]>
Commit | Line | Data |
---|---|---|
1 | /* | |
2 | * Copyright (c) 2009, 2010, 2011, 2012, 2013 Nicira, Inc. | |
3 | * | |
4 | * Licensed under the Apache License, Version 2.0 (the "License"); | |
5 | * you may not use this file except in compliance with the License. | |
6 | * You may obtain a copy of the License at: | |
7 | * | |
8 | * http://www.apache.org/licenses/LICENSE-2.0 | |
9 | * | |
10 | * Unless required by applicable law or agreed to in writing, software | |
11 | * distributed under the License is distributed on an "AS IS" BASIS, | |
12 | * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. | |
13 | * See the License for the specific language governing permissions and | |
14 | * limitations under the License. | |
15 | */ | |
16 | ||
17 | #include <config.h> | |
18 | ||
19 | #include "ofproto/ofproto-provider.h" | |
20 | ||
21 | #include <errno.h> | |
22 | ||
23 | #include "bfd.h" | |
24 | #include "bond.h" | |
25 | #include "bundle.h" | |
26 | #include "byte-order.h" | |
27 | #include "connmgr.h" | |
28 | #include "coverage.h" | |
29 | #include "cfm.h" | |
30 | #include "dpif.h" | |
31 | #include "dynamic-string.h" | |
32 | #include "fail-open.h" | |
33 | #include "hmapx.h" | |
34 | #include "lacp.h" | |
35 | #include "learn.h" | |
36 | #include "mac-learning.h" | |
37 | #include "meta-flow.h" | |
38 | #include "multipath.h" | |
39 | #include "netdev-vport.h" | |
40 | #include "netdev.h" | |
41 | #include "netlink.h" | |
42 | #include "nx-match.h" | |
43 | #include "odp-util.h" | |
44 | #include "ofp-util.h" | |
45 | #include "ofpbuf.h" | |
46 | #include "ofp-actions.h" | |
47 | #include "ofp-parse.h" | |
48 | #include "ofp-print.h" | |
49 | #include "ofproto-dpif-governor.h" | |
50 | #include "ofproto-dpif-ipfix.h" | |
51 | #include "ofproto-dpif-sflow.h" | |
52 | #include "poll-loop.h" | |
53 | #include "simap.h" | |
54 | #include "smap.h" | |
55 | #include "timer.h" | |
56 | #include "tunnel.h" | |
57 | #include "unaligned.h" | |
58 | #include "unixctl.h" | |
59 | #include "vlan-bitmap.h" | |
60 | #include "vlog.h" | |
61 | ||
62 | VLOG_DEFINE_THIS_MODULE(ofproto_dpif); | |
63 | ||
64 | COVERAGE_DEFINE(ofproto_dpif_expired); | |
65 | COVERAGE_DEFINE(ofproto_dpif_xlate); | |
66 | COVERAGE_DEFINE(facet_changed_rule); | |
67 | COVERAGE_DEFINE(facet_revalidate); | |
68 | COVERAGE_DEFINE(facet_unexpected); | |
69 | COVERAGE_DEFINE(facet_suppress); | |
70 | ||
71 | /* Maximum depth of flow table recursion (due to resubmit actions) in a | |
72 | * flow translation. */ | |
73 | #define MAX_RESUBMIT_RECURSION 64 | |
74 | ||
75 | /* Number of implemented OpenFlow tables. */ | |
76 | enum { N_TABLES = 255 }; | |
77 | enum { TBL_INTERNAL = N_TABLES - 1 }; /* Used for internal hidden rules. */ | |
78 | BUILD_ASSERT_DECL(N_TABLES >= 2 && N_TABLES <= 255); | |
79 | ||
80 | struct ofport_dpif; | |
81 | struct ofproto_dpif; | |
82 | struct flow_miss; | |
83 | struct facet; | |
84 | ||
85 | struct rule_dpif { | |
86 | struct rule up; | |
87 | ||
88 | /* These statistics: | |
89 | * | |
90 | * - Do include packets and bytes from facets that have been deleted or | |
91 | * whose own statistics have been folded into the rule. | |
92 | * | |
93 | * - Do include packets and bytes sent "by hand" that were accounted to | |
94 | * the rule without any facet being involved (this is a rare corner | |
95 | * case in rule_execute()). | |
96 | * | |
97 | * - Do not include packet or bytes that can be obtained from any facet's | |
98 | * packet_count or byte_count member or that can be obtained from the | |
99 | * datapath by, e.g., dpif_flow_get() for any subfacet. | |
100 | */ | |
101 | uint64_t packet_count; /* Number of packets received. */ | |
102 | uint64_t byte_count; /* Number of bytes received. */ | |
103 | ||
104 | tag_type tag; /* Caches rule_calculate_tag() result. */ | |
105 | ||
106 | struct list facets; /* List of "struct facet"s. */ | |
107 | }; | |
108 | ||
109 | static struct rule_dpif *rule_dpif_cast(const struct rule *rule) | |
110 | { | |
111 | return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL; | |
112 | } | |
113 | ||
114 | static struct rule_dpif *rule_dpif_lookup(struct ofproto_dpif *, | |
115 | const struct flow *); | |
116 | static struct rule_dpif *rule_dpif_lookup__(struct ofproto_dpif *, | |
117 | const struct flow *, | |
118 | uint8_t table); | |
119 | static struct rule_dpif *rule_dpif_miss_rule(struct ofproto_dpif *ofproto, | |
120 | const struct flow *flow); | |
121 | ||
122 | static void rule_credit_stats(struct rule_dpif *, | |
123 | const struct dpif_flow_stats *); | |
124 | static void flow_push_stats(struct facet *, const struct dpif_flow_stats *); | |
125 | static tag_type rule_calculate_tag(const struct flow *, | |
126 | const struct minimask *, uint32_t basis); | |
127 | static void rule_invalidate(const struct rule_dpif *); | |
128 | ||
129 | #define MAX_MIRRORS 32 | |
130 | typedef uint32_t mirror_mask_t; | |
131 | #define MIRROR_MASK_C(X) UINT32_C(X) | |
132 | BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS); | |
133 | struct ofmirror { | |
134 | struct ofproto_dpif *ofproto; /* Owning ofproto. */ | |
135 | size_t idx; /* In ofproto's "mirrors" array. */ | |
136 | void *aux; /* Key supplied by ofproto's client. */ | |
137 | char *name; /* Identifier for log messages. */ | |
138 | ||
139 | /* Selection criteria. */ | |
140 | struct hmapx srcs; /* Contains "struct ofbundle *"s. */ | |
141 | struct hmapx dsts; /* Contains "struct ofbundle *"s. */ | |
142 | unsigned long *vlans; /* Bitmap of chosen VLANs, NULL selects all. */ | |
143 | ||
144 | /* Output (exactly one of out == NULL and out_vlan == -1 is true). */ | |
145 | struct ofbundle *out; /* Output port or NULL. */ | |
146 | int out_vlan; /* Output VLAN or -1. */ | |
147 | mirror_mask_t dup_mirrors; /* Bitmap of mirrors with the same output. */ | |
148 | ||
149 | /* Counters. */ | |
150 | int64_t packet_count; /* Number of packets sent. */ | |
151 | int64_t byte_count; /* Number of bytes sent. */ | |
152 | }; | |
153 | ||
154 | static void mirror_destroy(struct ofmirror *); | |
155 | static void update_mirror_stats(struct ofproto_dpif *ofproto, | |
156 | mirror_mask_t mirrors, | |
157 | uint64_t packets, uint64_t bytes); | |
158 | ||
159 | struct ofbundle { | |
160 | struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */ | |
161 | struct ofproto_dpif *ofproto; /* Owning ofproto. */ | |
162 | void *aux; /* Key supplied by ofproto's client. */ | |
163 | char *name; /* Identifier for log messages. */ | |
164 | ||
165 | /* Configuration. */ | |
166 | struct list ports; /* Contains "struct ofport"s. */ | |
167 | enum port_vlan_mode vlan_mode; /* VLAN mode */ | |
168 | int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */ | |
169 | unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1. | |
170 | * NULL if all VLANs are trunked. */ | |
171 | struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */ | |
172 | struct bond *bond; /* Nonnull iff more than one port. */ | |
173 | bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */ | |
174 | ||
175 | /* Status. */ | |
176 | bool floodable; /* True if no port has OFPUTIL_PC_NO_FLOOD set. */ | |
177 | ||
178 | /* Port mirroring info. */ | |
179 | mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */ | |
180 | mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */ | |
181 | mirror_mask_t mirror_out; /* Mirrors that output to this bundle. */ | |
182 | }; | |
183 | ||
184 | static void bundle_remove(struct ofport *); | |
185 | static void bundle_update(struct ofbundle *); | |
186 | static void bundle_destroy(struct ofbundle *); | |
187 | static void bundle_del_port(struct ofport_dpif *); | |
188 | static void bundle_run(struct ofbundle *); | |
189 | static void bundle_wait(struct ofbundle *); | |
190 | static struct ofbundle *lookup_input_bundle(const struct ofproto_dpif *, | |
191 | uint16_t in_port, bool warn, | |
192 | struct ofport_dpif **in_ofportp); | |
193 | ||
194 | /* A controller may use OFPP_NONE as the ingress port to indicate that | |
195 | * it did not arrive on a "real" port. 'ofpp_none_bundle' exists for | |
196 | * when an input bundle is needed for validation (e.g., mirroring or | |
197 | * OFPP_NORMAL processing). It is not connected to an 'ofproto' or have | |
198 | * any 'port' structs, so care must be taken when dealing with it. */ | |
199 | static struct ofbundle ofpp_none_bundle = { | |
200 | .name = "OFPP_NONE", | |
201 | .vlan_mode = PORT_VLAN_TRUNK | |
202 | }; | |
203 | ||
204 | static void stp_run(struct ofproto_dpif *ofproto); | |
205 | static void stp_wait(struct ofproto_dpif *ofproto); | |
206 | static int set_stp_port(struct ofport *, | |
207 | const struct ofproto_port_stp_settings *); | |
208 | ||
209 | static bool ofbundle_includes_vlan(const struct ofbundle *, uint16_t vlan); | |
210 | ||
211 | struct action_xlate_ctx { | |
212 | /* action_xlate_ctx_init() initializes these members. */ | |
213 | ||
214 | /* The ofproto. */ | |
215 | struct ofproto_dpif *ofproto; | |
216 | ||
217 | /* Flow to which the OpenFlow actions apply. xlate_actions() will modify | |
218 | * this flow when actions change header fields. */ | |
219 | struct flow flow; | |
220 | ||
221 | /* stack for the push and pop actions. | |
222 | * Each stack element is of the type "union mf_subvalue". */ | |
223 | struct ofpbuf stack; | |
224 | union mf_subvalue init_stack[1024 / sizeof(union mf_subvalue)]; | |
225 | ||
226 | /* The packet corresponding to 'flow', or a null pointer if we are | |
227 | * revalidating without a packet to refer to. */ | |
228 | const struct ofpbuf *packet; | |
229 | ||
230 | /* Should OFPP_NORMAL update the MAC learning table? Should "learn" | |
231 | * actions update the flow table? | |
232 | * | |
233 | * We want to update these tables if we are actually processing a packet, | |
234 | * or if we are accounting for packets that the datapath has processed, but | |
235 | * not if we are just revalidating. */ | |
236 | bool may_learn; | |
237 | ||
238 | /* The rule that we are currently translating, or NULL. */ | |
239 | struct rule_dpif *rule; | |
240 | ||
241 | /* Union of the set of TCP flags seen so far in this flow. (Used only by | |
242 | * NXAST_FIN_TIMEOUT. Set to zero to avoid updating updating rules' | |
243 | * timeouts.) */ | |
244 | uint8_t tcp_flags; | |
245 | ||
246 | /* If nonnull, flow translation calls this function just before executing a | |
247 | * resubmit or OFPP_TABLE action. In addition, disables logging of traces | |
248 | * when the recursion depth is exceeded. | |
249 | * | |
250 | * 'rule' is the rule being submitted into. It will be null if the | |
251 | * resubmit or OFPP_TABLE action didn't find a matching rule. | |
252 | * | |
253 | * This is normally null so the client has to set it manually after | |
254 | * calling action_xlate_ctx_init(). */ | |
255 | void (*resubmit_hook)(struct action_xlate_ctx *, struct rule_dpif *rule); | |
256 | ||
257 | /* If nonnull, flow translation calls this function to report some | |
258 | * significant decision, e.g. to explain why OFPP_NORMAL translation | |
259 | * dropped a packet. */ | |
260 | void (*report_hook)(struct action_xlate_ctx *, const char *s); | |
261 | ||
262 | /* If nonnull, flow translation credits the specified statistics to each | |
263 | * rule reached through a resubmit or OFPP_TABLE action. | |
264 | * | |
265 | * This is normally null so the client has to set it manually after | |
266 | * calling action_xlate_ctx_init(). */ | |
267 | const struct dpif_flow_stats *resubmit_stats; | |
268 | ||
269 | /* xlate_actions() initializes and uses these members. The client might want | |
270 | * to look at them after it returns. */ | |
271 | ||
272 | struct ofpbuf *odp_actions; /* Datapath actions. */ | |
273 | tag_type tags; /* Tags associated with actions. */ | |
274 | enum slow_path_reason slow; /* 0 if fast path may be used. */ | |
275 | bool has_learn; /* Actions include NXAST_LEARN? */ | |
276 | bool has_normal; /* Actions output to OFPP_NORMAL? */ | |
277 | bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */ | |
278 | uint16_t nf_output_iface; /* Output interface index for NetFlow. */ | |
279 | mirror_mask_t mirrors; /* Bitmap of associated mirrors. */ | |
280 | ||
281 | /* xlate_actions() initializes and uses these members, but the client has no | |
282 | * reason to look at them. */ | |
283 | ||
284 | int recurse; /* Recursion level, via xlate_table_action. */ | |
285 | bool max_resubmit_trigger; /* Recursed too deeply during translation. */ | |
286 | struct flow base_flow; /* Flow at the last commit. */ | |
287 | uint32_t orig_skb_priority; /* Priority when packet arrived. */ | |
288 | uint8_t table_id; /* OpenFlow table ID where flow was found. */ | |
289 | uint32_t sflow_n_outputs; /* Number of output ports. */ | |
290 | uint32_t sflow_odp_port; /* Output port for composing sFlow action. */ | |
291 | uint16_t user_cookie_offset;/* Used for user_action_cookie fixup. */ | |
292 | bool exit; /* No further actions should be processed. */ | |
293 | }; | |
294 | ||
295 | /* Initial values of fields of the packet that may be changed during | |
296 | * flow processing and needed later. */ | |
297 | struct initial_vals { | |
298 | /* This is the value of vlan_tci in the packet as actually received from | |
299 | * dpif. This is the same as the facet's flow.vlan_tci unless the packet | |
300 | * was received via a VLAN splinter. In that case, this value is 0 | |
301 | * (because the packet as actually received from the dpif had no 802.1Q | |
302 | * tag) but the facet's flow.vlan_tci is set to the VLAN that the splinter | |
303 | * represents. | |
304 | * | |
305 | * This member should be removed when the VLAN splinters feature is no | |
306 | * longer needed. */ | |
307 | ovs_be16 vlan_tci; | |
308 | }; | |
309 | ||
310 | static void action_xlate_ctx_init(struct action_xlate_ctx *, | |
311 | struct ofproto_dpif *, const struct flow *, | |
312 | const struct initial_vals *initial_vals, | |
313 | struct rule_dpif *, | |
314 | uint8_t tcp_flags, const struct ofpbuf *); | |
315 | static void xlate_actions(struct action_xlate_ctx *, | |
316 | const struct ofpact *ofpacts, size_t ofpacts_len, | |
317 | struct ofpbuf *odp_actions); | |
318 | static void xlate_actions_for_side_effects(struct action_xlate_ctx *, | |
319 | const struct ofpact *ofpacts, | |
320 | size_t ofpacts_len); | |
321 | static void xlate_table_action(struct action_xlate_ctx *, uint16_t in_port, | |
322 | uint8_t table_id, bool may_packet_in); | |
323 | ||
324 | static size_t put_userspace_action(const struct ofproto_dpif *, | |
325 | struct ofpbuf *odp_actions, | |
326 | const struct flow *, | |
327 | const union user_action_cookie *, | |
328 | const size_t); | |
329 | ||
330 | static void compose_slow_path(const struct ofproto_dpif *, const struct flow *, | |
331 | enum slow_path_reason, | |
332 | uint64_t *stub, size_t stub_size, | |
333 | const struct nlattr **actionsp, | |
334 | size_t *actions_lenp); | |
335 | ||
336 | static void xlate_report(struct action_xlate_ctx *ctx, const char *s); | |
337 | ||
338 | /* A subfacet (see "struct subfacet" below) has three possible installation | |
339 | * states: | |
340 | * | |
341 | * - SF_NOT_INSTALLED: Not installed in the datapath. This will only be the | |
342 | * case just after the subfacet is created, just before the subfacet is | |
343 | * destroyed, or if the datapath returns an error when we try to install a | |
344 | * subfacet. | |
345 | * | |
346 | * - SF_FAST_PATH: The subfacet's actions are installed in the datapath. | |
347 | * | |
348 | * - SF_SLOW_PATH: An action that sends every packet for the subfacet through | |
349 | * ofproto_dpif is installed in the datapath. | |
350 | */ | |
351 | enum subfacet_path { | |
352 | SF_NOT_INSTALLED, /* No datapath flow for this subfacet. */ | |
353 | SF_FAST_PATH, /* Full actions are installed. */ | |
354 | SF_SLOW_PATH, /* Send-to-userspace action is installed. */ | |
355 | }; | |
356 | ||
357 | static const char *subfacet_path_to_string(enum subfacet_path); | |
358 | ||
359 | /* A dpif flow and actions associated with a facet. | |
360 | * | |
361 | * See also the large comment on struct facet. */ | |
362 | struct subfacet { | |
363 | /* Owners. */ | |
364 | struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */ | |
365 | struct list list_node; /* In struct facet's 'facets' list. */ | |
366 | struct facet *facet; /* Owning facet. */ | |
367 | ||
368 | enum odp_key_fitness key_fitness; | |
369 | struct nlattr *key; | |
370 | int key_len; | |
371 | ||
372 | long long int used; /* Time last used; time created if not used. */ | |
373 | long long int created; /* Time created. */ | |
374 | ||
375 | uint64_t dp_packet_count; /* Last known packet count in the datapath. */ | |
376 | uint64_t dp_byte_count; /* Last known byte count in the datapath. */ | |
377 | ||
378 | /* Datapath actions. | |
379 | * | |
380 | * These should be essentially identical for every subfacet in a facet, but | |
381 | * may differ in trivial ways due to VLAN splinters. */ | |
382 | size_t actions_len; /* Number of bytes in actions[]. */ | |
383 | struct nlattr *actions; /* Datapath actions. */ | |
384 | ||
385 | enum slow_path_reason slow; /* 0 if fast path may be used. */ | |
386 | enum subfacet_path path; /* Installed in datapath? */ | |
387 | ||
388 | /* Initial values of the packet that may be needed later. */ | |
389 | struct initial_vals initial_vals; | |
390 | ||
391 | /* Datapath port the packet arrived on. This is needed to remove | |
392 | * flows for ports that are no longer part of the bridge. Since the | |
393 | * flow definition only has the OpenFlow port number and the port is | |
394 | * no longer part of the bridge, we can't determine the datapath port | |
395 | * number needed to delete the flow from the datapath. */ | |
396 | uint32_t odp_in_port; | |
397 | }; | |
398 | ||
399 | #define SUBFACET_DESTROY_MAX_BATCH 50 | |
400 | ||
401 | static struct subfacet *subfacet_create(struct facet *, struct flow_miss *miss, | |
402 | long long int now); | |
403 | static struct subfacet *subfacet_find(struct ofproto_dpif *, | |
404 | const struct nlattr *key, size_t key_len, | |
405 | uint32_t key_hash); | |
406 | static void subfacet_destroy(struct subfacet *); | |
407 | static void subfacet_destroy__(struct subfacet *); | |
408 | static void subfacet_destroy_batch(struct ofproto_dpif *, | |
409 | struct subfacet **, int n); | |
410 | static void subfacet_reset_dp_stats(struct subfacet *, | |
411 | struct dpif_flow_stats *); | |
412 | static void subfacet_update_time(struct subfacet *, long long int used); | |
413 | static void subfacet_update_stats(struct subfacet *, | |
414 | const struct dpif_flow_stats *); | |
415 | static void subfacet_make_actions(struct subfacet *, | |
416 | const struct ofpbuf *packet, | |
417 | struct ofpbuf *odp_actions); | |
418 | static int subfacet_install(struct subfacet *, | |
419 | const struct nlattr *actions, size_t actions_len, | |
420 | struct dpif_flow_stats *, enum slow_path_reason); | |
421 | static void subfacet_uninstall(struct subfacet *); | |
422 | ||
423 | static enum subfacet_path subfacet_want_path(enum slow_path_reason); | |
424 | ||
425 | /* An exact-match instantiation of an OpenFlow flow. | |
426 | * | |
427 | * A facet associates a "struct flow", which represents the Open vSwitch | |
428 | * userspace idea of an exact-match flow, with one or more subfacets. Each | |
429 | * subfacet tracks the datapath's idea of the exact-match flow equivalent to | |
430 | * the facet. When the kernel module (or other dpif implementation) and Open | |
431 | * vSwitch userspace agree on the definition of a flow key, there is exactly | |
432 | * one subfacet per facet. If the dpif implementation supports more-specific | |
433 | * flow matching than userspace, however, a facet can have more than one | |
434 | * subfacet, each of which corresponds to some distinction in flow that | |
435 | * userspace simply doesn't understand. | |
436 | * | |
437 | * Flow expiration works in terms of subfacets, so a facet must have at least | |
438 | * one subfacet or it will never expire, leaking memory. */ | |
439 | struct facet { | |
440 | /* Owners. */ | |
441 | struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */ | |
442 | struct list list_node; /* In owning rule's 'facets' list. */ | |
443 | struct rule_dpif *rule; /* Owning rule. */ | |
444 | ||
445 | /* Owned data. */ | |
446 | struct list subfacets; | |
447 | long long int used; /* Time last used; time created if not used. */ | |
448 | ||
449 | /* Key. */ | |
450 | struct flow flow; | |
451 | ||
452 | /* These statistics: | |
453 | * | |
454 | * - Do include packets and bytes sent "by hand", e.g. with | |
455 | * dpif_execute(). | |
456 | * | |
457 | * - Do include packets and bytes that were obtained from the datapath | |
458 | * when a subfacet's statistics were reset (e.g. dpif_flow_put() with | |
459 | * DPIF_FP_ZERO_STATS). | |
460 | * | |
461 | * - Do not include packets or bytes that can be obtained from the | |
462 | * datapath for any existing subfacet. | |
463 | */ | |
464 | uint64_t packet_count; /* Number of packets received. */ | |
465 | uint64_t byte_count; /* Number of bytes received. */ | |
466 | ||
467 | /* Resubmit statistics. */ | |
468 | uint64_t prev_packet_count; /* Number of packets from last stats push. */ | |
469 | uint64_t prev_byte_count; /* Number of bytes from last stats push. */ | |
470 | long long int prev_used; /* Used time from last stats push. */ | |
471 | ||
472 | /* Accounting. */ | |
473 | uint64_t accounted_bytes; /* Bytes processed by facet_account(). */ | |
474 | struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */ | |
475 | uint8_t tcp_flags; /* TCP flags seen for this 'rule'. */ | |
476 | ||
477 | /* Properties of datapath actions. | |
478 | * | |
479 | * Every subfacet has its own actions because actions can differ slightly | |
480 | * between splintered and non-splintered subfacets due to the VLAN tag | |
481 | * being initially different (present vs. absent). All of them have these | |
482 | * properties in common so we just store one copy of them here. */ | |
483 | bool has_learn; /* Actions include NXAST_LEARN? */ | |
484 | bool has_normal; /* Actions output to OFPP_NORMAL? */ | |
485 | bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */ | |
486 | tag_type tags; /* Tags that would require revalidation. */ | |
487 | mirror_mask_t mirrors; /* Bitmap of dependent mirrors. */ | |
488 | ||
489 | /* Storage for a single subfacet, to reduce malloc() time and space | |
490 | * overhead. (A facet always has at least one subfacet and in the common | |
491 | * case has exactly one subfacet. However, 'one_subfacet' may not | |
492 | * always be valid, since it could have been removed after newer | |
493 | * subfacets were pushed onto the 'subfacets' list.) */ | |
494 | struct subfacet one_subfacet; | |
495 | ||
496 | long long int learn_rl; /* Rate limiter for facet_learn(). */ | |
497 | }; | |
498 | ||
499 | static struct facet *facet_create(struct rule_dpif *, | |
500 | const struct flow *, uint32_t hash); | |
501 | static void facet_remove(struct facet *); | |
502 | static void facet_free(struct facet *); | |
503 | ||
504 | static struct facet *facet_find(struct ofproto_dpif *, | |
505 | const struct flow *, uint32_t hash); | |
506 | static struct facet *facet_lookup_valid(struct ofproto_dpif *, | |
507 | const struct flow *, uint32_t hash); | |
508 | static void facet_revalidate(struct facet *); | |
509 | static bool facet_check_consistency(struct facet *); | |
510 | ||
511 | static void facet_flush_stats(struct facet *); | |
512 | ||
513 | static void facet_update_time(struct facet *, long long int used); | |
514 | static void facet_reset_counters(struct facet *); | |
515 | static void facet_push_stats(struct facet *); | |
516 | static void facet_learn(struct facet *); | |
517 | static void facet_account(struct facet *); | |
518 | static void push_all_stats(void); | |
519 | ||
520 | static struct subfacet *facet_get_subfacet(struct facet *); | |
521 | ||
522 | static bool facet_is_controller_flow(struct facet *); | |
523 | ||
524 | struct ofport_dpif { | |
525 | struct hmap_node odp_port_node; /* In dpif_backer's "odp_to_ofport_map". */ | |
526 | struct ofport up; | |
527 | ||
528 | uint32_t odp_port; | |
529 | struct ofbundle *bundle; /* Bundle that contains this port, if any. */ | |
530 | struct list bundle_node; /* In struct ofbundle's "ports" list. */ | |
531 | struct cfm *cfm; /* Connectivity Fault Management, if any. */ | |
532 | struct bfd *bfd; /* BFD, if any. */ | |
533 | tag_type tag; /* Tag associated with this port. */ | |
534 | bool may_enable; /* May be enabled in bonds. */ | |
535 | long long int carrier_seq; /* Carrier status changes. */ | |
536 | struct tnl_port *tnl_port; /* Tunnel handle, or null. */ | |
537 | ||
538 | /* Spanning tree. */ | |
539 | struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */ | |
540 | enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */ | |
541 | long long int stp_state_entered; | |
542 | ||
543 | struct hmap priorities; /* Map of attached 'priority_to_dscp's. */ | |
544 | ||
545 | /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.) | |
546 | * | |
547 | * This is deprecated. It is only for compatibility with broken device | |
548 | * drivers in old versions of Linux that do not properly support VLANs when | |
549 | * VLAN devices are not used. When broken device drivers are no longer in | |
550 | * widespread use, we will delete these interfaces. */ | |
551 | uint16_t realdev_ofp_port; | |
552 | int vlandev_vid; | |
553 | }; | |
554 | ||
555 | /* Node in 'ofport_dpif''s 'priorities' map. Used to maintain a map from | |
556 | * 'priority' (the datapath's term for QoS queue) to the dscp bits which all | |
557 | * traffic egressing the 'ofport' with that priority should be marked with. */ | |
558 | struct priority_to_dscp { | |
559 | struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'priorities' map. */ | |
560 | uint32_t priority; /* Priority of this queue (see struct flow). */ | |
561 | ||
562 | uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */ | |
563 | }; | |
564 | ||
565 | /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.) | |
566 | * | |
567 | * This is deprecated. It is only for compatibility with broken device drivers | |
568 | * in old versions of Linux that do not properly support VLANs when VLAN | |
569 | * devices are not used. When broken device drivers are no longer in | |
570 | * widespread use, we will delete these interfaces. */ | |
571 | struct vlan_splinter { | |
572 | struct hmap_node realdev_vid_node; | |
573 | struct hmap_node vlandev_node; | |
574 | uint16_t realdev_ofp_port; | |
575 | uint16_t vlandev_ofp_port; | |
576 | int vid; | |
577 | }; | |
578 | ||
579 | static uint32_t vsp_realdev_to_vlandev(const struct ofproto_dpif *, | |
580 | uint32_t realdev, ovs_be16 vlan_tci); | |
581 | static bool vsp_adjust_flow(const struct ofproto_dpif *, struct flow *); | |
582 | static void vsp_remove(struct ofport_dpif *); | |
583 | static void vsp_add(struct ofport_dpif *, uint16_t realdev_ofp_port, int vid); | |
584 | ||
585 | static uint32_t ofp_port_to_odp_port(const struct ofproto_dpif *, | |
586 | uint16_t ofp_port); | |
587 | static uint16_t odp_port_to_ofp_port(const struct ofproto_dpif *, | |
588 | uint32_t odp_port); | |
589 | ||
590 | static struct ofport_dpif * | |
591 | ofport_dpif_cast(const struct ofport *ofport) | |
592 | { | |
593 | return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL; | |
594 | } | |
595 | ||
596 | static void port_run(struct ofport_dpif *); | |
597 | static void port_run_fast(struct ofport_dpif *); | |
598 | static void port_wait(struct ofport_dpif *); | |
599 | static int set_cfm(struct ofport *, const struct cfm_settings *); | |
600 | static void ofport_clear_priorities(struct ofport_dpif *); | |
601 | static void run_fast_rl(void); | |
602 | ||
603 | struct dpif_completion { | |
604 | struct list list_node; | |
605 | struct ofoperation *op; | |
606 | }; | |
607 | ||
608 | /* Extra information about a classifier table. | |
609 | * Currently used just for optimized flow revalidation. */ | |
610 | struct table_dpif { | |
611 | /* If either of these is nonnull, then this table has a form that allows | |
612 | * flows to be tagged to avoid revalidating most flows for the most common | |
613 | * kinds of flow table changes. */ | |
614 | struct cls_table *catchall_table; /* Table that wildcards all fields. */ | |
615 | struct cls_table *other_table; /* Table with any other wildcard set. */ | |
616 | uint32_t basis; /* Keeps each table's tags separate. */ | |
617 | }; | |
618 | ||
619 | /* Reasons that we might need to revalidate every facet, and corresponding | |
620 | * coverage counters. | |
621 | * | |
622 | * A value of 0 means that there is no need to revalidate. | |
623 | * | |
624 | * It would be nice to have some cleaner way to integrate with coverage | |
625 | * counters, but with only a few reasons I guess this is good enough for | |
626 | * now. */ | |
627 | enum revalidate_reason { | |
628 | REV_RECONFIGURE = 1, /* Switch configuration changed. */ | |
629 | REV_STP, /* Spanning tree protocol port status change. */ | |
630 | REV_PORT_TOGGLED, /* Port enabled or disabled by CFM, LACP, ...*/ | |
631 | REV_FLOW_TABLE, /* Flow table changed. */ | |
632 | REV_INCONSISTENCY /* Facet self-check failed. */ | |
633 | }; | |
634 | COVERAGE_DEFINE(rev_reconfigure); | |
635 | COVERAGE_DEFINE(rev_stp); | |
636 | COVERAGE_DEFINE(rev_port_toggled); | |
637 | COVERAGE_DEFINE(rev_flow_table); | |
638 | COVERAGE_DEFINE(rev_inconsistency); | |
639 | ||
640 | /* Drop keys are odp flow keys which have drop flows installed in the kernel. | |
641 | * These are datapath flows which have no associated ofproto, if they did we | |
642 | * would use facets. */ | |
643 | struct drop_key { | |
644 | struct hmap_node hmap_node; | |
645 | struct nlattr *key; | |
646 | size_t key_len; | |
647 | }; | |
648 | ||
649 | /* All datapaths of a given type share a single dpif backer instance. */ | |
650 | struct dpif_backer { | |
651 | char *type; | |
652 | int refcount; | |
653 | struct dpif *dpif; | |
654 | struct timer next_expiration; | |
655 | struct hmap odp_to_ofport_map; /* ODP port to ofport mapping. */ | |
656 | ||
657 | struct simap tnl_backers; /* Set of dpif ports backing tunnels. */ | |
658 | ||
659 | /* Facet revalidation flags applying to facets which use this backer. */ | |
660 | enum revalidate_reason need_revalidate; /* Revalidate every facet. */ | |
661 | struct tag_set revalidate_set; /* Revalidate only matching facets. */ | |
662 | ||
663 | struct hmap drop_keys; /* Set of dropped odp keys. */ | |
664 | }; | |
665 | ||
666 | /* All existing ofproto_backer instances, indexed by ofproto->up.type. */ | |
667 | static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers); | |
668 | ||
669 | static void drop_key_clear(struct dpif_backer *); | |
670 | static struct ofport_dpif * | |
671 | odp_port_to_ofport(const struct dpif_backer *, uint32_t odp_port); | |
672 | ||
673 | static void dpif_stats_update_hit_count(struct ofproto_dpif *ofproto, | |
674 | uint64_t delta); | |
675 | struct avg_subfacet_rates { | |
676 | double add_rate; /* Moving average of new flows created per minute. */ | |
677 | double del_rate; /* Moving average of flows deleted per minute. */ | |
678 | }; | |
679 | static void show_dp_rates(struct ds *ds, const char *heading, | |
680 | const struct avg_subfacet_rates *rates); | |
681 | static void exp_mavg(double *avg, int base, double new); | |
682 | ||
683 | struct ofproto_dpif { | |
684 | struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */ | |
685 | struct ofproto up; | |
686 | struct dpif_backer *backer; | |
687 | ||
688 | /* Special OpenFlow rules. */ | |
689 | struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */ | |
690 | struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */ | |
691 | ||
692 | /* Statistics. */ | |
693 | uint64_t n_matches; | |
694 | ||
695 | /* Bridging. */ | |
696 | struct netflow *netflow; | |
697 | struct dpif_sflow *sflow; | |
698 | struct dpif_ipfix *ipfix; | |
699 | struct hmap bundles; /* Contains "struct ofbundle"s. */ | |
700 | struct mac_learning *ml; | |
701 | struct ofmirror *mirrors[MAX_MIRRORS]; | |
702 | bool has_mirrors; | |
703 | bool has_bonded_bundles; | |
704 | ||
705 | /* Facets. */ | |
706 | struct hmap facets; | |
707 | struct hmap subfacets; | |
708 | struct governor *governor; | |
709 | long long int consistency_rl; | |
710 | ||
711 | /* Revalidation. */ | |
712 | struct table_dpif tables[N_TABLES]; | |
713 | ||
714 | /* Support for debugging async flow mods. */ | |
715 | struct list completions; | |
716 | ||
717 | bool has_bundle_action; /* True when the first bundle action appears. */ | |
718 | struct netdev_stats stats; /* To account packets generated and consumed in | |
719 | * userspace. */ | |
720 | ||
721 | /* Spanning tree. */ | |
722 | struct stp *stp; | |
723 | long long int stp_last_tick; | |
724 | ||
725 | /* VLAN splinters. */ | |
726 | struct hmap realdev_vid_map; /* (realdev,vid) -> vlandev. */ | |
727 | struct hmap vlandev_map; /* vlandev -> (realdev,vid). */ | |
728 | ||
729 | /* Ports. */ | |
730 | struct sset ports; /* Set of standard port names. */ | |
731 | struct sset ghost_ports; /* Ports with no datapath port. */ | |
732 | struct sset port_poll_set; /* Queued names for port_poll() reply. */ | |
733 | int port_poll_errno; /* Last errno for port_poll() reply. */ | |
734 | ||
735 | /* Per ofproto's dpif stats. */ | |
736 | uint64_t n_hit; | |
737 | uint64_t n_missed; | |
738 | ||
739 | /* Subfacet statistics. | |
740 | * | |
741 | * These keep track of the total number of subfacets added and deleted and | |
742 | * flow life span. They are useful for computing the flow rates stats | |
743 | * exposed via "ovs-appctl dpif/show". The goal is to learn about | |
744 | * traffic patterns in ways that we can use later to improve Open vSwitch | |
745 | * performance in new situations. */ | |
746 | long long int created; /* Time when it is created. */ | |
747 | unsigned int max_n_subfacet; /* Maximum number of flows */ | |
748 | ||
749 | /* The average number of subfacets... */ | |
750 | struct avg_subfacet_rates hourly; /* ...over the last hour. */ | |
751 | struct avg_subfacet_rates daily; /* ...over the last day. */ | |
752 | long long int last_minute; /* Last time 'hourly' was updated. */ | |
753 | ||
754 | /* Number of subfacets added or deleted since 'last_minute'. */ | |
755 | unsigned int subfacet_add_count; | |
756 | unsigned int subfacet_del_count; | |
757 | ||
758 | /* Number of subfacets added or deleted from 'created' to 'last_minute.' */ | |
759 | unsigned long long int total_subfacet_add_count; | |
760 | unsigned long long int total_subfacet_del_count; | |
761 | ||
762 | /* Sum of the number of milliseconds that each subfacet existed, | |
763 | * over the subfacets that have been added and then later deleted. */ | |
764 | unsigned long long int total_subfacet_life_span; | |
765 | ||
766 | /* Incremented by the number of currently existing subfacets, each | |
767 | * time we pull statistics from the kernel. */ | |
768 | unsigned long long int total_subfacet_count; | |
769 | ||
770 | /* Number of times we pull statistics from the kernel. */ | |
771 | unsigned long long int n_update_stats; | |
772 | }; | |
773 | static unsigned long long int avg_subfacet_life_span( | |
774 | const struct ofproto_dpif *); | |
775 | static double avg_subfacet_count(const struct ofproto_dpif *ofproto); | |
776 | static void update_moving_averages(struct ofproto_dpif *ofproto); | |
777 | static void dpif_stats_update_hit_count(struct ofproto_dpif *ofproto, | |
778 | uint64_t delta); | |
779 | static void update_max_subfacet_count(struct ofproto_dpif *ofproto); | |
780 | ||
781 | /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only | |
782 | * for debugging the asynchronous flow_mod implementation.) */ | |
783 | static bool clogged; | |
784 | ||
785 | /* All existing ofproto_dpif instances, indexed by ->up.name. */ | |
786 | static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs); | |
787 | ||
788 | static void ofproto_dpif_unixctl_init(void); | |
789 | ||
790 | static struct ofproto_dpif * | |
791 | ofproto_dpif_cast(const struct ofproto *ofproto) | |
792 | { | |
793 | ovs_assert(ofproto->ofproto_class == &ofproto_dpif_class); | |
794 | return CONTAINER_OF(ofproto, struct ofproto_dpif, up); | |
795 | } | |
796 | ||
797 | static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *, | |
798 | uint16_t ofp_port); | |
799 | static struct ofport_dpif *get_odp_port(const struct ofproto_dpif *, | |
800 | uint32_t odp_port); | |
801 | static void ofproto_trace(struct ofproto_dpif *, const struct flow *, | |
802 | const struct ofpbuf *, | |
803 | const struct initial_vals *, struct ds *); | |
804 | ||
805 | /* Packet processing. */ | |
806 | static void update_learning_table(struct ofproto_dpif *, | |
807 | const struct flow *, int vlan, | |
808 | struct ofbundle *); | |
809 | /* Upcalls. */ | |
810 | #define FLOW_MISS_MAX_BATCH 50 | |
811 | static int handle_upcalls(struct dpif_backer *, unsigned int max_batch); | |
812 | ||
813 | /* Flow expiration. */ | |
814 | static int expire(struct dpif_backer *); | |
815 | ||
816 | /* NetFlow. */ | |
817 | static void send_netflow_active_timeouts(struct ofproto_dpif *); | |
818 | ||
819 | /* Utilities. */ | |
820 | static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet); | |
821 | static size_t compose_sflow_action(const struct ofproto_dpif *, | |
822 | struct ofpbuf *odp_actions, | |
823 | const struct flow *, uint32_t odp_port); | |
824 | static void compose_ipfix_action(const struct ofproto_dpif *, | |
825 | struct ofpbuf *odp_actions, | |
826 | const struct flow *); | |
827 | static void add_mirror_actions(struct action_xlate_ctx *ctx, | |
828 | const struct flow *flow); | |
829 | /* Global variables. */ | |
830 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
831 | ||
832 | /* Initial mappings of port to bridge mappings. */ | |
833 | static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports); | |
834 | \f | |
835 | /* Factory functions. */ | |
836 | ||
837 | static void | |
838 | init(const struct shash *iface_hints) | |
839 | { | |
840 | struct shash_node *node; | |
841 | ||
842 | /* Make a local copy, since we don't own 'iface_hints' elements. */ | |
843 | SHASH_FOR_EACH(node, iface_hints) { | |
844 | const struct iface_hint *orig_hint = node->data; | |
845 | struct iface_hint *new_hint = xmalloc(sizeof *new_hint); | |
846 | ||
847 | new_hint->br_name = xstrdup(orig_hint->br_name); | |
848 | new_hint->br_type = xstrdup(orig_hint->br_type); | |
849 | new_hint->ofp_port = orig_hint->ofp_port; | |
850 | ||
851 | shash_add(&init_ofp_ports, node->name, new_hint); | |
852 | } | |
853 | } | |
854 | ||
855 | static void | |
856 | enumerate_types(struct sset *types) | |
857 | { | |
858 | dp_enumerate_types(types); | |
859 | } | |
860 | ||
861 | static int | |
862 | enumerate_names(const char *type, struct sset *names) | |
863 | { | |
864 | struct ofproto_dpif *ofproto; | |
865 | ||
866 | sset_clear(names); | |
867 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
868 | if (strcmp(type, ofproto->up.type)) { | |
869 | continue; | |
870 | } | |
871 | sset_add(names, ofproto->up.name); | |
872 | } | |
873 | ||
874 | return 0; | |
875 | } | |
876 | ||
877 | static int | |
878 | del(const char *type, const char *name) | |
879 | { | |
880 | struct dpif *dpif; | |
881 | int error; | |
882 | ||
883 | error = dpif_open(name, type, &dpif); | |
884 | if (!error) { | |
885 | error = dpif_delete(dpif); | |
886 | dpif_close(dpif); | |
887 | } | |
888 | return error; | |
889 | } | |
890 | \f | |
891 | static const char * | |
892 | port_open_type(const char *datapath_type, const char *port_type) | |
893 | { | |
894 | return dpif_port_open_type(datapath_type, port_type); | |
895 | } | |
896 | ||
897 | /* Type functions. */ | |
898 | ||
899 | static struct ofproto_dpif * | |
900 | lookup_ofproto_dpif_by_port_name(const char *name) | |
901 | { | |
902 | struct ofproto_dpif *ofproto; | |
903 | ||
904 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
905 | if (sset_contains(&ofproto->ports, name)) { | |
906 | return ofproto; | |
907 | } | |
908 | } | |
909 | ||
910 | return NULL; | |
911 | } | |
912 | ||
913 | static int | |
914 | type_run(const char *type) | |
915 | { | |
916 | static long long int push_timer = LLONG_MIN; | |
917 | struct dpif_backer *backer; | |
918 | char *devname; | |
919 | int error; | |
920 | ||
921 | backer = shash_find_data(&all_dpif_backers, type); | |
922 | if (!backer) { | |
923 | /* This is not necessarily a problem, since backers are only | |
924 | * created on demand. */ | |
925 | return 0; | |
926 | } | |
927 | ||
928 | dpif_run(backer->dpif); | |
929 | ||
930 | /* The most natural place to push facet statistics is when they're pulled | |
931 | * from the datapath. However, when there are many flows in the datapath, | |
932 | * this expensive operation can occur so frequently, that it reduces our | |
933 | * ability to quickly set up flows. To reduce the cost, we push statistics | |
934 | * here instead. */ | |
935 | if (time_msec() > push_timer) { | |
936 | push_timer = time_msec() + 2000; | |
937 | push_all_stats(); | |
938 | } | |
939 | ||
940 | if (backer->need_revalidate | |
941 | || !tag_set_is_empty(&backer->revalidate_set)) { | |
942 | struct tag_set revalidate_set = backer->revalidate_set; | |
943 | bool need_revalidate = backer->need_revalidate; | |
944 | struct ofproto_dpif *ofproto; | |
945 | struct simap_node *node; | |
946 | struct simap tmp_backers; | |
947 | ||
948 | /* Handle tunnel garbage collection. */ | |
949 | simap_init(&tmp_backers); | |
950 | simap_swap(&backer->tnl_backers, &tmp_backers); | |
