
5475 | memcpy(ctx->flow.dl_src, ofpact_get_SET_ETH_SRC(a)->mac, | |
5476 | ETH_ADDR_LEN); | |
5477 | break; | |
5478 | ||
5479 | case OFPACT_SET_ETH_DST: | |
5480 | memcpy(ctx->flow.dl_dst, ofpact_get_SET_ETH_DST(a)->mac, | |
5481 | ETH_ADDR_LEN); | |
5482 | break; | |
5483 | ||
5484 | case OFPACT_SET_IPV4_SRC: | |
5485 | ctx->flow.nw_src = ofpact_get_SET_IPV4_SRC(a)->ipv4; | |
5486 | break; | |
5487 | ||
5488 | case OFPACT_SET_IPV4_DST: | |
5489 | ctx->flow.nw_dst = ofpact_get_SET_IPV4_DST(a)->ipv4; | |
5490 | break; | |
5491 | ||
5492 | case OFPACT_SET_IPV4_DSCP: | |
5493 | /* OpenFlow 1.0 only supports IPv4. */ | |
5494 | if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) { | |
5495 | ctx->flow.nw_tos &= ~IP_DSCP_MASK; | |
5496 | ctx->flow.nw_tos |= ofpact_get_SET_IPV4_DSCP(a)->dscp; | |
5497 | } | |
5498 | break; | |
5499 | ||
5500 | case OFPACT_SET_L4_SRC_PORT: | |
5501 | ctx->flow.tp_src = htons(ofpact_get_SET_L4_SRC_PORT(a)->port); | |
5502 | break; | |
5503 | ||
5504 | case OFPACT_SET_L4_DST_PORT: | |
5505 | ctx->flow.tp_dst = htons(ofpact_get_SET_L4_DST_PORT(a)->port); | |
5506 | break; | |
5507 | ||
5508 | case OFPACT_RESUBMIT: | |
5509 | xlate_ofpact_resubmit(ctx, ofpact_get_RESUBMIT(a)); | |
5510 | break; | |
5511 | ||
5512 | case OFPACT_SET_TUNNEL: | |
5513 | ctx->flow.tun_id = htonll(ofpact_get_SET_TUNNEL(a)->tun_id); | |
5514 | break; | |
5515 | ||
5516 | case OFPACT_SET_QUEUE: | |
5517 | xlate_set_queue_action(ctx, ofpact_get_SET_QUEUE(a)->queue_id); | |
5518 | break; | |
5519 | ||
5520 | case OFPACT_POP_QUEUE: | |
5521 | ctx->flow.skb_priority = ctx->orig_skb_priority; | |
5522 | break; | |
5523 | ||
5524 | case OFPACT_REG_MOVE: | |
5525 | nxm_execute_reg_move(ofpact_get_REG_MOVE(a), &ctx->flow); | |
5526 | break; | |
5527 | ||
5528 | case OFPACT_REG_LOAD: | |
5529 | nxm_execute_reg_load(ofpact_get_REG_LOAD(a), &ctx->flow); | |
5530 | break; | |
5531 | ||
5532 | case OFPACT_DEC_TTL: | |
5533 | if (compose_dec_ttl(ctx)) { | |
5534 | goto out; | |
5535 | } | |
5536 | break; | |
5537 | ||
5538 | case OFPACT_NOTE: | |
5539 | /* Nothing to do. */ | |
5540 | break; | |
5541 | ||
5542 | case OFPACT_MULTIPATH: | |
5543 | multipath_execute(ofpact_get_MULTIPATH(a), &ctx->flow); | |
5544 | break; | |
5545 | ||
5546 | case OFPACT_AUTOPATH: | |
5547 | xlate_autopath(ctx, ofpact_get_AUTOPATH(a)); | |
5548 | break; | |
5549 | ||
5550 | case OFPACT_BUNDLE: | |
5551 | ctx->ofproto->has_bundle_action = true; | |
5552 | xlate_bundle_action(ctx, ofpact_get_BUNDLE(a)); | |
5553 | break; | |
5554 | ||
5555 | case OFPACT_OUTPUT_REG: | |
5556 | xlate_output_reg_action(ctx, ofpact_get_OUTPUT_REG(a)); | |
5557 | break; | |
5558 | ||
5559 | case OFPACT_LEARN: | |
5560 | ctx->has_learn = true; | |
5561 | if (ctx->may_learn) { | |
5562 | xlate_learn_action(ctx, ofpact_get_LEARN(a)); | |
5563 | } | |
5564 | break; | |
5565 | ||
5566 | case OFPACT_EXIT: | |
5567 | ctx->exit = true; | |
5568 | break; | |
5569 | ||
5570 | case OFPACT_FIN_TIMEOUT: | |
5571 | ctx->has_fin_timeout = true; | |
5572 | xlate_fin_timeout(ctx, ofpact_get_FIN_TIMEOUT(a)); | |
5573 | break; | |
5574 | } | |
5575 | } | |
5576 | ||
5577 | out: | |
5578 | /* We've let OFPP_NORMAL and the learning action look at the packet, | |
5579 | * so drop it now if forwarding is disabled. */ | |
5580 | if (port && !stp_forward_in_state(port->stp_state)) { | |
5581 | ofpbuf_clear(ctx->odp_actions); | |
5582 | add_sflow_action(ctx); | |
5583 | } | |
5584 | if (ctx->rule) { | |
5585 | ctx->rule->up.evictable = was_evictable; | |
5586 | } | |
5587 | } | |
5588 | ||
5589 | static void | |
5590 | action_xlate_ctx_init(struct action_xlate_ctx *ctx, | |
5591 | struct ofproto_dpif *ofproto, const struct flow *flow, | |
5592 | ovs_be16 initial_tci, struct rule_dpif *rule, | |
5593 | uint8_t tcp_flags, const struct ofpbuf *packet) | |
5594 | { | |
5595 | ctx->ofproto = ofproto; | |
5596 | ctx->flow = *flow; | |
5597 | ctx->base_flow = ctx->flow; | |
5598 | ctx->base_flow.tun_id = 0; | |
5599 | ctx->base_flow.vlan_tci = initial_tci; | |
5600 | ctx->rule = rule; | |
5601 | ctx->packet = packet; | |
5602 | ctx->may_learn = packet != NULL; | |
5603 | ctx->tcp_flags = tcp_flags; | |
5604 | ctx->resubmit_hook = NULL; | |
5605 | ctx->report_hook = NULL; | |
5606 | ctx->resubmit_stats = NULL; | |
5607 | } | |
5608 | ||
5609 | /* Translates the 'ofpacts_len' bytes of "struct ofpacts" starting at 'ofpacts' | |
5610 | * into datapath actions in 'odp_actions', using 'ctx'. */ | |
5611 | static void | |
5612 | xlate_actions(struct action_xlate_ctx *ctx, | |
5613 | const struct ofpact *ofpacts, size_t ofpacts_len, | |
5614 | struct ofpbuf *odp_actions) | |
5615 | { | |
5616 | /* Normally false. Set to true if we ever hit MAX_RESUBMIT_RECURSION, so | |
5617 | * that in the future we always keep a copy of the original flow for | |
5618 | * tracing purposes. */ | |
5619 | static bool hit_resubmit_limit; | |
5620 | ||
5621 | enum slow_path_reason special; | |
5622 | ||
5623 | COVERAGE_INC(ofproto_dpif_xlate); | |
5624 | ||
5625 | ofpbuf_clear(odp_actions); | |
5626 | ofpbuf_reserve(odp_actions, NL_A_U32_SIZE); | |
5627 | ||
5628 | ctx->odp_actions = odp_actions; | |
5629 | ctx->tags = 0; | |
5630 | ctx->slow = 0; | |
5631 | ctx->has_learn = false; | |
5632 | ctx->has_normal = false; | |
5633 | ctx->has_fin_timeout = false; | |
5634 | ctx->nf_output_iface = NF_OUT_DROP; | |
5635 | ctx->mirrors = 0; | |
5636 | ctx->recurse = 0; | |
5637 | ctx->max_resubmit_trigger = false; | |
5638 | ctx->orig_skb_priority = ctx->flow.skb_priority; | |
5639 | ctx->table_id = 0; | |
5640 | ctx->exit = false; | |
5641 | ||
5642 | if (ctx->ofproto->has_mirrors || hit_resubmit_limit) { | |
5643 | /* Do this conditionally because the copy is expensive enough that it | |
5644 | * shows up in profiles. | |
5645 | * | |
5646 | * We keep orig_flow in 'ctx' only because I couldn't make GCC 4.4 | |
5647 | * believe that I wasn't using it without initializing it if I kept it | |
5648 | * in a local variable. */ | |
5649 | ctx->orig_flow = ctx->flow; | |
5650 | } | |
5651 | ||
5652 | if (ctx->flow.nw_frag & FLOW_NW_FRAG_ANY) { | |
5653 | switch (ctx->ofproto->up.frag_handling) { | |
5654 | case OFPC_FRAG_NORMAL: | |
5655 | /* We must pretend that transport ports are unavailable. */ | |
5656 | ctx->flow.tp_src = ctx->base_flow.tp_src = htons(0); | |
5657 | ctx->flow.tp_dst = ctx->base_flow.tp_dst = htons(0); | |
5658 | break; | |
5659 | ||
5660 | case OFPC_FRAG_DROP: | |
5661 | return; | |
5662 | ||
5663 | case OFPC_FRAG_REASM: | |
5664 | NOT_REACHED(); | |
5665 | ||
5666 | case OFPC_FRAG_NX_MATCH: | |
5667 | /* Nothing to do. */ | |
5668 | break; | |
5669 | ||
5670 | case OFPC_INVALID_TTL_TO_CONTROLLER: | |
5671 | NOT_REACHED(); | |
5672 | } | |
5673 | } | |
5674 | ||
5675 | special = process_special(ctx->ofproto, &ctx->flow, ctx->packet); | |
5676 | if (special) { | |
5677 | ctx->slow |= special; | |
5678 | } else { | |
5679 | static struct vlog_rate_limit trace_rl = VLOG_RATE_LIMIT_INIT(1, 1); | |
5680 | ovs_be16 initial_tci = ctx->base_flow.vlan_tci; | |
5681 | ||
5682 | add_sflow_action(ctx); | |
5683 | do_xlate_actions(ofpacts, ofpacts_len, ctx); | |
5684 | ||
5685 | if (ctx->max_resubmit_trigger && !ctx->resubmit_hook) { | |
5686 | if (!hit_resubmit_limit) { | |
5687 | /* We didn't record the original flow. Make sure we do from | |
5688 | * now on. */ | |
5689 | hit_resubmit_limit = true; | |
5690 | } else if (!VLOG_DROP_ERR(&trace_rl)) { | |
5691 | struct ds ds = DS_EMPTY_INITIALIZER; | |
5692 | ||
5693 | ofproto_trace(ctx->ofproto, &ctx->orig_flow, ctx->packet, | |
5694 | initial_tci, &ds); | |
5695 | VLOG_ERR("Trace triggered by excessive resubmit " | |
5696 | "recursion:\n%s", ds_cstr(&ds)); | |
5697 | ds_destroy(&ds); | |
5698 | } | |
5699 | } | |
5700 | ||
5701 | if (!connmgr_may_set_up_flow(ctx->ofproto->up.connmgr, &ctx->flow, | |
5702 | ctx->odp_actions->data, | |
5703 | ctx->odp_actions->size)) { | |
5704 | ctx->slow |= SLOW_IN_BAND; | |
5705 | if (ctx->packet | |
5706 | && connmgr_msg_in_hook(ctx->ofproto->up.connmgr, &ctx->flow, | |
5707 | ctx->packet)) { | |
5708 | compose_output_action(ctx, OFPP_LOCAL); | |
5709 | } | |
5710 | } | |
5711 | if (ctx->ofproto->has_mirrors) { | |
5712 | add_mirror_actions(ctx, &ctx->orig_flow); | |
5713 | } | |
5714 | fix_sflow_action(ctx); | |
5715 | } | |
5716 | } | |
5717 | ||
5718 | /* Translates the 'ofpacts_len' bytes of "struct ofpact"s starting at 'ofpacts' | |
5719 | * into datapath actions, using 'ctx', and discards the datapath actions. */ | |
5720 | static void | |
5721 | xlate_actions_for_side_effects(struct action_xlate_ctx *ctx, | |
5722 | const struct ofpact *ofpacts, | |
5723 | size_t ofpacts_len) | |
5724 | { | |
5725 | uint64_t odp_actions_stub[1024 / 8]; | |
5726 | struct ofpbuf odp_actions; | |
5727 | ||
5728 | ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub); | |
5729 | xlate_actions(ctx, ofpacts, ofpacts_len, &odp_actions); | |
5730 | ofpbuf_uninit(&odp_actions); | |
5731 | } | |
5732 | ||
5733 | static void | |
5734 | xlate_report(struct action_xlate_ctx *ctx, const char *s) | |
5735 | { | |
5736 | if (ctx->report_hook) { | |
5737 | ctx->report_hook(ctx, s); | |
5738 | } | |
5739 | } | |
5740 | \f | |
5741 | /* OFPP_NORMAL implementation. */ | |
5742 | ||
5743 | static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *); | |
5744 | ||
5745 | /* Given 'vid', the VID obtained from the 802.1Q header that was received as | |
5746 | * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle', | |
5747 | * the bundle on which the packet was received, returns the VLAN to which the | |
5748 | * packet belongs. | |
5749 | * | |
5750 | * Both 'vid' and the return value are in the range 0...4095. */ | |
5751 | static uint16_t | |
5752 | input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid) | |
5753 | { | |
5754 | switch (in_bundle->vlan_mode) { | |
5755 | case PORT_VLAN_ACCESS: | |
5756 | return in_bundle->vlan; | |
5757 | break; | |
5758 | ||
5759 | case PORT_VLAN_TRUNK: | |
5760 | return vid; | |
5761 | ||
5762 | case PORT_VLAN_NATIVE_UNTAGGED: | |
5763 | case PORT_VLAN_NATIVE_TAGGED: | |
5764 | return vid ? vid : in_bundle->vlan; | |
5765 | ||
5766 | default: | |
5767 | NOT_REACHED(); | |
5768 | } | |
5769 | } | |
5770 | ||
5771 | /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'. | |
5772 | * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs | |
5773 | * a warning. | |
5774 | * | |
5775 | * 'vid' should be the VID obtained from the 802.1Q header that was received as | |
5776 | * part of a packet (specify 0 if there was no 802.1Q header), in the range | |
5777 | * 0...4095. */ | |
5778 | static bool | |
5779 | input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn) | |
5780 | { | |
5781 | /* Allow any VID on the OFPP_NONE port. */ | |
5782 | if (in_bundle == &ofpp_none_bundle) { | |
5783 | return true; | |
5784 | } | |
5785 | ||
5786 | switch (in_bundle->vlan_mode) { | |
5787 | case PORT_VLAN_ACCESS: | |
5788 | if (vid) { | |
5789 | if (warn) { | |
5790 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
5791 | VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged " | |
5792 | "packet received on port %s configured as VLAN " | |
5793 | "%"PRIu16" access port", | |
5794 | in_bundle->ofproto->up.name, vid, | |
5795 | in_bundle->name, in_bundle->vlan); | |
5796 | } | |
5797 | return false; | |
5798 | } | |
5799 | return true; | |
5800 | ||
5801 | case PORT_VLAN_NATIVE_UNTAGGED: | |
5802 | case PORT_VLAN_NATIVE_TAGGED: | |
5803 | if (!vid) { | |
5804 | /* Port must always carry its native VLAN. */ | |
5805 | return true; | |
5806 | } | |
5807 | /* Fall through. */ | |
5808 | case PORT_VLAN_TRUNK: | |
5809 | if (!ofbundle_includes_vlan(in_bundle, vid)) { | |
5810 | if (warn) { | |
5811 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
5812 | VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet " | |
5813 | "received on port %s not configured for trunking " | |
5814 | "VLAN %"PRIu16, | |
5815 | in_bundle->ofproto->up.name, vid, | |
5816 | in_bundle->name, vid); | |
5817 | } | |
5818 | return false; | |
5819 | } | |
5820 | return true; | |
5821 | ||
5822 | default: | |
5823 | NOT_REACHED(); | |
5824 | } | |
5825 | ||
5826 | } | |
5827 | ||
5828 | /* Given 'vlan', the VLAN that a packet belongs to, and | |
5829 | * 'out_bundle', a bundle on which the packet is to be output, returns the VID | |
5830 | * that should be included in the 802.1Q header. (If the return value is 0, | |
5831 | * then the 802.1Q header should only be included in the packet if there is a | |
5832 | * nonzero PCP.) | |
5833 | * | |
5834 | * Both 'vlan' and the return value are in the range 0...4095. */ | |
5835 | static uint16_t | |
5836 | output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan) | |
5837 | { | |
5838 | switch (out_bundle->vlan_mode) { | |
5839 | case PORT_VLAN_ACCESS: | |
5840 | return 0; | |
5841 | ||
5842 | case PORT_VLAN_TRUNK: | |
5843 | case PORT_VLAN_NATIVE_TAGGED: | |
5844 | return vlan; | |
5845 | ||
5846 | case PORT_VLAN_NATIVE_UNTAGGED: | |
5847 | return vlan == out_bundle->vlan ? 0 : vlan; | |
5848 | ||
5849 | default: | |
5850 | NOT_REACHED(); | |
5851 | } | |
5852 | } | |
5853 | ||
5854 | static void | |
5855 | output_normal(struct action_xlate_ctx *ctx, const struct ofbundle *out_bundle, | |
5856 | uint16_t vlan) | |
5857 | { | |
5858 | struct ofport_dpif *port; | |
5859 | uint16_t vid; | |
5860 | ovs_be16 tci, old_tci; | |
5861 | ||
5862 | vid = output_vlan_to_vid(out_bundle, vlan); | |
5863 | if (!out_bundle->bond) { | |
5864 | port = ofbundle_get_a_port(out_bundle); | |
5865 | } else { | |
5866 | port = bond_choose_output_slave(out_bundle->bond, &ctx->flow, | |
5867 | vid, &ctx->tags); | |
5868 | if (!port) { | |
5869 | /* No slaves enabled, so drop packet. */ | |
5870 | return; | |
5871 | } | |
5872 | } | |
5873 | ||
5874 | old_tci = ctx->flow.vlan_tci; | |
5875 | tci = htons(vid); | |
5876 | if (tci || out_bundle->use_priority_tags) { | |
5877 | tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK); | |
5878 | if (tci) { | |
5879 | tci |= htons(VLAN_CFI); | |
5880 | } | |
5881 | } | |
5882 | ctx->flow.vlan_tci = tci; | |
5883 | ||
5884 | compose_output_action(ctx, port->up.ofp_port); | |
5885 | ctx->flow.vlan_tci = old_tci; | |
5886 | } | |
5887 | ||
5888 | static int | |
5889 | mirror_mask_ffs(mirror_mask_t mask) | |
5890 | { | |
5891 | BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask)); | |
5892 | return ffs(mask); | |
5893 | } | |
5894 | ||
5895 | static bool | |
5896 | ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan) | |
5897 | { | |
5898 | return (bundle->vlan_mode != PORT_VLAN_ACCESS | |
5899 | && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan))); | |
5900 | } | |
5901 | ||
5902 | static bool | |
5903 | ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan) | |
5904 | { | |
5905 | return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan); | |
5906 | } | |
5907 | ||
5908 | /* Returns an arbitrary interface within 'bundle'. */ | |
5909 | static struct ofport_dpif * | |
5910 | ofbundle_get_a_port(const struct ofbundle *bundle) | |
5911 | { | |
5912 | return CONTAINER_OF(list_front(&bundle->ports), | |
5913 | struct ofport_dpif, bundle_node); | |
5914 | } | |
5915 | ||
5916 | static bool | |
5917 | vlan_is_mirrored(const struct ofmirror *m, int vlan) | |
5918 | { | |
5919 | return !m->vlans || bitmap_is_set(m->vlans, vlan); | |
5920 | } | |
5921 | ||
5922 | static void | |
5923 | add_mirror_actions(struct action_xlate_ctx *ctx, const struct flow *orig_flow) | |
5924 | { | |
5925 | struct ofproto_dpif *ofproto = ctx->ofproto; | |
5926 | mirror_mask_t mirrors; | |
5927 | struct ofbundle *in_bundle; | |
5928 | uint16_t vlan; | |
5929 | uint16_t vid; | |