951 | ||
952 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
953 | struct ofport_dpif *iter; | |
954 | ||
955 | if (backer != ofproto->backer) { | |
956 | continue; | |
957 | } | |
958 | ||
959 | HMAP_FOR_EACH (iter, up.hmap_node, &ofproto->up.ports) { | |
960 | const char *dp_port; | |
961 | ||
962 | if (!iter->tnl_port) { | |
963 | continue; | |
964 | } | |
965 | ||
966 | dp_port = netdev_vport_get_dpif_port(iter->up.netdev); | |
967 | node = simap_find(&tmp_backers, dp_port); | |
968 | if (node) { | |
969 | simap_put(&backer->tnl_backers, dp_port, node->data); | |
970 | simap_delete(&tmp_backers, node); | |
971 | node = simap_find(&backer->tnl_backers, dp_port); | |
972 | } else { | |
973 | node = simap_find(&backer->tnl_backers, dp_port); | |
974 | if (!node) { | |
975 | uint32_t odp_port = UINT32_MAX; | |
976 | ||
977 | if (!dpif_port_add(backer->dpif, iter->up.netdev, | |
978 | &odp_port)) { | |
979 | simap_put(&backer->tnl_backers, dp_port, odp_port); | |
980 | node = simap_find(&backer->tnl_backers, dp_port); | |
981 | } | |
982 | } | |
983 | } | |
984 | ||
985 | iter->odp_port = node ? node->data : OVSP_NONE; | |
986 | if (tnl_port_reconfigure(&iter->up, iter->odp_port, | |
987 | &iter->tnl_port)) { | |
988 | backer->need_revalidate = REV_RECONFIGURE; | |
989 | } | |
990 | } | |
991 | } | |
992 | ||
993 | SIMAP_FOR_EACH (node, &tmp_backers) { | |
994 | dpif_port_del(backer->dpif, node->data); | |
995 | } | |
996 | simap_destroy(&tmp_backers); | |
997 | ||
998 | switch (backer->need_revalidate) { | |
999 | case REV_RECONFIGURE: COVERAGE_INC(rev_reconfigure); break; | |
1000 | case REV_STP: COVERAGE_INC(rev_stp); break; | |
1001 | case REV_PORT_TOGGLED: COVERAGE_INC(rev_port_toggled); break; | |
1002 | case REV_FLOW_TABLE: COVERAGE_INC(rev_flow_table); break; | |
1003 | case REV_INCONSISTENCY: COVERAGE_INC(rev_inconsistency); break; | |
1004 | } | |
1005 | ||
1006 | if (backer->need_revalidate) { | |
1007 | /* Clear the drop_keys in case we should now be accepting some | |
1008 | * formerly dropped flows. */ | |
1009 | drop_key_clear(backer); | |
1010 | } | |
1011 | ||
1012 | /* Clear the revalidation flags. */ | |
1013 | tag_set_init(&backer->revalidate_set); | |
1014 | backer->need_revalidate = 0; | |
1015 | ||
1016 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
1017 | struct facet *facet, *next; | |
1018 | ||
1019 | if (ofproto->backer != backer) { | |
1020 | continue; | |
1021 | } | |
1022 | ||
1023 | HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &ofproto->facets) { | |
1024 | if (need_revalidate | |
1025 | || tag_set_intersects(&revalidate_set, facet->tags)) { | |
1026 | facet_revalidate(facet); | |
1027 | run_fast_rl(); | |
1028 | } | |
1029 | } | |
1030 | } | |
1031 | } | |
1032 | ||
1033 | if (timer_expired(&backer->next_expiration)) { | |
1034 | int delay = expire(backer); | |
1035 | timer_set_duration(&backer->next_expiration, delay); | |
1036 | } | |
1037 | ||
1038 | /* Check for port changes in the dpif. */ | |
1039 | while ((error = dpif_port_poll(backer->dpif, &devname)) == 0) { | |
1040 | struct ofproto_dpif *ofproto; | |
1041 | struct dpif_port port; | |
1042 | ||
1043 | /* Don't report on the datapath's device. */ | |
1044 | if (!strcmp(devname, dpif_base_name(backer->dpif))) { | |
1045 | goto next; | |
1046 | } | |
1047 | ||
1048 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, | |
1049 | &all_ofproto_dpifs) { | |
1050 | if (simap_contains(&ofproto->backer->tnl_backers, devname)) { | |
1051 | goto next; | |
1052 | } | |
1053 | } | |
1054 | ||
1055 | ofproto = lookup_ofproto_dpif_by_port_name(devname); | |
1056 | if (dpif_port_query_by_name(backer->dpif, devname, &port)) { | |
1057 | /* The port was removed. If we know the datapath, | |
1058 | * report it through poll_set(). If we don't, it may be | |
1059 | * notifying us of a removal we initiated, so ignore it. | |
1060 | * If there's a pending ENOBUFS, let it stand, since | |
1061 | * everything will be reevaluated. */ | |
1062 | if (ofproto && ofproto->port_poll_errno != ENOBUFS) { | |
1063 | sset_add(&ofproto->port_poll_set, devname); | |
1064 | ofproto->port_poll_errno = 0; | |
1065 | } | |
1066 | } else if (!ofproto) { | |
1067 | /* The port was added, but we don't know with which | |
1068 | * ofproto we should associate it. Delete it. */ | |
1069 | dpif_port_del(backer->dpif, port.port_no); | |
1070 | } | |
1071 | dpif_port_destroy(&port); | |
1072 | ||
1073 | next: | |
1074 | free(devname); | |
1075 | } | |
1076 | ||
1077 | if (error != EAGAIN) { | |
1078 | struct ofproto_dpif *ofproto; | |
1079 | ||
1080 | /* There was some sort of error, so propagate it to all | |
1081 | * ofprotos that use this backer. */ | |
1082 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, | |
1083 | &all_ofproto_dpifs) { | |
1084 | if (ofproto->backer == backer) { | |
1085 | sset_clear(&ofproto->port_poll_set); | |
1086 | ofproto->port_poll_errno = error; | |
1087 | } | |
1088 | } | |
1089 | } | |
1090 | ||
1091 | return 0; | |
1092 | } | |
1093 | ||
1094 | static int | |
1095 | dpif_backer_run_fast(struct dpif_backer *backer, int max_batch) | |
1096 | { | |
1097 | unsigned int work; | |
1098 | ||
1099 | /* Handle one or more batches of upcalls, until there's nothing left to do | |
1100 | * or until we do a fixed total amount of work. | |
1101 | * | |
1102 | * We do work in batches because it can be much cheaper to set up a number | |
1103 | * of flows and fire off their patches all at once. We do multiple batches | |
1104 | * because in some cases handling a packet can cause another packet to be | |
1105 | * queued almost immediately as part of the return flow. Both | |
1106 | * optimizations can make major improvements on some benchmarks and | |
1107 | * presumably for real traffic as well. */ | |
1108 | work = 0; | |
1109 | while (work < max_batch) { | |
1110 | int retval = handle_upcalls(backer, max_batch - work); | |
1111 | if (retval <= 0) { | |
1112 | return -retval; | |
1113 | } | |
1114 | work += retval; | |
1115 | } | |
1116 | ||
1117 | return 0; | |
1118 | } | |
1119 | ||
1120 | static int | |
1121 | type_run_fast(const char *type) | |
1122 | { | |
1123 | struct dpif_backer *backer; | |
1124 | ||
1125 | backer = shash_find_data(&all_dpif_backers, type); | |
1126 | if (!backer) { | |
1127 | /* This is not necessarily a problem, since backers are only | |
1128 | * created on demand. */ | |
1129 | return 0; | |
1130 | } | |
1131 | ||
1132 | return dpif_backer_run_fast(backer, FLOW_MISS_MAX_BATCH); | |
1133 | } | |
1134 | ||
1135 | static void | |
1136 | run_fast_rl(void) | |
1137 | { | |
1138 | static long long int port_rl = LLONG_MIN; | |
1139 | static unsigned int backer_rl = 0; | |
1140 | ||
1141 | if (time_msec() >= port_rl) { | |
1142 | struct ofproto_dpif *ofproto; | |
1143 | struct ofport_dpif *ofport; | |
1144 | ||
1145 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
1146 | ||
1147 | HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) { | |
1148 | port_run_fast(ofport); | |
1149 | } | |
1150 | } | |
1151 | port_rl = time_msec() + 200; | |
1152 | } | |
1153 | ||
1154 | /* XXX: We have to be careful not to do too much work in this function. If | |
1155 | * we call dpif_backer_run_fast() too often, or with too large a batch, | |
1156 | * performance improves signifcantly, but at a cost. It's possible for the | |
1157 | * number of flows in the datapath to increase without bound, and for poll | |
1158 | * loops to take 10s of seconds. The correct solution to this problem, | |
1159 | * long term, is to separate flow miss handling into it's own thread so it | |
1160 | * isn't affected by revalidations, and expirations. Until then, this is | |
1161 | * the best we can do. */ | |
1162 | if (++backer_rl >= 10) { | |
1163 | struct shash_node *node; | |
1164 | ||
1165 | backer_rl = 0; | |
1166 | SHASH_FOR_EACH (node, &all_dpif_backers) { | |
1167 | dpif_backer_run_fast(node->data, 1); | |
1168 | } | |
1169 | } | |
1170 | } | |
1171 | ||
1172 | static void | |
1173 | type_wait(const char *type) | |
1174 | { | |
1175 | struct dpif_backer *backer; | |
1176 | ||
1177 | backer = shash_find_data(&all_dpif_backers, type); | |
1178 | if (!backer) { | |
1179 | /* This is not necessarily a problem, since backers are only | |
1180 | * created on demand. */ | |
1181 | return; | |
1182 | } | |
1183 | ||
1184 | timer_wait(&backer->next_expiration); | |
1185 | } | |
1186 | \f | |
1187 | /* Basic life-cycle. */ | |
1188 | ||
1189 | static int add_internal_flows(struct ofproto_dpif *); | |
1190 | ||
1191 | static struct ofproto * | |
1192 | alloc(void) | |
1193 | { | |
1194 | struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto); | |
1195 | return &ofproto->up; | |
1196 | } | |
1197 | ||
1198 | static void | |
1199 | dealloc(struct ofproto *ofproto_) | |
1200 | { | |
1201 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1202 | free(ofproto); | |
1203 | } | |
1204 | ||
1205 | static void | |
1206 | close_dpif_backer(struct dpif_backer *backer) | |
1207 | { | |
1208 | struct shash_node *node; | |
1209 | ||
1210 | ovs_assert(backer->refcount > 0); | |
1211 | ||
1212 | if (--backer->refcount) { | |
1213 | return; | |
1214 | } | |
1215 | ||
1216 | drop_key_clear(backer); | |
1217 | hmap_destroy(&backer->drop_keys); | |
1218 | ||
1219 | simap_destroy(&backer->tnl_backers); | |
1220 | hmap_destroy(&backer->odp_to_ofport_map); | |
1221 | node = shash_find(&all_dpif_backers, backer->type); | |
1222 | free(backer->type); | |
1223 | shash_delete(&all_dpif_backers, node); | |
1224 | dpif_close(backer->dpif); | |
1225 | ||
1226 | free(backer); | |
1227 | } | |
1228 | ||
1229 | /* Datapath port slated for removal from datapath. */ | |
1230 | struct odp_garbage { | |
1231 | struct list list_node; | |
1232 | uint32_t odp_port; | |
1233 | }; | |
1234 | ||
1235 | static int | |
1236 | open_dpif_backer(const char *type, struct dpif_backer **backerp) | |
1237 | { | |
1238 | struct dpif_backer *backer; | |
1239 | struct dpif_port_dump port_dump; | |
1240 | struct dpif_port port; | |
1241 | struct shash_node *node; | |
1242 | struct list garbage_list; | |
1243 | struct odp_garbage *garbage, *next; | |
1244 | struct sset names; | |
1245 | char *backer_name; | |
1246 | const char *name; | |
1247 | int error; | |
1248 | ||
1249 | backer = shash_find_data(&all_dpif_backers, type); | |
1250 | if (backer) { | |
1251 | backer->refcount++; | |
1252 | *backerp = backer; | |
1253 | return 0; | |
1254 | } | |
1255 | ||
1256 | backer_name = xasprintf("ovs-%s", type); | |
1257 | ||
1258 | /* Remove any existing datapaths, since we assume we're the only | |
1259 | * userspace controlling the datapath. */ | |
1260 | sset_init(&names); | |
1261 | dp_enumerate_names(type, &names); | |
1262 | SSET_FOR_EACH(name, &names) { | |
1263 | struct dpif *old_dpif; | |
1264 | ||
1265 | /* Don't remove our backer if it exists. */ | |
1266 | if (!strcmp(name, backer_name)) { | |
1267 | continue; | |
1268 | } | |
1269 | ||
1270 | if (dpif_open(name, type, &old_dpif)) { | |
1271 | VLOG_WARN("couldn't open old datapath %s to remove it", name); | |
1272 | } else { | |
1273 | dpif_delete(old_dpif); | |
1274 | dpif_close(old_dpif); | |
1275 | } | |
1276 | } | |
1277 | sset_destroy(&names); | |
1278 | ||
1279 | backer = xmalloc(sizeof *backer); | |
1280 | ||
1281 | error = dpif_create_and_open(backer_name, type, &backer->dpif); | |
1282 | free(backer_name); | |
1283 | if (error) { | |
1284 | VLOG_ERR("failed to open datapath of type %s: %s", type, | |
1285 | strerror(error)); | |
1286 | free(backer); | |
1287 | return error; | |
1288 | } | |
1289 | ||
1290 | backer->type = xstrdup(type); | |
1291 | backer->refcount = 1; | |
1292 | hmap_init(&backer->odp_to_ofport_map); | |
1293 | hmap_init(&backer->drop_keys); | |
1294 | timer_set_duration(&backer->next_expiration, 1000); | |
1295 | backer->need_revalidate = 0; | |
1296 | simap_init(&backer->tnl_backers); | |
1297 | tag_set_init(&backer->revalidate_set); | |
1298 | *backerp = backer; | |
1299 | ||
1300 | dpif_flow_flush(backer->dpif); | |
1301 | ||
1302 | /* Loop through the ports already on the datapath and remove any | |
1303 | * that we don't need anymore. */ | |
1304 | list_init(&garbage_list); | |
1305 | dpif_port_dump_start(&port_dump, backer->dpif); | |
1306 | while (dpif_port_dump_next(&port_dump, &port)) { | |
1307 | node = shash_find(&init_ofp_ports, port.name); | |
1308 | if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) { | |
1309 | garbage = xmalloc(sizeof *garbage); | |
1310 | garbage->odp_port = port.port_no; | |
1311 | list_push_front(&garbage_list, &garbage->list_node); | |
1312 | } | |
1313 | } | |
1314 | dpif_port_dump_done(&port_dump); | |
1315 | ||
1316 | LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) { | |
1317 | dpif_port_del(backer->dpif, garbage->odp_port); | |
1318 | list_remove(&garbage->list_node); | |
1319 | free(garbage); | |
1320 | } | |
1321 | ||
1322 | shash_add(&all_dpif_backers, type, backer); | |
1323 | ||
1324 | error = dpif_recv_set(backer->dpif, true); | |
1325 | if (error) { | |
1326 | VLOG_ERR("failed to listen on datapath of type %s: %s", | |
1327 | type, strerror(error)); | |
1328 | close_dpif_backer(backer); | |
1329 | return error; | |
1330 | } | |
1331 | ||
1332 | return error; | |
1333 | } | |
1334 | ||
1335 | static int | |
1336 | construct(struct ofproto *ofproto_) | |
1337 | { | |
1338 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1339 | struct shash_node *node, *next; | |
1340 | int max_ports; | |
1341 | int error; | |
1342 | int i; | |
1343 | ||
1344 | error = open_dpif_backer(ofproto->up.type, &ofproto->backer); | |
1345 | if (error) { | |
1346 | return error; | |
1347 | } | |
1348 | ||
1349 | max_ports = dpif_get_max_ports(ofproto->backer->dpif); | |
1350 | ofproto_init_max_ports(ofproto_, MIN(max_ports, OFPP_MAX)); | |
1351 | ||
1352 | ofproto->n_matches = 0; | |
1353 | ||
1354 | ofproto->netflow = NULL; | |
1355 | ofproto->sflow = NULL; | |
1356 | ofproto->ipfix = NULL; | |
1357 | ofproto->stp = NULL; | |
1358 | hmap_init(&ofproto->bundles); | |
1359 | ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME); | |
1360 | for (i = 0; i < MAX_MIRRORS; i++) { | |
1361 | ofproto->mirrors[i] = NULL; | |
1362 | } | |
1363 | ofproto->has_bonded_bundles = false; | |
1364 | ||
1365 | hmap_init(&ofproto->facets); | |
1366 | hmap_init(&ofproto->subfacets); | |
1367 | ofproto->governor = NULL; | |
1368 | ofproto->consistency_rl = LLONG_MIN; | |
1369 | ||
1370 | for (i = 0; i < N_TABLES; i++) { | |
1371 | struct table_dpif *table = &ofproto->tables[i]; | |
1372 | ||
1373 | table->catchall_table = NULL; | |
1374 | table->other_table = NULL; | |
1375 | table->basis = random_uint32(); | |
1376 | } | |
1377 | ||
1378 | list_init(&ofproto->completions); | |
1379 | ||
1380 | ofproto_dpif_unixctl_init(); | |
1381 | ||
1382 | ofproto->has_mirrors = false; | |
1383 | ofproto->has_bundle_action = false; | |
1384 | ||
1385 | hmap_init(&ofproto->vlandev_map); | |
1386 | hmap_init(&ofproto->realdev_vid_map); | |
1387 | ||
1388 | sset_init(&ofproto->ports); | |
1389 | sset_init(&ofproto->ghost_ports); | |
1390 | sset_init(&ofproto->port_poll_set); | |
1391 | ofproto->port_poll_errno = 0; | |
1392 | ||
1393 | SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) { | |
1394 | struct iface_hint *iface_hint = node->data; | |
1395 | ||
1396 | if (!strcmp(iface_hint->br_name, ofproto->up.name)) { | |
1397 | /* Check if the datapath already has this port. */ | |
1398 | if (dpif_port_exists(ofproto->backer->dpif, node->name)) { | |
1399 | sset_add(&ofproto->ports, node->name); | |
1400 | } | |
1401 | ||
1402 | free(iface_hint->br_name); | |
1403 | free(iface_hint->br_type); | |
1404 | free(iface_hint); | |
1405 | shash_delete(&init_ofp_ports, node); | |
1406 | } | |
1407 | } | |
1408 | ||
1409 | hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node, | |
1410 | hash_string(ofproto->up.name, 0)); | |
1411 | memset(&ofproto->stats, 0, sizeof ofproto->stats); | |
1412 | ||
1413 | ofproto_init_tables(ofproto_, N_TABLES); | |
1414 | error = add_internal_flows(ofproto); | |
1415 | ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY; | |
1416 | ||
1417 | ofproto->n_hit = 0; | |
1418 | ofproto->n_missed = 0; | |
1419 | ||
1420 | ofproto->max_n_subfacet = 0; | |
1421 | ofproto->created = time_msec(); | |
1422 | ofproto->last_minute = ofproto->created; | |
1423 | memset(&ofproto->hourly, 0, sizeof ofproto->hourly); | |
1424 | memset(&ofproto->daily, 0, sizeof ofproto->daily); | |
1425 | ofproto->subfacet_add_count = 0; | |
1426 | ofproto->subfacet_del_count = 0; | |
1427 | ofproto->total_subfacet_add_count = 0; | |
1428 | ofproto->total_subfacet_del_count = 0; | |
1429 | ofproto->total_subfacet_life_span = 0; | |
1430 | ofproto->total_subfacet_count = 0; | |
1431 | ofproto->n_update_stats = 0; | |
1432 | ||
1433 | return error; | |
1434 | } | |
1435 | ||
1436 | static int | |
1437 | add_internal_flow(struct ofproto_dpif *ofproto, int id, | |
1438 | const struct ofpbuf *ofpacts, struct rule_dpif **rulep) | |
1439 | { | |
1440 | struct ofputil_flow_mod fm; | |
1441 | int error; | |
1442 | ||
1443 | match_init_catchall(&fm.match); | |
1444 | fm.priority = 0; | |
1445 | match_set_reg(&fm.match, 0, id); | |
1446 | fm.new_cookie = htonll(0); | |
1447 | fm.cookie = htonll(0); | |
1448 | fm.cookie_mask = htonll(0); | |
1449 | fm.table_id = TBL_INTERNAL; | |
1450 | fm.command = OFPFC_ADD; | |
1451 | fm.idle_timeout = 0; | |
1452 | fm.hard_timeout = 0; | |
1453 | fm.buffer_id = 0; | |
1454 | fm.out_port = 0; | |
1455 | fm.flags = 0; | |
1456 | fm.ofpacts = ofpacts->data; | |
1457 | fm.ofpacts_len = ofpacts->size; | |
1458 | ||
1459 | error = ofproto_flow_mod(&ofproto->up, &fm); | |
1460 | if (error) { | |
1461 | VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)", | |
1462 | id, ofperr_to_string(error)); | |
1463 | return error; | |
1464 | } | |
1465 | ||
1466 | *rulep = rule_dpif_lookup__(ofproto, &fm.match.flow, TBL_INTERNAL); | |
1467 | ovs_assert(*rulep != NULL); | |
1468 | ||
1469 | return 0; | |
1470 | } | |
1471 | ||
1472 | static int | |
1473 | add_internal_flows(struct ofproto_dpif *ofproto) | |
1474 | { | |
1475 | struct ofpact_controller *controller; | |
1476 | uint64_t ofpacts_stub[128 / 8]; | |
1477 | struct ofpbuf ofpacts; | |
1478 | int error; | |
1479 | int id; | |
1480 | ||
1481 | ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub); | |
1482 | id = 1; | |
1483 | ||
1484 | controller = ofpact_put_CONTROLLER(&ofpacts); | |
1485 | controller->max_len = UINT16_MAX; | |
1486 | controller->controller_id = 0; | |
1487 | controller->reason = OFPR_NO_MATCH; | |
1488 | ofpact_pad(&ofpacts); | |
1489 | ||
1490 | error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule); | |
1491 | if (error) { | |
1492 | return error; | |
1493 | } | |
1494 | ||
1495 | ofpbuf_clear(&ofpacts); | |
1496 | error = add_internal_flow(ofproto, id++, &ofpacts, | |
1497 | &ofproto->no_packet_in_rule); | |
1498 | return error; | |
1499 | } | |
1500 | ||
1501 | static void | |
1502 | complete_operations(struct ofproto_dpif *ofproto) | |
1503 | { | |
1504 | struct dpif_completion *c, *next; | |
1505 | ||
1506 | LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) { | |
1507 | ofoperation_complete(c->op, 0); | |
1508 | list_remove(&c->list_node); | |
1509 | free(c); | |
1510 | } | |
1511 | } | |
1512 | ||
1513 | static void | |
1514 | destruct(struct ofproto *ofproto_) | |
1515 | { | |
1516 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1517 | struct rule_dpif *rule, *next_rule; | |
1518 | struct oftable *table; | |
1519 | int i; | |
1520 | ||
1521 | hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node); | |
1522 | complete_operations(ofproto); | |
1523 | ||
1524 | OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) { | |
1525 | struct cls_cursor cursor; | |
1526 | ||
1527 | cls_cursor_init(&cursor, &table->cls, NULL); | |
1528 | CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) { | |
1529 | ofproto_rule_destroy(&rule->up); | |
1530 | } | |
1531 | } | |
1532 | ||
1533 | for (i = 0; i < MAX_MIRRORS; i++) { | |
1534 | mirror_destroy(ofproto->mirrors[i]); | |
1535 | } | |
1536 | ||
1537 | netflow_destroy(ofproto->netflow); | |
1538 | dpif_sflow_destroy(ofproto->sflow); | |
1539 | hmap_destroy(&ofproto->bundles); | |
1540 | mac_learning_destroy(ofproto->ml); | |
1541 | ||
1542 | hmap_destroy(&ofproto->facets); | |
1543 | hmap_destroy(&ofproto->subfacets); | |
1544 | governor_destroy(ofproto->governor); | |
1545 | ||
1546 | hmap_destroy(&ofproto->vlandev_map); | |
1547 | hmap_destroy(&ofproto->realdev_vid_map); | |
1548 | ||
1549 | sset_destroy(&ofproto->ports); | |
1550 | sset_destroy(&ofproto->ghost_ports); | |
1551 | sset_destroy(&ofproto->port_poll_set); | |
1552 | ||
1553 | close_dpif_backer(ofproto->backer); | |
1554 | } | |
1555 | ||
1556 | static int | |
1557 | run_fast(struct ofproto *ofproto_) | |
1558 | { | |
1559 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1560 | struct ofport_dpif *ofport; | |
1561 | ||
1562 | HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) { | |
1563 | port_run_fast(ofport); | |
1564 | } | |
1565 | ||
1566 | return 0; | |
1567 | } | |
1568 | ||
1569 | static int | |
1570 | run(struct ofproto *ofproto_) | |
1571 | { | |
1572 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1573 | struct ofport_dpif *ofport; | |
1574 | struct ofbundle *bundle; | |
1575 | int error; | |
1576 | ||
1577 | if (!clogged) { | |
1578 | complete_operations(ofproto); | |
1579 | } | |
1580 | ||
1581 | error = run_fast(ofproto_); | |
1582 | if (error) { | |
1583 | return error; | |
1584 | } | |
1585 | ||
1586 | if (ofproto->netflow) { | |
1587 | if (netflow_run(ofproto->netflow)) { | |
1588 | send_netflow_active_timeouts(ofproto); | |
1589 | } | |
1590 | } | |
1591 | if (ofproto->sflow) { | |
1592 | dpif_sflow_run(ofproto->sflow); | |
1593 | } | |
1594 | ||
1595 | HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) { | |
1596 | port_run(ofport); | |
1597 | } | |
1598 | HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { | |
1599 | bundle_run(bundle); | |
1600 | } | |
1601 | ||
1602 | stp_run(ofproto); | |
1603 | mac_learning_run(ofproto->ml, &ofproto->backer->revalidate_set); | |
1604 | ||
1605 | /* Check the consistency of a random facet, to aid debugging. */ | |
1606 | if (time_msec() >= ofproto->consistency_rl | |
1607 | && !hmap_is_empty(&ofproto->facets) | |
1608 | && !ofproto->backer->need_revalidate) { | |
1609 | struct facet *facet; | |
1610 | ||
1611 | ofproto->consistency_rl = time_msec() + 250; | |
1612 | ||
1613 | facet = CONTAINER_OF(hmap_random_node(&ofproto->facets), | |
1614 | struct facet, hmap_node); | |
1615 | if (!tag_set_intersects(&ofproto->backer->revalidate_set, | |
1616 | facet->tags)) { | |
1617 | if (!facet_check_consistency(facet)) { | |
1618 | ofproto->backer->need_revalidate = REV_INCONSISTENCY; | |
1619 | } | |
1620 | } | |
1621 | } | |
1622 | ||
1623 | if (ofproto->governor) { | |
1624 | size_t n_subfacets; | |
1625 | ||
1626 | governor_run(ofproto->governor); | |
1627 | ||
1628 | /* If the governor has shrunk to its minimum size and the number of | |
1629 | * subfacets has dwindled, then drop the governor entirely. | |
1630 | * | |
1631 | * For hysteresis, the number of subfacets to drop the governor is | |
1632 | * smaller than the number needed to trigger its creation. */ | |
1633 | n_subfacets = hmap_count(&ofproto->subfacets); | |
1634 | if (n_subfacets * 4 < ofproto->up.flow_eviction_threshold | |
1635 | && governor_is_idle(ofproto->governor)) { | |
1636 | governor_destroy(ofproto->governor); | |
1637 | ofproto->governor = NULL; | |
1638 | } | |
1639 | } | |
1640 | ||
1641 | return 0; | |
1642 | } | |
1643 | ||
1644 | static void | |
1645 | wait(struct ofproto *ofproto_) | |
1646 | { | |
1647 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1648 | struct ofport_dpif *ofport; | |
1649 | struct ofbundle *bundle; | |
1650 | ||
1651 | if (!clogged && !list_is_empty(&ofproto->completions)) { | |
1652 | poll_immediate_wake(); | |
1653 | } | |
1654 | ||
1655 | dpif_wait(ofproto->backer->dpif); | |
1656 | dpif_recv_wait(ofproto->backer->dpif); | |
1657 | if (ofproto->sflow) { | |
1658 | dpif_sflow_wait(ofproto->sflow); | |
1659 | } | |
1660 | if (!tag_set_is_empty(&ofproto->backer->revalidate_set)) { | |
1661 | poll_immediate_wake(); | |
1662 | } | |
1663 | HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) { | |
1664 | port_wait(ofport); | |
1665 | } | |
1666 | HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { | |
1667 | bundle_wait(bundle); | |
1668 | } | |
1669 | if (ofproto->netflow) { | |
1670 | netflow_wait(ofproto->netflow); | |
1671 | } | |
1672 | mac_learning_wait(ofproto->ml); | |
1673 | stp_wait(ofproto); | |
1674 | if (ofproto->backer->need_revalidate) { | |
1675 | /* Shouldn't happen, but if it does just go around again. */ | |
1676 | VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()"); | |
1677 | poll_immediate_wake(); | |
1678 | } | |
1679 | if (ofproto->governor) { | |
1680 | governor_wait(ofproto->governor); | |
1681 | } | |
1682 | } | |
1683 | ||
1684 | static void | |
1685 | get_memory_usage(const struct ofproto *ofproto_, struct simap *usage) | |
1686 | { | |
1687 | const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1688 | ||
1689 | simap_increase(usage, "facets", hmap_count(&ofproto->facets)); | |
1690 | simap_increase(usage, "subfacets", hmap_count(&ofproto->subfacets)); | |
1691 | } | |
1692 | ||
1693 | static void | |
1694 | flush(struct ofproto *ofproto_) | |
1695 | { | |
1696 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1697 | struct subfacet *subfacet, *next_subfacet; | |
1698 | struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH]; | |
1699 | int n_batch; | |
1700 | ||
1701 | n_batch = 0; | |
1702 | HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node, | |
1703 | &ofproto->subfacets) { | |
1704 | if (subfacet->path != SF_NOT_INSTALLED) { | |
1705 | batch[n_batch++] = subfacet; | |
1706 | if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) { | |
1707 | subfacet_destroy_batch(ofproto, batch, n_batch); | |
1708 | n_batch = 0; | |
1709 | } | |
1710 | } else { | |
1711 | subfacet_destroy(subfacet); | |
1712 | } | |
1713 | } | |
1714 | ||
1715 | if (n_batch > 0) { | |
1716 | subfacet_destroy_batch(ofproto, batch, n_batch); | |
1717 | } | |
1718 | } | |
1719 | ||
1720 | static void | |
1721 | get_features(struct ofproto *ofproto_ OVS_UNUSED, | |
1722 | bool *arp_match_ip, enum ofputil_action_bitmap *actions) | |
1723 | { | |
1724 | *arp_match_ip = true; | |
1725 | *actions = (OFPUTIL_A_OUTPUT | | |
1726 | OFPUTIL_A_SET_VLAN_VID | | |
1727 | OFPUTIL_A_SET_VLAN_PCP | | |
1728 | OFPUTIL_A_STRIP_VLAN | | |
1729 | OFPUTIL_A_SET_DL_SRC | | |
1730 | OFPUTIL_A_SET_DL_DST | | |
1731 | OFPUTIL_A_SET_NW_SRC | | |
1732 | OFPUTIL_A_SET_NW_DST | | |
1733 | OFPUTIL_A_SET_NW_TOS | | |
1734 | OFPUTIL_A_SET_TP_SRC | | |
1735 | OFPUTIL_A_SET_TP_DST | | |
1736 | OFPUTIL_A_ENQUEUE); | |
1737 | } | |
1738 | ||
1739 | static void | |
1740 | get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots) | |
1741 | { | |
1742 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1743 | struct dpif_dp_stats s; | |
1744 | ||
1745 | strcpy(ots->name, "classifier"); | |
1746 | ||
1747 | dpif_get_dp_stats(ofproto->backer->dpif, &s); | |
1748 | ||
1749 | ots->lookup_count = htonll(s.n_hit + s.n_missed); | |
1750 | ots->matched_count = htonll(s.n_hit + ofproto->n_matches); | |
1751 | } | |
1752 | ||
1753 | static struct ofport * | |
1754 | port_alloc(void) | |
1755 | { | |
1756 | struct ofport_dpif *port = xmalloc(sizeof *port); | |
1757 | return &port->up; | |
1758 | } | |
1759 | ||
1760 | static void | |
1761 | port_dealloc(struct ofport *port_) | |
1762 | { | |
1763 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
1764 | free(port); | |
1765 | } | |
1766 | ||
1767 | static int | |
1768 | port_construct(struct ofport *port_) | |
1769 | { | |
1770 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
1771 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto); | |
1772 | const struct netdev *netdev = port->up.netdev; | |
1773 | struct dpif_port dpif_port; | |
1774 | int error; | |
1775 | ||
1776 | ofproto->backer->need_revalidate = REV_RECONFIGURE; | |
1777 | port->bundle = NULL; | |
1778 | port->cfm = NULL; | |
1779 | port->bfd = NULL; | |
1780 | port->tag = tag_create_random(); | |
1781 | port->may_enable = true; | |
1782 | port->stp_port = NULL; | |
1783 | port->stp_state = STP_DISABLED; | |
1784 | port->tnl_port = NULL; | |
1785 | hmap_init(&port->priorities); | |
1786 | port->realdev_ofp_port = 0; | |
1787 | port->vlandev_vid = 0; | |
1788 | port->carrier_seq = netdev_get_carrier_resets(netdev); | |
1789 | ||
1790 | if (netdev_vport_is_patch(netdev)) { | |
1791 | /* By bailing out here, we don't submit the port to the sFlow module | |
1792 | * to be considered for counter polling export. This is correct | |
1793 | * because the patch port represents an interface that sFlow considers | |
1794 | * to be "internal" to the switch as a whole, and therefore not an | |
1795 | * candidate for counter polling. */ | |
1796 | port->odp_port = OVSP_NONE; | |
1797 | return 0; | |
1798 | } | |
1799 | ||
1800 | error = dpif_port_query_by_name(ofproto->backer->dpif, | |
1801 | netdev_vport_get_dpif_port(netdev), | |
1802 | &dpif_port); | |
1803 | if (error) { | |
1804 | return error; | |
1805 | } | |
1806 | ||
1807 | port->odp_port = dpif_port.port_no; | |
1808 | ||
1809 | if (netdev_get_tunnel_config(netdev)) { | |
1810 | port->tnl_port = tnl_port_add(&port->up, port->odp_port); | |
1811 | } else { | |
1812 | /* Sanity-check that a mapping doesn't already exist. This | |
1813 | * shouldn't happen for non-tunnel ports. */ | |
1814 | if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) { | |
1815 | VLOG_ERR("port %s already has an OpenFlow port number", | |
1816 | dpif_port.name); | |
1817 | dpif_port_destroy(&dpif_port); | |
1818 | return EBUSY; | |
1819 | } | |
1820 | ||
1821 | hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node, | |
1822 | hash_int(port->odp_port, 0)); | |
1823 | } | |
1824 | dpif_port_destroy(&dpif_port); | |
1825 | ||
1826 | if (ofproto->sflow) { | |
1827 | dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port); | |
1828 | } | |
1829 | ||
1830 | return 0; | |
1831 | } | |
1832 | ||
1833 | static void | |
1834 | port_destruct(struct ofport *port_) | |
1835 | { | |
1836 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
1837 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto); | |
1838 | const char *dp_port_name = netdev_vport_get_dpif_port(port->up.netdev); | |
1839 | const char *devname = netdev_get_name(port->up.netdev); | |
1840 | ||
1841 | if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) { | |
1842 | /* The underlying device is still there, so delete it. This | |
1843 | * happens when the ofproto is being destroyed, since the caller | |
1844 | * assumes that removal of attached ports will happen as part of | |
1845 | * destruction. */ | |
1846 | if (!port->tnl_port) { | |
1847 | dpif_port_del(ofproto->backer->dpif, port->odp_port); | |
1848 | } | |
1849 | ofproto->backer->need_revalidate = REV_RECONFIGURE; | |
1850 | } | |
1851 | ||
1852 | if (port->odp_port != OVSP_NONE && !port->tnl_port) { | |
1853 | hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node); | |
1854 | } | |
1855 | ||
1856 | tnl_port_del(port->tnl_port); | |
1857 | sset_find_and_delete(&ofproto->ports, devname); | |
1858 | sset_find_and_delete(&ofproto->ghost_ports, devname); | |
1859 | ofproto->backer->need_revalidate = REV_RECONFIGURE; | |
1860 | bundle_remove(port_); | |
1861 | set_cfm(port_, NULL); | |
1862 | if (ofproto->sflow) { | |
1863 | dpif_sflow_del_port(ofproto->sflow, port->odp_port); | |
1864 | } | |
1865 | ||
1866 | ofport_clear_priorities(port); | |
1867 | hmap_destroy(&port->priorities); | |
1868 | } | |
1869 | ||
1870 | static void | |
1871 | port_modified(struct ofport *port_) | |
1872 | { | |
1873 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
1874 | ||
1875 | if (port->bundle && port->bundle->bond) { | |
1876 | bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev); | |
1877 | } | |
1878 | } | |
1879 | ||
1880 | static void | |
1881 | port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config) | |
1882 | { | |
1883 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
1884 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto); | |
1885 | enum ofputil_port_config changed = old_config ^ port->up.pp.config; | |
1886 | ||
1887 | if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP | | |
1888 | OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD | | |
1889 | OFPUTIL_PC_NO_PACKET_IN)) { | |
1890 | ofproto->backer->need_revalidate = REV_RECONFIGURE; | |
1891 | ||
1892 | if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) { | |
1893 | bundle_update(port->bundle); | |
1894 | } | |
1895 | } | |
1896 | } | |
1897 | ||
1898 | static int | |
1899 | set_sflow(struct ofproto *ofproto_, | |
1900 | const struct ofproto_sflow_options *sflow_options) | |
1901 | { | |
1902 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1903 | struct dpif_sflow *ds = ofproto->sflow; | |
1904 | ||
1905 | if (sflow_options) { | |
1906 | if (!ds) { | |
1907 | struct ofport_dpif *ofport; | |
1908 | ||
1909 | ds = ofproto->sflow = dpif_sflow_create(); | |
1910 | HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) { | |
1911 | dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port); | |
1912 | } | |
1913 | ofproto->backer->need_revalidate = REV_RECONFIGURE; | |
1914 | } | |
1915 | dpif_sflow_set_options(ds, sflow_options); | |
1916 | } else { | |
1917 | if (ds) { | |
1918 | dpif_sflow_destroy(ds); | |
1919 | ofproto->backer->need_revalidate = REV_RECONFIGURE; | |
1920 | ofproto->sflow = NULL; | |
1921 | } | |
1922 | } | |
1923 | return 0; | |
1924 | } | |
1925 | ||
1926 | static int | |
1927 | set_ipfix( | |
1928 | struct ofproto *ofproto_, | |
1929 | const struct ofproto_ipfix_bridge_exporter_options *bridge_exporter_options, | |
1930 | const struct ofproto_ipfix_flow_exporter_options *flow_exporters_options, | |
1931 | size_t n_flow_exporters_options) | |
1932 | { | |
1933 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1934 | struct dpif_ipfix *di = ofproto->ipfix; | |
1935 | ||
1936 | if (bridge_exporter_options || flow_exporters_options) { | |
1937 | if (!di) { | |
1938 | di = ofproto->ipfix = dpif_ipfix_create(); | |
1939 | } | |
1940 | dpif_ipfix_set_options( | |
1941 | di, bridge_exporter_options, flow_exporters_options, | |
1942 | n_flow_exporters_options); | |
1943 | } else { | |
1944 | if (di) { | |
1945 | dpif_ipfix_destroy(di); | |
1946 | ofproto->ipfix = NULL; | |
1947 | } | |
1948 | } | |
1949 | return 0; | |
1950 | } | |
1951 | ||
1952 | static int | |
1953 | set_cfm(struct ofport *ofport_, const struct cfm_settings *s) | |
1954 | { | |
1955 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
1956 | int error; | |
1957 | ||
1958 | if (!s) { | |
1959 | error = 0; | |
1960 | } else { | |
1961 | if (!ofport->cfm) { | |
1962 | struct ofproto_dpif *ofproto; | |
1963 | ||
1964 | ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
1965 | ofproto->backer->need_revalidate = REV_RECONFIGURE; | |
1966 | ofport->cfm = cfm_create(netdev_get_name(ofport->up.netdev)); | |
1967 | } | |
1968 | ||
1969 | if (cfm_configure(ofport->cfm, s)) { | |
1970 | return 0; | |
1971 | } | |
1972 | ||
1973 | error = EINVAL; | |
1974 | } | |
1975 | cfm_destroy(ofport->cfm); | |
1976 | ofport->cfm = NULL; | |
1977 | return error; | |
1978 | } | |
1979 | ||
1980 | static bool | |
1981 | get_cfm_status(const struct ofport *ofport_, | |
1982 | struct ofproto_cfm_status *status) | |
1983 | { | |
1984 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
1985 | ||
1986 | if (ofport->cfm) { | |
1987 | status->faults = cfm_get_fault(ofport->cfm); | |
1988 | status->remote_opstate = cfm_get_opup(ofport->cfm); | |
1989 | status->health = cfm_get_health(ofport->cfm); | |
1990 | cfm_get_remote_mpids(ofport->cfm, &status->rmps, &status->n_rmps); | |
1991 | return true; | |
1992 | } else { | |
1993 | return false; | |
1994 | } | |
1995 | } | |
1996 | ||
1997 | static int | |
1998 | set_bfd(struct ofport *ofport_, const struct smap *cfg) | |
1999 | { | |
2000 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto); | |
2001 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
2002 | struct bfd *old; | |
2003 | ||
2004 | old = ofport->bfd; | |
2005 | ofport->bfd = bfd_configure(old, netdev_get_name(ofport->up.netdev), cfg); | |
2006 | if (ofport->bfd != old) { | |
2007 | ofproto->backer->need_revalidate = REV_RECONFIGURE; | |
2008 | } | |
2009 | ||
2010 | return 0; | |
2011 | } | |
2012 | ||
2013 | static int | |
2014 | get_bfd_status(struct ofport *ofport_, struct smap *smap) | |
2015 | { | |
2016 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
2017 | ||
2018 | if (ofport->bfd) { | |
2019 | bfd_get_status(ofport->bfd, smap); | |
2020 | return 0; | |
2021 | } else { | |
2022 | return ENOENT; | |
2023 | } | |
2024 | } | |
2025 | \f | |
2026 | /* Spanning Tree. */ | |
2027 | ||
2028 | static void | |
2029 | send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_) | |
2030 | { | |
2031 | struct ofproto_dpif *ofproto = ofproto_; | |
2032 | struct stp_port *sp = stp_get_port(ofproto->stp, port_num); | |
2033 | struct ofport_dpif *ofport; | |
2034 | ||
2035 | ofport = stp_port_get_aux(sp); | |
2036 | if (!ofport) { | |
2037 | VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d", | |
2038 | ofproto->up.name, port_num); | |
2039 | } else { | |
2040 | struct eth_header *eth = pkt->l2; | |
2041 | ||
2042 | netdev_get_etheraddr(ofport->up.netdev, eth->eth_src); | |
2043 | if (eth_addr_is_zero(eth->eth_src)) { | |
2044 | VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d " | |
2045 | "with unknown MAC", ofproto->up.name, port_num); | |
2046 | } else { | |
2047 | send_packet(ofport, pkt); | |
2048 | } | |
2049 | } | |
2050 | ofpbuf_delete(pkt); | |
2051 | } | |
2052 | ||
2053 | /* Configures STP on 'ofproto_' using the settings defined in 's'. */ | |
2054 | static int | |
2055 | set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s) | |
2056 | { | |
2057 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2058 | ||
2059 | /* Only revalidate flows if the configuration changed. */ | |
2060 | if (!s != !ofproto->stp) { | |
2061 | ofproto->backer->need_revalidate = REV_RECONFIGURE; | |
2062 | } | |
2063 | ||
2064 | if (s) { | |
2065 | if (!ofproto->stp) { | |
2066 | ofproto->stp = stp_create(ofproto_->name, s->system_id, | |
2067 | send_bpdu_cb, ofproto); | |
2068 | ofproto->stp_last_tick = time_msec(); | |
2069 | } | |
2070 | ||
2071 | stp_set_bridge_id(ofproto->stp, s->system_id); | |
2072 | stp_set_bridge_priority(ofproto->stp, s->priority); | |
2073 | stp_set_hello_time(ofproto->stp, s->hello_time); | |
2074 | stp_set_max_age(ofproto->stp, s->max_age); | |
2075 | stp_set_forward_delay(ofproto->stp, s->fwd_delay); | |
2076 | } else { | |
2077 | struct ofport *ofport; | |
2078 | ||
2079 | HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) { | |
2080 | set_stp_port(ofport, NULL); | |
2081 | } | |
2082 | ||
2083 | stp_destroy(ofproto->stp); | |
2084 | ofproto->stp = NULL; | |
2085 | } | |
2086 | ||
2087 | return 0; | |
2088 | } | |
2089 | ||
2090 | static int | |
2091 | get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s) | |
2092 | { | |
2093 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2094 | ||
2095 | if (ofproto->stp) { | |
2096 | s->enabled = true; | |
2097 | s->bridge_id = stp_get_bridge_id(ofproto->stp); | |
2098 | s->designated_root = stp_get_designated_root(ofproto->stp); | |
2099 | s->root_path_cost = stp_get_root_path_cost(ofproto->stp); | |
2100 | } else { | |
2101 | s->enabled = false; | |
2102 | } | |
2103 | ||
2104 | return 0; | |
2105 | } | |
2106 | ||
2107 | static void | |
2108 | update_stp_port_state(struct ofport_dpif *ofport) | |
2109 | { | |
2110 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
2111 | enum stp_state state; | |
2112 | ||
2113 | /* Figure out new state. */ | |
2114 | state = ofport->stp_port ? stp_port_get_state(ofport->stp_port) | |
2115 | : STP_DISABLED; | |
2116 | ||
2117 | /* Update state. */ | |
2118 | if (ofport->stp_state != state) { | |
2119 | enum ofputil_port_state of_state; | |
2120 | bool fwd_change; | |
2121 | ||
2122 | VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s", | |
2123 | netdev_get_name(ofport->up.netdev), | |
2124 | stp_state_name(ofport->stp_state), | |
2125 | stp_state_name(state)); | |
2126 | if (stp_learn_in_state(ofport->stp_state) | |
2127 | != stp_learn_in_state(state)) { | |
2128 | /* xxx Learning action flows should also be flushed. */ | |
2129 | mac_learning_flush(ofproto->ml, | |
2130 | &ofproto->backer->revalidate_set); | |
2131 | } | |
2132 | fwd_change = stp_forward_in_state(ofport->stp_state) | |
2133 | != stp_forward_in_state(state); | |
2134 | ||
2135 | ofproto->backer->need_revalidate = REV_STP; | |
2136 | ofport->stp_state = state; | |
2137 | ofport->stp_state_entered = time_msec(); | |
2138 | ||
2139 | if (fwd_change && ofport->bundle) { | |
2140 | bundle_update(ofport->bundle); | |
2141 | } | |
2142 | ||
2143 | /* Update the STP state bits in the OpenFlow port description. */ | |
2144 | of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK; | |
2145 | of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN | |
2146 | : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN | |
2147 | : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD | |
2148 | : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK | |
2149 | : 0); | |
2150 | ofproto_port_set_state(&ofport->up, of_state); | |
2151 | } | |
2152 | } | |
2153 | ||
2154 | /* Configures STP on 'ofport_' using the settings defined in 's'. The | |
2155 | * caller is responsible for assigning STP port numbers and ensuring | |