5930 | const struct nlattr *a; | |
5931 | size_t left; | |
5932 | ||
5933 | in_bundle = lookup_input_bundle(ctx->ofproto, orig_flow->in_port, | |
5934 | ctx->packet != NULL, NULL); | |
5935 | if (!in_bundle) { | |
5936 | return; | |
5937 | } | |
5938 | mirrors = in_bundle->src_mirrors; | |
5939 | ||
5940 | /* Drop frames on bundles reserved for mirroring. */ | |
5941 | if (in_bundle->mirror_out) { | |
5942 | if (ctx->packet != NULL) { | |
5943 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
5944 | VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port " | |
5945 | "%s, which is reserved exclusively for mirroring", | |
5946 | ctx->ofproto->up.name, in_bundle->name); | |
5947 | } | |
5948 | return; | |
5949 | } | |
5950 | ||
5951 | /* Check VLAN. */ | |
5952 | vid = vlan_tci_to_vid(orig_flow->vlan_tci); | |
5953 | if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) { | |
5954 | return; | |
5955 | } | |
5956 | vlan = input_vid_to_vlan(in_bundle, vid); | |
5957 | ||
5958 | /* Look at the output ports to check for destination selections. */ | |
5959 | ||
5960 | NL_ATTR_FOR_EACH (a, left, ctx->odp_actions->data, | |
5961 | ctx->odp_actions->size) { | |
5962 | enum ovs_action_attr type = nl_attr_type(a); | |
5963 | struct ofport_dpif *ofport; | |
5964 | ||
5965 | if (type != OVS_ACTION_ATTR_OUTPUT) { | |
5966 | continue; | |
5967 | } | |
5968 | ||
5969 | ofport = get_odp_port(ofproto, nl_attr_get_u32(a)); | |
5970 | if (ofport && ofport->bundle) { | |
5971 | mirrors |= ofport->bundle->dst_mirrors; | |
5972 | } | |
5973 | } | |
5974 | ||
5975 | if (!mirrors) { | |
5976 | return; | |
5977 | } | |
5978 | ||
5979 | /* Restore the original packet before adding the mirror actions. */ | |
5980 | ctx->flow = *orig_flow; | |
5981 | ||
5982 | while (mirrors) { | |
5983 | struct ofmirror *m; | |
5984 | ||
5985 | m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1]; | |
5986 | ||
5987 | if (!vlan_is_mirrored(m, vlan)) { | |
5988 | mirrors &= mirrors - 1; | |
5989 | continue; | |
5990 | } | |
5991 | ||
5992 | mirrors &= ~m->dup_mirrors; | |
5993 | ctx->mirrors |= m->dup_mirrors; | |
5994 | if (m->out) { | |
5995 | output_normal(ctx, m->out, vlan); | |
5996 | } else if (vlan != m->out_vlan | |
5997 | && !eth_addr_is_reserved(orig_flow->dl_dst)) { | |
5998 | struct ofbundle *bundle; | |
5999 | ||
6000 | HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { | |
6001 | if (ofbundle_includes_vlan(bundle, m->out_vlan) | |
6002 | && !bundle->mirror_out) { | |
6003 | output_normal(ctx, bundle, m->out_vlan); | |
6004 | } | |
6005 | } | |
6006 | } | |
6007 | } | |
6008 | } | |
6009 | ||
6010 | static void | |
6011 | update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors, | |
6012 | uint64_t packets, uint64_t bytes) | |
6013 | { | |
6014 | if (!mirrors) { | |
6015 | return; | |
6016 | } | |
6017 | ||
6018 | for (; mirrors; mirrors &= mirrors - 1) { | |
6019 | struct ofmirror *m; | |
6020 | ||
6021 | m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1]; | |
6022 | ||
6023 | if (!m) { | |
6024 | /* In normal circumstances 'm' will not be NULL. However, | |
6025 | * if mirrors are reconfigured, we can temporarily get out | |
6026 | * of sync in facet_revalidate(). We could "correct" the | |
6027 | * mirror list before reaching here, but doing that would | |
6028 | * not properly account the traffic stats we've currently | |
6029 | * accumulated for previous mirror configuration. */ | |
6030 | continue; | |
6031 | } | |
6032 | ||
6033 | m->packet_count += packets; | |
6034 | m->byte_count += bytes; | |
6035 | } | |
6036 | } | |
6037 | ||
6038 | /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after | |
6039 | * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to | |
6040 | * indicate this; newer upstream kernels use gratuitous ARP requests. */ | |
6041 | static bool | |
6042 | is_gratuitous_arp(const struct flow *flow) | |
6043 | { | |
6044 | return (flow->dl_type == htons(ETH_TYPE_ARP) | |
6045 | && eth_addr_is_broadcast(flow->dl_dst) | |
6046 | && (flow->nw_proto == ARP_OP_REPLY | |
6047 | || (flow->nw_proto == ARP_OP_REQUEST | |
6048 | && flow->nw_src == flow->nw_dst))); | |
6049 | } | |
6050 | ||
6051 | static void | |
6052 | update_learning_table(struct ofproto_dpif *ofproto, | |
6053 | const struct flow *flow, int vlan, | |
6054 | struct ofbundle *in_bundle) | |
6055 | { | |
6056 | struct mac_entry *mac; | |
6057 | ||
6058 | /* Don't learn the OFPP_NONE port. */ | |
6059 | if (in_bundle == &ofpp_none_bundle) { | |
6060 | return; | |
6061 | } | |
6062 | ||
6063 | if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) { | |
6064 | return; | |
6065 | } | |
6066 | ||
6067 | mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan); | |
6068 | if (is_gratuitous_arp(flow)) { | |
6069 | /* We don't want to learn from gratuitous ARP packets that are | |
6070 | * reflected back over bond slaves so we lock the learning table. */ | |
6071 | if (!in_bundle->bond) { | |
6072 | mac_entry_set_grat_arp_lock(mac); | |
6073 | } else if (mac_entry_is_grat_arp_locked(mac)) { | |
6074 | return; | |
6075 | } | |
6076 | } | |
6077 | ||
6078 | if (mac_entry_is_new(mac) || mac->port.p != in_bundle) { | |
6079 | /* The log messages here could actually be useful in debugging, | |
6080 | * so keep the rate limit relatively high. */ | |
6081 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300); | |
6082 | VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is " | |
6083 | "on port %s in VLAN %d", | |
6084 | ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src), | |
6085 | in_bundle->name, vlan); | |
6086 | ||
6087 | mac->port.p = in_bundle; | |
6088 | tag_set_add(&ofproto->revalidate_set, | |
6089 | mac_learning_changed(ofproto->ml, mac)); | |
6090 | } | |
6091 | } | |
6092 | ||
6093 | static struct ofbundle * | |
6094 | lookup_input_bundle(const struct ofproto_dpif *ofproto, uint16_t in_port, | |
6095 | bool warn, struct ofport_dpif **in_ofportp) | |
6096 | { | |
6097 | struct ofport_dpif *ofport; | |
6098 | ||
6099 | /* Find the port and bundle for the received packet. */ | |
6100 | ofport = get_ofp_port(ofproto, in_port); | |
6101 | if (in_ofportp) { | |
6102 | *in_ofportp = ofport; | |
6103 | } | |
6104 | if (ofport && ofport->bundle) { | |
6105 | return ofport->bundle; | |
6106 | } | |
6107 | ||
6108 | /* Special-case OFPP_NONE, which a controller may use as the ingress | |
6109 | * port for traffic that it is sourcing. */ | |
6110 | if (in_port == OFPP_NONE) { | |
6111 | return &ofpp_none_bundle; | |
6112 | } | |
6113 | ||
6114 | /* Odd. A few possible reasons here: | |
6115 | * | |
6116 | * - We deleted a port but there are still a few packets queued up | |
6117 | * from it. | |
6118 | * | |
6119 | * - Someone externally added a port (e.g. "ovs-dpctl add-if") that | |
6120 | * we don't know about. | |
6121 | * | |
6122 | * - The ofproto client didn't configure the port as part of a bundle. | |
6123 | * This is particularly likely to happen if a packet was received on the | |
6124 | * port after it was created, but before the client had a chance to | |
6125 | * configure its bundle. | |
6126 | */ | |
6127 | if (warn) { | |
6128 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
6129 | ||
6130 | VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown " | |
6131 | "port %"PRIu16, ofproto->up.name, in_port); | |
6132 | } | |
6133 | return NULL; | |
6134 | } | |
6135 | ||
6136 | /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or | |
6137 | * dropped. Returns true if they may be forwarded, false if they should be | |
6138 | * dropped. | |
6139 | * | |
6140 | * 'in_port' must be the ofport_dpif that corresponds to flow->in_port. | |
6141 | * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull). | |
6142 | * | |
6143 | * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as | |
6144 | * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as | |
6145 | * checked by input_vid_is_valid(). | |
6146 | * | |
6147 | * May also add tags to '*tags', although the current implementation only does | |
6148 | * so in one special case. | |
6149 | */ | |
6150 | static bool | |
6151 | is_admissible(struct action_xlate_ctx *ctx, struct ofport_dpif *in_port, | |
6152 | uint16_t vlan) | |
6153 | { | |
6154 | struct ofproto_dpif *ofproto = ctx->ofproto; | |
6155 | struct flow *flow = &ctx->flow; | |
6156 | struct ofbundle *in_bundle = in_port->bundle; | |
6157 | ||
6158 | /* Drop frames for reserved multicast addresses | |
6159 | * only if forward_bpdu option is absent. */ | |
6160 | if (!ofproto->up.forward_bpdu && eth_addr_is_reserved(flow->dl_dst)) { | |
6161 | xlate_report(ctx, "packet has reserved destination MAC, dropping"); | |
6162 | return false; | |
6163 | } | |
6164 | ||
6165 | if (in_bundle->bond) { | |
6166 | struct mac_entry *mac; | |
6167 | ||
6168 | switch (bond_check_admissibility(in_bundle->bond, in_port, | |
6169 | flow->dl_dst, &ctx->tags)) { | |
6170 | case BV_ACCEPT: | |
6171 | break; | |
6172 | ||
6173 | case BV_DROP: | |
6174 | xlate_report(ctx, "bonding refused admissibility, dropping"); | |
6175 | return false; | |
6176 | ||
6177 | case BV_DROP_IF_MOVED: | |
6178 | mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL); | |
6179 | if (mac && mac->port.p != in_bundle && | |
6180 | (!is_gratuitous_arp(flow) | |
6181 | || mac_entry_is_grat_arp_locked(mac))) { | |
6182 | xlate_report(ctx, "SLB bond thinks this packet looped back, " | |
6183 | "dropping"); | |
6184 | return false; | |
6185 | } | |
6186 | break; | |
6187 | } | |
6188 | } | |
6189 | ||
6190 | return true; | |
6191 | } | |
6192 | ||
6193 | static void | |
6194 | xlate_normal(struct action_xlate_ctx *ctx) | |
6195 | { | |
6196 | struct ofport_dpif *in_port; | |
6197 | struct ofbundle *in_bundle; | |
6198 | struct mac_entry *mac; | |
6199 | uint16_t vlan; | |
6200 | uint16_t vid; | |
6201 | ||
6202 | ctx->has_normal = true; | |
6203 | ||
6204 | in_bundle = lookup_input_bundle(ctx->ofproto, ctx->flow.in_port, | |
6205 | ctx->packet != NULL, &in_port); | |
6206 | if (!in_bundle) { | |
6207 | xlate_report(ctx, "no input bundle, dropping"); | |
6208 | return; | |
6209 | } | |
6210 | ||
6211 | /* Drop malformed frames. */ | |
6212 | if (ctx->flow.dl_type == htons(ETH_TYPE_VLAN) && | |
6213 | !(ctx->flow.vlan_tci & htons(VLAN_CFI))) { | |
6214 | if (ctx->packet != NULL) { | |
6215 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
6216 | VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial " | |
6217 | "VLAN tag received on port %s", | |
6218 | ctx->ofproto->up.name, in_bundle->name); | |
6219 | } | |
6220 | xlate_report(ctx, "partial VLAN tag, dropping"); | |
6221 | return; | |
6222 | } | |
6223 | ||
6224 | /* Drop frames on bundles reserved for mirroring. */ | |
6225 | if (in_bundle->mirror_out) { | |
6226 | if (ctx->packet != NULL) { | |
6227 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
6228 | VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port " | |
6229 | "%s, which is reserved exclusively for mirroring", | |
6230 | ctx->ofproto->up.name, in_bundle->name); | |
6231 | } | |
6232 | xlate_report(ctx, "input port is mirror output port, dropping"); | |
6233 | return; | |
6234 | } | |
6235 | ||
6236 | /* Check VLAN. */ | |
6237 | vid = vlan_tci_to_vid(ctx->flow.vlan_tci); | |
6238 | if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) { | |
6239 | xlate_report(ctx, "disallowed VLAN VID for this input port, dropping"); | |
6240 | return; | |
6241 | } | |
6242 | vlan = input_vid_to_vlan(in_bundle, vid); | |
6243 | ||
6244 | /* Check other admissibility requirements. */ | |
6245 | if (in_port && !is_admissible(ctx, in_port, vlan)) { | |
6246 | return; | |
6247 | } | |
6248 | ||
6249 | /* Learn source MAC. */ | |
6250 | if (ctx->may_learn) { | |
6251 | update_learning_table(ctx->ofproto, &ctx->flow, vlan, in_bundle); | |
6252 | } | |
6253 | ||
6254 | /* Determine output bundle. */ | |
6255 | mac = mac_learning_lookup(ctx->ofproto->ml, ctx->flow.dl_dst, vlan, | |
6256 | &ctx->tags); | |
6257 | if (mac) { | |
6258 | if (mac->port.p != in_bundle) { | |
6259 | xlate_report(ctx, "forwarding to learned port"); | |
6260 | output_normal(ctx, mac->port.p, vlan); | |
6261 | } else { | |
6262 | xlate_report(ctx, "learned port is input port, dropping"); | |
6263 | } | |
6264 | } else { | |
6265 | struct ofbundle *bundle; | |
6266 | ||
6267 | xlate_report(ctx, "no learned MAC for destination, flooding"); | |
6268 | HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) { | |
6269 | if (bundle != in_bundle | |
6270 | && ofbundle_includes_vlan(bundle, vlan) | |
6271 | && bundle->floodable | |
6272 | && !bundle->mirror_out) { | |
6273 | output_normal(ctx, bundle, vlan); | |
6274 | } | |
6275 | } | |
6276 | ctx->nf_output_iface = NF_OUT_FLOOD; | |
6277 | } | |
6278 | } | |
6279 | \f | |
6280 | /* Optimized flow revalidation. | |
6281 | * | |
6282 | * It's a difficult problem, in general, to tell which facets need to have | |
6283 | * their actions recalculated whenever the OpenFlow flow table changes. We | |
6284 | * don't try to solve that general problem: for most kinds of OpenFlow flow | |
6285 | * table changes, we recalculate the actions for every facet. This is | |
6286 | * relatively expensive, but it's good enough if the OpenFlow flow table | |
6287 | * doesn't change very often. | |
6288 | * | |
6289 | * However, we can expect one particular kind of OpenFlow flow table change to | |
6290 | * happen frequently: changes caused by MAC learning. To avoid wasting a lot | |
6291 | * of CPU on revalidating every facet whenever MAC learning modifies the flow | |
6292 | * table, we add a special case that applies to flow tables in which every rule | |
6293 | * has the same form (that is, the same wildcards), except that the table is | |
6294 | * also allowed to have a single "catch-all" flow that matches all packets. We | |
6295 | * optimize this case by tagging all of the facets that resubmit into the table | |
6296 | * and invalidating the same tag whenever a flow changes in that table. The | |
6297 | * end result is that we revalidate just the facets that need it (and sometimes | |
6298 | * a few more, but not all of the facets or even all of the facets that | |
6299 | * resubmit to the table modified by MAC learning). */ | |
6300 | ||
6301 | /* Calculates the tag to use for 'flow' and wildcards 'wc' when it is inserted | |
6302 | * into an OpenFlow table with the given 'basis'. */ | |
6303 | static tag_type | |
6304 | rule_calculate_tag(const struct flow *flow, const struct flow_wildcards *wc, | |
6305 | uint32_t secret) | |
6306 | { | |
6307 | if (flow_wildcards_is_catchall(wc)) { | |
6308 | return 0; | |
6309 | } else { | |
6310 | struct flow tag_flow = *flow; | |
6311 | flow_zero_wildcards(&tag_flow, wc); | |
6312 | return tag_create_deterministic(flow_hash(&tag_flow, secret)); | |
6313 | } | |
6314 | } | |
6315 | ||
6316 | /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the | |
6317 | * taggability of that table. | |
6318 | * | |
6319 | * This function must be called after *each* change to a flow table. If you | |
6320 | * skip calling it on some changes then the pointer comparisons at the end can | |
6321 | * be invalid if you get unlucky. For example, if a flow removal causes a | |
6322 | * cls_table to be destroyed and then a flow insertion causes a cls_table with | |
6323 | * different wildcards to be created with the same address, then this function | |
6324 | * will incorrectly skip revalidation. */ | |
6325 | static void | |
6326 | table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id) | |
6327 | { | |
6328 | struct table_dpif *table = &ofproto->tables[table_id]; | |
6329 | const struct oftable *oftable = &ofproto->up.tables[table_id]; | |
6330 | struct cls_table *catchall, *other; | |