2156 | * there are no duplicates. */ | |
2157 | static int | |
2158 | set_stp_port(struct ofport *ofport_, | |
2159 | const struct ofproto_port_stp_settings *s) | |
2160 | { | |
2161 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
2162 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
2163 | struct stp_port *sp = ofport->stp_port; | |
2164 | ||
2165 | if (!s || !s->enable) { | |
2166 | if (sp) { | |
2167 | ofport->stp_port = NULL; | |
2168 | stp_port_disable(sp); | |
2169 | update_stp_port_state(ofport); | |
2170 | } | |
2171 | return 0; | |
2172 | } else if (sp && stp_port_no(sp) != s->port_num | |
2173 | && ofport == stp_port_get_aux(sp)) { | |
2174 | /* The port-id changed, so disable the old one if it's not | |
2175 | * already in use by another port. */ | |
2176 | stp_port_disable(sp); | |
2177 | } | |
2178 | ||
2179 | sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num); | |
2180 | stp_port_enable(sp); | |
2181 | ||
2182 | stp_port_set_aux(sp, ofport); | |
2183 | stp_port_set_priority(sp, s->priority); | |
2184 | stp_port_set_path_cost(sp, s->path_cost); | |
2185 | ||
2186 | update_stp_port_state(ofport); | |
2187 | ||
2188 | return 0; | |
2189 | } | |
2190 | ||
2191 | static int | |
2192 | get_stp_port_status(struct ofport *ofport_, | |
2193 | struct ofproto_port_stp_status *s) | |
2194 | { | |
2195 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
2196 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
2197 | struct stp_port *sp = ofport->stp_port; | |
2198 | ||
2199 | if (!ofproto->stp || !sp) { | |
2200 | s->enabled = false; | |
2201 | return 0; | |
2202 | } | |
2203 | ||
2204 | s->enabled = true; | |
2205 | s->port_id = stp_port_get_id(sp); | |
2206 | s->state = stp_port_get_state(sp); | |
2207 | s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000; | |
2208 | s->role = stp_port_get_role(sp); | |
2209 | stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count); | |
2210 | ||
2211 | return 0; | |
2212 | } | |
2213 | ||
2214 | static void | |
2215 | stp_run(struct ofproto_dpif *ofproto) | |
2216 | { | |
2217 | if (ofproto->stp) { | |
2218 | long long int now = time_msec(); | |
2219 | long long int elapsed = now - ofproto->stp_last_tick; | |
2220 | struct stp_port *sp; | |
2221 | ||
2222 | if (elapsed > 0) { | |
2223 | stp_tick(ofproto->stp, MIN(INT_MAX, elapsed)); | |
2224 | ofproto->stp_last_tick = now; | |
2225 | } | |
2226 | while (stp_get_changed_port(ofproto->stp, &sp)) { | |
2227 | struct ofport_dpif *ofport = stp_port_get_aux(sp); | |
2228 | ||
2229 | if (ofport) { | |
2230 | update_stp_port_state(ofport); | |
2231 | } | |
2232 | } | |
2233 | ||
2234 | if (stp_check_and_reset_fdb_flush(ofproto->stp)) { | |
2235 | mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set); | |
2236 | } | |
2237 | } | |
2238 | } | |
2239 | ||
2240 | static void | |
2241 | stp_wait(struct ofproto_dpif *ofproto) | |
2242 | { | |
2243 | if (ofproto->stp) { | |
2244 | poll_timer_wait(1000); | |
2245 | } | |
2246 | } | |
2247 | ||
2248 | /* Returns true if STP should process 'flow'. */ | |
2249 | static bool | |
2250 | stp_should_process_flow(const struct flow *flow) | |
2251 | { | |
2252 | return eth_addr_equals(flow->dl_dst, eth_addr_stp); | |
2253 | } | |
2254 | ||
2255 | static void | |
2256 | stp_process_packet(const struct ofport_dpif *ofport, | |
2257 | const struct ofpbuf *packet) | |
2258 | { | |
2259 | struct ofpbuf payload = *packet; | |
2260 | struct eth_header *eth = payload.data; | |
2261 | struct stp_port *sp = ofport->stp_port; | |
2262 | ||
2263 | /* Sink packets on ports that have STP disabled when the bridge has | |
2264 | * STP enabled. */ | |
2265 | if (!sp || stp_port_get_state(sp) == STP_DISABLED) { | |
2266 | return; | |
2267 | } | |
2268 | ||
2269 | /* Trim off padding on payload. */ | |
2270 | if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) { | |
2271 | payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN; | |
2272 | } | |
2273 | ||
2274 | if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) { | |
2275 | stp_received_bpdu(sp, payload.data, payload.size); | |
2276 | } | |
2277 | } | |
2278 | \f | |
2279 | static struct priority_to_dscp * | |
2280 | get_priority(const struct ofport_dpif *ofport, uint32_t priority) | |
2281 | { | |
2282 | struct priority_to_dscp *pdscp; | |
2283 | uint32_t hash; | |
2284 | ||
2285 | hash = hash_int(priority, 0); | |
2286 | HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) { | |
2287 | if (pdscp->priority == priority) { | |
2288 | return pdscp; | |
2289 | } | |
2290 | } | |
2291 | return NULL; | |
2292 | } | |
2293 | ||
2294 | static void | |
2295 | ofport_clear_priorities(struct ofport_dpif *ofport) | |
2296 | { | |
2297 | struct priority_to_dscp *pdscp, *next; | |
2298 | ||
2299 | HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) { | |
2300 | hmap_remove(&ofport->priorities, &pdscp->hmap_node); | |
2301 | free(pdscp); | |
2302 | } | |
2303 | } | |
2304 | ||
2305 | static int | |
2306 | set_queues(struct ofport *ofport_, | |
2307 | const struct ofproto_port_queue *qdscp_list, | |
2308 | size_t n_qdscp) | |
2309 | { | |
2310 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
2311 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
2312 | struct hmap new = HMAP_INITIALIZER(&new); | |
2313 | size_t i; | |
2314 | ||
2315 | for (i = 0; i < n_qdscp; i++) { | |
2316 | struct priority_to_dscp *pdscp; | |
2317 | uint32_t priority; | |
2318 | uint8_t dscp; | |
2319 | ||
2320 | dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK; | |
2321 | if (dpif_queue_to_priority(ofproto->backer->dpif, qdscp_list[i].queue, | |
2322 | &priority)) { | |
2323 | continue; | |
2324 | } | |
2325 | ||
2326 | pdscp = get_priority(ofport, priority); | |
2327 | if (pdscp) { | |
2328 | hmap_remove(&ofport->priorities, &pdscp->hmap_node); | |
2329 | } else { | |
2330 | pdscp = xmalloc(sizeof *pdscp); | |
2331 | pdscp->priority = priority; | |
2332 | pdscp->dscp = dscp; | |
2333 | ofproto->backer->need_revalidate = REV_RECONFIGURE; | |
2334 | } | |
2335 | ||
2336 | if (pdscp->dscp != dscp) { | |
2337 | pdscp->dscp = dscp; | |
2338 | ofproto->backer->need_revalidate = REV_RECONFIGURE; | |
2339 | } | |
2340 | ||
2341 | hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0)); | |
2342 | } | |
2343 | ||
2344 | if (!hmap_is_empty(&ofport->priorities)) { | |
2345 | ofport_clear_priorities(ofport); | |
2346 | ofproto->backer->need_revalidate = REV_RECONFIGURE; | |
2347 | } | |
2348 | ||
2349 | hmap_swap(&new, &ofport->priorities); | |
2350 | hmap_destroy(&new); | |
2351 | ||
2352 | return 0; | |
2353 | } | |
2354 | \f | |
2355 | /* Bundles. */ | |
2356 | ||
2357 | /* Expires all MAC learning entries associated with 'bundle' and forces its | |
2358 | * ofproto to revalidate every flow. | |
2359 | * | |
2360 | * Normally MAC learning entries are removed only from the ofproto associated | |
2361 | * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries | |
2362 | * are removed from every ofproto. When patch ports and SLB bonds are in use | |
2363 | * and a VM migration happens and the gratuitous ARPs are somehow lost, this | |
2364 | * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate | |
2365 | * with the host from which it migrated. */ | |
2366 | static void | |
2367 | bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos) | |
2368 | { | |
2369 | struct ofproto_dpif *ofproto = bundle->ofproto; | |
2370 | struct mac_learning *ml = ofproto->ml; | |
2371 | struct mac_entry *mac, *next_mac; | |
2372 | ||
2373 | ofproto->backer->need_revalidate = REV_RECONFIGURE; | |
2374 | LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) { | |
2375 | if (mac->port.p == bundle) { | |
2376 | if (all_ofprotos) { | |
2377 | struct ofproto_dpif *o; | |
2378 | ||
2379 | HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
2380 | if (o != ofproto) { | |
2381 | struct mac_entry *e; | |
2382 | ||
2383 | e = mac_learning_lookup(o->ml, mac->mac, mac->vlan, | |
2384 | NULL); | |
2385 | if (e) { | |
2386 | mac_learning_expire(o->ml, e); | |
2387 | } | |
2388 | } | |
2389 | } | |
2390 | } | |
2391 | ||
2392 | mac_learning_expire(ml, mac); | |
2393 | } | |
2394 | } | |
2395 | } | |
2396 | ||
2397 | static struct ofbundle * | |
2398 | bundle_lookup(const struct ofproto_dpif *ofproto, void *aux) | |
2399 | { | |
2400 | struct ofbundle *bundle; | |
2401 | ||
2402 | HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0), | |
2403 | &ofproto->bundles) { | |
2404 | if (bundle->aux == aux) { | |
2405 | return bundle; | |
2406 | } | |
2407 | } | |
2408 | return NULL; | |
2409 | } | |
2410 | ||
2411 | /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the | |
2412 | * ones that are found to 'bundles'. */ | |
2413 | static void | |
2414 | bundle_lookup_multiple(struct ofproto_dpif *ofproto, | |
2415 | void **auxes, size_t n_auxes, | |
2416 | struct hmapx *bundles) | |
2417 | { | |
2418 | size_t i; | |
2419 | ||
2420 | hmapx_init(bundles); | |
2421 | for (i = 0; i < n_auxes; i++) { | |
2422 | struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]); | |
2423 | if (bundle) { | |
2424 | hmapx_add(bundles, bundle); | |
2425 | } | |
2426 | } | |
2427 | } | |
2428 | ||
2429 | static void | |
2430 | bundle_update(struct ofbundle *bundle) | |
2431 | { | |
2432 | struct ofport_dpif *port; | |
2433 | ||
2434 | bundle->floodable = true; | |
2435 | LIST_FOR_EACH (port, bundle_node, &bundle->ports) { | |
2436 | if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD | |
2437 | || !stp_forward_in_state(port->stp_state)) { | |
2438 | bundle->floodable = false; | |
2439 | break; | |
2440 | } | |
2441 | } | |
2442 | } | |
2443 | ||
2444 | static void | |
2445 | bundle_del_port(struct ofport_dpif *port) | |
2446 | { | |
2447 | struct ofbundle *bundle = port->bundle; | |
2448 | ||
2449 | bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE; | |
2450 | ||
2451 | list_remove(&port->bundle_node); | |
2452 | port->bundle = NULL; | |
2453 | ||
2454 | if (bundle->lacp) { | |
2455 | lacp_slave_unregister(bundle->lacp, port); | |
2456 | } | |
2457 | if (bundle->bond) { | |
2458 | bond_slave_unregister(bundle->bond, port); | |
2459 | } | |
2460 | ||
2461 | bundle_update(bundle); | |
2462 | } | |
2463 | ||
2464 | static bool | |
2465 | bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port, | |
2466 | struct lacp_slave_settings *lacp) | |
2467 | { | |
2468 | struct ofport_dpif *port; | |
2469 | ||
2470 | port = get_ofp_port(bundle->ofproto, ofp_port); | |
2471 | if (!port) { | |
2472 | return false; | |
2473 | } | |
2474 | ||
2475 | if (port->bundle != bundle) { | |
2476 | bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE; | |
2477 | if (port->bundle) { | |
2478 | bundle_del_port(port); | |
2479 | } | |
2480 | ||
2481 | port->bundle = bundle; | |
2482 | list_push_back(&bundle->ports, &port->bundle_node); | |
2483 | if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD | |
2484 | || !stp_forward_in_state(port->stp_state)) { | |
2485 | bundle->floodable = false; | |
2486 | } | |
2487 | } | |
2488 | if (lacp) { | |
2489 | bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE; | |
2490 | lacp_slave_register(bundle->lacp, port, lacp); | |
2491 | } | |
2492 | ||
2493 | return true; | |
2494 | } | |
2495 | ||
2496 | static void | |
2497 | bundle_destroy(struct ofbundle *bundle) | |
2498 | { | |
2499 | struct ofproto_dpif *ofproto; | |
2500 | struct ofport_dpif *port, *next_port; | |
2501 | int i; | |
2502 | ||
2503 | if (!bundle) { | |
2504 | return; | |
2505 | } | |
2506 | ||
2507 | ofproto = bundle->ofproto; | |
2508 | for (i = 0; i < MAX_MIRRORS; i++) { | |
2509 | struct ofmirror *m = ofproto->mirrors[i]; | |
2510 | if (m) { | |
2511 | if (m->out == bundle) { | |
2512 | mirror_destroy(m); | |
2513 | } else if (hmapx_find_and_delete(&m->srcs, bundle) | |
2514 | || hmapx_find_and_delete(&m->dsts, bundle)) { | |
2515 | ofproto->backer->need_revalidate = REV_RECONFIGURE; | |
2516 | } | |
2517 | } | |
2518 | } | |
2519 | ||
2520 | LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) { | |
2521 | bundle_del_port(port); | |
2522 | } | |
2523 | ||
2524 | bundle_flush_macs(bundle, true); | |
2525 | hmap_remove(&ofproto->bundles, &bundle->hmap_node); | |
2526 | free(bundle->name); | |
2527 | free(bundle->trunks); | |
2528 | lacp_destroy(bundle->lacp); | |
2529 | bond_destroy(bundle->bond); | |
2530 | free(bundle); | |
2531 | } | |
2532 | ||
2533 | static int | |
2534 | bundle_set(struct ofproto *ofproto_, void *aux, | |
2535 | const struct ofproto_bundle_settings *s) | |
2536 | { | |
2537 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2538 | bool need_flush = false; | |
2539 | struct ofport_dpif *port; | |
2540 | struct ofbundle *bundle; | |
2541 | unsigned long *trunks; | |
2542 | int vlan; | |
2543 | size_t i; | |
2544 | bool ok; | |
2545 | ||
2546 | if (!s) { | |
2547 | bundle_destroy(bundle_lookup(ofproto, aux)); | |
2548 | return 0; | |
2549 | } | |
2550 | ||
2551 | ovs_assert(s->n_slaves == 1 || s->bond != NULL); | |
2552 | ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL)); | |
2553 | ||
2554 | bundle = bundle_lookup(ofproto, aux); | |
2555 | if (!bundle) { | |
2556 | bundle = xmalloc(sizeof *bundle); | |
2557 | ||
2558 | bundle->ofproto = ofproto; | |
2559 | hmap_insert(&ofproto->bundles, &bundle->hmap_node, | |
2560 | hash_pointer(aux, 0)); | |
2561 | bundle->aux = aux; | |
2562 | bundle->name = NULL; | |
2563 | ||
2564 | list_init(&bundle->ports); | |
2565 | bundle->vlan_mode = PORT_VLAN_TRUNK; | |
2566 | bundle->vlan = -1; | |
2567 | bundle->trunks = NULL; | |
2568 | bundle->use_priority_tags = s->use_priority_tags; | |
2569 | bundle->lacp = NULL; | |
2570 | bundle->bond = NULL; | |
2571 | ||
2572 | bundle->floodable = true; | |
2573 | ||
2574 | bundle->src_mirrors = 0; | |
2575 | bundle->dst_mirrors = 0; | |
2576 | bundle->mirror_out = 0; | |
2577 | } | |
2578 | ||
2579 | if (!bundle->name || strcmp(s->name, bundle->name)) { | |
2580 | free(bundle->name); | |
2581 | bundle->name = xstrdup(s->name); | |
2582 | } | |
2583 | ||
2584 | /* LACP. */ | |
2585 | if (s->lacp) { | |
2586 | if (!bundle->lacp) { | |
2587 | ofproto->backer->need_revalidate = REV_RECONFIGURE; | |
2588 | bundle->lacp = lacp_create(); | |
2589 | } | |
2590 | lacp_configure(bundle->lacp, s->lacp); | |
2591 | } else { | |
2592 | lacp_destroy(bundle->lacp); | |
2593 | bundle->lacp = NULL; | |
2594 | } | |
2595 | ||
2596 | /* Update set of ports. */ | |
2597 | ok = true; | |
2598 | for (i = 0; i < s->n_slaves; i++) { | |
2599 | if (!bundle_add_port(bundle, s->slaves[i], | |
2600 | s->lacp ? &s->lacp_slaves[i] : NULL)) { | |
2601 | ok = false; | |
2602 | } | |
2603 | } | |
2604 | if (!ok || list_size(&bundle->ports) != s->n_slaves) { | |
2605 | struct ofport_dpif *next_port; | |
2606 | ||
2607 | LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) { | |
2608 | for (i = 0; i < s->n_slaves; i++) { | |
2609 | if (s->slaves[i] == port->up.ofp_port) { | |
2610 | goto found; | |
2611 | } | |
2612 | } | |
2613 | ||
2614 | bundle_del_port(port); | |
2615 | found: ; | |
2616 | } | |
2617 | } | |
2618 | ovs_assert(list_size(&bundle->ports) <= s->n_slaves); | |
2619 | ||
2620 | if (list_is_empty(&bundle->ports)) { | |
2621 | bundle_destroy(bundle); | |
2622 | return EINVAL; | |
2623 | } | |
2624 | ||
2625 | /* Set VLAN tagging mode */ | |
2626 | if (s->vlan_mode != bundle->vlan_mode | |
2627 | || s->use_priority_tags != bundle->use_priority_tags) { | |
2628 | bundle->vlan_mode = s->vlan_mode; | |
2629 | bundle->use_priority_tags = s->use_priority_tags; | |
2630 | need_flush = true; | |
2631 | } | |
2632 | ||
2633 | /* Set VLAN tag. */ | |
2634 | vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1 | |
2635 | : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan | |
2636 | : 0); | |
2637 | if (vlan != bundle->vlan) { | |
2638 | bundle->vlan = vlan; | |
2639 | need_flush = true; | |
2640 | } | |
2641 | ||
2642 | /* Get trunked VLANs. */ | |
2643 | switch (s->vlan_mode) { | |
2644 | case PORT_VLAN_ACCESS: | |
2645 | trunks = NULL; | |
2646 | break; | |
2647 | ||
2648 | case PORT_VLAN_TRUNK: | |
2649 | trunks = CONST_CAST(unsigned long *, s->trunks); | |
2650 | break; | |
2651 | ||
2652 | case PORT_VLAN_NATIVE_UNTAGGED: | |
2653 | case PORT_VLAN_NATIVE_TAGGED: | |
2654 | if (vlan != 0 && (!s->trunks | |
2655 | || !bitmap_is_set(s->trunks, vlan) | |
2656 | || bitmap_is_set(s->trunks, 0))) { | |
2657 | /* Force trunking the native VLAN and prohibit trunking VLAN 0. */ | |
2658 | if (s->trunks) { | |
2659 | trunks = bitmap_clone(s->trunks, 4096); | |
2660 | } else { | |
2661 | trunks = bitmap_allocate1(4096); | |
2662 | } | |
2663 | bitmap_set1(trunks, vlan); | |
2664 | bitmap_set0(trunks, 0); | |
2665 | } else { | |
2666 | trunks = CONST_CAST(unsigned long *, s->trunks); | |
2667 | } | |
2668 | break; | |
2669 | ||
2670 | default: | |
2671 | NOT_REACHED(); | |
2672 | } | |
2673 | if (!vlan_bitmap_equal(trunks, bundle->trunks)) { | |
2674 | free(bundle->trunks); | |
2675 | if (trunks == s->trunks) { | |
2676 | bundle->trunks = vlan_bitmap_clone(trunks); | |
2677 | } else { | |
2678 | bundle->trunks = trunks; | |
2679 | trunks = NULL; | |
2680 | } | |
2681 | need_flush = true; | |
2682 | } | |
2683 | if (trunks != s->trunks) { | |
2684 | free(trunks); | |
2685 | } | |
2686 | ||
2687 | /* Bonding. */ | |
2688 | if (!list_is_short(&bundle->ports)) { | |
2689 | bundle->ofproto->has_bonded_bundles = true; | |
2690 | if (bundle->bond) { | |
2691 | if (bond_reconfigure(bundle->bond, s->bond)) { | |
2692 | ofproto->backer->need_revalidate = REV_RECONFIGURE; | |
2693 | } | |
2694 | } else { | |
2695 | bundle->bond = bond_create(s->bond); | |
2696 | ofproto->backer->need_revalidate = REV_RECONFIGURE; | |
2697 | } | |
2698 | ||
2699 | LIST_FOR_EACH (port, bundle_node, &bundle->ports) { | |
2700 | bond_slave_register(bundle->bond, port, port->up.netdev); | |
2701 | } | |
2702 | } else { | |
2703 | bond_destroy(bundle->bond); | |
2704 | bundle->bond = NULL; | |
2705 | } | |
2706 | ||
2707 | /* If we changed something that would affect MAC learning, un-learn | |
2708 | * everything on this port and force flow revalidation. */ | |
2709 | if (need_flush) { | |
2710 | bundle_flush_macs(bundle, false); | |
2711 | } | |
2712 | ||
2713 | return 0; | |
2714 | } | |
2715 | ||
2716 | static void | |
2717 | bundle_remove(struct ofport *port_) | |
2718 | { | |
2719 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
2720 | struct ofbundle *bundle = port->bundle; | |
2721 | ||
2722 | if (bundle) { | |
2723 | bundle_del_port(port); | |
2724 | if (list_is_empty(&bundle->ports)) { | |
2725 | bundle_destroy(bundle); | |
2726 | } else if (list_is_short(&bundle->ports)) { | |
2727 | bond_destroy(bundle->bond); | |
2728 | bundle->bond = NULL; | |
2729 | } | |
2730 | } | |
2731 | } | |
2732 | ||
2733 | static void | |
2734 | send_pdu_cb(void *port_, const void *pdu, size_t pdu_size) | |
2735 | { | |
2736 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10); | |
2737 | struct ofport_dpif *port = port_; | |
2738 | uint8_t ea[ETH_ADDR_LEN]; | |
2739 | int error; | |
2740 | ||
2741 | error = netdev_get_etheraddr(port->up.netdev, ea); | |
2742 | if (!error) { | |
2743 | struct ofpbuf packet; | |
2744 | void *packet_pdu; | |
2745 | ||
2746 | ofpbuf_init(&packet, 0); | |
2747 | packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP, | |
2748 | pdu_size); | |
2749 | memcpy(packet_pdu, pdu, pdu_size); | |
2750 | ||
2751 | send_packet(port, &packet); | |
2752 | ofpbuf_uninit(&packet); | |
2753 | } else { | |
2754 | VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface " | |
2755 | "%s (%s)", port->bundle->name, | |
2756 | netdev_get_name(port->up.netdev), strerror(error)); | |
2757 | } | |
2758 | } | |
2759 | ||
2760 | static void | |
2761 | bundle_send_learning_packets(struct ofbundle *bundle) | |
2762 | { | |
2763 | struct ofproto_dpif *ofproto = bundle->ofproto; | |
2764 | int error, n_packets, n_errors; | |
2765 | struct mac_entry *e; | |
2766 | ||
2767 | error = n_packets = n_errors = 0; | |
2768 | LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) { | |
2769 | if (e->port.p != bundle) { | |
2770 | struct ofpbuf *learning_packet; | |
2771 | struct ofport_dpif *port; | |
2772 | void *port_void; | |
2773 | int ret; | |
2774 | ||
2775 | /* The assignment to "port" is unnecessary but makes "grep"ing for | |
2776 | * struct ofport_dpif more effective. */ | |
2777 | learning_packet = bond_compose_learning_packet(bundle->bond, | |
2778 | e->mac, e->vlan, | |
2779 | &port_void); | |
2780 | port = port_void; | |
2781 | ret = send_packet(port, learning_packet); | |
2782 | ofpbuf_delete(learning_packet); | |
2783 | if (ret) { | |
2784 | error = ret; | |
2785 | n_errors++; | |
2786 | } | |
2787 | n_packets++; | |
2788 | } | |
2789 | } | |
2790 | ||
2791 | if (n_errors) { | |
2792 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
2793 | VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning " | |
2794 | "packets, last error was: %s", | |
2795 | bundle->name, n_errors, n_packets, strerror(error)); | |
2796 | } else { | |
2797 | VLOG_DBG("bond %s: sent %d gratuitous learning packets", | |
2798 | bundle->name, n_packets); | |
2799 | } | |
2800 | } | |
2801 | ||
2802 | static void | |
2803 | bundle_run(struct ofbundle *bundle) | |
2804 | { | |
2805 | if (bundle->lacp) { | |
2806 | lacp_run(bundle->lacp, send_pdu_cb); | |
2807 | } | |
2808 | if (bundle->bond) { | |
2809 | struct ofport_dpif *port; | |
2810 | ||
2811 | LIST_FOR_EACH (port, bundle_node, &bundle->ports) { | |
2812 | bond_slave_set_may_enable(bundle->bond, port, port->may_enable); | |
2813 | } | |
2814 | ||
2815 | bond_run(bundle->bond, &bundle->ofproto->backer->revalidate_set, | |
2816 | lacp_status(bundle->lacp)); | |
2817 | if (bond_should_send_learning_packets(bundle->bond)) { | |
2818 | bundle_send_learning_packets(bundle); | |
2819 | } | |
2820 | } | |
2821 | } | |
2822 | ||
2823 | static void | |
2824 | bundle_wait(struct ofbundle *bundle) | |
2825 | { | |
2826 | if (bundle->lacp) { | |
2827 | lacp_wait(bundle->lacp); | |
2828 | } | |
2829 | if (bundle->bond) { | |
2830 | bond_wait(bundle->bond); | |
2831 | } | |
2832 | } | |
2833 | \f | |
2834 | /* Mirrors. */ | |
2835 | ||
2836 | static int | |
2837 | mirror_scan(struct ofproto_dpif *ofproto) | |
2838 | { | |
2839 | int idx; | |
2840 | ||
2841 | for (idx = 0; idx < MAX_MIRRORS; idx++) { | |
2842 | if (!ofproto->mirrors[idx]) { | |
2843 | return idx; | |
2844 | } | |
2845 | } | |
2846 | return -1; | |
2847 | } | |
2848 | ||
2849 | static struct ofmirror * | |
2850 | mirror_lookup(struct ofproto_dpif *ofproto, void *aux) | |
2851 | { | |
2852 | int i; | |
2853 | ||
2854 | for (i = 0; i < MAX_MIRRORS; i++) { | |
2855 | struct ofmirror *mirror = ofproto->mirrors[i]; | |
2856 | if (mirror && mirror->aux == aux) { | |
2857 | return mirror; | |
2858 | } | |
2859 | } | |
2860 | ||
2861 | return NULL; | |
2862 | } | |
2863 | ||
2864 | /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */ | |
2865 | static void | |
2866 | mirror_update_dups(struct ofproto_dpif *ofproto) | |
2867 | { | |
2868 | int i; | |
2869 | ||
2870 | for (i = 0; i < MAX_MIRRORS; i++) { | |
2871 | struct ofmirror *m = ofproto->mirrors[i]; | |
2872 | ||
2873 | if (m) { | |
2874 | m->dup_mirrors = MIRROR_MASK_C(1) << i; | |
2875 | } | |
2876 | } | |
2877 | ||
2878 | for (i = 0; i < MAX_MIRRORS; i++) { | |
2879 | struct ofmirror *m1 = ofproto->mirrors[i]; | |
2880 | int j; | |
2881 | ||
2882 | if (!m1) { | |
2883 | continue; | |
2884 | } | |
2885 | ||
2886 | for (j = i + 1; j < MAX_MIRRORS; j++) { | |
2887 | struct ofmirror *m2 = ofproto->mirrors[j]; | |
2888 | ||
2889 | if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) { | |
2890 | m1->dup_mirrors |= MIRROR_MASK_C(1) << j; | |
2891 | m2->dup_mirrors |= m1->dup_mirrors; | |
2892 | } | |
2893 | } | |
2894 | } | |
2895 | } | |
2896 | ||
2897 | static int | |
2898 | mirror_set(struct ofproto *ofproto_, void *aux, | |
2899 | const struct ofproto_mirror_settings *s) | |
2900 | { | |
2901 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2902 | mirror_mask_t mirror_bit; | |
2903 | struct ofbundle *bundle; | |
2904 | struct ofmirror *mirror; | |
2905 | struct ofbundle *out; | |
2906 | struct hmapx srcs; /* Contains "struct ofbundle *"s. */ | |
2907 | struct hmapx dsts; /* Contains "struct ofbundle *"s. */ | |
2908 | int out_vlan; | |
2909 | ||
2910 | mirror = mirror_lookup(ofproto, aux); | |
2911 | if (!s) { | |
2912 | mirror_destroy(mirror); | |
2913 | return 0; | |
2914 | } | |
2915 | if (!mirror) { | |
2916 | int idx; | |
2917 | ||
2918 | idx = mirror_scan(ofproto); | |
2919 | if (idx < 0) { | |
2920 | VLOG_WARN("bridge %s: maximum of %d port mirrors reached, " | |
2921 | "cannot create %s", | |
2922 | ofproto->up.name, MAX_MIRRORS, s->name); | |
2923 | return EFBIG; | |
2924 | } | |
2925 | ||
2926 | mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror); | |
2927 | mirror->ofproto = ofproto; | |
2928 | mirror->idx = idx; | |
2929 | mirror->aux = aux; | |
2930 | mirror->out_vlan = -1; | |
2931 | mirror->name = NULL; | |
2932 | } | |
2933 | ||
2934 | if (!mirror->name || strcmp(s->name, mirror->name)) { | |
2935 | free(mirror->name); | |
2936 | mirror->name = xstrdup(s->name); | |
2937 | } | |
2938 | ||
2939 | /* Get the new configuration. */ | |
2940 | if (s->out_bundle) { | |
2941 | out = bundle_lookup(ofproto, s->out_bundle); | |
2942 | if (!out) { | |
2943 | mirror_destroy(mirror); | |
2944 | return EINVAL; | |
2945 | } | |
2946 | out_vlan = -1; | |
2947 | } else { | |
2948 | out = NULL; | |
2949 | out_vlan = s->out_vlan; | |
2950 | } | |
2951 | bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs); | |
2952 | bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts); | |
2953 | ||
2954 | /* If the configuration has not changed, do nothing. */ | |
2955 | if (hmapx_equals(&srcs, &mirror->srcs) | |
2956 | && hmapx_equals(&dsts, &mirror->dsts) | |
2957 | && vlan_bitmap_equal(mirror->vlans, s->src_vlans) | |
2958 | && mirror->out == out | |
2959 | && mirror->out_vlan == out_vlan) | |
2960 | { | |
2961 | hmapx_destroy(&srcs); | |
2962 | hmapx_destroy(&dsts); | |
2963 | return 0; | |
2964 | } | |
2965 | ||
2966 | hmapx_swap(&srcs, &mirror->srcs); | |
2967 | hmapx_destroy(&srcs); | |
2968 | ||
2969 | hmapx_swap(&dsts, &mirror->dsts); | |
2970 | hmapx_destroy(&dsts); | |
2971 | ||
2972 | free(mirror->vlans); | |
2973 | mirror->vlans = vlan_bitmap_clone(s->src_vlans); | |
2974 | ||
2975 | mirror->out = out; | |
2976 | mirror->out_vlan = out_vlan; | |
2977 | ||
2978 | /* Update bundles. */ | |
2979 | mirror_bit = MIRROR_MASK_C(1) << mirror->idx; | |
2980 | HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) { | |
2981 | if (hmapx_contains(&mirror->srcs, bundle)) { | |
2982 | bundle->src_mirrors |= mirror_bit; | |
2983 | } else { | |
2984 | bundle->src_mirrors &= ~mirror_bit; | |
2985 | } | |
2986 | ||
2987 | if (hmapx_contains(&mirror->dsts, bundle)) { | |
2988 | bundle->dst_mirrors |= mirror_bit; | |
2989 | } else { | |
2990 | bundle->dst_mirrors &= ~mirror_bit; | |
2991 | } | |
2992 | ||
2993 | if (mirror->out == bundle) { | |
2994 | bundle->mirror_out |= mirror_bit; | |
2995 | } else { | |
2996 | bundle->mirror_out &= ~mirror_bit; | |
2997 | } | |
2998 | } | |
2999 | ||
3000 | ofproto->backer->need_revalidate = REV_RECONFIGURE; | |
3001 | ofproto->has_mirrors = true; | |
3002 | mac_learning_flush(ofproto->ml, | |
3003 | &ofproto->backer->revalidate_set); | |
3004 | mirror_update_dups(ofproto); | |
3005 | ||
3006 | return 0; | |
3007 | } | |
3008 | ||
3009 | static void | |
3010 | mirror_destroy(struct ofmirror *mirror) | |
3011 | { | |
3012 | struct ofproto_dpif *ofproto; | |
3013 | mirror_mask_t mirror_bit; | |
3014 | struct ofbundle *bundle; | |
3015 | int i; | |
3016 | ||
3017 | if (!mirror) { | |
3018 | return; | |
3019 | } | |
3020 | ||
3021 | ofproto = mirror->ofproto; | |
3022 | ofproto->backer->need_revalidate = REV_RECONFIGURE; | |
3023 | mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set); | |
3024 | ||
3025 | mirror_bit = MIRROR_MASK_C(1) << mirror->idx; | |
3026 | HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { | |
3027 | bundle->src_mirrors &= ~mirror_bit; | |
3028 | bundle->dst_mirrors &= ~mirror_bit; | |
3029 | bundle->mirror_out &= ~mirror_bit; | |
3030 | } | |
3031 | ||
3032 | hmapx_destroy(&mirror->srcs); | |
3033 | hmapx_destroy(&mirror->dsts); | |
3034 | free(mirror->vlans); | |
3035 | ||
3036 | ofproto->mirrors[mirror->idx] = NULL; | |
3037 | free(mirror->name); | |
3038 | free(mirror); | |
3039 | ||
3040 | mirror_update_dups(ofproto); | |
3041 | ||
3042 | ofproto->has_mirrors = false; | |
3043 | for (i = 0; i < MAX_MIRRORS; i++) { | |
3044 | if (ofproto->mirrors[i]) { | |
3045 | ofproto->has_mirrors = true; | |
3046 | break; | |
3047 | } | |
3048 | } | |
3049 | } | |
3050 | ||
3051 | static int | |
3052 | mirror_get_stats(struct ofproto *ofproto_, void *aux, | |
3053 | uint64_t *packets, uint64_t *bytes) | |
3054 | { | |
3055 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
3056 | struct ofmirror *mirror = mirror_lookup(ofproto, aux); | |
3057 | ||
3058 | if (!mirror) { | |
3059 | *packets = *bytes = UINT64_MAX; | |
3060 | return 0; | |
3061 | } | |
3062 | ||
3063 | push_all_stats(); | |
3064 | ||
3065 | *packets = mirror->packet_count; | |
3066 | *bytes = mirror->byte_count; | |
3067 | ||
3068 | return 0; | |
3069 | } | |
3070 | ||
3071 | static int | |
3072 | set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans) | |
3073 | { | |
3074 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
3075 | if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) { | |
3076 | mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set); | |
3077 | } | |
3078 | return 0; | |
3079 | } | |
3080 | ||
3081 | static bool | |
3082 | is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux) | |
3083 | { | |
3084 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
3085 | struct ofbundle *bundle = bundle_lookup(ofproto, aux); | |
3086 | return bundle && bundle->mirror_out != 0; | |
3087 | } | |
3088 | ||
3089 | static void | |
3090 | forward_bpdu_changed(struct ofproto *ofproto_) | |
3091 | { | |
3092 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
3093 | ofproto->backer->need_revalidate = REV_RECONFIGURE; | |
3094 | } | |
3095 | ||
3096 | static void | |
3097 | set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time, | |
3098 | size_t max_entries) | |
3099 | { | |
3100 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
3101 | mac_learning_set_idle_time(ofproto->ml, idle_time); | |
3102 | mac_learning_set_max_entries(ofproto->ml, max_entries); | |
3103 | } | |
3104 | \f | |
3105 | /* Ports. */ | |
3106 | ||
3107 | static struct ofport_dpif * | |
3108 | get_ofp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port) | |
3109 | { | |
3110 | struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port); | |
3111 | return ofport ? ofport_dpif_cast(ofport) : NULL; | |
3112 | } | |
3113 | ||
3114 | static struct ofport_dpif * | |
3115 | get_odp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port) | |
3116 | { | |
3117 | struct ofport_dpif *port = odp_port_to_ofport(ofproto->backer, odp_port); | |
3118 | return port && &ofproto->up == port->up.ofproto ? port : NULL; | |
3119 | } | |
3120 | ||
3121 | static void | |
3122 | ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto, | |
3123 | struct ofproto_port *ofproto_port, | |
3124 | struct dpif_port *dpif_port) | |
3125 | { | |
3126 | ofproto_port->name = dpif_port->name; | |
3127 | ofproto_port->type = dpif_port->type; | |
3128 | ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no); | |
3129 | } | |
3130 | ||
3131 | static struct ofport_dpif * | |
3132 | ofport_get_peer(const struct ofport_dpif *ofport_dpif) | |
3133 | { | |
3134 | const struct ofproto_dpif *ofproto; | |
3135 | const char *peer; | |
3136 | ||
3137 | peer = netdev_vport_patch_peer(ofport_dpif->up.netdev); | |
3138 | if (!peer) { | |
3139 | return NULL; | |
3140 | } | |
3141 | ||
3142 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
3143 | struct ofport *ofport; | |
3144 | ||
3145 | ofport = shash_find_data(&ofproto->up.port_by_name, peer); | |
3146 | if (ofport && ofport->ofproto->ofproto_class == &ofproto_dpif_class) { | |
3147 | return ofport_dpif_cast(ofport); | |
3148 | } | |
3149 | } | |
3150 | return NULL; | |
3151 | } | |
3152 | ||
3153 | static void | |
3154 | port_run_fast(struct ofport_dpif *ofport) | |
3155 | { | |
3156 | if (ofport->cfm && cfm_should_send_ccm(ofport->cfm)) { | |
3157 | struct ofpbuf packet; | |
3158 | ||
3159 | ofpbuf_init(&packet, 0); | |
3160 | cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr); | |
3161 | send_packet(ofport, &packet); | |
3162 | ofpbuf_uninit(&packet); | |
3163 | } | |
3164 | ||
3165 | if (ofport->bfd && bfd_should_send_packet(ofport->bfd)) { | |
3166 | struct ofpbuf packet; | |
3167 | ||
3168 | ofpbuf_init(&packet, 0); | |
3169 | bfd_put_packet(ofport->bfd, &packet, ofport->up.pp.hw_addr); | |
3170 | send_packet(ofport, &packet); | |
3171 | ofpbuf_uninit(&packet); | |
3172 | } | |
3173 | } | |
3174 | ||
3175 | static void | |
3176 | port_run(struct ofport_dpif *ofport) | |
3177 | { | |
3178 | long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev); | |
3179 | bool carrier_changed = carrier_seq != ofport->carrier_seq; | |
3180 | bool enable = netdev_get_carrier(ofport->up.netdev); | |
3181 | ||
3182 | ofport->carrier_seq = carrier_seq; | |
3183 | ||
3184 | port_run_fast(ofport); | |
3185 | ||
3186 | if (ofport->tnl_port | |
3187 | && tnl_port_reconfigure(&ofport->up, ofport->odp_port, | |
3188 | &ofport->tnl_port)) { | |
3189 | ofproto_dpif_cast(ofport->up.ofproto)->backer->need_revalidate = true; | |
3190 | } | |
3191 | ||
3192 | if (ofport->cfm) { | |
3193 | int cfm_opup = cfm_get_opup(ofport->cfm); | |
3194 | ||
3195 | cfm_run(ofport->cfm); | |
3196 | enable = enable && !cfm_get_fault(ofport->cfm); | |
3197 | ||
3198 | if (cfm_opup >= 0) { | |
3199 | enable = enable && cfm_opup; | |
3200 | } | |
3201 | } | |
3202 | ||
3203 | if (ofport->bfd) { | |
3204 | bfd_run(ofport->bfd); | |
3205 | enable = enable && bfd_forwarding(ofport->bfd); | |
3206 | } | |
3207 | ||
3208 | if (ofport->bundle) { | |
3209 | enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport); | |
3210 | if (carrier_changed) { | |
3211 | lacp_slave_carrier_changed(ofport->bundle->lacp, ofport); | |
3212 | } | |
3213 | } | |
3214 | ||
3215 | if (ofport->may_enable != enable) { | |
3216 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
3217 | ||
3218 | if (ofproto->has_bundle_action) { | |
3219 | ofproto->backer->need_revalidate = REV_PORT_TOGGLED; | |
3220 | } | |
3221 | } | |
3222 | ||
3223 | ofport->may_enable = enable; | |
3224 | } | |
3225 | ||
3226 | static void | |
3227 | port_wait(struct ofport_dpif *ofport) | |
3228 | { | |
3229 | if (ofport->cfm) { | |
3230 | cfm_wait(ofport->cfm); | |
3231 | } | |
3232 | ||
3233 | if (ofport->bfd) { | |
3234 | bfd_wait(ofport->bfd); | |
3235 | } | |
3236 | } | |
3237 | ||
3238 | static int | |
3239 | port_query_by_name(const struct ofproto *ofproto_, const char *devname, | |
3240 | struct ofproto_port *ofproto_port) | |
3241 | { | |
3242 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
3243 | struct dpif_port dpif_port; | |
3244 | int error; | |
3245 | ||
3246 | if (sset_contains(&ofproto->ghost_ports, devname)) { | |
3247 | const char *type = netdev_get_type_from_name(devname); | |
3248 | ||
3249 | /* We may be called before ofproto->up.port_by_name is populated with | |
3250 | * the appropriate ofport. For this reason, we must get the name and | |
3251 | * type from the netdev layer directly. */ | |
3252 | if (type) { | |
3253 | const struct ofport *ofport; | |
3254 | ||
3255 | ofport = shash_find_data(&ofproto->up.port_by_name, devname); | |
3256 | ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE; | |
3257 | ofproto_port->name = xstrdup(devname); | |
3258 | ofproto_port->type = xstrdup(type); | |
3259 | return 0; | |
3260 | } | |
3261 | return ENODEV; | |
3262 | } | |
3263 | ||
3264 | if (!sset_contains(&ofproto->ports, devname)) { | |
3265 | return ENODEV; | |
3266 | } | |
3267 | error = dpif_port_query_by_name(ofproto->backer->dpif, | |
3268 | devname, &dpif_port); | |
3269 | if (!error) { | |
3270 | ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port); | |
3271 | } | |
3272 | return error; | |
3273 | } | |
3274 | ||
3275 | static int | |
3276 | port_add(struct ofproto *ofproto_, struct netdev *netdev) | |
3277 | { | |
3278 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
3279 | const char *dp_port_name = netdev_vport_get_dpif_port(netdev); | |
3280 | const char *devname = netdev_get_name(netdev); | |
3281 | ||
3282 | if (netdev_vport_is_patch(netdev)) { | |
3283 | sset_add(&ofproto->ghost_ports, netdev_get_name(netdev)); | |
3284 | return 0; | |
3285 | } | |
3286 | ||
3287 | if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) { | |
3288 | uint32_t port_no = UINT32_MAX; | |
3289 | int error; | |
3290 | ||
3291 | error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no); | |
3292 | if (error) { | |
3293 | return error; | |
3294 | } | |
3295 | if (netdev_get_tunnel_config(netdev)) { | |
3296 | simap_put(&ofproto->backer->tnl_backers, dp_port_name, port_no); | |
3297 | } | |
3298 | } | |
3299 | ||
3300 | if (netdev_get_tunnel_config(netdev)) { | |
3301 | sset_add(&ofproto->ghost_ports, devname); | |
3302 | } else { | |
3303 | sset_add(&ofproto->ports, devname); | |
3304 | } | |
3305 | return 0; | |
3306 | } | |
3307 | ||
3308 | static int | |
3309 | port_del(struct ofproto *ofproto_, uint16_t ofp_port) | |
3310 | { | |
3311 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
3312 | struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port); | |
3313 | int error = 0; | |
3314 | ||
3315 | if (!ofport) { | |
3316 | return 0; | |
3317 | } | |
3318 | ||
3319 | sset_find_and_delete(&ofproto->ghost_ports, | |
3320 | netdev_get_name(ofport->up.netdev)); | |
3321 | ofproto->backer->need_revalidate = REV_RECONFIGURE; | |
3322 | if (!ofport->tnl_port) { | |
3323 | error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port); | |
3324 | if (!error) { | |
3325 | /* The caller is going to close ofport->up.netdev. If this is a | |
3326 | * bonded port, then the bond is using that netdev, so remove it | |
3327 | * from the bond. The client will need to reconfigure everything | |
3328 | * after deleting ports, so then the slave will get re-added. */ | |
3329 | bundle_remove(&ofport->up); | |
3330 | } | |
3331 | } | |
3332 | return error; | |
3333 | } | |
3334 | ||
3335 | static int | |
3336 | port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats) | |
3337 | { | |
3338 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
3339 | int error; | |
3340 | ||
3341 | push_all_stats(); | |
3342 | ||
3343 | error = netdev_get_stats(ofport->up.netdev, stats); | |
3344 | ||
3345 | if (!error && ofport_->ofp_port == OFPP_LOCAL) { | |
3346 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
3347 | ||
3348 | /* ofproto->stats.tx_packets represents packets that we created | |
3349 | * internally and sent to some port (e.g. packets sent with | |
3350 | * send_packet()). Account for them as if they had come from | |
3351 | * OFPP_LOCAL and got forwarded. */ | |
3352 | ||
3353 | if (stats->rx_packets != UINT64_MAX) { | |
3354 | stats->rx_packets += ofproto->stats.tx_packets; | |
3355 | } | |
3356 | ||
3357 | if (stats->rx_bytes != UINT64_MAX) { | |
3358 | stats->rx_bytes += ofproto->stats.tx_bytes; | |
3359 | } | |
3360 | ||
3361 | /* ofproto->stats.rx_packets represents packets that were received on | |
3362 | * some port and we processed internally and dropped (e.g. STP). | |
3363 | * Account for them as if they had been forwarded to OFPP_LOCAL. */ | |
3364 | ||
3365 | if (stats->tx_packets != UINT64_MAX) { | |
3366 | stats->tx_packets += ofproto->stats.rx_packets; | |
3367 | } | |
3368 | ||
3369 | if (stats->tx_bytes != UINT64_MAX) { | |
3370 | stats->tx_bytes += ofproto->stats.rx_bytes; | |
3371 | } | |
3372 | } | |
3373 | ||
3374 | return error; | |
3375 | } | |
3376 | ||
3377 | struct port_dump_state { | |
3378 | uint32_t bucket; | |
3379 | uint32_t offset; | |
3380 | bool ghost; | |
3381 | ||
3382 | struct ofproto_port port; | |
3383 | bool has_port; | |
3384 | }; | |
3385 | ||
3386 | static int | |
3387 | port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep) | |
3388 | { | |
3389 | *statep = xzalloc(sizeof(struct port_dump_state)); | |
3390 | return 0; | |
3391 | } | |
3392 | ||
3393 | static int | |
3394 | port_dump_next(const struct ofproto *ofproto_, void *state_, | |
3395 | struct ofproto_port *port) | |
3396 | { | |
3397 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
3398 | struct port_dump_state *state = state_; | |
3399 | const struct sset *sset; | |
3400 | struct sset_node *node; | |
3401 | ||
3402 | if (state->has_port) { | |
3403 | ofproto_port_destroy(&state->port); | |
3404 | state->has_port = false; | |
3405 | } | |
3406 | sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports; | |
3407 | while ((node = sset_at_position(sset, &state->bucket, &state->offset))) { | |
3408 | int error; | |
3409 | ||
3410 | error = port_query_by_name(ofproto_, node->name, &state->port); | |
3411 | if (!error) { | |
3412 | *port = state->port; | |
3413 | state->has_port = true; | |
3414 | return 0; | |
3415 | } else if (error != ENODEV) { | |
3416 | return error; | |
3417 | } | |
3418 | } | |
3419 | ||
3420 | if (!state->ghost) { | |
3421 | state->ghost = true; | |
3422 | state->bucket = 0; | |
3423 | state->offset = 0; | |
3424 | return port_dump_next(ofproto_, state_, port); | |
3425 | } | |
3426 | ||
3427 | return EOF; | |
3428 | } | |
3429 | ||
3430 | static int | |
3431 | port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_) | |
3432 | { | |
3433 | struct port_dump_state *state = state_; | |
3434 | ||
3435 | if (state->has_port) { | |
3436 | ofproto_port_destroy(&state->port); | |
3437 | } | |
3438 | free(state); | |
3439 | return 0; | |
3440 | } | |
3441 | ||
3442 | static int | |
3443 | port_poll(const struct ofproto *ofproto_, char **devnamep) | |
3444 | { | |
3445 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
3446 | ||
3447 | if (ofproto->port_poll_errno) { | |
3448 | int error = ofproto->port_poll_errno; | |
3449 | ofproto->port_poll_errno = 0; | |
3450 | return error; | |
3451 | } | |
3452 | ||
3453 | if (sset_is_empty(&ofproto->port_poll_set)) { | |
3454 | return EAGAIN; | |
3455 | } | |
3456 | ||
3457 | *devnamep = sset_pop(&ofproto->port_poll_set); | |
3458 | return 0; | |
3459 | } | |
3460 | ||
3461 | static void | |
3462 | port_poll_wait(const struct ofproto *ofproto_) | |
3463 | { | |
3464 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
3465 | dpif_port_poll_wait(ofproto->backer->dpif); | |
3466 | } | |
3467 | ||
3468 | static int | |
3469 | port_is_lacp_current(const struct ofport *ofport_) | |
3470 | { | |
3471 | const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
3472 | return (ofport->bundle && ofport->bundle->lacp | |
3473 | ? lacp_slave_is_current(ofport->bundle->lacp, ofport) | |
3474 | : -1); | |
3475 | } | |
3476 | \f | |
3477 | /* Upcall handling. */ | |
3478 | ||
3479 | /* Flow miss batching. | |
3480 | * | |
3481 | * Some dpifs implement operations faster when you hand them off in a batch. | |
3482 | * To allow batching, "struct flow_miss" queues the dpif-related work needed | |
3483 | * for a given flow. Each "struct flow_miss" corresponds to sending one or | |
3484 | * more packets, plus possibly installing the flow in the dpif. | |
3485 | * | |
3486 | * So far we only batch the operations that affect flow setup time the most. | |
3487 | * It's possible to batch more than that, but the benefit might be minimal. */ | |
3488 | struct flow_miss { | |
3489 | struct hmap_node hmap_node; | |
3490 | struct ofproto_dpif *ofproto; | |
3491 | struct flow flow; | |
3492 | enum odp_key_fitness key_fitness; | |
3493 | const struct nlattr *key; | |
3494 | size_t key_len; | |
3495 | struct initial_vals initial_vals; | |
3496 | struct list packets; | |
3497 | enum dpif_upcall_type upcall_type; | |
3498 | uint32_t odp_in_port; | |
3499 | }; | |
3500 | ||
3501 | struct flow_miss_op { | |
3502 | struct dpif_op dpif_op; | |
3503 | void *garbage; /* Pointer to pass to free(), NULL if none. */ | |
3504 | uint64_t stub[1024 / 8]; /* Temporary buffer. */ | |
3505 | }; | |
3506 | ||
3507 | /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each | |
3508 | * OpenFlow controller as necessary according to their individual | |
3509 | * configurations. */ | |
3510 | static void | |
3511 | send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet, | |
3512 | const struct flow *flow) | |
3513 | { | |
3514 | struct ofputil_packet_in pin; | |
3515 | ||
3516 | pin.packet = packet->data; | |
3517 | pin.packet_len = packet->size; | |
3518 | pin.reason = OFPR_NO_MATCH; | |
3519 | pin.controller_id = 0; | |
3520 | ||
3521 | pin.table_id = 0; | |
3522 | pin.cookie = 0; | |
3523 | ||
3524 | pin.send_len = 0; /* not used for flow table misses */ | |
3525 | ||
3526 | flow_get_metadata(flow, &pin.fmd); | |
3527 | ||
3528 | connmgr_send_packet_in(ofproto->up.connmgr, &pin); | |
3529 | } | |
3530 | ||
3531 | static enum slow_path_reason | |
3532 | process_special(struct ofproto_dpif *ofproto, const struct flow *flow, | |
3533 | const struct ofport_dpif *ofport, const struct ofpbuf *packet) | |
3534 | { | |
3535 | if (!ofport) { | |
3536 | return 0; | |
3537 | } else if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) { | |
3538 | if (packet) { | |
3539 | cfm_process_heartbeat(ofport->cfm, packet); | |
3540 | } | |
3541 | return SLOW_CFM; | |
3542 | } else if (ofport->bfd && bfd_should_process_flow(flow)) { | |
3543 | if (packet) { | |
3544 | bfd_process_packet(ofport->bfd, flow, packet); | |
3545 | } | |
3546 | return SLOW_BFD; | |
3547 | } else if (ofport->bundle && ofport->bundle->lacp | |
3548 | && flow->dl_type == htons(ETH_TYPE_LACP)) { | |
3549 | if (packet) { | |
3550 | lacp_process_packet(ofport->bundle->lacp, ofport, packet); | |
3551 | } | |
3552 | return SLOW_LACP; | |
3553 | } else if (ofproto->stp && stp_should_process_flow(flow)) { | |
3554 | if (packet) { | |
3555 | stp_process_packet(ofport, packet); | |
3556 | } | |
3557 | return SLOW_STP; | |
3558 | } else { | |
3559 | return 0; | |
3560 | } | |
3561 | } | |
3562 | ||
3563 | static struct flow_miss * | |
3564 | flow_miss_find(struct hmap *todo, const struct ofproto_dpif *ofproto, | |
3565 | const struct flow *flow, uint32_t hash) | |
3566 | { | |
3567 | struct flow_miss *miss; | |
3568 | ||
3569 | HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) { | |
3570 | if (miss->ofproto == ofproto && flow_equal(&miss->flow, flow)) { | |
3571 | return miss; | |
3572 | } | |
3573 | } | |
3574 | ||
3575 | return NULL; | |
3576 | } | |
3577 | ||
3578 | /* Partially Initializes 'op' as an "execute" operation for 'miss' and | |
3579 | * 'packet'. The caller must initialize op->actions and op->actions_len. If | |
3580 | * 'miss' is associated with a subfacet the caller must also initialize the | |
3581 | * returned op->subfacet, and if anything needs to be freed after processing | |
3582 | * the op, the caller must initialize op->garbage also. */ | |
3583 | static void | |
3584 | init_flow_miss_execute_op(struct flow_miss *miss, struct ofpbuf *packet, | |
3585 | struct flow_miss_op *op) | |
3586 | { | |
3587 | if (miss->flow.vlan_tci != miss->initial_vals.vlan_tci) { | |
3588 | /* This packet was received on a VLAN splinter port. We | |
3589 | * added a VLAN to the packet to make the packet resemble | |
3590 | * the flow, but the actions were composed assuming that | |
3591 | * the packet contained no VLAN. So, we must remove the | |
3592 | * VLAN header from the packet before trying to execute the | |
3593 | * actions. */ | |
3594 | eth_pop_vlan(packet); | |
3595 | } | |
3596 | ||
3597 | op->garbage = NULL; | |
3598 | op->dpif_op.type = DPIF_OP_EXECUTE; | |
3599 | op->dpif_op.u.execute.key = miss->key; | |
3600 | op->dpif_op.u.execute.key_len = miss->key_len; | |
3601 | op->dpif_op.u.execute.packet = packet; | |
3602 | } | |
3603 | ||
3604 | /* Helper for handle_flow_miss_without_facet() and | |
3605 | * handle_flow_miss_with_facet(). */ | |
3606 | static void | |
3607 | handle_flow_miss_common(struct rule_dpif *rule, | |
3608 | struct ofpbuf *packet, const struct flow *flow) | |
3609 | { | |
3610 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
3611 | ||
3612 | ofproto->n_matches++; | |
3613 | ||
3614 | if (rule->up.cr.priority == FAIL_OPEN_PRIORITY) { | |
3615 | /* | |
3616 | * Extra-special case for fail-open mode. | |
3617 | * | |
3618 | * We are in fail-open mode and the packet matched the fail-open | |
3619 | * rule, but we are connected to a controller too. We should send | |
3620 | * the packet up to the controller in the hope that it will try to | |
3621 | * set up a flow and thereby allow us to exit fail-open. | |
3622 | * | |
3623 | * See the top-level comment in fail-open.c for more information. | |
3624 | */ | |
3625 | send_packet_in_miss(ofproto, packet, flow); | |
3626 | } | |
3627 | } | |
3628 | ||
3629 | /* Figures out whether a flow that missed in 'ofproto', whose details are in | |
3630 | * 'miss', is likely to be worth tracking in detail in userspace and (usually) | |
3631 | * installing a datapath flow. The answer is usually "yes" (a return value of | |
3632 | * true). However, for short flows the cost of bookkeeping is much higher than | |
3633 | * the benefits, so when the datapath holds a large number of flows we impose | |
3634 | * some heuristics to decide which flows are likely to be worth tracking. */ | |
3635 | static bool | |
3636 | flow_miss_should_make_facet(struct ofproto_dpif *ofproto, | |
3637 | struct flow_miss *miss, uint32_t hash) | |
3638 | { | |
3639 | if (!ofproto->governor) { | |
3640 | size_t n_subfacets; | |
3641 | ||
3642 | n_subfacets = hmap_count(&ofproto->subfacets); | |
3643 | if (n_subfacets * 2 <= ofproto->up.flow_eviction_threshold) { | |
3644 | return true; | |
3645 | } | |
3646 | ||
3647 | ofproto->governor = governor_create(ofproto->up.name); | |
3648 | } | |
3649 | ||
3650 | return governor_should_install_flow(ofproto->governor, hash, | |
3651 | list_size(&miss->packets)); | |
3652 | } | |
3653 | ||
3654 | /* Handles 'miss', which matches 'rule', without creating a facet or subfacet | |
3655 | * or creating any datapath flow. May add an "execute" operation to 'ops' and | |
3656 | * increment '*n_ops'. */ | |
3657 | static void | |
3658 | handle_flow_miss_without_facet(struct flow_miss *miss, | |
3659 | struct rule_dpif *rule, | |
3660 | struct flow_miss_op *ops, size_t *n_ops) | |
3661 | { | |
3662 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
3663 | long long int now = time_msec(); | |
3664 | struct action_xlate_ctx ctx; | |
3665 | struct ofpbuf *packet; | |
3666 | ||
3667 | LIST_FOR_EACH (packet, list_node, &miss->packets) { | |
3668 | struct flow_miss_op *op = &ops[*n_ops]; | |
3669 | struct dpif_flow_stats stats; | |
3670 | struct ofpbuf odp_actions; | |
3671 | ||
3672 | COVERAGE_INC(facet_suppress); | |
3673 | ||
3674 | ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub); | |
3675 | ||
3676 | dpif_flow_stats_extract(&miss->flow, packet, now, &stats); | |
3677 | rule_credit_stats(rule, &stats); | |
3678 | ||
3679 | action_xlate_ctx_init(&ctx, ofproto, &miss->flow, | |
3680 | &miss->initial_vals, rule, 0, packet); | |
3681 | ctx.resubmit_stats = &stats; | |
3682 | xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, | |
3683 | &odp_actions); | |
3684 | ||
3685 | if (odp_actions.size) { | |
3686 | struct dpif_execute *execute = &op->dpif_op.u.execute; | |
3687 | ||
3688 | init_flow_miss_execute_op(miss, packet, op); | |
3689 | execute->actions = odp_actions.data; | |
3690 | execute->actions_len = odp_actions.size; | |
3691 | op->garbage = ofpbuf_get_uninit_pointer(&odp_actions); | |
3692 | ||
3693 | (*n_ops)++; | |
3694 | } else { | |
3695 | ofpbuf_uninit(&odp_actions); | |
3696 | } | |
3697 | } | |
3698 | } | |
3699 | ||
3700 | /* Handles 'miss', which matches 'facet'. May add any required datapath | |
3701 | * operations to 'ops', incrementing '*n_ops' for each new op. | |
3702 | * | |
3703 | * All of the packets in 'miss' are considered to have arrived at time 'now'. | |
3704 | * This is really important only for new facets: if we just called time_msec() | |
3705 | * here, then the new subfacet or its packets could look (occasionally) as | |
3706 | * though it was used some time after the facet was used. That can make a | |
3707 | * one-packet flow look like it has a nonzero duration, which looks odd in | |
3708 | * e.g. NetFlow statistics. */ | |
3709 | static void | |
3710 | handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet, | |
3711 | long long int now, | |
3712 | struct flow_miss_op *ops, size_t *n_ops) | |
3713 | { | |
3714 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); | |
3715 | enum subfacet_path want_path; | |
3716 | struct subfacet *subfacet; | |
3717 | struct ofpbuf *packet; | |
3718 | ||
3719 | subfacet = subfacet_create(facet, miss, now); | |
3720 | ||
3721 | LIST_FOR_EACH (packet, list_node, &miss->packets) { | |
3722 | struct flow_miss_op *op = &ops[*n_ops]; | |
3723 | struct dpif_flow_stats stats; | |
3724 | struct ofpbuf odp_actions; | |
3725 | ||
3726 | handle_flow_miss_common(facet->rule, packet, &miss->flow); | |
3727 | ||
3728 | ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub); | |
3729 | if (!subfacet->actions || subfacet->slow) { | |
3730 | subfacet_make_actions(subfacet, packet, &odp_actions); | |
3731 | } | |
3732 | ||
3733 | dpif_flow_stats_extract(&facet->flow, packet, now, &stats); | |
3734 | subfacet_update_stats(subfacet, &stats); | |
3735 | ||
3736 | if (subfacet->actions_len) { | |
3737 | struct dpif_execute *execute = &op->dpif_op.u.execute; | |
3738 | ||
3739 | init_flow_miss_execute_op(miss, packet, op); | |
3740 | if (!subfacet->slow) { | |
3741 | execute->actions = subfacet->actions; | |
3742 | execute->actions_len = subfacet->actions_len; | |
3743 | ofpbuf_uninit(&odp_actions); | |
3744 | } else { | |
3745 | execute->actions = odp_actions.data; | |
3746 | execute->actions_len = odp_actions.size; | |
3747 | op->garbage = ofpbuf_get_uninit_pointer(&odp_actions); | |
3748 | } | |
3749 | ||
3750 | (*n_ops)++; | |
3751 | } else { | |
3752 | ofpbuf_uninit(&odp_actions); | |
3753 | } | |
3754 | } | |
3755 | ||
3756 | want_path = subfacet_want_path(subfacet->slow); | |
3757 | if (miss->upcall_type == DPIF_UC_MISS || subfacet->path != want_path) { | |
3758 | struct flow_miss_op *op = &ops[(*n_ops)++]; | |
3759 | struct dpif_flow_put *put = &op->dpif_op.u.flow_put; | |
3760 | ||
3761 | subfacet->path = want_path; | |
3762 | ||
3763 | op->garbage = NULL; | |
3764 | op->dpif_op.type = DPIF_OP_FLOW_PUT; | |
3765 | put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY; | |
3766 | put->key = miss->key; | |
3767 | put->key_len = miss->key_len; | |
3768 | if (want_path == SF_FAST_PATH) { | |
3769 | put->actions = subfacet->actions; | |
3770 | put->actions_len = subfacet->actions_len; | |
3771 | } else { | |
3772 | compose_slow_path(ofproto, &facet->flow, subfacet->slow, | |
3773 | op->stub, sizeof op->stub, | |
3774 | &put->actions, &put->actions_len); | |
3775 | } | |
3776 | put->stats = NULL; | |
3777 | } | |
3778 | } | |
3779 | ||
3780 | /* Handles flow miss 'miss'. May add any required datapath operations | |
3781 | * to 'ops', incrementing '*n_ops' for each new op. */ | |
3782 | static void | |
3783 | handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops, | |
3784 | size_t *n_ops) | |
3785 | { | |
3786 | struct ofproto_dpif *ofproto = miss->ofproto; | |
3787 | struct facet *facet; | |
3788 | long long int now; | |
3789 | uint32_t hash; | |
3790 | ||
3791 | /* The caller must ensure that miss->hmap_node.hash contains | |
3792 | * flow_hash(miss->flow, 0). */ | |
3793 | hash = miss->hmap_node.hash; | |
3794 | ||
3795 | facet = facet_lookup_valid(ofproto, &miss->flow, hash); | |
3796 | if (!facet) { | |
3797 | struct rule_dpif *rule = rule_dpif_lookup(ofproto, &miss->flow); | |
3798 | ||
3799 | if (!flow_miss_should_make_facet(ofproto, miss, hash)) { | |
3800 | handle_flow_miss_without_facet(miss, rule, ops, n_ops); | |
3801 | return; | |
3802 | } | |
3803 | ||
3804 | facet = facet_create(rule, &miss->flow, hash); | |
3805 | now = facet->used; | |
3806 | } else { | |
3807 | now = time_msec(); | |
3808 | } | |
3809 | handle_flow_miss_with_facet(miss, facet, now, ops, n_ops); | |
3810 | } | |
3811 | ||
3812 | static struct drop_key * | |
3813 | drop_key_lookup(const struct dpif_backer *backer, const struct nlattr *key, | |
3814 | size_t key_len) | |
3815 | { | |
3816 | struct drop_key *drop_key; | |
3817 | ||
3818 | HMAP_FOR_EACH_WITH_HASH (drop_key, hmap_node, hash_bytes(key, key_len, 0), | |
3819 | &backer->drop_keys) { | |
3820 | if (drop_key->key_len == key_len | |
3821 | && !memcmp(drop_key->key, key, key_len)) { | |
3822 | return drop_key; | |
3823 | } | |
3824 | } | |
3825 | return NULL; | |
3826 | } | |
3827 | ||
3828 | static void | |
3829 | drop_key_clear(struct dpif_backer *backer) | |
3830 | { | |
3831 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15); | |
3832 | struct drop_key *drop_key, *next; | |
3833 | ||
3834 | HMAP_FOR_EACH_SAFE (drop_key, next, hmap_node, &backer->drop_keys) { | |
3835 | int error; | |
3836 | ||
3837 | error = dpif_flow_del(backer->dpif, drop_key->key, drop_key->key_len, | |
3838 | NULL); | |
3839 | if (error && !VLOG_DROP_WARN(&rl)) { | |
3840 | struct ds ds = DS_EMPTY_INITIALIZER; | |
3841 | odp_flow_key_format(drop_key->key, drop_key->key_len, &ds); | |
3842 | VLOG_WARN("Failed to delete drop key (%s) (%s)", strerror(error), | |
3843 | ds_cstr(&ds)); | |
3844 | ds_destroy(&ds); | |
3845 | } | |
3846 | ||
3847 | hmap_remove(&backer->drop_keys, &drop_key->hmap_node); | |
3848 | free(drop_key->key); | |
3849 | free(drop_key); | |
3850 | } | |
3851 | } | |
3852 | ||
3853 | /* Given a datpath, packet, and flow metadata ('backer', 'packet', and 'key' | |
3854 | * respectively), populates 'flow' with the result of odp_flow_key_to_flow(). | |
3855 | * Optionally, if nonnull, populates 'fitnessp' with the fitness of 'flow' as | |
3856 | * returned by odp_flow_key_to_flow(). Also, optionally populates 'ofproto' | |
3857 | * with the ofproto_dpif, and 'odp_in_port' with the datapath in_port, that | |
3858 | * 'packet' ingressed. | |
3859 | * | |
3860 | * If 'ofproto' is nonnull, requires 'flow''s in_port to exist. Otherwise sets | |
3861 | * 'flow''s in_port to OFPP_NONE. | |
3862 | * | |
3863 | * This function does post-processing on data returned from | |
3864 | * odp_flow_key_to_flow() to help make VLAN splinters transparent to the rest | |
3865 | * of the upcall processing logic. In particular, if the extracted in_port is | |
3866 | * a VLAN splinter port, it replaces flow->in_port by the "real" port, sets | |
3867 | * flow->vlan_tci correctly for the VLAN of the VLAN splinter port, and pushes | |
3868 | * a VLAN header onto 'packet' (if it is nonnull). | |
3869 | * | |
3870 | * Optionally, if 'initial_vals' is nonnull, sets 'initial_vals->vlan_tci' | |
3871 | * to the VLAN TCI with which the packet was really received, that is, the | |
3872 | * actual VLAN TCI extracted by odp_flow_key_to_flow(). (This differs from | |
3873 | * the value returned in flow->vlan_tci only for packets received on | |
3874 | * VLAN splinters.) | |
3875 | * | |
3876 | * Similarly, this function also includes some logic to help with tunnels. It | |
3877 | * may modify 'flow' as necessary to make the tunneling implementation | |
3878 | * transparent to the upcall processing logic. | |
3879 | * | |
3880 | * Returns 0 if successful, ENODEV if the parsed flow has no associated ofport, | |
3881 | * or some other positive errno if there are other problems. */ | |
3882 | static int | |
3883 | ofproto_receive(const struct dpif_backer *backer, struct ofpbuf *packet, | |
3884 | const struct nlattr *key, size_t key_len, | |
3885 | struct flow *flow, enum odp_key_fitness *fitnessp, | |
3886 | struct ofproto_dpif **ofproto, uint32_t *odp_in_port, | |
3887 | struct initial_vals *initial_vals) | |
3888 | { | |
3889 | const struct ofport_dpif *port; | |
3890 | enum odp_key_fitness fitness; | |
3891 | int error = ENODEV; | |
3892 | ||
3893 | fitness = odp_flow_key_to_flow(key, key_len, flow); | |
3894 | if (fitness == ODP_FIT_ERROR) { | |
3895 | error = EINVAL; | |
3896 | goto exit; | |
3897 | } | |
3898 | ||
3899 | if (initial_vals) { | |
3900 | initial_vals->vlan_tci = flow->vlan_tci; | |
3901 | } | |
3902 | ||
3903 | if (odp_in_port) { | |
3904 | *odp_in_port = flow->in_port; | |
3905 | } | |
3906 | ||
3907 | port = (tnl_port_should_receive(flow) | |
3908 | ? ofport_dpif_cast(tnl_port_receive(flow)) | |
3909 | : odp_port_to_ofport(backer, flow->in_port)); | |
3910 | flow->in_port = port ? port->up.ofp_port : OFPP_NONE; | |
3911 | if (!port) { | |
3912 | goto exit; | |
3913 | } | |
3914 | ||
3915 | /* XXX: Since the tunnel module is not scoped per backer, for a tunnel port | |
3916 | * it's theoretically possible that we'll receive an ofport belonging to an | |
3917 | * entirely different datapath. In practice, this can't happen because no | |
3918 | * platforms has two separate datapaths which each support tunneling. */ | |
3919 | ovs_assert(ofproto_dpif_cast(port->up.ofproto)->backer == backer); | |
3920 | ||
3921 | if (vsp_adjust_flow(ofproto_dpif_cast(port->up.ofproto), flow)) { | |
3922 | if (packet) { | |
3923 | /* Make the packet resemble the flow, so that it gets sent to | |
3924 | * an OpenFlow controller properly, so that it looks correct | |
3925 | * for sFlow, and so that flow_extract() will get the correct | |
3926 | * vlan_tci if it is called on 'packet'. | |
3927 | * | |
3928 | * The allocated space inside 'packet' probably also contains | |
3929 | * 'key', that is, both 'packet' and 'key' are probably part of | |
3930 | * a struct dpif_upcall (see the large comment on that | |
3931 | * structure definition), so pushing data on 'packet' is in | |
3932 | * general not a good idea since it could overwrite 'key' or | |
3933 | * free it as a side effect. However, it's OK in this special | |
3934 | * case because we know that 'packet' is inside a Netlink | |
3935 | * attribute: pushing 4 bytes will just overwrite the 4-byte | |
3936 | * "struct nlattr", which is fine since we don't need that | |
3937 | * header anymore. */ | |
3938 | eth_push_vlan(packet, flow->vlan_tci); | |
3939 | } | |
3940 | /* We can't reproduce 'key' from 'flow'. */ | |
3941 | fitness = fitness == ODP_FIT_PERFECT ? ODP_FIT_TOO_MUCH : fitness; | |
3942 | } | |
3943 | error = 0; | |
3944 | ||
3945 | if (ofproto) { | |
3946 | *ofproto = ofproto_dpif_cast(port->up.ofproto); | |
3947 | } | |
3948 | ||
3949 | exit: | |
3950 | if (fitnessp) { | |
3951 | *fitnessp = fitness; | |
3952 | } | |
3953 | return error; | |
3954 | } | |
3955 | ||
3956 | static void | |
3957 | handle_miss_upcalls(struct dpif_backer *backer, struct dpif_upcall *upcalls, | |
3958 | size_t n_upcalls) | |
3959 | { | |
3960 | struct dpif_upcall *upcall; | |
3961 | struct flow_miss *miss; | |
3962 | struct flow_miss misses[FLOW_MISS_MAX_BATCH]; | |
3963 | struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2]; | |
3964 | struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2]; | |
3965 | struct hmap todo; | |
3966 | int n_misses; | |
3967 | size_t n_ops; | |
3968 | size_t i; | |
3969 | ||
3970 | if (!n_upcalls) { | |
3971 | return; | |
3972 | } | |
3973 | ||
3974 | /* Construct the to-do list. | |
3975 | * | |
3976 | * This just amounts to extracting the flow from each packet and sticking | |
3977 | * the packets that have the same flow in the same "flow_miss" structure so | |
3978 | * that we can process them together. */ | |
3979 | hmap_init(&todo); | |
3980 | n_misses = 0; | |
3981 | for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) { | |
3982 | struct flow_miss *miss = &misses[n_misses]; | |
3983 | struct flow_miss *existing_miss; | |
3984 | struct ofproto_dpif *ofproto; | |
3985 | uint32_t odp_in_port; | |
3986 | struct flow flow; | |
3987 | uint32_t hash; | |
3988 | int error; | |
3989 | ||
3990 | error = ofproto_receive(backer, upcall->packet, upcall->key, | |
3991 | upcall->key_len, &flow, &miss->key_fitness, | |
3992 | &ofproto, &odp_in_port, &miss->initial_vals); | |
3993 | if (error == ENODEV) { | |
3994 | struct drop_key *drop_key; | |
3995 | ||
3996 | /* Received packet on port for which we couldn't associate | |
3997 | * an ofproto. This can happen if a port is removed while | |
3998 | * traffic is being received. Print a rate-limited message | |
3999 | * in case it happens frequently. Install a drop flow so | |
4000 | * that future packets of the flow are inexpensively dropped | |
4001 | * in the kernel. */ | |
4002 | VLOG_INFO_RL(&rl, "received packet on unassociated port %"PRIu32, | |
4003 | flow.in_port); | |
4004 | ||
4005 | drop_key = drop_key_lookup(backer, upcall->key, upcall->key_len); | |
4006 | if (!drop_key) { | |
4007 | drop_key = xmalloc(sizeof *drop_key); | |
4008 | drop_key->key = xmemdup(upcall->key, upcall->key_len); | |
4009 | drop_key->key_len = upcall->key_len; | |
4010 | ||
4011 | hmap_insert(&backer->drop_keys, &drop_key->hmap_node, | |
4012 | hash_bytes(drop_key->key, drop_key->key_len, 0)); | |
4013 | dpif_flow_put(backer->dpif, DPIF_FP_CREATE | DPIF_FP_MODIFY, | |
4014 | drop_key->key, drop_key->key_len, NULL, 0, NULL); | |
4015 | } | |
4016 | continue; | |
4017 | } | |
4018 | if (error) { | |
4019 | continue; | |
4020 | } | |
4021 | ||
4022 | ofproto->n_missed++; | |
4023 | flow_extract(upcall->packet, flow.skb_priority, flow.skb_mark, | |
4024 | &flow.tunnel, flow.in_port, &miss->flow); | |
4025 | ||
4026 | /* Add other packets to a to-do list. */ | |
4027 | hash = flow_hash(&miss->flow, 0); | |
4028 | existing_miss = flow_miss_find(&todo, ofproto, &miss->flow, hash); | |
4029 | if (!existing_miss) { | |
4030 | hmap_insert(&todo, &miss->hmap_node, hash); | |
4031 | miss->ofproto = ofproto; | |
4032 | miss->key = upcall->key; | |
4033 | miss->key_len = upcall->key_len; | |
4034 | miss->upcall_type = upcall->type; | |
4035 | miss->odp_in_port = odp_in_port; | |
4036 | list_init(&miss->packets); | |
4037 | ||
4038 | n_misses++; | |
4039 | } else { | |
4040 | miss = existing_miss; | |
4041 | } | |
4042 | list_push_back(&miss->packets, &upcall->packet->list_node); | |
4043 | } | |
4044 | ||
4045 | /* Process each element in the to-do list, constructing the set of | |
4046 | * operations to batch. */ | |
4047 | n_ops = 0; | |
4048 | HMAP_FOR_EACH (miss, hmap_node, &todo) { | |
4049 | handle_flow_miss(miss, flow_miss_ops, &n_ops); | |
4050 | } | |
4051 | ovs_assert(n_ops <= ARRAY_SIZE(flow_miss_ops)); | |
4052 | ||
4053 | /* Execute batch. */ | |
4054 | for (i = 0; i < n_ops; i++) { | |
4055 | dpif_ops[i] = &flow_miss_ops[i].dpif_op; | |
4056 | } | |
4057 | dpif_operate(backer->dpif, dpif_ops, n_ops); | |
4058 | ||
4059 | /* Free memory. */ | |
4060 | for (i = 0; i < n_ops; i++) { | |
4061 | free(flow_miss_ops[i].garbage); | |
4062 | } | |
4063 | hmap_destroy(&todo); | |
4064 | } | |
4065 | ||
4066 | static enum { SFLOW_UPCALL, MISS_UPCALL, BAD_UPCALL, FLOW_SAMPLE_UPCALL, | |
4067 | IPFIX_UPCALL } | |
4068 | classify_upcall(const struct dpif_upcall *upcall) | |
4069 | { | |
4070 | size_t userdata_len; | |
4071 | union user_action_cookie cookie; | |
4072 | ||
4073 | /* First look at the upcall type. */ | |
4074 | switch (upcall->type) { | |
4075 | case DPIF_UC_ACTION: | |
4076 | break; | |
4077 | ||
4078 | case DPIF_UC_MISS: | |
4079 | return MISS_UPCALL; | |
4080 | ||
4081 | case DPIF_N_UC_TYPES: | |
4082 | default: | |
4083 | VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type); | |
4084 | return BAD_UPCALL; | |
4085 | } | |
4086 | ||
4087 | /* "action" upcalls need a closer look. */ | |
4088 | if (!upcall->userdata) { | |
4089 | VLOG_WARN_RL(&rl, "action upcall missing cookie"); | |
4090 | return BAD_UPCALL; | |
4091 | } | |
4092 | userdata_len = nl_attr_get_size(upcall->userdata); | |
4093 | if (userdata_len < sizeof cookie.type | |
4094 | || userdata_len > sizeof cookie) { | |
4095 | VLOG_WARN_RL(&rl, "action upcall cookie has unexpected size %zu", | |
4096 | userdata_len); | |
4097 | return BAD_UPCALL; | |
4098 | } | |
4099 | memset(&cookie, 0, sizeof cookie); | |
4100 | memcpy(&cookie, nl_attr_get(upcall->userdata), userdata_len); | |
4101 | if (userdata_len == sizeof cookie.sflow | |
4102 | && cookie.type == USER_ACTION_COOKIE_SFLOW) { | |
4103 | return SFLOW_UPCALL; | |
4104 | } else if (userdata_len == sizeof cookie.slow_path | |
4105 | && cookie.type == USER_ACTION_COOKIE_SLOW_PATH) { | |
4106 | return MISS_UPCALL; | |
4107 | } else if (userdata_len == sizeof cookie.flow_sample | |
4108 | && cookie.type == USER_ACTION_COOKIE_FLOW_SAMPLE) { | |
4109 | return FLOW_SAMPLE_UPCALL; | |
4110 | } else if (userdata_len == sizeof cookie.ipfix | |
4111 | && cookie.type == USER_ACTION_COOKIE_IPFIX) { | |
4112 | return IPFIX_UPCALL; | |
4113 | } else { | |
4114 | VLOG_WARN_RL(&rl, "invalid user cookie of type %"PRIu16 | |
4115 | " and size %zu", cookie.type, userdata_len); | |
4116 | return BAD_UPCALL; | |
4117 | } | |
4118 | } | |
4119 | ||
4120 | static void | |
4121 | handle_sflow_upcall(struct dpif_backer *backer, | |
4122 | const struct dpif_upcall *upcall) | |
4123 | { | |
4124 | struct ofproto_dpif *ofproto; | |
4125 | union user_action_cookie cookie; | |
4126 | struct flow flow; | |
4127 | uint32_t odp_in_port; | |
4128 | ||
4129 | if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len, | |
4130 | &flow, NULL, &ofproto, &odp_in_port, NULL) | |
4131 | || !ofproto->sflow) { | |
4132 | return; | |
4133 | } | |
4134 | ||
4135 | memset(&cookie, 0, sizeof cookie); | |
4136 | memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.sflow); | |
4137 | dpif_sflow_received(ofproto->sflow, upcall->packet, &flow, | |
4138 | odp_in_port, &cookie); | |
4139 | } | |
4140 | ||
4141 | static void | |
4142 | handle_flow_sample_upcall(struct dpif_backer *backer, | |
4143 | const struct dpif_upcall *upcall) | |
4144 | { | |
4145 | struct ofproto_dpif *ofproto; | |
4146 | union user_action_cookie cookie; | |
4147 | struct flow flow; | |
4148 | ||
4149 | if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len, | |
4150 | &flow, NULL, &ofproto, NULL, NULL) | |
4151 | || !ofproto->ipfix) { | |
4152 | return; | |
4153 | } | |
4154 | ||
4155 | memset(&cookie, 0, sizeof cookie); | |
4156 | memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.flow_sample); | |
4157 | ||
4158 | /* The flow reflects exactly the contents of the packet. Sample | |
4159 | * the packet using it. */ | |
4160 | dpif_ipfix_flow_sample(ofproto->ipfix, upcall->packet, &flow, | |
4161 | cookie.flow_sample.collector_set_id, | |
4162 | cookie.flow_sample.probability, | |
4163 | cookie.flow_sample.obs_domain_id, | |
4164 | cookie.flow_sample.obs_point_id); | |
4165 | } | |
4166 | ||
4167 | static void | |
4168 | handle_ipfix_upcall(struct dpif_backer *backer, | |
4169 | const struct dpif_upcall *upcall) | |
4170 | { | |
4171 | struct ofproto_dpif *ofproto; | |
4172 | struct flow flow; | |
4173 | ||
4174 | if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len, | |
4175 | &flow, NULL, &ofproto, NULL, NULL) | |
4176 | || !ofproto->ipfix) { | |
4177 | return; | |
4178 | } | |
4179 | ||
4180 | /* The flow reflects exactly the contents of the packet. Sample | |
4181 | * the packet using it. */ | |
4182 | dpif_ipfix_bridge_sample(ofproto->ipfix, upcall->packet, &flow); | |
4183 | } | |
4184 | ||
4185 | static int | |
4186 | handle_upcalls(struct dpif_backer *backer, unsigned int max_batch) | |
4187 | { | |
4188 | struct dpif_upcall misses[FLOW_MISS_MAX_BATCH]; | |
4189 | struct ofpbuf miss_bufs[FLOW_MISS_MAX_BATCH]; | |
4190 | uint64_t miss_buf_stubs[FLOW_MISS_MAX_BATCH][4096 / 8]; | |
4191 | int n_processed; | |
4192 | int n_misses; | |
4193 | int i; | |
4194 | ||
4195 | ovs_assert(max_batch <= FLOW_MISS_MAX_BATCH); | |
4196 | ||
4197 | n_misses = 0; | |
4198 | for (n_processed = 0; n_processed < max_batch; n_processed++) { | |
4199 | struct dpif_upcall *upcall = &misses[n_misses]; | |
4200 | struct ofpbuf *buf = &miss_bufs[n_misses]; | |
4201 | int error; | |
4202 | ||
4203 | ofpbuf_use_stub(buf, miss_buf_stubs[n_misses], | |
4204 | sizeof miss_buf_stubs[n_misses]); | |
4205 | error = dpif_recv(backer->dpif, upcall, buf); | |
4206 | if (error) { | |
4207 | ofpbuf_uninit(buf); | |
4208 | break; | |
4209 | } | |
4210 | ||
4211 | switch (classify_upcall(upcall)) { | |
4212 | case MISS_UPCALL: | |
4213 | /* Handle it later. */ | |
4214 | n_misses++; | |
4215 | break; | |
4216 | ||
4217 | case SFLOW_UPCALL: | |
4218 | handle_sflow_upcall(backer, upcall); | |
4219 | ofpbuf_uninit(buf); | |
4220 | break; | |
4221 | ||
4222 | case FLOW_SAMPLE_UPCALL: | |
4223 | handle_flow_sample_upcall(backer, upcall); | |
4224 | ofpbuf_uninit(buf); | |
4225 | break; | |
4226 | ||
4227 | case IPFIX_UPCALL: | |
4228 | handle_ipfix_upcall(backer, upcall); | |
4229 | ofpbuf_uninit(buf); | |
4230 | break; | |
4231 | ||
4232 | case BAD_UPCALL: | |
4233 | ofpbuf_uninit(buf); | |
4234 | break; | |
4235 | } | |
4236 | } | |
4237 | ||
4238 | /* Handle deferred MISS_UPCALL processing. */ | |
4239 | handle_miss_upcalls(backer, misses, n_misses); | |
4240 | for (i = 0; i < n_misses; i++) { | |
4241 | ofpbuf_uninit(&miss_bufs[i]); | |
4242 | } | |
4243 | ||
4244 | return n_processed; | |
4245 | } | |
4246 | \f | |
4247 | /* Flow expiration. */ | |
4248 | ||
4249 | static int subfacet_max_idle(const struct ofproto_dpif *); | |
4250 | static void update_stats(struct dpif_backer *); | |
4251 | static void rule_expire(struct rule_dpif *); | |
4252 | static void expire_subfacets(struct ofproto_dpif *, int dp_max_idle); | |
4253 | ||
4254 | /* This function is called periodically by run(). Its job is to collect | |
4255 | * updates for the flows that have been installed into the datapath, most | |
4256 | * importantly when they last were used, and then use that information to | |
4257 | * expire flows that have not been used recently. | |
4258 | * | |
4259 | * Returns the number of milliseconds after which it should be called again. */ | |
4260 | static int | |
4261 | expire(struct dpif_backer *backer) | |
4262 | { | |
4263 | struct ofproto_dpif *ofproto; | |
4264 | int max_idle = INT32_MAX; | |
4265 | ||
4266 | /* Periodically clear out the drop keys in an effort to keep them | |
4267 | * relatively few. */ | |
4268 | drop_key_clear(backer); | |
4269 | ||
4270 | /* Update stats for each flow in the backer. */ | |
4271 | update_stats(backer); | |
4272 | ||
4273 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
4274 | struct rule *rule, *next_rule; | |
4275 | int dp_max_idle; | |
4276 | ||
4277 | if (ofproto->backer != backer) { | |
4278 | continue; | |
4279 | } | |
4280 | ||
4281 | /* Keep track of the max number of flows per ofproto_dpif. */ | |
4282 | update_max_subfacet_count(ofproto); | |
4283 | ||
4284 | /* Expire subfacets that have been idle too long. */ | |
4285 | dp_max_idle = subfacet_max_idle(ofproto); | |
4286 | expire_subfacets(ofproto, dp_max_idle); | |
4287 | ||
4288 | max_idle = MIN(max_idle, dp_max_idle); | |
4289 | ||
4290 | /* Expire OpenFlow flows whose idle_timeout or hard_timeout | |
4291 | * has passed. */ | |
4292 | LIST_FOR_EACH_SAFE (rule, next_rule, expirable, | |
4293 | &ofproto->up.expirable) { | |
4294 | rule_expire(rule_dpif_cast(rule)); | |
4295 | } | |
4296 | ||
4297 | /* All outstanding data in existing flows has been accounted, so it's a | |
4298 | * good time to do bond rebalancing. */ | |
4299 | if (ofproto->has_bonded_bundles) { | |
4300 | struct ofbundle *bundle; | |
4301 | ||
4302 | HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { | |
4303 | if (bundle->bond) { | |
4304 | bond_rebalance(bundle->bond, &backer->revalidate_set); | |
4305 | } | |
4306 | } | |
4307 | } | |
4308 | } | |
4309 | ||
4310 | return MIN(max_idle, 1000); | |
4311 | } | |
4312 | ||
4313 | /* Updates flow table statistics given that the datapath just reported 'stats' | |
4314 | * as 'subfacet''s statistics. */ | |
4315 | static void | |
4316 | update_subfacet_stats(struct subfacet *subfacet, | |
4317 | const struct dpif_flow_stats *stats) | |
4318 | { | |
4319 | struct facet *facet = subfacet->facet; | |
4320 | ||
4321 | if (stats->n_packets >= subfacet->dp_packet_count) { | |
4322 | uint64_t extra = stats->n_packets - subfacet->dp_packet_count; | |
4323 | facet->packet_count += extra; | |
4324 | } else { | |
4325 | VLOG_WARN_RL(&rl, "unexpected packet count from the datapath"); | |
4326 | } | |
4327 | ||
4328 | if (stats->n_bytes >= subfacet->dp_byte_count) { | |
4329 | facet->byte_count += stats->n_bytes - subfacet->dp_byte_count; | |
4330 | } else { | |
4331 | VLOG_WARN_RL(&rl, "unexpected byte count from datapath"); | |
4332 | } | |
4333 | ||
4334 | subfacet->dp_packet_count = stats->n_packets; | |
4335 | subfacet->dp_byte_count = stats->n_bytes; | |
4336 | ||
4337 | facet->tcp_flags |= stats->tcp_flags; | |
4338 | ||
4339 | subfacet_update_time(subfacet, stats->used); | |
4340 | if (facet->accounted_bytes < facet->byte_count) { | |
4341 | facet_learn(facet); | |
4342 | facet_account(facet); | |
4343 | facet->accounted_bytes = facet->byte_count; | |
4344 | } | |
4345 | } | |
4346 | ||
4347 | /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing | |
4348 | * about, or a flow that shouldn't be installed but was anyway. Delete it. */ | |
4349 | static void | |
4350 | delete_unexpected_flow(struct ofproto_dpif *ofproto, | |
4351 | const struct nlattr *key, size_t key_len) | |
4352 | { | |
4353 | if (!VLOG_DROP_WARN(&rl)) { | |
4354 | struct ds s; | |
4355 | ||
4356 | ds_init(&s); | |
4357 | odp_flow_key_format(key, key_len, &s); | |
4358 | VLOG_WARN("unexpected flow on %s: %s", ofproto->up.name, ds_cstr(&s)); | |
4359 | ds_destroy(&s); | |
4360 | } | |
4361 | ||
4362 | COVERAGE_INC(facet_unexpected); | |
4363 | dpif_flow_del(ofproto->backer->dpif, key, key_len, NULL); | |
4364 | } | |
4365 | ||
4366 | /* Update 'packet_count', 'byte_count', and 'used' members of installed facets. | |
4367 | * | |
4368 | * This function also pushes statistics updates to rules which each facet | |
4369 | * resubmits into. Generally these statistics will be accurate. However, if a | |
4370 | * facet changes the rule it resubmits into at some time in between | |
4371 | * update_stats() runs, it is possible that statistics accrued to the | |
4372 | * old rule will be incorrectly attributed to the new rule. This could be | |
4373 | * avoided by calling update_stats() whenever rules are created or | |
4374 | * deleted. However, the performance impact of making so many calls to the | |
4375 | * datapath do not justify the benefit of having perfectly accurate statistics. | |
4376 | * | |
4377 | * In addition, this function maintains per ofproto flow hit counts. The patch | |
4378 | * port is not treated specially. e.g. A packet ingress from br0 patched into | |
4379 | * br1 will increase the hit count of br0 by 1, however, does not affect | |
4380 | * the hit or miss counts of br1. | |
4381 | */ | |
4382 | static void | |
4383 | update_stats(struct dpif_backer *backer) | |
4384 | { | |
4385 | const struct dpif_flow_stats *stats; | |
4386 | struct dpif_flow_dump dump; | |
4387 | const struct nlattr *key; | |
4388 | struct ofproto_dpif *ofproto; | |
4389 | size_t key_len; | |
4390 | ||
4391 | dpif_flow_dump_start(&dump, backer->dpif); | |
4392 | while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) { | |
4393 | struct flow flow; | |
4394 | struct subfacet *subfacet; | |
4395 | struct ofport_dpif *ofport; | |
4396 | uint32_t key_hash; | |
4397 | ||
4398 | if (ofproto_receive(backer, NULL, key, key_len, &flow, NULL, &ofproto, | |
4399 | NULL, NULL)) { | |
4400 | continue; | |
4401 | } | |
4402 | ||
4403 | ofproto->total_subfacet_count += hmap_count(&ofproto->subfacets); | |
4404 | ofproto->n_update_stats++; | |
4405 | ||
4406 | ofport = get_ofp_port(ofproto, flow.in_port); | |
4407 | if (ofport && ofport->tnl_port) { | |
4408 | netdev_vport_inc_rx(ofport->up.netdev, stats); | |
4409 | } | |
4410 | ||
4411 | key_hash = odp_flow_key_hash(key, key_len); | |
4412 | subfacet = subfacet_find(ofproto, key, key_len, key_hash); | |
4413 | switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) { | |
4414 | case SF_FAST_PATH: | |
4415 | /* Update ofproto_dpif's hit count. */ | |
4416 | if (stats->n_packets > subfacet->dp_packet_count) { | |
4417 | uint64_t delta = stats->n_packets - subfacet->dp_packet_count; | |
4418 | dpif_stats_update_hit_count(ofproto, delta); | |
4419 | } | |
4420 | ||
4421 | update_subfacet_stats(subfacet, stats); | |
4422 | break; | |
4423 | ||
4424 | case SF_SLOW_PATH: | |
4425 | /* Stats are updated per-packet. */ | |
4426 | break; | |
4427 | ||
4428 | case SF_NOT_INSTALLED: | |
4429 | default: | |
4430 | delete_unexpected_flow(ofproto, key, key_len); | |
4431 | break; | |
4432 | } | |
4433 | run_fast_rl(); | |
4434 | } | |
4435 | dpif_flow_dump_done(&dump); | |
4436 | ||
4437 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
4438 | update_moving_averages(ofproto); | |
4439 | } | |
4440 | ||
4441 | } | |
4442 | ||
4443 | /* Calculates and returns the number of milliseconds of idle time after which | |
4444 | * subfacets should expire from the datapath. When a subfacet expires, we fold | |
4445 | * its statistics into its facet, and when a facet's last subfacet expires, we | |
4446 | * fold its statistic into its rule. */ | |
4447 | static int | |
4448 | subfacet_max_idle(const struct ofproto_dpif *ofproto) | |
4449 | { | |
4450 | /* | |
4451 | * Idle time histogram. | |
4452 | * | |
4453 | * Most of the time a switch has a relatively small number of subfacets. | |
4454 | * When this is the case we might as well keep statistics for all of them | |
4455 | * in userspace and to cache them in the kernel datapath for performance as | |
4456 | * well. | |
4457 | * | |
4458 | * As the number of subfacets increases, the memory required to maintain | |
4459 | * statistics about them in userspace and in the kernel becomes | |
4460 | * significant. However, with a large number of subfacets it is likely | |
4461 | * that only a few of them are "heavy hitters" that consume a large amount | |
4462 | * of bandwidth. At this point, only heavy hitters are worth caching in | |
4463 | * the kernel and maintaining in userspaces; other subfacets we can | |
4464 | * discard. | |
4465 | * | |
4466 | * The technique used to compute the idle time is to build a histogram with | |
4467 | * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet | |
4468 | * that is installed in the kernel gets dropped in the appropriate bucket. | |
4469 | * After the histogram has been built, we compute the cutoff so that only | |
4470 | * the most-recently-used 1% of subfacets (but at least | |
4471 | * ofproto->up.flow_eviction_threshold flows) are kept cached. At least | |
4472 | * the most-recently-used bucket of subfacets is kept, so actually an | |
4473 | * arbitrary number of subfacets can be kept in any given expiration run | |
4474 | * (though the next run will delete most of those unless they receive | |
4475 | * additional data). | |
4476 | * | |
4477 | * This requires a second pass through the subfacets, in addition to the | |
4478 | * pass made by update_stats(), because the former function never looks at | |
4479 | * uninstallable subfacets. | |
4480 | */ | |
4481 | enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) }; | |
4482 | enum { N_BUCKETS = 5000 / BUCKET_WIDTH }; | |
4483 | int buckets[N_BUCKETS] = { 0 }; | |