6331 | struct cls_table *t; | |
6332 | ||
6333 | catchall = other = NULL; | |
6334 | ||
6335 | switch (hmap_count(&oftable->cls.tables)) { | |
6336 | case 0: | |
6337 | /* We could tag this OpenFlow table but it would make the logic a | |
6338 | * little harder and it's a corner case that doesn't seem worth it | |
6339 | * yet. */ | |
6340 | break; | |
6341 | ||
6342 | case 1: | |
6343 | case 2: | |
6344 | HMAP_FOR_EACH (t, hmap_node, &oftable->cls.tables) { | |
6345 | if (cls_table_is_catchall(t)) { | |
6346 | catchall = t; | |
6347 | } else if (!other) { | |
6348 | other = t; | |
6349 | } else { | |
6350 | /* Indicate that we can't tag this by setting both tables to | |
6351 | * NULL. (We know that 'catchall' is already NULL.) */ | |
6352 | other = NULL; | |
6353 | } | |
6354 | } | |
6355 | break; | |
6356 | ||
6357 | default: | |
6358 | /* Can't tag this table. */ | |
6359 | break; | |
6360 | } | |
6361 | ||
6362 | if (table->catchall_table != catchall || table->other_table != other) { | |
6363 | table->catchall_table = catchall; | |
6364 | table->other_table = other; | |
6365 | ofproto->need_revalidate = REV_FLOW_TABLE; | |
6366 | } | |
6367 | } | |
6368 | ||
6369 | /* Given 'rule' that has changed in some way (either it is a rule being | |
6370 | * inserted, a rule being deleted, or a rule whose actions are being | |
6371 | * modified), marks facets for revalidation to ensure that packets will be | |
6372 | * forwarded correctly according to the new state of the flow table. | |
6373 | * | |
6374 | * This function must be called after *each* change to a flow table. See | |
6375 | * the comment on table_update_taggable() for more information. */ | |
6376 | static void | |
6377 | rule_invalidate(const struct rule_dpif *rule) | |
6378 | { | |
6379 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
6380 | ||
6381 | table_update_taggable(ofproto, rule->up.table_id); | |
6382 | ||
6383 | if (!ofproto->need_revalidate) { | |
6384 | struct table_dpif *table = &ofproto->tables[rule->up.table_id]; | |
6385 | ||
6386 | if (table->other_table && rule->tag) { | |
6387 | tag_set_add(&ofproto->revalidate_set, rule->tag); | |
6388 | } else { | |
6389 | ofproto->need_revalidate = REV_FLOW_TABLE; | |
6390 | } | |
6391 | } | |
6392 | } | |
6393 | \f | |
6394 | static bool | |
6395 | set_frag_handling(struct ofproto *ofproto_, | |
6396 | enum ofp_config_flags frag_handling) | |
6397 | { | |
6398 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
6399 | ||
6400 | if (frag_handling != OFPC_FRAG_REASM) { | |
6401 | ofproto->need_revalidate = REV_RECONFIGURE; | |
6402 | return true; | |
6403 | } else { | |
6404 | return false; | |
6405 | } | |
6406 | } | |
6407 | ||
6408 | static enum ofperr | |
6409 | packet_out(struct ofproto *ofproto_, struct ofpbuf *packet, | |
6410 | const struct flow *flow, | |
6411 | const struct ofpact *ofpacts, size_t ofpacts_len) | |
6412 | { | |
6413 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
6414 | enum ofperr error; | |
6415 | ||
6416 | if (flow->in_port >= ofproto->max_ports && flow->in_port < OFPP_MAX) { | |
6417 | return OFPERR_NXBRC_BAD_IN_PORT; | |
6418 | } | |
6419 | ||
6420 | error = ofpacts_check(ofpacts, ofpacts_len, flow, ofproto->max_ports); | |
6421 | if (!error) { | |
6422 | struct odputil_keybuf keybuf; | |
6423 | struct dpif_flow_stats stats; | |
6424 | ||
6425 | struct ofpbuf key; | |
6426 | ||
6427 | struct action_xlate_ctx ctx; | |
6428 | uint64_t odp_actions_stub[1024 / 8]; | |
6429 | struct ofpbuf odp_actions; | |
6430 | ||
6431 | ofpbuf_use_stack(&key, &keybuf, sizeof keybuf); | |
6432 | odp_flow_key_from_flow(&key, flow); | |
6433 | ||
6434 | dpif_flow_stats_extract(flow, packet, &stats); | |
6435 | ||
6436 | action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci, NULL, | |
6437 | packet_get_tcp_flags(packet, flow), packet); | |
6438 | ctx.resubmit_stats = &stats; | |
6439 | ||
6440 | ofpbuf_use_stub(&odp_actions, | |
6441 | odp_actions_stub, sizeof odp_actions_stub); | |
6442 | xlate_actions(&ctx, ofpacts, ofpacts_len, &odp_actions); | |
6443 | dpif_execute(ofproto->dpif, key.data, key.size, | |
6444 | odp_actions.data, odp_actions.size, packet); | |
6445 | ofpbuf_uninit(&odp_actions); | |
6446 | } | |
6447 | return error; | |
6448 | } | |
6449 | \f | |
6450 | /* NetFlow. */ | |
6451 | ||
6452 | static int | |
6453 | set_netflow(struct ofproto *ofproto_, | |
6454 | const struct netflow_options *netflow_options) | |
6455 | { | |
6456 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
6457 | ||
6458 | if (netflow_options) { | |
6459 | if (!ofproto->netflow) { | |
6460 | ofproto->netflow = netflow_create(); | |
6461 | } | |
6462 | return netflow_set_options(ofproto->netflow, netflow_options); | |
6463 | } else { | |
6464 | netflow_destroy(ofproto->netflow); | |
6465 | ofproto->netflow = NULL; | |
6466 | return 0; | |
6467 | } | |
6468 | } | |
6469 | ||
6470 | static void | |
6471 | get_netflow_ids(const struct ofproto *ofproto_, | |
6472 | uint8_t *engine_type, uint8_t *engine_id) | |
6473 | { | |
6474 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
6475 | ||
6476 | dpif_get_netflow_ids(ofproto->dpif, engine_type, engine_id); | |
6477 | } | |
6478 | ||
6479 | static void | |
6480 | send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet) | |
6481 | { | |
6482 | if (!facet_is_controller_flow(facet) && | |
6483 | netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) { | |
6484 | struct subfacet *subfacet; | |
6485 | struct ofexpired expired; | |
6486 | ||
6487 | LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) { | |
6488 | if (subfacet->path == SF_FAST_PATH) { | |
6489 | struct dpif_flow_stats stats; | |
6490 | ||
6491 | subfacet_reinstall(subfacet, &stats); | |
6492 | subfacet_update_stats(subfacet, &stats); | |
6493 | } | |
6494 | } | |
6495 | ||
6496 | expired.flow = facet->flow; | |
6497 | expired.packet_count = facet->packet_count; | |
6498 | expired.byte_count = facet->byte_count; | |
6499 | expired.used = facet->used; | |
6500 | netflow_expire(ofproto->netflow, &facet->nf_flow, &expired); | |
6501 | } | |
6502 | } | |
6503 | ||
6504 | static void | |
6505 | send_netflow_active_timeouts(struct ofproto_dpif *ofproto) | |
6506 | { | |
6507 | struct facet *facet; | |
6508 | ||
6509 | HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) { | |
6510 | send_active_timeout(ofproto, facet); | |
6511 | } | |
6512 | } | |
6513 | \f | |
6514 | static struct ofproto_dpif * | |
6515 | ofproto_dpif_lookup(const char *name) | |
6516 | { | |
6517 | struct ofproto_dpif *ofproto; | |
6518 | ||
6519 | HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node, | |
6520 | hash_string(name, 0), &all_ofproto_dpifs) { | |
6521 | if (!strcmp(ofproto->up.name, name)) { | |
6522 | return ofproto; | |
6523 | } | |
6524 | } | |
6525 | return NULL; | |
6526 | } | |
6527 | ||
6528 | static void | |
6529 | ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc, | |
6530 | const char *argv[], void *aux OVS_UNUSED) | |
6531 | { | |
6532 | struct ofproto_dpif *ofproto; | |
6533 | ||
6534 | if (argc > 1) { | |
6535 | ofproto = ofproto_dpif_lookup(argv[1]); | |
6536 | if (!ofproto) { | |
6537 | unixctl_command_reply_error(conn, "no such bridge"); | |
6538 | return; | |
6539 | } | |
6540 | mac_learning_flush(ofproto->ml, &ofproto->revalidate_set); | |
6541 | } else { | |
6542 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
6543 | mac_learning_flush(ofproto->ml, &ofproto->revalidate_set); | |
6544 | } | |
6545 | } | |
6546 | ||
6547 | unixctl_command_reply(conn, "table successfully flushed"); | |
6548 | } | |
6549 | ||
6550 | static void | |
6551 | ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED, | |
6552 | const char *argv[], void *aux OVS_UNUSED) | |
6553 | { | |
6554 | struct ds ds = DS_EMPTY_INITIALIZER; | |
6555 | const struct ofproto_dpif *ofproto; | |
6556 | const struct mac_entry *e; | |
6557 | ||
6558 | ofproto = ofproto_dpif_lookup(argv[1]); | |
6559 | if (!ofproto) { | |
6560 | unixctl_command_reply_error(conn, "no such bridge"); | |
6561 | return; | |
6562 | } | |
6563 | ||
6564 | ds_put_cstr(&ds, " port VLAN MAC Age\n"); | |
6565 | LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) { | |
6566 | struct ofbundle *bundle = e->port.p; | |
6567 | ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n", | |
6568 | ofbundle_get_a_port(bundle)->odp_port, | |
6569 | e->vlan, ETH_ADDR_ARGS(e->mac), | |
6570 | mac_entry_age(ofproto->ml, e)); | |
6571 | } | |
6572 | unixctl_command_reply(conn, ds_cstr(&ds)); | |
6573 | ds_destroy(&ds); | |
6574 | } | |
6575 | ||
6576 | struct trace_ctx { | |
6577 | struct action_xlate_ctx ctx; | |
6578 | struct flow flow; | |
6579 | struct ds *result; | |
6580 | }; | |
6581 | ||
6582 | static void | |
6583 | trace_format_rule(struct ds *result, uint8_t table_id, int level, | |
6584 | const struct rule_dpif *rule) | |
6585 | { | |
6586 | ds_put_char_multiple(result, '\t', level); | |
6587 | if (!rule) { | |
6588 | ds_put_cstr(result, "No match\n"); | |
6589 | return; | |
6590 | } | |
6591 | ||
6592 | ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ", | |
6593 | table_id, ntohll(rule->up.flow_cookie)); | |
6594 | cls_rule_format(&rule->up.cr, result); | |
6595 | ds_put_char(result, '\n'); | |
6596 | ||
6597 | ds_put_char_multiple(result, '\t', level); | |
6598 | ds_put_cstr(result, "OpenFlow "); | |
6599 | ofpacts_format(rule->up.ofpacts, rule->up.ofpacts_len, result); | |
6600 | ds_put_char(result, '\n'); | |
6601 | } | |
6602 | ||
6603 | static void | |
6604 | trace_format_flow(struct ds *result, int level, const char *title, | |
6605 | struct trace_ctx *trace) | |
6606 | { | |
6607 | ds_put_char_multiple(result, '\t', level); | |
6608 | ds_put_format(result, "%s: ", title); | |
6609 | if (flow_equal(&trace->ctx.flow, &trace->flow)) { | |
6610 | ds_put_cstr(result, "unchanged"); | |
6611 | } else { | |
6612 | flow_format(result, &trace->ctx.flow); | |
6613 | trace->flow = trace->ctx.flow; | |
6614 | } | |
6615 | ds_put_char(result, '\n'); | |
6616 | } | |
6617 | ||
6618 | static void | |
6619 | trace_format_regs(struct ds *result, int level, const char *title, | |
6620 | struct trace_ctx *trace) | |
6621 | { | |
6622 | size_t i; | |
6623 | ||
6624 | ds_put_char_multiple(result, '\t', level); | |
6625 | ds_put_format(result, "%s:", title); | |
6626 | for (i = 0; i < FLOW_N_REGS; i++) { | |
6627 | ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]); | |
6628 | } | |
6629 | ds_put_char(result, '\n'); | |
6630 | } | |
6631 | ||
6632 | static void | |
6633 | trace_format_odp(struct ds *result, int level, const char *title, | |
6634 | struct trace_ctx *trace) | |
6635 | { | |
6636 | struct ofpbuf *odp_actions = trace->ctx.odp_actions; | |
6637 | ||
6638 | ds_put_char_multiple(result, '\t', level); | |
6639 | ds_put_format(result, "%s: ", title); | |
6640 | format_odp_actions(result, odp_actions->data, odp_actions->size); | |
6641 | ds_put_char(result, '\n'); | |
6642 | } | |
6643 | ||
6644 | static void | |
6645 | trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule) | |
6646 | { | |
6647 | struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx); | |
6648 | struct ds *result = trace->result; | |
6649 | ||
6650 | ds_put_char(result, '\n'); | |
6651 | trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace); | |
6652 | trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace); | |
6653 | trace_format_odp(result, ctx->recurse + 1, "Resubmitted odp", trace); | |
6654 | trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule); | |
6655 | } | |
6656 | ||
6657 | static void | |
6658 | trace_report(struct action_xlate_ctx *ctx, const char *s) | |
6659 | { | |
6660 | struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx); | |
6661 | struct ds *result = trace->result; | |
6662 | ||
6663 | ds_put_char_multiple(result, '\t', ctx->recurse); | |
6664 | ds_put_cstr(result, s); | |
6665 | ds_put_char(result, '\n'); | |
6666 | } | |
6667 | ||
6668 | static void | |
6669 | ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[], | |
6670 | void *aux OVS_UNUSED) | |
6671 | { | |
6672 | const char *dpname = argv[1]; | |
6673 | struct ofproto_dpif *ofproto; | |
6674 | struct ofpbuf odp_key; | |
6675 | struct ofpbuf *packet; | |
6676 | ovs_be16 initial_tci; | |
6677 | struct ds result; | |
6678 | struct flow flow; | |
6679 | char *s; | |
6680 | ||
6681 | packet = NULL; | |
6682 | ofpbuf_init(&odp_key, 0); | |
6683 | ds_init(&result); | |
6684 | ||
6685 | ofproto = ofproto_dpif_lookup(dpname); | |
6686 | if (!ofproto) { | |
6687 | unixctl_command_reply_error(conn, "Unknown ofproto (use ofproto/list " | |
6688 | "for help)"); | |
6689 | goto exit; | |
6690 | } | |
6691 | if (argc == 3 || (argc == 4 && !strcmp(argv[3], "-generate"))) { | |
6692 | /* ofproto/trace dpname flow [-generate] */ | |
6693 | const char *flow_s = argv[2]; | |
6694 | const char *generate_s = argv[3]; | |
6695 | ||
6696 | /* Allow 'flow_s' to be either a datapath flow or an OpenFlow-like | |
6697 | * flow. We guess which type it is based on whether 'flow_s' contains | |
6698 | * an '(', since a datapath flow always contains '(') but an | |
6699 | * OpenFlow-like flow should not (in fact it's allowed but I believe | |
6700 | * that's not documented anywhere). | |
6701 | * | |
6702 | * An alternative would be to try to parse 'flow_s' both ways, but then | |
6703 | * it would be tricky giving a sensible error message. After all, do | |
6704 | * you just say "syntax error" or do you present both error messages? | |
6705 | * Both choices seem lousy. */ | |
6706 | if (strchr(flow_s, '(')) { | |
6707 | int error; | |
6708 | ||
6709 | /* Convert string to datapath key. */ | |
6710 | ofpbuf_init(&odp_key, 0); | |
6711 | error = odp_flow_key_from_string(flow_s, NULL, &odp_key); | |
6712 | if (error) { | |
6713 | unixctl_command_reply_error(conn, "Bad flow syntax"); | |
6714 | goto exit; | |
6715 | } | |
6716 | ||
6717 | /* Convert odp_key to flow. */ | |
6718 | error = ofproto_dpif_extract_flow_key(ofproto, odp_key.data, | |
6719 | odp_key.size, &flow, | |
6720 | &initial_tci, NULL); | |
6721 | if (error == ODP_FIT_ERROR) { | |
6722 | unixctl_command_reply_error(conn, "Invalid flow"); | |
6723 | goto exit; | |
6724 | } | |
6725 | } else { | |
6726 | char *error_s; | |
6727 | ||
6728 | error_s = parse_ofp_exact_flow(&flow, argv[2]); | |
6729 | if (error_s) { | |
6730 | unixctl_command_reply_error(conn, error_s); | |
6731 | free(error_s); | |
6732 | goto exit; | |
6733 | } | |
6734 | ||
6735 | initial_tci = flow.vlan_tci; | |
6736 | vsp_adjust_flow(ofproto, &flow); | |
6737 | } | |
6738 | ||
6739 | /* Generate a packet, if requested. */ | |
6740 | if (generate_s) { | |
6741 | packet = ofpbuf_new(0); | |
6742 | flow_compose(packet, &flow); | |
6743 | } | |
6744 | } else if (argc == 6) { | |
6745 | /* ofproto/trace dpname priority tun_id in_port packet */ | |
6746 | const char *priority_s = argv[2]; | |
6747 | const char *tun_id_s = argv[3]; | |
6748 | const char *in_port_s = argv[4]; | |
6749 | const char *packet_s = argv[5]; | |
6750 | uint16_t in_port = ofp_port_to_odp_port(atoi(in_port_s)); | |
6751 | ovs_be64 tun_id = htonll(strtoull(tun_id_s, NULL, 0)); | |
6752 | uint32_t priority = atoi(priority_s); | |
6753 | const char *msg; | |
6754 | ||
6755 | msg = eth_from_hex(packet_s, &packet); | |
6756 | if (msg) { | |
6757 | unixctl_command_reply_error(conn, msg); | |
6758 | goto exit; | |
6759 | } | |
6760 | ||
6761 | ds_put_cstr(&result, "Packet: "); | |
6762 | s = ofp_packet_to_string(packet->data, packet->size); | |
6763 | ds_put_cstr(&result, s); | |
6764 | free(s); | |
6765 | ||
6766 | flow_extract(packet, priority, tun_id, in_port, &flow); | |
6767 | initial_tci = flow.vlan_tci; | |