4484 | int total, subtotal, bucket; | |
4485 | struct subfacet *subfacet; | |
4486 | long long int now; | |
4487 | int i; | |
4488 | ||
4489 | total = hmap_count(&ofproto->subfacets); | |
4490 | if (total <= ofproto->up.flow_eviction_threshold) { | |
4491 | return N_BUCKETS * BUCKET_WIDTH; | |
4492 | } | |
4493 | ||
4494 | /* Build histogram. */ | |
4495 | now = time_msec(); | |
4496 | HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) { | |
4497 | long long int idle = now - subfacet->used; | |
4498 | int bucket = (idle <= 0 ? 0 | |
4499 | : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1 | |
4500 | : (unsigned int) idle / BUCKET_WIDTH); | |
4501 | buckets[bucket]++; | |
4502 | } | |
4503 | ||
4504 | /* Find the first bucket whose flows should be expired. */ | |
4505 | subtotal = bucket = 0; | |
4506 | do { | |
4507 | subtotal += buckets[bucket++]; | |
4508 | } while (bucket < N_BUCKETS && | |
4509 | subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100)); | |
4510 | ||
4511 | if (VLOG_IS_DBG_ENABLED()) { | |
4512 | struct ds s; | |
4513 | ||
4514 | ds_init(&s); | |
4515 | ds_put_cstr(&s, "keep"); | |
4516 | for (i = 0; i < N_BUCKETS; i++) { | |
4517 | if (i == bucket) { | |
4518 | ds_put_cstr(&s, ", drop"); | |
4519 | } | |
4520 | if (buckets[i]) { | |
4521 | ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]); | |
4522 | } | |
4523 | } | |
4524 | VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s)); | |
4525 | ds_destroy(&s); | |
4526 | } | |
4527 | ||
4528 | return bucket * BUCKET_WIDTH; | |
4529 | } | |
4530 | ||
4531 | static void | |
4532 | expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle) | |
4533 | { | |
4534 | /* Cutoff time for most flows. */ | |
4535 | long long int normal_cutoff = time_msec() - dp_max_idle; | |
4536 | ||
4537 | /* We really want to keep flows for special protocols around, so use a more | |
4538 | * conservative cutoff. */ | |
4539 | long long int special_cutoff = time_msec() - 10000; | |
4540 | ||
4541 | struct subfacet *subfacet, *next_subfacet; | |
4542 | struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH]; | |
4543 | int n_batch; | |
4544 | ||
4545 | n_batch = 0; | |
4546 | HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node, | |
4547 | &ofproto->subfacets) { | |
4548 | long long int cutoff; | |
4549 | ||
4550 | cutoff = (subfacet->slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP | SLOW_STP) | |
4551 | ? special_cutoff | |
4552 | : normal_cutoff); | |
4553 | if (subfacet->used < cutoff) { | |
4554 | if (subfacet->path != SF_NOT_INSTALLED) { | |
4555 | batch[n_batch++] = subfacet; | |
4556 | if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) { | |
4557 | subfacet_destroy_batch(ofproto, batch, n_batch); | |
4558 | n_batch = 0; | |
4559 | } | |
4560 | } else { | |
4561 | subfacet_destroy(subfacet); | |
4562 | } | |
4563 | } | |
4564 | } | |
4565 | ||
4566 | if (n_batch > 0) { | |
4567 | subfacet_destroy_batch(ofproto, batch, n_batch); | |
4568 | } | |
4569 | } | |
4570 | ||
4571 | /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules, | |
4572 | * then delete it entirely. */ | |
4573 | static void | |
4574 | rule_expire(struct rule_dpif *rule) | |
4575 | { | |
4576 | struct facet *facet, *next_facet; | |
4577 | long long int now; | |
4578 | uint8_t reason; | |
4579 | ||
4580 | if (rule->up.pending) { | |
4581 | /* We'll have to expire it later. */ | |
4582 | return; | |
4583 | } | |
4584 | ||
4585 | /* Has 'rule' expired? */ | |
4586 | now = time_msec(); | |
4587 | if (rule->up.hard_timeout | |
4588 | && now > rule->up.modified + rule->up.hard_timeout * 1000) { | |
4589 | reason = OFPRR_HARD_TIMEOUT; | |
4590 | } else if (rule->up.idle_timeout | |
4591 | && now > rule->up.used + rule->up.idle_timeout * 1000) { | |
4592 | reason = OFPRR_IDLE_TIMEOUT; | |
4593 | } else { | |
4594 | return; | |
4595 | } | |
4596 | ||
4597 | COVERAGE_INC(ofproto_dpif_expired); | |
4598 | ||
4599 | /* Update stats. (This is a no-op if the rule expired due to an idle | |
4600 | * timeout, because that only happens when the rule has no facets left.) */ | |
4601 | LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) { | |
4602 | facet_remove(facet); | |
4603 | } | |
4604 | ||
4605 | /* Get rid of the rule. */ | |
4606 | ofproto_rule_expire(&rule->up, reason); | |
4607 | } | |
4608 | \f | |
4609 | /* Facets. */ | |
4610 | ||
4611 | /* Creates and returns a new facet owned by 'rule', given a 'flow'. | |
4612 | * | |
4613 | * The caller must already have determined that no facet with an identical | |
4614 | * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in | |
4615 | * the ofproto's classifier table. | |
4616 | * | |
4617 | * 'hash' must be the return value of flow_hash(flow, 0). | |
4618 | * | |
4619 | * The facet will initially have no subfacets. The caller should create (at | |
4620 | * least) one subfacet with subfacet_create(). */ | |
4621 | static struct facet * | |
4622 | facet_create(struct rule_dpif *rule, const struct flow *flow, uint32_t hash) | |
4623 | { | |
4624 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
4625 | struct facet *facet; | |
4626 | ||
4627 | facet = xzalloc(sizeof *facet); | |
4628 | facet->used = time_msec(); | |
4629 | hmap_insert(&ofproto->facets, &facet->hmap_node, hash); | |
4630 | list_push_back(&rule->facets, &facet->list_node); | |
4631 | facet->rule = rule; | |
4632 | facet->flow = *flow; | |
4633 | list_init(&facet->subfacets); | |
4634 | netflow_flow_init(&facet->nf_flow); | |
4635 | netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used); | |
4636 | ||
4637 | facet->learn_rl = time_msec() + 500; | |
4638 | ||
4639 | return facet; | |
4640 | } | |
4641 | ||
4642 | static void | |
4643 | facet_free(struct facet *facet) | |
4644 | { | |
4645 | free(facet); | |
4646 | } | |
4647 | ||
4648 | /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on | |
4649 | * 'packet', which arrived on 'in_port'. */ | |
4650 | static bool | |
4651 | execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow, | |
4652 | const struct nlattr *odp_actions, size_t actions_len, | |
4653 | struct ofpbuf *packet) | |
4654 | { | |
4655 | struct odputil_keybuf keybuf; | |
4656 | struct ofpbuf key; | |
4657 | int error; | |
4658 | ||
4659 | ofpbuf_use_stack(&key, &keybuf, sizeof keybuf); | |
4660 | odp_flow_key_from_flow(&key, flow, | |
4661 | ofp_port_to_odp_port(ofproto, flow->in_port)); | |
4662 | ||
4663 | error = dpif_execute(ofproto->backer->dpif, key.data, key.size, | |
4664 | odp_actions, actions_len, packet); | |
4665 | return !error; | |
4666 | } | |
4667 | ||
4668 | /* Remove 'facet' from 'ofproto' and free up the associated memory: | |
4669 | * | |
4670 | * - If 'facet' was installed in the datapath, uninstalls it and updates its | |
4671 | * rule's statistics, via subfacet_uninstall(). | |
4672 | * | |
4673 | * - Removes 'facet' from its rule and from ofproto->facets. | |
4674 | */ | |
4675 | static void | |
4676 | facet_remove(struct facet *facet) | |
4677 | { | |
4678 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); | |
4679 | struct subfacet *subfacet, *next_subfacet; | |
4680 | ||
4681 | ovs_assert(!list_is_empty(&facet->subfacets)); | |
4682 | ||
4683 | /* First uninstall all of the subfacets to get final statistics. */ | |
4684 | LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) { | |
4685 | subfacet_uninstall(subfacet); | |
4686 | } | |
4687 | ||
4688 | /* Flush the final stats to the rule. | |
4689 | * | |
4690 | * This might require us to have at least one subfacet around so that we | |
4691 | * can use its actions for accounting in facet_account(), which is why we | |
4692 | * have uninstalled but not yet destroyed the subfacets. */ | |
4693 | facet_flush_stats(facet); | |
4694 | ||
4695 | /* Now we're really all done so destroy everything. */ | |
4696 | LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node, | |
4697 | &facet->subfacets) { | |
4698 | subfacet_destroy__(subfacet); | |
4699 | } | |
4700 | hmap_remove(&ofproto->facets, &facet->hmap_node); | |
4701 | list_remove(&facet->list_node); | |
4702 | facet_free(facet); | |
4703 | } | |
4704 | ||
4705 | /* Feed information from 'facet' back into the learning table to keep it in | |
4706 | * sync with what is actually flowing through the datapath. */ | |
4707 | static void | |
4708 | facet_learn(struct facet *facet) | |
4709 | { | |
4710 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); | |
4711 | struct subfacet *subfacet= CONTAINER_OF(list_front(&facet->subfacets), | |
4712 | struct subfacet, list_node); | |
4713 | long long int now = time_msec(); | |
4714 | struct action_xlate_ctx ctx; | |
4715 | ||
4716 | if (!facet->has_fin_timeout && now < facet->learn_rl) { | |
4717 | return; | |
4718 | } | |
4719 | ||
4720 | facet->learn_rl = now + 500; | |
4721 | ||
4722 | if (!facet->has_learn | |
4723 | && !facet->has_normal | |
4724 | && (!facet->has_fin_timeout | |
4725 | || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) { | |
4726 | return; | |
4727 | } | |
4728 | ||
4729 | action_xlate_ctx_init(&ctx, ofproto, &facet->flow, | |
4730 | &subfacet->initial_vals, | |
4731 | facet->rule, facet->tcp_flags, NULL); | |
4732 | ctx.may_learn = true; | |
4733 | xlate_actions_for_side_effects(&ctx, facet->rule->up.ofpacts, | |
4734 | facet->rule->up.ofpacts_len); | |
4735 | } | |
4736 | ||
4737 | static void | |
4738 | facet_account(struct facet *facet) | |
4739 | { | |
4740 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); | |
4741 | struct subfacet *subfacet = facet_get_subfacet(facet); | |
4742 | const struct nlattr *a; | |
4743 | unsigned int left; | |
4744 | ovs_be16 vlan_tci; | |
4745 | uint64_t n_bytes; | |
4746 | ||
4747 | if (!facet->has_normal || !ofproto->has_bonded_bundles) { | |
4748 | return; | |
4749 | } | |
4750 | n_bytes = facet->byte_count - facet->accounted_bytes; | |
4751 | ||
4752 | /* This loop feeds byte counters to bond_account() for rebalancing to use | |
4753 | * as a basis. We also need to track the actual VLAN on which the packet | |
4754 | * is going to be sent to ensure that it matches the one passed to | |
4755 | * bond_choose_output_slave(). (Otherwise, we will account to the wrong | |
4756 | * hash bucket.) | |
4757 | * | |
4758 | * We use the actions from an arbitrary subfacet because they should all | |
4759 | * be equally valid for our purpose. */ | |
4760 | vlan_tci = facet->flow.vlan_tci; | |
4761 | NL_ATTR_FOR_EACH_UNSAFE (a, left, | |
4762 | subfacet->actions, subfacet->actions_len) { | |
4763 | const struct ovs_action_push_vlan *vlan; | |
4764 | struct ofport_dpif *port; | |
4765 | ||
4766 | switch (nl_attr_type(a)) { | |
4767 | case OVS_ACTION_ATTR_OUTPUT: | |
4768 | port = get_odp_port(ofproto, nl_attr_get_u32(a)); | |
4769 | if (port && port->bundle && port->bundle->bond) { | |
4770 | bond_account(port->bundle->bond, &facet->flow, | |
4771 | vlan_tci_to_vid(vlan_tci), n_bytes); | |
4772 | } | |
4773 | break; | |
4774 | ||
4775 | case OVS_ACTION_ATTR_POP_VLAN: | |
4776 | vlan_tci = htons(0); | |
4777 | break; | |
4778 | ||
4779 | case OVS_ACTION_ATTR_PUSH_VLAN: | |
4780 | vlan = nl_attr_get(a); | |
4781 | vlan_tci = vlan->vlan_tci; | |
4782 | break; | |
4783 | } | |
4784 | } | |
4785 | } | |
4786 | ||
4787 | /* Returns true if the only action for 'facet' is to send to the controller. | |
4788 | * (We don't report NetFlow expiration messages for such facets because they | |
4789 | * are just part of the control logic for the network, not real traffic). */ | |
4790 | static bool | |
4791 | facet_is_controller_flow(struct facet *facet) | |
4792 | { | |
4793 | if (facet) { | |
4794 | const struct rule *rule = &facet->rule->up; | |
4795 | const struct ofpact *ofpacts = rule->ofpacts; | |
4796 | size_t ofpacts_len = rule->ofpacts_len; | |
4797 | ||
4798 | if (ofpacts_len > 0 && | |
4799 | ofpacts->type == OFPACT_CONTROLLER && | |
4800 | ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len)) { | |
4801 | return true; | |
4802 | } | |
4803 | } | |
4804 | return false; | |
4805 | } | |
4806 | ||
4807 | /* Folds all of 'facet''s statistics into its rule. Also updates the | |
4808 | * accounting ofhook and emits a NetFlow expiration if appropriate. All of | |
4809 | * 'facet''s statistics in the datapath should have been zeroed and folded into | |
4810 | * its packet and byte counts before this function is called. */ | |
4811 | static void | |
4812 | facet_flush_stats(struct facet *facet) | |
4813 | { | |
4814 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); | |
4815 | struct subfacet *subfacet; | |
4816 | ||
4817 | LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) { | |
4818 | ovs_assert(!subfacet->dp_byte_count); | |
4819 | ovs_assert(!subfacet->dp_packet_count); | |
4820 | } | |
4821 | ||
4822 | facet_push_stats(facet); | |
4823 | if (facet->accounted_bytes < facet->byte_count) { | |
4824 | facet_account(facet); | |
4825 | facet->accounted_bytes = facet->byte_count; | |
4826 | } | |
4827 | ||
4828 | if (ofproto->netflow && !facet_is_controller_flow(facet)) { | |
4829 | struct ofexpired expired; | |
4830 | expired.flow = facet->flow; | |
4831 | expired.packet_count = facet->packet_count; | |
4832 | expired.byte_count = facet->byte_count; | |
4833 | expired.used = facet->used; | |
4834 | netflow_expire(ofproto->netflow, &facet->nf_flow, &expired); | |
4835 | } | |
4836 | ||
4837 | /* Reset counters to prevent double counting if 'facet' ever gets | |
4838 | * reinstalled. */ | |
4839 | facet_reset_counters(facet); | |
4840 | ||
4841 | netflow_flow_clear(&facet->nf_flow); | |
4842 | facet->tcp_flags = 0; | |
4843 | } | |
4844 | ||
4845 | /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'. | |
4846 | * Returns it if found, otherwise a null pointer. | |
4847 | * | |
4848 | * 'hash' must be the return value of flow_hash(flow, 0). | |
4849 | * | |
4850 | * The returned facet might need revalidation; use facet_lookup_valid() | |
4851 | * instead if that is important. */ | |
4852 | static struct facet * | |
4853 | facet_find(struct ofproto_dpif *ofproto, | |
4854 | const struct flow *flow, uint32_t hash) | |
4855 | { | |
4856 | struct facet *facet; | |
4857 | ||
4858 | HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, hash, &ofproto->facets) { | |
4859 | if (flow_equal(flow, &facet->flow)) { | |
4860 | return facet; | |
4861 | } | |
4862 | } | |
4863 | ||
4864 | return NULL; | |
4865 | } | |
4866 | ||
4867 | /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'. | |
4868 | * Returns it if found, otherwise a null pointer. | |
4869 | * | |
4870 | * 'hash' must be the return value of flow_hash(flow, 0). | |
4871 | * | |
4872 | * The returned facet is guaranteed to be valid. */ | |
4873 | static struct facet * | |
4874 | facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow, | |
4875 | uint32_t hash) | |
4876 | { | |
4877 | struct facet *facet; | |
4878 | ||
4879 | facet = facet_find(ofproto, flow, hash); | |
4880 | if (facet | |
4881 | && (ofproto->backer->need_revalidate | |
4882 | || tag_set_intersects(&ofproto->backer->revalidate_set, | |
4883 | facet->tags))) { | |
4884 | facet_revalidate(facet); | |
4885 | ||
4886 | /* facet_revalidate() may have destroyed 'facet'. */ | |
4887 | facet = facet_find(ofproto, flow, hash); | |
4888 | } | |
4889 | ||
4890 | return facet; | |
4891 | } | |
4892 | ||
4893 | /* Return a subfacet from 'facet'. A facet consists of one or more | |
4894 | * subfacets, and this function returns one of them. */ | |
4895 | static struct subfacet *facet_get_subfacet(struct facet *facet) | |
4896 | { | |
4897 | return CONTAINER_OF(list_front(&facet->subfacets), struct subfacet, | |
4898 | list_node); | |
4899 | } | |
4900 | ||
4901 | static const char * | |
4902 | subfacet_path_to_string(enum subfacet_path path) | |
4903 | { | |
4904 | switch (path) { | |
4905 | case SF_NOT_INSTALLED: | |
4906 | return "not installed"; | |
4907 | case SF_FAST_PATH: | |
4908 | return "in fast path"; | |
4909 | case SF_SLOW_PATH: | |
4910 | return "in slow path"; | |
4911 | default: | |
4912 | return "<error>"; | |
4913 | } | |
4914 | } | |
4915 | ||
4916 | /* Returns the path in which a subfacet should be installed if its 'slow' | |
4917 | * member has the specified value. */ | |
4918 | static enum subfacet_path | |
4919 | subfacet_want_path(enum slow_path_reason slow) | |
4920 | { | |
4921 | return slow ? SF_SLOW_PATH : SF_FAST_PATH; | |
4922 | } | |
4923 | ||
4924 | /* Returns true if 'subfacet' needs to have its datapath flow updated, | |
4925 | * supposing that its actions have been recalculated as 'want_actions' and that | |
4926 | * 'slow' is nonzero iff 'subfacet' should be in the slow path. */ | |
4927 | static bool | |
4928 | subfacet_should_install(struct subfacet *subfacet, enum slow_path_reason slow, | |
4929 | const struct ofpbuf *want_actions) | |
4930 | { | |
4931 | enum subfacet_path want_path = subfacet_want_path(slow); | |
4932 | return (want_path != subfacet->path | |
4933 | || (want_path == SF_FAST_PATH | |
4934 | && (subfacet->actions_len != want_actions->size | |
4935 | || memcmp(subfacet->actions, want_actions->data, | |
4936 | subfacet->actions_len)))); | |
4937 | } | |
4938 | ||
4939 | static bool | |
4940 | facet_check_consistency(struct facet *facet) | |
4941 | { | |
4942 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15); | |
4943 | ||
4944 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); | |
4945 | ||
4946 | uint64_t odp_actions_stub[1024 / 8]; | |
4947 | struct ofpbuf odp_actions; | |
4948 | ||
4949 | struct rule_dpif *rule; | |
4950 | struct subfacet *subfacet; | |
4951 | bool may_log = false; | |
4952 | bool ok; | |
4953 | ||
4954 | /* Check the rule for consistency. */ | |
4955 | rule = rule_dpif_lookup(ofproto, &facet->flow); | |
4956 | ok = rule == facet->rule; | |
4957 | if (!ok) { | |
4958 | may_log = !VLOG_DROP_WARN(&rl); | |
4959 | if (may_log) { | |
4960 | struct ds s; | |
4961 | ||
4962 | ds_init(&s); | |
4963 | flow_format(&s, &facet->flow); | |
4964 | ds_put_format(&s, ": facet associated with wrong rule (was " | |
4965 | "table=%"PRIu8",", facet->rule->up.table_id); | |
4966 | cls_rule_format(&facet->rule->up.cr, &s); | |
4967 | ds_put_format(&s, ") (should have been table=%"PRIu8",", | |
4968 | rule->up.table_id); | |
4969 | cls_rule_format(&rule->up.cr, &s); | |
4970 | ds_put_char(&s, ')'); | |
4971 | ||
4972 | VLOG_WARN("%s", ds_cstr(&s)); | |
4973 | ds_destroy(&s); | |
4974 | } | |
4975 | } | |
4976 | ||
4977 | /* Check the datapath actions for consistency. */ | |
4978 | ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub); | |
4979 | LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) { | |
4980 | enum subfacet_path want_path; | |
4981 | struct action_xlate_ctx ctx; | |
4982 | struct ds s; | |
4983 | ||
4984 | action_xlate_ctx_init(&ctx, ofproto, &facet->flow, | |
4985 | &subfacet->initial_vals, rule, 0, NULL); | |
4986 | xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, | |
4987 | &odp_actions); | |
4988 | ||
4989 | if (subfacet->path == SF_NOT_INSTALLED) { | |
4990 | /* This only happens if the datapath reported an error when we | |
4991 | * tried to install the flow. Don't flag another error here. */ | |
4992 | continue; | |
4993 | } | |
4994 | ||
4995 | want_path = subfacet_want_path(subfacet->slow); | |
4996 | if (want_path == SF_SLOW_PATH && subfacet->path == SF_SLOW_PATH) { | |
4997 | /* The actions for slow-path flows may legitimately vary from one | |
4998 | * packet to the next. We're done. */ | |
4999 | continue; | |
5000 | } | |
5001 | ||
5002 | if (!subfacet_should_install(subfacet, subfacet->slow, &odp_actions)) { | |
5003 | continue; | |
5004 | } | |
5005 | ||
5006 | /* Inconsistency! */ | |
5007 | if (ok) { | |
5008 | may_log = !VLOG_DROP_WARN(&rl); | |
5009 | ok = false; | |
5010 | } | |
5011 | if (!may_log) { | |
5012 | /* Rate-limited, skip reporting. */ | |
5013 | continue; | |
5014 | } | |
5015 | ||
5016 | ds_init(&s); | |
5017 | odp_flow_key_format(subfacet->key, subfacet->key_len, &s); | |
5018 | ||
5019 | ds_put_cstr(&s, ": inconsistency in subfacet"); | |
5020 | if (want_path != subfacet->path) { | |
5021 | enum odp_key_fitness fitness = subfacet->key_fitness; | |
5022 | ||
5023 | ds_put_format(&s, " (%s, fitness=%s)", | |
5024 | subfacet_path_to_string(subfacet->path), | |
5025 | odp_key_fitness_to_string(fitness)); | |
5026 | ds_put_format(&s, " (should have been %s)", | |
5027 | subfacet_path_to_string(want_path)); | |
5028 | } else if (want_path == SF_FAST_PATH) { | |
5029 | ds_put_cstr(&s, " (actions were: "); | |
5030 | format_odp_actions(&s, subfacet->actions, | |
5031 | subfacet->actions_len); | |
5032 | ds_put_cstr(&s, ") (correct actions: "); | |
5033 | format_odp_actions(&s, odp_actions.data, odp_actions.size); | |
5034 | ds_put_char(&s, ')'); | |
5035 | } else { | |
5036 | ds_put_cstr(&s, " (actions: "); | |
5037 | format_odp_actions(&s, subfacet->actions, | |
5038 | subfacet->actions_len); | |
5039 | ds_put_char(&s, ')'); | |
5040 | } | |
5041 | VLOG_WARN("%s", ds_cstr(&s)); | |
5042 | ds_destroy(&s); | |
5043 | } | |
5044 | ofpbuf_uninit(&odp_actions); | |
5045 | ||
5046 | return ok; | |
5047 | } | |
5048 | ||
5049 | /* Re-searches the classifier for 'facet': | |
5050 | * | |
5051 | * - If the rule found is different from 'facet''s current rule, moves | |
5052 | * 'facet' to the new rule and recompiles its actions. | |
5053 | * | |
5054 | * - If the rule found is the same as 'facet''s current rule, leaves 'facet' | |
5055 | * where it is and recompiles its actions anyway. | |
5056 | * | |
5057 | * - If any of 'facet''s subfacets correspond to a new flow according to | |
5058 | * ofproto_receive(), 'facet' is removed. */ | |
5059 | static void | |
5060 | facet_revalidate(struct facet *facet) | |
5061 | { | |
5062 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); | |
5063 | struct actions { | |
5064 | struct nlattr *odp_actions; | |
5065 | size_t actions_len; | |
5066 | }; | |
5067 | struct actions *new_actions; | |
5068 | ||
5069 | struct action_xlate_ctx ctx; | |
5070 | uint64_t odp_actions_stub[1024 / 8]; | |
5071 | struct ofpbuf odp_actions; | |
5072 | ||
5073 | struct rule_dpif *new_rule; | |
5074 | struct subfacet *subfacet; | |
5075 | int i; | |
5076 | ||
5077 | COVERAGE_INC(facet_revalidate); | |
5078 | ||
5079 | /* Check that child subfacets still correspond to this facet. Tunnel | |
5080 | * configuration changes could cause a subfacet's OpenFlow in_port to | |
5081 | * change. */ | |
5082 | LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) { | |
5083 | struct ofproto_dpif *recv_ofproto; | |
5084 | struct flow recv_flow; | |
5085 | int error; | |
5086 | ||
5087 | error = ofproto_receive(ofproto->backer, NULL, subfacet->key, | |
5088 | subfacet->key_len, &recv_flow, NULL, | |
5089 | &recv_ofproto, NULL, NULL); | |
5090 | if (error | |
5091 | || recv_ofproto != ofproto | |
5092 | || memcmp(&recv_flow, &facet->flow, sizeof recv_flow)) { | |
5093 | facet_remove(facet); | |
5094 | return; | |
5095 | } | |
5096 | } | |
5097 | ||
5098 | new_rule = rule_dpif_lookup(ofproto, &facet->flow); | |
5099 | ||
5100 | /* Calculate new datapath actions. | |
5101 | * | |
5102 | * We do not modify any 'facet' state yet, because we might need to, e.g., | |
5103 | * emit a NetFlow expiration and, if so, we need to have the old state | |
5104 | * around to properly compose it. */ | |
5105 | ||
5106 | /* If the datapath actions changed or the installability changed, | |
5107 | * then we need to talk to the datapath. */ | |
5108 | i = 0; | |
5109 | new_actions = NULL; | |
5110 | memset(&ctx, 0, sizeof ctx); | |
5111 | ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub); | |
5112 | LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) { | |
5113 | enum slow_path_reason slow; | |
5114 | ||
5115 | action_xlate_ctx_init(&ctx, ofproto, &facet->flow, | |
5116 | &subfacet->initial_vals, new_rule, 0, NULL); | |
5117 | xlate_actions(&ctx, new_rule->up.ofpacts, new_rule->up.ofpacts_len, | |
5118 | &odp_actions); | |
5119 | ||
5120 | slow = (subfacet->slow & SLOW_MATCH) | ctx.slow; | |
5121 | if (subfacet_should_install(subfacet, slow, &odp_actions)) { | |
5122 | struct dpif_flow_stats stats; | |
5123 | ||
5124 | subfacet_install(subfacet, | |
5125 | odp_actions.data, odp_actions.size, &stats, slow); | |
5126 | subfacet_update_stats(subfacet, &stats); | |
5127 | ||
5128 | if (!new_actions) { | |
5129 | new_actions = xcalloc(list_size(&facet->subfacets), | |
5130 | sizeof *new_actions); | |
5131 | } | |
5132 | new_actions[i].odp_actions = xmemdup(odp_actions.data, | |
5133 | odp_actions.size); | |
5134 | new_actions[i].actions_len = odp_actions.size; | |
5135 | } | |
5136 | ||
5137 | i++; | |
5138 | } | |
5139 | ofpbuf_uninit(&odp_actions); | |
5140 | ||
5141 | if (new_actions) { | |
5142 | facet_flush_stats(facet); | |
5143 | } | |
5144 | ||
5145 | /* Update 'facet' now that we've taken care of all the old state. */ | |
5146 | facet->tags = ctx.tags; | |
5147 | facet->nf_flow.output_iface = ctx.nf_output_iface; | |
5148 | facet->has_learn = ctx.has_learn; | |
5149 | facet->has_normal = ctx.has_normal; | |
5150 | facet->has_fin_timeout = ctx.has_fin_timeout; | |
5151 | facet->mirrors = ctx.mirrors; | |
5152 | ||
5153 | i = 0; | |
5154 | LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) { | |
5155 | subfacet->slow = (subfacet->slow & SLOW_MATCH) | ctx.slow; | |
5156 | ||
5157 | if (new_actions && new_actions[i].odp_actions) { | |
5158 | free(subfacet->actions); | |
5159 | subfacet->actions = new_actions[i].odp_actions; | |
5160 | subfacet->actions_len = new_actions[i].actions_len; | |
5161 | } | |
5162 | i++; | |
5163 | } | |
5164 | free(new_actions); | |
5165 | ||
5166 | if (facet->rule != new_rule) { | |
5167 | COVERAGE_INC(facet_changed_rule); | |
5168 | list_remove(&facet->list_node); | |
5169 | list_push_back(&new_rule->facets, &facet->list_node); | |
5170 | facet->rule = new_rule; | |
5171 | facet->used = new_rule->up.created; | |
5172 | facet->prev_used = facet->used; | |
5173 | } | |
5174 | } | |
5175 | ||
5176 | /* Updates 'facet''s used time. Caller is responsible for calling | |
5177 | * facet_push_stats() to update the flows which 'facet' resubmits into. */ | |
5178 | static void | |
5179 | facet_update_time(struct facet *facet, long long int used) | |
5180 | { | |
5181 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); | |
5182 | if (used > facet->used) { | |
5183 | facet->used = used; | |
5184 | ofproto_rule_update_used(&facet->rule->up, used); | |
5185 | netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used); | |
5186 | } | |
5187 | } | |
5188 | ||
5189 | static void | |
5190 | facet_reset_counters(struct facet *facet) | |
5191 | { | |
5192 | facet->packet_count = 0; | |
5193 | facet->byte_count = 0; | |
5194 | facet->prev_packet_count = 0; | |
5195 | facet->prev_byte_count = 0; | |
5196 | facet->accounted_bytes = 0; | |
5197 | } | |
5198 | ||
5199 | static void | |
5200 | facet_push_stats(struct facet *facet) | |
5201 | { | |
5202 | struct dpif_flow_stats stats; | |
5203 | ||
5204 | ovs_assert(facet->packet_count >= facet->prev_packet_count); | |
5205 | ovs_assert(facet->byte_count >= facet->prev_byte_count); | |
5206 | ovs_assert(facet->used >= facet->prev_used); | |
5207 | ||
5208 | stats.n_packets = facet->packet_count - facet->prev_packet_count; | |
5209 | stats.n_bytes = facet->byte_count - facet->prev_byte_count; | |
5210 | stats.used = facet->used; | |
5211 | stats.tcp_flags = 0; | |
5212 | ||
5213 | if (stats.n_packets || stats.n_bytes || facet->used > facet->prev_used) { | |
5214 | facet->prev_packet_count = facet->packet_count; | |
5215 | facet->prev_byte_count = facet->byte_count; | |
5216 | facet->prev_used = facet->used; | |
5217 | ||
5218 | rule_credit_stats(facet->rule, &stats); | |
5219 | flow_push_stats(facet, &stats); | |
5220 | ||
5221 | update_mirror_stats(ofproto_dpif_cast(facet->rule->up.ofproto), | |
5222 | facet->mirrors, stats.n_packets, stats.n_bytes); | |
5223 | } | |
5224 | } | |
5225 | ||
5226 | static void | |
5227 | push_all_stats__(bool run_fast) | |
5228 | { | |
5229 | static long long int rl = LLONG_MIN; | |
5230 | struct ofproto_dpif *ofproto; | |
5231 | ||
5232 | if (time_msec() < rl) { | |
5233 | return; | |
5234 | } | |
5235 | ||
5236 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
5237 | struct facet *facet; | |
5238 | ||
5239 | HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) { | |
5240 | facet_push_stats(facet); | |
5241 | if (run_fast) { | |
5242 | run_fast_rl(); | |
5243 | } | |
5244 | } | |
5245 | } | |
5246 | ||
5247 | rl = time_msec() + 100; | |
5248 | } | |
5249 | ||
5250 | static void | |
5251 | push_all_stats(void) | |
5252 | { | |
5253 | push_all_stats__(true); | |
5254 | } | |
5255 | ||
5256 | static void | |
5257 | rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats) | |
5258 | { | |
5259 | rule->packet_count += stats->n_packets; | |
5260 | rule->byte_count += stats->n_bytes; | |
5261 | ofproto_rule_update_used(&rule->up, stats->used); | |
5262 | } | |
5263 | ||
5264 | /* Pushes flow statistics to the rules which 'facet->flow' resubmits | |
5265 | * into given 'facet->rule''s actions and mirrors. */ | |
5266 | static void | |
5267 | flow_push_stats(struct facet *facet, const struct dpif_flow_stats *stats) | |
5268 | { | |
5269 | struct rule_dpif *rule = facet->rule; | |
5270 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
5271 | struct subfacet *subfacet = facet_get_subfacet(facet); | |
5272 | struct action_xlate_ctx ctx; | |
5273 | ||
5274 | ofproto_rule_update_used(&rule->up, stats->used); | |
5275 | ||
5276 | action_xlate_ctx_init(&ctx, ofproto, &facet->flow, | |
5277 | &subfacet->initial_vals, rule, 0, NULL); | |
5278 | ctx.resubmit_stats = stats; | |
5279 | xlate_actions_for_side_effects(&ctx, rule->up.ofpacts, | |
5280 | rule->up.ofpacts_len); | |
5281 | } | |
5282 | \f | |
5283 | /* Subfacets. */ | |
5284 | ||
5285 | static struct subfacet * | |
5286 | subfacet_find(struct ofproto_dpif *ofproto, | |
5287 | const struct nlattr *key, size_t key_len, uint32_t key_hash) | |
5288 | { | |
5289 | struct subfacet *subfacet; | |
5290 | ||
5291 | HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash, | |
5292 | &ofproto->subfacets) { | |
5293 | if (subfacet->key_len == key_len | |
5294 | && !memcmp(key, subfacet->key, key_len)) { | |
5295 | return subfacet; | |
5296 | } | |
5297 | } | |
5298 | ||
5299 | return NULL; | |
5300 | } | |
5301 | ||
5302 | /* Searches 'facet' (within 'ofproto') for a subfacet with the specified | |
5303 | * 'key_fitness', 'key', and 'key_len' members in 'miss'. Returns the | |
5304 | * existing subfacet if there is one, otherwise creates and returns a | |
5305 | * new subfacet. | |
5306 | * | |
5307 | * If the returned subfacet is new, then subfacet->actions will be NULL, in | |
5308 | * which case the caller must populate the actions with | |
5309 | * subfacet_make_actions(). */ | |
5310 | static struct subfacet * | |
5311 | subfacet_create(struct facet *facet, struct flow_miss *miss, | |
5312 | long long int now) | |
5313 | { | |
5314 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); | |
5315 | enum odp_key_fitness key_fitness = miss->key_fitness; | |
5316 | const struct nlattr *key = miss->key; | |
5317 | size_t key_len = miss->key_len; | |
5318 | uint32_t key_hash; | |
5319 | struct subfacet *subfacet; | |
5320 | ||
5321 | key_hash = odp_flow_key_hash(key, key_len); | |
5322 | ||
5323 | if (list_is_empty(&facet->subfacets)) { | |
5324 | subfacet = &facet->one_subfacet; | |
5325 | } else { | |
5326 | subfacet = subfacet_find(ofproto, key, key_len, key_hash); | |
5327 | if (subfacet) { | |
5328 | if (subfacet->facet == facet) { | |
5329 | return subfacet; | |
5330 | } | |
5331 | ||
5332 | /* This shouldn't happen. */ | |
5333 | VLOG_ERR_RL(&rl, "subfacet with wrong facet"); | |
5334 | subfacet_destroy(subfacet); | |
5335 | } | |
5336 | ||
5337 | subfacet = xmalloc(sizeof *subfacet); | |
5338 | } | |
5339 | ||
5340 | hmap_insert(&ofproto->subfacets, &subfacet->hmap_node, key_hash); | |
5341 | list_push_back(&facet->subfacets, &subfacet->list_node); | |
5342 | subfacet->facet = facet; | |
5343 | subfacet->key_fitness = key_fitness; | |
5344 | subfacet->key = xmemdup(key, key_len); | |
5345 | subfacet->key_len = key_len; | |
5346 | subfacet->used = now; | |
5347 | subfacet->created = now; | |
5348 | subfacet->dp_packet_count = 0; | |
5349 | subfacet->dp_byte_count = 0; | |
5350 | subfacet->actions_len = 0; | |
5351 | subfacet->actions = NULL; | |
5352 | subfacet->slow = (subfacet->key_fitness == ODP_FIT_TOO_LITTLE | |
5353 | ? SLOW_MATCH | |
5354 | : 0); | |
5355 | subfacet->path = SF_NOT_INSTALLED; | |
5356 | subfacet->initial_vals = miss->initial_vals; | |
5357 | subfacet->odp_in_port = miss->odp_in_port; | |
5358 | ||
5359 | ofproto->subfacet_add_count++; | |
5360 | return subfacet; | |
5361 | } | |
5362 | ||
5363 | /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from | |
5364 | * its facet within 'ofproto', and frees it. */ | |
5365 | static void | |
5366 | subfacet_destroy__(struct subfacet *subfacet) | |
5367 | { | |
5368 | struct facet *facet = subfacet->facet; | |
5369 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); | |
5370 | ||
5371 | /* Update ofproto stats before uninstall the subfacet. */ | |
5372 | ofproto->subfacet_del_count++; | |
5373 | ofproto->total_subfacet_life_span += (time_msec() - subfacet->created); | |
5374 | ||
5375 | subfacet_uninstall(subfacet); | |
5376 | hmap_remove(&ofproto->subfacets, &subfacet->hmap_node); | |
5377 | list_remove(&subfacet->list_node); | |
5378 | free(subfacet->key); | |
5379 | free(subfacet->actions); | |
5380 | if (subfacet != &facet->one_subfacet) { | |
5381 | free(subfacet); | |
5382 | } | |
5383 | } | |
5384 | ||
5385 | /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the | |
5386 | * last remaining subfacet in its facet destroys the facet too. */ | |
5387 | static void | |
5388 | subfacet_destroy(struct subfacet *subfacet) | |
5389 | { | |
5390 | struct facet *facet = subfacet->facet; | |
5391 | ||
5392 | if (list_is_singleton(&facet->subfacets)) { | |
5393 | /* facet_remove() needs at least one subfacet (it will remove it). */ | |
5394 | facet_remove(facet); | |
5395 | } else { | |
5396 | subfacet_destroy__(subfacet); | |
5397 | } | |
5398 | } | |
5399 | ||
5400 | static void | |
5401 | subfacet_destroy_batch(struct ofproto_dpif *ofproto, | |
5402 | struct subfacet **subfacets, int n) | |
5403 | { | |
5404 | struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH]; | |
5405 | struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH]; | |
5406 | struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH]; | |
5407 | int i; | |
5408 | ||
5409 | for (i = 0; i < n; i++) { | |
5410 | ops[i].type = DPIF_OP_FLOW_DEL; | |
5411 | ops[i].u.flow_del.key = subfacets[i]->key; | |
5412 | ops[i].u.flow_del.key_len = subfacets[i]->key_len; | |
5413 | ops[i].u.flow_del.stats = &stats[i]; | |
5414 | opsp[i] = &ops[i]; | |
5415 | } | |
5416 | ||
5417 | dpif_operate(ofproto->backer->dpif, opsp, n); | |
5418 | for (i = 0; i < n; i++) { | |
5419 | subfacet_reset_dp_stats(subfacets[i], &stats[i]); | |
5420 | subfacets[i]->path = SF_NOT_INSTALLED; | |
5421 | subfacet_destroy(subfacets[i]); | |
5422 | run_fast_rl(); | |
5423 | } | |
5424 | } | |
5425 | ||
5426 | /* Composes the datapath actions for 'subfacet' based on its rule's actions. | |
5427 | * Translates the actions into 'odp_actions', which the caller must have | |
5428 | * initialized and is responsible for uninitializing. */ | |
5429 | static void | |
5430 | subfacet_make_actions(struct subfacet *subfacet, const struct ofpbuf *packet, | |
5431 | struct ofpbuf *odp_actions) | |
5432 | { | |
5433 | struct facet *facet = subfacet->facet; | |
5434 | struct rule_dpif *rule = facet->rule; | |
5435 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
5436 | ||
5437 | struct action_xlate_ctx ctx; | |
5438 | ||
5439 | action_xlate_ctx_init(&ctx, ofproto, &facet->flow, | |
5440 | &subfacet->initial_vals, rule, 0, packet); | |
5441 | xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, odp_actions); | |
5442 | facet->tags = ctx.tags; | |
5443 | facet->has_learn = ctx.has_learn; | |
5444 | facet->has_normal = ctx.has_normal; | |
5445 | facet->has_fin_timeout = ctx.has_fin_timeout; | |
5446 | facet->nf_flow.output_iface = ctx.nf_output_iface; | |
5447 | facet->mirrors = ctx.mirrors; | |
5448 | ||
5449 | subfacet->slow = (subfacet->slow & SLOW_MATCH) | ctx.slow; | |
5450 | if (subfacet->actions_len != odp_actions->size | |
5451 | || memcmp(subfacet->actions, odp_actions->data, odp_actions->size)) { | |
5452 | free(subfacet->actions); | |
5453 | subfacet->actions_len = odp_actions->size; | |
5454 | subfacet->actions = xmemdup(odp_actions->data, odp_actions->size); | |
5455 | } | |
5456 | } | |
5457 | ||
5458 | /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len' | |
5459 | * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters | |
5460 | * in the datapath will be zeroed and 'stats' will be updated with traffic new | |
5461 | * since 'subfacet' was last updated. | |
5462 | * | |
5463 | * Returns 0 if successful, otherwise a positive errno value. */ | |
5464 | static int | |
5465 | subfacet_install(struct subfacet *subfacet, | |
5466 | const struct nlattr *actions, size_t actions_len, | |
5467 | struct dpif_flow_stats *stats, | |
5468 | enum slow_path_reason slow) | |
5469 | { | |
5470 | struct facet *facet = subfacet->facet; | |
5471 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); | |
5472 | enum subfacet_path path = subfacet_want_path(slow); | |
5473 | uint64_t slow_path_stub[128 / 8]; | |
5474 | enum dpif_flow_put_flags flags; | |
5475 | int ret; | |
5476 | ||
5477 | flags = DPIF_FP_CREATE | DPIF_FP_MODIFY; | |
5478 | if (stats) { | |
5479 | flags |= DPIF_FP_ZERO_STATS; | |
5480 | } | |
5481 | ||
5482 | if (path == SF_SLOW_PATH) { | |
5483 | compose_slow_path(ofproto, &facet->flow, slow, | |
5484 | slow_path_stub, sizeof slow_path_stub, | |
5485 | &actions, &actions_len); | |
5486 | } | |
5487 | ||
5488 | ret = dpif_flow_put(ofproto->backer->dpif, flags, subfacet->key, | |
5489 | subfacet->key_len, actions, actions_len, stats); | |
5490 | ||
5491 | if (stats) { | |
5492 | subfacet_reset_dp_stats(subfacet, stats); | |
5493 | } | |
5494 | ||
5495 | if (!ret) { | |
5496 | subfacet->path = path; | |
5497 | } | |
5498 | return ret; | |
5499 | } | |
5500 | ||
5501 | static int | |
5502 | subfacet_reinstall(struct subfacet *subfacet, struct dpif_flow_stats *stats) | |
5503 | { | |
5504 | return subfacet_install(subfacet, subfacet->actions, subfacet->actions_len, | |
5505 | stats, subfacet->slow); | |
5506 | } | |
5507 | ||
5508 | /* If 'subfacet' is installed in the datapath, uninstalls it. */ | |
5509 | static void | |
5510 | subfacet_uninstall(struct subfacet *subfacet) | |
5511 | { | |
5512 | if (subfacet->path != SF_NOT_INSTALLED) { | |
5513 | struct rule_dpif *rule = subfacet->facet->rule; | |
5514 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
5515 | struct dpif_flow_stats stats; | |
5516 | int error; | |
5517 | ||
5518 | error = dpif_flow_del(ofproto->backer->dpif, subfacet->key, | |
5519 | subfacet->key_len, &stats); | |
5520 | subfacet_reset_dp_stats(subfacet, &stats); | |
5521 | if (!error) { | |
5522 | subfacet_update_stats(subfacet, &stats); | |
5523 | } | |
5524 | subfacet->path = SF_NOT_INSTALLED; | |
5525 | } else { | |
5526 | ovs_assert(subfacet->dp_packet_count == 0); | |
5527 | ovs_assert(subfacet->dp_byte_count == 0); | |
5528 | } | |
5529 | } | |
5530 | ||
5531 | /* Resets 'subfacet''s datapath statistics counters. This should be called | |
5532 | * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is | |
5533 | * non-null, it should contain the statistics returned by dpif when 'subfacet' | |
5534 | * was reset in the datapath. 'stats' will be modified to include only | |
5535 | * statistics new since 'subfacet' was last updated. */ | |
5536 | static void | |
5537 | subfacet_reset_dp_stats(struct subfacet *subfacet, | |
5538 | struct dpif_flow_stats *stats) | |
5539 | { | |
5540 | if (stats | |
5541 | && subfacet->dp_packet_count <= stats->n_packets | |
5542 | && subfacet->dp_byte_count <= stats->n_bytes) { | |
5543 | stats->n_packets -= subfacet->dp_packet_count; | |
5544 | stats->n_bytes -= subfacet->dp_byte_count; | |
5545 | } | |
5546 | ||
5547 | subfacet->dp_packet_count = 0; | |
5548 | subfacet->dp_byte_count = 0; | |
5549 | } | |
5550 | ||
5551 | /* Updates 'subfacet''s used time. The caller is responsible for calling | |
5552 | * facet_push_stats() to update the flows which 'subfacet' resubmits into. */ | |
5553 | static void | |
5554 | subfacet_update_time(struct subfacet *subfacet, long long int used) | |
5555 | { | |
5556 | if (used > subfacet->used) { | |
5557 | subfacet->used = used; | |
5558 | facet_update_time(subfacet->facet, used); | |
5559 | } | |
5560 | } | |
5561 | ||
5562 | /* Folds the statistics from 'stats' into the counters in 'subfacet'. | |
5563 | * | |
5564 | * Because of the meaning of a subfacet's counters, it only makes sense to do | |
5565 | * this if 'stats' are not tracked in the datapath, that is, if 'stats' | |
5566 | * represents a packet that was sent by hand or if it represents statistics | |
5567 | * that have been cleared out of the datapath. */ | |
5568 | static void | |
5569 | subfacet_update_stats(struct subfacet *subfacet, | |
5570 | const struct dpif_flow_stats *stats) | |
5571 | { | |
5572 | if (stats->n_packets || stats->used > subfacet->used) { | |
5573 | struct facet *facet = subfacet->facet; | |
5574 | ||
5575 | subfacet_update_time(subfacet, stats->used); | |
5576 | facet->packet_count += stats->n_packets; | |
5577 | facet->byte_count += stats->n_bytes; | |
5578 | facet->tcp_flags |= stats->tcp_flags; | |
5579 | netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags); | |
5580 | } | |
5581 | } | |
5582 | \f | |
5583 | /* Rules. */ | |
5584 | ||
5585 | static struct rule_dpif * | |
5586 | rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow) | |
5587 | { | |
5588 | struct rule_dpif *rule; | |
5589 | ||
5590 | rule = rule_dpif_lookup__(ofproto, flow, 0); | |
5591 | if (rule) { | |
5592 | return rule; | |
5593 | } | |
5594 | ||
5595 | return rule_dpif_miss_rule(ofproto, flow); | |
5596 | } | |
5597 | ||
5598 | static struct rule_dpif * | |
5599 | rule_dpif_lookup__(struct ofproto_dpif *ofproto, const struct flow *flow, | |
5600 | uint8_t table_id) | |
5601 | { | |
5602 | struct cls_rule *cls_rule; | |
5603 | struct classifier *cls; | |
5604 | ||
5605 | if (table_id >= N_TABLES) { | |
5606 | return NULL; | |
5607 | } | |
5608 | ||
5609 | cls = &ofproto->up.tables[table_id].cls; | |
5610 | if (flow->nw_frag & FLOW_NW_FRAG_ANY | |
5611 | && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) { | |
5612 | /* For OFPC_NORMAL frag_handling, we must pretend that transport ports | |
5613 | * are unavailable. */ | |
5614 | struct flow ofpc_normal_flow = *flow; | |
5615 | ofpc_normal_flow.tp_src = htons(0); | |
5616 | ofpc_normal_flow.tp_dst = htons(0); | |
5617 | cls_rule = classifier_lookup(cls, &ofpc_normal_flow); | |
5618 | } else { | |
5619 | cls_rule = classifier_lookup(cls, flow); | |
5620 | } | |
5621 | return rule_dpif_cast(rule_from_cls_rule(cls_rule)); | |
5622 | } | |
5623 | ||
5624 | static struct rule_dpif * | |
5625 | rule_dpif_miss_rule(struct ofproto_dpif *ofproto, const struct flow *flow) | |
5626 | { | |
5627 | struct ofport_dpif *port; | |
5628 | ||
5629 | port = get_ofp_port(ofproto, flow->in_port); | |
5630 | if (!port) { | |
5631 | VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, flow->in_port); | |
5632 | return ofproto->miss_rule; | |
5633 | } | |
5634 | ||
5635 | if (port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN) { | |
5636 | return ofproto->no_packet_in_rule; | |
5637 | } | |
5638 | return ofproto->miss_rule; | |
5639 | } | |
5640 | ||
5641 | static void | |
5642 | complete_operation(struct rule_dpif *rule) | |
5643 | { | |
5644 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
5645 | ||
5646 | rule_invalidate(rule); | |
5647 | if (clogged) { | |
5648 | struct dpif_completion *c = xmalloc(sizeof *c); | |
5649 | c->op = rule->up.pending; | |
5650 | list_push_back(&ofproto->completions, &c->list_node); | |
5651 | } else { | |
5652 | ofoperation_complete(rule->up.pending, 0); | |
5653 | } | |
5654 | } | |
5655 | ||
5656 | static struct rule * | |
5657 | rule_alloc(void) | |
5658 | { | |
5659 | struct rule_dpif *rule = xmalloc(sizeof *rule); | |
5660 | return &rule->up; | |
5661 | } | |
5662 | ||
5663 | static void | |
5664 | rule_dealloc(struct rule *rule_) | |
5665 | { | |
5666 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
5667 | free(rule); | |
5668 | } | |
5669 | ||
5670 | static enum ofperr | |
5671 | rule_construct(struct rule *rule_) | |
5672 | { | |
5673 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
5674 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
5675 | struct rule_dpif *victim; | |
5676 | uint8_t table_id; | |
5677 | ||
5678 | rule->packet_count = 0; | |
5679 | rule->byte_count = 0; | |
5680 | ||
5681 | victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending)); | |
5682 | if (victim && !list_is_empty(&victim->facets)) { | |
5683 | struct facet *facet; | |
5684 | ||
5685 | rule->facets = victim->facets; | |
5686 | list_moved(&rule->facets); | |
5687 | LIST_FOR_EACH (facet, list_node, &rule->facets) { | |
5688 | /* XXX: We're only clearing our local counters here. It's possible | |
5689 | * that quite a few packets are unaccounted for in the datapath | |
5690 | * statistics. These will be accounted to the new rule instead of | |
5691 | * cleared as required. This could be fixed by clearing out the | |
5692 | * datapath statistics for this facet, but currently it doesn't | |
5693 | * seem worth it. */ | |
5694 | facet_reset_counters(facet); | |
5695 | facet->rule = rule; | |
5696 | } | |
5697 | } else { | |
5698 | /* Must avoid list_moved() in this case. */ | |
5699 | list_init(&rule->facets); | |
5700 | } | |
5701 | ||
5702 | table_id = rule->up.table_id; | |
5703 | if (victim) { | |
5704 | rule->tag = victim->tag; | |
5705 | } else if (table_id == 0) { | |
5706 | rule->tag = 0; | |
5707 | } else { | |
5708 | struct flow flow; | |
5709 | ||
5710 | miniflow_expand(&rule->up.cr.match.flow, &flow); | |
5711 | rule->tag = rule_calculate_tag(&flow, &rule->up.cr.match.mask, | |
5712 | ofproto->tables[table_id].basis); | |
5713 | } | |
5714 | ||
5715 | complete_operation(rule); | |
5716 | return 0; | |
5717 | } | |
5718 | ||
5719 | static void | |
5720 | rule_destruct(struct rule *rule_) | |
5721 | { | |
5722 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
5723 | struct facet *facet, *next_facet; | |
5724 | ||
5725 | LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) { | |
5726 | facet_revalidate(facet); | |
5727 | } | |
5728 | ||
5729 | complete_operation(rule); | |
5730 | } | |
5731 | ||
5732 | static void | |
5733 | rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes) | |
5734 | { | |
5735 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
5736 | ||
5737 | /* push_all_stats() can handle flow misses which, when using the learn | |
5738 | * action, can cause rules to be added and deleted. This can corrupt our | |
5739 | * caller's datastructures which assume that rule_get_stats() doesn't have | |
5740 | * an impact on the flow table. To be safe, we disable miss handling. */ | |
5741 | push_all_stats__(false); | |
5742 | ||
5743 | /* Start from historical data for 'rule' itself that are no longer tracked | |
5744 | * in facets. This counts, for example, facets that have expired. */ | |
5745 | *packets = rule->packet_count; | |
5746 | *bytes = rule->byte_count; | |
5747 | } | |
5748 | ||
5749 | static void | |
5750 | rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow, | |
5751 | struct ofpbuf *packet) | |
5752 | { | |
5753 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
5754 | struct initial_vals initial_vals; | |
5755 | struct dpif_flow_stats stats; | |
5756 | struct action_xlate_ctx ctx; | |
5757 | uint64_t odp_actions_stub[1024 / 8]; | |
5758 | struct ofpbuf odp_actions; | |
5759 | ||
5760 | dpif_flow_stats_extract(flow, packet, time_msec(), &stats); | |
5761 | rule_credit_stats(rule, &stats); | |
5762 | ||
5763 | initial_vals.vlan_tci = flow->vlan_tci; | |
5764 | ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub); | |
5765 | action_xlate_ctx_init(&ctx, ofproto, flow, &initial_vals, | |
5766 | rule, stats.tcp_flags, packet); | |
5767 | ctx.resubmit_stats = &stats; | |
5768 | xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, &odp_actions); | |
5769 | ||
5770 | execute_odp_actions(ofproto, flow, odp_actions.data, | |
5771 | odp_actions.size, packet); | |
5772 | ||
5773 | ofpbuf_uninit(&odp_actions); | |
5774 | } | |
5775 | ||
5776 | static enum ofperr | |
5777 | rule_execute(struct rule *rule, const struct flow *flow, | |
5778 | struct ofpbuf *packet) | |
5779 | { | |
5780 | rule_dpif_execute(rule_dpif_cast(rule), flow, packet); | |
5781 | ofpbuf_delete(packet); | |
5782 | return 0; | |
5783 | } | |
5784 | ||
5785 | static void | |
5786 | rule_modify_actions(struct rule *rule_) | |
5787 | { | |
5788 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
5789 | ||
5790 | complete_operation(rule); | |
5791 | } | |
5792 | \f | |
5793 | /* Sends 'packet' out 'ofport'. | |
5794 | * May modify 'packet'. | |
5795 | * Returns 0 if successful, otherwise a positive errno value. */ | |
5796 | static int | |
5797 | send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet) | |
5798 | { | |
5799 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
5800 | uint64_t odp_actions_stub[1024 / 8]; | |
5801 | struct ofpbuf key, odp_actions; | |
5802 | struct dpif_flow_stats stats; | |
5803 | struct odputil_keybuf keybuf; | |
5804 | struct ofpact_output output; | |
5805 | struct action_xlate_ctx ctx; | |
5806 | struct flow flow; | |
5807 | int error; | |
5808 | ||
5809 | ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub); | |
5810 | ofpbuf_use_stack(&key, &keybuf, sizeof keybuf); | |
5811 | ||
5812 | /* Use OFPP_NONE as the in_port to avoid special packet processing. */ | |
5813 | flow_extract(packet, 0, 0, NULL, OFPP_NONE, &flow); | |
5814 | odp_flow_key_from_flow(&key, &flow, ofp_port_to_odp_port(ofproto, | |
5815 | OFPP_LOCAL)); | |
5816 | dpif_flow_stats_extract(&flow, packet, time_msec(), &stats); | |
5817 | ||
5818 | ofpact_init(&output.ofpact, OFPACT_OUTPUT, sizeof output); | |
5819 | output.port = ofport->up.ofp_port; | |
5820 | output.max_len = 0; | |
5821 | ||
5822 | action_xlate_ctx_init(&ctx, ofproto, &flow, NULL, NULL, 0, packet); | |
5823 | ctx.resubmit_stats = &stats; | |
5824 | xlate_actions(&ctx, &output.ofpact, sizeof output, &odp_actions); | |
5825 | ||
5826 | error = dpif_execute(ofproto->backer->dpif, | |
5827 | key.data, key.size, | |
5828 | odp_actions.data, odp_actions.size, | |
5829 | packet); | |
5830 | ofpbuf_uninit(&odp_actions); | |
5831 | ||
5832 | if (error) { | |
5833 | VLOG_WARN_RL(&rl, "%s: failed to send packet on port %s (%s)", | |
5834 | ofproto->up.name, netdev_get_name(ofport->up.netdev), | |
5835 | strerror(error)); | |
5836 | } | |
5837 | ||
5838 | ofproto->stats.tx_packets++; | |
5839 | ofproto->stats.tx_bytes += packet->size; | |
5840 | return error; | |
5841 | } | |
5842 | \f | |
5843 | /* OpenFlow to datapath action translation. */ | |
5844 | ||
5845 | static bool may_receive(const struct ofport_dpif *, struct action_xlate_ctx *); | |
5846 | static void do_xlate_actions(const struct ofpact *, size_t ofpacts_len, | |
5847 | struct action_xlate_ctx *); | |
5848 | static void xlate_normal(struct action_xlate_ctx *); | |
5849 | ||
5850 | /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'. | |
5851 | * The action will state 'slow' as the reason that the action is in the slow | |
5852 | * path. (This is purely informational: it allows a human viewing "ovs-dpctl | |
5853 | * dump-flows" output to see why a flow is in the slow path.) | |
5854 | * | |
5855 | * The 'stub_size' bytes in 'stub' will be used to store the action. | |
5856 | * 'stub_size' must be large enough for the action. | |
5857 | * | |
5858 | * The action and its size will be stored in '*actionsp' and '*actions_lenp', | |
5859 | * respectively. */ | |
5860 | static void | |
5861 | compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow, | |
5862 | enum slow_path_reason slow, | |
5863 | uint64_t *stub, size_t stub_size, | |
5864 | const struct nlattr **actionsp, size_t *actions_lenp) | |
5865 | { | |
5866 | union user_action_cookie cookie; | |
5867 | struct ofpbuf buf; | |
5868 | ||
5869 | cookie.type = USER_ACTION_COOKIE_SLOW_PATH; | |
5870 | cookie.slow_path.unused = 0; | |
5871 | cookie.slow_path.reason = slow; | |
5872 | ||
5873 | ofpbuf_use_stack(&buf, stub, stub_size); | |
5874 | if (slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP | SLOW_STP)) { | |
5875 | uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif, UINT32_MAX); | |
5876 | odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf); | |
5877 | } else { | |
5878 | put_userspace_action(ofproto, &buf, flow, &cookie, | |
5879 | sizeof cookie.slow_path); | |
5880 | } | |
5881 | *actionsp = buf.data; | |
5882 | *actions_lenp = buf.size; | |
5883 | } | |
5884 | ||
5885 | static size_t | |
5886 | put_userspace_action(const struct ofproto_dpif *ofproto, | |
5887 | struct ofpbuf *odp_actions, | |
5888 | const struct flow *flow, | |
5889 | const union user_action_cookie *cookie, | |
5890 | const size_t cookie_size) | |
5891 | { | |
5892 | uint32_t pid; | |
5893 | ||
5894 | pid = dpif_port_get_pid(ofproto->backer->dpif, | |
5895 | ofp_port_to_odp_port(ofproto, flow->in_port)); | |
5896 | ||
5897 | return odp_put_userspace_action(pid, cookie, cookie_size, odp_actions); | |
5898 | } | |
5899 | ||
5900 | /* Compose SAMPLE action for sFlow or IPFIX. The given probability is | |
5901 | * the number of packets out of UINT32_MAX to sample. The given | |
5902 | * cookie is passed back in the callback for each sampled packet. | |
5903 | */ | |
5904 | static size_t | |
5905 | compose_sample_action(const struct ofproto_dpif *ofproto, | |
5906 | struct ofpbuf *odp_actions, | |
5907 | const struct flow *flow, | |
5908 | const uint32_t probability, | |
5909 | const union user_action_cookie *cookie, | |
5910 | const size_t cookie_size) | |
5911 | { | |
5912 | size_t sample_offset, actions_offset; | |
5913 | int cookie_offset; | |
5914 | ||
5915 | sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE); | |
5916 | ||
5917 | nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability); | |
5918 | ||
5919 | actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS); | |
5920 | cookie_offset = put_userspace_action(ofproto, odp_actions, flow, cookie, | |
5921 | cookie_size); | |
5922 | ||
5923 | nl_msg_end_nested(odp_actions, actions_offset); | |
5924 | nl_msg_end_nested(odp_actions, sample_offset); | |
5925 | return cookie_offset; | |
5926 | } | |
5927 | ||
5928 | static void | |
5929 | compose_sflow_cookie(const struct ofproto_dpif *ofproto, | |
5930 | ovs_be16 vlan_tci, uint32_t odp_port, | |
5931 | unsigned int n_outputs, union user_action_cookie *cookie) | |
5932 | { | |
5933 | int ifindex; | |
5934 | ||
5935 | cookie->type = USER_ACTION_COOKIE_SFLOW; | |
5936 | cookie->sflow.vlan_tci = vlan_tci; | |
5937 | ||
5938 | /* See http://www.sflow.org/sflow_version_5.txt (search for "Input/output | |
5939 | * port information") for the interpretation of cookie->output. */ | |
5940 | switch (n_outputs) { | |
5941 | case 0: | |
5942 | /* 0x40000000 | 256 means "packet dropped for unknown reason". */ | |
5943 | cookie->sflow.output = 0x40000000 | 256; | |
5944 | break; | |
5945 | ||
5946 | case 1: | |
5947 | ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port); | |
5948 | if (ifindex) { | |
5949 | cookie->sflow.output = ifindex; | |
5950 | break; | |
5951 | } | |
5952 | /* Fall through. */ | |
5953 | default: | |
5954 | /* 0x80000000 means "multiple output ports. */ | |
5955 | cookie->sflow.output = 0x80000000 | n_outputs; | |
5956 | break; | |
5957 | } | |
5958 | } | |
5959 | ||
5960 | /* Compose SAMPLE action for sFlow bridge sampling. */ | |
5961 | static size_t | |
5962 | compose_sflow_action(const struct ofproto_dpif *ofproto, | |
5963 | struct ofpbuf *odp_actions, | |
5964 | const struct flow *flow, | |
5965 | uint32_t odp_port) | |
5966 | { | |
5967 | uint32_t probability; | |
5968 | union user_action_cookie cookie; | |
5969 | ||
5970 | if (!ofproto->sflow || flow->in_port == OFPP_NONE) { | |
5971 | return 0; | |
5972 | } | |
5973 | ||
5974 | probability = dpif_sflow_get_probability(ofproto->sflow); | |
5975 | compose_sflow_cookie(ofproto, htons(0), odp_port, | |
5976 | odp_port == OVSP_NONE ? 0 : 1, &cookie); | |
5977 | ||
5978 | return compose_sample_action(ofproto, odp_actions, flow, probability, | |
5979 | &cookie, sizeof cookie.sflow); | |
5980 | } | |
5981 | ||
5982 | static void | |
5983 | compose_flow_sample_cookie(uint16_t probability, uint32_t collector_set_id, | |
5984 | uint32_t obs_domain_id, uint32_t obs_point_id, | |
5985 | union user_action_cookie *cookie) | |
5986 | { | |
5987 | cookie->type = USER_ACTION_COOKIE_FLOW_SAMPLE; | |
5988 | cookie->flow_sample.probability = probability; | |
5989 | cookie->flow_sample.collector_set_id = collector_set_id; | |
5990 | cookie->flow_sample.obs_domain_id = obs_domain_id; | |
5991 | cookie->flow_sample.obs_point_id = obs_point_id; | |
5992 | } | |
5993 | ||
5994 | static void | |
5995 | compose_ipfix_cookie(union user_action_cookie *cookie) | |
5996 | { | |
5997 | cookie->type = USER_ACTION_COOKIE_IPFIX; | |
5998 | } | |
5999 | ||
6000 | /* Compose SAMPLE action for IPFIX bridge sampling. */ | |
6001 | static void | |
6002 | compose_ipfix_action(const struct ofproto_dpif *ofproto, | |
6003 | struct ofpbuf *odp_actions, | |
6004 | const struct flow *flow) | |
6005 | { | |
6006 | uint32_t probability; | |
6007 | union user_action_cookie cookie; | |
6008 | ||
6009 | if (!ofproto->ipfix || flow->in_port == OFPP_NONE) { | |
6010 | return; | |
6011 | } | |
6012 | ||
6013 | probability = dpif_ipfix_get_bridge_exporter_probability(ofproto->ipfix); | |
6014 | compose_ipfix_cookie(&cookie); | |
6015 | ||
6016 | compose_sample_action(ofproto, odp_actions, flow, probability, | |
6017 | &cookie, sizeof cookie.ipfix); | |
6018 | } | |
6019 | ||
6020 | /* SAMPLE action for sFlow must be first action in any given list of | |
6021 | * actions. At this point we do not have all information required to | |
6022 | * build it. So try to build sample action as complete as possible. */ | |
6023 | static void | |
6024 | add_sflow_action(struct action_xlate_ctx *ctx) | |
6025 | { | |
6026 | ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto, | |
6027 | ctx->odp_actions, | |
6028 | &ctx->flow, OVSP_NONE); | |
6029 | ctx->sflow_odp_port = 0; | |
6030 | ctx->sflow_n_outputs = 0; | |
6031 | } | |
6032 | ||
6033 | /* SAMPLE action for IPFIX must be 1st or 2nd action in any given list | |
6034 | * of actions, eventually after the SAMPLE action for sFlow. */ | |
6035 | static void | |
6036 | add_ipfix_action(struct action_xlate_ctx *ctx) | |
6037 | { | |
6038 | compose_ipfix_action(ctx->ofproto, ctx->odp_actions, &ctx->flow); | |
6039 | } | |
6040 | ||
6041 | /* Fix SAMPLE action according to data collected while composing ODP actions. | |
6042 | * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested | |
6043 | * USERSPACE action's user-cookie which is required for sflow. */ | |
6044 | static void | |
6045 | fix_sflow_action(struct action_xlate_ctx *ctx) | |
6046 | { | |
6047 | const struct flow *base = &ctx->base_flow; | |
6048 | union user_action_cookie *cookie; | |
6049 | ||
6050 | if (!ctx->user_cookie_offset) { | |
6051 | return; | |
6052 | } | |
6053 | ||
6054 | cookie = ofpbuf_at(ctx->odp_actions, ctx->user_cookie_offset, | |
6055 | sizeof cookie->sflow); | |
6056 | ovs_assert(cookie->type == USER_ACTION_COOKIE_SFLOW); | |
6057 | ||
6058 | compose_sflow_cookie(ctx->ofproto, base->vlan_tci, | |
6059 | ctx->sflow_odp_port, ctx->sflow_n_outputs, cookie); | |
6060 | } | |
6061 | ||
6062 | static void | |
6063 | compose_output_action__(struct action_xlate_ctx *ctx, uint16_t ofp_port, | |
6064 | bool check_stp) | |
6065 | { | |
6066 | const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port); | |
6067 | ovs_be16 flow_vlan_tci; | |
6068 | uint32_t flow_skb_mark; | |
6069 | uint8_t flow_nw_tos; | |
6070 | struct priority_to_dscp *pdscp; | |
6071 | uint32_t out_port, odp_port; | |
6072 | ||
6073 | /* If 'struct flow' gets additional metadata, we'll need to zero it out | |
6074 | * before traversing a patch port. */ | |
6075 | BUILD_ASSERT_DECL(FLOW_WC_SEQ == 20); | |
6076 | ||
6077 | if (!ofport) { | |
6078 | xlate_report(ctx, "Nonexistent output port"); | |
6079 | return; | |
6080 | } else if (ofport->up.pp.config & OFPUTIL_PC_NO_FWD) { | |
6081 | xlate_report(ctx, "OFPPC_NO_FWD set, skipping output"); | |
6082 | return; | |
6083 | } else if (check_stp && !stp_forward_in_state(ofport->stp_state)) { | |
6084 | xlate_report(ctx, "STP not in forwarding state, skipping output"); | |
6085 | return; | |
6086 | } | |
6087 | ||
6088 | if (netdev_vport_is_patch(ofport->up.netdev)) { | |
6089 | struct ofport_dpif *peer = ofport_get_peer(ofport); | |
6090 | struct flow old_flow = ctx->flow; | |
6091 | const struct ofproto_dpif *peer_ofproto; | |
6092 | enum slow_path_reason special; | |
6093 | struct ofport_dpif *in_port; | |
6094 | ||
6095 | if (!peer) { | |
6096 | xlate_report(ctx, "Nonexistent patch port peer"); | |
6097 | return; | |
6098 | } | |
6099 | ||
6100 | peer_ofproto = ofproto_dpif_cast(peer->up.ofproto); | |
6101 | if (peer_ofproto->backer != ctx->ofproto->backer) { | |
6102 | xlate_report(ctx, "Patch port peer on a different datapath"); | |
6103 | return; | |
6104 | } | |
6105 | ||
6106 | ctx->ofproto = ofproto_dpif_cast(peer->up.ofproto); | |
6107 | ctx->flow.in_port = peer->up.ofp_port; | |
6108 | ctx->flow.metadata = htonll(0); | |
6109 | memset(&ctx->flow.tunnel, 0, sizeof ctx->flow.tunnel); | |
6110 | memset(ctx->flow.regs, 0, sizeof ctx->flow.regs); | |
6111 | ||
6112 | in_port = get_ofp_port(ctx->ofproto, ctx->flow.in_port); | |
6113 | special = process_special(ctx->ofproto, &ctx->flow, in_port, | |
6114 | ctx->packet); | |
6115 | if (special) { | |
6116 | ctx->slow |= special; | |
6117 | } else if (!in_port || may_receive(in_port, ctx)) { | |
6118 | if (!in_port || stp_forward_in_state(in_port->stp_state)) { | |
6119 | xlate_table_action(ctx, ctx->flow.in_port, 0, true); | |
6120 | } else { | |
6121 | /* Forwarding is disabled by STP. Let OFPP_NORMAL and the | |
6122 | * learning action look at the packet, then drop it. */ | |
6123 | struct flow old_base_flow = ctx->base_flow; | |
6124 | size_t old_size = ctx->odp_actions->size; | |
6125 | xlate_table_action(ctx, ctx->flow.in_port, 0, true); | |
6126 | ctx->base_flow = old_base_flow; | |
6127 | ctx->odp_actions->size = old_size; | |
6128 | } | |
6129 | } | |
6130 | ||
6131 | ctx->flow = old_flow; | |
6132 | ctx->ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
6133 | ||
6134 | if (ctx->resubmit_stats) { | |
6135 | netdev_vport_inc_tx(ofport->up.netdev, ctx->resubmit_stats); | |
6136 | netdev_vport_inc_rx(peer->up.netdev, ctx->resubmit_stats); | |
6137 | } | |
6138 | ||
6139 | return; | |
6140 | } | |
6141 | ||
6142 | flow_vlan_tci = ctx->flow.vlan_tci; | |
6143 | flow_skb_mark = ctx->flow.skb_mark; | |
6144 | flow_nw_tos = ctx->flow.nw_tos; | |
6145 | ||
6146 | pdscp = get_priority(ofport, ctx->flow.skb_priority); | |
6147 | if (pdscp) { | |
6148 | ctx->flow.nw_tos &= ~IP_DSCP_MASK; | |
6149 | ctx->flow.nw_tos |= pdscp->dscp; | |
6150 | } | |
6151 | ||
6152 | if (ofport->tnl_port) { | |
6153 | /* Save tunnel metadata so that changes made due to | |
6154 | * the Logical (tunnel) Port are not visible for any further | |
6155 | * matches, while explicit set actions on tunnel metadata are. | |
6156 | */ | |
6157 | struct flow_tnl flow_tnl = ctx->flow.tunnel; | |
6158 | odp_port = tnl_port_send(ofport->tnl_port, &ctx->flow); | |
6159 | if (odp_port == OVSP_NONE) { | |
6160 | xlate_report(ctx, "Tunneling decided against output"); | |
6161 | goto out; /* restore flow_nw_tos */ | |
6162 | } | |
6163 | ||
6164 | if (ctx->resubmit_stats) { | |
6165 | netdev_vport_inc_tx(ofport->up.netdev, ctx->resubmit_stats); | |
6166 | } | |
6167 | out_port = odp_port; | |
6168 | commit_odp_tunnel_action(&ctx->flow, &ctx->base_flow, | |
6169 | ctx->odp_actions); | |
6170 | ctx->flow.tunnel = flow_tnl; /* Restore tunnel metadata */ | |
6171 | } else { | |
6172 | odp_port = ofport->odp_port; | |
6173 | out_port = vsp_realdev_to_vlandev(ctx->ofproto, odp_port, | |
6174 | ctx->flow.vlan_tci); | |
6175 | if (out_port != odp_port) { | |
6176 | ctx->flow.vlan_tci = htons(0); | |
6177 | } | |
6178 | ctx->flow.skb_mark &= ~IPSEC_MARK; | |
6179 | } | |
6180 | commit_odp_actions(&ctx->flow, &ctx->base_flow, ctx->odp_actions); | |
6181 | nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_OUTPUT, out_port); | |
6182 | ||
6183 | ctx->sflow_odp_port = odp_port; | |
6184 | ctx->sflow_n_outputs++; | |
6185 | ctx->nf_output_iface = ofp_port; | |
6186 | ||
6187 | /* Restore flow */ | |
6188 | ctx->flow.vlan_tci = flow_vlan_tci; | |
6189 | ctx->flow.skb_mark = flow_skb_mark; | |
6190 | out: | |
6191 | ctx->flow.nw_tos = flow_nw_tos; | |
6192 | } | |
6193 | ||
6194 | static void | |
6195 | compose_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port) | |
6196 | { | |
6197 | compose_output_action__(ctx, ofp_port, true); | |
6198 | } | |
6199 | ||
6200 | static void | |
6201 | tag_the_flow(struct action_xlate_ctx *ctx, struct rule_dpif *rule) | |
6202 | { | |
6203 | struct ofproto_dpif *ofproto = ctx->ofproto; | |
6204 | uint8_t table_id = ctx->table_id; | |
6205 | ||
6206 | if (table_id > 0 && table_id < N_TABLES) { | |
6207 | struct table_dpif *table = &ofproto->tables[table_id]; | |
6208 | if (table->other_table) { | |
6209 | ctx->tags |= (rule && rule->tag | |
6210 | ? rule->tag | |
6211 | : rule_calculate_tag(&ctx->flow, | |
6212 | &table->other_table->mask, | |
6213 | table->basis)); | |
6214 | } | |
6215 | } | |
6216 | } | |
6217 | ||
6218 | /* Common rule processing in one place to avoid duplicating code. */ | |
6219 | static struct rule_dpif * | |
6220 | ctx_rule_hooks(struct action_xlate_ctx *ctx, struct rule_dpif *rule, | |
6221 | bool may_packet_in) | |
6222 | { | |
6223 | if (ctx->resubmit_hook) { | |
6224 | ctx->resubmit_hook(ctx, rule); | |
6225 | } | |
6226 | if (rule == NULL && may_packet_in) { | |
6227 | /* XXX | |
6228 | * check if table configuration flags | |
6229 | * OFPTC_TABLE_MISS_CONTROLLER, default. | |
6230 | * OFPTC_TABLE_MISS_CONTINUE, | |
6231 | * OFPTC_TABLE_MISS_DROP | |
6232 | * When OF1.0, OFPTC_TABLE_MISS_CONTINUE is used. What to do? | |
6233 | */ | |
6234 | rule = rule_dpif_miss_rule(ctx->ofproto, &ctx->flow); | |
6235 | } | |
6236 | if (rule && ctx->resubmit_stats) { | |
6237 | rule_credit_stats(rule, ctx->resubmit_stats); | |
6238 | } | |
6239 | return rule; | |
6240 | } | |
6241 | ||
6242 | static void | |
6243 | xlate_table_action(struct action_xlate_ctx *ctx, | |
6244 | uint16_t in_port, uint8_t table_id, bool may_packet_in) | |
6245 | { | |
6246 | if (ctx->recurse < MAX_RESUBMIT_RECURSION) { | |
6247 | struct rule_dpif *rule; | |
6248 | uint16_t old_in_port = ctx->flow.in_port; | |
6249 | uint8_t old_table_id = ctx->table_id; | |
6250 | ||
6251 | ctx->table_id = table_id; | |
6252 | ||
6253 | /* Look up a flow with 'in_port' as the input port. */ | |
6254 | ctx->flow.in_port = in_port; | |
6255 | rule = rule_dpif_lookup__(ctx->ofproto, &ctx->flow, table_id); | |
6256 | ||
6257 | tag_the_flow(ctx, rule); | |
6258 | ||
6259 | /* Restore the original input port. Otherwise OFPP_NORMAL and | |
6260 | * OFPP_IN_PORT will have surprising behavior. */ | |
6261 | ctx->flow.in_port = old_in_port; | |
6262 | ||
6263 | rule = ctx_rule_hooks(ctx, rule, may_packet_in); | |
6264 | ||
6265 | if (rule) { | |
6266 | struct rule_dpif *old_rule = ctx->rule; | |
6267 | ||
6268 | ctx->recurse++; | |
6269 | ctx->rule = rule; | |
6270 | do_xlate_actions(rule->up.ofpacts, rule->up.ofpacts_len, ctx); | |
6271 | ctx->rule = old_rule; | |
6272 | ctx->recurse--; | |
6273 | } | |
6274 | ||
6275 | ctx->table_id = old_table_id; | |
6276 | } else { | |
6277 | static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1); | |
6278 | ||
6279 | VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times", | |
6280 | MAX_RESUBMIT_RECURSION); | |
6281 | ctx->max_resubmit_trigger = true; | |
6282 | } | |
6283 | } | |
6284 | ||
6285 | static void | |
6286 | xlate_ofpact_resubmit(struct action_xlate_ctx *ctx, | |
6287 | const struct ofpact_resubmit *resubmit) | |
6288 | { | |
6289 | uint16_t in_port; | |
6290 | uint8_t table_id; | |
6291 | ||
6292 | in_port = resubmit->in_port; | |
6293 | if (in_port == OFPP_IN_PORT) { | |
6294 | in_port = ctx->flow.in_port; | |
6295 | } | |
6296 | ||
6297 | table_id = resubmit->table_id; | |
6298 | if (table_id == 255) { | |
6299 | table_id = ctx->table_id; | |
6300 | } | |
6301 | ||
6302 | xlate_table_action(ctx, in_port, table_id, false); | |
6303 | } | |
6304 | ||
6305 | static void | |
6306 | flood_packets(struct action_xlate_ctx *ctx, bool all) | |
6307 | { | |
6308 | struct ofport_dpif *ofport; | |
6309 | ||
6310 | HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) { | |
6311 | uint16_t ofp_port = ofport->up.ofp_port; | |
6312 | ||
6313 | if (ofp_port == ctx->flow.in_port) { | |
6314 | continue; | |
6315 | } | |
6316 | ||
6317 | if (all) { | |
6318 | compose_output_action__(ctx, ofp_port, false); | |
6319 | } else if (!(ofport->up.pp.config & OFPUTIL_PC_NO_FLOOD)) { | |
6320 | compose_output_action(ctx, ofp_port); | |
6321 | } | |
6322 | } | |
6323 | ||
6324 | ctx->nf_output_iface = NF_OUT_FLOOD; | |
6325 | } | |
6326 | ||
6327 | static void | |
6328 | execute_controller_action(struct action_xlate_ctx *ctx, int len, | |
6329 | enum ofp_packet_in_reason reason, | |
6330 | uint16_t controller_id) | |
6331 | { | |
6332 | struct ofputil_packet_in pin; | |
6333 | struct ofpbuf *packet; | |
6334 | ||
6335 | ctx->slow |= SLOW_CONTROLLER; | |
6336 | if (!ctx->packet) { | |
6337 | return; | |
6338 | } | |
6339 | ||
6340 | packet = ofpbuf_clone(ctx->packet); | |
6341 | ||
6342 | if (packet->l2 && packet->l3) { | |
6343 | struct eth_header *eh; | |
6344 | uint16_t mpls_depth; | |
6345 | ||
6346 | eth_pop_vlan(packet); | |
6347 | eh = packet->l2; | |
6348 | ||
6349 | memcpy(eh->eth_src, ctx->flow.dl_src, sizeof eh->eth_src); | |
6350 | memcpy(eh->eth_dst, ctx->flow.dl_dst, sizeof eh->eth_dst); | |
6351 | ||
6352 | if (ctx->flow.vlan_tci & htons(VLAN_CFI)) { | |
6353 | eth_push_vlan(packet, ctx->flow.vlan_tci); | |
6354 | } | |
6355 | ||
6356 | mpls_depth = eth_mpls_depth(packet); | |
6357 | ||
6358 | if (mpls_depth < ctx->flow.mpls_depth) { | |
6359 | push_mpls(packet, ctx->flow.dl_type, ctx->flow.mpls_lse); | |
6360 | } else if (mpls_depth > ctx->flow.mpls_depth) { | |
6361 | pop_mpls(packet, ctx->flow.dl_type); | |
6362 | } else if (mpls_depth) { | |
6363 | set_mpls_lse(packet, ctx->flow.mpls_lse); | |
6364 | } | |
6365 | ||
6366 | if (packet->l4) { | |
6367 | if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) { | |
6368 | packet_set_ipv4(packet, ctx->flow.nw_src, ctx->flow.nw_dst, | |
6369 | ctx->flow.nw_tos, ctx->flow.nw_ttl); | |
6370 | } | |
6371 | ||
6372 | if (packet->l7) { | |
6373 | if (ctx->flow.nw_proto == IPPROTO_TCP) { | |
6374 | packet_set_tcp_port(packet, ctx->flow.tp_src, | |
6375 | ctx->flow.tp_dst); | |
6376 | } else if (ctx->flow.nw_proto == IPPROTO_UDP) { | |
6377 | packet_set_udp_port(packet, ctx->flow.tp_src, | |
6378 | ctx->flow.tp_dst); | |
6379 | } | |
6380 | } | |
6381 | } | |
6382 | } | |
6383 | ||
6384 | pin.packet = packet->data; | |
6385 | pin.packet_len = packet->size; | |
6386 | pin.reason = reason; | |
6387 | pin.controller_id = controller_id; | |
6388 | pin.table_id = ctx->table_id; | |
6389 | pin.cookie = ctx->rule ? ctx->rule->up.flow_cookie : 0; | |
6390 | ||
6391 | pin.send_len = len; | |
6392 | flow_get_metadata(&ctx->flow, &pin.fmd); | |
6393 | ||
6394 | connmgr_send_packet_in(ctx->ofproto->up.connmgr, &pin); | |
6395 | ofpbuf_delete(packet); | |
6396 | } | |
6397 | ||
6398 | static void | |
6399 | execute_mpls_push_action(struct action_xlate_ctx *ctx, ovs_be16 eth_type) | |
6400 | { | |
6401 | ovs_assert(eth_type_mpls(eth_type)); | |
6402 | ||
6403 | if (ctx->base_flow.mpls_depth) { | |
6404 | ctx->flow.mpls_lse &= ~htonl(MPLS_BOS_MASK); | |
6405 | ctx->flow.mpls_depth++; | |
6406 | } else { | |
6407 | ovs_be32 label; | |
6408 | uint8_t tc, ttl; | |
6409 | ||
6410 | if (ctx->flow.dl_type == htons(ETH_TYPE_IPV6)) { | |
6411 | label = htonl(0x2); /* IPV6 Explicit Null. */ | |
6412 | } else { | |
6413 | label = htonl(0x0); /* IPV4 Explicit Null. */ | |
6414 | } | |
6415 | tc = (ctx->flow.nw_tos & IP_DSCP_MASK) >> 2; | |
6416 | ttl = ctx->flow.nw_ttl ? ctx->flow.nw_ttl : 0x40; | |
6417 | ctx->flow.mpls_lse = set_mpls_lse_values(ttl, tc, 1, label); | |
6418 | ctx->flow.mpls_depth = 1; | |
6419 | } | |
6420 | ctx->flow.dl_type = eth_type; | |
6421 | } | |
6422 | ||
6423 | static void | |
6424 | execute_mpls_pop_action(struct action_xlate_ctx *ctx, ovs_be16 eth_type) | |
6425 | { | |
6426 | ovs_assert(eth_type_mpls(ctx->flow.dl_type)); | |
6427 | ovs_assert(!eth_type_mpls(eth_type)); | |
6428 | ||
6429 | if (ctx->flow.mpls_depth) { | |
6430 | ctx->flow.mpls_depth--; | |
6431 | ctx->flow.mpls_lse = htonl(0); | |
6432 | if (!ctx->flow.mpls_depth) { | |
6433 | ctx->flow.dl_type = eth_type; | |
6434 | } | |
6435 | } | |
6436 | } | |
6437 | ||
6438 | static bool | |
6439 | compose_dec_ttl(struct action_xlate_ctx *ctx, struct ofpact_cnt_ids *ids) | |
6440 | { | |
6441 | if (ctx->flow.dl_type != htons(ETH_TYPE_IP) && | |
6442 | ctx->flow.dl_type != htons(ETH_TYPE_IPV6)) { | |
6443 | return false; | |
6444 | } | |
6445 | ||
6446 | if (ctx->flow.nw_ttl > 1) { | |
6447 | ctx->flow.nw_ttl--; | |
6448 | return false; | |
6449 | } else { | |
6450 | size_t i; | |
6451 | ||
6452 | for (i = 0; i < ids->n_controllers; i++) { | |
6453 | execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL, | |
6454 | ids->cnt_ids[i]); | |
6455 | } | |
6456 | ||
6457 | /* Stop processing for current table. */ | |
6458 | return true; | |
6459 | } | |
6460 | } | |
6461 | ||
6462 | static bool | |
6463 | execute_set_mpls_ttl_action(struct action_xlate_ctx *ctx, uint8_t ttl) | |
6464 | { | |
6465 | if (!eth_type_mpls(ctx->flow.dl_type)) { | |
6466 | return true; | |
6467 | } | |
6468 | ||
6469 | set_mpls_lse_ttl(&ctx->flow.mpls_lse, ttl); | |
6470 | return false; | |
6471 | } | |
6472 | ||
6473 | static bool | |
6474 | execute_dec_mpls_ttl_action(struct action_xlate_ctx *ctx) | |
6475 | { | |
6476 | uint8_t ttl = mpls_lse_to_ttl(ctx->flow.mpls_lse); | |
6477 | ||
6478 | if (!eth_type_mpls(ctx->flow.dl_type)) { | |
6479 | return false; | |
6480 | } | |
6481 | ||
6482 | if (ttl > 1) { | |
6483 | ttl--; | |
6484 | set_mpls_lse_ttl(&ctx->flow.mpls_lse, ttl); | |
6485 | return false; | |
6486 | } else { | |
6487 | execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL, 0); | |
6488 | ||
6489 | /* Stop processing for current table. */ | |
6490 | return true; | |
6491 | } | |
6492 | } | |
6493 | ||
6494 | static void | |
6495 | xlate_output_action(struct action_xlate_ctx *ctx, | |
6496 | uint16_t port, uint16_t max_len, bool may_packet_in) | |
6497 | { | |
6498 | uint16_t prev_nf_output_iface = ctx->nf_output_iface; | |
6499 | ||
6500 | ctx->nf_output_iface = NF_OUT_DROP; | |
6501 | ||
6502 | switch (port) { | |
6503 | case OFPP_IN_PORT: | |
6504 | compose_output_action(ctx, ctx->flow.in_port); | |
6505 | break; | |
6506 | case OFPP_TABLE: | |
6507 | xlate_table_action(ctx, ctx->flow.in_port, 0, may_packet_in); | |
6508 | break; | |
6509 | case OFPP_NORMAL: | |
6510 | xlate_normal(ctx); | |
6511 | break; | |
6512 | case OFPP_FLOOD: | |
6513 | flood_packets(ctx, false); | |
6514 | break; | |
6515 | case OFPP_ALL: | |
6516 | flood_packets(ctx, true); | |
6517 | break; | |
6518 | case OFPP_CONTROLLER: | |
6519 | execute_controller_action(ctx, max_len, OFPR_ACTION, 0); | |
6520 | break; | |
6521 | case OFPP_NONE: | |
6522 | break; | |
6523 | case OFPP_LOCAL: | |
6524 | default: | |
6525 | if (port != ctx->flow.in_port) { | |
6526 | compose_output_action(ctx, port); | |
6527 | } else { | |
6528 | xlate_report(ctx, "skipping output to input port"); | |
6529 | } | |
6530 | break; | |
6531 | } | |
6532 | ||
6533 | if (prev_nf_output_iface == NF_OUT_FLOOD) { | |
6534 | ctx->nf_output_iface = NF_OUT_FLOOD; | |
6535 | } else if (ctx->nf_output_iface == NF_OUT_DROP) { | |
6536 | ctx->nf_output_iface = prev_nf_output_iface; | |
6537 | } else if (prev_nf_output_iface != NF_OUT_DROP && | |
6538 | ctx->nf_output_iface != NF_OUT_FLOOD) { | |
6539 | ctx->nf_output_iface = NF_OUT_MULTI; | |
6540 | } | |
6541 | } | |
6542 | ||
6543 | static void | |
6544 | xlate_output_reg_action(struct action_xlate_ctx *ctx, | |
6545 | const struct ofpact_output_reg *or) | |
6546 | { | |
6547 | uint64_t port = mf_get_subfield(&or->src, &ctx->flow); | |
6548 | if (port <= UINT16_MAX) { | |
6549 | xlate_output_action(ctx, port, or->max_len, false); | |
6550 | } | |
6551 | } | |
6552 | ||
6553 | static void | |
6554 | xlate_enqueue_action(struct action_xlate_ctx *ctx, | |
6555 | const struct ofpact_enqueue *enqueue) | |
6556 | { | |
6557 | uint16_t ofp_port = enqueue->port; | |
6558 | uint32_t queue_id = enqueue->queue; | |
6559 | uint32_t flow_priority, priority; | |
6560 | int error; | |
6561 | ||
6562 | /* Translate queue to priority. */ | |
6563 | error = dpif_queue_to_priority(ctx->ofproto->backer->dpif, | |
6564 | queue_id, &priority); | |
6565 | if (error) { | |
6566 | /* Fall back to ordinary output action. */ | |
6567 | xlate_output_action(ctx, enqueue->port, 0, false); | |
6568 | return; | |
6569 | } | |
6570 | ||
6571 | /* Check output port. */ | |
6572 | if (ofp_port == OFPP_IN_PORT) { | |
6573 | ofp_port = ctx->flow.in_port; | |
6574 | } else if (ofp_port == ctx->flow.in_port) { | |
6575 | return; | |
6576 | } | |
6577 | ||
6578 | /* Add datapath actions. */ | |
6579 | flow_priority = ctx->flow.skb_priority; | |
6580 | ctx->flow.skb_priority = priority; | |
6581 | compose_output_action(ctx, ofp_port); | |
6582 | ctx->flow.skb_priority = flow_priority; | |
6583 | ||
6584 | /* Update NetFlow output port. */ | |
6585 | if (ctx->nf_output_iface == NF_OUT_DROP) { | |
6586 | ctx->nf_output_iface = ofp_port; | |
6587 | } else if (ctx->nf_output_iface != NF_OUT_FLOOD) { | |
6588 | ctx->nf_output_iface = NF_OUT_MULTI; | |
6589 | } | |
6590 | } | |
6591 | ||
6592 | static void | |
6593 | xlate_set_queue_action(struct action_xlate_ctx *ctx, uint32_t queue_id) | |
6594 | { | |
6595 | uint32_t skb_priority; | |
6596 | ||
6597 | if (!dpif_queue_to_priority(ctx->ofproto->backer->dpif, | |
6598 | queue_id, &skb_priority)) { | |
6599 | ctx->flow.skb_priority = skb_priority; | |
6600 | } else { | |
6601 | /* Couldn't translate queue to a priority. Nothing to do. A warning | |
6602 | * has already been logged. */ | |
6603 | } | |
6604 | } | |
6605 | ||
6606 | static bool | |
6607 | slave_enabled_cb(uint16_t ofp_port, void *ofproto_) | |
6608 | { | |
6609 | struct ofproto_dpif *ofproto = ofproto_; | |
6610 | struct ofport_dpif *port; | |
6611 | ||
6612 | switch (ofp_port) { | |
6613 | case OFPP_IN_PORT: | |
6614 | case OFPP_TABLE: | |
6615 | case OFPP_NORMAL: | |
6616 | case OFPP_FLOOD: | |
6617 | case OFPP_ALL: | |
6618 | case OFPP_NONE: | |
6619 | return true; | |
6620 | case OFPP_CONTROLLER: /* Not supported by the bundle action. */ | |
6621 | return false; | |
6622 | default: | |
6623 | port = get_ofp_port(ofproto, ofp_port); | |
6624 | return port ? port->may_enable : false; | |
6625 | } | |
6626 | } | |
6627 | ||
6628 | static void | |
6629 | xlate_bundle_action(struct action_xlate_ctx *ctx, | |
6630 | const struct ofpact_bundle *bundle) | |
6631 | { | |
6632 | uint16_t port; | |
6633 | ||
6634 | port = bundle_execute(bundle, &ctx->flow, slave_enabled_cb, ctx->ofproto); | |
6635 | if (bundle->dst.field) { | |
6636 | nxm_reg_load(&bundle->dst, port, &ctx->flow); | |
6637 | } else { | |
6638 | xlate_output_action(ctx, port, 0, false); | |
6639 | } | |
6640 | } | |
6641 | ||
6642 | static void | |
6643 | xlate_learn_action(struct action_xlate_ctx *ctx, | |
6644 | const struct ofpact_learn *learn) | |
6645 | { | |
6646 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1); | |
6647 | struct ofputil_flow_mod fm; | |
6648 | uint64_t ofpacts_stub[1024 / 8]; | |
6649 | struct ofpbuf ofpacts; | |
6650 | int error; | |
6651 | ||
6652 | ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub); | |
6653 | learn_execute(learn, &ctx->flow, &fm, &ofpacts); | |
6654 | ||
6655 | error = ofproto_flow_mod(&ctx->ofproto->up, &fm); | |
6656 | if (error && !VLOG_DROP_WARN(&rl)) { | |
6657 | VLOG_WARN("learning action failed to modify flow table (%s)", | |
6658 | ofperr_get_name(error)); | |
6659 | } | |
6660 | ||
6661 | ofpbuf_uninit(&ofpacts); | |
6662 | } | |
6663 | ||
6664 | /* Reduces '*timeout' to no more than 'max'. A value of zero in either case | |
6665 | * means "infinite". */ | |
6666 | static void | |
6667 | reduce_timeout(uint16_t max, uint16_t *timeout) | |
6668 | { | |
6669 | if (max && (!*timeout || *timeout > max)) { | |
6670 | *timeout = max; | |
6671 | } | |
6672 | } | |
6673 | ||
6674 | static void | |
6675 | xlate_fin_timeout(struct action_xlate_ctx *ctx, | |
6676 | const struct ofpact_fin_timeout *oft) | |
6677 | { | |
6678 | if (ctx->tcp_flags & (TCP_FIN | TCP_RST) && ctx->rule) { | |
6679 | struct rule_dpif *rule = ctx->rule; | |
6680 | ||
6681 | reduce_timeout(oft->fin_idle_timeout, &rule->up.idle_timeout); | |
6682 | reduce_timeout(oft->fin_hard_timeout, &rule->up.hard_timeout); | |
6683 | } | |
6684 | } | |
6685 | ||
6686 | static void | |
6687 | xlate_sample_action(struct action_xlate_ctx *ctx, | |
6688 | const struct ofpact_sample *os) | |
6689 | { | |
6690 | union user_action_cookie cookie; | |
6691 | /* Scale the probability from 16-bit to 32-bit while representing | |
6692 | * the same percentage. */ | |
6693 | uint32_t probability = (os->probability << 16) | os->probability; | |
6694 | ||
6695 | commit_odp_actions(&ctx->flow, &ctx->base_flow, ctx->odp_actions); | |
6696 | ||
6697 | compose_flow_sample_cookie(os->probability, os->collector_set_id, | |
6698 | os->obs_domain_id, os->obs_point_id, &cookie); | |
6699 | compose_sample_action(ctx->ofproto, ctx->odp_actions, &ctx->flow, | |
6700 | probability, &cookie, sizeof cookie.flow_sample); | |
6701 | } | |
6702 | ||
6703 | static bool | |
6704 | may_receive(const struct ofport_dpif *port, struct action_xlate_ctx *ctx) | |
6705 | { | |
6706 | if (port->up.pp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp) | |
6707 | ? OFPUTIL_PC_NO_RECV_STP | |
6708 | : OFPUTIL_PC_NO_RECV)) { | |
6709 | return false; | |
6710 | } | |
6711 | ||
6712 | /* Only drop packets here if both forwarding and learning are | |
6713 | * disabled. If just learning is enabled, we need to have | |
6714 | * OFPP_NORMAL and the learning action have a look at the packet | |
6715 | * before we can drop it. */ | |
6716 | if (!stp_forward_in_state(port->stp_state) | |
6717 | && !stp_learn_in_state(port->stp_state)) { | |
6718 | return false; | |
6719 | } | |
6720 | ||
6721 | return true; | |
6722 | } | |
6723 | ||
6724 | static bool | |
6725 | tunnel_ecn_ok(struct action_xlate_ctx *ctx) | |
6726 | { | |
6727 | if (is_ip_any(&ctx->base_flow) | |
6728 | && (ctx->flow.tunnel.ip_tos & IP_ECN_MASK) == IP_ECN_CE) { | |
6729 | if ((ctx->base_flow.nw_tos & IP_ECN_MASK) == IP_ECN_NOT_ECT) { | |
6730 | VLOG_WARN_RL(&rl, "dropping tunnel packet marked ECN CE" | |