6768 | } else { | |
6769 | unixctl_command_reply_error(conn, "Bad command syntax"); | |
6770 | goto exit; | |
6771 | } | |
6772 | ||
6773 | ofproto_trace(ofproto, &flow, packet, initial_tci, &result); | |
6774 | unixctl_command_reply(conn, ds_cstr(&result)); | |
6775 | ||
6776 | exit: | |
6777 | ds_destroy(&result); | |
6778 | ofpbuf_delete(packet); | |
6779 | ofpbuf_uninit(&odp_key); | |
6780 | } | |
6781 | ||
6782 | static void | |
6783 | ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow, | |
6784 | const struct ofpbuf *packet, ovs_be16 initial_tci, | |
6785 | struct ds *ds) | |
6786 | { | |
6787 | struct rule_dpif *rule; | |
6788 | ||
6789 | ds_put_cstr(ds, "Flow: "); | |
6790 | flow_format(ds, flow); | |
6791 | ds_put_char(ds, '\n'); | |
6792 | ||
6793 | rule = rule_dpif_lookup(ofproto, flow); | |
6794 | ||
6795 | trace_format_rule(ds, 0, 0, rule); | |
6796 | if (rule == ofproto->miss_rule) { | |
6797 | ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n"); | |
6798 | } else if (rule == ofproto->no_packet_in_rule) { | |
6799 | ds_put_cstr(ds, "\nNo match, packets dropped because " | |
6800 | "OFPPC_NO_PACKET_IN is set on in_port.\n"); | |
6801 | } | |
6802 | ||
6803 | if (rule) { | |
6804 | uint64_t odp_actions_stub[1024 / 8]; | |
6805 | struct ofpbuf odp_actions; | |
6806 | ||
6807 | struct trace_ctx trace; | |
6808 | uint8_t tcp_flags; | |
6809 | ||
6810 | tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0; | |
6811 | trace.result = ds; | |
6812 | trace.flow = *flow; | |
6813 | ofpbuf_use_stub(&odp_actions, | |
6814 | odp_actions_stub, sizeof odp_actions_stub); | |
6815 | action_xlate_ctx_init(&trace.ctx, ofproto, flow, initial_tci, | |
6816 | rule, tcp_flags, packet); | |
6817 | trace.ctx.resubmit_hook = trace_resubmit; | |
6818 | trace.ctx.report_hook = trace_report; | |
6819 | xlate_actions(&trace.ctx, rule->up.ofpacts, rule->up.ofpacts_len, | |
6820 | &odp_actions); | |
6821 | ||
6822 | ds_put_char(ds, '\n'); | |
6823 | trace_format_flow(ds, 0, "Final flow", &trace); | |
6824 | ds_put_cstr(ds, "Datapath actions: "); | |
6825 | format_odp_actions(ds, odp_actions.data, odp_actions.size); | |
6826 | ofpbuf_uninit(&odp_actions); | |
6827 | ||
6828 | if (trace.ctx.slow) { | |
6829 | enum slow_path_reason slow; | |
6830 | ||
6831 | ds_put_cstr(ds, "\nThis flow is handled by the userspace " | |
6832 | "slow path because it:"); | |
6833 | for (slow = trace.ctx.slow; slow; ) { | |
6834 | enum slow_path_reason bit = rightmost_1bit(slow); | |
6835 | ||
6836 | switch (bit) { | |
6837 | case SLOW_CFM: | |
6838 | ds_put_cstr(ds, "\n\t- Consists of CFM packets."); | |
6839 | break; | |
6840 | case SLOW_LACP: | |
6841 | ds_put_cstr(ds, "\n\t- Consists of LACP packets."); | |
6842 | break; | |
6843 | case SLOW_STP: | |
6844 | ds_put_cstr(ds, "\n\t- Consists of STP packets."); | |
6845 | break; | |
6846 | case SLOW_IN_BAND: | |
6847 | ds_put_cstr(ds, "\n\t- Needs in-band special case " | |
6848 | "processing."); | |
6849 | if (!packet) { | |
6850 | ds_put_cstr(ds, "\n\t (The datapath actions are " | |
6851 | "incomplete--for complete actions, " | |
6852 | "please supply a packet.)"); | |
6853 | } | |
6854 | break; | |
6855 | case SLOW_CONTROLLER: | |
6856 | ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages " | |
6857 | "to the OpenFlow controller."); | |
6858 | break; | |
6859 | case SLOW_MATCH: | |
6860 | ds_put_cstr(ds, "\n\t- Needs more specific matching " | |
6861 | "than the datapath supports."); | |
6862 | break; | |
6863 | } | |
6864 | ||
6865 | slow &= ~bit; | |
6866 | } | |
6867 | ||
6868 | if (slow & ~SLOW_MATCH) { | |
6869 | ds_put_cstr(ds, "\nThe datapath actions above do not reflect " | |
6870 | "the special slow-path processing."); | |
6871 | } | |
6872 | } | |
6873 | } | |
6874 | } | |
6875 | ||
6876 | static void | |
6877 | ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED, | |
6878 | const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED) | |
6879 | { | |
6880 | clogged = true; | |
6881 | unixctl_command_reply(conn, NULL); | |
6882 | } | |
6883 | ||
6884 | static void | |
6885 | ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED, | |
6886 | const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED) | |
6887 | { | |
6888 | clogged = false; | |
6889 | unixctl_command_reply(conn, NULL); | |
6890 | } | |
6891 | ||
6892 | /* Runs a self-check of flow translations in 'ofproto'. Appends a message to | |
6893 | * 'reply' describing the results. */ | |
6894 | static void | |
6895 | ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply) | |
6896 | { | |
6897 | struct facet *facet; | |
6898 | int errors; | |
6899 | ||
6900 | errors = 0; | |
6901 | HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) { | |
6902 | if (!facet_check_consistency(facet)) { | |
6903 | errors++; | |
6904 | } | |
6905 | } | |
6906 | if (errors) { | |
6907 | ofproto->need_revalidate = REV_INCONSISTENCY; | |
6908 | } | |
6909 | ||
6910 | if (errors) { | |
6911 | ds_put_format(reply, "%s: self-check failed (%d errors)\n", | |
6912 | ofproto->up.name, errors); | |
6913 | } else { | |
6914 | ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name); | |
6915 | } | |
6916 | } | |
6917 | ||
6918 | static void | |
6919 | ofproto_dpif_self_check(struct unixctl_conn *conn, | |
6920 | int argc, const char *argv[], void *aux OVS_UNUSED) | |
6921 | { | |
6922 | struct ds reply = DS_EMPTY_INITIALIZER; | |
6923 | struct ofproto_dpif *ofproto; | |
6924 | ||
6925 | if (argc > 1) { | |
6926 | ofproto = ofproto_dpif_lookup(argv[1]); | |
6927 | if (!ofproto) { | |
6928 | unixctl_command_reply_error(conn, "Unknown ofproto (use " | |
6929 | "ofproto/list for help)"); | |
6930 | return; | |
6931 | } | |
6932 | ofproto_dpif_self_check__(ofproto, &reply); | |
6933 | } else { | |
6934 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
6935 | ofproto_dpif_self_check__(ofproto, &reply); | |
6936 | } | |
6937 | } | |
6938 | ||
6939 | unixctl_command_reply(conn, ds_cstr(&reply)); | |
6940 | ds_destroy(&reply); | |
6941 | } | |
6942 | ||
6943 | static void | |
6944 | ofproto_dpif_unixctl_init(void) | |
6945 | { | |
6946 | static bool registered; | |
6947 | if (registered) { | |
6948 | return; | |
6949 | } | |
6950 | registered = true; | |
6951 | ||
6952 | unixctl_command_register( | |
6953 | "ofproto/trace", | |
6954 | "bridge {tun_id in_port packet | odp_flow [-generate]}", | |
6955 | 2, 5, ofproto_unixctl_trace, NULL); | |
6956 | unixctl_command_register("fdb/flush", "[bridge]", 0, 1, | |
6957 | ofproto_unixctl_fdb_flush, NULL); | |
6958 | unixctl_command_register("fdb/show", "bridge", 1, 1, | |
6959 | ofproto_unixctl_fdb_show, NULL); | |
6960 | unixctl_command_register("ofproto/clog", "", 0, 0, | |
6961 | ofproto_dpif_clog, NULL); | |
6962 | unixctl_command_register("ofproto/unclog", "", 0, 0, | |
6963 | ofproto_dpif_unclog, NULL); | |
6964 | unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1, | |
6965 | ofproto_dpif_self_check, NULL); | |
6966 | } | |
6967 | \f | |
6968 | /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.) | |
6969 | * | |
6970 | * This is deprecated. It is only for compatibility with broken device drivers | |
6971 | * in old versions of Linux that do not properly support VLANs when VLAN | |
6972 | * devices are not used. When broken device drivers are no longer in | |
6973 | * widespread use, we will delete these interfaces. */ | |
6974 | ||
6975 | static int | |
6976 | set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid) | |
6977 | { | |
6978 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto); | |
6979 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
6980 | ||
6981 | if (realdev_ofp_port == ofport->realdev_ofp_port | |
6982 | && vid == ofport->vlandev_vid) { | |
6983 | return 0; | |
6984 | } | |
6985 | ||
6986 | ofproto->need_revalidate = REV_RECONFIGURE; | |
6987 | ||
6988 | if (ofport->realdev_ofp_port) { | |
6989 | vsp_remove(ofport); | |
6990 | } | |
6991 | if (realdev_ofp_port && ofport->bundle) { | |
6992 | /* vlandevs are enslaved to their realdevs, so they are not allowed to | |
6993 | * themselves be part of a bundle. */ | |
6994 | bundle_set(ofport->up.ofproto, ofport->bundle, NULL); | |
6995 | } | |
6996 | ||
6997 | ofport->realdev_ofp_port = realdev_ofp_port; | |
6998 | ofport->vlandev_vid = vid; | |
6999 | ||
7000 | if (realdev_ofp_port) { | |
7001 | vsp_add(ofport, realdev_ofp_port, vid); | |
7002 | } | |
7003 | ||
7004 | return 0; | |
7005 | } | |
7006 | ||
7007 | static uint32_t | |
7008 | hash_realdev_vid(uint16_t realdev_ofp_port, int vid) | |
7009 | { | |
7010 | return hash_2words(realdev_ofp_port, vid); | |
7011 | } | |
7012 | ||
7013 | /* Returns the ODP port number of the Linux VLAN device that corresponds to | |
7014 | * 'vlan_tci' on the network device with port number 'realdev_odp_port' in | |
7015 | * 'ofproto'. For example, given 'realdev_odp_port' of eth0 and 'vlan_tci' 9, | |
7016 | * it would return the port number of eth0.9. | |
7017 | * | |
7018 | * Unless VLAN splinters are enabled for port 'realdev_odp_port', this | |
7019 | * function just returns its 'realdev_odp_port' argument. */ | |
7020 | static uint32_t | |
7021 | vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto, | |
7022 | uint32_t realdev_odp_port, ovs_be16 vlan_tci) | |
7023 | { | |
7024 | if (!hmap_is_empty(&ofproto->realdev_vid_map)) { | |
7025 | uint16_t realdev_ofp_port = odp_port_to_ofp_port(realdev_odp_port); | |
7026 | int vid = vlan_tci_to_vid(vlan_tci); | |
7027 | const struct vlan_splinter *vsp; | |
7028 | ||
7029 | HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node, | |
7030 | hash_realdev_vid(realdev_ofp_port, vid), | |
7031 | &ofproto->realdev_vid_map) { | |
7032 | if (vsp->realdev_ofp_port == realdev_ofp_port | |
7033 | && vsp->vid == vid) { | |
7034 | return ofp_port_to_odp_port(vsp->vlandev_ofp_port); | |
7035 | } | |
7036 | } | |
7037 | } | |
7038 | return realdev_odp_port; | |
7039 | } | |
7040 | ||
7041 | static struct vlan_splinter * | |
7042 | vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port) | |
7043 | { | |
7044 | struct vlan_splinter *vsp; | |
7045 | ||
7046 | HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, hash_int(vlandev_ofp_port, 0), | |
7047 | &ofproto->vlandev_map) { | |
7048 | if (vsp->vlandev_ofp_port == vlandev_ofp_port) { | |
7049 | return vsp; | |
7050 | } | |
7051 | } | |
7052 | ||
7053 | return NULL; | |
7054 | } | |
7055 | ||
7056 | /* Returns the OpenFlow port number of the "real" device underlying the Linux | |
7057 | * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the | |
7058 | * VLAN VID of the Linux VLAN device in '*vid'. For example, given | |
7059 | * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of | |
7060 | * eth0 and store 9 in '*vid'. | |
7061 | * | |
7062 | * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux | |
7063 | * VLAN device. Unless VLAN splinters are enabled, this is what this function | |
7064 | * always does.*/ | |
7065 | static uint16_t | |
7066 | vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto, | |
7067 | uint16_t vlandev_ofp_port, int *vid) | |
7068 | { | |
7069 | if (!hmap_is_empty(&ofproto->vlandev_map)) { | |
7070 | const struct vlan_splinter *vsp; | |
7071 | ||
7072 | vsp = vlandev_find(ofproto, vlandev_ofp_port); | |
7073 | if (vsp) { | |
7074 | if (vid) { | |
7075 | *vid = vsp->vid; | |
7076 | } | |
7077 | return vsp->realdev_ofp_port; | |
7078 | } | |
7079 | } | |
7080 | return 0; | |
7081 | } | |
7082 | ||
7083 | /* Given 'flow', a flow representing a packet received on 'ofproto', checks | |
7084 | * whether 'flow->in_port' represents a Linux VLAN device. If so, changes | |
7085 | * 'flow->in_port' to the "real" device backing the VLAN device, sets | |
7086 | * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is | |
7087 | * always the case unless VLAN splinters are enabled), returns false without | |
7088 | * making any changes. */ | |
7089 | static bool | |
7090 | vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow) | |
7091 | { | |
7092 | uint16_t realdev; | |
7093 | int vid; | |
7094 | ||
7095 | realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port, &vid); | |
7096 | if (!realdev) { | |
7097 | return false; | |
7098 | } | |
7099 | ||
7100 | /* Cause the flow to be processed as if it came in on the real device with | |
7101 | * the VLAN device's VLAN ID. */ | |
7102 | flow->in_port = realdev; | |
7103 | flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI); | |
7104 | return true; | |
7105 | } | |
7106 | ||
7107 | static void | |
7108 | vsp_remove(struct ofport_dpif *port) | |
7109 | { | |
7110 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto); | |
7111 | struct vlan_splinter *vsp; | |
7112 | ||
7113 | vsp = vlandev_find(ofproto, port->up.ofp_port); | |
7114 | if (vsp) { | |
7115 | hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node); | |
7116 | hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node); | |
7117 | free(vsp); | |
7118 | ||
7119 | port->realdev_ofp_port = 0; | |
7120 | } else { | |
7121 | VLOG_ERR("missing vlan device record"); | |
7122 | } | |
7123 | } | |
7124 | ||
7125 | static void | |
7126 | vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid) | |
7127 | { | |
7128 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto); | |
7129 | ||
7130 | if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL) | |
7131 | && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid)) | |
7132 | == realdev_ofp_port)) { | |
7133 | struct vlan_splinter *vsp; | |
7134 | ||
7135 | vsp = xmalloc(sizeof *vsp); | |
7136 | hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node, | |
7137 | hash_int(port->up.ofp_port, 0)); | |
7138 | hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node, | |
7139 | hash_realdev_vid(realdev_ofp_port, vid)); | |
7140 | vsp->realdev_ofp_port = realdev_ofp_port; | |
7141 | vsp->vlandev_ofp_port = port->up.ofp_port; | |
7142 | vsp->vid = vid; | |
7143 | ||
7144 | port->realdev_ofp_port = realdev_ofp_port; | |
7145 | } else { | |
7146 | VLOG_ERR("duplicate vlan device record"); | |
7147 | } | |
7148 | } | |
7149 | \f | |
7150 | const struct ofproto_class ofproto_dpif_class = { | |
7151 | enumerate_types, | |
7152 | enumerate_names, | |
7153 | del, | |
7154 | alloc, | |
7155 | construct, | |
7156 | destruct, | |
7157 | dealloc, | |
7158 | run, | |
7159 | run_fast, | |
7160 | wait, | |
7161 | get_memory_usage, | |
7162 | flush, | |
7163 | get_features, | |
7164 | get_tables, | |
7165 | port_alloc, | |
7166 | port_construct, | |
7167 | port_destruct, | |
7168 | port_dealloc, | |
7169 | port_modified, | |
7170 | port_reconfigured, | |
7171 | port_query_by_name, | |
7172 | port_add, | |
7173 | port_del, | |
7174 | port_get_stats, | |
7175 | port_dump_start, | |
7176 | port_dump_next, | |
7177 | port_dump_done, | |
7178 | port_poll, | |
7179 | port_poll_wait, | |
7180 | port_is_lacp_current, | |
7181 | NULL, /* rule_choose_table */ | |
7182 | rule_alloc, | |
7183 | rule_construct, | |
7184 | rule_destruct, | |
7185 | rule_dealloc, | |
7186 | rule_get_stats, | |
7187 | rule_execute, | |
7188 | rule_modify_actions, | |
7189 | set_frag_handling, | |
7190 | packet_out, | |
7191 | set_netflow, | |
7192 | get_netflow_ids, | |
7193 | set_sflow, | |
7194 | set_cfm, | |
7195 | get_cfm_fault, | |
7196 | get_cfm_opup, | |
7197 | get_cfm_remote_mpids, | |
7198 | get_cfm_health, | |
7199 | set_stp, | |
7200 | get_stp_status, | |
7201 | set_stp_port, | |
7202 | get_stp_port_status, | |
7203 | set_queues, | |
7204 | bundle_set, | |
7205 | bundle_remove, | |
7206 | mirror_set, | |
7207 | mirror_get_stats, | |
7208 | set_flood_vlans, | |
7209 | is_mirror_output_bundle, | |
7210 | forward_bpdu_changed, | |
7211 | set_mac_idle_time, | |
7212 | set_realdev, | |
7213 | }; |