6731 | " but is not ECN capable"); | |
6732 | return false; | |
6733 | } else { | |
6734 | /* Set the ECN CE value in the tunneled packet. */ | |
6735 | ctx->flow.nw_tos |= IP_ECN_CE; | |
6736 | } | |
6737 | } | |
6738 | ||
6739 | return true; | |
6740 | } | |
6741 | ||
6742 | static void | |
6743 | do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len, | |
6744 | struct action_xlate_ctx *ctx) | |
6745 | { | |
6746 | bool was_evictable = true; | |
6747 | const struct ofpact *a; | |
6748 | ||
6749 | if (ctx->rule) { | |
6750 | /* Don't let the rule we're working on get evicted underneath us. */ | |
6751 | was_evictable = ctx->rule->up.evictable; | |
6752 | ctx->rule->up.evictable = false; | |
6753 | } | |
6754 | ||
6755 | do_xlate_actions_again: | |
6756 | OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) { | |
6757 | struct ofpact_controller *controller; | |
6758 | const struct ofpact_metadata *metadata; | |
6759 | ||
6760 | if (ctx->exit) { | |
6761 | break; | |
6762 | } | |
6763 | ||
6764 | switch (a->type) { | |
6765 | case OFPACT_OUTPUT: | |
6766 | xlate_output_action(ctx, ofpact_get_OUTPUT(a)->port, | |
6767 | ofpact_get_OUTPUT(a)->max_len, true); | |
6768 | break; | |
6769 | ||
6770 | case OFPACT_CONTROLLER: | |
6771 | controller = ofpact_get_CONTROLLER(a); | |
6772 | execute_controller_action(ctx, controller->max_len, | |
6773 | controller->reason, | |
6774 | controller->controller_id); | |
6775 | break; | |
6776 | ||
6777 | case OFPACT_ENQUEUE: | |
6778 | xlate_enqueue_action(ctx, ofpact_get_ENQUEUE(a)); | |
6779 | break; | |
6780 | ||
6781 | case OFPACT_SET_VLAN_VID: | |
6782 | ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK); | |
6783 | ctx->flow.vlan_tci |= (htons(ofpact_get_SET_VLAN_VID(a)->vlan_vid) | |
6784 | | htons(VLAN_CFI)); | |
6785 | break; | |
6786 | ||
6787 | case OFPACT_SET_VLAN_PCP: | |
6788 | ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK); | |
6789 | ctx->flow.vlan_tci |= htons((ofpact_get_SET_VLAN_PCP(a)->vlan_pcp | |
6790 | << VLAN_PCP_SHIFT) | |
6791 | | VLAN_CFI); | |
6792 | break; | |
6793 | ||
6794 | case OFPACT_STRIP_VLAN: | |
6795 | ctx->flow.vlan_tci = htons(0); | |
6796 | break; | |
6797 | ||
6798 | case OFPACT_PUSH_VLAN: | |
6799 | /* XXX 802.1AD(QinQ) */ | |
6800 | ctx->flow.vlan_tci = htons(VLAN_CFI); | |
6801 | break; | |
6802 | ||
6803 | case OFPACT_SET_ETH_SRC: | |
6804 | memcpy(ctx->flow.dl_src, ofpact_get_SET_ETH_SRC(a)->mac, | |
6805 | ETH_ADDR_LEN); | |
6806 | break; | |
6807 | ||
6808 | case OFPACT_SET_ETH_DST: | |
6809 | memcpy(ctx->flow.dl_dst, ofpact_get_SET_ETH_DST(a)->mac, | |
6810 | ETH_ADDR_LEN); | |
6811 | break; | |
6812 | ||
6813 | case OFPACT_SET_IPV4_SRC: | |
6814 | if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) { | |
6815 | ctx->flow.nw_src = ofpact_get_SET_IPV4_SRC(a)->ipv4; | |
6816 | } | |
6817 | break; | |
6818 | ||
6819 | case OFPACT_SET_IPV4_DST: | |
6820 | if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) { | |
6821 | ctx->flow.nw_dst = ofpact_get_SET_IPV4_DST(a)->ipv4; | |
6822 | } | |
6823 | break; | |
6824 | ||
6825 | case OFPACT_SET_IPV4_DSCP: | |
6826 | /* OpenFlow 1.0 only supports IPv4. */ | |
6827 | if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) { | |
6828 | ctx->flow.nw_tos &= ~IP_DSCP_MASK; | |
6829 | ctx->flow.nw_tos |= ofpact_get_SET_IPV4_DSCP(a)->dscp; | |
6830 | } | |
6831 | break; | |
6832 | ||
6833 | case OFPACT_SET_L4_SRC_PORT: | |
6834 | if (is_ip_any(&ctx->flow)) { | |
6835 | ctx->flow.tp_src = htons(ofpact_get_SET_L4_SRC_PORT(a)->port); | |
6836 | } | |
6837 | break; | |
6838 | ||
6839 | case OFPACT_SET_L4_DST_PORT: | |
6840 | if (is_ip_any(&ctx->flow)) { | |
6841 | ctx->flow.tp_dst = htons(ofpact_get_SET_L4_DST_PORT(a)->port); | |
6842 | } | |
6843 | break; | |
6844 | ||
6845 | case OFPACT_RESUBMIT: | |
6846 | xlate_ofpact_resubmit(ctx, ofpact_get_RESUBMIT(a)); | |
6847 | break; | |
6848 | ||
6849 | case OFPACT_SET_TUNNEL: | |
6850 | ctx->flow.tunnel.tun_id = htonll(ofpact_get_SET_TUNNEL(a)->tun_id); | |
6851 | break; | |
6852 | ||
6853 | case OFPACT_SET_QUEUE: | |
6854 | xlate_set_queue_action(ctx, ofpact_get_SET_QUEUE(a)->queue_id); | |
6855 | break; | |
6856 | ||
6857 | case OFPACT_POP_QUEUE: | |
6858 | ctx->flow.skb_priority = ctx->orig_skb_priority; | |
6859 | break; | |
6860 | ||
6861 | case OFPACT_REG_MOVE: | |
6862 | nxm_execute_reg_move(ofpact_get_REG_MOVE(a), &ctx->flow); | |
6863 | break; | |
6864 | ||
6865 | case OFPACT_REG_LOAD: | |
6866 | nxm_execute_reg_load(ofpact_get_REG_LOAD(a), &ctx->flow); | |
6867 | break; | |
6868 | ||
6869 | case OFPACT_STACK_PUSH: | |
6870 | nxm_execute_stack_push(ofpact_get_STACK_PUSH(a), &ctx->flow, | |
6871 | &ctx->stack); | |
6872 | break; | |
6873 | ||
6874 | case OFPACT_STACK_POP: | |
6875 | nxm_execute_stack_pop(ofpact_get_STACK_POP(a), &ctx->flow, | |
6876 | &ctx->stack); | |
6877 | break; | |
6878 | ||
6879 | case OFPACT_PUSH_MPLS: | |
6880 | execute_mpls_push_action(ctx, ofpact_get_PUSH_MPLS(a)->ethertype); | |
6881 | break; | |
6882 | ||
6883 | case OFPACT_POP_MPLS: | |
6884 | execute_mpls_pop_action(ctx, ofpact_get_POP_MPLS(a)->ethertype); | |
6885 | break; | |
6886 | ||
6887 | case OFPACT_SET_MPLS_TTL: | |
6888 | if (execute_set_mpls_ttl_action(ctx, ofpact_get_SET_MPLS_TTL(a)->ttl)) { | |
6889 | goto out; | |
6890 | } | |
6891 | break; | |
6892 | ||
6893 | case OFPACT_DEC_MPLS_TTL: | |
6894 | if (execute_dec_mpls_ttl_action(ctx)) { | |
6895 | goto out; | |
6896 | } | |
6897 | break; | |
6898 | ||
6899 | case OFPACT_DEC_TTL: | |
6900 | if (compose_dec_ttl(ctx, ofpact_get_DEC_TTL(a))) { | |
6901 | goto out; | |
6902 | } | |
6903 | break; | |
6904 | ||
6905 | case OFPACT_NOTE: | |
6906 | /* Nothing to do. */ | |
6907 | break; | |
6908 | ||
6909 | case OFPACT_MULTIPATH: | |
6910 | multipath_execute(ofpact_get_MULTIPATH(a), &ctx->flow); | |
6911 | break; | |
6912 | ||
6913 | case OFPACT_BUNDLE: | |
6914 | ctx->ofproto->has_bundle_action = true; | |
6915 | xlate_bundle_action(ctx, ofpact_get_BUNDLE(a)); | |
6916 | break; | |
6917 | ||
6918 | case OFPACT_OUTPUT_REG: | |
6919 | xlate_output_reg_action(ctx, ofpact_get_OUTPUT_REG(a)); | |
6920 | break; | |
6921 | ||
6922 | case OFPACT_LEARN: | |
6923 | ctx->has_learn = true; | |
6924 | if (ctx->may_learn) { | |
6925 | xlate_learn_action(ctx, ofpact_get_LEARN(a)); | |
6926 | } | |
6927 | break; | |
6928 | ||
6929 | case OFPACT_EXIT: | |
6930 | ctx->exit = true; | |
6931 | break; | |
6932 | ||
6933 | case OFPACT_FIN_TIMEOUT: | |
6934 | ctx->has_fin_timeout = true; | |
6935 | xlate_fin_timeout(ctx, ofpact_get_FIN_TIMEOUT(a)); | |
6936 | break; | |
6937 | ||
6938 | case OFPACT_CLEAR_ACTIONS: | |
6939 | /* XXX | |
6940 | * Nothing to do because writa-actions is not supported for now. | |
6941 | * When writa-actions is supported, clear-actions also must | |
6942 | * be supported at the same time. | |
6943 | */ | |
6944 | break; | |
6945 | ||
6946 | case OFPACT_WRITE_METADATA: | |
6947 | metadata = ofpact_get_WRITE_METADATA(a); | |
6948 | ctx->flow.metadata &= ~metadata->mask; | |
6949 | ctx->flow.metadata |= metadata->metadata & metadata->mask; | |
6950 | break; | |
6951 | ||
6952 | case OFPACT_GOTO_TABLE: { | |
6953 | /* It is assumed that goto-table is the last action. */ | |
6954 | struct ofpact_goto_table *ogt = ofpact_get_GOTO_TABLE(a); | |
6955 | struct rule_dpif *rule; | |
6956 | ||
6957 | ovs_assert(ctx->table_id < ogt->table_id); | |
6958 | ||
6959 | ctx->table_id = ogt->table_id; | |
6960 | ||
6961 | /* Look up a flow from the new table. */ | |
6962 | rule = rule_dpif_lookup__(ctx->ofproto, &ctx->flow, ctx->table_id); | |
6963 | ||
6964 | tag_the_flow(ctx, rule); | |
6965 | ||
6966 | rule = ctx_rule_hooks(ctx, rule, true); | |
6967 | ||
6968 | if (rule) { | |
6969 | if (ctx->rule) { | |
6970 | ctx->rule->up.evictable = was_evictable; | |
6971 | } | |
6972 | ctx->rule = rule; | |
6973 | was_evictable = rule->up.evictable; | |
6974 | rule->up.evictable = false; | |
6975 | ||
6976 | /* Tail recursion removal. */ | |
6977 | ofpacts = rule->up.ofpacts; | |
6978 | ofpacts_len = rule->up.ofpacts_len; | |
6979 | goto do_xlate_actions_again; | |
6980 | } | |
6981 | break; | |
6982 | } | |
6983 | ||
6984 | case OFPACT_SAMPLE: | |
6985 | xlate_sample_action(ctx, ofpact_get_SAMPLE(a)); | |
6986 | break; | |
6987 | } | |
6988 | } | |
6989 | ||
6990 | out: | |
6991 | if (ctx->rule) { | |
6992 | ctx->rule->up.evictable = was_evictable; | |
6993 | } | |
6994 | } | |
6995 | ||
6996 | static void | |
6997 | action_xlate_ctx_init(struct action_xlate_ctx *ctx, | |
6998 | struct ofproto_dpif *ofproto, const struct flow *flow, | |
6999 | const struct initial_vals *initial_vals, | |
7000 | struct rule_dpif *rule, | |
7001 | uint8_t tcp_flags, const struct ofpbuf *packet) | |
7002 | { | |
7003 | /* Flow initialization rules: | |
7004 | * - 'base_flow' must match the kernel's view of the packet at the | |
7005 | * time that action processing starts. 'flow' represents any | |
7006 | * transformations we wish to make through actions. | |
7007 | * - By default 'base_flow' and 'flow' are the same since the input | |
7008 | * packet matches the output before any actions are applied. | |
7009 | * - When using VLAN splinters, 'base_flow''s VLAN is set to the value | |
7010 | * of the received packet as seen by the kernel. If we later output | |
7011 | * to another device without any modifications this will cause us to | |
7012 | * insert a new tag since the original one was stripped off by the | |
7013 | * VLAN device. | |
7014 | * - Tunnel metadata as received is retained in 'flow'. This allows | |
7015 | * tunnel metadata matching also in later tables. | |
7016 | * Since a kernel action for setting the tunnel metadata will only be | |
7017 | * generated with actual tunnel output, changing the tunnel metadata | |
7018 | * values in 'flow' (such as tun_id) will only have effect with a later | |
7019 | * tunnel output action. | |
7020 | * - Tunnel 'base_flow' is completely cleared since that is what the | |
7021 | * kernel does. If we wish to maintain the original values an action | |
7022 | * needs to be generated. */ | |
7023 | ||
7024 | ctx->ofproto = ofproto; | |
7025 | ctx->flow = *flow; | |
7026 | ctx->base_flow = ctx->flow; | |
7027 | memset(&ctx->base_flow.tunnel, 0, sizeof ctx->base_flow.tunnel); | |
7028 | ctx->rule = rule; | |
7029 | ctx->packet = packet; | |
7030 | ctx->may_learn = packet != NULL; | |
7031 | ctx->tcp_flags = tcp_flags; | |
7032 | ctx->resubmit_hook = NULL; | |
7033 | ctx->report_hook = NULL; | |
7034 | ctx->resubmit_stats = NULL; | |
7035 | ||
7036 | if (initial_vals) { | |
7037 | ctx->base_flow.vlan_tci = initial_vals->vlan_tci; | |
7038 | } | |
7039 | } | |
7040 | ||
7041 | /* Translates the 'ofpacts_len' bytes of "struct ofpacts" starting at 'ofpacts' | |
7042 | * into datapath actions in 'odp_actions', using 'ctx'. */ | |
7043 | static void | |
7044 | xlate_actions(struct action_xlate_ctx *ctx, | |
7045 | const struct ofpact *ofpacts, size_t ofpacts_len, | |
7046 | struct ofpbuf *odp_actions) | |
7047 | { | |
7048 | /* Normally false. Set to true if we ever hit MAX_RESUBMIT_RECURSION, so | |
7049 | * that in the future we always keep a copy of the original flow for | |
7050 | * tracing purposes. */ | |
7051 | static bool hit_resubmit_limit; | |
7052 | ||
7053 | enum slow_path_reason special; | |
7054 | struct ofport_dpif *in_port; | |
7055 | struct flow orig_flow; | |
7056 | ||
7057 | COVERAGE_INC(ofproto_dpif_xlate); | |
7058 | ||
7059 | ofpbuf_clear(odp_actions); | |
7060 | ofpbuf_reserve(odp_actions, NL_A_U32_SIZE); | |
7061 | ||
7062 | ctx->odp_actions = odp_actions; | |
7063 | ctx->tags = 0; | |
7064 | ctx->slow = 0; | |
7065 | ctx->has_learn = false; | |
7066 | ctx->has_normal = false; | |
7067 | ctx->has_fin_timeout = false; | |
7068 | ctx->nf_output_iface = NF_OUT_DROP; | |
7069 | ctx->mirrors = 0; | |
7070 | ctx->recurse = 0; | |
7071 | ctx->max_resubmit_trigger = false; | |
7072 | ctx->orig_skb_priority = ctx->flow.skb_priority; | |
7073 | ctx->table_id = 0; | |
7074 | ctx->exit = false; | |
7075 | ||
7076 | ofpbuf_use_stub(&ctx->stack, ctx->init_stack, sizeof ctx->init_stack); | |
7077 | ||
7078 | if (ctx->ofproto->has_mirrors || hit_resubmit_limit) { | |
7079 | /* Do this conditionally because the copy is expensive enough that it | |
7080 | * shows up in profiles. */ | |
7081 | orig_flow = ctx->flow; | |
7082 | } | |
7083 | ||
7084 | if (ctx->flow.nw_frag & FLOW_NW_FRAG_ANY) { | |
7085 | switch (ctx->ofproto->up.frag_handling) { | |
7086 | case OFPC_FRAG_NORMAL: | |
7087 | /* We must pretend that transport ports are unavailable. */ | |
7088 | ctx->flow.tp_src = ctx->base_flow.tp_src = htons(0); | |
7089 | ctx->flow.tp_dst = ctx->base_flow.tp_dst = htons(0); | |
7090 | break; | |
7091 | ||
7092 | case OFPC_FRAG_DROP: | |
7093 | return; | |
7094 | ||
7095 | case OFPC_FRAG_REASM: | |
7096 | NOT_REACHED(); | |
7097 | ||
7098 | case OFPC_FRAG_NX_MATCH: | |
7099 | /* Nothing to do. */ | |
7100 | break; | |
7101 | ||
7102 | case OFPC_INVALID_TTL_TO_CONTROLLER: | |
7103 | NOT_REACHED(); | |
7104 | } | |
7105 | } | |
7106 | ||
7107 | in_port = get_ofp_port(ctx->ofproto, ctx->flow.in_port); | |
7108 | special = process_special(ctx->ofproto, &ctx->flow, in_port, ctx->packet); | |
7109 | if (special) { | |
7110 | ctx->slow |= special; | |
7111 | } else { | |
7112 | static struct vlog_rate_limit trace_rl = VLOG_RATE_LIMIT_INIT(1, 1); | |
7113 | struct initial_vals initial_vals; | |
7114 | size_t sample_actions_len; | |
7115 | uint32_t local_odp_port; | |
7116 | ||
7117 | initial_vals.vlan_tci = ctx->base_flow.vlan_tci; | |
7118 | ||
7119 | add_sflow_action(ctx); | |
7120 | add_ipfix_action(ctx); | |
7121 | sample_actions_len = ctx->odp_actions->size; | |
7122 | ||
7123 | if (tunnel_ecn_ok(ctx) && (!in_port || may_receive(in_port, ctx))) { | |
7124 | do_xlate_actions(ofpacts, ofpacts_len, ctx); | |
7125 | ||
7126 | /* We've let OFPP_NORMAL and the learning action look at the | |
7127 | * packet, so drop it now if forwarding is disabled. */ | |
7128 | if (in_port && !stp_forward_in_state(in_port->stp_state)) { | |
7129 | ctx->odp_actions->size = sample_actions_len; | |
7130 | } | |
7131 | } | |
7132 | ||
7133 | if (ctx->max_resubmit_trigger && !ctx->resubmit_hook) { | |
7134 | if (!hit_resubmit_limit) { | |
7135 | /* We didn't record the original flow. Make sure we do from | |
7136 | * now on. */ | |
7137 | hit_resubmit_limit = true; | |
7138 | } else if (!VLOG_DROP_ERR(&trace_rl)) { | |
7139 | struct ds ds = DS_EMPTY_INITIALIZER; | |
7140 | ||
7141 | ofproto_trace(ctx->ofproto, &orig_flow, ctx->packet, | |
7142 | &initial_vals, &ds); | |
7143 | VLOG_ERR("Trace triggered by excessive resubmit " | |
7144 | "recursion:\n%s", ds_cstr(&ds)); | |
7145 | ds_destroy(&ds); | |
7146 | } | |
7147 | } | |
7148 | ||
7149 | local_odp_port = ofp_port_to_odp_port(ctx->ofproto, OFPP_LOCAL); | |
7150 | if (!connmgr_may_set_up_flow(ctx->ofproto->up.connmgr, &ctx->flow, | |
7151 | local_odp_port, | |
7152 | ctx->odp_actions->data, | |
7153 | ctx->odp_actions->size)) { | |
7154 | ctx->slow |= SLOW_IN_BAND; | |
7155 | if (ctx->packet | |
7156 | && connmgr_msg_in_hook(ctx->ofproto->up.connmgr, &ctx->flow, | |
7157 | ctx->packet)) { | |
7158 | compose_output_action(ctx, OFPP_LOCAL); | |
7159 | } | |
7160 | } | |
7161 | if (ctx->ofproto->has_mirrors) { | |
7162 | add_mirror_actions(ctx, &orig_flow); | |
7163 | } | |
7164 | fix_sflow_action(ctx); | |
7165 | } | |
7166 | ||
7167 | ofpbuf_uninit(&ctx->stack); | |
7168 | } | |
7169 | ||
7170 | /* Translates the 'ofpacts_len' bytes of "struct ofpact"s starting at 'ofpacts' | |
7171 | * into datapath actions, using 'ctx', and discards the datapath actions. */ | |
7172 | static void | |
7173 | xlate_actions_for_side_effects(struct action_xlate_ctx *ctx, | |
7174 | const struct ofpact *ofpacts, | |
7175 | size_t ofpacts_len) | |
7176 | { | |
7177 | uint64_t odp_actions_stub[1024 / 8]; | |
7178 | struct ofpbuf odp_actions; | |
7179 | ||
7180 | ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub); | |
7181 | xlate_actions(ctx, ofpacts, ofpacts_len, &odp_actions); | |
7182 | ofpbuf_uninit(&odp_actions); | |
7183 | } | |
7184 | ||
7185 | static void | |
7186 | xlate_report(struct action_xlate_ctx *ctx, const char *s) | |
7187 | { | |
7188 | if (ctx->report_hook) { | |
7189 | ctx->report_hook(ctx, s); | |
7190 | } | |
7191 | } | |
7192 | \f | |
7193 | /* OFPP_NORMAL implementation. */ | |
7194 | ||
7195 | static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *); | |
7196 | ||
7197 | /* Given 'vid', the VID obtained from the 802.1Q header that was received as | |
7198 | * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle', | |
7199 | * the bundle on which the packet was received, returns the VLAN to which the | |
7200 | * packet belongs. | |
7201 | * | |
7202 | * Both 'vid' and the return value are in the range 0...4095. */ | |
7203 | static uint16_t | |
7204 | input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid) | |
7205 | { | |
7206 | switch (in_bundle->vlan_mode) { | |
7207 | case PORT_VLAN_ACCESS: | |
7208 | return in_bundle->vlan; | |
7209 | break; | |
7210 | ||
7211 | case PORT_VLAN_TRUNK: | |
7212 | return vid; | |
7213 | ||
7214 | case PORT_VLAN_NATIVE_UNTAGGED: | |
7215 | case PORT_VLAN_NATIVE_TAGGED: | |
7216 | return vid ? vid : in_bundle->vlan; | |
7217 | ||
7218 | default: | |
7219 | NOT_REACHED(); | |
7220 | } | |
7221 | } | |
7222 | ||
7223 | /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'. | |
7224 | * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs | |
7225 | * a warning. | |
7226 | * | |
7227 | * 'vid' should be the VID obtained from the 802.1Q header that was received as | |
7228 | * part of a packet (specify 0 if there was no 802.1Q header), in the range | |
7229 | * 0...4095. */ | |
7230 | static bool | |
7231 | input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn) | |
7232 | { | |
7233 | /* Allow any VID on the OFPP_NONE port. */ | |
7234 | if (in_bundle == &ofpp_none_bundle) { | |
7235 | return true; | |
7236 | } | |
7237 | ||
7238 | switch (in_bundle->vlan_mode) { | |
7239 | case PORT_VLAN_ACCESS: | |
7240 | if (vid) { | |
7241 | if (warn) { | |
7242 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
7243 | VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged " | |
7244 | "packet received on port %s configured as VLAN " | |
7245 | "%"PRIu16" access port", | |
7246 | in_bundle->ofproto->up.name, vid, | |
7247 | in_bundle->name, in_bundle->vlan); | |
7248 | } | |
7249 | return false; | |
7250 | } | |
7251 | return true; | |
7252 | ||
7253 | case PORT_VLAN_NATIVE_UNTAGGED: | |
7254 | case PORT_VLAN_NATIVE_TAGGED: | |
7255 | if (!vid) { | |
7256 | /* Port must always carry its native VLAN. */ | |
7257 | return true; | |
7258 | } | |
7259 | /* Fall through. */ | |
7260 | case PORT_VLAN_TRUNK: | |
7261 | if (!ofbundle_includes_vlan(in_bundle, vid)) { | |
7262 | if (warn) { | |
7263 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
7264 | VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet " | |
7265 | "received on port %s not configured for trunking " | |
7266 | "VLAN %"PRIu16, | |
7267 | in_bundle->ofproto->up.name, vid, | |
7268 | in_bundle->name, vid); | |
7269 | } | |
7270 | return false; | |
7271 | } | |
7272 | return true; | |
7273 | ||
7274 | default: | |
7275 | NOT_REACHED(); | |
7276 | } | |
7277 | ||
7278 | } | |
7279 | ||
7280 | /* Given 'vlan', the VLAN that a packet belongs to, and | |
7281 | * 'out_bundle', a bundle on which the packet is to be output, returns the VID | |
7282 | * that should be included in the 802.1Q header. (If the return value is 0, | |
7283 | * then the 802.1Q header should only be included in the packet if there is a | |
7284 | * nonzero PCP.) | |
7285 | * | |
7286 | * Both 'vlan' and the return value are in the range 0...4095. */ | |
7287 | static uint16_t | |
7288 | output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan) | |
7289 | { | |
7290 | switch (out_bundle->vlan_mode) { | |
7291 | case PORT_VLAN_ACCESS: | |
7292 | return 0; | |
7293 | ||
7294 | case PORT_VLAN_TRUNK: | |
7295 | case PORT_VLAN_NATIVE_TAGGED: | |
7296 | return vlan; | |
7297 | ||
7298 | case PORT_VLAN_NATIVE_UNTAGGED: | |
7299 | return vlan == out_bundle->vlan ? 0 : vlan; | |
7300 | ||
7301 | default: | |
7302 | NOT_REACHED(); | |
7303 | } | |
7304 | } | |
7305 | ||
7306 | static void | |
7307 | output_normal(struct action_xlate_ctx *ctx, const struct ofbundle *out_bundle, | |
7308 | uint16_t vlan) | |
7309 | { | |
7310 | struct ofport_dpif *port; | |
7311 | uint16_t vid; | |
7312 | ovs_be16 tci, old_tci; | |
7313 | ||
7314 | vid = output_vlan_to_vid(out_bundle, vlan); | |
7315 | if (!out_bundle->bond) { | |
7316 | port = ofbundle_get_a_port(out_bundle); | |
7317 | } else { | |
7318 | port = bond_choose_output_slave(out_bundle->bond, &ctx->flow, | |
7319 | vid, &ctx->tags); | |
7320 | if (!port) { | |
7321 | /* No slaves enabled, so drop packet. */ | |
7322 | return; | |
7323 | } | |
7324 | } | |
7325 | ||
7326 | old_tci = ctx->flow.vlan_tci; | |
7327 | tci = htons(vid); | |
7328 | if (tci || out_bundle->use_priority_tags) { | |
7329 | tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK); | |
7330 | if (tci) { | |
7331 | tci |= htons(VLAN_CFI); | |
7332 | } | |
7333 | } | |
7334 | ctx->flow.vlan_tci = tci; | |
7335 | ||
7336 | compose_output_action(ctx, port->up.ofp_port); | |
7337 | ctx->flow.vlan_tci = old_tci; | |
7338 | } | |
7339 | ||
7340 | static int | |
7341 | mirror_mask_ffs(mirror_mask_t mask) | |
7342 | { | |
7343 | BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask)); | |
7344 | return ffs(mask); | |
7345 | } | |
7346 | ||
7347 | static bool | |
7348 | ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan) | |
7349 | { | |
7350 | return (bundle->vlan_mode != PORT_VLAN_ACCESS | |
7351 | && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan))); | |
7352 | } | |
7353 | ||
7354 | static bool | |
7355 | ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan) | |
7356 | { | |
7357 | return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan); | |
7358 | } | |
7359 | ||
7360 | /* Returns an arbitrary interface within 'bundle'. */ | |
7361 | static struct ofport_dpif * | |
7362 | ofbundle_get_a_port(const struct ofbundle *bundle) | |
7363 | { | |
7364 | return CONTAINER_OF(list_front(&bundle->ports), | |
7365 | struct ofport_dpif, bundle_node); | |
7366 | } | |
7367 | ||
7368 | static bool | |
7369 | vlan_is_mirrored(const struct ofmirror *m, int vlan) | |
7370 | { | |
7371 | return !m->vlans || bitmap_is_set(m->vlans, vlan); | |
7372 | } | |
7373 | ||
7374 | static void | |
7375 | add_mirror_actions(struct action_xlate_ctx *ctx, const struct flow *orig_flow) | |
7376 | { | |
7377 | struct ofproto_dpif *ofproto = ctx->ofproto; | |
7378 | mirror_mask_t mirrors; | |
7379 | struct ofbundle *in_bundle; | |
7380 | uint16_t vlan; | |
7381 | uint16_t vid; | |
7382 | const struct nlattr *a; | |
7383 | size_t left; | |
7384 | ||
7385 | in_bundle = lookup_input_bundle(ctx->ofproto, orig_flow->in_port, | |
7386 | ctx->packet != NULL, NULL); | |
7387 | if (!in_bundle) { | |
7388 | return; | |
7389 | } | |
7390 | mirrors = in_bundle->src_mirrors; | |
7391 | ||
7392 | /* Drop frames on bundles reserved for mirroring. */ | |
7393 | if (in_bundle->mirror_out) { | |
7394 | if (ctx->packet != NULL) { | |
7395 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
7396 | VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port " | |
7397 | "%s, which is reserved exclusively for mirroring", | |
7398 | ctx->ofproto->up.name, in_bundle->name); | |
7399 | } | |
7400 | return; | |
7401 | } | |
7402 | ||
7403 | /* Check VLAN. */ | |
7404 | vid = vlan_tci_to_vid(orig_flow->vlan_tci); | |
7405 | if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) { | |
7406 | return; | |
7407 | } | |
7408 | vlan = input_vid_to_vlan(in_bundle, vid); | |
7409 | ||
7410 | /* Look at the output ports to check for destination selections. */ | |
7411 | ||
7412 | NL_ATTR_FOR_EACH (a, left, ctx->odp_actions->data, | |
7413 | ctx->odp_actions->size) { | |
7414 | enum ovs_action_attr type = nl_attr_type(a); | |
7415 | struct ofport_dpif *ofport; | |
7416 | ||
7417 | if (type != OVS_ACTION_ATTR_OUTPUT) { | |
7418 | continue; | |
7419 | } | |
7420 | ||
7421 | ofport = get_odp_port(ofproto, nl_attr_get_u32(a)); | |
7422 | if (ofport && ofport->bundle) { | |
7423 | mirrors |= ofport->bundle->dst_mirrors; | |
7424 | } | |
7425 | } | |
7426 | ||
7427 | if (!mirrors) { | |
7428 | return; | |
7429 | } | |
7430 | ||
7431 | /* Restore the original packet before adding the mirror actions. */ | |
7432 | ctx->flow = *orig_flow; | |
7433 | ||
7434 | while (mirrors) { | |
7435 | struct ofmirror *m; | |
7436 | ||
7437 | m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1]; | |
7438 | ||
7439 | if (!vlan_is_mirrored(m, vlan)) { | |
7440 | mirrors = zero_rightmost_1bit(mirrors); | |
7441 | continue; | |
7442 | } | |
7443 | ||
7444 | mirrors &= ~m->dup_mirrors; | |
7445 | ctx->mirrors |= m->dup_mirrors; | |
7446 | if (m->out) { | |
7447 | output_normal(ctx, m->out, vlan); | |
7448 | } else if (vlan != m->out_vlan | |
7449 | && !eth_addr_is_reserved(orig_flow->dl_dst)) { | |
7450 | struct ofbundle *bundle; | |
7451 | ||
7452 | HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { | |
7453 | if (ofbundle_includes_vlan(bundle, m->out_vlan) | |
7454 | && !bundle->mirror_out) { | |
7455 | output_normal(ctx, bundle, m->out_vlan); | |
7456 | } | |
7457 | } | |
7458 | } | |
7459 | } | |
7460 | } | |
7461 | ||
7462 | static void | |
7463 | update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors, | |
7464 | uint64_t packets, uint64_t bytes) | |
7465 | { | |
7466 | if (!mirrors) { | |
7467 | return; | |
7468 | } | |
7469 | ||
7470 | for (; mirrors; mirrors = zero_rightmost_1bit(mirrors)) { | |
7471 | struct ofmirror *m; | |
7472 | ||
7473 | m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1]; | |
7474 | ||
7475 | if (!m) { | |
7476 | /* In normal circumstances 'm' will not be NULL. However, | |
7477 | * if mirrors are reconfigured, we can temporarily get out | |
7478 | * of sync in facet_revalidate(). We could "correct" the | |
7479 | * mirror list before reaching here, but doing that would | |
7480 | * not properly account the traffic stats we've currently | |
7481 | * accumulated for previous mirror configuration. */ | |
7482 | continue; | |
7483 | } | |
7484 | ||
7485 | m->packet_count += packets; | |
7486 | m->byte_count += bytes; | |
7487 | } | |
7488 | } | |
7489 | ||
7490 | /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after | |
7491 | * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to | |
7492 | * indicate this; newer upstream kernels use gratuitous ARP requests. */ | |
7493 | static bool | |
7494 | is_gratuitous_arp(const struct flow *flow) | |
7495 | { | |
7496 | return (flow->dl_type == htons(ETH_TYPE_ARP) | |
7497 | && eth_addr_is_broadcast(flow->dl_dst) | |
7498 | && (flow->nw_proto == ARP_OP_REPLY | |
7499 | || (flow->nw_proto == ARP_OP_REQUEST | |
7500 | && flow->nw_src == flow->nw_dst))); | |
7501 | } | |
7502 | ||
7503 | static void | |
7504 | update_learning_table(struct ofproto_dpif *ofproto, | |
7505 | const struct flow *flow, int vlan, | |
7506 | struct ofbundle *in_bundle) | |
7507 | { | |
7508 | struct mac_entry *mac; | |
7509 | ||
7510 | /* Don't learn the OFPP_NONE port. */ | |
7511 | if (in_bundle == &ofpp_none_bundle) { | |
7512 | return; | |
7513 | } | |
7514 | ||
7515 | if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) { | |
7516 | return; | |
7517 | } | |
7518 | ||
7519 | mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan); | |
7520 | if (is_gratuitous_arp(flow)) { | |
7521 | /* We don't want to learn from gratuitous ARP packets that are | |
7522 | * reflected back over bond slaves so we lock the learning table. */ | |
7523 | if (!in_bundle->bond) { | |
7524 | mac_entry_set_grat_arp_lock(mac); | |
7525 | } else if (mac_entry_is_grat_arp_locked(mac)) { | |
7526 | return; | |
7527 | } | |
7528 | } | |
7529 | ||
7530 | if (mac_entry_is_new(mac) || mac->port.p != in_bundle) { | |
7531 | /* The log messages here could actually be useful in debugging, | |
7532 | * so keep the rate limit relatively high. */ | |
7533 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300); | |
7534 | VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is " | |
7535 | "on port %s in VLAN %d", | |
7536 | ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src), | |
7537 | in_bundle->name, vlan); | |
7538 | ||
7539 | mac->port.p = in_bundle; | |
7540 | tag_set_add(&ofproto->backer->revalidate_set, | |
7541 | mac_learning_changed(ofproto->ml, mac)); | |
7542 | } | |
7543 | } | |
7544 | ||
7545 | static struct ofbundle * | |
7546 | lookup_input_bundle(const struct ofproto_dpif *ofproto, uint16_t in_port, | |
7547 | bool warn, struct ofport_dpif **in_ofportp) | |
7548 | { | |
7549 | struct ofport_dpif *ofport; | |
7550 | ||
7551 | /* Find the port and bundle for the received packet. */ | |
7552 | ofport = get_ofp_port(ofproto, in_port); | |
7553 | if (in_ofportp) { | |
7554 | *in_ofportp = ofport; | |
7555 | } | |
7556 | if (ofport && ofport->bundle) { | |
7557 | return ofport->bundle; | |
7558 | } | |
7559 | ||
7560 | /* Special-case OFPP_NONE, which a controller may use as the ingress | |
7561 | * port for traffic that it is sourcing. */ | |
7562 | if (in_port == OFPP_NONE) { | |
7563 | return &ofpp_none_bundle; | |
7564 | } | |
7565 | ||
7566 | /* Odd. A few possible reasons here: | |
7567 | * | |
7568 | * - We deleted a port but there are still a few packets queued up | |
7569 | * from it. | |
7570 | * | |
7571 | * - Someone externally added a port (e.g. "ovs-dpctl add-if") that | |
7572 | * we don't know about. | |
7573 | * | |
7574 | * - The ofproto client didn't configure the port as part of a bundle. | |
7575 | * This is particularly likely to happen if a packet was received on the | |
7576 | * port after it was created, but before the client had a chance to | |
7577 | * configure its bundle. | |
7578 | */ | |
7579 | if (warn) { | |
7580 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
7581 | ||
7582 | VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown " | |
7583 | "port %"PRIu16, ofproto->up.name, in_port); | |
7584 | } | |
7585 | return NULL; | |
7586 | } | |
7587 | ||
7588 | /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or | |
7589 | * dropped. Returns true if they may be forwarded, false if they should be | |
7590 | * dropped. | |
7591 | * | |
7592 | * 'in_port' must be the ofport_dpif that corresponds to flow->in_port. | |
7593 | * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull). | |
7594 | * | |
7595 | * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as | |
7596 | * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as | |
7597 | * checked by input_vid_is_valid(). | |
7598 | * | |
7599 | * May also add tags to '*tags', although the current implementation only does | |
7600 | * so in one special case. | |
7601 | */ | |
7602 | static bool | |
7603 | is_admissible(struct action_xlate_ctx *ctx, struct ofport_dpif *in_port, | |
7604 | uint16_t vlan) | |
7605 | { | |
7606 | struct ofproto_dpif *ofproto = ctx->ofproto; | |
7607 | struct flow *flow = &ctx->flow; | |
7608 | struct ofbundle *in_bundle = in_port->bundle; | |
7609 | ||
7610 | /* Drop frames for reserved multicast addresses | |
7611 | * only if forward_bpdu option is absent. */ | |
7612 | if (!ofproto->up.forward_bpdu && eth_addr_is_reserved(flow->dl_dst)) { | |
7613 | xlate_report(ctx, "packet has reserved destination MAC, dropping"); | |
7614 | return false; | |
7615 | } | |
7616 | ||
7617 | if (in_bundle->bond) { | |
7618 | struct mac_entry *mac; | |
7619 | ||
7620 | switch (bond_check_admissibility(in_bundle->bond, in_port, | |
7621 | flow->dl_dst, &ctx->tags)) { | |
7622 | case BV_ACCEPT: | |
7623 | break; | |
7624 | ||
7625 | case BV_DROP: | |
7626 | xlate_report(ctx, "bonding refused admissibility, dropping"); | |
7627 | return false; | |
7628 | ||
7629 | case BV_DROP_IF_MOVED: | |
7630 | mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL); | |
7631 | if (mac && mac->port.p != in_bundle && | |
7632 | (!is_gratuitous_arp(flow) | |
7633 | || mac_entry_is_grat_arp_locked(mac))) { | |
7634 | xlate_report(ctx, "SLB bond thinks this packet looped back, " | |
7635 | "dropping"); | |
7636 | return false; | |
7637 | } | |
7638 | break; | |
7639 | } | |
7640 | } | |
7641 | ||
7642 | return true; | |
7643 | } | |
7644 | ||
7645 | static void | |
7646 | xlate_normal(struct action_xlate_ctx *ctx) | |
7647 | { | |
7648 | struct ofport_dpif *in_port; | |
7649 | struct ofbundle *in_bundle; | |
7650 | struct mac_entry *mac; | |
7651 | uint16_t vlan; | |
7652 | uint16_t vid; | |
7653 | ||
7654 | ctx->has_normal = true; | |
7655 | ||
7656 | in_bundle = lookup_input_bundle(ctx->ofproto, ctx->flow.in_port, | |
7657 | ctx->packet != NULL, &in_port); | |
7658 | if (!in_bundle) { | |
7659 | xlate_report(ctx, "no input bundle, dropping"); | |
7660 | return; | |
7661 | } | |
7662 | ||
7663 | /* Drop malformed frames. */ | |
7664 | if (ctx->flow.dl_type == htons(ETH_TYPE_VLAN) && | |
7665 | !(ctx->flow.vlan_tci & htons(VLAN_CFI))) { | |
7666 | if (ctx->packet != NULL) { | |
7667 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
7668 | VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial " | |
7669 | "VLAN tag received on port %s", | |
7670 | ctx->ofproto->up.name, in_bundle->name); | |
7671 | } | |
7672 | xlate_report(ctx, "partial VLAN tag, dropping"); | |
7673 | return; | |
7674 | } | |
7675 | ||
7676 | /* Drop frames on bundles reserved for mirroring. */ | |
7677 | if (in_bundle->mirror_out) { | |
7678 | if (ctx->packet != NULL) { | |
7679 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
7680 | VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port " | |
7681 | "%s, which is reserved exclusively for mirroring", | |
7682 | ctx->ofproto->up.name, in_bundle->name); | |
7683 | } | |
7684 | xlate_report(ctx, "input port is mirror output port, dropping"); | |
7685 | return; | |
7686 | } | |
7687 | ||
7688 | /* Check VLAN. */ | |
7689 | vid = vlan_tci_to_vid(ctx->flow.vlan_tci); | |
7690 | if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) { | |
7691 | xlate_report(ctx, "disallowed VLAN VID for this input port, dropping"); | |
7692 | return; | |
7693 | } | |
7694 | vlan = input_vid_to_vlan(in_bundle, vid); | |
7695 | ||
7696 | /* Check other admissibility requirements. */ | |
7697 | if (in_port && !is_admissible(ctx, in_port, vlan)) { | |
7698 | return; | |
7699 | } | |
7700 | ||
7701 | /* Learn source MAC. */ | |
7702 | if (ctx->may_learn) { | |
7703 | update_learning_table(ctx->ofproto, &ctx->flow, vlan, in_bundle); | |
7704 | } | |
7705 | ||
7706 | /* Determine output bundle. */ | |
7707 | mac = mac_learning_lookup(ctx->ofproto->ml, ctx->flow.dl_dst, vlan, | |
7708 | &ctx->tags); | |
7709 | if (mac) { | |
7710 | if (mac->port.p != in_bundle) { | |
7711 | xlate_report(ctx, "forwarding to learned port"); | |
7712 | output_normal(ctx, mac->port.p, vlan); | |
7713 | } else { | |
7714 | xlate_report(ctx, "learned port is input port, dropping"); | |
7715 | } | |
7716 | } else { | |
7717 | struct ofbundle *bundle; | |
7718 | ||
7719 | xlate_report(ctx, "no learned MAC for destination, flooding"); | |
7720 | HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) { | |
7721 | if (bundle != in_bundle | |
7722 | && ofbundle_includes_vlan(bundle, vlan) | |
7723 | && bundle->floodable | |
7724 | && !bundle->mirror_out) { | |
7725 | output_normal(ctx, bundle, vlan); | |
7726 | } | |
7727 | } | |
7728 | ctx->nf_output_iface = NF_OUT_FLOOD; | |
7729 | } | |
7730 | } | |
7731 | \f | |
7732 | /* Optimized flow revalidation. | |
7733 | * | |
7734 | * It's a difficult problem, in general, to tell which facets need to have | |
7735 | * their actions recalculated whenever the OpenFlow flow table changes. We | |
7736 | * don't try to solve that general problem: for most kinds of OpenFlow flow | |
7737 | * table changes, we recalculate the actions for every facet. This is | |
7738 | * relatively expensive, but it's good enough if the OpenFlow flow table | |
7739 | * doesn't change very often. | |
7740 | * | |
7741 | * However, we can expect one particular kind of OpenFlow flow table change to | |
7742 | * happen frequently: changes caused by MAC learning. To avoid wasting a lot | |
7743 | * of CPU on revalidating every facet whenever MAC learning modifies the flow | |
7744 | * table, we add a special case that applies to flow tables in which every rule | |
7745 | * has the same form (that is, the same wildcards), except that the table is | |
7746 | * also allowed to have a single "catch-all" flow that matches all packets. We | |
7747 | * optimize this case by tagging all of the facets that resubmit into the table | |
7748 | * and invalidating the same tag whenever a flow changes in that table. The | |
7749 | * end result is that we revalidate just the facets that need it (and sometimes | |
7750 | * a few more, but not all of the facets or even all of the facets that | |
7751 | * resubmit to the table modified by MAC learning). */ | |
7752 | ||
7753 | /* Calculates the tag to use for 'flow' and mask 'mask' when it is inserted | |
7754 | * into an OpenFlow table with the given 'basis'. */ | |
7755 | static tag_type | |
7756 | rule_calculate_tag(const struct flow *flow, const struct minimask *mask, | |
7757 | uint32_t secret) | |
7758 | { | |
7759 | if (minimask_is_catchall(mask)) { | |
7760 | return 0; | |
7761 | } else { | |
7762 | uint32_t hash = flow_hash_in_minimask(flow, mask, secret); | |
7763 | return tag_create_deterministic(hash); | |
7764 | } | |
7765 | } | |
7766 | ||
7767 | /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the | |
7768 | * taggability of that table. | |
7769 | * | |
7770 | * This function must be called after *each* change to a flow table. If you | |
7771 | * skip calling it on some changes then the pointer comparisons at the end can | |
7772 | * be invalid if you get unlucky. For example, if a flow removal causes a | |
7773 | * cls_table to be destroyed and then a flow insertion causes a cls_table with | |
7774 | * different wildcards to be created with the same address, then this function | |
7775 | * will incorrectly skip revalidation. */ | |
7776 | static void | |
7777 | table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id) | |
7778 | { | |
7779 | struct table_dpif *table = &ofproto->tables[table_id]; | |
7780 | const struct oftable *oftable = &ofproto->up.tables[table_id]; | |
7781 | struct cls_table *catchall, *other; | |
7782 | struct cls_table *t; | |
7783 | ||
7784 | catchall = other = NULL; | |
7785 | ||
7786 | switch (hmap_count(&oftable->cls.tables)) { | |
7787 | case 0: | |
7788 | /* We could tag this OpenFlow table but it would make the logic a | |
7789 | * little harder and it's a corner case that doesn't seem worth it | |
7790 | * yet. */ | |
7791 | break; | |
7792 | ||
7793 | case 1: | |
7794 | case 2: | |
7795 | HMAP_FOR_EACH (t, hmap_node, &oftable->cls.tables) { | |
7796 | if (cls_table_is_catchall(t)) { | |
7797 | catchall = t; | |
7798 | } else if (!other) { | |
7799 | other = t; | |
7800 | } else { | |
7801 | /* Indicate that we can't tag this by setting both tables to | |
7802 | * NULL. (We know that 'catchall' is already NULL.) */ | |
7803 | other = NULL; | |
7804 | } | |
7805 | } | |
7806 | break; | |
7807 | ||
7808 | default: | |
7809 | /* Can't tag this table. */ | |
7810 | break; | |
7811 | } | |
7812 | ||
7813 | if (table->catchall_table != catchall || table->other_table != other) { | |
7814 | table->catchall_table = catchall; | |
7815 | table->other_table = other; | |
7816 | ofproto->backer->need_revalidate = REV_FLOW_TABLE; | |
7817 | } | |
7818 | } | |
7819 | ||
7820 | /* Given 'rule' that has changed in some way (either it is a rule being | |
7821 | * inserted, a rule being deleted, or a rule whose actions are being | |
7822 | * modified), marks facets for revalidation to ensure that packets will be | |
7823 | * forwarded correctly according to the new state of the flow table. | |
7824 | * | |
7825 | * This function must be called after *each* change to a flow table. See | |
7826 | * the comment on table_update_taggable() for more information. */ | |
7827 | static void | |
7828 | rule_invalidate(const struct rule_dpif *rule) | |
7829 | { | |
7830 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
7831 | ||
7832 | table_update_taggable(ofproto, rule->up.table_id); | |
7833 | ||
7834 | if (!ofproto->backer->need_revalidate) { | |
7835 | struct table_dpif *table = &ofproto->tables[rule->up.table_id]; | |
7836 | ||
7837 | if (table->other_table && rule->tag) { | |
7838 | tag_set_add(&ofproto->backer->revalidate_set, rule->tag); | |
7839 | } else { | |
7840 | ofproto->backer->need_revalidate = REV_FLOW_TABLE; | |
7841 | } | |
7842 | } | |
7843 | } | |
7844 | \f | |
7845 | static bool | |
7846 | set_frag_handling(struct ofproto *ofproto_, | |
7847 | enum ofp_config_flags frag_handling) | |
7848 | { | |
7849 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
7850 | if (frag_handling != OFPC_FRAG_REASM) { | |
7851 | ofproto->backer->need_revalidate = REV_RECONFIGURE; | |
7852 | return true; | |
7853 | } else { | |
7854 | return false; | |
7855 | } | |
7856 | } | |
7857 | ||
7858 | static enum ofperr | |
7859 | packet_out(struct ofproto *ofproto_, struct ofpbuf *packet, | |
7860 | const struct flow *flow, | |
7861 | const struct ofpact *ofpacts, size_t ofpacts_len) | |
7862 | { | |
7863 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
7864 | struct initial_vals initial_vals; | |
7865 | struct odputil_keybuf keybuf; | |
7866 | struct dpif_flow_stats stats; | |
7867 | ||
7868 | struct ofpbuf key; | |
7869 | ||
7870 | struct action_xlate_ctx ctx; | |
7871 | uint64_t odp_actions_stub[1024 / 8]; | |
7872 | struct ofpbuf odp_actions; | |
7873 | ||
7874 | ofpbuf_use_stack(&key, &keybuf, sizeof keybuf); | |
7875 | odp_flow_key_from_flow(&key, flow, | |
7876 | ofp_port_to_odp_port(ofproto, flow->in_port)); | |
7877 | ||
7878 | dpif_flow_stats_extract(flow, packet, time_msec(), &stats); | |
7879 | ||
7880 | initial_vals.vlan_tci = flow->vlan_tci; | |
7881 | action_xlate_ctx_init(&ctx, ofproto, flow, &initial_vals, NULL, | |
7882 | packet_get_tcp_flags(packet, flow), packet); | |
7883 | ctx.resubmit_stats = &stats; | |
7884 | ||
7885 | ofpbuf_use_stub(&odp_actions, | |
7886 | odp_actions_stub, sizeof odp_actions_stub); | |
7887 | xlate_actions(&ctx, ofpacts, ofpacts_len, &odp_actions); | |
7888 | dpif_execute(ofproto->backer->dpif, key.data, key.size, | |
7889 | odp_actions.data, odp_actions.size, packet); | |
7890 | ofpbuf_uninit(&odp_actions); | |
7891 | ||
7892 | return 0; | |
7893 | } | |
7894 | \f | |
7895 | /* NetFlow. */ | |
7896 | ||
7897 | static int | |
7898 | set_netflow(struct ofproto *ofproto_, | |
7899 | const struct netflow_options *netflow_options) | |
7900 | { | |
7901 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
7902 | ||
7903 | if (netflow_options) { | |
7904 | if (!ofproto->netflow) { | |
7905 | ofproto->netflow = netflow_create(); | |
7906 | } | |
7907 | return netflow_set_options(ofproto->netflow, netflow_options); | |
7908 | } else { | |
7909 | netflow_destroy(ofproto->netflow); | |
7910 | ofproto->netflow = NULL; | |
7911 | return 0; | |
7912 | } | |
7913 | } | |
7914 | ||
7915 | static void | |
7916 | get_netflow_ids(const struct ofproto *ofproto_, | |
7917 | uint8_t *engine_type, uint8_t *engine_id) | |
7918 | { | |
7919 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
7920 | ||
7921 | dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id); | |
7922 | } | |
7923 | ||
7924 | static void | |
7925 | send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet) | |
7926 | { | |
7927 | if (!facet_is_controller_flow(facet) && | |
7928 | netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) { | |
7929 | struct subfacet *subfacet; | |
7930 | struct ofexpired expired; | |
7931 | ||
7932 | LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) { | |
7933 | if (subfacet->path == SF_FAST_PATH) { | |
7934 | struct dpif_flow_stats stats; | |
7935 | ||
7936 | subfacet_reinstall(subfacet, &stats); | |
7937 | subfacet_update_stats(subfacet, &stats); | |
7938 | } | |
7939 | } | |
7940 | ||
7941 | expired.flow = facet->flow; | |
7942 | expired.packet_count = facet->packet_count; | |
7943 | expired.byte_count = facet->byte_count; | |
7944 | expired.used = facet->used; | |
7945 | netflow_expire(ofproto->netflow, &facet->nf_flow, &expired); | |
7946 | } | |
7947 | } | |
7948 | ||
7949 | static void | |
7950 | send_netflow_active_timeouts(struct ofproto_dpif *ofproto) | |
7951 | { | |
7952 | struct facet *facet; | |
7953 | ||
7954 | HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) { | |
7955 | send_active_timeout(ofproto, facet); | |
7956 | } | |
7957 | } | |
7958 | \f | |
7959 | static struct ofproto_dpif * | |
7960 | ofproto_dpif_lookup(const char *name) | |
7961 | { | |
7962 | struct ofproto_dpif *ofproto; | |
7963 | ||
7964 | HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node, | |
7965 | hash_string(name, 0), &all_ofproto_dpifs) { | |
7966 | if (!strcmp(ofproto->up.name, name)) { | |
7967 | return ofproto; | |
7968 | } | |
7969 | } | |
7970 | return NULL; | |
7971 | } | |
7972 | ||
7973 | static void | |
7974 | ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc, | |
7975 | const char *argv[], void *aux OVS_UNUSED) | |
7976 | { | |
7977 | struct ofproto_dpif *ofproto; | |
7978 | ||
7979 | if (argc > 1) { | |
7980 | ofproto = ofproto_dpif_lookup(argv[1]); | |
7981 | if (!ofproto) { | |
7982 | unixctl_command_reply_error(conn, "no such bridge"); | |
7983 | return; | |
7984 | } | |
7985 | mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set); | |
7986 | } else { | |
7987 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
7988 | mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set); | |
7989 | } | |
7990 | } | |
7991 | ||
7992 | unixctl_command_reply(conn, "table successfully flushed"); | |
7993 | } | |
7994 | ||
7995 | static void | |
7996 | ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED, | |
7997 | const char *argv[], void *aux OVS_UNUSED) | |
7998 | { | |
7999 | struct ds ds = DS_EMPTY_INITIALIZER; | |
8000 | const struct ofproto_dpif *ofproto; | |
8001 | const struct mac_entry *e; | |
8002 | ||
8003 | ofproto = ofproto_dpif_lookup(argv[1]); | |
8004 | if (!ofproto) { | |
8005 | unixctl_command_reply_error(conn, "no such bridge"); | |
8006 | return; | |
8007 | } | |
8008 | ||
8009 | ds_put_cstr(&ds, " port VLAN MAC Age\n"); | |
8010 | LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) { | |
8011 | struct ofbundle *bundle = e->port.p; | |
8012 | ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n", | |
8013 | ofbundle_get_a_port(bundle)->odp_port, | |
8014 | e->vlan, ETH_ADDR_ARGS(e->mac), | |
8015 | mac_entry_age(ofproto->ml, e)); | |
8016 | } | |
8017 | unixctl_command_reply(conn, ds_cstr(&ds)); | |
8018 | ds_destroy(&ds); | |
8019 | } | |
8020 | ||
8021 | struct trace_ctx { | |
8022 | struct action_xlate_ctx ctx; | |
8023 | struct flow flow; | |
8024 | struct ds *result; | |
8025 | }; | |
8026 | ||
8027 | static void | |
8028 | trace_format_rule(struct ds *result, uint8_t table_id, int level, | |
8029 | const struct rule_dpif *rule) | |
8030 | { | |
8031 | ds_put_char_multiple(result, '\t', level); | |
8032 | if (!rule) { | |
8033 | ds_put_cstr(result, "No match\n"); | |
8034 | return; | |
8035 | } | |
8036 | ||
8037 | ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ", | |
8038 | table_id, ntohll(rule->up.flow_cookie)); | |
8039 | cls_rule_format(&rule->up.cr, result); | |
8040 | ds_put_char(result, '\n'); | |
8041 | ||
8042 | ds_put_char_multiple(result, '\t', level); | |
8043 | ds_put_cstr(result, "OpenFlow "); | |
8044 | ofpacts_format(rule->up.ofpacts, rule->up.ofpacts_len, result); | |
8045 | ds_put_char(result, '\n'); | |
8046 | } | |
8047 | ||
8048 | static void | |
8049 | trace_format_flow(struct ds *result, int level, const char *title, | |
8050 | struct trace_ctx *trace) | |
8051 | { | |
8052 | ds_put_char_multiple(result, '\t', level); | |
8053 | ds_put_format(result, "%s: ", title); | |
8054 | if (flow_equal(&trace->ctx.flow, &trace->flow)) { | |
8055 | ds_put_cstr(result, "unchanged"); | |
8056 | } else { | |
8057 | flow_format(result, &trace->ctx.flow); | |
8058 | trace->flow = trace->ctx.flow; | |
8059 | } | |
8060 | ds_put_char(result, '\n'); | |
8061 | } | |
8062 | ||
8063 | static void | |
8064 | trace_format_regs(struct ds *result, int level, const char *title, | |
8065 | struct trace_ctx *trace) | |
8066 | { | |
8067 | size_t i; | |
8068 | ||
8069 | ds_put_char_multiple(result, '\t', level); | |
8070 | ds_put_format(result, "%s:", title); | |
8071 | for (i = 0; i < FLOW_N_REGS; i++) { | |
8072 | ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]); | |
8073 | } | |
8074 | ds_put_char(result, '\n'); | |
8075 | } | |
8076 | ||
8077 | static void | |
8078 | trace_format_odp(struct ds *result, int level, const char *title, | |
8079 | struct trace_ctx *trace) | |
8080 | { | |
8081 | struct ofpbuf *odp_actions = trace->ctx.odp_actions; | |
8082 | ||
8083 | ds_put_char_multiple(result, '\t', level); | |
8084 | ds_put_format(result, "%s: ", title); | |
8085 | format_odp_actions(result, odp_actions->data, odp_actions->size); | |
8086 | ds_put_char(result, '\n'); | |
8087 | } | |
8088 | ||
8089 | static void | |
8090 | trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule) | |
8091 | { | |
8092 | struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx); | |
8093 | struct ds *result = trace->result; | |
8094 | ||
8095 | ds_put_char(result, '\n'); | |
8096 | trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace); | |
8097 | trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace); | |
8098 | trace_format_odp(result, ctx->recurse + 1, "Resubmitted odp", trace); | |
8099 | trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule); | |
8100 | } | |
8101 | ||
8102 | static void | |
8103 | trace_report(struct action_xlate_ctx *ctx, const char *s) | |
8104 | { | |
8105 | struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx); | |
8106 | struct ds *result = trace->result; | |
8107 | ||
8108 | ds_put_char_multiple(result, '\t', ctx->recurse); | |
8109 | ds_put_cstr(result, s); | |
8110 | ds_put_char(result, '\n'); | |
8111 | } | |
8112 | ||
8113 | static void | |
8114 | ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[], | |
8115 | void *aux OVS_UNUSED) | |
8116 | { | |
8117 | const char *dpname = argv[1]; | |
8118 | struct ofproto_dpif *ofproto; | |
8119 | struct ofpbuf odp_key; | |
8120 | struct ofpbuf *packet; | |
8121 | struct initial_vals initial_vals; | |
8122 | struct ds result; | |
8123 | struct flow flow; | |
8124 | char *s; | |
8125 | ||
8126 | packet = NULL; | |
8127 | ofpbuf_init(&odp_key, 0); | |
8128 | ds_init(&result); | |
8129 | ||
8130 | ofproto = ofproto_dpif_lookup(dpname); | |
8131 | if (!ofproto) { | |
8132 | unixctl_command_reply_error(conn, "Unknown ofproto (use ofproto/list " | |
8133 | "for help)"); | |
8134 | goto exit; | |
8135 | } | |
8136 | if (argc == 3 || (argc == 4 && !strcmp(argv[3], "-generate"))) { | |
8137 | /* ofproto/trace dpname flow [-generate] */ | |
8138 | const char *flow_s = argv[2]; | |
8139 | const char *generate_s = argv[3]; | |
8140 | ||
8141 | /* Allow 'flow_s' to be either a datapath flow or an OpenFlow-like | |
8142 | * flow. We guess which type it is based on whether 'flow_s' contains | |
8143 | * an '(', since a datapath flow always contains '(') but an | |
8144 | * OpenFlow-like flow should not (in fact it's allowed but I believe | |
8145 | * that's not documented anywhere). | |
8146 | * | |
8147 | * An alternative would be to try to parse 'flow_s' both ways, but then | |
8148 | * it would be tricky giving a sensible error message. After all, do | |
8149 | * you just say "syntax error" or do you present both error messages? | |
8150 | * Both choices seem lousy. */ | |
8151 | if (strchr(flow_s, '(')) { | |
8152 | int error; | |
8153 | ||
8154 | /* Convert string to datapath key. */ | |
8155 | ofpbuf_init(&odp_key, 0); | |
8156 | error = odp_flow_key_from_string(flow_s, NULL, &odp_key); | |
8157 | if (error) { | |
8158 | unixctl_command_reply_error(conn, "Bad flow syntax"); | |
8159 | goto exit; | |
8160 | } | |
8161 | ||
8162 | /* The user might have specified the wrong ofproto but within the | |
8163 | * same backer. That's OK, ofproto_receive() can find the right | |
8164 | * one for us. */ | |
8165 | if (ofproto_receive(ofproto->backer, NULL, odp_key.data, | |
8166 | odp_key.size, &flow, NULL, &ofproto, NULL, | |
8167 | &initial_vals)) { | |
8168 | unixctl_command_reply_error(conn, "Invalid flow"); | |
8169 | goto exit; | |
8170 | } | |
8171 | ds_put_format(&result, "Bridge: %s\n", ofproto->up.name); | |
8172 | } else { | |
8173 | char *error_s; | |
8174 | ||
8175 | error_s = parse_ofp_exact_flow(&flow, argv[2]); | |
8176 | if (error_s) { | |
8177 | unixctl_command_reply_error(conn, error_s); | |
8178 | free(error_s); | |
8179 | goto exit; | |
8180 | } | |
8181 | ||
8182 | initial_vals.vlan_tci = flow.vlan_tci; | |
8183 | } | |
8184 | ||
8185 | /* Generate a packet, if requested. */ | |
8186 | if (generate_s) { | |
8187 | packet = ofpbuf_new(0); | |
8188 | flow_compose(packet, &flow); | |
8189 | } | |
8190 | } else if (argc == 7) { | |
8191 | /* ofproto/trace dpname priority tun_id in_port mark packet */ | |
8192 | const char *priority_s = argv[2]; | |
8193 | const char *tun_id_s = argv[3]; | |
8194 | const char *in_port_s = argv[4]; | |
8195 | const char *mark_s = argv[5]; | |
8196 | const char *packet_s = argv[6]; | |
8197 | uint32_t in_port = atoi(in_port_s); | |
8198 | ovs_be64 tun_id = htonll(strtoull(tun_id_s, NULL, 0)); | |
8199 | uint32_t priority = atoi(priority_s); | |
8200 | uint32_t mark = atoi(mark_s); | |
8201 | const char *msg; | |
8202 | ||
8203 | msg = eth_from_hex(packet_s, &packet); | |
8204 | if (msg) { | |
8205 | unixctl_command_reply_error(conn, msg); | |
8206 | goto exit; | |
8207 | } | |
8208 | ||
8209 | ds_put_cstr(&result, "Packet: "); | |
8210 | s = ofp_packet_to_string(packet->data, packet->size); | |
8211 | ds_put_cstr(&result, s); | |
8212 | free(s); | |
8213 | ||
8214 | flow_extract(packet, priority, mark, NULL, in_port, &flow); | |
8215 | flow.tunnel.tun_id = tun_id; | |
8216 | initial_vals.vlan_tci = flow.vlan_tci; | |
8217 | } else { | |
8218 | unixctl_command_reply_error(conn, "Bad command syntax"); | |
8219 | goto exit; | |
8220 | } | |
8221 | ||
8222 | ofproto_trace(ofproto, &flow, packet, &initial_vals, &result); | |
8223 | unixctl_command_reply(conn, ds_cstr(&result)); | |
8224 | ||
8225 | exit: | |
8226 | ds_destroy(&result); | |
8227 | ofpbuf_delete(packet); | |
8228 | ofpbuf_uninit(&odp_key); | |
8229 | } | |
8230 | ||
8231 | static void | |
8232 | ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow, | |
8233 | const struct ofpbuf *packet, | |
8234 | const struct initial_vals *initial_vals, struct ds *ds) | |
8235 | { | |
8236 | struct rule_dpif *rule; | |
8237 | ||
8238 | ds_put_cstr(ds, "Flow: "); | |
8239 | flow_format(ds, flow); | |
8240 | ds_put_char(ds, '\n'); | |
8241 | ||
8242 | rule = rule_dpif_lookup(ofproto, flow); | |
8243 | ||
8244 | trace_format_rule(ds, 0, 0, rule); | |
8245 | if (rule == ofproto->miss_rule) { | |
8246 | ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n"); | |
8247 | } else if (rule == ofproto->no_packet_in_rule) { | |
8248 | ds_put_cstr(ds, "\nNo match, packets dropped because " | |
8249 | "OFPPC_NO_PACKET_IN is set on in_port.\n"); | |
8250 | } | |
8251 | ||
8252 | if (rule) { | |
8253 | uint64_t odp_actions_stub[1024 / 8]; | |
8254 | struct ofpbuf odp_actions; | |
8255 | ||
8256 | struct trace_ctx trace; | |
8257 | uint8_t tcp_flags; | |
8258 | ||
8259 | tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0; | |
8260 | trace.result = ds; | |
8261 | trace.flow = *flow; | |
8262 | ofpbuf_use_stub(&odp_actions, | |
8263 | odp_actions_stub, sizeof odp_actions_stub); | |
8264 | action_xlate_ctx_init(&trace.ctx, ofproto, flow, initial_vals, | |
8265 | rule, tcp_flags, packet); | |
8266 | trace.ctx.resubmit_hook = trace_resubmit; | |
8267 | trace.ctx.report_hook = trace_report; | |
8268 | xlate_actions(&trace.ctx, rule->up.ofpacts, rule->up.ofpacts_len, | |
8269 | &odp_actions); | |
8270 | ||
8271 | ds_put_char(ds, '\n'); | |
8272 | trace_format_flow(ds, 0, "Final flow", &trace); | |
8273 | ds_put_cstr(ds, "Datapath actions: "); | |
8274 | format_odp_actions(ds, odp_actions.data, odp_actions.size); | |
8275 | ofpbuf_uninit(&odp_actions); | |
8276 | ||
8277 | if (trace.ctx.slow) { | |
8278 | enum slow_path_reason slow; | |
8279 | ||
8280 | ds_put_cstr(ds, "\nThis flow is handled by the userspace " | |
8281 | "slow path because it:"); | |
8282 | for (slow = trace.ctx.slow; slow; ) { | |
8283 | enum slow_path_reason bit = rightmost_1bit(slow); | |
8284 | ||
8285 | switch (bit) { | |
8286 | case SLOW_CFM: | |
8287 | ds_put_cstr(ds, "\n\t- Consists of CFM packets."); | |
8288 | break; | |
8289 | case SLOW_LACP: | |
8290 | ds_put_cstr(ds, "\n\t- Consists of LACP packets."); | |
8291 | break; | |
8292 | case SLOW_STP: | |
8293 | ds_put_cstr(ds, "\n\t- Consists of STP packets."); | |
8294 | break; | |
8295 | case SLOW_BFD: | |
8296 | ds_put_cstr(ds, "\n\t- Consists of BFD packets."); | |
8297 | break; | |
8298 | case SLOW_IN_BAND: | |
8299 | ds_put_cstr(ds, "\n\t- Needs in-band special case " | |
8300 | "processing."); | |
8301 | if (!packet) { | |
8302 | ds_put_cstr(ds, "\n\t (The datapath actions are " | |
8303 | "incomplete--for complete actions, " | |
8304 | "please supply a packet.)"); | |
8305 | } | |
8306 | break; | |
8307 | case SLOW_CONTROLLER: | |
8308 | ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages " | |
8309 | "to the OpenFlow controller."); | |
8310 | break; | |
8311 | case SLOW_MATCH: | |
8312 | ds_put_cstr(ds, "\n\t- Needs more specific matching " | |
8313 | "than the datapath supports."); | |
8314 | break; | |
8315 | } | |
8316 | ||
8317 | slow &= ~bit; | |
8318 | } | |
8319 | ||
8320 | if (slow & ~SLOW_MATCH) { | |
8321 | ds_put_cstr(ds, "\nThe datapath actions above do not reflect " | |
8322 | "the special slow-path processing."); | |
8323 | } | |
8324 | } | |
8325 | } | |
8326 | } | |
8327 | ||
8328 | static void | |
8329 | ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED, | |
8330 | const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED) | |
8331 | { | |
8332 | clogged = true; | |
8333 | unixctl_command_reply(conn, NULL); | |
8334 | } | |
8335 | ||
8336 | static void | |
8337 | ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED, | |
8338 | const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED) | |
8339 | { | |
8340 | clogged = false; | |
8341 | unixctl_command_reply(conn, NULL); | |
8342 | } | |
8343 | ||
8344 | /* Runs a self-check of flow translations in 'ofproto'. Appends a message to | |
8345 | * 'reply' describing the results. */ | |
8346 | static void | |
8347 | ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply) | |
8348 | { | |
8349 | struct facet *facet; | |
8350 | int errors; | |
8351 | ||
8352 | errors = 0; | |
8353 | HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) { | |
8354 | if (!facet_check_consistency(facet)) { | |
8355 | errors++; | |
8356 | } | |
8357 | } | |
8358 | if (errors) { | |
8359 | ofproto->backer->need_revalidate = REV_INCONSISTENCY; | |
8360 | } | |
8361 | ||
8362 | if (errors) { | |
8363 | ds_put_format(reply, "%s: self-check failed (%d errors)\n", | |
8364 | ofproto->up.name, errors); | |
8365 | } else { | |
8366 | ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name); | |
8367 | } | |
8368 | } | |
8369 | ||
8370 | static void | |
8371 | ofproto_dpif_self_check(struct unixctl_conn *conn, | |
8372 | int argc, const char *argv[], void *aux OVS_UNUSED) | |
8373 | { | |
8374 | struct ds reply = DS_EMPTY_INITIALIZER; | |
8375 | struct ofproto_dpif *ofproto; | |
8376 | ||
8377 | if (argc > 1) { | |
8378 | ofproto = ofproto_dpif_lookup(argv[1]); | |
8379 | if (!ofproto) { | |
8380 | unixctl_command_reply_error(conn, "Unknown ofproto (use " | |
8381 | "ofproto/list for help)"); | |
8382 | return; | |
8383 | } | |
8384 | ofproto_dpif_self_check__(ofproto, &reply); | |
8385 | } else { | |
8386 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
8387 | ofproto_dpif_self_check__(ofproto, &reply); | |
8388 | } | |
8389 | } | |
8390 | ||
8391 | unixctl_command_reply(conn, ds_cstr(&reply)); | |
8392 | ds_destroy(&reply); | |
8393 | } | |
8394 | ||
8395 | /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list | |
8396 | * of the 'ofproto_shash' nodes. It is the responsibility of the caller | |
8397 | * to destroy 'ofproto_shash' and free the returned value. */ | |
8398 | static const struct shash_node ** | |
8399 | get_ofprotos(struct shash *ofproto_shash) | |
8400 | { | |
8401 | const struct ofproto_dpif *ofproto; | |
8402 | ||
8403 | HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { | |
8404 | char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name); | |
8405 | shash_add_nocopy(ofproto_shash, name, ofproto); | |
8406 | } | |
8407 | ||
8408 | return shash_sort(ofproto_shash); | |
8409 | } | |
8410 | ||
8411 | static void | |
8412 | ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED, | |
8413 | const char *argv[] OVS_UNUSED, | |
8414 | void *aux OVS_UNUSED) | |
8415 | { | |
8416 | struct ds ds = DS_EMPTY_INITIALIZER; | |
8417 | struct shash ofproto_shash; | |
8418 | const struct shash_node **sorted_ofprotos; | |
8419 | int i; | |
8420 | ||
8421 | shash_init(&ofproto_shash); | |
8422 | sorted_ofprotos = get_ofprotos(&ofproto_shash); | |
8423 | for (i = 0; i < shash_count(&ofproto_shash); i++) { | |
8424 | const struct shash_node *node = sorted_ofprotos[i]; | |
8425 | ds_put_format(&ds, "%s\n", node->name); | |
8426 | } | |
8427 | ||
8428 | shash_destroy(&ofproto_shash); | |
8429 | free(sorted_ofprotos); | |
8430 | ||
8431 | unixctl_command_reply(conn, ds_cstr(&ds)); | |
8432 | ds_destroy(&ds); | |
8433 | } | |
8434 | ||
8435 | static void | |
8436 | show_dp_format(const struct ofproto_dpif *ofproto, struct ds *ds) | |
8437 | { | |
8438 | const struct shash_node **ports; | |
8439 | int i; | |
8440 | struct avg_subfacet_rates lifetime; | |
8441 | unsigned long long int minutes; | |
8442 | const int min_ms = 60 * 1000; /* milliseconds in one minute. */ | |
8443 | ||
8444 | minutes = (time_msec() - ofproto->created) / min_ms; | |
8445 | ||
8446 | if (minutes > 0) { | |
8447 | lifetime.add_rate = (double)ofproto->total_subfacet_add_count | |
8448 | / minutes; | |
8449 | lifetime.del_rate = (double)ofproto->total_subfacet_del_count | |
8450 | / minutes; | |
8451 | }else { | |
8452 | lifetime.add_rate = 0.0; | |
8453 | lifetime.del_rate = 0.0; | |
8454 | } | |
8455 | ||
8456 | ds_put_format(ds, "%s (%s):\n", ofproto->up.name, | |
8457 | dpif_name(ofproto->backer->dpif)); | |
8458 | ds_put_format(ds, | |
8459 | "\tlookups: hit:%"PRIu64" missed:%"PRIu64"\n", | |
8460 | ofproto->n_hit, ofproto->n_missed); | |
8461 | ds_put_format(ds, "\tflows: cur: %zu, avg: %5.3f, max: %d," | |
8462 | " life span: %llu(ms)\n", | |
8463 | hmap_count(&ofproto->subfacets), | |
8464 | avg_subfacet_count(ofproto), | |
8465 | ofproto->max_n_subfacet, | |
8466 | avg_subfacet_life_span(ofproto)); | |
8467 | if (minutes >= 60) { | |
8468 | show_dp_rates(ds, "\t\thourly avg:", &ofproto->hourly); | |
8469 | } | |
8470 | if (minutes >= 60 * 24) { | |
8471 | show_dp_rates(ds, "\t\tdaily avg:", &ofproto->daily); | |
8472 | } | |
8473 | show_dp_rates(ds, "\t\toverall avg:", &lifetime); | |
8474 | ||
8475 | ports = shash_sort(&ofproto->up.port_by_name); | |
8476 | for (i = 0; i < shash_count(&ofproto->up.port_by_name); i++) { | |
8477 | const struct shash_node *node = ports[i]; | |
8478 | struct ofport *ofport = node->data; | |
8479 | const char *name = netdev_get_name(ofport->netdev); | |
8480 | const char *type = netdev_get_type(ofport->netdev); | |
8481 | uint32_t odp_port; | |
8482 | ||
8483 | ds_put_format(ds, "\t%s %u/", name, ofport->ofp_port); | |
8484 | ||
8485 | odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port); | |
8486 | if (odp_port != OVSP_NONE) { | |
8487 | ds_put_format(ds, "%"PRIu32":", odp_port); | |
8488 | } else { | |
8489 | ds_put_cstr(ds, "none:"); | |
8490 | } | |
8491 | ||
8492 | if (strcmp(type, "system")) { | |
8493 | struct netdev *netdev; | |
8494 | int error; | |
8495 | ||
8496 | ds_put_format(ds, " (%s", type); | |
8497 | ||
8498 | error = netdev_open(name, type, &netdev); | |
8499 | if (!error) { | |
8500 | struct smap config; | |
8501 | ||
8502 | smap_init(&config); | |
8503 | error = netdev_get_config(netdev, &config); | |
8504 | if (!error) { | |
8505 | const struct smap_node **nodes; | |
8506 | size_t i; | |
8507 | ||
8508 | nodes = smap_sort(&config); | |
8509 | for (i = 0; i < smap_count(&config); i++) { | |
8510 | const struct smap_node *node = nodes[i]; | |
8511 | ds_put_format(ds, "%c %s=%s", i ? ',' : ':', | |
8512 | node->key, node->value); | |
8513 | } | |
8514 | free(nodes); | |
8515 | } | |
8516 | smap_destroy(&config); | |
8517 | ||
8518 | netdev_close(netdev); | |
8519 | } | |
8520 | ds_put_char(ds, ')'); | |
8521 | } | |
8522 | ds_put_char(ds, '\n'); | |
8523 | } | |
8524 | free(ports); | |
8525 | } | |
8526 | ||
8527 | static void | |
8528 | ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc, | |
8529 | const char *argv[], void *aux OVS_UNUSED) | |
8530 | { | |
8531 | struct ds ds = DS_EMPTY_INITIALIZER; | |
8532 | const struct ofproto_dpif *ofproto; | |
8533 | ||
8534 | if (argc > 1) { | |
8535 | int i; | |
8536 | for (i = 1; i < argc; i++) { | |
8537 | ofproto = ofproto_dpif_lookup(argv[i]); | |
8538 | if (!ofproto) { | |
8539 | ds_put_format(&ds, "Unknown bridge %s (use dpif/dump-dps " | |
8540 | "for help)", argv[i]); | |
8541 | unixctl_command_reply_error(conn, ds_cstr(&ds)); | |
8542 | return; | |
8543 | } | |
8544 | show_dp_format(ofproto, &ds); | |
8545 | } | |
8546 | } else { | |
8547 | struct shash ofproto_shash; | |
8548 | const struct shash_node **sorted_ofprotos; | |
8549 | int i; | |
8550 | ||
8551 | shash_init(&ofproto_shash); | |
8552 | sorted_ofprotos = get_ofprotos(&ofproto_shash); | |
8553 | for (i = 0; i < shash_count(&ofproto_shash); i++) { | |
8554 | const struct shash_node *node = sorted_ofprotos[i]; | |
8555 | show_dp_format(node->data, &ds); | |
8556 | } | |
8557 | ||
8558 | shash_destroy(&ofproto_shash); | |
8559 | free(sorted_ofprotos); | |
8560 | } | |
8561 | ||
8562 | unixctl_command_reply(conn, ds_cstr(&ds)); | |
8563 | ds_destroy(&ds); | |
8564 | } | |
8565 | ||
8566 | static void | |
8567 | ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn, | |
8568 | int argc OVS_UNUSED, const char *argv[], | |
8569 | void *aux OVS_UNUSED) | |
8570 | { | |
8571 | struct ds ds = DS_EMPTY_INITIALIZER; | |
8572 | const struct ofproto_dpif *ofproto; | |
8573 | struct subfacet *subfacet; | |
8574 | ||
8575 | ofproto = ofproto_dpif_lookup(argv[1]); | |
8576 | if (!ofproto) { | |
8577 | unixctl_command_reply_error(conn, "no such bridge"); | |
8578 | return; | |
8579 | } | |
8580 | ||
8581 | update_stats(ofproto->backer); | |
8582 | ||
8583 | HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) { | |
8584 | odp_flow_key_format(subfacet->key, subfacet->key_len, &ds); | |
8585 | ||
8586 | ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:", | |
8587 | subfacet->dp_packet_count, subfacet->dp_byte_count); | |
8588 | if (subfacet->used) { | |
8589 | ds_put_format(&ds, "%.3fs", | |
8590 | (time_msec() - subfacet->used) / 1000.0); | |
8591 | } else { | |
8592 | ds_put_format(&ds, "never"); | |
8593 | } | |
8594 | if (subfacet->facet->tcp_flags) { | |
8595 | ds_put_cstr(&ds, ", flags:"); | |
8596 | packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags); | |
8597 | } | |
8598 | ||
8599 | ds_put_cstr(&ds, ", actions:"); | |
8600 | if (subfacet->slow) { | |
8601 | uint64_t slow_path_stub[128 / 8]; | |
8602 | const struct nlattr *actions; | |
8603 | size_t actions_len; | |
8604 | ||
8605 | compose_slow_path(ofproto, &subfacet->facet->flow, subfacet->slow, | |
8606 | slow_path_stub, sizeof slow_path_stub, | |
8607 | &actions, &actions_len); | |
8608 | format_odp_actions(&ds, actions, actions_len); | |
8609 | } else { | |
8610 | format_odp_actions(&ds, subfacet->actions, subfacet->actions_len); | |
8611 | } | |
8612 | ds_put_char(&ds, '\n'); | |
8613 | } | |
8614 | ||
8615 | unixctl_command_reply(conn, ds_cstr(&ds)); | |
8616 | ds_destroy(&ds); | |
8617 | } | |
8618 | ||
8619 | static void | |
8620 | ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn, | |
8621 | int argc OVS_UNUSED, const char *argv[], | |
8622 | void *aux OVS_UNUSED) | |
8623 | { | |
8624 | struct ds ds = DS_EMPTY_INITIALIZER; | |
8625 | struct ofproto_dpif *ofproto; | |
8626 | ||
8627 | ofproto = ofproto_dpif_lookup(argv[1]); | |
8628 | if (!ofproto) { | |
8629 | unixctl_command_reply_error(conn, "no such bridge"); | |
8630 | return; | |
8631 | } | |
8632 | ||
8633 | flush(&ofproto->up); | |
8634 | ||
8635 | unixctl_command_reply(conn, ds_cstr(&ds)); | |
8636 | ds_destroy(&ds); | |
8637 | } | |
8638 | ||
8639 | static void | |
8640 | ofproto_dpif_unixctl_init(void) | |
8641 | { | |
8642 | static bool registered; | |
8643 | if (registered) { | |
8644 | return; | |
8645 | } | |
8646 | registered = true; | |
8647 | ||
8648 | unixctl_command_register( | |
8649 | "ofproto/trace", | |
8650 | "bridge {priority tun_id in_port mark packet | odp_flow [-generate]}", | |
8651 | 2, 6, ofproto_unixctl_trace, NULL); | |
8652 | unixctl_command_register("fdb/flush", "[bridge]", 0, 1, | |
8653 | ofproto_unixctl_fdb_flush, NULL); | |
8654 | unixctl_command_register("fdb/show", "bridge", 1, 1, | |
8655 | ofproto_unixctl_fdb_show, NULL); | |
8656 | unixctl_command_register("ofproto/clog", "", 0, 0, | |
8657 | ofproto_dpif_clog, NULL); | |
8658 | unixctl_command_register("ofproto/unclog", "", 0, 0, | |
8659 | ofproto_dpif_unclog, NULL); | |
8660 | unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1, | |
8661 | ofproto_dpif_self_check, NULL); | |
8662 | unixctl_command_register("dpif/dump-dps", "", 0, 0, | |
8663 | ofproto_unixctl_dpif_dump_dps, NULL); | |
8664 | unixctl_command_register("dpif/show", "[bridge]", 0, INT_MAX, | |
8665 | ofproto_unixctl_dpif_show, NULL); | |
8666 | unixctl_command_register("dpif/dump-flows", "bridge", 1, 1, | |
8667 | ofproto_unixctl_dpif_dump_flows, NULL); | |
8668 | unixctl_command_register("dpif/del-flows", "bridge", 1, 1, | |
8669 | ofproto_unixctl_dpif_del_flows, NULL); | |
8670 | } | |
8671 | \f | |
8672 | /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.) | |
8673 | * | |
8674 | * This is deprecated. It is only for compatibility with broken device drivers | |
8675 | * in old versions of Linux that do not properly support VLANs when VLAN | |
8676 | * devices are not used. When broken device drivers are no longer in | |
8677 | * widespread use, we will delete these interfaces. */ | |
8678 | ||
8679 | static int | |
8680 | set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid) | |
8681 | { | |
8682 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto); | |
8683 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
8684 | ||
8685 | if (realdev_ofp_port == ofport->realdev_ofp_port | |
8686 | && vid == ofport->vlandev_vid) { | |
8687 | return 0; | |
8688 | } | |
8689 | ||
8690 | ofproto->backer->need_revalidate = REV_RECONFIGURE; | |
8691 | ||
8692 | if (ofport->realdev_ofp_port) { | |
8693 | vsp_remove(ofport); | |
8694 | } | |
8695 | if (realdev_ofp_port && ofport->bundle) { | |
8696 | /* vlandevs are enslaved to their realdevs, so they are not allowed to | |
8697 | * themselves be part of a bundle. */ | |
8698 | bundle_set(ofport->up.ofproto, ofport->bundle, NULL); | |
8699 | } | |
8700 | ||
8701 | ofport->realdev_ofp_port = realdev_ofp_port; | |
8702 | ofport->vlandev_vid = vid; | |
8703 | ||
8704 | if (realdev_ofp_port) { | |
8705 | vsp_add(ofport, realdev_ofp_port, vid); | |
8706 | } | |
8707 | ||
8708 | return 0; | |
8709 | } | |
8710 | ||
8711 | static uint32_t | |
8712 | hash_realdev_vid(uint16_t realdev_ofp_port, int vid) | |
8713 | { | |
8714 | return hash_2words(realdev_ofp_port, vid); | |
8715 | } | |
8716 | ||
8717 | /* Returns the ODP port number of the Linux VLAN device that corresponds to | |
8718 | * 'vlan_tci' on the network device with port number 'realdev_odp_port' in | |
8719 | * 'ofproto'. For example, given 'realdev_odp_port' of eth0 and 'vlan_tci' 9, | |
8720 | * it would return the port number of eth0.9. | |
8721 | * | |
8722 | * Unless VLAN splinters are enabled for port 'realdev_odp_port', this | |
8723 | * function just returns its 'realdev_odp_port' argument. */ | |
8724 | static uint32_t | |
8725 | vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto, | |
8726 | uint32_t realdev_odp_port, ovs_be16 vlan_tci) | |
8727 | { | |
8728 | if (!hmap_is_empty(&ofproto->realdev_vid_map)) { | |
8729 | uint16_t realdev_ofp_port; | |
8730 | int vid = vlan_tci_to_vid(vlan_tci); | |
8731 | const struct vlan_splinter *vsp; | |
8732 | ||
8733 | realdev_ofp_port = odp_port_to_ofp_port(ofproto, realdev_odp_port); | |
8734 | HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node, | |
8735 | hash_realdev_vid(realdev_ofp_port, vid), | |
8736 | &ofproto->realdev_vid_map) { | |
8737 | if (vsp->realdev_ofp_port == realdev_ofp_port | |
8738 | && vsp->vid == vid) { | |
8739 | return ofp_port_to_odp_port(ofproto, vsp->vlandev_ofp_port); | |
8740 | } | |
8741 | } | |
8742 | } | |
8743 | return realdev_odp_port; | |
8744 | } | |
8745 | ||
8746 | static struct vlan_splinter * | |
8747 | vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port) | |
8748 | { | |
8749 | struct vlan_splinter *vsp; | |
8750 | ||
8751 | HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, hash_int(vlandev_ofp_port, 0), | |
8752 | &ofproto->vlandev_map) { | |
8753 | if (vsp->vlandev_ofp_port == vlandev_ofp_port) { | |
8754 | return vsp; | |
8755 | } | |
8756 | } | |
8757 | ||
8758 | return NULL; | |
8759 | } | |
8760 | ||
8761 | /* Returns the OpenFlow port number of the "real" device underlying the Linux | |
8762 | * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the | |
8763 | * VLAN VID of the Linux VLAN device in '*vid'. For example, given | |
8764 | * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of | |
8765 | * eth0 and store 9 in '*vid'. | |
8766 | * | |
8767 | * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux | |
8768 | * VLAN device. Unless VLAN splinters are enabled, this is what this function | |
8769 | * always does.*/ | |
8770 | static uint16_t | |
8771 | vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto, | |
8772 | uint16_t vlandev_ofp_port, int *vid) | |
8773 | { | |
8774 | if (!hmap_is_empty(&ofproto->vlandev_map)) { | |
8775 | const struct vlan_splinter *vsp; | |
8776 | ||
8777 | vsp = vlandev_find(ofproto, vlandev_ofp_port); | |
8778 | if (vsp) { | |
8779 | if (vid) { | |
8780 | *vid = vsp->vid; | |
8781 | } | |
8782 | return vsp->realdev_ofp_port; | |
8783 | } | |
8784 | } | |
8785 | return 0; | |
8786 | } | |
8787 | ||
8788 | /* Given 'flow', a flow representing a packet received on 'ofproto', checks | |
8789 | * whether 'flow->in_port' represents a Linux VLAN device. If so, changes | |
8790 | * 'flow->in_port' to the "real" device backing the VLAN device, sets | |
8791 | * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is | |
8792 | * always the case unless VLAN splinters are enabled), returns false without | |
8793 | * making any changes. */ | |
8794 | static bool | |
8795 | vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow) | |
8796 | { | |
8797 | uint16_t realdev; | |
8798 | int vid; | |
8799 | ||
8800 | realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port, &vid); | |
8801 | if (!realdev) { | |
8802 | return false; | |
8803 | } | |
8804 | ||
8805 | /* Cause the flow to be processed as if it came in on the real device with | |
8806 | * the VLAN device's VLAN ID. */ | |
8807 | flow->in_port = realdev; | |
8808 | flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI); | |
8809 | return true; | |
8810 | } | |
8811 | ||
8812 | static void | |
8813 | vsp_remove(struct ofport_dpif *port) | |
8814 | { | |
8815 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto); | |
8816 | struct vlan_splinter *vsp; | |
8817 | ||
8818 | vsp = vlandev_find(ofproto, port->up.ofp_port); | |
8819 | if (vsp) { | |
8820 | hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node); | |
8821 | hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node); | |
8822 | free(vsp); | |
8823 | ||
8824 | port->realdev_ofp_port = 0; | |
8825 | } else { | |
8826 | VLOG_ERR("missing vlan device record"); | |
8827 | } | |
8828 | } | |
8829 | ||
8830 | static void | |
8831 | vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid) | |
8832 | { | |
8833 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto); | |
8834 | ||
8835 | if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL) | |
8836 | && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid)) | |
8837 | == realdev_ofp_port)) { | |
8838 | struct vlan_splinter *vsp; | |
8839 | ||
8840 | vsp = xmalloc(sizeof *vsp); | |
8841 | hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node, | |
8842 | hash_int(port->up.ofp_port, 0)); | |
8843 | hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node, | |
8844 | hash_realdev_vid(realdev_ofp_port, vid)); | |
8845 | vsp->realdev_ofp_port = realdev_ofp_port; | |
8846 | vsp->vlandev_ofp_port = port->up.ofp_port; | |
8847 | vsp->vid = vid; | |
8848 | ||
8849 | port->realdev_ofp_port = realdev_ofp_port; | |
8850 | } else { | |
8851 | VLOG_ERR("duplicate vlan device record"); | |
8852 | } | |
8853 | } | |
8854 | ||
8855 | static uint32_t | |
8856 | ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port) | |
8857 | { | |
8858 | const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port); | |
8859 | return ofport ? ofport->odp_port : OVSP_NONE; | |
8860 | } | |
8861 | ||
8862 | static struct ofport_dpif * | |
8863 | odp_port_to_ofport(const struct dpif_backer *backer, uint32_t odp_port) | |
8864 | { | |
8865 | struct ofport_dpif *port; | |
8866 | ||
8867 | HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node, | |
8868 | hash_int(odp_port, 0), | |
8869 | &backer->odp_to_ofport_map) { | |
8870 | if (port->odp_port == odp_port) { | |
8871 | return port; | |
8872 | } | |
8873 | } | |
8874 | ||
8875 | return NULL; | |
8876 | } | |
8877 | ||
8878 | static uint16_t | |
8879 | odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port) | |
8880 | { | |
8881 | struct ofport_dpif *port; | |
8882 | ||
8883 | port = odp_port_to_ofport(ofproto->backer, odp_port); | |
8884 | if (port && &ofproto->up == port->up.ofproto) { | |
8885 | return port->up.ofp_port; | |
8886 | } else { | |
8887 | return OFPP_NONE; | |
8888 | } | |
8889 | } | |
8890 | static unsigned long long int | |
8891 | avg_subfacet_life_span(const struct ofproto_dpif *ofproto) | |
8892 | { | |
8893 | unsigned long long int dc; | |
8894 | unsigned long long int avg; | |
8895 | ||
8896 | dc = ofproto->total_subfacet_del_count + ofproto->subfacet_del_count; | |
8897 | avg = dc ? ofproto->total_subfacet_life_span / dc : 0; | |
8898 | ||
8899 | return avg; | |
8900 | } | |
8901 | ||
8902 | static double | |
8903 | avg_subfacet_count(const struct ofproto_dpif *ofproto) | |
8904 | { | |
8905 | double avg_c = 0.0; | |
8906 | ||
8907 | if (ofproto->n_update_stats) { | |
8908 | avg_c = (double)ofproto->total_subfacet_count | |
8909 | / ofproto->n_update_stats; | |
8910 | } | |
8911 | ||
8912 | return avg_c; | |
8913 | } | |
8914 | ||
8915 | static void | |
8916 | show_dp_rates(struct ds *ds, const char *heading, | |
8917 | const struct avg_subfacet_rates *rates) | |
8918 | { | |
8919 | ds_put_format(ds, "%s add rate: %5.3f/min, del rate: %5.3f/min\n", | |
8920 | heading, rates->add_rate, rates->del_rate); | |
8921 | } | |
8922 | ||
8923 | static void | |
8924 | update_max_subfacet_count(struct ofproto_dpif *ofproto) | |
8925 | { | |
8926 | ofproto->max_n_subfacet = MAX(ofproto->max_n_subfacet, | |
8927 | hmap_count(&ofproto->subfacets)); | |
8928 | } | |
8929 | ||
8930 | /* Compute exponentially weighted moving average, adding 'new' as the newest, | |
8931 | * most heavily weighted element. 'base' designates the rate of decay: after | |
8932 | * 'base' further updates, 'new''s weight in the EWMA decays to about 1/e | |
8933 | * (about .37). */ | |
8934 | static void | |
8935 | exp_mavg(double *avg, int base, double new) | |
8936 | { | |
8937 | *avg = (*avg * (base - 1) + new) / base; | |
8938 | } | |
8939 | ||
8940 | static void | |
8941 | update_moving_averages(struct ofproto_dpif *ofproto) | |
8942 | { | |
8943 | const int min_ms = 60 * 1000; /* milliseconds in one minute. */ | |
8944 | ||
8945 | /* Update hourly averages on the minute boundaries. */ | |
8946 | if (time_msec() - ofproto->last_minute >= min_ms) { | |
8947 | exp_mavg(&ofproto->hourly.add_rate, 60, ofproto->subfacet_add_count); | |
8948 | exp_mavg(&ofproto->hourly.del_rate, 60, ofproto->subfacet_del_count); | |
8949 | ||
8950 | /* Update daily averages on the hour boundaries. */ | |
8951 | if ((ofproto->last_minute - ofproto->created) / min_ms % 60 == 59) { | |
8952 | exp_mavg(&ofproto->daily.add_rate, 24, ofproto->hourly.add_rate); | |
8953 | exp_mavg(&ofproto->daily.del_rate, 24, ofproto->hourly.del_rate); | |
8954 | } | |
8955 | ||
8956 | ofproto->total_subfacet_add_count += ofproto->subfacet_add_count; | |
8957 | ofproto->total_subfacet_del_count += ofproto->subfacet_del_count; | |
8958 | ofproto->subfacet_add_count = 0; | |
8959 | ofproto->subfacet_del_count = 0; | |
8960 | ofproto->last_minute += min_ms; | |
8961 | } | |
8962 | } | |
8963 | ||
8964 | static void | |
8965 | dpif_stats_update_hit_count(struct ofproto_dpif *ofproto, uint64_t delta) | |
8966 | { | |
8967 | ofproto->n_hit += delta; | |
8968 | } | |
8969 | ||
8970 | const struct ofproto_class ofproto_dpif_class = { | |
8971 | init, | |
8972 | enumerate_types, | |
8973 | enumerate_names, | |
8974 | del, | |
8975 | port_open_type, | |
8976 | type_run, | |
8977 | type_run_fast, | |
8978 | type_wait, | |
8979 | alloc, | |
8980 | construct, | |
8981 | destruct, | |
8982 | dealloc, | |
8983 | run, | |
8984 | run_fast, | |
8985 | wait, | |
8986 | get_memory_usage, | |
8987 | flush, | |
8988 | get_features, | |
8989 | get_tables, | |
8990 | port_alloc, | |
8991 | port_construct, | |
8992 | port_destruct, | |
8993 | port_dealloc, | |
8994 | port_modified, | |
8995 | port_reconfigured, | |
8996 | port_query_by_name, | |
8997 | port_add, | |
8998 | port_del, | |
8999 | port_get_stats, | |
9000 | port_dump_start, | |
9001 | port_dump_next, | |
9002 | port_dump_done, | |
9003 | port_poll, | |
9004 | port_poll_wait, | |
9005 | port_is_lacp_current, | |
9006 | NULL, /* rule_choose_table */ | |
9007 | rule_alloc, | |
9008 | rule_construct, | |
9009 | rule_destruct, | |
9010 | rule_dealloc, | |
9011 | rule_get_stats, | |
9012 | rule_execute, | |
9013 | rule_modify_actions, | |
9014 | set_frag_handling, | |
9015 | packet_out, | |
9016 | set_netflow, | |
9017 | get_netflow_ids, | |
9018 | set_sflow, | |
9019 | set_ipfix, | |
9020 | set_cfm, | |
9021 | get_cfm_status, | |
9022 | set_bfd, | |
9023 | get_bfd_status, | |
9024 | set_stp, | |
9025 | get_stp_status, | |
9026 | set_stp_port, | |
9027 | get_stp_port_status, | |
9028 | set_queues, | |
9029 | bundle_set, | |
9030 | bundle_remove, | |
9031 | mirror_set, | |
9032 | mirror_get_stats, | |
9033 | set_flood_vlans, | |
9034 | is_mirror_output_bundle, | |
9035 | forward_bpdu_changed, | |
9036 | set_mac_table_config, | |
9037 | set_realdev, | |
9038 | }; |